University of Southern California Dissertations and Theses

Determinants of the age at natural menopause: The multiethnic cohort

(USC Thesis Other)

Determinants of the age at natural menopause: The multiethnic cohort

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content DETERMINANTS OF THE AGE AT NATURAL MENOPAUSE: THE MULTIETHNIC COHORT by Katherine DeLellis Henderson A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (EPIDEMIOLOGY) August 2005 Copyright 2005 Katherine DeLellis Henderson Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3196812 Copyright 2005 by Henderson, Katherine DeLellis All rights reserved. INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3196812 Copyright 2006 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DEDICATION This dissertation is dedicated to my family. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ACKNOWLEDGEMENTS I would like to take a moment to thank the members of my committee for mentoring me through this process and giving me the opportunity to work with these data. Thanks also go to my colleagues in the USC Department of Preventive Medicine for creating a healthy and collaborative learning environment. Special thanks go to the participants in the Hawaii and Los Angeles Multiethnic Cohort, for their continuing generosity. Finally, thank you to my family, and my husband in particular, for their incredible patience and support. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. TABLE OF CONTENTS DEDICATION........................................................................................ ii ACKNOWLEGEMENTS..................................................................... iii LIST OF TABLES.................................................................................. vi LIST OF FIGURES.............................................................................. viii ABSTRACT............................................................................................ ix INTRODUCTION TO THE DISSERTATION................................... 1 PART I. LITERATURE REVIEW Introduction to the Literature Review................................................ 7 Chapter 1. Descriptive epidemiology of the age at natural m enopause...................................................................... 14 Chapter 2. Critical evaluation of the observational/ analytic epidemiologic studies of the age at natural menopause....................................................................... 17 a. Cross-sectional studies..................................... 18 b. Prospective studies...........................................43 c. Genetic Studies.................................................. 70 PART II. DATA ANALYSES Chapter 3. Determinants of the age at natural menopause: The Multiethnic Cohort................................................. 81 Chapter 4. Haplotype-based association analysis for three candidate genes and age at natural menopause: CYP17, HSD17B1 and IGF1.......................................... 108 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. PART III. GRANT PROPOSAL Chapter 5. Introduction to the Grant Proposal.............................. 136 Chapter 6. The Grant Proposal......................................................... 165 ALPHABETIZED BIBLIOGRAPHY.................................. 201 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF TABLES 1. Cross-sectional studies of determinants of age at natural menopause...................................................................................... 31 2. Prospective studies of determinants of age at natural menopause...................................................................................... 46 3. Genetic studies of determinants of age at natural menopause...................................................................................... 76 4. Distribution of characteristics by racial/ ethnic group, and univariate Cox proportional hazards ratios for all subjects.......... 95 5. Multivariate Cox proportional hazards ratios of age at natural menopause............................................................................................ 96 6. Multivariate Cox proportional hazards ratios of age at natural menopause, by race/ethnicity...........................................................97 7. Minimum detectable relative risks with 80% power (1 control per case, 350 controls).....................................................117 8. The Haplotype Approach in Brief..............................................124 9. IGF1 haplotype effects by block for all races combined......... 128 10. CYP17 haplotype effects by block for all races combined 129 11. HSD17B1 genotype effects for all races combined................ 130 12. HSD17B1 haplotype effects for all races combined............... 131 13. Example of haplotype characterization across two linked SNPs, in four hypothetical subjects.................................................143 vi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Grant Proposal Tables 1. Minimum Detectable Relative Risks with 80% Power (N=100,000)..........................................................................................178 2. Minimum Detectable Relative Risks with 80% Power (1 Control per Case, 500 Controls)...................................................181 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF FIGURES 1. Block structure and position of htSNPs for the three candidate genes in this study [(A) IGF1, (B) CYP17 and (C) HSD17B1].............................................................................................122 2. The three inhibin genes and their five dimers........................... 138 3. Candidate gene positions, sizes and visuals of chormosomal location........................................................................148 4. Detailed representation of (A) INHa, (B) INH(3A, (C) INHpB and (D) FST gene regions, with location of working SNPs in region, known adjacent genes and conservation across species [created via the UCSC genome browser (http: / / www. genome, ucsc .edu / cgibin / hgGate way]..................151 5. Preliminary block structure for (A) INHa, (B) INHpA, (C) INHpB and (D) FST............................ 157 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ABSTRACT Epidemiological studies have shown that delayed age at natural menopause (age 55+) confers a roughly 2-fold increased risk of breast cancer, compared to women with early menopause (< 45). Identification of factors that determine the timing of natural menopause should provide insight into the complex etiology of breast cancer. The epidemiological literature on the determinants of age at natural menopause has verified few determinants of age at natural menopause. This literature, in combination with results from a cross-sectional analysis of data from the Hawaii and Los Angeles Multiethnic Cohort Study (MEC) showing that age at natural menopause differs in a statistically significant manner across racial/ ethnic groups even after adjustment for covariates, provides support for the hypothesis that age at natural menopause is driven by a combination of reproductive, lifestyle and genetic factors. We selected three genes in two candidate pathways for investigation using a haplotype-based approach to test for possible association with age at natural menopause and identified several associations which we intend to follow up in future, larger studies. We also submitted a postdoctoral fellowship grant application entitled "Reproductive, lifestyle and genetic determinants of the age at natural menopause: The Multiethnic Cohort". ix Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We proposed to perform a study of the reproductive and lifestyle correlates of age at natural menopause in the California Teachers Study Cohort. We also proposed to extend our genetic analysis in the MEC by applying the haplotype-based approach to genes in a novel candidate pathway, the inhibin/ activin pathway. These investigations have provided insight into this complex phenotype, and illuminated the need for further study. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. INTRODUCTION TO THE DISSERTATION This dissertation has been prepared in accordance with the requirements of the Doctor of Philosophy (Ph.D.) degree in Epidemiology of the Department of Preventive Medicine at the Keck School of Medicine of the University of Southern California. This dissertation is comprised of three components: Part I) a review paper on an area of epidemiological research; Part II) an independent and complete data analysis of data arising from an on-going epidemiological study; and Part III) a grant application to carry out a new epidemiological study. This dissertation presents an investigation of the environmental, reproductive and genetic determinants of the age at natural menopause among women participating in the MEC. Specific aims and rationale for each component of this dissertation are described below. For purposes of this dissertation, it is critical to understand the meaning of the term menopause, and to differentiate between natural and surgical menopause. Menopause is defined as the permanent cessation of menstruation, which is caused by a severe decline in ovarian follicular function. If menstruation ceases with normal aging, it is termed a natural menopause, and in the United States the average age at natural 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menopause is roughly 51.4 years (The North American Menopause Society 2002). Menopause may also occur as a result of the surgical removal of both ovaries; this is referred to as a surgical menopause, and is often accompanied by hysterectomy. Menopause can additionally occur as a result of medical treatment, such as chemotherapy or radiation treatment. Natural menopause is considered the event of interest for purposes of this dissertation. The foundation of the rationale for study of the age at natural menopause is its importance as a breast cancer risk factor. In the United States the average life expectancy for a woman has risen from 48 to 79 years over the period from 1900 to 2000 (NCHS 2005). Consequently, in the next century, healthcare infrastructures will face the impact of greater numbers experiencing health issues related to the menopausal transition and to the postmenopausal period. In particular, risk for breast cancer, the most common and second most deadly cancer among women in the United States (ACS 2003), is clearly associated with age of menopause. In 1972, Trichopoulos et al. (Trichopoulos, MacMahon et al. 1972) reported that women with late natural menopause (age 55+) had two times the risk for breast cancer compared to women experiencing natural menopause before age 45. In 1983 Pike et al. (Pike, Krailo et al. 1983) proposed a model for breast cancer risk that reflected the hormonal exposure to breast 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. tissue over the life-span, in which the age-related increase in risk for breast cancer decelerated distinctly at the menopause. This age-incidence curve and the role of age at natural menopause as a breast cancer risk factor have since been established firmly by further study [for review see (Bernstein 2002; Hankinson, Colditz et al. 2004)]. Part I is comprised of a comprehensive literature review on the topics covered in this dissertation and is divided into two sections. The descriptive epidemiology of the age at natural menopause is presented in Section 1. The second section of Part I is a critical evaluation of the observational and analytic epidemiologic studies of the age at natural menopause. The aim of presenting the descriptive epidemiology of the age at natural menopause is to provide basic descriptive data on the age at natural menopause in the United States and the world. Gaining perspective on the distribution of the age at natural menopause across national and racial/ethnic boundaries not only provides a foundation for comparison of data from the MEC, but illuminates the clear need for further study. The critical evaluation of the observational and analytic epidemiologic studies of the age at natural menopause will describe and critique published studies within the major study design categories, including cross-sectional, prospective, and genetic studies. The genetic 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. studies include early observational studies, twin studies, single nucleotide polymorphism association studies and linkage studies. Evidence will be evaluated with respect to the hypothesis that age at natural menopause is a complex trait, driven by a combination of factors including lifestyle, reproductive and genetic factors. Part II is an independent and complete analysis of data arising from an on-going epidemiological study, the MEC study. Part II is composed of two sections. Section 1 is entitled, "Determinants of age at natural menopause: The Multiethnic Cohort." This section shares results of an investigation of the determinants of the age at natural menopause in the MEC. A survival analysis approach was used to assess which lifestyle and reproductive factors, as self-reported on the MEC baseline questionnaire, were associated with age at natural menopause. Importantly, the results of this investigation serve to validate the use of MEC data for study of this phenotype. In addition, this investigation was intended to contribute to the literature on the determinants of the age at natural menopause in assessing whether an independent effect of racial/ethnic group exists in age at natural menopause in the MEC. The results of this analysis provide support for the hypothesis that age at natural menopause is a complex trait governed by a set of reproductive, lifestyle and genetic factors. 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The investigation of the role of common genetic variants in determination of age at natural menopause using data from the MEC is discussed in Part II, Section 2. We report the results of a haplotype-based association analysis of previously characterized candidate genes (CYP17, HSD17B1 and IGF1) and age at natural menopause in the MEC. We have used a haplotype-based approach to evaluate evidence for possible association between common variation in three candidate genes and age at natural menopause. Part III of this dissertation is comprised of two sections, an introduction to the grant proposal and the grant proposal itself. The introduction to the grant proposal covers three important aspects of the proposal, the rationale for the candidate gene selection, a detailed overview of the proposed methodology, and the preliminary data that has been accumulated to support this proposal. The grant proposal, which has been submitted for consideration for a Susan G. Komen Breast Cancer Foundation postdoctoral award, was entitled "Reproductive, lifestyle and genetic determinants of the age at natural menopause." The aims of the grant proposal were: 1) to investigate the association between reproductive and lifestyle factors and age at natural menopause in a large cohort of women, the California Teacher's Study (CTS), and 2) to determine whether sequence variation within candidate genes in the 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. inhibin/activin pathway, m odulates the timing of natural menopause in the MEC. An unprecedented num ber of women are exceeding menopausal age, and postmenopausal life-spans are increasing. Age at natural menopause has increasingly important implications for postmenopausal health. In particular, the association between late natural menopause and increased risk for breast cancer is an important public health concern, and identification of the factors that determine timing of natural menopause, including any common genetic variants, will provide insight into the complex etiology of this important relationship. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. PART I. LITERATURE REVIEW Introduction to the literature review: A brief history of reproductive medicine and the biology of the menopause. The hum an reproductive system has been a matter of practical and academic interest for centuries. A Hindu text written between the 14th and 13th century B.C., the Manava-Dharma-Sastra, described two of the prominent contradictory theories on how human offspring arise. One theory held that humans were created from the uniting of two seeds, one from each parent, each containing both masculine and feminine principles. The second theory was that the whole seed was contributed by the male, and the female contributed the nurturing vessel. The continuation of this debate is evident in the 5th century B.C., in writings of members of the Grecian Phythagorean School. As the science of anatomy developed, the biology of human reproduction progressed. Vesalius (-1514-1564 A.D.), described the female uterine tubes in detail and compared them with the semen-conveying ducts of the male. Their true function was not understood until 1672 when the Dutch scientist Reinier De Graaf (1641-1673) described the presence of follicles, fallopian tubes, and the source of the follicles which we now call ovaries. (Alexandre 2001). 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In 1787 a London surgeon named John Hunter removed one ovary from a sow and observed its subsequent reproductive performance, comparing it with that of a sow with two ovaries. Hunter observed that the sow with one ovary produced approximately one half of the number of offspring as did the sow with both ovaries (Finn 2001). In 1958 McLaren and Biggers repeated and expanded on John Hunter's 1787 sow experiment in mice. In agreement with Hunter's findings, hemiovariectomised mice gave birth on average to half as many offspring as intact animals (Biggers, Finn et al. 1962; Biggers, Finn et al. 1962). McLaren and Biggers also observed a large variation in the age at which female mice stopped producing litters, as well as in the number and size of litters produced (Biggers, Finn et al. 1962; Biggers, Finn et al. 1962). A distinct pattern of litter size over time was observed in mice. The second litter was bigger than the first litter. A plateau period occurred in the middle of the reproductive lifespan when litter size was maximal and constant. After this plateau period followed a period of reproductive decline in which litters grew progressively smaller. Mice then experienced a period of reproductive senescence before death. The hemiovariectomised mice produced litters that were 75% of the size of intact animals during the plateau period, and their reproductive life spans were shorter. Autopsies showed that termination of litter production was 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. not due to lack of ovulation or fertilization but rather to the inability of the uterus to support an embryo. Estrus cycling and ovulation were evident, although progressively more irregular over time, well past the cessation of litter production (Finn 2001). The end of the hum an reproductive life span came about for different reasons. Unlike in mice, hum an ovaries became depleted of ova about two thirds of the way through life (Gosden, Laing et al. 1983). With the depletion of ova, ovulation ceased, the uterus ceased cycling, menstruation ceased and the menopause occurred. Several theories have arisen as to the evolutionary cause of early reproductive senescence experienced by humans. One theory centers on longevity models based on body size. Animals with a smaller body size have a larger surface area to volume ratio, and thus m ust maintain higher metabolism to support homeostasis. Animals with higher metabolisms produce more oxidative free radicals which cause DNA damage, protein damage and eventually somatic breakdown. Humans may live longer than would be predicted by body size due to a reduced dependence on physiological processes for homeostasis, lower metabolisms are associated with decreased production of oxidative free radicals, less DNA damage and longer life. Thus the menopause may be the result of a recent extension of the human lifespan, because though humans are living 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. longer, our reproductive life spans are limited by a finite oocyte number (Finn 2001). Other researchers contend that the menopause arose as a result of natural selection favoring females who ceased reproducing and focused energy on acquiring resources and protecting existing offspring. This is similar to another argument which centers on the impact of the mother. In evolution, certain species exchanged encephalization for altriciality, meaning they evolved by increasing brain capacity, but in exchange sacrificed infant independence from the mother. This exchange led to evolutionary selection for mothers who stopped reproducing, ensuring survival of existing offspring [for review see (Peccei 1995)]. During the last century the menopause has been characterized as a complex biological event driven by a network of pathways in the female hum an involving the central nervous, endocrine and reproductive systems. Defined generally as the endpoint of a woman's reproductive period of life, the menopause technically refers to the cessation of ovulatory function. The term menopause is derived from the two Latin words for month (mensis) and cessation (pausis). It is now known that primordial follicles are present at week 16 of hum an gestation. By week 20 the number of developing follicles reaches its lifetime peak of between 6 and 7 million. Through attrition, thought to 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. be driven by an apoptotic process called atresia, this num ber declines to around 1 million at birth, and 300,000 at puberty (Block 1952; Block 1953). At puberty, follicle attrition begins to occur through a second process, entry into the cycle of growth. A follicle recruited into the growth cycle may either grow to gain dominance over the other follicles in its cohort, becoming the ovulatory follicle, or undergo atresia. Through these processes, the follicle pool becomes depleted with age, and it has been hypothesized that it is this follicle attrition which causes menstrual cycles to cease permanently by age 45-54 in well nourished societies (Gosden 1987; Faddy and Gosden 1996). It is thought that the rate of follicle loss actually accelerates around age 37, leaving the follicle pool virtually exhausted by age 51 (Richardson, Senikas et al. 1987; Faddy, Gosden et al. 1992; Gougeon, Ecochard et al. 1994). Thus, it has been hypothesized that the age at which the menopause occurs should be driven by two things: 1) the original number of high quality oocytes in the follicle pool at some critical time point, and 2) the rate of atresia to which follicles are subjected (Cramer, Xu et al. 1995). Neither of these factors is well understood, however some insight has been gained in recent years, particularly in the area of follicle atresia. Follicle atresia has been shown to occur through a highly regulated apoptotic process (Hsueh, Billig et al. 1994). It is thought that 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the "normal" path for a follicle is to undergo atresia, unless it is rescued by certain survival factors. Follicle rescue is accomplished by hormones like follicle stimulating hormone (FSH), as well as by growth factors such as insulin-like growth factor 1 (IGF1) and activin [see (Gougeon 1998)]. Non growing follicles are thought to be recruited into the growth phase at a rate of 10 to 30 per day by some still unidentified process (Driancourt, Cahill et al. 1985), at which point they begin to grow, eventually acquiring dependency on hormonal support from FSH (Gougeon 1984). During each normal reproductive cycle, FSH rises during the early follicular phase, stimulating growth in these small growing follicles. FSH stimulates follicular secretion of estradiol by upregulating secretion of androgens by theca cells and inducing the aromatase enzyme in granulosa cells (Armstrong, Goff et al. 1979). FSH induces follicle expression of FSH receptors (Richards 1980), which is thought to play a critical role in the dominance attainment of one preovulatory follicle. As FSH levels decline in the mid to late follicular phase, growing follicles compete for FSH. The preovulatory follicle has some advantage, such as a more effective FSH receptor or tighter control of gene regulation; the dominant follicle has the ability to synthesize high levels of estradiol, enabling continued growth in an increasingly FSH-poor environment. Non-preovulatory follicles are those which cannot compete. They cease to develop and undergo atresia. 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Meanwhile the preovulatory or dom inant follicle is able to mature fully, secreting high levels of estradiol, and once the biological threshold for circulating concentration of estradiol is reached, positive feedback is exerted on the hypothalamus, stimulating the leutinizing hormone (LH) surge and ovulation [for review see (Gougeon 1996)]. The prevailing view of the cause of menopause has been centered on the exhaustion of ovarian follicles. Recent work has illuminated some of the processes regulating the rate of this exhaustion. In particular, the hypothalamic-pituitary-ovarian axis has been shown to be a key regulator of the normal reproductive cycle. The exact etiology of menopausal timing, though still unknown, is certain to become clearer as factors such as the inhibins, activin and follistatin are characterized more fully. 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Literature Review: Section 1. Descriptive epidemiology of the age at natural menopause. A review of the epidemiologic literature on determinants of age at natural menopause provides perspective on the distribution of the trait across national and racial/ethnic boundaries. This provides a foundation for comparison of data from current studies in the MEC, and illuminates the clear need for further study. The age-related decline in female fertility is thought to be relatively consistent among all races and ethnicities. In the developed world, studies have shown that fecundity begins to decline around age 35, and in populations of women in which social behavior and contraception have not interfered with "natural" reproduction patterns, the birth of the last child occurs roughly ten years prior to the menopause [for review see (te Velde and Pearson 2002)]. The perimenopause, or point in time when periods become irregular, has been shown to occur at a mean of roughly 45 to 46 years of age (Richardson, Senikas et al. 1987). The menopause itself has been reported to occur at a mean age of 50 to 51 years (Brambilla and McKinlay 1989; Whelan, Sandler et al. 1990; Bromberger, Matthews et al. 1997; van Noord, Dubas et al. 1997; Kato, Toniolo et al. 1998; Cooper, Sandler et al. 1999). In a large international 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. study of 18,997 women from 11 countries the mean age at natural menopause was 51 years of age, with a range of 49 to 52 years of age (Morabia and Costanza 1998). Most of the current worldwide data on age at natural menopause is from White women living in developed countries; however increasing interest in understanding the global distribution of age at natural menopause has produced a few small studies from the developing world. Data on women living in Africa, Central America and Asia have provided some means by which to compare the age at natural menopause across racial/ethnic boundaries. Two studies of women living in South Africa and Ghana have reported that African women have an earlier mean age at natural menopause than the mean for Whites (Frere 1971; Kwawukume, Ghosh et al. 1993), however these reports were from small study populations, as was another African study that was done in Nigeria. This study found the mean age at natural menopause among a group of 563 Nigerian women of Yoruba descent to be 48.4 (Aina 1992). Women of Mexican descent were also found to have an earlier mean age at natural menopause when compared to the mean for White women (Garcia Vela, Nava et al. 1987). Among Asian women, Tamada et al. reported that Japanese women had a median age at natural menopause of 50.5 (Tamada and Iwasaki 1995), and Ismael showed that Malaysian women, 70% of 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. whom were of Malay descent, had a median age at natural menopause of 50.5 (Ismael 1994). These studies, while intended to provide information on the distribution of age at natural menopause in diverse populations, were generally done on small groups of women, used diverse sampling and analytic methods and cannot be used to make direct comparisons across racial/ethnic groups w ithout rigorous quantitative assessment. Information on age at natural menopause among globally diverse populations is sparse, however from the data that does exist; we might conclude that the age at natural menopause has a rough mean of 50 years of age. This dialogue is also informed by the detailed discussion of the cross-sectional, prospective and genetic studies looking at factors associated with age at natural menopause that follows. 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Literature Review: Section 2. Critical evaluation of the observational/analytic epidemiological studies of the age at natural menopause. Menopause is a complex event that is becoming more of a public health issue as life expectancies increase. More women will be living more years post menopause, when they are at increased risk for a host of late-life diseases, one of which is breast cancer. A critical evaluation of the epidemiologic studies of age at natural menopause will shed needed light on this important phenotype. In this section, the major studies within three epidemiological study design areas will be described and critically evaluated. These include cross-sectional, prospective, and genetic studies. Within the section on genetic studies we will evaluate the development of the literature investigating the hypothesis that age at natural menopause is a heritable trait. These studies primarily include twin studies, single nucleotide polymorphism association studies and linkage studies, one of which was a genome-wide scan. This section will provide epidemiologic background for discussing the current study of age at natural menopause. Existing literature on determinants of age at natural menopause will be presented and critiqued for strengths and limitations. Overall, evidence will support the 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. hypothesis that age at natural menopause is a complex trait, driven by a combination of factors, including lifestyle, reproductive and genetic factors. Cross sectional studies of determinants of age at natural menopause. This section presents results of a comprehensive review of the cross- sectional epidemiological literature on determinants of the age of natural menopause. The Ovid Technologies, Inc. search engine was used to search the MEDLINE database covering journal articles published from 1966 through the second week of March 2005. The search strategy was to request all cross-sectional hum an studies, written in the English language, with a topic of the "menopause," or with any one of three keywords: menopause, natural menopause or age of natural menopause. In addition, a search for papers referencing the MacMahon and Worcester (MacMahon and Worcester 1966) study using the Science Citation Index, was cross- referenced with the results of the MEDLINE survey. Of the roughly 200 articles initially identified, thirteen studies were determined to address the determinants of the age at natural menopause using a cross-sectional study design. 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In 1966 the National Center for Health Statistics, an office of the U.S. Department of Health, Education and Welfare, published a cross- sectional study of predictors of age at menopause in the United States from 1960-1962, based on a nationwide probability sample of women aged 18-79 years. The age at natural menopause was examined in relation to race, marital status, parity, family income, geographic region, height, and skin-fold measurement. Up to that time there had been few analyses of the effects of demographic and physiologic variables on age at natural menopause. Three studies had previously reported an association between increasing parity and increasing age at natural menopause, but a fourth study reported finding no such association. Socioeconomic status had previously been investigated as a possible determinant of age at natural menopause in one study, with no significant result; and the potential for age at menarche to be associated with age at menopause had already become a controversial issue [see (MacMahon and Worcester 1966)]. MacMahon and Worcester intended to provide estimates of the age distribution natural menopause for the civilian population of the United States, and carefully investigate the association between age at natural menopause and demographic, socioeconomic, and health-related factors using rigorous statistical methods and large numbers. For example, the questionnaire asked very simple questions such as (1) "Have 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. periods stopped?", and (2) "IF YES: Age when periods stopped The authors noted potential sources of bias in the analysis. One of these was the tendency of answers to the question of age at nature menopause to cluster at ages ending in 5 and 0. In order to prevent this bias, the authors ignored answers to question (2), instead using only answers from question (1), and expressed the number of women at each age who reported that they had had a natural menopause, as a percentage of the number who had not had an operative menopause by that age. This method has become known as the "status quo" methodology. While this method set a new standard for cross-sectional studies, the number of women in the analysis fell somewhat short of ideal. Initially 4211 women were selected to represent the civilian, non-institutional population of the United States between 18 and 79 years of age. However, 15% failed to respond to the survey, 2211 women were premenopausal, and 473 had had a surgical menopause, leaving only 897 women who had experienced a natural menopausal. While still large compared to most previous studies, this number was not large enough to provide power for complex multivariate regressions. In the MacMahon study, the association between nulliparity and age at natural menopause was directionally consistent with previous studies, in that nulliparous women appeared to experience natural 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menopause earlier than more parous women, but this association did not reach statistical significance. There was a suggestion, in these data, that age at natural menopause was earlier among African Americans versus Whites. In addition, Whites of low socioeconomic status had an earlier natural menopause than more affluent Whites. However the only significant difference found was between women in the leanest and heaviest skin-fold measurement categories. Leaner women reported earlier natural menopause (MacMahon and Worcester 1966). Because of its population-based subject selection and methodology, this study became the model for future cross-sectional investigation into the determinants of age at natural menopause. See Table 1 for a summary of cross-sectional studies discussed in this section. In June, 1977 Jick et al. (Jick, Porter et al. 1977) published results of an important study on cross-sectional data on two sets of hospital inpatients. The first group was from 24 Boston area hospitals, and the second was from a group of hospitals in seven countries, including the United States, Canada, New Zealand, Scotland, Germany, Israel and Italy. From the total numbers of 25,000 and 32,000 women in each study group, the number of naturally postmenopausal women between ages 44 and 53 with known smoking status were 2143 and 1391 respectively; these women were included in this analysis. They calculated the age-specific 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. proportions of women who were naturally postmenopausal for current smokers and non-smokers, and reported that the proportion of women who were naturally postmenopausal was higher among current smokers than among non-smokers. In addition, they reported that within each age category, there was a progressive increase in the proportion of naturally postmenopausal women with increased intensity of smoking. For example, in the age group 50 to 51 years, the percentages of naturally postmenopausal women were 74, 81 and 87% for never, 1 /2 pack per day and > 1 pack per day smokers respectively. Jick et al. reported that the effect of potential confounders on these results was minimal. Confounders examined included parity, marital status, type of admission, first discharge diagnosis, coffee, tea and alcohol intake. This study was the first to report an association between cigarette smoking and natural menopause. In early 1983, Krailo and Pike (Krailo and Pike 1983) published a discussion of the approach of MacMahon and Worcester (MacMahon and Worcester 1966), focusing on the impact of two major potential sources of bias in the cross-sectional study design. One potential source of bias was the high proportion of women undergoing surgical menopause, which could have biased the observable distribution of age at natural menopause in the study sample. The other potential source of bias was in the 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. reporting of age at natural menopause by women, who it seems were more likely to report an age of natural menopause ending in zero or five, a phenomenon known as terminal digit clustering (MacMahon and Worcester 1966). Krailo and Pike compared a competing risks approach to the "status quo" approach used by MacMahon. Suspecting possible bias from self-reported age at natural menopause, MacMahon and Worcester ignored those answers and instead relied only on the yes/no answers given. However they also chose to exclude women with a surgical menopause, which Krailo and Pike criticized in that it could have produced a biased distribution of age at natural menopause in the study sample. Krailo and Pike described a competing risks approach in which natural and surgical menopause were treated as separate outcomes. They concluded that in comparisons of the calculated age at natural menopause between different populations from previous studies, results were likely to be confounded very little by differential rates of surgical menopause. Nevertheless, Krailo and Pike raised important concerns, and ones which would convince the field that any analysis of age at natural menopause should be done with attention to such issues. Later in 1983 a study was published by Hoel et al. (Hoel, Wakabayashi et al. 1983) which examined whether distribution or trends in menarche, age at first birth, age at menopause or weight could explain 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. observed higher incidence of breast cancer in the United States versus the incidence in Japan. Subjects for this study were selected from the lowest exposure group of the cohort of atomic bomb survivors in Nagasaki and Hiroshima, Japan. This study used the life table derived distribution of reported age at natural menopause, and a competing risks analysis (Krailo and Pike 1983) to calculate the median age at natural menopause in consecutive birth cohorts, reporting that the median age at natural menopause increased 1.2 years between the 1880-1889 to 1910-1924 birth cohorts among Japanese (p<0.01). The medians were not different between birth cohorts in the U.S. sample, however there was a much smaller distribution of birth years represented among the U.S. women in this study (1880-1905). This study also reported a positive association between body weight and age at natural menopause. A small study on the effect of exposure to passive smoking on age at natural menopause, among 261 controls, provided further support for the smoking effect. In this study a conditional logistic regression approach, analogous to methods appropriate for a cohort study, were used. The event of interest was a natural menopause. Women who were premenopausal or had a surgical menopause were censored at age at questionnaire or at age of surgery respectively. Using this method, covariates could be investigated; in this study, race, education, year of 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. birth, and alcohol intake were not significant independent predictors of age at natural menopause. However the exposure of interest was found to be important. Women who were current smokers had a median age at natural menopause of 49.7 years, while non smokers had a median of 50.8. In addition, among non-smokers, women whose spouses smoked had a median age at natural menopause of 49.8 years, while non-smokers whose spouses did not smoke had a median of 51.9 years. The differences were not huge, though they were statistically significant. Stanford, Hartge et al. (Stanford, Hartge et al. 1987) reported the results of a similar analysis on a larger subject group. Subjects for this study were selected randomly from among the controls of a breast cancer case/control study. The final study group consisted of 983 premenopausal women, 1091 surgically menopausal women and 1423 naturally menopausal women. Again, a life-table approach was used, in which premenopausal and surgically menopausal women were censored at date of interview or date of surgery. Factors found to be significant in the univariate model were tested in a multivariate proportional hazards regression model. This study investigated a large number of factors, such as race (African American vs. White), age at menarche, age at first live birth, number of live births, regularity of menstrual periods, history of breast feeding, family history of breast cancer, height, weight, body mass 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. index (BMI), smoking, income, and education. In the final model, in order of relative significance, were num ber of live births, regularity of menstrual periods, income and education. Median age at natural menopause differed by about four months between African Americans and Whites in this study. This difference was not statistically significant, however it was in the same direction as in the MacMahon and Worcester study, in which the difference was also not significant. The effect of smoking was very small and not significant in this study, although again directionally consistent with previous studies. By this point, the prevailing view of the cause of m enopause was follicle pool depletion, and factors which were hypothesized to impact the number of follicles or the rate of follicle loss were logical candidates for association with age at natural menopause. The concept of incessant ovulation, which had been proposed as a predictor of ovarian cancer risk, was then extended to the age at natural menopause. The hypothesis, as it pertained to the natural menopause, was that factors which promoted regularity or frequency of ovulatory cycles would promote follicle loss and thus earlier natural menopause. In an effort to address this specific hypothesis, Cramer et al. (Cramer, Xu et al. 1995) investigated the reproductive histories of 344 "case" women, who had experienced early (< 46) natural menopause and 344 "controls", who were either still 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menstruating or had experienced a natural menopause after age 46. Using this kind of case/control approach, incorporating a logistic regression analytic technique, Cramer et al. found that early menarche, shorter menstrual cycles, and fewer pregnancies with live birth were associated with early natural menopause. Thus, results did seem to support the hypothesis that factors which led to more ovulatory cycles were associated with early natural menopause. The concept that age at natural menopause might be driven by differences in sociodemographic factors was targeted for study by Luoto et al. in 1994 (Luoto, Kaprio et al. 1994). This was an analysis of data from a questionnaire sent to a random sample of 2000 Finnish women aged 45 to 64 years. 1713 women responded and 1505 women, with complete data on critical variables, were included in the analysis. Life tables and proportional hazards regression were used here, with censoring of premenopausal and surgically menopausal women at age at questionnaire or age at surgery. In addition, women using oral contraceptives (OCs) or postmenopausal hormone replacement therapy (HRT) were censored at the age of questionnaire. In these data, women in white collar occupations, and those with more education were likely to have later natural menopause, even after adjustment in the multivariate model for smoking, any hormone use and BMI. 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In another study of factors associated with onset of menopause, Torgerson et al. (Torgerson, Avenell et al. 1994) used logistic regression with early natural menopause as the outcome of interest, to test for possible association with age, smoking, maternal menopausal age, parity, social class, meat intake and alcohol intake among women living in Aberdeen, Scotland. In a nod to the growing concern that differences in how women were defined as postmenopausal were creating bias in previous studies, Torgerson et al. chose to use two definitions in order to assess whether different definitions would have an impact on findings. Women were defined as naturally postmenopausal after (1) 6 and (2) 12 months of spontaneous amenorrhea. In this study women reporting use of OCs or HRT were excluded. In addition, women reporting a surgical menopause were excluded from this study. In the final study group of women aged 45 to 49 there were 1227 premenopausal women and 238 women who had experienced an early natural menopause. In a multivariate logistic regression, the following factors were associated with early natural menopause: older age, more smoking, lower age of maternal menopause, lower parity, lower social class, no alcohol use, and less meat consumption. This study marked the first report of the association of lower age of maternal menopause with early natural menopause, an important step in the development of the literature on the determinants of 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. age at natural menopause. Using 6 or 12 months of amenorrhea as the definition of menopause did not change the results of this study. From this point forward, the concept that age at natural menopause may be a heritable trait is a m atter of growing interest in the literature, especially after the first classical twin study was published on this topic in 1998 (Snieder, MacGregor et al. 1998), which showed in quantitative terms that age at natural menopause is likely to be a highly heritable trait. However observational studies based on prevalence data such as these cross-sectional studies often have no data to assess this aspect. In the study by Gold et al (Gold, Bromberger et al. 2001), a number of demographic and lifestyle factors were studied in a large multiethnic sample of women from seven study centers in the United States, the Study of Women's Health Across the Nation (SWAN). A total of 14,620 women were included in the study sample, from five racial/ ethnic groups. Each study center recruited Whites in addition to one other racial/ethnic group. African Americans were recruited from Boston, Chicago, Detroit and Pittsburgh. Chinese were recruited in Oakland and Japanese women were recruited in Los Angeles. The Hispanics, recruited in Newark, were a very diverse group; they were comprised of women with Mexican, Central American, South American, Cuban, Dominican, Puerto Rican, and Spanish ancestry. Analytical methods included use of the Cox 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. proportional hazards regression, treating surgically menopausal and premenopausal women as censored as described previously. In this study, HRT users were considered in two ways, first by exclusion and then by censoring at the age they started using hormones. Though results by each method were not different, the latter method was used for reporting because it provided a slightly better model fit. Current smoking, less education, being separated, widowed or divorced, non employment, and having a history of heart disease were associated with earlier natural menopause; while higher parity, OC use and Japanese race/ethnicity were associated with later natural menopause (Whites were baseline). An analysis of data from participants in the Third National Health and Nutrition Examination Survey (NHANES III) allowed for an innovative assessment of menopausal status, circulating FSH measurement greater than 20 IU /liter (Cooper, Baird et al. 2001). The purpose of this study was to assess the association of race/ ethnicity, demographic and lifestyle factors with early menopause based either on self-reported menstrual cycle patterns, or on elevated FSH levels. Roughly equal numbers of African American (N=548), Hispanic (N=543) and White (N=576) subjects were included in this analysis. 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 1. Cross sectional studies of determinants of age at natural menopause No. Authors, year Methods N = Conclusions 1 MacMahon et al., 1966 Status Quo Methods, excluded women with surgical menopause (s.m.) 897 Only significant difference between leanest and heaviest women, by msmt. of skinfold. Geaner=earlier). Suggestion that AA earlier than W, and W of low SES earlier than more affluent W. 2 Jick et al., 1977 Status Quo Methods by smoking status, excluded women with s.m. 3534 Proportion of women who were postmenopausal (naturally) was higher among current smokers then non-smokers. Significant dose-response wtih smoking density. 3 Hoel et al., 1983 Competing Risks Analysis. Lfe-table derived distribution of reported age at natural menopause (n.m.). s.m. considered as independent competing risk. 21,000 0), 3,600 (W) Median age at natural menopause increased 1.2 years between extreme birth cohorts (1880- 1889 to 1910 to 1924) among J. Significant positive association between body weight and age at natural menopause. 4 Everson et al., 1986 Life-table derived distribution, "survival analysis" approach, event = n.m., premenopausal or surgically menopausal censored at age at qx or age at surgery respectively. 261 Race, education, year of birth, alcohol intake were not important, but passive smoking was important. 5 Stanford et al., 1987 Life-table derived distribution, "proportional hazards regression" approach, event = n.m., premenopausal or surgically menopausal censored at age at qx or age at surgery respectively. 983 pm, 1091 sm, 1423 nm Number of live births, regularity of menstrual periods, income and education were important. AA earlier than W by 4 months (not sig). No effect with smoking or BMI. 6 Cramer et al., 1995 Case/control approach (outcome = early menopause), logistic regression. 344 ca, 344 co Early menarche, shorter cycles, fewer pregnancies with live birth = early nm (Adjusted for smoking and BMI). 7 Luoto et al., 1994 Life-table derived distribution, "proportional hazards regression" approach, event = n.m., premenopausal or surgically menopausal censored at age at qx or age at surgery. *Also, women usine OCs or HRT were censored at aee at ax. 1505 Higher occupation, more education=later (adjusted for smoking, any hormone use, BMI and age at first full term pregnancy) 8 Torgerson et al., 1994 Logistic regression to predict menopausal status, early natural menopause = positive outcome. Women using OCs or HRT were excluded. *Also tested 2 definitions of n.m. (6+, 12+) 1227 pm, 238 nm (aged 45- 49) Early n.m. was associated with: older age, more smoking, lower age at maternal menopause, lower parity, lower social class, no alcohol use, less meat consumption. No difference in results by definition of n.m. 9 Gold et al., 2001 Life-table derived distribution, "Cox Prop Hazards Regression" *Also tested treatment of HRT users: 1) exclusion, 2) censoring at age started using hormones. SWAN study, N=14,620 Early n.m. was associated with current smoking, less education, being separated/ divorced/ widowed, non-employment, history of heart disease. Late n.m. was associated with higher parity, OC use, Japanese race/ethnicity. HRT methods not different, latter gave better model fit. 10 Cooper et al., 2001 Compared self reports with FSH msmts. Logistic regression with early nm outcome of interest, model adjusted for age, smoking, and unilateral oophorectomy. NHANES III. 548 AA, 543 L ,576 W Higher BMI was protective for early nm using FSH criteria, but not menstrual cycle self- reports. No association with ethnicity, education, age at menarche, number of live births, or OC use. Though AA suggestion earllier vs. W. 11 Lawlor et al., 2003 Multiple linear regressions used, age at n.m. continuous. Excluded HRT users and s.m. The main question was whether early life SES = early n.m. 3513 Most of the ten indicators of adverse SES were associated with early n.m. (adj for smoking and BMI) 12 Akldna et al, 2004 Multiple linear regressions used, age at n.m. continuous. Excluded women with s.m. HRT use ignored? 219 Pesticide levels in serum were significantly negatively associated with age at n.m. A logistic regression, adjusted for age, and smoking was performed, treating early menopause as the outcome of interest. Results of this study showed that higher BMI was associated w ith lower likelihood of elevated FSH, but this association was not seen with the self- reported menstrual measure of menopausal status. More exercise was associated with a lower likelihood of being postmenopausal on the basis of menstrual and hormonal measures. Early natural menopause was not significantly associated with ethnicity, education, age at menarche, number of live births, or OC use. African Americans showed some tendency to experience natural menopause earlier than Whites, however this was not significant as previously stated. One important limitation of this study was that the day of cycle was not controlled for during blood draw, detracting from the accuracy of the FSH measurement scheme and possibly producing a non-differential misclassification problem. To address this problem the authors excluded women with an LH:FSH ratio above 2.0, which would decrease the likelihood of misinterpreting an ovulation-related increase in FSH. This type of study was expensive, and thus not practical on a larger scale, however it did highlight the problem that in women with high BMI, some misclassification in menopausal status may occur, if status is determined based on menstrual cycle characteristics. The authors speculated that obese women may be 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. classified as menopausal, while remaining hormonally premenopausal. This may account for why an association between higher BMI and later age at natural menopause has not been seen consistently across studies. The question of whether age at natural menopause is a heritable trait is an important one, and will be considered further below, however two additional cross-sectional studies, which did not address this issue directly, warrant mention. In 2003, Lawlor et al. (Lawlor, Ebrahim et al. 2003) published a study assessing the possible association between indicators of adverse SES across the entire life course with early age at natural menopause. This data was based on the theory that nutritional factors throughout life, particularly in childhood, may affect the follicle pool and thus the age at natural menopause. Of ten indicators of adverse SES available, most were associated with early natural menopause. The magnitude of these associations were not changed by adjustment for other reproductive factors, but were attenuated to some degree by smoking and BMI. Finally, in this recent climate of environmental activism it is interesting to note one additional study. Akkina et al. (Akkina, Reif et al. 2004) studied the association between age at natural menopause and exposure to organochlorine pesticides (OCP) in a small group of 219 Hispanic women. Analysis of variance was used to assess mean age at Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. natural menopause among women by category of serum OCP level. Pesticide levels, measured in serum, were significantly negatively associated with age at natural menopause in these data. Women in the highest category of three pesticide types had 5.7, 3.4 and 5.2 years earlier natural menopause, respectively, versus the lowest exposed group for each type of pesticide. These are very large differences in light of the relatively small differences reported previously, and further work in this area is warranted. These cross-sectional studies have offered much valuable information regarding the factors associated with age at natural menopause. Factors such as smoking and parity seem well established after thirty years of cross-sectional study. However these cross-sectional studies have failed to confirm or reject whether factors such as age at menarche, BMI, and race/ ethnicity are important in determining age at natural menopause. In part this is because of small sample sizes and diverse statistical methods. But even if all studies were large, and all methods were robust, a cross-sectional study could not prove causality because of the lack of appropriate temporal relationship between exposures and outcome. 34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Critique of existing cross-sectional literature on determinants of age at natural menopause. Studies investigating the age at natural menopause have been plagued historically by a range of epidemiologic study design issues which have hindered the proper assessment of the determinants of the age at natural menopause and of the relative ages at natural menopause across racial/ ethnic strata. First, classification of the postmenopausal woman has been done inconsistently across studies. Investigators have defined a woman as postmenopausal when she has experienced an interval of at least 6, 9 or 12 months free of menstruation. Menopausal age is then defined as current age minus the menstruation-free months to estimate the age at which the last menstrual period occurred. This lower age is then used in computation and thus the median age in some studies is the median age at which the last menstrual period occurred. In other studies the median menopausal age is the age at which 50 percent of women actually recognized that menstruation had ceased. Next, many studies have estimated menopausal age from samples of convenience using hospital patients as study subjects. Studies of healthy women are critical to proper age at natural menopause assessment and avoidance of selection bias. Bias is an issue when consistent recall of 35 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the age at menopause is suspect, perhaps in studies requiring self- reporting via a questionnaire or an interview. There is evidence that naturally menopausal women are more prone to terminal digit clustering than are women reporting a surgical menopause, for example. Another challenge in ascertainment of age at natural menopause results from difficulty in assessment of type of menopause and thus, age at surgical menopause. Technically, surgical menopause results from the removal of both ovaries. However many women cannot accurately report the type of surgery she has undergone. If a woman who has undergone a bilateral oophorectomy, but still has her uterus, is placed on hormone replacement therapy, she may experience bleeding, and as a result incorrectly report that she is premenopausal. Conversely, a woman may report having a surgical menopause but still have one or even two ovaries, having undergone a unilateral oophorectomy or simple hysterectomy. Investigators have dealt with these issues in various ways, some by excluding all current hormone users, and some by censoring an HRT user's age at menopause at her reported age when she began using the hormones. These issues have resulted in complication of interpretation of results and reduced comparability across studies. Two final important limitations of studies in this field are the lack of data on demographic and physiologic variables which might have an effect on the age at natural 36 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menopause and the paucity of racially and ethnically diverse subjects in most studies. The MacMahon study (MacMahon and Worcester 1966) was the first study to actively address many of these issues. The problem of samples of convenience was addressed by using a nationwide probability sampling scheme, intended to provide an accurate assessment of age at natural menopause that would be generalizable to civilian, non institutionalized women living in the United States. In addition, in using what would be called the "status quo" methodology, they did not directly address the potential biases resulting from exclusion of women with surgical menopause or HRT users, but they did set the stage for methodology that would better address these problems. The application of a competing risks analysis (Krailo and Pike 1983) to the analysis of cross-sectional age at natural menopause data, provided a methodological framework that many studies would build on (Hoel, Wakabayashi et al. 1983; Everson, Sandler et al. 1986; Stanford, Hartge et al. 1987; Luoto, Kaprio et al. 1994; Gold, Bromberger et al. 2001). The study by Hoel et al. (Hoel, Wakabayashi et al. 1983) demonstrated successful use an important methodological development, the competing risks analysis, and thereby addressed the issue of possible selection bias created by exclusion of women who had undergone a surgical 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menopause. This is an important issue because of the large proportion of women, up to 30% in some groups, who have experienced a surgical menopause in the developed world. Exclusion of these women from analyses of age at natural menopause may result in a biased distribution of age at natural menopause in the study sample. This would be particularly important in multiethnic studies because rates of surgical menopause have been shown to vary with racial/ethnic group. By implementing the competing risks analysis and censoring women who had a surgical menopause at their age at surgery, studies were maximizing the amount of data that could be gleaned from the sample and gaining a more representative distribution of the phenotype. This methodology set a new standard in this field. Another study using a version of the competing risks analysis . Everson et al. (Everson, Sandler et al. 1986) constructed a life-table derived distribution and used a "survival analysis" approach to regression. The event of interest was the natural menopause. Premenopausal or surgically menopausal women were censored at the age at questionnaire or age at surgery respectively. Very similar methods were used by Stanford et al. (Stanford, Hartge et al. 1987), and this was a much larger, multiethnic study. Luoto et al. (Luoto, Kaprio et al. 1994) used similar methodology, the Cox proportional hazards regression, and addressed the treatment of 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. HRT users by censoring them at the age at questionnaire. The issue of how to deal with HRT users had not been previously addressed nor investigated in a consistent manner, and this study highlighted this fact. Gold et al. (Gold, Bromberger et al. 2001) then addressed the issue directly by first excluding and then censoring HRT users, at age at questionnaire, and comparing results. The fact that there was no difference between results using either method indicated that HRT use was not a serious problem in these studies. The study by Gold et al. also found that Japanese, but no other ethnic group studied had a significant difference from Whites in age at natural menopause. Japanese women had later natural menopause versus White women after adjusting for other covariates. The Latinas did not differ significantly from Whites, but the Latina group was a very diverse group, leaving room for debate about the meaning of these findings. A handful of studies used a case/control design to investigate the factors associated with age at natural menopause (Cramer, Xu et al. 1995) (Torgerson, Avenell et al. 1994; Cooper, Baird et al. 2001). All three evaluated the risk of early natural menopause. Torgerson et al. tested for difference in results by definition of menopause, 6+ months of amenorrhea versus 12+ months, and found no significant difference in results using the two methods. Cooper et al. addressed the issue of definition of menstrual 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. status, but in a different way. Self-reported menstrual status based on menstrual cycling characteristics were compared w ith menstrual status defined by serum FSH measurements. Cooper et al. reported that higher BMI was protective for early natural menopause using the FSH criteria, which was in line with previous studies. But they saw no relationship between BMI and early natural menopause based on the self report data. This indicates that obese women, who may experience obesity induced amenorrhea, may report themselves to be postmenopausal, when in fact they are hormonally premenopausal. For this reason, particularly in heavy populations, BMI might not appear to extend the reproductive phase. Other studies, such as those by Lawlor (Lawlor, Ebrahim et al. 2003) and Akkina (Akkina, Reif et al. 2004) used less ideal methods, but are worth mentioning in light of results. Using multiple linear regressions, Lawlor reported that adverse socioeconomic conditions early in life were associated with early natural menopause, after adjusting for smoking and BMI. This was a unique study, difficult to perform because of the type of data required; it offered an insight into the impact of nutrition and or hardship during early life on reproduction. In the Akkina study (Akkina, Reif et al. 2004), pesticide levels in serum were significantly negatively associated, via linear regression with age at 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. natural menopause. These two studies focused on two specific environmental factors, and though methods were not ideal, they offer important insights and raise questions for further investigation. In summary, in these cross-sectional studies of age at natural menopause there were three m ain approaches used. The first was the status quo method, first used by MacMahon and Worcester, and on which Krailo an Pike built the second approach used in these studies, the competing risks analysis. This competing risks analysis was then used in various forms by several studies as described above (Hoel, Wakabayashi et al. 1983; Everson, Sandler et al. 1986; Stanford, Hartge et al. 1987; Luoto, Kaprio et al. 1994; Gold, Bromberger et al. 2001). By using this competing risks methodology, investigators addressed several of the important methodologic issues. For example they censored women who had had a surgical menopause at the age at surgery, thereby gaining a more representative distribution of the phenotype than if they had excluded those women. Three other studies used a case/control design to investigate the factors associated with age at natural menopause as described above (Cramer, Xu et al. 1995) (Torgerson, Avenell et al. 1994; Cooper, Baird et al. 2001). In summary, methods used in these cross-sectional studies have some consistencies, particularly among the set of studies using the 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. competing risks approach. However many design factors vary across studies, such as selection scheme, number of subjects, and choice of potential covariates or confounders tested. But we can gain some sense of the important factors in determination of age at natural menopause from these studies. Smoking and parity are the most consistent factors. There is also a suggestion that BMI is important, but the investigation of BMI may have been impeded by misclassification of menstrual status due to obesity related amenorrhea. Some effect of socioeconomic status is also apparent; however the factors underlying this association remain to be identified. A racial/ethnic effect was suggested but not confirmed. However the suggestion that maternal age at natural menopause predicts the age at natural menopause of the offspring, provides support for the underlying hypothesis that age at natural menopause is a heritable trait, in lieu of significance levels for racial/ ethnic effects. Overall, the relationship between these factors and age at natural menopause cannot be confirmed by the cross-sectional study design due to the main limitation of the design. A causal relationship between covariates and outcomes cannot be ascertained with certainty due to the lack of appropriate temporal relationship between exposures and outcome. 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Prospective studies on determinants of age at natural menopause. In 1980 the World Health Organization (WHO) sponsored a meeting on the menopause that ultimately served to reveal how little was known globally about the determinants of its timing (WHO 1981). Work on this topic accomplished up to that point had been primarily cross-sectional in design and results had been inconsistent. One of the main themes derived from the 1980 meeting was the encouragement of future carefully designed prospective studies. Through a comprehensive search of the literature a number of publications were located on a variety of topics on issues related to the menopause, but only a few described carefully designed prospective studies and even fewer treated age at natural menopause as the outcome of interest. This section presents results of a comprehensive review of the prospective epidemiological literature on determinants of the age of natural menopause. The Ovid Technologies, Inc. search engine was the tool used to search through the MEDLINE database covering journal articles published from 1966 through the second week of March 2005. My strategy was to request all prospective human studies, written in the English language, with a topic of the "menopause", or with any one of three keywords: menopause, natural menopause or age of natural 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menopause. Of the roughly 600 articles matching these criteria, fifteen studies were identified which prospectively addressed the determinants of the age at natural menopause. Two of these will not be addressed further as they were carried out among small populations of women with Down's syndrome and hence have findings which are not generalizable to any greater population (Cosgrave, Tyrrell et al. 1999; Seltzer, Schupf et al. 2001). Among the thirteen studies that met the criteria for discussion here, those being prospective in design with outcome or event equal to natural menopause, differences in study design were notable. Several studies aimed to address one or two specific associations, while others tested associations with a larger number of familial and/or reproductive factors. Statistical methods reflected the variety in design detail. Finally an important limitation to this literature is that the populations under study are markedly homogeneous. The first of these studies was published in 1983 by Willett and colleagues who responded to the WHO mandate by publishing a paper on data from their seven year-old Nurses Health Study addressing the question of whether cigarette smoking and/or premenopausal weight were independent determinants of the age of menopause among participants (Willett, Stampfer et al. 1983). The Nurses Health Study was 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. started in 1976 when Willett and colleagues sent a health-related questionnaire to all married, female registered nurses, aged 30-55 residing in eleven American states, identified through the State Boards of Nursing and the American Nurses' Association. 78,678 of 121,964 responders reported that their periods had not stopped permanently. Two years later 69,906 of the 78,678 premenopausal women responded to a second questionnaire, of which 2864 reported having had a surgical menopause and were excluded from analysis in the paper. After exclusion of a few more women for radiation-induced menopause and missing menopausal status, 66,663 women remained in the analysis in this paper. In light of limited follow-up time, these authors calculated two-year cumulative incidence rates of natural menopause, rate-ratios and rate differences to quantify the magnitude of the associations between exposures and the outcome, and found that age itself was the strongest predictor of age at menopause. They also found that current smokers had an elevated risk of becoming postmenopausal vs. never-smokers and that within each quintile of relative body weight current smokers were more likely to become postmenopausal. Conversely lean women were at higher risk for menopause than heavier women, a finding which was only partially explained by smoking, and so they concluded that both cigarette smoking 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 2. Prospective studies of determinants of age at natural menopause Authors, year Cohort Description N = Conclusions 1 Willett et al., 1983 The Nurses Health Study, U.S.A. 66,663 Age, smoking and low rel weight were independently associated with higher rates of menopause. However the trend in relative weight was limited to smokers. So at all ages lean smokers experience the highest rate of menopause. 2 Brambilla and McKinlay, 1989 The Massachusetts Women's Health Study, U.S.A. 1945 Smoking, low education and low annual household income were associated with early menopause. 3 Whelan et al., 1990 The Menstruation and Reproductive History Study*, U.S.A. 561 M enstrual cycle length, gravidity and parity (but not age at menarche or ever breast feeding) were associated with age at natural menopause. 4 McKinlay et al., 1992 The Massachusetts Women's Health Study, U.S.A. 2570 The multivariate model with education, parity and smoking provided the best fit to the data. 5 Bromberger et al., 1997 A prospective investigation of the effects of menopause on behavioral and biologic cardiovascular risk factors, U.S.A. 185 African American race, current irregular cycles, current smoking, dieting, nulliparity, regular between 20-35, <= high school education were associated w ith early natural menopause. 6 Torgerson, D. et al, 1997 A prospective study to investigate whether a number of nutritional/fam ilial factors were associated with menopausal age, UK 805 1) There was a strong familial association in menopausal age. 2) Moderate consumption of alcohol was associated with delayed menopause. 7 van Noord, P. et al., 1997 The Doorlopend Onderzoek Morbiditeit/M ortaliteit [DOM] project, The Netherlands 4686 In the multivariate analysis, there was an association with age, fecundity, currnt smoking, SES, age at first OC use (among OC users). All variables in model explain less than 10% of total variation in outcome. 8 Nilsson, P. et al., 1997 The Glostrup Population Study, Denmark 349 In this study there was an association with smoking, and high insulin levels at age 40, which m ight be a marker for norm al female sex hormone physiology. 9 Kato et al., 1998 New York University Women's Health Study, U.S.A. 4694 Factors other than smoking were significant in this study: parity and bmi. 10 Cooper, G. et al., 1999 The Menstruation and Reproductive History Study*, U.S.A. 760 C urrent smoking is important, not former smoking. There was little indication of a smoking dose-response relationship. There was no association with passive smoke exposure. 11 Hardy, R. et al., 2000 The Medical Research Council National Survey of Health and Development (MRC NSHD), Britain 1544 Smoking was significant, not SES nor BMI. 12 Nagata, C. et al., 2000 The Takayama Study, Takayama, Gifu, Japan 1130 Green and yellow veggie intake were associated with later menopause, and carotene intake was borderline significant. 13 Akahoshi, M. et al., 2002 Radiation Effects Research Foundation Follow-up Program, Nagasaki, Japan 1136 High BMI was associated with later menopause, but serum cholesterol and SBP were not. os and low relative weight are independently associated with higher rates of natural menopause. The authors also tested for associations and ruled out confounding by age at menarche, height, history of hypertension and history of diabetes. In their final multivariate model current smoking, number of cigarettes smoked daily, relative weight (in quintiles) and nulliparity were statistically significant independent predictors. Limitations to this study include its homogeneous, overwhelmingly White, middle-class population base, its exclusion of women w ith a surgical menopause and the lack of information regarding postmenopausal HRT utilization. Information on HRT use is im portant in instances such as the following: A woman who started using HRT before she experienced a natural menopause may have experienced HRT induced bleeding, and thus may have reported herself to be premenopausal longer than she would have had she not been using HRT. Even in light of these limitations, this study was an important first step in understanding the drivers of the timing of menopause and the results have held up well over time. Brambilla and McKinlay (Brambilla and McKinlay 1989) carried out a prospective study on the age at menopause which addressed some of the design concerns in the Willett study. This cohort was started between 1980 and 1981 with cross-sectional survey sent to a random sample of 47 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. female, primarily White, Massachusetts residents between the ages of 45 and 55. 8050 women responded (77%) and 2565 women who met the eligibility criteria and were sent a baseline study questionnaire, agreed to participate. Women were considered eligible, or premenopausal, if they had menstruated at least once in the three months prior to filling out the survey and had not had a procedure which would induce iatrogenic menopause such as a hysterectomy. An exclusion of 551 women occurred because they had become postmenopausal since the initial cross-sectional survey. Enrolled subjects were interviewed over the phone six times at nine month intervals over three years to assess menopausal status. The authors considered natural and surgical menopause as competing risks and modeled their conditional probabilities for each year. In univariate models Brambilla and McKinlay found that smoking, less education and low relative household income were associated with early age at natural menopause. In further analysis it became clear that education and income were highly correlated with each other and also with smoking, and thus were not significant covariates in multivariate models when smoking was in the model. Analysis of quantity smoked did not result in a significant trend in these data, a finding that contrasted with those of Willett et al. (Willett, Stampfer et al. 1983). An investigation for confounding by location, parity, marital status, weight, height, BMI, oral contraceptive use 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. or HRT use produced no evidence that any of these covariates was confounding the smoking, education or income associations. These were covariates which had been studied, with inconsistent results, in previous non-prospective studies, and several of which Willett et al. also addressed. In fact the authors explained in the discussion that "an unknown or incompletely controlled confounder" may have created the association between weight and age at menopause in the study by Willett et al. So the question of an association with weight had not been answered, but it seemed as though the effect of smoking on the age of menopause had now been twice-confirmed. The question remained as to whether menopause becomes increasingly earlier with increasing smoking intensity or duration. At this point it appeared as though few lifestyle or reproductive factors could be unquestionably linked to age of menopause. These authors again studied a small homogeneous population. The next prospective study addressing the determinants of menopausal age was an analysis of data that was collected for a slightly different focus. The paper by Whelan et al (Whelan, Sandler et al. 1990) described results of an analysis of determinants of the age at natural menopause in data from The Menstruation and Reproductive History Study begun by Dr. Alan Treloar in 1934 at the University of Minnesota. This study was intended to be a prospective study of menstrual cycle 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. variability, and involved the enrollment of 3962 women in total, primarily during two periods of recruitment; the first in 1934 tol939 and the second in 1960 to 1964. Baseline and yearly follow-up questionnaires were used to extract information about age at menarche, menstrual history, reproductive history, exogenous hormone use, surgery and hospitalizations. In addition participants recorded daily menstrual data. Whelan et al, in intending to analyze determinants of natural menopause, limited their analysis to the 561 women who enrolled before age 25, contributed data through at least age 44, if menopausal experienced a menopause between 44 and 56 years of age and who did not have a lapse in participation equal to or more than 2 years. A discrete-time survival model was employed which is a logistic regression modeling the odds of having a natural menopause for each year of age over the interval 44 to 56. Surgical menopause was treated as a censoring event and as a competing risk. Life table techniques were used to derive mean ages of menopause for various strata. This study had three main findings: 1) women with shorter cycle lengths had an earlier natural menopause, 2) gravidity and higher parity were associated with later age at natural menopause and 3) age at menarche was not associated with the age at natural menopause after adjustment for parity and cycle length. Having a first full-term pregnancy after age 30 seemed to be associated with having an earlier 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menopause, however the trend between increasing age at first full term pregnancy and age at menopause was not apparent. Having a history of spontaneous abortion or history of ever breast feeding was not associated with the event. The authors saw no change in their results upon exclusion of subjects with a surgical menopause or subjects who had used HRT. Unfortunately, neither data on smoking nor body size characteristics were available for analysis. Again subjects were mainly White and socio economically similar in that they were college-educated. After five years of follow-up in the Massachusetts W omen's Health Study, McKinlay, Brambilla and Posner (McKinlay, Brambilla et al. 1992) reported results from an analysis of factors associated with onset of natural menopause among 1178 women who were clearly premenopausal on the baseline survey, and who were aged 45-55 as of January 1,1982. Natural menopause was defined as spontaneous amenorrhea lasting 12 or more months and covariates studies included age, education, parity, BMI, and smoking. HRT users were excluded from this study, but the authors commented that the number of HRT users was small and their exclusion had no effect on results. In a multivariate grouped survival model, only smoking was significantly associated with timing of natural menopause transition. 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In 1997, Bromberger and colleagues published results of their analysis in the American Journal of Epidemiology (Bromberger, Matthews et al. 1997). This was a well-designed, though small, study, and was the first prospective study to report a difference in median age at menopause between two racial/ethnic groups, Caucasians and African-Americans. This study included 185 women, 19 of which were African-American. These 185 women were members of a larger population-based sample of women recruited between 1983 and 1985 for a prospective study of the effects of menopause on risk factors for cardiovascular disease. These 185 women were healthy American women between ages 42 to 50, who were clearly premenopausal. Clearly premenopausal was defined as having menstruated at least once in the previous three months and not using HRT or any other medication which might interfere with determination of menopausal status. Bromberger et al. further limited the subject group to those women less than 47.5 years of age in order to enhance the credibility of his group as truly premenopausal. Subjects were given an in-home baseline interview on gynecologic history, health history, smoking behavior and eating behavior. Gynecologic items included whether they ever used oral contraceptives, their number of pregnancies and live births, the regularity of their menses between ages 20 and 35, and if regular, the number of days between cycles. In a follow-up clinical assessment trained 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. staff measured weight and height, administered 24-hour food recall questionnaires and self-report questionnaires that had questions on physical activity, psychosocial stress, symptoms and personality. Menopausal status, hormone use and hysterectomy status were reported by subjects on a monthly basis via postcards for between seven and nine years. If a subject reported that she had become postmenopausal she was given another physical examination similar to the baseline clinical assessment. In the analysis women who had experienced a surgical menopause were treated as having a different outcome than those with a natural menopause; they were censored at the date of surgery. Otherwise censoring occurred for premenopausal women at drop-out time or study end-date September 1,1992. Kaplan-Meier product-limit survival analysis produced estimates of median age at menopause for the whole group and then for strata of possible determinants in a univariate manner. The logrank test was used to assess differences between median ages in these univariate analyses. The factors which proved in the univariate analysis to have significant differences in median age at menopause across levels were then tested for significance in multivariate Cox proportional hazards models. These included cycle regularity, ethnicity, smoking, dietary restriction, education and nulliparity. Age at menopause was not 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. significantly associated with life stress, physical activity or BMI, although a non-significant trend was apparent between decreasing weight and later menopause. Interactions of ethnicity with certain covariates were also assessed via the Cox model. Results supported the three previous studies in that smoking was associated with an advance in age at natural menopause, although data was not available to assess a trend effect in intensity or duration of smoking. The multivariate analysis showed that having a history of regular cycle length between the ages of 20 and 35, being of African- American racial/ethnic origin (vs. White), being on a restricted diet, having a large weight gain (> 27 pounds) since age 20, and having significant life stress if African-American were all statistically significantly associated with having an earlier menopause. Physical activity was not found to be associated with a woman's age at menopause, however the authors commented that none of these women was more than a moderate exerciser, and thus these women may not have represented a fair distribution of the physical activity spectrum. This study was strong in that it used prospective, monthly assessment of menopausal status; it tested a good number of possible covariates using appropriate statistical methods and had a longer follow- up period than any previous study. However 30% of eligible women 54 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. were excluded from the analysis because their use of HRT interfered with assessment of menopausal status. Further analysis showed that these excluded women were leaner and more likely to use oral contraceptives than those included. But since these factors did not appear to be associated with the outcome it is not clear that this would bias their results. It was also interesting that more African-Americans were excluded from the study at outset for ineligibility due to hypertension, diabetes or hysterectomy leaving only 19 meeting baseline eligibility criteria. This was the first prospective study to draw attention to the marked racial/ethnic differences in the experience of the menopause. The next prospective study (Torgerson, Thomas et al. 1997) was a prospective follow-up to a baseline cross-sectional survey published previously (Torgerson, Avenell et al. 1994). A random sample of women aged 45 to 49, who were identified through a population health registry, were invited to attend a free osteoporosis screening program between January 1991 and April 1992. The two-thousand two hundred and forty women who attended, 82% of those invited, were subsequently sent a questionnaire regarding osteoporosis risk factors. From this questionnaire menopausal status was determined, and the 1227 premenopausal women defined as having menstruated at least once in the prior six months, were sent follow-up postal questionnaires two years later that addressed HRT 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. use and menopausal status. 178 of the 983 responders to this follow-up questionnaire were excluded from the analysis because they were HRT users for whom menopausal status could not be clearly determined, leaving 805 subjects for further analysis. F-tests were used for continuous variables in this analysis to test for differences between means across three menopausal groups; premenopausal, peri-menopausal and postmenopausal. Peri-menopausal women were those who reported having erratic menstruation within the last six months. Proportional odds regression was used for multivariate modeling. 'Results of the univariate analysis showed that age, age of maternal menopause, alcohol consumption and social class were significantly associated with menopausal status. Tobacco consumption was borderline significant (p=0.06), a fact which the authors address in the discussion by explaining that in their data smoking was correlated with HRT use and thus many smokers were excluded from the analysis. In multivariate modeling, maternal menopausal age, alcohol consumption and tobacco consumption exceeded the p<0.05 significance level. Other factors that were tested but not found to be associated in either univariate or multivariate analyses were parity and meat consumption. This study was significant in that it was the first prospective study to implicate a familial role in determination of the age at menopause. 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Limitations included racial/ ethnic and socioeconomic homogeneity, exclusion of HRT users, lacking of information about surgical menopause and limited multivariate investigation of covariates due to their intent to study only those variables which had been significantly associated with menopausal status in their previous analysis. This led to the preclusion of an analysis of weight or BMI as a predictor. The limitations present in this study illustrate several of the recurrent issues with this literature as a whole up to this point; inconsistent and potentially biasing treatment of HRT users and women with a surgical menopause, lack of consistency in covariate analyses, small numbers, short follow-up time and limited generalizability due to narrow subject selection. Another 1997 study on Dutch women has some interesting findings, but was not devoid of such problems as have already been highlighted. This study by van Noord et al. (van Noord, Dubas et al. 1997) was a prospective cohort study intended to investigate predictors of the age at natural menopause. These authors were interested in whether the age at menarche, smoking, weight, height, BMI, fecundity (in years), parity, age at first child birth, socioeconomic status or use of any OC were independently associated with the age at menopause. The design for this study involved taking data from questionnaires given to women at each of five rounds of clinical assessment during a breast cancer screening 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. program between 1975 and 1984. These women were born between 1911 and 1925, making them between the ages of 50 to 75 at baseline. The basic design was therefore not a true prospective study of the age at natural menopause, in which only women considered at risk for the event, or clearly premenopausal, would be enrolled at baseline and then followed prospectively over time. In fact it is not clear how many women were postmenopausal at baseline, and how many experienced menopause between 1975 and 1984. While conclusions as to cause and effect associations would be invalid in this case, the study was interesting in light of previous work. Statistical methods reflected design conditions. Women were classified as OC users, which included women taking HRT, or non-users, and analyses were carried out on each group, as well as on all women combined. Non-users were then dichotomized into early (<49, N=1267) vs. late (>=53, N=1144) natural menopause groups, and for purposes of a univariate analysis, a difference in means for each possible predictor was assessed by t-test. Higher parity, higher socioeconomic status, less smoking, and having fewer siblings were found to be significantly associated with later age at natural menopause. A Pearson correlation coefficient calculated between age at menarche and age at natural menopause was not statistically significant and the multivariate linear 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. regression performed on the non-OC user group indicated that age, fecundity, current smoking and socioeconomic status was each associated with the age at natural menopause. So, while its design reflected many of the same limitations that have been previously discussed including population homogeneity, exclusion of HRT users and women with a surgical menopause, this study supported several previously implicated factors such as smoking and parity. These data failed to confirm an association between age at natural menopause with weight or BMI. Importantly, it also discounted the hypothesis that age at menarche and age at menopause are associated and stated that the amount of variation in age at natural menopause explained by the combined variables the the multivariate analysis was very low (<10%). This implicated as yet undiscovered inherited, maternal and/or early life factors in the determining an individual's age at natural menopause. Another prospective study published in 1997 with a more valid prospective design, but similarly ethnically narrow subject group, came from a group in Denmark (Nilsson, Moller et al. 1997). In this study 630 women were examined as part of a larger community based cohort, and then invited to participate in a follow-up examination in 1987-1988. 493 women were examined for a second time and of those 271 reported that 59 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. they were postmenopausal. 78 were excluded from further analysis because they reported having had a surgical menopause. Of the 232 premenopausal women 62 were excluded because they were HRT users. Fasting bloods and answers to a self-administered questionnaire were taken at age 40. Univariate associations were investigated via comparisons of means for three categories of the age at natural menopause (40-45,46-51 and 51+), calculation of Spearman's linear correlation coefficients for continuous variables or chi-squared test for testing differences between proportions of categorical variables. Several multiple regression analyses were accomplished in which the age at natural menopause was treated as the dependent variable and different groups of independent variables were included. Results of the univariate analysis showed that high serum insulin, high blood uric acid, better expiratory lung function, less smoking, higher social class and fewer subjective feelings of tiredness were associated with later menopause. No association with age at natural menopause was found with lipid levels, glucose levels, alcohol intake or coffee intake. Through a series of multivariate models it was concluded that smoking, insulin and uric acid were independently associated with the outcome, however it was not clear that these analyses were performed in a systematic, objective manner. For instance the authors stated in the 60 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. discussion that high fasting insulin levels might have represented a marker for increased weight; so it is unclear whether they tested for the possible confounding of the insulin effect by weight. The next American study was published in 1998 by Kato and colleagues working at the New York University (Kato, Toniolo et al. 1998). This study included a subset of participants in the New York University Women's Health Study. The original cohort was established between 1985 and 1991 to study the role of hormones and diet in cancer etiology. It was made up of 15,785 women 34 to 65 years old who were not hormone users and had not been recently pregnant or nursing. Subjects were followed by postal questionnaire every 18 to 24 months and of the 7357 women who were premenopausal at cohort baseline, 6141 were successfully followed up. 941 of those had experienced a surgical menopause or cancer diagnosis during follow-up. 237 had experienced a natural menopause before enrollment date and 269 reported using HRT before start of menopause. Each of these three groups of women was excluded from Kato's analysis, leaving 4694. Statistical analysis was performed via the Cox proportional hazards model first univariately, and then multivariately for those factors at least marginally significant on their own. 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Later menopause was associated univariately w ith being parous, Jewish (vs. Catholic), having a higher BMI, and never smoking. No statistically significant associations were found between age at natural menopause and ethnicity, age at menarche, family history of breast cancer or education. Multivariate analyses supported the conclusion that each of the univariately significant factors may have an independent association with the age at natural menopause since estimates of risk ratios were similar to those found in the univariate analyses. This study was limited by its exclusions and its use of a screening population, but it did have decent numbers to address the ethnicity question, finding no significant differences, but reporting that Blacks had slightly later age at natural menopause vs. Caucasians. This is opposite to other studies which had addressed this question (MacMahon and Worcester 1966; Bromberger, Matthews et al. 1997; Cooper, Baird et al. 2001). In 1999 Cooper, Sandler and Bohlig (Cooper, Sandler et al. 1999) published results from a study focused on the relationship between smoking and the age at natural menopause, using the women from the first round of recruitment for the Menstruation and Reproductive History Study at the University of Minnesota. Data from a larger subject pool from this study was analyzed in the Whelan et at. study (Whelan, Sandler et al. 1990; Sandler, Bohlig et al. 1999). In 1990 and 1991, the investigators 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. attempted to re-contact all of the subjects who had been enrolled between 1934 and 1939 and who participated for a minimum of five years. 943 of 1134 were successfully located and 716 filled out a questionnaire themselves, while another 158 questionnaires were filled out by proxy. This questionnaire was designed to attain detailed information about the individual's active and passive smoking history for each decade of life between ages 10 and 79, as well as supplement the menopausal status information for those women who stopped recording menstrual data for the original study before their menopause. 104 women were missing critical data and thus excluded from analysis. 217 women reported having had a surgical menopause and were treated as a separate outcome, while 10 women were excluded from all analyses because they reported having undergone menopause for some other reason. First an analysis of variance on the 543 naturally menopausal women was performed to find the mean age at menopause among smoking strata. Then a Cox proportional hazards regression was carried out on the 543 plus the 217 with surgical menopause, in which the 217 with a surgical menopause contributed data until their date of surgery, at which point they were censored. Women were defined as active smokers if they smoked at the age menopause occurred. 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Active smokers, but not former smokers or women exposed to environmental tobacco smoke, experienced an earlier age at natural menopause than never smokers. In addition, no trend was observed between increasing smoking intensity and earlier menopause among current smokers. Finally, adjustment for BMI at age 30 did not alter the results. While this study does confirm the finding that current smokers had an earlier menopause, their finding no dose-response effect contributes to the inconsistency of that effect. Some limitations of this study include small numbers, limited covariate data and retrospectively collected smoking data. The detailed and careful investigation of smoking was not included throughout the prospective follow-up of these women. The next study was focused on defining the roles of cigarette smoking, BMI and SES in age at menopause. This study addressed both age at onset of perimenopause and the age at natural menopause in a group of British women followed since their birth during the same week in 1946 as participants in the nationally representative cohort of men and women, the Medical Research Council National Survey for Health and Development (MRC NSHD). This analysis by Hardy et al. (Hardy, Kuh et al. 2000) was performed on data from follow-up through 1996 (age 50) of the 1572 women who had responded to at least one of the annual postal questionnaires sent since 1993 (age 47). These questionnaires had been 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. sent to the 1778 original cohort members with whom the researchers were still in contact in 1993. Data from the original cohort follow-up protocol, which was designed to contact each participant nineteen times between birth and age 43, was used to categorize women on levels for variables of interest. Items of interest for their potential direct effect on the age at menopause such as smoking status, height, weight, occupation of self (and of spouse when applicable), occupation of father during childhood, marital status and education were attained at age 36, which the authors stated was before any peri-menopausal symptoms for all but four women. Information on three potential confounders, parity, shortest and longest menstrual cycles over year of age 42, was gained at age 43. Menstrual status was then assessed by age 50 in the follow-up questionnaires sent annually from 1993 to 1996 for the current study. Cox proportional hazards models were utilized to calculate relative risk estimates for each potential risk factor univariately. Adjusted models and models with interactions terms were also investigated. While endpoints of perimenopause and menopause were investigated, only the results for the menopause analysis will be discussed here. Follow-up was treated as censored if the event was not clearly natural menopause, including surgical menopause, start of HRT use or last questionnaire returned. 26 women were excluded from the analysis for iatrogenic menopause, 65 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. current oral contraceptive use or incomplete information. Results showed that smokers had an earlier menopause than non-smokers, BMI had no effect on the age at menopause and of the socioeconomic indicators only marital status had an association, but this effect was clearly confounded by parity. A study published in 2000 by a Japanese group [Takatsuka, 2000 #293] moved ahead into the question of diet and the age at menopause while controlling for age, total energy intake, BMI, smoking and age at menarche. In August 1998,1500 of 6324 baseline-premenopausal, female participants in the Takayama Study of diet and cancer (Shimizu 1996) were randomly selected for recruitment into this study of diet and age at menopause. These 1500 were sent a questionnaire asking detailed questions on menstrual status to supplement data obtained during a baseline 1992 demographic, lifestyle and food-frequency questionnaire. Cox proportional hazards models provided estimates of relative risk in this study. 28 women with surgical menopause were censored at their date of surgery as were 8 women whose periods had stopped due to radiation or medication. After exclusions, 1130 of the 1196 responders were included in the analysis. Univariate analyses showed that BMI, years of smoking and age at menarche were significantly associated with the age at menopause. 66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Further analyses adjusted for these factors, although the authors stated that the hazards ratios obtained were similar to those adjusted only for age. 20 different food and nutrient intake variables were investigated, such as total energy (g), total protein (g), total fat (g), calcium (g), vitamin A (IU#), carotene (mg), green/ yellow vegetables (g), and soy products (g) among others. Data showed that women with higher green/ yellow vegetable intake had significantly delayed incidence of natural menopause. Carotene intake was borderline significant in the same direction. Additional adjustments for parity, education, history of chronic disease such as diabetes, ischemic heart disease and hypertension did not alter the results substantially nor did exclusion of the 132 HRT users. Akahoshi et al. (Akahoshi, Soda et al. 2002) recently published an analysis of the prospective data from the cohort of atomic bomb survivors on Nagasaki, Japan. 4190 female residents of Nagasaki, Japan have been followed since their initial examination in 1958-1959 as part of the Radiation Effects Research Foundation. Biennial physical examinations included measurements of height, weight, sitting blood pressure, serum cholesterol and the extraction of information about menopausal status and type. Only those 493 women who were physically examined at age 40-41, and who later experienced a natural menopause were included in this analysis. These women were then categorized into three groups of BMI, 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. blood pressure and serum cholesterol: upper 25%, m iddle 50% and lower 25%. A comparison of the mean age at natural menopause across groups was made using the ANOVA and longitudinal models provided estimates of relative risk. In summary the results suggested that BMI, but not blood pressure or serum cholesterol, measured premenopausally, is associated with age at natural menopause. One important limitation of this study is its lack of data on potential confounders, including smoking and parity. Another is the great number of women excluded from the analysis due to the statistical methods employed. Critique of existing prospective literature on determinants of age at natural menopause. The prospective studies of the determinants of the age at natural menopause are a heterogeneous group. In general, the prospective design was hinged on the analysis of data from women who were premenopausal at baseline. These women were then followed for various lengths of time, and the data was analyzed for factors associated with age at natural menopause in a variety of ways. The predominant method of analysis was the Cox proportional hazards regression. This method, with natural menopause as the event of interest, was used in eight of the thirteen studies discussed here (Brambilla and McKinlay 1989; 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Whelan, Sandler et al. 1990; Bromberger, Matthews et al. 1997; Torgerson, Thomas et al. 1997; Kato, Toniolo et al. 1998; Cooper, Sandler et al. 1999; Hardy, Kuh et al. 2000; Nagata, Takatsuka et al. 2000). Of those eight, two studies (Torgerson, Thomas et al. 1997; Kato, Toniolo et al. 1998) excluded women with a surgical menopause from the analysis. In the remaining six studies, women with a surgical menopause were censored at the age at surgery. Treatment of HRT users was also not consistent across studies. In fact in the studies by Bromberger et al. (Bromberger, Matthews et al. 1997) and Cooper et al. (Cooper, Sandler et al. 1999), HRT users were either excluded or not specifically addressed in the methods section. In one of the remaining four studies (Brambilla and McKinlay 1989), HRT use was treated as a covariate and was not found to be associated with age at natural menopause. In the other three studies, HRT use was considered as a separate outcome and HRT users were censored at the time they started HRT use. The results of these three studies, considered to be the most comprehensively analyzed of those discussed, serve to summarize the overall findings of the prospective literature. Overall, the results from the prospective studies are similar to those from the cross sectional studies. Smoking and parity are strongly associated with age at natural menopause. Weight or BMI is not consistently associated with age at natural menopause, although studies 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. which do see an association find it in the direction consistent with the cross-sectional literature: low BMI and early natural menopause. Other factors such as age at menarche, oral contraceptive use, alcohol use, education, socioeconomic status, dietary factors and racial/ethnic group are not consistent in the prospective literature. There is suggestion that there is a fam ilial aspect to age at natural menopause through comparison of the ages at natural menopause of m other/daughter pairs (Torgerson 1997), and in this prospective assessment the issue of recall bias is more effectively dismissed. The genetics of the age at natural menopause. Much excellent work has been accomplished to investigate the reproductive and lifestyle factors which may be involved in determining the age at natural menopause. Factors such as smoking and parity have been established as risk factors for early natural menopause. However, estimates of the variance in age at natural menopause that is explained by such factors are low, < 10% (van Noord, Dubas et al. 1997). This fact, in combination with observational studies implicating familial factors in age at natural menopause, has spurred serious consideration of the hypothesis that age at natural menopause is a genetically determined trait. 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Evidence for a familial association of premature ovarian failure was investigated in the early 1980's. Idiopathic premature ovarian failure, meaning spontaneous ovarian failure occurring before age 40, is thought to be reflective of early ovary depletion of follicles. A known genetic disorder has been identified which leads to rapid atresia and POF. In addition maternal, gestational galactosemia has been shown to lead to POF in the offspring [for review see (Santoro 2003)]. In one study (Mattison, Evans et al. 1984) among five families with several cases of idiopathic premature ovarian failure, meaning that which cannot be attributed to any known disorder, pedigree analysis supported transmission of POF through either autosomal dominant or X-linked dominant means. Karyotype analysis in these individuals revealed no informative chromosomal changes. These genetic analytic methods are indicative of those used in the early studies investigating general association between heritability and phenotype. More than ten years later, Cramer et al. (Cramer, Xu et al. 1995) performed a case-control study looking at risk of early natural menopause among women selected from a previously discussed cross-sectional study (Cramer, Xu et al. 1995). 344 "cases", considered in this study to be women with early natural menopause (mean age at natural menopause = 42.2), and an equal number of age-matched "controls" were selected. 71 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Controls were comprised of women who were still premenopausal or had a natural menopause in the normal range (>46). A logistic regression was performed in order to estimate the risk of early natural menopause among women who did and did not have a family history of early natural menopause. After adjustment for smoking, education, parity and BMI, the odds ratio for risk of early natural menopause due to family history of natural menopause (< 46) in a mother, sister, aunt or grandmother was 6.1 (95% Cl, 3.9 to 9.4). Although these results were of large magnitude and were highly significant, one caveat was that recall bias could have played a role in this study if women with early natural menopause are more likely to accurately report a relative's early age at natural menopause. Another study for which this type of bias may have been an issue, reported an association between age at natural menopause and maternal age at natural menopause (Torgerson, Thomas et al. 1997). Women with premature (< 40) or early (40 to 45) natural menopause reported having lower maternal menopausal ages than women with menopause at age 45 or more. Mean maternal menopausal ages among the 551 eligible participants were found to be 43.8,45.4 and 48.4 for premature, early and normal women respectively (p<0.0001). However, of the random population sample of 2399 women between the ages of 45 to 54 living near Aberdeen, Scotland, 1758 women participated, and only 1081 were eligible 72 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. for the current analysis because they were not HRT users and they had experienced a natural menopause. Furthermore, only 51% of the 1081 knew the menopausal age of their mothers. So, while the analytic methods were strong in this study, the possibility that recall bias was acting could not be dismissed. The bellwether study was published in 1998 by Snieder et al (Snieder, MacGregor et al. 1998). In this study, a maximum likelihood model-fitting approach to data from 256 monozygotic (MZ) and 269 dizygotic (DZ) twin pairs was used, which yielded estimates of the variance components for additive genetic variance, dominance genetic variance, and shared (or common) environmental variance. These individual variance components were then divided by the total variance to produce the different standardized components of variance, of which the additive genetic variance/total variance is equal to the heritability quotient, h2. A model including additive genetic and unique environmental variance components produced the most parsimonious fit to the data, yielding a heritability estimate of 63% (95% Cl, 0.53-0.71). In a comment on the Sneider paper (Snieder, MacGregor et al. 1998) in the Lancet, Treloar et al. (Treloar, Do et al. 1998) provided further supporting evidence for the heritability of age at natural menopause. From a preliminary analysis of data from a volunteer twin registry in 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Australia, the estimate of heritability was reported to be between 31 and 53%. The formal analysis of the data from the Australian twin sample was published in 2000 (Do, Broom et al. 2000). This sample was much larger than the Sneider set, with 1355 MZ and 827 DZ twin pairs. Three different statistical approaches were used to assess these data, a path analysis approach, a generalized estimating equation (GEE) approach and a Bayesian analytic approach. Under all three methods the results suggested that additive genetic effects were most important in determination of age at natural menopause and the estimate of heritability was 53% (GEE estimate). The final twin study in the literature came from de Bruin et al. (de Bruin, Bovenhuis et al. 2001). This small study, with 22 MZ and 37 DZ pairs, produced a heritability estimate of 71%. The model with additive genetic and environmental factors provided the best fit to the data. Addition of shared (common) environmental effects did not significantly improve the fit. These twin studies are clear support for the hypothesis that age at natural menopause is influenced by genetic factors, in addition to reproductive and lifestyle factors such as parity and smoking. With this new paradigm firmly in place, the scientific machine turned its attention to identification of the genetic factors influencing age 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. at natural menopause (Weel, Uitterlinden et al. 1999). The first set of studies, in line with development of the field, took a candidate gene, polymorphism association study approach. Weel et al. reported that the PP genotype, which denotes homozygous absence of restriction site, of an anonymous PvuII restriction fragment length polymorphism (RFLP) in intron 1 of the estrogen receptor gene was associated with a 1.1 year earlier onset of natural menopause, vs. the pp genotype (Weel, Uitterlinden et al. 1999). Hefler et al. (Hefler, W orda et al. 2002) chose a candidate gene through evidence from a mouse model implicating deficiency in endothelial nitric oxide synthase (eNOS), encoded by the Nos3 gene, in early menopause onset. In this study, a 27-base pair (bp) tandem repeat polymorphism was measured by PCR amplification using oligonucleotide flanking primers. No association was found between Nos3 genotype and age at natural menopause. A second study from this group of investigators on the same gene looked for association between two different single nucleotide polymorphisms (SNPs) and age at natural menopause. Choice of the two SNPs for investigation in this study reflects the emerging emphasis on "functional" variants. The first SNP was a missense variant which had been investigated in a number of human 75 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 3. Genetic studies of determinants of age at natural menopause. No. Authors, year Methods N = Conclusions 1 Mattison et al., 1984 Pedigree and karyotype analysis. Five familes with IPOF Supported transmission of IPOF through either autosomal dominant of X-linked dominant means. No karyotypic changes found. 2 Cramer et al., 1995 Case/control approach (outcome = early menopause), logistic regression. 344 cases and 344 controls The odds ratio for risk of early natural menopause due to family history of early n.m. (<46) in a mother, sister, aunt or grandmother was 6.1 (95% Cl, 3.9-9.4). Adjusted for smoking, education, parity and BM I. 3 Torgerson et al., 1997 Mutiple linear regression, excluded HRT and s.m. 551 (51% ) There was a significant linear association between self-reported menopausal age and maternal menopausal age, even after adjusting for potential covariates, such as current age and smoking status. 4 Snieder et al., 1998 Maximum likelihood model fitting to twin data. 256 MZ and 269 DZ A model including additive genetic and unique environmental variance components produced the best fit to data, yielding a heritability estimate (ft2) of 63% (95% Cl, 0.53-0.71) 5 Doetal.,2000 Three statistical approaches used, twin data. 1355 MZ and 827 DZ Under all three methods the results suggested that additive genetic effects were most important in determination of age at n.m. The estimate of heritability was 53% 6 de Bruin et al., 2001 Three statistical approaches used, twin data. 22 MZ and 37 DZ The model with additive genetic and environmental factors provided best fit to data. Estimate of heritability of between 71 and 72% . 7 Weel et al., 1999 Candidate gene polymorphism association study. 900 PP genotype in estrogen receptor gene associated with early n.m. 8 Hefler et al., 2002 27 bp tandem repeat in NOS3, ANOVA used. 91 No association was found between Nos3 genotype and age at n.m. 9 Worda et al., 2004 2 SNPs in N O S3, ANOVA used. 87 No association was found with either SNP in this study. 10 Gorai et al., 2004 5 SNPs in 4 genes, ANOVA used. 317 No association was found between any SNP and age at n.m. in this study. 11 Kok et al., 2004 Sib-pair linkage analysis of FSHR region. 6 markers. 126 families In these data, there was no evidence for linkage of early age at natural menopause with markers in the FSHR region. 12 Van Asselt et al, 2004 Genome-wide linkage study. 417 markers. 165 sister pairs Two chromosomal regions were identified as showing suggestive linkage with age at natural menopause: 9q21.3 and Xp21.3. Several other loci showed moderate linkage - j os diseases, and the second SNP had been reported to influence gene transcription rates. Though clearly underpowered, this study did make mention of power, stating that w ith a desired power of 80%, they needed 40 subjects in each group to show a 3-year difference in age at natural menopause. To achieve these numbers they grouped heterozygote and homozygote recessive genotypes. No association was found with either SNP in this study. Increasing sophistication was achieved by investigation of a larger number of SNPs in several related genes. For example, Gorai et al. (Gorai, Tanaka et al. 2003) genotyped five SNPs in four estrogen metabolism genes (ERa, CYP17, CYP1A1, and COMT), among 317 postmenopausal Japanese women. No association was found between age at natural menopause and any genotype in these data. With the results of SNP association studies providing little consistent evidence in the literature, attention was given to the "linkage study" in an effort to more comprehensively study the genome. The first linkage-type study for age at natural menopause was performed by Kok and colleagues (Kok, van Asselt et al. 2004). In this study a candidate genome region was targeted for study, the region surrounding the follicle stimulating hormone receptor gene 77 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (FSHR). The design of this linkage study was a sib-pair based linkage analysis. Eligible families were those in which at least two sisters had a natural menopause in the upper or lower tail of the distribution for menopausal age. Six polymorphic markers were genotyped across the FSHR region and statistical analysis produced measures for shared alleles among families. In these data, there was no evidence for linkage of early age at natural menopause with markers in the FSHR region. A genome-wide linkage study with age at natural menopause was performed in 165 Dutch sister pairs (van Asselt, Kok et al. 2004). Genotype data from 417 markers was analyzed using standard linkage analysis methods, and two chromosomal regions were identified as showing suggestive linkage with age at natural menopause: 9q21.3 and Xp21.3. Several other loci showed moderate linkage and should perhaps also be targeted in future studies. Critique of existing genetic literature on determinants of age at natural menopause. Studies assessing the genetic contribution to age at natural menopause fall into four broad categories: 1) early observational studies (Mattison, Evans et al. 1984; Cramer, Xu et al. 1995; Torgerson, Thomas et al. 1997), 2) 78 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. tw in studies (Snieder, MacGregor et al. 1998; Do, Broom et al. 2000; de Bruin, Bovenhuis et al. 2001), 3) candidate gene association studies (Weel, Uitterlinden et al. 1999; Hefler, W orda et al. 2002; Gorai, Tanaka et al. 2003; Worda, Walch et al. 2004), and 4) linkage studies (Kok, van Asselt et al. 2004; van Asselt, Kok et al. 2004). The critical study was the tw in study by Snieder et al. (Snieder, MacGregor et al. 1998). In this study a model including additive genetic and unique environmental variance components produced the most parsimonious fit to the data, yielding a heritability estimate of 63%. These findings were supported by two more twin studies (Do, Broom et al. 2000; de Bruin, Bovenhuis et al. 2001). The next few studies, chronologically, took a candidate gene-polymorphism association study approach. Only one of these studies reported a significant result (Weel, Uitterlinden et al. 1999). The final group of studies used the linkage design. In one study a candidate region was targeted for study, the region surrounding FSHR, w ith no result (Kok, van Asselt et al. 2004), but the other study was a genome-wide linkage study in 165 sister pairs using data from 417 markers. This study identified two chromosomal regions showing linkage w ith age at natural menopause (van Asselt, Kok et al. 2004). In summary, these genetic studies suggest that determination of the age at natural menopause has a strong genetic component, yet 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. identification of those genetic components is a challenge. The review of the genetic literature on age at natural menopause is also an interesting window on the history of m odern genetic epidemiological study. Overall, this review of the literature illuminated the state of current knowledge about the determinants of age at natural menopause, a recognized breast cancer risk factor and strongly supported the hypothesis that age at natural menopause is determined by a set of reproductive, lifestyle and genetic factors which have only been partially defined. 80 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. PART II. DATA ANALYSES Section 1. Determinants of Age at Natural Menopause: The Multiethnic Cohort K . A. DeLellis*, B.E. Henderson, L. Kolonel and M.C. Pike University of Southern California/Norris Comprehensive Cancer Center Department of Preventive Medicine 1441 Eastlake Ave. MS 44 Los Angeles, California 90033 *To whom correspondence should be sent. This work was supported by National Cancer Institute grants CA63464 and CA54281. These ROl grants are respectively entitled "Genetic Susceptibility to Cancer in Multiethnic Cohorts" and "Multiethnic/Minority Cohort Study of Diet and Cancer." Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ABSTRACT Age at natural menopause has broad implications for a variety of postmenopausal health endpoints, and is of particular epidemiologic interest because of its role as an established risk factor for breast cancer. However for a variety of reasons the determinants of the age at natural menopause have eluded confirmation to a large extent. We investigated factors influencing the age at natural menopause in the five major racial/ethnic groups in the Hawaii and Los Angeles Multiethnic Cohort Study (MEC): Whites, Japanese, African Americans, Native Hawaiians and Latinas. We used a survival analysis approach to investigate the effect of race/ethnicity, and several lifestyle and reproductive variables for possible association with age at natural menopause. In a series of univariate analyses we found that race/ ethnicity, smoking, education, parity, age when periods became regular, BMI, year of birth, oral contraceptive use and age at first full-term pregnancy were significantly associated with age at natural menopause. In a multiple regression, with Whites as baseline, the following factors were statistically significantly associated with age at natural menopause: race/ethnicity, smoking, education, parity, year of birth, current body mass index and age when periods became regular. Natural menopause occurred earliest among 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Latinas. Furthermore, Latinas not born in the United States [non-US-born Latinas (L-NUS)] and those born in the United States [US-born Latinas (L- US)] were significantly different with respect to age at natural menopause, [ H R m u it iv a r ia t e (95% Cl): L-US = 1.228 (1.181-1.278) and L-US = 1.104 (1.064- 1.145)]. A change of 0.01 in the hazard ratio corresponds to a change of 0.030 years, thus the difference of 0.12 between L-NUS and L-US translates to an age difference of 0.122*(0.03/0.01)=0.366 years or 4.392 months. Latinas not born in the United States experience menopause roughly 4.4 months earlier than Latinas born in the United States, and roughly 8.7 months earlier than Whites. Age at natural menopause was latest among Japanese (J) [J = 0.939 (0.913-0.964). African Americans and Native Hawaiians (NH) did not differ significantly from Whites (W) in age at natural menopause in these data. Smoking advanced the age at natural menopause in a dose-dependent manner. Adjustment for all significant covariates did not significantly alter the magnitude of the ethnic-specific hazards ratios. These results support the hypothesis that the age at natural menopause is driven by a combination of reproductive, lifestyle and genetic factors. Keywords: age at natural menopause, race/ethnicity, smoking, genetics 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. INTRODUCTION An unprecedented number of wom en are exceeding menopausal age, and the age at which a woman experiences natural menopause has broad implications for her health in the postmenopausal period [for review see (Barrett-Connor 1993)]. Epidemiological studies have shown that an early natural menopause is associated w ith protection for several hormonally driven diseases, including breast cancer (Trichopoulos, MacMahon et al. 1972). The explanatory hypothesis for this association centers on the shorter length of exposure to reproductive levels of ovarian estrogens, proven breast mitogens (Pike, Krailo et al. 1983). The term menopause technically refers to the cessation of menses, and is derived from the Latin for m onth (mensis) and cessation (pausis), but the biologically relevant event of the natural menopause is age-related ovarian failure. Ovarian failure occurs roughly ten years after the end of fertility, when the complex and synchronous interaction between the central nervous, endocrine and reproductive systems regulating ovarian cycling can no longer sustain ovulation. The menopause is preceded by an average of three years of perimenopause, during which time the approaching menopause is evident in deviations from normal gonadotropin and steroid hormone cycles. Day three estrogen levels become increasingly erratic and decline, whereas day three follicle 84 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. stimulating hormone levels gradually rise to their eventual postmenopausal levels. These neuroendocrine changes are thought to be driven by a decline in follicle num ber below some biologically relevant threshold. Although the biological pathway driving the age at natural menopause has not yet been exactly defined, factors which determine the number of follicles in the original follicle pool, the rate of follicle atresia throughout female life a n d /o r follicle entry into growth phase are likely candidates [for review see (Finch 1990)]. Historically, epidemiological investigation of determinants of age at natural menopause has been centered on factors thought to prevent ovulation, as factors hypothesized to retard the pace of follicle attrition. These have included poor nutrition (Nagata, Takatsuka et al. 2000), obesity (MacMahon and Worcester 1966; Treloar, Do et al. 1998; Nagata, Takatsuka et al. 2000; Akahoshi, Soda et al. 2002), late age at menarche (MacMahon and Worcester 1966; Nagata, Takatsuka et al. 2000), high parity (MacMahon and Worcester 1966; WHO 1981; Kato, Toniolo et al. 1998), oral contraceptive use (Cramer, Xu et al. 1995) and sociodemographic factors (MacMahon and Worcester 1966; Brambilla and McKinlay 1989; Torgerson, Thomas et al. 1997). Factors such as alcohol intake (Torgerson, Thomas et al. 1997) and smoking (MacMahon and Worcester 1966; Willett, Stampfer et al. 1983; Brambilla and McKinlay 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1989; Bromberger, Matthews et al. 1997; Nilsson, Moller et al. 1997; Kato, Toniolo et al. 1998; Hardy, Kuh et al. 2000; Nagata, Takatsuka et al. 2000), which are hypothesized to negatively impact oocyte viability, have been proposed to accelerate reproductive aging. The investigation of a racial/ethnic effect on age at natural menopause has proven to be a controversial issue (Bromberger, Matthews et al. 1997; Kato, Toniolo et al. 1998). High parity has been shown to delay age at natural menopause (Whelan, Sandler et al. 1990; Kato, Toniolo et al. 1998), and smoking has been shown to accelerate the age at natural menopause (Willett, Stampfer et al. 1983; Nilsson, Moller et al. 1997). These are the most firmly established non-genetic determinants of age at natural menopause to date. In addition, results from several recent observational, twin and linkage studies have provided evidence that age at natural menopause is likely to be a genetically determined trait. Women with early menopause (<45 years) report significantly lower maternal menopausal age than women w ith menopause at later ages (45+ years)[12]; results from two twin studies have shown age at natural menopause to be a highly heritable trait (heritability~63%)[13,14]. In the current study we investigated the determinants of the age at natural menopause using data from a large, multiethnic cohort, the Hawaii and Los Angeles Multiethnic Cohort (MEC). Our goal was to 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. develop a predictive model for age at natural menopause and to carefully assess whether evidence existed for difference in age at natural menopause across racial/ethnic groups after adjustment for potentially confounding variables. SUBJECTS AND METHODS Study Subjects and Data Participants included in these analyses were selected from the MEC, a large multiethnic cohort study initiated with the purpose of investigating possible associations between diet and lifestyle factors, and cancer. Cohort recruitment occurred from 1993 to 1996 using driver's license files in Hawaii and Los Angeles, resulting in 215,251 men and women, aged 45 to 75 years at enrollment. Baseline data were collected on all cohort members via a mailed questionnaire, including medical history, family cancer history, diet, medication, physical activity, smoking, and reproductive history, including menopausal status and hormone use. Details about the MEC have been described previously (Kolonel, Henderson et al. 2000). 103,521 female cohort members belonged to one of the five major racial ethnic groups in the MEC [African-American (AA), Japanese (J), Native Hawaiian (NH), Latino (L) and White (W)], and had no prevalent 87 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. breast, ovarian or endometrial cancer. 730 women were excluded because they reported being younger than age 45. 11,630 women were excluded from all analyses because of missing data for variables show n in Table 1. 91,161 women were included in the current analyses. The postmenopausal state was defined in one of three ways: 1) periods had stopped naturally; 2) periods had not stopped and reported taking HRT; 3) periods had stopped due to hysterectomy or bilateral oophorectomy. Women in the first two categories were determined to have experienced a natural menopause. Women in category three were determ ined to have experienced a surgical menopause. Age at menopause was determined in the following ways for each of the three groups: 1) From the data reported on the baseline questionnaire it was possible to determine when a natural menopause occurred for subjects in the first group, as described above, w ithin four categories (<45,45-49,50-54 and 55+). We defined age at natural menopause to be the midpoint of the self-reported age-menopause category. However, if a woman was not yet older than the top age value of her self-reported age at natural menopause category, we coded her as premenopausal and censored her at the top age of the previous category. For example, a 51 year-old woman reporting a natural menopause between ages 45 and 49 was assigned an age at natural m enopause of 47.5. 88 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. However, a 47 year-old also reporting a natural menopause between ages 45 and 49 was coded as premenopausal and assigned a censored age of 50. For women in the second group above, those who started using HRT before their periods stopped naturally, we determined the age of menopause to be the age at which she reported started use of HRT, under the assumption that she initiated HRT to treat menopausal symptoms. For surgically menopausal women, the age at surgical menopause was determined to be age of surgery in a manner parallel to the age at natural menopause coding described above. Premenopausal and surgically menopausal women were censored. Premenopausal women were censored at the highest age value of the five-year category they had successfully lived through. For example, a premenopausal woman of age 53 was censored at 49, the highest value of the previous 45-49 category. Women who experienced a surgical menopause were censored at age of surgery in a manner parallel to age at natural menopause coding. The factors investigated for an association with age at natural menopause were race/ ethnicity, year of birth, smoking history, height, weight, BMI, age at menarche, age when periods became regular, age at first full-term pregnancy, parity, oral contraceptive use, education, hours of sleep/day, hours sitting/ day, hours of vigorous work per week, hours of sweat-inducing activity per week, family history of breast cancer, and 89 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. family history of ovarian cancer. Smoking history was defined by a six- level variable with an aim at separating women whose smoking might have been biologically relevant to their age at menopause (never smoker, smoker who quit before age 40 (smoked <=10 cigarettes per day), smoker who quit before age 40 (smoked >10 cigarettes per day), current smoker at age 40 (smoked <=10 cigarettes per day), current smoker at age 40 (smoked >10 cigarettes per day), and started smoking after age 40. Other variables were categorical variables derived directly from the baseline questionnaire. For example, age at menarche and age when periods became regular were defined as <=11,11-12,13-14, or 15-16 and 17+ with one additional level of "never" for the period-regularity question. Age at first full-term pregnancy had eight levels (Nulliparous, <15,15-17,18-20, 21-25, 26-30,31-35,36+). Parity had four levels (Nulliparous, 1, 2-3,4+ children). Trend analyses shown (see Tables 4 and 5) were conducted with ordinal variables, although results of trend analyses conducted with categorical continuous variables, in which women were given a value of the midpoint of the category, were not significantly different from those presented. A Cox proportional hazards regression was performed to determine which factors were statistically significantly associated with age at natural menopause. Cox regression was performed in SAS version 9.0, 90 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. using Proc Tphreg (SAS Institute, Inc., Cary, North Carolina). A hazard ratio greater than one indicated that age at natural menopause for the exposed group was earlier than in the unexposed group. Using the plot and list options of Proc Cox (Epilog Software, Pasadena, CA), we were able to impute the median ages at natural menopause for specified covariate levels and calculate that a change of 0.01 in the hazard ratio corresponded to a change of 0.03 years or roughly 11 days. Thus differences in the hazard ratio of 0.10 and 0.15 would translate to age differences of 0.3 and 0.45 years, or 3.65 and 5.48 months respectively. Variables showing statistically significant associations with age at natural menopause in the univariate analyses were subsequently tested for independent associations in a multivariate model. The final model represents the set of factors which resulted in a significant likelihood ratio test (p<0.05). Separate analyses stratified by racial/ethnic group, were also performed, and the final model reflects the set of factors providing consistent independent associations across racial/ ethnic strata among the 91,161 women with no missing data for the variables which were significant in the multivariate model. 91 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. RESULTS Table 4 illustrates that among women in our sample, 25.2% (N=22,992) reported themselves to be White (W), 19.5% (N=17,781) were African American (AA), 7.2% (N=6,541) were Native Hawaiian (NH), 27.8% (N=23,357) were Japanese (J), and 20.3% (N=18,490) were Latino (L). Of the Ls, 9,644 were born in the United States (L-US, 10.6% of total) and 8,846 were not born in the United States (L-NUS, 9.7% of total). AA women were older, heavier, had early first full-term pregnancies and were heavy smokers. They also had a higher proportion of women who had experienced a surgical menopause (42.5%) than any other racial/ethnic group. NH women were youngest, were heavy with respect to body mass index, and were heavy smokers. Japanese women were lean, smoked very little, had fewer children, and had later first full-term pregnancies. W women were in the center of the continuum with respect to many of these characteristics, although they were relatively heavy smokers and represented the highest percentage of nulliparous women. There was a striking difference between Latino women born in the United States and not born in the United States. This was most striking in the age of menarche data; L-NUS were late with respect to experiencing menarche and period regularity and L-US were early (data not shown). Thus, we decided to separate the two groups of Latinas for purposes of this 92 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. analysis. These characteristics were in line with what has been previously published (Pike, Kolonel et al. 2002). In univariate analyses we found that race/ ethnicity, smoking, education, parity, age at first full term pregnancy, year of birth, body mass index, age when periods became regular and oral contraceptive use were significantly associated with age at natural menopause. A multivariate regression on these factors showed that age at natural menopause was independently associated with race/ ethnicity, smoking, education, parity, year of birth, body mass index and age when periods became regular (see Table 5). After adjusting for all significant covariates, treating Ws as the reference group, L-NUSs experienced the earliest natural menopause (H R l-n u s = 1.228, 95% Cl = 1.181-1.278). L-USs also had earlier natural menopause vs. Ws (H R l-u s = 1.104, 95% Cl = 1.064-1.145). A As, NHs and Ws did not differ significantly in age at natural menopause. Japanese experienced the latest natural menopause among those ethnicities included in this analysis (HRj = 0.939,95% Cl = 0.913-0.964). Next to race/ethnicity, smoking was the strongest predictor of the age at natural menopause. We found a dose-response effect between smoking and age at natural menopause for both density and duration of smoking, and women who were current smokers at age 40 were likely to have a natural 93 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menopause before those who quit before age 40, regardless of smoking density. Women with more education, experienced the natural menopause later than women with less education (p, trend <0.0001). Of the reproductive factors studied, parity and age when periods became regular were the strongest predictors of age at natural menopause. Higher parity was associated with a delayed natural menopause (multivariate p(trend) <0.0001) and women whose periods became regular at age 17 or later had a hazards ratio of 0.908 (95% Cl = 0.876-0.941), which can be interpreted to mean that they experienced natural menopause roughly 3.2 months later than women with regular periods before age 11. Current BMI was the best measure of body size among those tested for association with age at natural menopause, and in the multivariate model, higher BMI remained independently associated with later natural menopause (p(trend) < 0.0001). Later year of birth was associated with delayed natural menopause (p(trend) < 0.0001). Inclusion of all variables in the model did not significantly alter the magnitude of the ethnic-specific hazards ratios. In Table 6, the results of a racial/ ethnically stratified regression were displayed. While the associations between these factors were generally consistent across ethnic groups, the effect of period regularity 94 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. was not consistent across racial/ethnic groups. This contrast is particularly striking in comparing the two groups of Latinos. The L-US showed the strongest relationship between late period regularity and age at natural menopause, but the L-NUS was not significant at any level. Table 4) D istribution of characteristics by racial/ethnic group and Cox proportional hazards ratios W AA N H J L-US L-NUS HRa N 22992 17781 6541 25357 9644 8846 % of total 25.22 19.51 7.18 27.82 10.58 9.70 HRC 1.000 1.004 1.002 0.930 1.096 1.238 (95% Cl) reference (0.974-1.034) (0.960-1.045) (0.906-0.954) (1.059-1.135) (1.197-1.281) M enopausal Status Premenopausal 23.7% 16.6% 30.7% 21.9% 14,4% 19.7% Natural 44.1% 40.9% 40.2% 52.6% 48.5% 55.0% Surgical 32.2% 42.5% 29.1% 25.5% 37.1% 25.3% Sm oking Never 44.6% 45.3% 44.5% 68.9% 58.1% 72.8% 1.000 Ex-smoker, quit before e 9.2% 12.7% 13.2% 7.1% 11.1% 8.2% 1.025 Ex-smoker, quit before e 18.4% 10.0% 18.1% 7.0% 5.5% 2.2% 1.121 Current sm oker at age 4 6.4% 10.3% 6.8% 5.3% 9.6% 6.4% 1.061 C urrent smoker at age 4 15.1% 9.3% 10.7% 6.6% 4.7% 1.8% 1.239 Started after 40 6.3% 12.4% 6.7% 5.1% 11.0% 8.6% 1.083 Education < High School 6.7% 12.8% 11.3% 9.5% 28.1% 65.7% 1.000 Finished High School 23.3% 27.0% 44.5% 31.8% 38.1% 12.6% 0.920 Finished Some College 33.5% 37.6% 28.5% 29.6% 24.7% 14.6% 0.868 College Graduate 36.5% 22.6% 15.7% 29.1% 9.1% 7.1% 0.844 Parity Nulliparous 16.5% 13.1% 7.8% 13.4% 7.7% 7.1% 1.000 1 Child 12.4% 15.8% 6.9% 11.4% 6.6% 7.4% 0.941 2 or 3 Children 47.9% 36.6% 35.7% 56.6% 37.1% 29.8% 0.877 4 or more Children 23.2% 34.5% 49.6% 18.6% 48.6% 55.7% 0.900 A ge at first fu ll term pregnancy Nulliparous 16.5% 13.1% 7.9% 13.5% 7.6% 7.1% 1.000 <26 61.5% 73.6% 80.4% 49.7% 79.5% 71.8% 0.890 26+ 22.0% 13.3% 11.7% 36.8% 12.9% 21.1% 0.899 Year of Birth <1930 32.1% 43.4% 21.5% 41.3% 34.2% 20.4% 1.000 1930-1935 17.6% 17.4% 18.0% 19.9% 26.4% 24.3% 0.946 1936-1940 16.9% 15.5% 16.9% 12.8% 19.0% 27.5% 0.951 1941-1945 19.4% 14.7% 24.1% 15.2% 12.7% 18.4% 0.884 1946 or later 14.0% 9.0% 19.5% 10.8% 7.7% 9.4% 0.826 BM I < 20 10.1% 6.3% 4.8% 17.3% 3.0% 5.1% 1.000 20-24.9 44.2% 22.5% 28.8% 52.9% 28.8% 29.0% 0.915 25-29.9 27.7% 35.0% 32.7% 23.3% 36.9% 40.8% 0.888 30+ 18.0% 36.2% 33.7% 6.5% 31.3% 25.1% 0.886 A ge w hen periods becam e regular <13 32.5% 34.7% 34.6% 30.1% 40.9% 21.7% 1.000 13-14 41.2% 37.8% 35.2% 37.2% 35.3% 35.6% 1.004 15-16 13.8% 15.4% 14.6% 16.9% 11.8% 24.6% 1.000 17+ 7.6% 7.5% 9.4% 11.0% 6.2% 10.7% 0.921 Never 4.9% 4.6% 6.2% 4.8% 5.8% 7.4% 0.950 O ral Contraceptive Use Never 46.3% 57.0% 53.3% 65.9% 59.8% 61.8% 1.000 <3 years 17.3% 15.0% 18.2% 12.8% 16.6% 19.4% 0.952 3-9 years 23.7% 17.8% 20.0% 15.1% 16.1% 14.1% 0.946 p, trend 12.7% 10.2 % 8.5% 6.2 % 7.5% Race/ethnicity adusted hazard ratio for risk of natural menopause b Race/ethnicity-adjusted p value for trend calculated using the ordinal variable 1 Crude hazard ratio for risk of natural menopause < 0.0001 < 0.0001 < 0.0001 < 0.0001 0.0001 4.7% 0.946 (0.911-0.981) <0.0001 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 5) Multivariate Cox proportional hazards ratios of age at natural menopause p, trendb HRa (95% Cl) Ethnicity W 1.000 reference AA 1.009 (0.978-1.041) NH 1.010 (0.967-1.055) J 0.939 (0.913-0.964) L-US 1.104 (1.064-1.145) L-NUS 1.228 (1.181-1.278) Smoking Never 1.000 reference Ex-smoker, quit before age 40 (<=10/ day) 1.030 (0.993-1.067) Ex-smoker, quit before age 40 (>10/day) 1.123 (1.085-1.162) Current smoker at age 40 (<=10/ day) 1.088 (1.048-1.130) Current smoker at age 40 (>10/day) 1.262 (1.218-1.308) Started after 40 1.061 (1.025-1.097) Education < High School 1.000 reference Finished High School 0.920 (0.893-0.969) Finished Some College 0.868 (0.842-0.895) College Graduate 0.843 (0.815-0.872) Parity Nulliparous 1.000 reference 1 Child 0.930 (0.893-0.969) 2 or 3 Children 0.871 (0.844-0.899) 4 or more Children 0.878 (0.849-0.907) Year of Birth <1930 1.000 reference 1930-1935 0.958 (0.935-0.981) 1936-1940 0.964 (0.937-0.991) 1941-1945 0.887 (0.851-0.925) 1946 or later 0.829 (0.770-0.892) BMI <20 1.000 reference 20-24.9 0.930 (0.901-0.961) 25-29.9 0.898 (0.868-0.929) 30+ 0.893 (0.860-0.927) Age when periods became regular <13 1.000 reference 13-14 0.998 (0.975-1.021) 15-16 0.983 (0.955-1.012) 17+ 0.908 (0.876-0.941) Never 0.937 (0.895-0.982) <0.0001 <0.0001 <0.0001 < 0.0001 <0.0001 Adjusted hazard ratio for risk of natural menopause (Adjusted for all variables present) ’ Multivariate p value for trend calculated using the ordinal variable 96 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 6) Multivariate Cox proportional hazards W ratios of age at natural menopause, by race/ethnicity AA NH J L-US L-NUS Smoking HR* (95% Cl) HR' (95% Cl) HR* (95% Cl) HR* (95% Cl) HR" (95% Cl) HR1 (95% Cl) Never 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference Ex-smoker, quit before age 40 (<=10/day) 1.035 (0.960-1.117) 1.060 (0.981-1.147) 1.200 (1.038-1.387) 0.997 (0.927-1.073) 1.029 (0.934-1.134) 0.985 (0.878-1.104) Ex-smoker, quit before age 40 (>10/ day) 1.160 (1.098-1.226) 1.072 (1.989-1.163) 1.263 (1.118-1.427) 1.091 (1.019-1.168) 1.113 (0.972-1.276) 1.113 (1.915-1.356) Current smoker at age 40 (<=10/day) 1.057 (0.977-1.145) 1.131 (1.043-1.226) 1.270 (1.110-1.454) 1.035 (0.954-1.123) 1.117 (1.009-1.238) 1.031 (0.921-1.154) Current smoker at age 40 (>10/ day) 1.258 (1.187-1.334) 1.231 (1.129-1.342) 1.367 (1.220-1.532) 1.275 (1.180-1.377) 1.311 (1.144-1.503) 1.248 (1.018-1.530) Started after 40 Education 1.055 (0.977-1.140) 1.071 (0.999-1.148) 1.154 (0.999-1.333) 1.059 (0.987-1.137) 1.088 (0.995-1.191) 1.036 (0.942-1.139) < High School 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference Finished High School 0.873 (0.806-0.946) 1.028 (0.954-1.107) 0.880 (0.783-0.989) 0.896 (0.847-0.948) 0.922 (0.859-0.989) 0.894 (0.815-0.980) Finished Some College 0.829 (0.766-0.896) 0.926 (0.861-0.997) 0.827 (0.725-0.944) 0.861 (0.810-0.915) 0.891 (0.820-0.967) 0.807 (0.719-0.907) College Graduate Parity 0.818 (0.757-0.885) 0.896 (0.827-0.972) 0.756 (0.649-0.882) 0.849 (0.796-0.905) 0.867 (0.773-0.972) 0.895 (0.821-0.976) Nulliparous 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1 Child 0.952 (0.882-1.027) 0.946 (0.862-1.038) 0.871 (0.703-1.079) 0.925 (0.861-0.994) 0.936 (0.805-1.088) 0.908 (0.787-1.048) 2 or 3 Children 0.880 (0.832-0.932) 0.926 (0.855-1.003) 0.775 (0.661-0.910) 0.860 (0.814-0.908) 0.820 (0.734-0.917) 0.856 (0.764-0.958) 4 or more Children Year of Birth 0.862 (0.862-0.809) 0.964 (0.890-1.044) 0.801 (0.686-0.935) 0.885 (0.833-0.942) 0.784 (0.703-0.875) 0.856 (0.768-0.954) <1930 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1930-1935 0.967 (0.920-1.016) 1.029 (0.970-1.092) 1.019 (0.920-1.128) 0.937 (0.898-0.977) 0.904 (0.844-0.969) 0.931 (0.865-1.002) 1936-1940 1.016 (0.962-1.074) 1.026 (0.956-1.101) 0.960 (0.857-1.074) 0.949 (0.898-1.002) 0.915 (0.843-0.993) 0.884 (0.820-0.952) 1941-1945 1.008 (0.934-1.089) 0.812 (0.728-0.907) 0.896 (0.775-1.036) 0.828 (0.756-0.906) 0.837 (0.733-0.956) 0.843 (0.756-0.940) 1946 or later BMI <20 1.018 (0.891-1.162) 0.816 (0.680-0.979) 0.930 (0.750-1.153) 0.773 (0.653-0.916) 0.637 (0.488-0.831) 0.698 (0.561-0.869) 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 20-24.9 0.907 (0.848-0.969) 0.930 (0.843-1.026) 0.804 (0.681-0.949) 0.955 (0.913-0.999) 0.948 (0.803-1.118) 0.896 (0.790-1.016) 25-29.9 0.890 (0.830-0.955) 0.888 (0.809-0.975) 0.776 (0.776-0.659) 0.929 (0.881-0.980) 0.925 (0.785-1.089) 0.849 (0.751-0.960) 30+ Age when periods became regular 0.869 (0.804-0.938) 0.844 (0.768-0.927) 0.814 (0.814-0.689) 0.921 (0.846-1.004) 0.946 (0.801-1.117) 0.881 (0.776-1.002) <13 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 1.000 reference 13-14 0.994 (0.949-1.041) 1.003 (0.949-1.060) 0.978 (0.891-1.073) 1.007 (0.964-1.051) 0.955 (0.894-1.020) 1.029 (0.951-1.113) 15-16 0.962 (0.962-0.904) 1.026 (0.958-1.100) 0.978 (0.872-1.096) 0.970 (0.921-1.021) 0.954 (0.870-1.045) 1.034 (0.952-1.124) 17+ 0.860 (0.794-0.931) 0.931 (0.849-1.021) 0.882 (0.763-1.021) 0.900 (0.847-0.957) 0.871 (0.771-0.984) 1.030 (0.928-1.143) Never 0.900 (0.815-0.995) 1.003 (0.885-1.136) 0.881 a Adjusted hazard ratio for risk of natural menopause (Adjusted for all variables present) (0.737-1.052) 0.939 (0.862-1.023) 0.839 (0.731-0.963) 1.048 (0.927-1.186) vo DISCUSSION Of the nearly 216,000 American women who were diagnosed with invasive breast cancer in 2004, approximately 65% occurred in women 55 years or older [1]. Several hereditary, reproductive and lifestyle factors have been established as breast cancer risk factors, including a personal or family history of breast cancer, high mammographic density, post menopausal obesity, use of combined hormone replacement therapy, nulliparity, late first full-term pregnancy, never breastfeeding, physical inactivity and experience of a long reproductive lifespan. Epidemiologic data have shown that women with relatively long exposures to high endogenous levels of estrogen and progesterone, like those which occur during the normal menstrual cycle, are at an increased risk for breast cancer. Thus, an early menarche and late menopause are associated with an increase in breast cancer risk. Identification of the determinants of late natural menopause is critical to understanding the complex etiology of breast cancer. Age at menopause is an increasingly important public health issue because longevity is increasing in developed countries. More women are exceeding menopausal age and facing risk for the conditions of the postmenopausal period, including breast cancer. The age at natural menopause appears to be normally distributed, with a range between 40 98 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to 58 years and a mean of 51.4 years (The North American Menopause Society 2002). This range may be due to real differences in age at natural menopause, driven by variation in an as yet poorly defined set of determinants, but it may also be attributed to a variety of methodological issues that have plagued studies on this phenotype. Methodological issues have included differing/problematic methods of classification of menopausal status, use of samples of convenience such as hospital patients, potentially inaccurate recall in retrospective assessment of age at menopause and covariate information, inconsistent treatment of HRT users, lack of data on demographic and physiological covariates and paucity of data on racially diverse populations [for review see (WHO 1981; Crawford 2000)]. In the current study we have attempted to address many of these issues by using a survival analysis approach to investigate a host of lifestyle and reproductive variables for possible association with age at natural menopause in the large, population-based Hawaii and Los Angeles Multiethnic Cohort Study. The Cox proportional hazards regression approach, in which age at natural menopause was the event of interest and women who had experienced a surgical menopause and premenopausal women were censored, allowed us to avoid excluding a large proportion of subjects, and utilize the maximum amount of information from the cohort. 99 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In this multivariate survival analysis of 91,161 women of five racial/ethnic groups in Hawaii and Los Angeles, late age at natural menopause was clearly, independently associated w ith Japanese race/ethnicity, never smoking, more education, increased parity, more recent year of birth, higher current body mass index and late period regularity. Although this analysis was limited by the categorical age at natural menopause variable, use of the survival analysis approach on this large, multiethnic sample provided ample power to detect small differences in the hazards ratios for age at natural menopause between strata. We were able to show a significant racial/ ethnic difference in age at natural menopause, independent of adjustment for potential confounders. These findings also confirm factors hypothesized to be associated with age at natural menopause which had heretofore remained controversial in the literature. Of all factors studied, smoking has been most consistently associated with the age of menopause (Willett, Stampfer et al. 1983). Smokers were first shown to experience menopause significantly earlier than non-smokers in 1949 (Bernhard 1949), an association which has since been widely confirmed (Bailey, Robinson et al. 1977; Jick, Porter et al. 1977; Lindquist and Bengtsson 1979; Kaufman, Slone et al. 1980; Andersen, Transbol et al. 1982; Gold, Bromberger et al. 2001). The results of the 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. current study, which showed a dose-response relationship between timing and amount of smoking and age at natural menopause, are in line with evidence from a quantitative review reporting summary Mantel-Haenszel odds ratios and 95% confidence intervals for the odds of having a natural menopause for current smokers vs. never smokers, current smokers vs. former smokers and former smokers vs. never smokers of: 1.9 (1.7-2.2), 1.7 (1.5-1.9) and 1.3 (1.0-1.7) (Midgette and Baron 1990). In addition, timing of conception studies have shown delayed conception with heavier smoking (Curtis, Savitz et al. 1997), and oocyte retrieval studies demonstrate that the heaviest smokers have the highest reduction in number of retrieved oocytes (Zenzes 2000). While much research has established that smoking has a significant deleterious effect on reproductive outcomes, the mechanism by which smoking advances the age at natural menopause is not fully understood. Animal studies have shown an increased rate of follicular depletion in mice exposed to benzo(a)pyrene (Shiromizu and Mattison 1984); and the alkaloid derivatives, nicotine and cotinine, impair meiotic spindle function and are responsible for the down-regulation of aromatase activity in the granulosa cells (see (Zenzes 2000) for review)(Barbieri, Gochberg et al. 1986; Barbieri, McShane et al. 1986). The biological mechanism by which smoking is associated with early natural menopause may thus be 101 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. multifaceted. Smoking seems to have a direct role through follicle pool depletion, perhaps via apoptosis triggered by detected DNA damage. But smoking may also have indirect effects, such as through the premature depression of estrogen levels, caused by the down-regulation of enzymes such as aromatase. Indeed, smokers have been shown to possess lower urinary luteal phase circulating estrogen levels compared to non-smokers (MacMahon, Trichopoulos et al. 1982). The association found here between increased education and late age at natural menopause is not novel. In a prospective study, Brambilla and McKinlay (Brambilla and McKinlay 1989) showed a univariate association between low education and early menopause, but this effect was not independent of smoking. Bromberger et al. (Bromberger, Matthews et al. 1997) also reported finding a significant association between less education and early age at menopause. However several other studies have examined this possible association and found no association (Kato, Toniolo et al. 1998; Hardy, Kuh et al. 2000; Nagata, Takatsuka et al. 2000). The heterogeneity in these studies, in terms of country of origin and racial/ethnic distribution makes comparison of these findings difficult, but the relationship between education and such factors as nutrition may indeed account for some biologically relevant effect on reproductive health. Other studies have looked at socioeconomic 102 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. status, another correlate of education, with mixed results. Torgerson et al (Torgerson, Thomas et al. 1997; van Noord, Dubas et al. 1997) found no association between husband's occupation and age at menopause, but van Noord et al. found a significant positive association using type of health insurance as a surrogate for SES (Torgerson, Thomas et al. 1997; van Noord, Dubas et al. 1997). This is a question particularly worthy of further study. We found statistically significant independent associations between late age at natural menopause, higher parity and later period regularity. These parity findings are not in line with the parity effect on breast cancer risk, however they are in line with the traditional oocyte attrition theory of the etiology of the natural menopause. According to this theory, the menopause occurs as the result of the depletion of follicles, and ovulation prevention might delay natural menopause, if by preventing ovulation the pace of follicle attrition is slowed. In this vein, increasing parity has been hypothesized to delay the menopause (MacMahon and Worcester 1966; WHO 1981; Kato, Toniolo et al. 1998), as has use of oral contraceptives and lactation. We have no data on breastfeeding for analysis in this study, however it was interesting that we saw no significant association between oral contraceptive use and age at natural menopause after adjustment for other covariates. In fact, recent evidence shows that depletion of the 103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. follicle pool continues regardless of ovulation status or oral contraceptive use [for review see (te Velde and Pearson 2002)]. Thus, it seems that the association between increasing parity and later age at natural menopause remains unexplained. Similarly, late age of period regularity, has been associated with late age at natural menopause in previous studies, mainly measured as age at menarche. Frisch et al. proposed that women with early menarche should have early natural menopause (Frisch and JW 1974), however this hypothesis has been recently contested (Whelan, Sandler et al. 1990; Snieder, MacGregor et al. 1998). In fact, since it is at puberty that the cyclical loss of a follicle cohort via normal reproductive cycling begins, it may simply be inferred that delay of this cycling will delay the eventual attrition of follicles, assuming the rate of attrition is relatively constant across individuals. Although much is not known about the biological processes involved in follicle attrition, current work promises to shed light on these topics, and thus will provide new insight into menopause biology. Higher BMI was associated with later age at natural menopause. It has been proposed that increased circulating levels of estrogen among women in the higher categories of BMI, may have the consequence of supporting ovary production of a dominant follicle. During the transition 104 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to menopause, estrogen levels become increasingly erratic and developing dominant follicle m ust compete for estrogen. Higher circulating estrogen levels may support successful achievement of dominance and ovulation through a short window of time. The existence of a secular trend in age at natural menopause has been previously reported. Van Noord et al. investigated the potential for a secular trend using linear regression among 4686 women born between 1911 and 1925 (van Noord, Dubas et al. 1997). They reported strong evidence for a slight secular trend between later birth year and later age at natural menopause. Although our subjects were largely born in later decades, we do find evidence for the existence of a secular trend linking later birth year with later age at natural menopause. Adjustment for all significant covariates did not significantly alter the magnitude of the ethnic-specific hazards ratios. Age at natural menopause was earliest among Latinos, particularly among Latinos not born in the United States. The difference between these two groups may be indicative of nutritional and physical activity differences in these two groups, a topic for future investigation in this population. Interestingly, these racial/ethnic relationships were not inconsistent with expectations based on the incidence of breast cancer in the MEC. Pike and colleagues found the lowest rates of breast cancer among L-NUSs, followed by L- 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. USs, AAs, Js, Ws and NHs. The observed adjusted relative rates were 0.60,0.77, 0.78, 0.99,1.00 and 1.33 respectively. These relative rates were adjusted for age and seven known breast cancer risk factors (Pike, Kolonel et al. 2002). The relatively late age at natural menopause among Js reported in the current study does not agree with the relative breast cancer incidence in the cohort, which is statistically equivalent to Ws. This finding is interesting in light of the fact that Js would be expected to have an 11% lower breast cancer risk compared to Whites on the basis of the risk factors as discussed in the paper by Pike et al. (Pike, Kolonel et al. 2002); these discrepancies warrant further investigation. Implications of the strong, independent racial/ ethnic effect reported in the current study are profound, particularly in light of recent evidence supporting the hypothesis that age at natural menopause is a highly heritable trait (Torgerson, Thomas et al. 1997; Snieder, MacGregor et al. 1998; Treloar, Do et al. 1998). Torgerson et al. reported finding an association between age at natural menopause and maternal age at natural menopause (Torgerson, Thomas et al. 1997). Women with premature (<40) and early (<45) natural menopause reported lower maternal menopausal ages than women with menopause at age 45 or more (normal). Mean maternal menopausal ages among 551 eligible participants were found to be 43.81,45.40 and 48.38 for premature, early 106 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and normal women respectively (p<0.0001). Sneider et al. showed, in a classic twin study, that age at natural menopause was 63% heritable (Snieder, MacGregor et al. 1998) and Treloar et al. reported that the age at menopause was 31 to 53% heritable (Treloar, Do et al. 1998). Spurred by this evidence, van Asselt et al. performed a genomewide linkage study with age at natural menopause as a continuous quantitative phenotype in Dutch sister pairs and found two loci, 9q21.3 and Xp21.3, with strong signals of linkage (LOD scores of 2.6 and 3.1 respectively). Twelve other loci had LOD scores >= 1.0 (van Asselt, Kok et al. 2004). In addition, studies have suggested associations with single nucleotide polymorphisms (Weel, Uitterlinden et al. 1999; van Asselt, Kok et al. 2003), which remain to be validated. These findings certainly suggest that determination of the age at natural menopause may have a genetic component. These findings provide support for the hypothesis that age at natural menopause is driven by a combination of reproductive, lifestyle and genetic factors. Based on these findings and on supporting evidence from recent twin, linkage and single SNP studies, we intend to investigate the potential for a genetic role in the determination of age at natural menopause in future studies in the MEC. 107 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. PART II. DATA ANALYSES Section 2. Haplotype-based association analysis for three candidate genes and age at natural menopause: IGF1, CYP17 and HSD17B1 K . A. DeLellis*, I. Cheng, C. Pearce, C. Haiman, B.E. Henderson and M.C. Pike University of Southern California/Norris Comprehensive Cancer Center Department of Preventive Medicine 1441 Eastlake Ave. MS 44 Los Angeles, California 90033 *To whom correspondence should be sent. This work was supported by National Cancer Institute grants CA63464 and CA54281. These ROl grants are respectively entitled "Genetic Susceptibility to Cancer in Multiethnic Cohorts" and "Multiethnic/Minority Cohort Study of Diet and Cancer." 108 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ABSTRACT Evidence from epidemiologic studies, including analysis of data from the Hawaii and Los Angeles Multiethnic Cohort Study, linkage studies, and single nucleotide polymorphism studies in candidate genes, has supported the hypothesis that that age at natural menopause is a heritable trait. In this study we set out to use a haplotype-based approach to investigate whether common variation in three candidate genes was associated with age at natural menopause among participants in the Hawaii and Los Angeles Multiethnic Cohort Study (MEC). A total of 697 MEC participants were genotyped for the haplotype-tagging single nucleotide polymorphisms (htSNPs) in three genes which have been previously characterized for haplotype analysis, and which are reasonable candidates for association with age at natural menopause (IGF1, CYP17, and HSD17B1). Subjects were selected from the two extreme groups of age at natural menopause (<45 and 55+). Eligible subjects reported no prevalent breast, ovarian or endometrial cancer and belonged to one of four major racial/ ethnic groups in the MEC, including White, African American, Japanese, and Latino. In light of our previous data showing that Latinos born in the United States (LUS) and Latinos born outside of the United States (LNUS) were statistically significantly different with 109 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. respect to age at natural menopause, we chose to sample equal numbers of these groups in this analysis. 70 subjects were selected from each of the ten age at natural menopause-racial/ethnic groups. 29,15, and 6 htSNPs were genotyped for IGF1, CYP17, and HSD17B1 respectively. Single SNP analyses were carried out first, via unconditional logistic regression treating late natural menopause as "case" status, for each individual htSNP. Haplotype copy estimates were then calculated based on unphased genotype data using an EM algorithm embedded in D. Stram's tagsnpsv2.exe program, by gene for each respective block. Haplotype estimates were entered as values for the independent variables for logistic regression. Block-specific global tests were performed to test whether addition of all haplotype variables to a model with age and race significantly improved the model. Then haplotypes were entered singly to the model to test for individual haplotype effects among subjects in this analysis. In IGF1, there was no consistent and significant association between common genetic variation and age at natural menopause. The strongest effect was found in block 4, in which haplotype 1 was associated, though not significantly, with increased risk for late natural menopause. In CYP17 the T27C (rs743572), which had been previously reported to be associated with risk for breast cancer, was included here as SNP 5 in block 1. Results for that SNP in these data were directionally 110 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. consistent with previous results, but did not reach statistical significance. In addition, results indicating risk associated with CYP17 block 1-SNP7, with the single haplotype carrying the minor (A) allele of the same htSNP, and findings in HSD17B1 which were suggestive of some relationship between the S312G missense SNP (rs605059), require follow up with larger numbers. From these data we cannot rule out the possibility of an association with these common genetic variants and age at natural menopause. The haplotype-based approach is a powerful approach by which to canvas gene loci for association with various phenotypes. We plan to utilize this approach in future, larger studies of genetic determinants of the age at natural menopause. I ll Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. INTRODUCTION The age at which a woman experiences natural menopause has important implications for her postmenopausal health. In particular, later age at menopause has been associated with increased risk for breast cancer. The mechanism for this effect is thought to be via a prolonged exposure of the breast tissue to premenopausal levels of endogenous estrogens. Thus it has been of interest to understand the determinants of age at natural menopause. However, the determinants of age at natural menopause have eluded confirmation to a large extent. In our analysis of data from almost 90,000 females participating in the MEC, we have confirmed the existence of statistically significant independent associations between late age at natural menopause and Japanese and African American race/ ethnicity, never smoking, more education, increased parity, late period regularity, higher current BMI, more recent year of birth and alcohol intake. Adjustment for all significant covariates did not significantly alter the magnitude of the ethnic-specific hazards ratios, and these data support the hypothesis that the age at natural menopause is driven by a combination of reproductive, lifestyle and genetic factors. Additional support for this hypothesis also comes from other recent studies. Torgerson et al. reported an association between age at natural menopause and maternal age at natural menopause. Women with 112 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. premature (<40) and early menopause (<45), reported significantly lower maternal m enopausal age than women with menopause at a "normal" age (45+) (Torgerson, Thomas et al. 1997). Results from recent twin studies have shown that age at natural menopause is likely to be a highly heritable trait (heritability(h2 )>60%) (Snieder, MacGregor et al. 1998; Treloar, Do et al. 1998; de Bruin, Bovenhuis et al. 2001). A genome-wide linkage study performed to identify quantitative trait loci for age at natural menopause identified two regions (one on chromosome 9 and one on the X chromosome) w ith significant linkage (LOD score 2.6 and 3.1 respectively)(van Asselt, Kok et al. 2004). In addition, an investigation of a SNP in the estrogen receptor gene was reported to be associated with onset of natural menopause (Weel, Uitterlinden et al. 1999). Based on the strong findings in the MEC and on supporting evidence from these genetic studies, we set out to determine whether variation in genes in two candidate pathways is associated with the age at natural menopause using both SNP and haplotype-based analytic approaches. The candidate pathways (GENES) under study are the insulin-like growth factor (IGF) pathway (IGF1) and the estrogen metabolism pathway (CYP17, HSD17B1). These three genes were chosen because of their status as characterized for haplotype analyses and their biological validity as candidate genes for association with age at natural 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. menopause. Both the estrogen metabolism and IGF pathways are involved in the complex and carefully timed interactions of the hypothalamic-ovarian-pituitary axis, the pacemaker of normal reproductive cycling, and variation in key genes in these pathways, such as IGF1, CYP17, and HSD17B1, may lead to variation in the gradual degradation of this axis and thereby alter the timing of natural menopause. Although these pathways are controlled by many genes interacting in a complex manner, this subset of genes encodes key enzymes in these pathways. IGF1 is the primary member of the insulin like growth factor pathway and has been shown not only to have involvement in cross talk with estrogen signaling pathways, but also has been shown to have direct effects on follicle development [for review see (Richards, Russell et al. 2002). Each pathway is a strong candidate for a role in determination of age at natural menopause. CYP17 and FISD17B1 are involved in estradiol biosynthesis, a physiologically active form of estrogen and a critical regulator of the normal menstrual cycle. SUBJECTS AND METHODS The Hawaii and Los Angeles Multiethnic Cohort is an important scientific resource by which to study potentially heritable traits, such as the age at natural menopause, among a large racially-heterogeneous population. 114 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Established between 1993 and 1996 from driver's license files in Hawaii and Los Angeles, the cohort includes 215,251 men and women, ages 45-75 years at the time of enrollment. The MEC primarily includes five major racial/ethnic groups: African-Americans, Japanese-Americans, native Hawaiians, Latino and Non-Latino Whites. Baseline data were collected on cohort members via a mailed questionnaire that contained five sections: (a) background, including medical history and family cancer history; (b) diet history; (c) medication use; (d) physical activity; and (e) female reproductive history, including the use of hormones. Blood was collected on a sub-cohort of about 5,000 randomly selected participants. The draw was completed in the morning, typically at the person's home, after informed consent was obtained. The participation rate for providing a blood sample was 66% for cohort controls. Details of the study have been published previously (Kolonel, Henderson et al. 2000). This study has been approved by the Institutional Review Boards of the University of Southern California and the University of Hawaii, as appropriate. Written informed consent was obtained from each case patient and each control subject before her blood was drawn. Subjects for this study were selected from the two extreme groups of age at natural menopause (<45 years of age and 55+ years of age). Eligible subjects reported no prevalent breast, ovarian or endometrial 115 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. cancer and belonged to one of four major racial/ ethnic groups in the MEC, including White, African American, Japanese, and Latino. In light of previous data showing a significant difference with respect to age at natural menopause between Latinos born in the United States (LUS) and Latinos born outside of the United States (LNUS), we sampled equal numbers of these two groups. Table 7 shows the results of power calculations based on 350 controls and 350 cases, which were performed using the Quanto program [Quanto version 0.4 (Gauderman 2002)]. Assumptions included a log additive model, an alpha of 5%, 80% power and a two-sided hypothesis test. Estimated minimum detectable relative risks for a SNP with a minor allele frequency of 10,15, 20 and 25% were 1.581,1.482,1.429 and 1.396 respectively. Estimated minimum detectable relative risks for haplotypes, calculated via a sample size inflation factor according to methods described by Haiman et al. (Haiman, Stram et al. 2003), with frequencies of 10,15,20 and 25% were 1.661,1.548,1.487 and 1.450. While underpowered to detect within-ethnicity effects, this study was adequately powered to detect differences within a reasonable range of odds ratios with all races combined. 116 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 7: Minimum detectable relative risks with 80% power (1 control per case, 350 controls) Allele / haplotype frequency SNP Analysis (RR) Haplotype Analysis (RR) 10% 1.581 1.661 15% 1.482 1.548 20% 1.429 1.487 25% 1.396 1.450 We selected 70 subjects from each of the ten age at natural menopause-racial/ ethnic groups. Blood was available for 537 subjects meeting these criteria. DNA was extracted from 534 of these samples with adequate biological specimen for use in this study, using the Qiagen DNA extraction kit (Qiagen, Inc., Valencia, CA). 163 subjects were selected from among the controls used for other MEC breast cancer haplotype analyses (refs). For these 163 subjects, the 50 htSNPs for this analysis were scheduled for genotyping as part of other independent breast cancer case- control analyses of these genes. After exclusion of three subjects for whom no genetic material was available for analysis, 697 healthy MEC participants were included in the current analyses. The three candidate genes chosen for study herein (IGF1, CYP17, and HSD17B1) have been previously characterized for haplotype analysis using the haplotype-based methods under development by members of 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the Department of Preventive Medicine at the USC/Keck School of Medicine. This USC/Keck haplotype approach is outlined in brief in Table 8 (Gabriel, Schaffner et al. 2002; Haiman, Stram et al. 2003). HAPLOTYPE-BASED APPROACH The chosen methodology for this candidate gene association study, looking for common genetic variation associated with age at natural menopause, was a haplotype-based approach. We used a two-stage analytic process. The first stage was gene and haplotype characterization. The second stage was haplotype-based association analysis. In order to characterize the gene region, a dense set of SNPs was genotyped across the locus. A survey was made for SNPs from 20 kb upstream through 10 kb downstream of the gene. SNPs were selected from public and private sources (http://w w w .ncbi.nlm .nih.gov/ projects/ SN P/ and h ttp :// www.celera.com). All known SNPs in the coding regions were selected and in cases where prior studies had implicated a relevant SNP with the disease under study, these were also included in the selection process. The goal was to identify one SNP per 2 to 5 kb with minor allele frequency >5%. The criteria for successful genotyping was that no SNP could have Hardy-Weinberg p-value <0.01 in more than one racial/ ethnic group. This genotyping was performed in a multiethnic panel (MEC 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. haplotype panel) of 349 participants in the Hawaii and Los Angeles Multiethnic Cohort who had no history of cancer: African American (N = 70), Hawaiian (N = 69), Japanese (N = 70), Latino (N = 70), and White (N = 70). This genotype data was then used to characterize the gene region for areas of tight linkage disequilibrium, a term used in population genetics to describe the situation in which two alleles preferentially co-segregate, meaning they have been inherited together more than would be predicted by chance. Haplotype blocks were defined as sizable regions over which there was little evidence for historical recombination and within which only a few common haplotypes were observed. According to the methods of Gabriel et al. (Gabriel, Schaffner et al. 2002) we used 90% confidence bounds of Lewontin's D' statistic (Lewontin 1964), a measure of allelic association between SNP pairs, to determine regions of high linkage disequilibrium. D' for each pair of common SNPs, only SNPs with a minor allele frequency >10% were considered, was calculated and SNP pairs with an upper bound above 0.98 and a lower bound above 0.70 were defined as demonstrating strong linkage disequilibrium. A haplotype block was defined as a region over which <5% of SNP pair linkage disequilibrium assessments showed strong evidence of historical recombination. A minimum of six to eight SNPs could define a block 119 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. based on data showing that this num ber is sufficient to identify the common haplotypes in a block (Gabriel, Schaffner et al. 2002). Once blocks were defined, within-block haplotype frequency estimates were generated using a partition-ligation variant (Qin, Niu et al. 2 0 0 2 ) of the Excoffier and Slatkin EM algorithm (Excoffier and Slatkin 1 9 9 5 ) and a subset of the original dense SNP panel was selected as haplotype tag SNPs (htSNPs) using the methods of Stram et al. (Stram, Haiman et al. 2 0 0 3 ) . This htSNP selection method used the haplotype frequency estimates to calculate the probability, for each subject, of having 0 , 1 or 2 copies of each common haplotype. This was summarized as R 2h , the coefficient of determination which represented the correlation between a set of "true" haplotype dose values and the haplotype dose estimates. Through an iterative process, a set of htSNPs was chosen so as to maximize the minimum value of R 2h over the common haplotypes ( > 5 % ) . HtSNPs were then genotyped on all subjects. The haplotypes were re-estimated by block using pooled data from all newly genotyped subjects [for review see (Thomas, Xie et al. 2004)]. The haplotype frequency expectations were then computed for all subjects and used in logistic regressions. 120 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The first step of the analytic process was to run a logistic regression testing the null hypothesis that each individual SNP had no association with outcome. The second step was to perform a global test by block on the null hypothesis that all of the common haplotypes had no association with the outcome. The third analytic step was to test the null hypothesis that having 1 or 2 copies of each common haplotype had no association with outcome, with reference of 0 copies of the common haplotype. Genotyping for haplotype development of these three genes was carried out through collaboration with the Broad Institute for Genome Research at the Massachusetts Institute of Technology. Preliminary work, involving steps one through five of the summarized haplotype approach described in Table 8, has been completed by members of this department. The block structure and position of htSNPs for each gene are shown in Figure 1. In IGF1, 64 SNPs spanning a 156 kb (23 kb upstream and 48 kb downstream) region were characterized. Four blocks of strong linkage disequilibrium were identified, and 29 htSNPs were selected for genotyping in phenotype-specific panels. For CYP17, a 49.5 kb region was parsed into 2 blocks, and 15 htSNPs were chosen to characterize the common haplotype variation (R2 h > 0.80). For HSD17B115 SNPs over a 58 kb region were characterized in the multiethnic panel and it was 121 permission of the copyright owner. Further reproduction prohibited without permission. determined that the entire gene region could be parsed into one block and six htSNPs were selected. Figure 1. Block structure and position of htSNPs for the three candidate genes in this study [A) IGF1, B) CYP17 and C) HSD17B1] A) IGF1 Base P o s itio n B1OCK4_10 B lO C k4_9 B l0 C k 4 _ 8 B lO C k4_7 B!O C k4_6 BTOCk4_5 B!O C k4_4 B 1 0C k4_3 B l o c k 4 _ 2 B l o c k 4 _ l B1O Ck3_10 B lO C k3_9 B l o c k 3 _ 8 B lO C k3_7 B lO C k3_6 B 10Ck3_5 B lOCk3_4 B lO C k3_3 B 10Ck3_2 B l o c k 3 _ l B ! o c k 2 _ 5 B!O C k2_4 B 10Ck2_3 B l o c k 2 _ 2 B l o c k 2 _ l B l o c k l _ 4 B l o c k l . 3 B l o c k l _ 2 B l o c k l . l BL0CK_4 BL0CK.3 BL0CK.2 BLOCK.1 IGF! c o n s e rv a tio n 1 0 1 3 0 0 0 0 0 ] 1013500001 IGF1 htSNPs 1014000001 IGF1 BLOCKS RefSeq Genes -H —I H u/C him p/M ouse/R at/D og/C hick/Fugu/Z fish M u ltiz A lignm ents & C o n serv atio n c h im p _______ dog d i a l___ mouse ahm li II I I r a t I I I k i t I I Jill chicken fugu z e b ra fis h li n M ' J f £ S ! L l M M H D ia a iiilfe M III I B M I I I II ■HI IlH 122 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. B) CYP17 Base P o s itio n Block2_5 Bl0Ck2.4 B!OCk2_3 B10Ck2_2 B lock2.1 B lo ck l_ ie B lo c k l.9 B1ocki_8 BlOCkl_7 BlOCkl_6 B lockl_5 B io c k i.4 B locki_3 B lockl_2 BlO C kl.l BL0CK.2 BLOCK.1 1 9 4 5 7 6 0 0 0 1 I 1645860001 1045900061 CYP17 htSNPs 1046000061 104610000! CYP17_BL0CKS RefSeq Genes I H + H H H u/C him p/M ouse/R at/D og/C hick/Fugu/Z fish M u ltiz A lignm ents & C o n serv atio n C onserv atio n chimp 1^1 dog imMHM iiinimli* afei i kkn MMnHiiiiiiii mouse I I t , | |, | „ j H H Ili fc k k k h l r e t , t , | M| feiiltllliiJll chicken ^ id U I U \ fugu ||j b i l l il l k z e b ra fis h hi II I k h y HIl niBiid ill i M uhiiH H I Ml H I i i II iU klkuiiM I M k I III I i l l 1 1 1 i l l iH liri I IU I I II I I I I 1 H C) HSD17B1 Base P o sitio n BlO C kl.l Blockl_2 B lockl_3 B lockl_4 B 10Ckl_5 B lockl_6 BLOCK.1 N A G L U HSD17B1 C O A SY M L X M L X M L X TBPIP TBPIP C onservation chimp I I 379450061 379506601 37955000] 37960060] 37965606] 37976006| 379756601 37936006] 17HSDB1 htSNPs i I 17HSDB1_BL0CK RefSea Genes ^IH -W H H - H H u/C him p/M ouse/R at/D og/C hick/Fugu/Z fish M u ltiz A lignm ents & C o n serv atio n m-i- dog | M l M id i I k iH Mil M h k h N i 1 j mouse j * 4 ill Ik HI M i r a t * k H * 4 Mil k M m chicken | 1 M i M fugu | i 1 1 i n m IU lu ■ z e b ra fis h | y|| h i l il i m u III II 1 kn i iw i kill II IU I l U l IH b llk l llk ll I U l I I llalil I I III I I I I Genotyping of the htSNPs for each gene was carried out on the natural menopause plates using the Taqman allelic discrimination assay 123 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (Applied Biosystems, Foster City, CA). Single SNP analyses, via unconditional logistic regression treating late natural menopause as "case" status, the direction in line with predicted breast cancer risk, were performed first for each individual htSNP. Haplotype copy estimates were then calculated based on unphased genotype data using an EM algorithm embedded in D. Stram's tagsnpsv2.exe program, by gene and block [available at Dr. Stram's website (h ttp ://w w w .rcf.usc.edu/ -stra m /tagSNPs.html)]. The haplotype estimates were entered as independent variables for logistic regression. Block-specific global tests were performed to test whether addition of all haplotype variables to a model with age and race significantly improved the model. Haplotypes were also entered singly to the model to test for individual haplotype effects. All analyses for this report were performed using the SAS system for Windows (SAS Institute Inc. Cary, North Carolina). Table 8: The Haplotype Approach in Brief (1) Mapping of each gene using data on public databases and Celera. Gene maps cover 20 kb upstream and 10 kb downstream of gene and illustrate the positions of all published SNPs, intron/ exon boundaries, putative regulatory regions, and areas of high hum an/m ouse homology. (2) Pick SNPs at a density of 3 kb for genotyping based on quality of source, and potential functionality. (3) Genotype SNPs in Haplotype Plate. This plate is composed of DNA from 249 healthy MEC participants, all women, 70 of each of 124 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the five major racial/ ethnic groups in the MEC (Only 69 Hawaiians). (4) Define blocks of linkage disequilibrium according to criteria outlined in Gabriel et al. (Gabriel, Schaffner et al. 2002) (5) Develop haplotypes, then pick haplotype tagging SNPs (htSNPs) according to methods outlined in Stram et al 2003 (Stram, Haiman et al. 2003) (6) Genotype htSNPs in phenotype specific plate. (7) Association analysis according to methods outlined in recent papers (Haiman, Stram et al. 2003; Stram 2004). RESULTS 1GF1 Genotyping of the htSNPs in IGF1, according to haplotype characterization of the IGF1 gene which has been described previously (Cheng et al., in preparation), resulted in all SNPs meeting Hardy- Weinberg equilibrium criteria except for SNP2 in block 2. The data for this SNP are under review, however at this time we have included this SNP in all analyses and results are presented for review. There was no strong evidence for difference in distribution of alleles between cases and controls in any htSNP. In looking at the best powered analysis, the comparison for all races combined, we see no consistent statistically significant association between any individual genotype and age at natural menopause. The results of the haplotype analysis are presented in Table 9. Odds ratios and 95% confidence intervals are shown for having one or two copies of each common haplotype compared to having zero 125 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. copies. There were six common haplotypes (>=5%) present in each of IGF1 blocks 1 and 2, and 11 common haplotypes in blocks 3 and 4. The strongest association was found in block 4, in which having one or two copies of haplotype 1 was associated with and increased risk for late age at natural menopause of 1.20 (95% Cl, 0.85,1.691) and 1.46 (95% Cl, 0.90, 2.36) respectively. CYP17 CYP17 htSNP genotyping quality was very high. No SNP had a Hardy-Weinberg p-value <0.01. The p-values for difference in allele frequency between cases and controls provided no evidence for difference in any CYP17 htSNP. In CYP17 the T27C (rs743572) SNP, was in block 1 (SNP 5). Results for that SNP in these data were directionally consistent with previous results, in that having copies of the minor allele was associated with risk for late age at natural menopause, however these results did not reach statistical significance. One SNP in CYP17 (SNP7 in block 1) was statistically significantly associated with age at natural menopause among all races combined and African Americans. Heterozygosity for SNP7 in block 1 was associated with risk for having a late natural menopause (OR(ad j, a il r a ces c o m b in e d )= 2 . 8 6 4 (95% CL 1.109, 7.393). The only haplotype carrying the minor (A) allele Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. for SNP7 in block 1, haplotype 7, was statistically significantly associated with risk for late age at natural menopause among all races combined (Table 10). HSD17B1 HSD17B1 htSNP genotyping quality was also very good. No SNP in this gene's single block produced a Hardy Weinberg p-value less than 0.01. HtSNPs 1,3,4 and 5 showed evidence for a difference in allele frequency between cases (late) and controls (early) among Latinos (data not shown). The single htSNP analysis showed that the minor alleles for htSNPs 1 and 4 conferred protection for late age at natural menopause. Whereas, the minor alleles for htSNPs 3 and 5 were associated with risk for late natural menopause. These directions were not consistent across racial/ethnic groups (see Table 11) and the risk effects did not translate into haplotype-risk effects, as seen in Table 12. No haplotype shows significant association with age at natural menopause. The most interesting effect in this gene was the analysis of the missense SNP S312G (rs605059) which showed some association with age at natural menopause. Having one or two copies of the minor allele of this missense SNP was associated with an increased risk of 1.17 (95% Cl, 0.81,1.70) or 1.331 (95% Cl, 0.86, 2.05) respectively. 127 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 9. IGF1 haplotype effects by block for all races combined IGFl BLOCK 4 A ll Races Adjusted3 Copy # OR (95% Cl) Haplotype 1 3322243121 GGCCCTGACA 0 1.00 1 1.196 (0.846,1.691) 2 1.458 (0.901, 2.360) Haplotype 2 3342243121 GGTCCTGACA 0 1.00 1 0.901 (0.380,2.137) 2 NA Haplotype 3 3342233321 GGTCCGGGCA 0 1.00 1 1.008 (0.698,1.456) 2 1.195 (0.302,4.734) Haplotype 4 3342233323 GGTCCGGGCG 0 1.00 1 0.786 (0.388,1.590) 2 NA Haplotype 5 3122243121 GACCCTGACA 0 1.00 1 1.510 (0.785,2.904) 2 NA Haplotype 6 3122242321 GACCCTCGCA 0 1.00 1 0.854 (0.310, 2.349) 2 NA Haplotype 7 3142232321 GATCCGCGCA 0 1.00 1 0.684 (0.439,1.064) 2 0.720 (0.131, 3.950) Haplotype 8 3142232323 GATCCGCGCG 0 1.00 1 0.765 (0.418,1.401) 2 NA Haplotype 9 3142432323 GATCTGCGCG 0 1.00 1 0.845 (0.475,1.503) 2 0.242 (0.048,1.224) Haplotype 10 4144232321 TATTCGCGCA 0 1.00 1 0.928 (0.589,1.463) 2 1.779 (0.404, 7.833) Haplotype 11 3342233341 GGTCCGGGTA 0 1.00 1 0.997 (0.359, 2.766) 2 NA 3 adjusted for age 128 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 10. CYP17 haplotype effects by block for all races combined CYP17 BLOCK 1 CYP17 BLOCK 2 All Races All Races Adjusted1 Adjusted1 Copy Copy # OR (95% Cl) # OR (95% Cl) Haplotype 1 4424333123 0 1.00 Haplotype 1 41113 0 1.00 1 1.019 (0.733,1.415) 1 1.079 (0.765,1.522) 2 1.084 (0.605,1.924) 2 1.594 (0.671,3.783) Haplotype 2 4422333123 0 1.00 Haplotype 2 21211 0 1.00 1 0.924 (0.399,2.137) 1 0.922 (0.660,1.288) 2 0.865 (0.050,14.915) 2 1.117 (0.454,2.748) Haplotype 3 4422333124 0 1.00 Haplotype 3 21231 0 1.00 1 0.929 (0.655,1.317) 1 0.881 (0.625,1.241) 2 2.062 (0.830,5.127) 2 1.084 (0.465,2.526) Haplotype 4 4444113133 0 1.00 Haplotype 4 21111 0 1.00 1 1.151(0.489,2.711) 1 1.098 (0.610,1.975) 2 NA 2 NA Haplotype 5 2444133333 0 1.00 Haplotype 5 21113 0 1.00 1 0.967 (0.673,1.389) 1 1.032 (0.730,1.460) 2 0.500 (0.187,1.339) 2 0.878 (0.464,1.664) Haplotype 6 2444113133 0 1.00 Haplotype 6 23211 0 1.00 1 0.990 (0.672,1.459) 1 1.006 (0.701,1.444) 2 1.230 (0.383,3.951) 2 1.262 (0.371,4.292) Haplotype 7 2444111133 0 1.00 1 2.744 (1.093,6.889) 2 NA Haplotype 8 4224333123 0 1.00 1 0.524 (0.278,0.989) 2 0.498 (0.078,3.190) Haplotype 9 2444133323 0 1.00 1 1.102 (0.441,2.755) 2 NA Haplotype 10 4444133333 0 1.00 1 1.081 (0.325,3.598) 2 NA Haplotype 11 2444113123 0 1.00 1 1.316 (0.498,3.479) 2 NA 1 Adjusted for age 129 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 11. HSD17B1 genotype effects for all races combined All status=2 status=l early late Crude Adjusted* controls n (% ) case n (% ) OR (95% Cl) OR (95% Cl) p-trend 44 T T 88 (25.5) 101 (29.5) 1.00 1.00 24 C T 178 (51.6) 175 (51.0) 0.857 (0.601,1.220) 0.872 (0.609,1.248) 22 C C 79 (22.9) 67 (19.5) 0.739 (0.479,1.140) 0.745 (0.479,1.158) 0.191 22 C C 202 (58.2) 184 (54.0) 1.00 1.00 12 A C 123 (35.5) 130 (38.1) 1.160 (0.845,1.594) 1.229 (0.881,1.716) 11 AA 22 (6.3) 27 (7.9) 1.347 (0.741, 2.448) 1.444 (0.773, 2.697) 0.138 33 G G 92 (26.5) 79 (22.9) 1.00 1.00 13 A G 177 (51.0) 175 (50.7) 1.151 (0.798,1.660) 1.171 (0.808,1.696) 11 AA 78 (22.5) 91 (26.4) 1.359 (0.887, 2.081) 1.331 (0.863, 2.053) 0.195 22 C C 93 (26.7) 109 (31.6) 1.00 1.00 12 A C 179 (51.3) 172 (49.9) 0.820 (0.580,1.160) 0.831 (0.584,1.183) 11 AA 77 (22.1) 64 (18.6) 0.709 (0.460,1.092) 0.710 (0.455,1.106) 0.124 11 AA 97 (27.8) 84 (24.4) 1.00 1.00 13 A G 181 (51.9) 178 (51.7) 1.136 (0.794,1.624) 1.150 (0.800,1.653) 33 G G 7 1 (20.3) 82 (23.8) 1.334 (0.866, 2.053) 1.316 (0.847, 2.045) 0.220 22 C C 118 (34.8) 132 (39.8) 1.00 1.00 24 C T 162 (47.8) 148 (44.6) 0.817 (0.585,1.140) 0.806 (0.570,1.141) 44 T T 59 (17.4) 52 (15.7) 0.788 (0.504,1.233) 0.746 (0.467,1.194) 0.171 130 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 12. HSD17B1 haplotype effects for all races combined 17HSDB1 BLOCK 1 All Races Adjusted* Copy # OR (95% Cl) Haplotype 1 223114 0 1.00 1 0.822 (0.581,1.163) 2 0.798 (0.497,1.280) Haplotype 2 421232 0 1.00 1 0.960 (0.670,1.373) 2 0.870 (0.438,1.728) Haplotype 3 411232 0 1.00 1 1.273 (0.907,1.787) 2 1.678 (0.888, 3.173) Haplotype 4 223112 0 1.00 1 1.047 (0.629,1.743) 2 NA Haplotype 5 423212 0 1.00 1 1.167 (0.590,2.309) 2 0.127 (0.004,4.251) Haplotype 6 421212 0 1.00 1 1.477 (0.551,3.958) 2 NA a Adjusted for age and ethnicity Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DISCUSSION Age at natural menopause is an important outcome for study because of its role as one of the best established breast cancer risk factors. Understanding of the determinants of age at natural menopause will greatly impact female reproductive medicine, perhaps leading to modes of breast cancer prevention. The factors which determine age at natural menopause are largely unknown. In a previous analysis of data from the MEC, we have confirmed that several reproductive and lifestyle factors are statistically significantly associated with age at natural menopause. In our analysis of data from almost 90,000 females participating in the Hawaii and Los Angeles Multiethnic Cohort Study (MEC), we confirmed the existence of statistically significant independent associations between late age at natural menopause and Japanese and African American race/ethnicity, never smoking, more education, increased parity, late period regularity, higher current BMI, more recent year of birth and alcohol intake (see Part II, Section 1). Adjustment for all significant covariates did not significantly alter the magnitude of the ethnic-specific hazards ratios. These data, in combination with recent work showing that age at natural menopause is a highly heritable trait (Torgerson, Thomas et al. 1997; Snieder, MacGregor et al. 1998; Treloar, Do et al. 1998; Weel, 132 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Uitterlinden et al. 1999; van Asselt, Kok et al. 2003; van Asselt, Kok et al. 2004), provide strong support for the hypothesis that age at natural menopause is likely to be driven by a combination of genetic, reproductive and lifestyle factors; the exact ingredients and mixture of which have yet to discerned. IGF1, CYP17 and HSD17B1 have been identified by our group as strong candidates for association with breast cancer risk and have thus been characterized for haplotype-based analysis in preparation for breast cancer case-control analyses. Details of this work will be published elsewhere (refs). As strong breast cancer candidate genes, these three genes are also excellent candidates for association with age at natural menopause and here we have presented the results of a haplotype-based approach to investigation of whether common variation in these genes was associated with age at natural menopause among participants in the MEC. We found some evidence for an association between several genetic variants with age at natural menopause in these data. In IGF1, there was no consistent and significant association between common genetic variation and age at natural menopause. The strongest effect was found in block 4, in which haplotype 1 was associated, though not significantly, with increased risk for late natural menopause. In CYP17 the T27C 133 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (rs743572), which had been previously reported to be associated with risk for breast cancer, was included here as SNP 5 in block 1. Results for that SNP in these data were directionally consistent w ith previous results, but did not reach statistical significance. In addition, results indicating risk associated with CYP17 block 1-SNP7, with the single haplotype carrying the minor (A) allele of the same htSNP, and findings in HSD17B1 which were suggestive of some relationship between the S312G missense SNP (rs605059), require follow up with larger numbers. From these data we cannot rule out the possibility of an association with these common genetic variants and age at natural menopause. One of the main limitations of this study was the small number of subjects, rendering the analysis underpowered for racially/ ethnically stratified analyses. Even with all races combined, the estimated minimum detectable relative risks for a SNP with a minor allele or haplotype frequency of 10% were 1.581 and 1.661 respectively (see Table 7). Thus, while underpowered to detect within-ethnicity effects, this study was adequately powered to detect differences within a reasonable range of odds ratios with all races combined. The haplotype-based approach is a powerful approach by which to test for associations between common genetic variations in candidate genes for association with traits such as age at natural menopause. We are 134 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. planning further work using this approach, with the aim of identifying the exact combination of genetic, reproductive and lifestyle factors that are important in determining the age at natural menopause. 135 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. PART III. GRANT PROPOSAL Section 1. Introduction to the Grant Proposal The impetus for writing the grant proposal entitled "Reproductive, lifestyle and genetic determinants of the age at natural menopause," which received a superior ranking from the review committee for the Susan G. Komen Breast Cancer Foundation postdoctoral award, was driven by the belief that by investigating the age at natural menopause, light would consequently be shed on the underlying etiology of breast cancer. The grant had two aims, the first aim was to seek out reproduction of previous findings in data from the Hawaii and Los Angeles Multiethnic Cohort Study, by investigating the association between reproductive and lifestyle factors and age at natural menopause in another large cohort of women, the California Teacher's Study. The second aim was to investigate the impact of genes on age at natural menopause by determining whether sequence variation in candidate genes in a novel candidate pathway modulates the timing of natural menopause. In light of space restrictions imposed by the grant guidelines, this introduction has been conceived in order to present a more detailed argument for relevance of candidate genes. In addition this introduction will present a more thorough 136 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. background on the chosen methodology for the second aim, employing illustrations from current preliminary data, with emphasis on that which has been obtained since the grant was submitted in August of 2004. The inhibin/activin pathway was chosen as the candidate gene pathway for this grant proposal. The inhibin/activin pathway is an excellent candidate pathway for possible association with age at natural menopause because of its involvement in regulation of the hypothalamic- pituitary-ovarian (HPO) axis, the pacemaker of normal reproductive cycling. A primary action involved in this HPO axis is the hypothalamic release of gonadotropin releasing hormone (GnRH) in a particular pulsatile manner. This GnRH pulse stimulates the pituitary to release the gonadotropin, FSH, into the circulation. The main target of FSH is the ovarian follicle, which is stimulated by FSH to grow, produce estrogen, and secrete inhibins and activins. Estrogen and the inhibins then signal back to the central nervous system to complete the neuroendocrine regulatory feed back loop, and are particularly important in regulating FSH secretion [for review see (Tong, Wallace et al. 2003)]. The inhibin/activin pathway is comprised of four primary genes, three inhibin genes and one gene for the binding protein follistatin. Figure 2 illustrates how the protein products of the three inhibin genes, INHa, INHpA and INH(3B, dimerize via disulfide linkages to form five 137 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. biologically active proteins. Hetero-dimerization of the a subunit with either of the two ( 3 subunits produces the two inhibin proteins, inhibin A (INHa -INHpA) and inhibin B (INHa -INH(3B). Dimerization of two ( 3 subunits gives rise to the activin proteins, activin A (INH|3A -INHpA), activin B (INHpB -INHpB) and activin AB (INHpA -INHpB). Figure 2. The three inhibin genes and their five dimers. Gene INHA INHBA INHBB Gene product (subunit) Inhibin A FSH - Inhibin B Activin A Activin FSH Activin AB The name "inhibin" was first proposed in a 1932 Science article by D. McCullagh to describe a water soluble testicular factor which could regulate pituitary function (McCullagh 1932). Inhibin activity was subsequently detected in bovine follicular fluid and rat granulosa cells (De Jong and Sharpe 1976; Erickson and Hsueh 1978) and shown to suppress FSH secretion (Franchimont, Verstraelen-Proyard et al. 1979). In 1985 bovine inhibin was isolated and found to be a protein composed of two disulphide-linked polypeptide chains (Robertson, Foulds et al. 1985). It 138 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. was then discovered that not only were there two forms of FSH suppressing inhibins, which shared an identical a-subunit linked to distinct p-subunits (Ling, Ying et al. 1985; Ling, Ying et al. 1986), but there were also FSH stimulating forms present, which were collectively named activin and shown to be dimers of the two p-subunits as described above (Ling, Ying et al. 1986; Vale, Rivier et al. 1986). A structurally distinct FSH suppressing polypeptide was subsequently isolated, which would be named follistatin (Robertson and Robertson 1987; Ueno, Ling et al. 1987). The complexity of this inhibin/activin system was becoming clearer as advances in assays were made, enabling investigators to distinguish between inhibin isoforms (Groome, Illingworth et al. 1996). Investigators proposed that inhibin/activin regulation of FSH was accomplished through variations, over the course of the normal menstrual cycle, in granulosa cell production of the respective inhibin subunits (Meunier, Cajander et al. 1988). In effect, FSH regulation was dependent on the ratio of inhibin: activin production, which in turn was dependent on subunit production. The FSH suppressing role of follistatin was also clarified, in that follistatin was found to be an activin binding protein, and thereby exert its effects on FSH by its ability to neutralize activin activity rather than suppress FSH in a direct manner (Nakamura, Takio et al. 1990; 139 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Kogawa, Nakamura et al. 1991). These four genes are critical elements in maintenance of the normal reproductive cycle. Variation in the structure of these genes may lead to important differences in function of the protein products. If such differences lead to differential regulation of FSH, this could impact the fragile regulation of the normal reproductive axis and thus effect the age at which the menopause occurs. One of the hallmark endocrine indicators of approaching natural menopause is the monotropic FSH rise (Sherman, West et al. 1976; Reyes, Winter et al. 1977). In a prospective study of serum FSH, estradiol, inhibin A and inhibin B, mean FSH levels were shown to begin increasing roughly two years before the menopause, increasing most rapidly in the ten months before the final menstrual period. Mean estradiol levels also began to decrease in an erratic manner about two years before the natural menopause, decreasing most rapidly at the time of final menstrual period. Both inhibin A and B decreased during the two years preceding the natural menopause, with the more pronounced decline of inhibin B beginning earlier than that of inhibin A (Burger, Dudley et al. 1999). It was hypothesized that the immediate cause of the monotropic FSH rise was a decrease in inhibin B caused by a decline in the number of preantral follicles, characteristic of the peri- menopausal ovary (Soules, Battaglia et al. 1998). The exact mechanism by 140 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. which the inhibin/activin pathway regulates FSH secretion is not yet fully elucidated, although much recent work has been centered on identifying and characterizing the receptors and downstream signaling pathways involved. It is clear, however, that the inhibin/activin system is closely tied to regulation of the HPO axis, and the functioning of the normal reproductive cycle. Thus the inhibin/ activin genes and the FST genes make excellent candidates for investigation in a candidate gene association study for determinants of the age at natural menopause. The chosen methodology for study of the inhibin/ activin genes and age at natural menopause was a haplotype-based candidate gene association study. This approach was developed to comprehensively test a candidate gene locus for possible association with a complex trait, in an indirect manner. The approach was meant for application to gene-disease relationships demonstrating a common variant-common disease profile [for review see (Hirschhorn, Pearce et al. 2003)]. In a traditional direct approach, an investigator w ould identify a candidate gene based on prior knowledge of the biological pathway involved. The investigator would then sequence the gene to search comprehensively for variants or perhaps increasingly, the investigator might rely on published SNPs in the public domain. Careful study would then be made of the gene structure and SNP characteristics to identify 141 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. candidate SNPs, those which would be likely to have an impact on gene or protein function, and then perform association studies on the best candidate SNPs. An indirect method would employ the principle of linkage disequilibrium to gain efficiency in the search for an important genetic variant. This is what the haplotype based approach attempts to do. A haplotype, as defined herein, is a set of alleles that have been inherited together as part of an intact chromosomal region, meaning that the set of alleles has remained intact for many generations. Because these alleles are linked, measurement of one of them will perform as a surrogate for the others. Theoretically then it would be possible to genotype a set of SNPs across a gene region, in a relatively small set of subjects, in order to characterize the extent of linkage between each SNP pair, and by doing so gain information about the extent of linkage along the chromosome. This information could then be used to choose a subset of SNPs, within each linked region, which could be genotyped in a larger sample to provide information about all linked SNPs. Haplotype characterization could be accomplished using this data w ith one important exception. Consider the instance, illustrated in Table 13, in which there are two linked SNPs rslOO and rs200 with alleles of A /C and G /T respectively. Subject 1, who is homozygous at both loci, has two 142 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. copies of the haplotype A-G. The haplotypes for Subjects 2 and 3 can also be determined with genotype data for the two loci. However the haplotype determination for subject 4, who is heterozygous at both loci, cannot be exactly determined with only genotype data. In order to determine empirically which alleles lie on the same chromosome, or the phase of these alleles, it would be necessary to have some genotype data from related individuals for these same two loci, such as from both of the subject's parents. This data would then enable the investigator to infer the haplotypes for subject 4. While an effective method for determining the haplotypes of double heterozygous individuals, familial studies are labor intensive and expensive to perform. Table 13. Example of haplotype characterization across two linked SNPs, in four hypothetical subjects rslOO genotype (A/C) rs200 genotype (G/T) haplotype subject 1 A /A G /G A-G A-G subject 2 A /C G /G A-G C-G subject 3 C /C G /T C-G C-T subject 4 A /C G /T A-G C-T or A-T C-G 143 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. An efficient alternative method has been developed to accomplish the haplotype determination in this instance. Excoffier and Slatkin (Excoffier and Slatkin 1995) developed an expectation-maximization (EM) algorithm to produce maximum-likelihood estimates of the haplotype frequencies for these individuals. In reality haplotypes are composed of several SNPs, rather than only two, and thus the complexity of haplotype estimation is much greater. The EM algorithm developed by Excoffier and Slatkin (Excoffier and Slatkin 1995) will produce estimates of haplotype frequencies for all subjects who are heterozygous at more than one locus. This algorithm, which is based on phase inference using the assumption of Hardy-Weinberg equilibrium, chooses the most probable haplotype assignment, given genotype data. We have developed a two-stage analytic process. The first stage is gene and haplotype characterization. The second stage is haplotype association study. The first stage begins with gene characterization and ends with haplotype tag SNP selection. First, in order to characterize the gene region, a dense set of SNPs are genotyped across the locus. We surveyed the SNPs from 20 kb upstream through 10 kb downstream of the gene. SNPs were selected from public and private sources (http://w w w .ncbi.nlm .nih.gov/projects/SN P/ and h ttp :// www.celera.com). All known SNPs in the coding regions were 144 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. selected. In cases where prior studies have implicated a relevant SNP with the disease under study, these are also included in the selection process. The density goal is to identify one SNP per 3 to 5 kb w ith minor allele frequency >5%. This genotyping is performed in a multiethnic panel (MEC haplotype panel) of 349 participants in the Hawaii and Los Angeles Multiethnic Cohort who had no history of cancer: African American (N = 70), Hawaiian (N = 69), Japanese (N = 70), Latino (N = 70), and White (N = 70). This genotype data is then used to characterize the gene region for areas of tight linkage. This is referred to as regions of linkage disequilibrium, a term used in population genetics to describe the situation in which two alleles preferentially co-segregate, meaning they are inherited together more than would be predicted by chance. Much work has been done to discern the best method for utilizing these principles for association studies. Daly et al. (Daly, Rioux et al. 2001) showed that genotype data could be used to formally define linkage disequilibrium blocks, and Gabriel et al. (Gabriel, Schaffner et al. 2002) described a method for characterization of haplotype patterns across a gene region by parsing the region into haplotype blocks. Haplotype blocks were defined as sizable regions over which there was little evidence for historical recombination and within which only a few 145 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. common haplotypes were observed. According the m ethods of Gabriel et al. (Gabriel, Schaffner et al. 2002) we used 90% confidence bounds of Lewontin's D' statistic (Lewontin 1964), a measure of allelic association between SNP pairs, to determine regions of high linkage disequilibrium. D1 for each pair of common SNPs, only SNPs with a minor allele frequency >10% were considered, was calculated and SNP pairs with an upper bound above 0.98 and a lower bound above 0.70 were defined as demonstrating strong linkage disequilibrium. A haplotype block was defined as a region over which <5% of SNP pair linkage disequilibrium assessments showed strong evidence of historical recombination. A minimum of six to eight SNPs could define a block based on data showing that this number is sufficient to identify the common haplotypes in a block (Gabriel, Schaffner et al. 2002). Once blocks were defined, within-block haplotype frequency estimates were generated using a partition-ligation variant (Qin, Niu et al. 2002) of the Excoffier and Slatkin EM algorithm (Excoffier and Slatkin 1995) and a subset of the original dense SNP panel was selected as haplotype tag SNPs (htSNPs) using the methods of Stram et al. (Stram, Haiman et al. 2003). This htSNP selection method uses the haplotype frequency estimates to calculate the probability, for each subject, of having 0,1 or 2 copies of each common haplotype. This is summarized as R2 h, the 146 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. coefficient of determination which represents the correlation between a set of "true" haplotype dose values and the haplotype dose estimates. Through an iterative process, a set of htSNPs is chosen so as to maximize the minimum value of R2 h over the common haplotypes (>5%). HtSNPs are then genotyped on a phenotype specific sample. The haplotypes are re-estimated by block using pooled data from all newly genotyped subjects [for review see (Thomas, Xie et al. 2004)]. The haplotype frequency expectations are then computed for all subjects and used in logistic regressions. The first step of the analytic process was to run a logistic regression testing the null hypothesis that each individual SNP has no association with outcome. The second step was to perform a global test by block on the null hypothesis that all of the common haplotypes have no association with the outcome. The third analytic step was to test the null hypothesis that having 1 or 2 copies of each common haplotype has no association with outcome, with reference of 0 copies of the common haplotype. In summary, the haplotype-based approach, as utilized here, has been developed to investigate whether common genetic variation in candidate genes, is associated with the outcome of interest. In the case of this grant proposal we applied this approach to the assessment of possible 147 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. association between genes in the inhibin/activin pathway and age at natural menopause. Preliminary data to characterize INHa, INH(3A, INH(3B and FST has been accomplished by myself, and thus the state of current w ork on these genes will be the focus of this section. For a brief background, the three inhibin genes, INHa, INHJ3A and INH(3B, have been localized to 2q33, 7pl5-13 and 2cen-ql3 (Barton, Yang-Feng et al. 1987; Barton, Yang- Feng et al. 1989), and Follistatin has been placed at 5qll.2 (Bondestam, Horelli-Kuitunen et al. 1999). Figure 3 depicts the position, size and chromosomal location of each of the four candidate genes selected for this grant proposal. These four genes, INHa, INH(3A, INH(3B and FST, are relatively small genes, with sizes, from transcription start site to transcription stop site, of 3,475,11,265,5,977 and 5,515 bases respectively. Figure 3. Candidate gene positions, sizes and visuals of chormosomal location INHa Chr2:220,262,458-220,265,932 size 3,474 bp. size 11,265 bp. INHPA Chr7:41,502,035-41,513,299 INHpB C h r2:120,819,228-120,825,204 Size 5,977 bp size 5,515 bp FST Chr5:52,812,173-52,817,687 148 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. INHa is the smallest gene of the four, w ith two exons over 3.4 kb. It is located distally on the long arm of chromosome 2. Through the iterative process by which SNPs were selected and genotyped on the MEC haplotype panel (Haiman, Stram et al. 2003), we successfully identified 21 SNPs across 24.7 kb which met working SNP criteria. Working SNPs were those for which an assay was successfully designed. The first few iterations of SNP genotyping were performed on the Sequenom mass spectrometry platform (Sequenom Inc., San Diego, CA). The last iteration of SNP genotyping was performed on the Illumina platform (Illumina, Inc., San Diego, CA). All genotyping of the MEC haplotype panel was performed via collaboration with the Broad Institute at MIT. Working SNPs were not monomorphic in more than one ethnic group, their genotyping results were of high quality, meaning they were not out of Hardy Weinberg equilibrium in more than one ethnic group, <= 1 repeats failed to match, and > 75% of samples were genotyped. Of 53 total SNPs selected in INHa, 6 SNPs failed design, 8 displayed poor genotyping results, and 18 were monomorphic in more than one racial/ethnic group. The remaining 21 SNPs are working SNPs as shown in Figure 4, Part A, resulting in an average SNP density of 1.7 SNPs per 2 kb. There are several known genes upstream of INHa, and the gene region shows a high 149 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. degree of homology across species (Figure 4), especially upstream and in exon regions. INH(3A, the largest of the four candidate genes at 11.3 kb, has three exons. We genotyped 33 SNPs over 62.5 kb. 6 SNPs failed design, 4 failed genotyping, 4 were monomorphic in more than one racial/ ethnic group and 33 are working SNPs. The average density of working SNPs across the region is 0.60 per 2 kb. Figure 4, Part B illustrates that there are no known genes in the larger INHp A gene region, and there is a high degree of homology within the immediate region of the INH(3A gene itself, which falls off with distance. This same pattern describes both INH(3B and FST (Figure 4, Parts C and D). INHpB is almost 6 kb, with two exons. In a 41.8 kb region surrounding this gene, we genotyped 46 SNPs. 4 SNPs failed design, 8 failed genotyping, 7 were monomorphic in more than one racial/ethnic group and 27 are working SNPs. The average density of working SNPs across the INHpB region is 1.29 per 2 kb. FST has three exons across 5.5 kb. Covering 20 kb upstream and 10 kb downstream we genotyped 46 SNPs over 24.6 kb. 8 SNPs failed design, 7 failed genotyping, 18 were monomorphic in more than one racial/ ethnic group and 13 are working SNPs. The average density of working SNPs in the FST region is 1.06 per 2 kb. 150 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 4. Detailed representation of INHa, INHpA, INHpB and FST gene regions, with location of working SNPs in region, known adjacent genes and conservation across species [created via the UCSC genome browser (h ttp ://www.genom e.ucsc.edu/ cgi-bin/hgGateway] Part A) INHa B ase P o s it io n 3 5 7 8 9 16 13 14 15 16 17 18 19 20 21 22 23 24 26 28 29 EX0N_1 CODING.START EX0N_2 CODING.STOP AB014557 AK023854 BC061909 BC607201 INHA C o n s e rv a t io n chim p | d og |__ mouse |£ | r a t [ c h ic k e n f u g u , z e p r a f is h | 2 2 0 2 5 0 0 0 0 ] 2 2 0 2 5 5 0 0 0 ] 2 2 0 2 6 0 0 0 0 | 2 2 0 2 6 5 0 0 0 ] 2 2 0 2 7 0 0 0 0 | 2 2 0 2 7 5 0 INHA w o rk in g SNPs IN H fl GENE CHARACTERISTICS I I I Known G enes <Nov 22, 0 4 ) B ased on SWISS-PROT, TrE M B L, mRNfl, a n d R e fs e q H u /C h im p /M o u s e /R a t/D o g /C h ic k /F u g u /Z fis h M u lt iz A lig n m e n ts & C o n s e r v a tio n K IM Ili II id u r n ii ill I 1 y h i I ■ l a 151 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Part B) INH(3A B ase P o s i t i o n 1 I 2 3 4 5 7 9 1 9 13 14 15 414900ft0| I I 415000001 I 415100001 415200001 INHBA W o rk in g SNPs 4 1 5 3 0 0 0 0 | 415400001 19 2 0 2 1 22 23 25 27 EX0N_3 COOING STOP E X 0N .2 CODING.START EXON_l INHBA C o n s e rv a t io n INHBfl GENE CHARACTERISTICS Known G enes <Nov 22, ©4> B ase d on SW ISS-PROT, TrEM BL, mRNA, t- - - - - - - - - - - - - - - - - - - - - + - a nd R efS eq H u /C h im p /M o u s e /R a t/D o g /C h ic k /F u g u /Z fis h M u lt iz A lig n m e n ts & C o n s e rv a tio n r a t f c h ic k e n ^ fu g u z e b r a f is h M ill I h h a W u IIM M il t I I * * i I II ch imp d og H i | M I ( m o u s e a | iridu M H II* • M l Part C) INHpB B ase P o s it io n 1 I 2 I 3 4 5 7 8 13 14 15 16 17 1 8 19 20 21 22 2 3 25 26 27 28 29 30 32 33 34 EXON_l CODING.START EX0N_2 CODING_STOP INHBB C o n s e r v a tio n ch imp m dog mouse 4 | r a t i i c h ic k e n fu g u z e b r a f is h 1208000001 1208100001 1208200001 INHBB W o rk in g SNPs 120830000) INHBB GENE CHARACTERISTICS Known Genes <Nov 2 2 , 04> B ased on SWISS-PROT, TrE M B L, mRNfl, a nd R efS eq H u /C h im p /M o u s e /R a t/O o g /C h ic k /F u g u /Z fis h M u lt iz A lig n m e n ts & C o n s e rv a tio n ■ M MM UlMMMMtftfi h i M il U 1 ib M i ill M M b ha i la H I M M l* b Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Part D) FST Base P o s it io n I 528000601 528050601 5 2 8 1 0 0 0 0 | 5 2 8 1 5 0 8 0 | 5 26200001 5 2 8 2 5 0 6 6 | FST W o rk in g SNPS 5 6 9 12 14 17 21 22 24 28 EXON.i CODING_START EXON_2 E X0N .3 EX0N_4 EX0N_5 EXON_6 EXON_7 CODING.STOP FST FST C o n s e rv a tio n c h im p | d og 11 m ouse J r a t j c h ic k e n fu g u z e b r a f is h FST GENE CHARACTERISTICS i Known Genes (Nov 22, 645 Based on SWISS-PROT, TrEMBL, m R N fl, and RefSec* « ■ ---l-l+H I----- I - I + - I Hu/Chimp/Mouse/Rat/Dog/Chick/Fugu/Zfish Multiz Alignments & Conservation I li i k U U liW I I li k t i ■ II kfl Ulkml Figure 5 illustrates the current status of gene characterization for INHa (Part A), INH(3A (Part B), INHpB (Part C) and FST (Part D). These plots were created using the confidence bounds plot option in the Haploview program (Barrett, Fry et al. 2005). The confidence bound plot illustrates in shades of grey, the degree of linkage disequilibrium between SNP pairs. A dark grey square denotes strong evidence of LD between the two intersecting SNPs. Light grey indicates that the genotype data is uninformative for assessing the LD between the two SNPs, and a white square indicates that there is strong evidence of recombination in the area between two SNPs. These definitions and the block definition criteria 153 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. used in this plot are derived from those of Gabriel et al. (Gabriel, Schaffner et al. 2002). The ethnic-specific confidence bounds plots (data not shown) showed that Hawaiians, Japanese, Latinos and Whites shared similar patterns of linkage disequilibrium across these gene regions. African Americans displayed shorter regions of linkage disequilibrium and smaller blocks. Figure 5 shows the confidence bounds plots for all non- African American subjects. Figure 5, Part A illustrates that the gene INHa falls between SNPs 20 and 26. The region upstream of INHa has two regions of strong linkage disequilibrium. A major portion of the 3' end of the gene itself is in strong linkage disequilibrium. However the region downstream of the gene has not yet been well characterized. Block 2, in Figure 5, Part A, is a complete block, as dictated by the Gabriel et al. criteria (Gabriel, Schaffner et al. 2002). There are at least 6 SNPs of minor allele frequency 10% or greater. Interblock regions in this gene are below the threshold for acceptable distance of 10 kb. In order to complete the characterization of this gene, all previously untried SNPs in the gene region itself and several SNPs in the 3' downstream region, to a distance of 10 kb downstream, have been ordered to attempt to fill in areas of uncertain linkage disequilibrium. 154 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 5, Part A shows that the gene INH0A falls between working SNPs 18 and 7. Note that the directionality for INH(3A is opposite to the other three genes; the 5' end of this gene is to the right on the Figure. The gene region itself and the 3' region (to the left) fall in one large region of strong linkage disequilibrium, referred to as block 1. There is an interblock region of approximately 7.4 kb, before block 2 begins with working SNP 19. Both blocks in this INH(3A gene region are complete blocks according to the Gabriel et al criteria (Gabriel, Schaffner et al. 2002). In order to complete the characterization of this gene, all SNPs in the gene region itself, and several SNPs between SNPs 15 and 19 have been ordered on the MEC haplotype panel. The current characterization of INHpB is shown in Figure 5, Part C. Five regions of strong linkage disequilibrium are shown designated as blocks 1 through 5. Blocks 1,2,3 and 5 are comprised of only 2,2,5 and 2 SNPs respectively, and thus are not complete blocks. However the gene region itself, which is between SNPs 22 and 25, is in a region of strong linkage disequilibrium comprised of 8 SNPs, and is thus a true block. SNPs have been ordered to complete characterization of the INHpB gene region in order to fill in the extreme 5' region, especially upstream of SNP 8 and between SNPs 15 and 17. In addition SNPs in the 3' region, 155 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. particularly several downstream of SNP 27 have been ordered to better define the block structure in this gene region. Figure 5, Part D shows the linkage disequilibrium profile of the follistatin gene region. According to strict interpretation of Gabriel et al criteria, this region falls into three blocks of strong linkage disequilibrium, blocks 1, 2 and 3. Blocks 1 and 2 have 3 and 2 working SNPs within their boundaries respectively, and are thus not fully characterized. However block 3 has 8 working SNPs within it's borders and can thus be considered complete. For FST completion, we have ordered SNPs that will help us to better define the 5' region, perhaps redefining blocks 1 and 2 as a single block. In addition we have ordered SNPs in the interblock region between SNPs 9 and 12 in an effort to better define the block boundaries and ensure we do not miss an important haplotype in either block. We have also ordered SNPs in the extreme 3 and 5' regions in order to cover the full distance from 20 kb upstream to 10 kb downstream, as is the goal. 156 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 5. Preliminary block structure for INHa, INH0A, INHpB and FST Part A) INHa INHA 18399, 5' 2902, 3 ' 3976 3129 157 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Part B) INHpA 11308, 3 ' 20071, 5 ' INHBA gene borders 158 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Part C) INHpB IN H B B 26666,5’ 8269 20 kb, 5' 2281 9149, 3' ' ’ ; v- v j ► gene borders 159 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Part D) FST 12971, 5 ' Block 2 (4 kl) Block 3 (10 k i) . 6 „ 9 1 12 . I I 17 „ 21 . 2? . 74 26 20 gene borders After these genes are fully characterized for regions of linkage disequilibrium, htSNP selection is possible for each block. As an example, the htSNP selection process has been run for the two complete blocks in INH(3A, one of the most fully characterized of the four genes. The htSNP selection process occurs within each complete block. Using the tagsnps.exe program available at D. Stram's website (h ttp ://w w w - rcf.usc.edu/~stram /) we used the genotype data from the MEC haplotype panel to select a preliminary set of htSNPs for blocks 1 and 2 of INHpA. 160 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The tagSNPs.exe program uses an EM algorithm to calculate the squared correlation coefficient, R2 h, a measure of correlation between the true haplotypes and their estimates (Stram, Haiman et al. 2003). We then select the minimum number of SNPs, which produce an R2 h >= 0.70 for all haplotypes with an estimated frequency of 5% or greater, to become htSNPs. The tagSNPs.exe program was first run on combined data from all 349 subjects on the MEC haplotype panel, denoted as T for "Total". In INHpA block 1, the smallest set of SNPs with an R2 h >=0.70 for T, was four SNPs, including SNPs 1,12,14 and 15. R2 hfor this set of four SNPs was 0.695. This lean set of SNPs was then "forced in" to the tagSNPs.exe selection program for each racial/ ethnic group individually to assess the adequacy of this set of htSNPs in each racial/ethnic group. In Hawaiians and Japanese, this minimum set of four SNPs was adequate, producing an R2 h= 0.738 and 0.913 respectively. In the Latinos and Whites, an additional SNP, SNP3, was needed to produce an R2 h greater than 0.70, and in African Americans four additional SNPs, SNPs 7, 9,10 and 13, were needed to obtain the minimum correlation coefficient. In an effort to minimize the number of htSNPs selected for this block, we then ran the five SNP set obtained from the Latinos and Whites (SNPs 1, 3,12,14 and 15) in the African American sample to determine whether SNP 3 might 161 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. replace one or more of the SNPs chosen in that group. Indeed, SNP 3 replaced SNP 10, and thus a set of eight SNPs was still needed to produce an R2 h >= 0.70. This set of eight SNPs was then forced in to the tagSNPs.exe program for each racial/ethnic group, to obtain the racial/ethnic specific R2 h. R2 h for the set of eight SNPs including SNPs 1, 3, 7,9,12,13,14 and 15, in the Japanese, Latino, Hawaiian, White and African Americans were 0.9991, 0.9907, 0.9869, 0.9243 and 0.7899 respectively. The final step of this process is to determine the proportion of total variation among common haplotypes has been captured by this set of htSNPs, by racial/ ethnic group. In Japanese there were three common haplotypes (>=5%), with a cumulative frequency of 95%. Latinos and Hawaiians there were four common haplotypes (>=5%) with cumulative frequencies of 89 and 86%. Whites demonstrated five common haplotypes with cumulative frequency of 88%. There were eight common haplotypes in the African American sample, which covered 83% of total genetic variation in this group. The htSNP selection for INHpA block 2 was performed in the same manner described for block 1. For block 2, four SNPs (19,21,23 and 25) were adequate to produce an R2 h >= 0.70 in all non-African Americans. In African Americans an additional four SNPs were needed. Thus the complete set of htSNPs for block 2 of INHpA includes SNPs 19, 20, 21, 23, 162 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 24, 25,26 and 27. Each non-African American group had four common haplotypes which described over 94% of the genetic diversity among all haplotypes. African Americans had six common haplotypes in block 2, and these six haplotypes represented 88% of all haplotypes. These methods will be applied to each block, once the four genes are fully characterized. Once htSNPs are selected for each block, they will be genotyped on the phenotype specific plates. The phenotype specific plates, relevant to this grant proposal, are plates constructed specifically to study the age at natural menopause as described in the grant proposal (Part D: Study Design/Aim 2 /Parts 1 and 2). These proposed plates are an expanded version of those created for Part II, Section 2 of this dissertation. The new plates will hold a larger number of samples for better power, and will also include subjects in the normal age at natural menopause category (age at natural menopause between ages 46 and 54) to serve as true controls. These plates will hold 1500 samples, 500 from each of three broad age at natural menopause categories (<44,45-54 and 55+). Subjects will be selected to obtain equal numbers of the same five racial/ethnic groups as were studied in Part II, Section 2 (White, African American, Japanese, Latino-USBorn and Latino-Not USBorn). The analytic approach will be parallel to that which has been described in Part II, Section 2 of this 163 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. dissertation, applied appropriately to the inhibin/ activin pathway. The most important modification will be to incorporate the three outcome groups. This introduction has provided background in support of candidate gene selection, a detailed overview of the proposed methodology for this work and a review of the gene characterization that has been accomplished to date. The inhibin/ activin pathway is an interesting candidate to study with respect to age at natural menopause, and the haplotype-based approach is a robust and flexible methodology by which to study the relationship. Gene characterization is nearly complete and with a solid plan for finishing that process, selection of the final set of htSNPs will be enabled. Genotyping of these htSNPs can be accomplished in a short period of time, via the Taqman allelic discrimination platform (Applied Biosystems, Foster City, CA) here in the USC Core genotyping facility. The analysis of genotype data within the framework of the haplotype approach will be greatly facilitated by the ongoing development of programs by colleagues in this department. This introduction was intended to provide context and background for the grant proposal that follows. 164 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. PART III. GRANT PROPOSAL Section 2. The Grant Proposal Title of Project: Reproductive, Lifestyle and Genetic Determinants of the Age at Natural Menopause The Susan G. Komen Breast Cancer Foundation Request for Application Forms for Postdoctoral Fellowship Research Award Sponsor: Institution: Address: City, State, Zip: Phone: Fax: Email: Postdoctoral Fellow: '05 Postdoctoral Fellow Address: City, State, Zip Code: Postdoctoral Fellow Phone: Postdoctoral Fellow Fax: Postdoctoral Fellow Email: Total Amount Requested: 165 Dr. Leslie Bernstein, Ph.D. Keck School of Medicine, USC 1441 Eastlake Avenue, Room 4449 Los Angeles, CA 90033-0804 323-865-0421 323-865-0128 lbern@usc.edu Katherine DeLellis, Ph.D. pending May Keck School of Medicine, USC Los Angeles, CA 90033 323-865-3995 323-865-0127 delellis@usc.edu 135,000.00 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LAY ABSTRACT: Reproductive, lifestyle and genetic factors have been hypothesized to be important determinants of a woman's age at natural menopause, but few factors have been confirmed. Understanding why women reach natural menopause at different ages is of growing public health importance because the age that menopause is reached affects a woman's risk of disease (e.g. breast cancer, cardiovascular disease, osteoporosis), and because a larger percent of American women is reaching menopausal age. We propose to study reproductive, lifestyle and genetic factors as potential determinants of age at natural menopause using data from two large, pre-existing cohort studies. The role of reproductive and lifestyle factors will be investigated using detailed questionnaire data from the California Teachers Study. Further, we will evaluate the association between six candidate genes in the inhibin/activin pathway using a haplotype-based approach using samples from participants enrolled in a Multiethnic Cohort in Hawaii and Los Angeles. For example, inhibin (INHBB) is considered an important candidate gene, because it codes for a protein involved in the regulation of the normal menstrual cycle. In the course of this study, we propose to develop a predictive model of age at natural menopause using reproductive and lifestyle characteristics and investigate the role of candidate genes in the inhibin pathway with age at natural menopause. 166 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This study will provide insight into the complex determinants of age at natural menopause. SCIENTIFIC ABSTRACT: Background: Age at natural menopause, a recognized breast cancer risk factor, is determined by a set of reproductive, lifestyle and genetic factors which have only been partially defined. Analysis of data from the Hawaii and Los Angeles Multiethnic Cohort (MEC) has shown that later age at natural menopause is associated with race/ ethnicity, and with several reproductive and lifestyle factors: Japanese ethnicity (Whites are baseline), smoking, higher parity, more education, heavier current weight, later age at first full-term pregnancy and later age at menarche. Objective/hypothesis: We propose to confirm and extend preliminary analyses of reproductive and lifestyle determinants of age at natural menopause in the California Teacher's Study (CTS), a cohort composed of participants with high breast cancer incidence rates and detailed questionnaire data on factors of interest to these analyses. Further, we propose to investigate the role of common genetic variants in determination of age at natural menopause using data from the MEC. The inhibin/ activin pathway is a strong candidate because of its role in regulation of the hypothalamic-pituitary-ovarian axis. We hypothesize 167 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. that genetic variants in the inhibin/ activin gene family may contribute to functional differences in the gradual degradation of this axis, and thereby alter the timing of natural menopause. Specific Aims: (1) Investigate the association between reproductive and lifestyle factors and age at natural menopause in the CTS. (2) Determine whether sequence variation within six candidate genes in the inhibin/activin pathway modulates the timing of natural menopause. Design and Methodology: (1) We will conduct a cohort analysis of the CTS to identify factors associated w ith age at natural menopause (2) We will utilize a haplotype-based approach to investigate the association between common genetic variation in candidate genes among 1500 MEC subjects selected from three age at natural menopause categories (<=44 years, 45-54,55+). Potential Outcomes and Benefits: An unprecedented number of women are now exceeding menopausal age and age at natural menopause has broad implications for postmenopausal health. In particular, epidemiological studies have shown that a late natural menopause is associated with increased risk for breast cancer. Identification of the factors that determine timing of natural menopause, including any common genetic variants, will provide insight into the complex etiology of this important risk factor and potentially allow for targeted breast cancer prevention efforts. 168 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A. BACKGROUND OF AGE AT NATURAL MENOPAUSE Of the nearly 216,000 American women who will be diagnosed with invasive breast cancer in 2004, approximately 65% will occur in women 55 years or older [1]. Several hereditary, reproductive and lifestyle factors are established breast cancer risk factors, including a personal or family history of breast cancer, high mammographic density, post-menopausal obesity, use of combined hormone replacement therapy, nulliparity, late first full-term pregnancy, never breastfeeding, physical inactivity and experience of a long reproductive lifespan. Epidemiologic data show that a woman with a relatively long exposure to high endogenous levels of estrogen and progesterone, like those which occur during the normal menstrual cycle, is at increased risk for breast cancer. Thus, an early menarche and late menopause are associated with an increase in breast cancer risk. Identifying the determinants of late natural menopause is critical to understanding the complex etiology of breast cancer. Natural Menopause, Biology and Epidemiology. The exact determinants of age at natural menopause are not known. In biological terms, normal human reproductive function depends on the synchronous interaction between physiologic pathways including the central nervous, endocrine and reproductive systems, specifically the hypothalamic-pituitary-ovarian (HPO) axis. With aging, carefully timed interactions decay, leading to 169 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ovulatory failure and cessation of menses, or menopause. It has been hypothesized that the depletion of ovarian follicles below a threshold value leads to natural menopause (for review[2]). Thus factors which affect follicle attrition would strongly influence the function of the hypothalamic-pituitary-ovarian (HPO) axis and determine age at natural menopause[3,4]. These factors include obesity[5-7], age at menarche[5, 6], parity[5, 8, 9], and oral contraceptive use[10]. Smoking[5, 11], which negatively impacts oocyte viability, may accelerate reproductive senescence. In addition, results from several recent twin, linkage and single nucleotide polymorphism (SNP) association studies have provided evidence that age at natural menopause is likely to be a genetically determined trait. Women with early(<45 years) m enopause report significantly lower maternal menopausal age than w om en with menopause at later ages (45+ years)[12]; results from two tw in studies have shown age at natural menopause to be a highly heritable trait (heritability~63%) [13, 14]. A strong candidate pathw ay for investigation of the heritable component of age at natural menopause: The inhibin/ activin pathway. The inhibin/ activin pathway is involved with the complex and carefully timed interactions of the HPO axis. Genetic variants in the inhibin/activin pathway may contribute to functional differences in the gradual degradation of this axis thereby altering the 170 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. timing of natural menopause. The inhibin protein is a heterodimer of an alpha subunit and one of two beta subunits, either betaA (inhibin A) or betaB (inhibin B). Dimerization of two beta subunits gives rise to activins (activin A (betaA-betaA), activin B (betaB-betaB) or activin AB (betaA- betaB)). Inhibins are secreted by the ovarian granulosa cells in response to FSH, and act in a negative endocrine feedback loop to suppress FSH secretion by the pituitary. Activins have the opposite effect, stimulating FSH secretion, primarily through local paracrine action at the pituitary. Activin signals through an activin receptor system, comprised of a type I and a type II receptor that cooperate to bind the ligand and stimulate downstream signaling. The final component of the system is a binding protein, follistatin, which has been shown to bind and deactivate activins. During the three years preceding menopause, circulating inhibins decline markedly. This is thought to contribute to the hallmark endocrine change of approaching menopause, the rise in follicle stimulating hormone. Differences in the efficiency of the inhibin system, which may be driven by inherited genetic variation, may modulate HPO axis function and determine the timing of natural menopause. The haplotype-based approach. The genome sequences of all humans are 99.9% identical. Single nucleotide changes account for the majority of the 0.1% variation, and are thought to be responsible for most phenotypic differences. The 171 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. hum an genome contains regions of linkage disequilibrium (LD) which have been inherited in haplotype blocks. The haplotypes in each block can be uniquely identified by genotyping a subset of SNPs. SNPs in this subset are called haplotype-tagging SNPs (htSNPs). Approximately 5-6 common haplotypes per block capture the common diversity (>=80%) in a block. HtSNPs may then be tested in association studies to investigate possible relationship between haplotypes and disease risk. The haplotype-based approach is an efficient tool for identifying genomic regions that may contain variants associated with common diseases [15]. Preliminary Data An analysis of data from the Hawaii and Los Angeles Multiethnic Cohort (MEC) shows that age at natural menopause is associated with race/ethnicity and several reproductive and lifestyle factors. The MEC was established through collaboration between the Cancer Research Center at the University of Hawaii and the University of Southern California (USC). The study's primary aim is to evaluate the dietary and other environmental contributions to the racial/ ethnic variability in cancer risk. The MEC has a large female component (N=118/441) of Japanese (J), Whites (W), African American (AA) and Latino (L). Participants between the ages of 45 and 75 years were recruited between 1993 and 1996 in Hawaii and Los Angeles. Baseline data, collected via a mailed questionnaire, included medical history, 172 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. family cancer history, diet history, medication use, physical activity, and reproductive history, including use of hormones and age at menarche. Collection of blood and urine samples from MEC participants began in July 1995 and is ongoing with the goal of collecting specimens from all willing members. Approximately 20,000 blood specimens have been collected from female cohort participants to date, and it is projected that 40,000 will be collected by 2005. Details of the study have been published previously [16]. A Cox proportional hazards regression approach was chosen to maximize the amount of information obtained from MEC data. Data from 84,567 women who completed the baseline questionnaire were analyzed. In this approach we assumed that all women entered the cohort at birth. We followed them to the event, natural menopuase, or censored them at the time of surgical menopause, or enrollment in the cohort, if still premenopausal. Univariate models evaluating potential determinants of age at natural menopause, including but not limited to ethnicity, smoking, parity, education, age at first full-term pregnancy, age at menarche, oral contraceptive use, body mass index (BMI), weight, and height, revealed that ethnicity was the strongest predictor of age at natural menopause. Compared to Ws, the age at natural menopause occurred earliest among Ls. In fact, Ls born outside of the United States (L-NUS) and Latinas born inside of the United States (L-US) had significantly different hazards ratios 1 7 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (HR) for age at natural menopause. L-NUS had the largest HR, or greatest risk for natural menopause at any age (HR (95% Cl): L-NUS = 1.27 (1.23- 1.31) and L-US = 1.14 (1.10-1.18)). Japanese women had the latest natural menopause of the racial/ethnic groups studied (J = 0.89 (0.87-0.92)). AAs did not differ significantly in age at natural menopause from Ws. These findings are parallel to those reported by Kato et al.[9] for A A, W and L women, and are consistent with expectations based on the ethnic-specific incidence rates of breast cancer in the MECJ17J. In a multivariate model Japanese ethnicity, smoking, higher parity, more education, heavier current weight, later age at first full-term pregnancy, and later age at menarche were inversely associated with age at natural menopause. The magnitude of the ethnic-specific hazard ratios did not change upon addition of covariates. These results have motivated the investigation of determinants of age at natural menopause in the California Teacher's Study (CTS), a cohort composed of participants with high breast cancer incidence rates, which cannot be fully explained by known lifestyle and reproductive risk factors (Aim 1). The CTS, established in 1995-1996, had a primary focus of investigating risk factors for the high breast cancer incidence rates of California teachers. 133,479 members of the California State Teacher's Retirement System responded to a baseline questionnaire and were enrolled in the cohort. The baseline CTS questionnaire included 174 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. questions covering a wide variety of experiences and exposures including medical history, menstrual and reproductive history, exogenous hormone use (oral contraceptives and hormone replacement therapy), height, weight, smoking history, alcohol intake and history of physical exercise activity. Two additional questionnaires have been sent to participants since the baseline questionnaire. These obtained information on passive smoke exposure, waist and hip measures, and updates on pregnancy history, exogenous hormone use and menopausal status. Details of the CTS are available elsewhere [18]. 41,394 CTS participants (31% of total, 58% of postmenopausal) reported experiencing a natural menopause on the baseline questionnaire. 18% of postmenopausal women experienced a surgical menopause in the CTS, 4% experienced menopause due to other reasons, and data on type of menopause is missing for 6% of women. The CTS will be a powerful tool for further investigation of the determinants of age at natural menopause. Preliminary results from the MEC have clarified the effect of covariates suggested in the literature and motivated investigation of determinants of age at natural menopause in another large cohort. Racial/ethnic differences in age at menopause in the MEC support the hypothesis that age at natural menopause may be in part genetically determined. In order to investigate this hypothesis, a haplotype-based approach will be used (Aim 2). Preliminary work, 175 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. involving steps one through three of the haplotype-based approach, has been performed for all six candidate genes in the inhibin/ activin pathway (see Study Design Section: Aim 2). For example, in INHA, a gene of size 3.4 kb on chromosome 2, 34 SNPs in a 40 kb region surrounding the locus have been genotyped in a multiethnic panel, called the haplotype plate (HP), which was designed for the purpose of delineating the haplotype blocks in candidate genes. 16 of the 34 SNPs tested in the HP are working SNPs, meaning that the assays genotyped with >75% success and met Hardy-Weinberg criteria. These working SNPs then become candidates for selection as htSNPs as will be described in detail in the Study Design Section for Aim 2. Additional SNPs are currently being evaluated in the HP to aid haplotype block definition in the six candidate genes included in this proposal. B. OBJECTIVE/HYPOTHESIS Hypothesis: Age at natural menopause is a complex trait driven by a combination of reproductive, environmental and genetic factors. Objectives: (1) To investigate the determinants of age at natural menopause in another large cohort of women, the CTS, for which more detailed data have been obtained on important potential covariates; (2) To utilize a candidate gene, haplotype-based approach to study whether 176 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. variation in genes of the inhibin/ activin pathway may be associated with age at natural menopause in the MEC. C. SPECIFIC AIMS AIM 1: Investigate association between reproductive and lifestyle factors and age at natural menopause in the CTS. AIM 2: Determine whether sequence variation within candidate genes from a relevant biological pathway, the inhibin/ activin pathway, modulates the timing of a woman's natural menopause. D. STUDY DESIGN AIM 1: Determinants of age at natural menopause in the California Teacher's Study 1. Study Design and Eligibility. Analysis of the factors associated with age at natural menopause in the CTS will be performed, using a cohort analysis approach. Data from CTS participants who meet the following eligibiltiy criteria will be analyzed: 1) no prevalent breast, ovarian or endometrial cancer; 2) no missing data on critical analytic variables. 2. Power and Sample Size Considerations. We will have excellent statistical power to find factors that influence the risk of natural menopause at any age in a Cox regression analysis. Table 1 shows the 177 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. minimum detectable relative risks for a range of exposure frequencies. These calculations were performed in POWER (Epicenter Software, Pasadena, CA) assuming alpha=0.05, beta=0.80 and a two-sided hypothesis test. Table 1: Minimum Detectable Relative Risks with 80% Power (N=100,000) Exposure Frequency RR 10% ' 1.06 20% 1.05 30% 1.04 Statistical Analysis. Cox proportional hazards regression will be used to investigate possible associations between reproductive and lifestyle factors with age at natural menopause in the CTS. Natural menopause will be the event of interest, and surgical and premenopausal women will be censored in a manner similar to that performed in the MEC. AIM 2: The inhibin/ activin gene pathway: A haplotype-based approach investigating potential association with age at natural menopause 1. Study Design and Eligibility. In the MEC, common genetic variation in six members of the inhibin/ activin gene family (INHA, INHbetaA, 178 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. INHbetaB, FST, ACVR1 and ACVR2) will be investigated, using a haplotype-based approach, for possible association with age at natural menopause. The haplotype-based approach consists of six steps: (1) Select SNPs at a density of 3 kilobases (kb) on quality of source (i.e. double-hit SNPs in dbSNP), and potential functionality, to cover 20 kb upstream through 10 kb downstream of gene and illustrate positions of all published SNPs, intron/exon boundaries, putative regulatory regions, and areas of high hum an/ mouse homology; (2) Genotype selected SNPs in Haplotype Plate (HP). The HP is composed of DNAs from 280 healthy female MEC participants, 70 of each of the four major racial/ ethnic groups in this study plus 10% repeats; (3) Initial analysis of genotyping data to determine linkage disequilibrium (LD) patterns across candidate genes; (4) Frame haplotype blocks using the methods of Gabriel et al.[19]. (4) Within these blocks htSNPs will be identified using the method of Stram et al.[20]; (5) Genotype selected htSNPs in natural menopause sample. Natural menopause sample consists of 1500 MEC participants who reported experiencing a natural menopause. The 1500 women includes 500 women from each of three age at natural menopause categories (<44, 45-54,55+). Within each age at natural menopause category, 100 women were selected from each of five racial/ ethnic groups (W, A A, J, L-US, L- NUS). Only women who reported no history of breast, endometrial or 179 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ovarian cancer on the baseline questionnaire were eligible for this natural menopause sample. (6) Analyze data (see AIM 2: Statistical Analysis). 2. Power and Sample Size Considerations. For the haplotype-based approach to evaluate the role of genetic variants in the inhibin/ activin pathway in the age of natural menopause within the MEC, a study design will be employed that is well powered to detect differences in age at natural menopause between three age at natural menopause groups (<=44,45-54, 55+). Table 2 illustrates the minimum detectable relative risks for comparison of either extreme age at natural menopause category (<=44 or 55+) with "normal" (45-54). Column A shows relative risks for an analysis of a single SNP (based on allele frequency) and column B shows relative risks for an analysis of a haplotype (based on haplotype frequency, adjusted by a sample size inflation factor according to methods described in (15)). This study is underpowered to investigate racially- stratified haplotype associations with age at natural menopause; however, we have capacity to expand the study in future work if preliminary racially stratified analyses strongly suggest the presence of any race- specific effects. These power calculations were performed using the Quanto program (Quanto version 0.4,[21]). Assumptions included a log additive model, an alpha of 5% and a two-sided hypothesis test. These 180 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. relative risks are within the expected range. Over 3500 MEC subjects currently meet the eligibility criteria outlined, demonstrating that sufficient subjects exist to perform the proposed study. Table 2: Minimum Detectable Relative Risks with 80% Power (1 Control per Case, 500 Controls) A. SNP Analysis B. Haplotype Analysis Allele/Haplotvpe Frequency RR RR 10% ' 1.47 1.54 15% 1.39 1.45 20% 1.35 1.40 25% 1,32 1.37 Statistical Analysis. HtSNP genotypes and estimated haplotypes and their interactions with associated factors (eg. smoking) will be investigated using the method described in Haiman et al.[15]. Possible approaches include logistic regression comparing early or late natural menopause vs. normal age of at natural menopause, polychotomous logistic regression, or Cox regression as was done in the covariate analysis above, treating age at natural menopause as the outcome. 4. The Laboratory Approach. All remaining genotyping will be accomplished through ongoing collaborations with investigators at the Massachusetts Institute of Technology (MIT) and here at USC. Genotyping of the HP was performed 181 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. by MALDI-TOF mass spectroscopy (Sequenom, Inc.) at MIT. Additional genotyping, to fill in areas of poor LD is currently being done on the Illumina platform at MIT. Genotyping of htSNPs in the natural menopause sample is being performed using the Taqman Allelic Discrimination assay at the USC-MEC Genotyping Laboratory using the ABI7900 Sequence Detection System (Applied Biosystems, Foster City, CA).E. POTENTIAL OUTCOMES AND BENEFITS OF THE RESEARCH An unprecedented number of women are now postmenopausal, and the age at natural menopause has broad implications for postmenopausal health. In particular, epidemiological studies have shown that an early menopause is associated with lower risk of several hormonally driven diseases, including breast cancer[22]. Identification of the factors that determine timing of natural menopause, including any common genetic variants, will provide insight into the complex etiology of this important breast cancer risk factor, shed light on the racial/ ethnic differences in breast cancer incidence and potentially allow for targeted breast cancer prevention efforts. 182 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. F. TIMELINE Year One 6/05-12/05 1/06-5/06 Year Two 6/06-12/06 1/07-5/07 Year Three 6/07-12/07 1/08-5/08 AIM I: A) Clean/Analyze CTS data B) Prepare tables/manuscripts X X X X X X AIM 2: A) HtSNP selection genotyping B) Analyze genetic data C) Prepare tables/manuscripts X X X X X X X X G. DISSEMINATION PLAN The CTS manuscript entitled "Determinants of the age at natural menopause: The California Teacher's Study" will be submitted to Fertility and Sterility in an effort to reach the clinical community. Papers resulting from the haplotype-based approach to the inhibin/ activin pathway will be submitted to Human Molecular Genetics, in light of the positive response to the paper by Haiman et al.[15]. I plan to submit an abstract to the annual conference of the American Association for Cancer Research, and I would be pleased to present, if invited, at a Susan G. Komen Breast Cancer Foundation conference. References 1. (SEER), S., Epidemiology, and End Results Program (www.seer.cancer.gov), Public Use Data (1973-2001), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch. 183 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2. Finch, C.E., Longevity, Senescence, and the Genome. The John D. and Catherine T. MacArthur Foundation Series on Mental Health and Development. 1990, Chicago: The University of Chicago Press. 922.3. 3. Karagas, M.R., J. Kelsey, and V. McGuire, Cancers of the Female Reproductive System, in Menopause: Biology and Pathology, R. Lobo, J. Kelsey, and R. Marcus, Editors. 2000, Academic Press: San Diego, p. 359- 381.4. 4. Rebar, R.W., Premature Ovarian Failure, in Menopause: Biology and Pathobiology, R.A. Lobo, J. Kelsey, and R. Marcus, Editors. 2000, Academic Press: San Diego, p. 135-146.5. 5. MacMahon, B. and J. Worcester, Age at menopause. United States— 1960-1962. Vital Health Stat 1,1966.11(19): p. 1-20.6. Nagata, C„ et al„ Association of diet with the onset of menopause in Japanese women. American Journal of Epidemiology, 2000.152(9): p. 863-7.7. 6. Akahoshi, M., et al., The effects of body mass index on age at menopause. International Journal of Obesity & Related Metabolic Disorders, 2002. 26(7): p. 961-8.8. 7. WHO, Research on the Menopause: Report of a WHO Scientific Group. 1981, World Health Organization: Geneva.9. 184 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8. Kato, I., et al., Prospective Study of Factors Influencing the Onset of Natural Menopause. Journal of Clinical Epidemiology, 1998. 51(12): p. 1271. 9. Cramer, D.W., H. Xu, and B.L. Harlow, Does "incessant" ovulation increase risk for early menopause? American Journal of Obstetrics & Gynecology, 1995.172(2 Pt 1): p. 568-73.11. 10. Brambilla, D.J. and S.M. McKinlay, A prospective study of factors affecting age at menopause, [erratum appears in J Clin Epidemiol 1990;43(5):537.J. Journal of Clinical Epidemiology, 1989. 42(11): p. 1031. 11. Torgerson, D.J., et al., Alcohol consumption and age of maternal menopause are associated with menopause onset. Maturitas, 1997. 26(1): p. 21-5.13. 12. Treloar, S. A., K.A. Do, and N.G. Martin, Genetic influences on the age at menopause. Lancet, 1998. 352(9134): p. 1084-5.14. 13. Snieder, H., A.J. MacGregor, and T.D. Spector, Genes control the cessation of a woman's reproductive life: a twin study of hysterectomy and age at menopause. J Clin Endocrinol Metab, 1998. 83(6): p. 1875-80.15. 14. Haiman, C.A., et al., A comprehensive haplotype analysis of CYP19 and breast cancer risk: the Multiethnic Cohort. Human Molecular Genetics, 2003.12(20): p. 2679-92.16. 1 8 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 15. Kolonel, L.N., et al., A multiethnic cohort in Hawaii and Los Angeles: baseline characteristics. American Journal of Epidemiology, 2000.151(4): p. 346-57.17. 16. Pike, M.C., et al., Breast cancer in a multiethnic cohort in Hawaii and Los Angeles: risk factor-adjusted incidence in Japanese equals and in Hawaiians exceeds that in Whites. Cancer Epidemiol Biomarkers Prev, 2002.11(9): p. 795-800.18. 17. Bernstein, L., et al., High breast cancer incidence rates among California teachers: results from the California Teachers Study (United States). Cancer Causes & Control, 2002.13(7): p. 625-35.19. 18. Gabriel, S.B., et al., The structure of haplotype blocks in the human genome. Science, 2002. 296(5576): p. 2225-9.20. Stram, D.O., et al., Choosing haplotype-tagging SNPS based on unphased genotype data using a preliminary sample of unrelated subjects with an example from the Multiethnic Cohort Study. Hum Hered, 2003. 55(1): p. 27-36.21. 19. Gauderman, W.J., Sample size requirements for association studies of gene-gene interaction. Am J Epidemiol, 2002.155(5): p. 478-84.22. 20. Trichopoulos, D., B. MacMahon, and P. Cole, Menopause and breast cancer risk. J Natl Cancer Inst, 1972. 48(3): p. 605-13. 186 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. BUDGETFORM Budget Category Detailed Budget for Entire Budget Period ITEMIZE ALL CATEGORIES IN BUDGET JUSTIFICATION Year One Year Two Y ear Three Dollar Amount Requested (U.S. currency, omit cents) TOTAL Personnel Costs* (Itemize in Budget Justification) 44,095.00 44,095.00 44.095.00 132.285.00 Supplies (Itemize in Budget Justification) 0.00 0.00 0.00 0.00 Tuition 0.00 0.00 0.00 0 0 0 Travel (Itemize in Budget Justification) 905.00 905.00 905.00 2,715.00 Other Expenses (Itemize in Budget Justification) 0.00 0.00 0.00 0.00 Total Funding Request (Cannot exceed $45,000 per year ) 45,000.00 45,000.00 45.000.00 135.000.00 *Postdoctoral Fellow must have a 100% effort on project. Indicate Fringe Benefit rate in Budget Justification. The current postdoctoral award will provide salary support for analysis of data for this specific outcome, age at natural menopause. Completion of genotyping is supported by the Genetic Susceptibility to Cancer in Multiethnic Cohorts (NCI R01 CA 63464) and Candidate Gene Association Studies of Ovarian Cancer (CACRP 2110170). The fringe benefit rate is 34% for the period 7/1/04-6/30/05. From 7/1/05-6/30/06 and beyond, the fringe benefit rate is estimated to be 34.75%. PERSONNEL Dr. Leslie Bernstein (Sponsor): Involved in all aspects of the study in an 187 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. advisory role. Ms. Katherine DeLellis (Postdoctoral Fellow, 100%) has coordinated selection of study samples as part of pre-doctoral work and will be responsible for oversight of laboratory activities including preparation of master and diluation plates of samples, preparation of reaction mixes and reactions using the robotic platforms and use of Taqman for genotyping. Genotyping is being accomplished through an ongoing research study of hormones and cancer. Primary roles will be as data analyst, statistician and primary author of manuscripts. TRAVEL A travel budget is requested for one trip per year for presentation of results at a conference such as a Susan G. Komen Breast Cancer Foundation conference. Estimated cost of travel is $500.00 (airfare) + $100.00 x 3 (hotel/night) + $35.00 x 3 (meal allowance/day) = $905.00. 188 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Letter of Support August 2,2004 Re: Katherine DeLellis' Application for a Susan G. Komen Foundation Postdoctoral Fellowship I write this letter to provide my strong support for Ms. Katherine DeLellis' application for a postdoctoral fellowship award. I am extremely pleased to serve as the Sponsor for Ms. DeLellis' postdoctoral studies. Ms DeLellis will receive her PhD in epidemiology from the Department of Preventive Medicine at the Keck School of Medicine of the University of Southern California (USC) in early Fall, 2004. As a PhD student, she has worked intensively with Drs. Sue Ingles, Malcolm Pike, Chris Haiman, Dan Stram and Brian Henderson at USC. In her postdoctoral fellowship, she will expand her interests in the determinants of age at menopause as a critical risk factor for breast cancer. She has come to me with her ideas regarding the genetics of menopause which I find fascinating. She also wants to explore further the determinants of menopause in our California Teachers Study cohort. All of the ideas in her proposal are her own - and I look forward to guiding her through her postdoctoral fellowship years, learning with her about the genetic determinants of age at menopause, and helping her to navigate the complex datasets we have developed for 189 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the California Teachers Study. I am certain that her ideas will expand to encompass other genetically determined "traits" that have bearing on breast cancer risk through her proposed research projects. Ms. DeLellis and I met initially a year ago when she asked me to consider taking her on as a postdoctoral fellow when she graduated with her PhD. At the time we met, she was initiating several avenues of research for her dissertation and as part of her responsibilities as a research assistant was working on the Multiethnic Cohort (MEC) Study. As the year has progressed, Ms. DeLellis and I have met to discuss her ideas which have become focused on the determinants of age at menopause. Ms. DeLellis is trained not only in epidemiology, but also in laboratory methods of molecular biology, so that she can carry out the experimental work needed to conduct Aim 2 of her proposed research program. As a biostatistician by training, I am always alert to the abilities of students and colleagues in the area of biostatistics and data analysis. I find that Ms. DeLellis meets my high standards regarding statistical analytic approaches, in large part, I am certain due to her collaborations during her training with Drs. Pike and Stram in analyses of the MEC. It has been clear to me through my prior meetings with Ms. DeLellis that her abilities and prior training present a unique opportunity for both of us to work together on the complex problem she has posed. Age at menopause is a critical risk factor for 190 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. breast cancer; breast cancer risk increases rapidly throughout a woman's reproductive years and the earlier that a woman enters the menopausal period of her life, the lower is her risk of breast cancer. Age-specific incidence curves in all populations reflect the fact that following menopause, breast cancer risk levels off. Furthermore, supplementation with a combined hormone regimen of estrogen and progestin following menopause confirms the important role of these reproductive hormones in determining a woman's risk. Yet we know very little about what determines a woman's age at natural menopause. If certain lifestyle factors predict later menopause, these are amenable to change and will likely act to reduce a woman's risk of breast cancer. But there is also suggestive evidence regarding heritability of age at menopause from twin and other family studies. It is important for us to know what role genetics plays in the timing of this event. The genetic pathways proposed by Ms. DeLellis have not been studied with regard to age at menopause; in fact, only now is work being done to determine the haplotypes in the critical genes she has proposed to study. These pathways have an impact on the functioning of the hypothalamic pituitary axis which governs many aspects of a woman's hormonal exposure and thus, may also have some more direct impact on breast cancer risk. Part of my prior research has focused on direct hormonal determinants of breast cancer risk, as well as 191 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. determinants of age at menarche, another important breast cancer risk factor. This is why I am so intrigued by the research proposed by Ms. DeLellis and look forward to working w ith her for the next 3 years. I believe that Ms. DeLellis' proposal shows the high promise that she has for a future in breast cancer research. As indicated below (see description of my research program), I have broad interests in the etiology and prevention of breast cancer and am enthusiastic about the area of breast cancer research that Ms. DeLellis will pursue. The Norris Comprehensive Cancer Center at USC has a formal Breast Cancer Research Program and I co-direct this program with Michael Press, MD, PhD, a pathologist and molecular biologist. This program has weekly conferences (eachThursday) at which both clinical and research discussions take place. At least monthly, a program member presents the results of his or her ongoing breast cancer research. Ms. DeLellis will attend these meetings and as she progresses with her own research will be encouraged to present her results at these conferences. Ms. DeLellis will also attend weekly grand rounds at the Cancer Center and the ongoing biweekly seminar series within the Department of Preventive Medicine. In addition, both the USC Institute of Genetic Medicine and the Zilkha Neurogenetic Institute have regular seminar series which bring outstanding scientists to USC who are working on important and novel 192 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. issues in genetics. Ms. DeLellis will also have the opportunity of interacting with other members of the Department of Preventive Medicine, many of whom have interests in the etiology of breast cancer and will maintain her relationship with the MEC, while developing a relationship with the investigators involved, with me, in the California Teachers Study. The Department of Preventive Medicine offers graduate training in epidemiology, biometry and health behavior research that is provided by an excellent and internationally recognized faculty of epidemiologists, biostatisticians, geneticists, psychologists, and psychometricians. As her research findings develop, I will see to it that Ms. DeLellis has the opportunity of presenting these at key meetings related to breast cancer and cancer prevention including the annual San Antonio Breast Cancer Symposium, the annual AACR Cancer Prevention Meeting, the Society for Epidemiologic Research and the American Society for Preventive Oncology. Ms. DeLellis will be provided office space in an office suite on the third floor of the Norris Cancer Center Tower Building. The computer facilities available to Ms. DeLellis include networks within each of the research programs in the Department of Preventive Medicine that can be accessed directly from personal computers at the Norris Cancer Center via ethernet. A variety of statistical software packages (such as SAS, Epilog, Epicure) are available 193 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and many of us in the department have the expertise to provide guidance and assistance to post-doctoral fellows in constructing and testing complex statistical models. Although I was trained as a biostatistician, I have pursued a career in cancer epidemiology. My main research interest throughout my career has been the etiology and prevention of breast cancer. During the 1980s, much of my research focused on role of endogenous hormones (particularly estrogens and estrogen binding fractions) and breast cancer risk. I conducted numerous studies assessing differences in circulating hormone levels of women at high and at low risk of breast cancer and among breast cancer patients with localized disease and among well-matched controls. In the late 1980s, I conducted a study of the effects of moderate exercise activity on menstrual cycle patterns of adolescent girls which led to my major effort to determine whether participation in exercise activity would reduce women's risk of breast cancer and to identify determinants of age at menarche. I am one of the founders of the California Teacher Study and direct all data analysis of the cohort. I currently have ongoing studies on etiology and pathogenesis, secondary outcomes following breast cancer treatment, factors affecting breast cancer prognosis and quality of life issues for the breast cancer survivor. I have extensive experience as an advisor of PhD students and postdoctoral fellows. My office and Ms. DeLellis' office are quite 194 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. close to each other which will facilitate our interactions. As needed, we can interact on a daily basis. I find that this sort of availability is necessary in working with graduate and post-doctoral students on their research projects. This allows me always to be up-to-date on the fellow's progress and to assist when problems arise. In addition, I will schedule a one hour weekly meeting with Ms. DeLellis to review her progress and to develop strategies for addressing her work. Other longer working sessions will be scheduled as she progresses w ith her research. I also plan to include Ms. DeLellis in any meetings that I have related to my other breast cancer research studies, so that she can participate in the ongoing development of new studies and management of current studies. My breast cancer research staff function as a team, providing support and solutions to problems based on each individual's expertise. Ms. DeLellis will therefore gain much experience in the conduct and management of epidemiologic and follow-up studies of breast cancer. My goal in serving as Ms. DeLellis' sponsor during her postdoctoral fellowship is to provide her with the necessary training and skills to function as an independent breast cancer epidemiologist in an academic setting. I will encourage her to continue to develop her own research ideas, to write research proposals for further studies and to seek external funding for her research. Sincerely, 195 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Leslie Bernstein, Ph.D. Professor, Preventive Medicine AFLAC, Inc., Chair in Cancer Research Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. August 2,2004 Katherine DeLellis, M.S. Department of Preventive Medicine Keck School of Medicine 1441 Eastlake Avenue, RM 4411 Los Angeles, CA 90033 Dear Katherine, It is my pleasure to confirm my willingness to serve as a mentor on your grant proposal entitled "Reproductive, Lifestyle and Genetic Determinants of the Age of Natural Menopause" for the Susan G. Komen Breast Cancer Foundation Postdoctoral Fellowship Grant Application. It is our hope that this project will definitively answer some of the questions about the determinants of the age of natural menopause and shed light on the potential role of genetics on this understudied topic. As a statistician with enormous interest (and also some experience) in the general methodological issues related to your study (including haplotype discovery & effect estimation, etc) I look forward to guiding you on this project and have no doubt that we will carry this project to successful completion. Sincerely, 1 9 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Dan Stram Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Letter of Support August 10,2004 Sue Ingles Assistant Professor of Preventive Medicine Director, Program in Molecular Epidemiology USC/Norris Cancer Center Room 64191441 Eastlake Ave, MS 44 Los Angeles CA 90089 Zip 90033 323-865-0498; FAX 323-865-0473 Lab: 323-865-0472 I write this letter to provide my strong support for Katherine DeLellis' application for a postdoctoral fellowship award. I have been able to observe Ms. DeLellis' development since she started her graduate studies at USC. She completed the laboratory work necessary for her M.S. research in my laboratory. Although she came to my lab with little laboratory experience, she quickly became quite proficient in the laboratory skills necessary for her research. Subsequent to her M.S. degree, I served as her advisor for her Ph.D. research on the genetics of menopause in the Multiethnic Cohort Study. Her dissertation work has given her the opportunity to expand her knowledge of molecular genetics as well as statistical analysis. She has received excellent mentoring in data analyses by Drs. Pike and Stram, and has completed some rather 199 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. complicated data analyses. In short, she has an excellent background in epidemiology, molecular biology, and statistical analysis that will allow her to successfully complete her proposed research. Furthermore, under Dr. Bernstein's direction she will receive excellent training and support that will allow her to develop into an independent breast cancer researcher. Sincerely, Dr. Sue Ingles 200 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ALPHABETIZED BIBLIOGRAPHY Aina, A. 0. (1992). "An investigation into the climacteric in Nigerians." Journal of the Medical Association of Thailand 75(3): 168-72. Akahoshi, M., M. Soda, et al. (2002). "The effects of body mass index on age at menopause." International Journal of Obesity & Related Metabolic Disorders 26(7): 961-8. Akkina, J., J. Reif, et al. (2004). "Age at natural menopause and exposure to organochlorine pesticides in Hispanic women." Journal of Toxicology & Environmental Health Part A 67(18): 1407-22. Alexandre, H. (2001). "A history of mammalian embryological research." International Journal of Developmental Biology 45(3): 457-67. Andersen, F. S., I. Transbol, et al. (1982). "Is cigarette smoking a promotor of the menopause?" Acta Medica Scandinavica 212(3): 137-9. Armstrong, D., A. Goff, et al. (1979). Regulation of follicular estrogen biosynthesis. Ovarian Follicular Development and Function. A. Midgley Jr and W. Sadler Jr. New York, Raven Press: 169-182. Bailey, A., D. Robinson, et al. (1977). "Smoking and age of natural menopause." Lancet 2(8040): 722. Barbieri, R. L., J. Gochberg, et al. (1986). "Nicotine, cotinine, and anabasine inhibit aromatase in human trophoblast in vitro." Journal of Clinical Investigation 77(6): 1727-33. Barbieri, R. L., P. M. McShane, et al. (1986). "Constituents of cigarette smoke inhibit human granulosa cell aromatase embryotoxicity of benzo(a)pyrene and some of its synthetic derivatives in Swiss mice." Fertility & Sterility 46(2): 232-6. Barrett, J., B. Fry, et al. (2005). "Haploview: analysis and visualization of LD and haplotype maps." Bioinformatics. 201 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Barrett-Connor, E. (1993). "Epidemiology and the menopause: a global overview." International Journal of Fertility & Menopausal Studies 38 Suppl 1: 6-14. Barton, D., T. Yang-Feng, et al. (1987). "INHA, INHBA and INHBB, the loci for the three subunits of inhibin, mapped in mouse and man. (Abstract)." Cytogenet. Cell Genet. 46: 578 only. Barton, D. E., T. L. Yang-Feng, et al. (1989). "Mapping of genes for inhibin subunits alpha, beta A, and beta B on human and mouse chromosomes and studies of jsd mice level of expression and chromosome mapping of the mouse cholecystokinin gene: implications for murine models of genetic obesity." Genomics 5(1): 91-9. Bernhard, P. (1949). "Certain injurious effects of cigarette smoking on women." Mediziniche Monatsschrift 3: 58-60. Bernstein, L. (2002). "Epidemiology of endocrine-related risk factors for breast cancer." Journal of Mammary Gland Biology & Neoplasia 7(1): 3-15. Biggers, J. D., C. A. Finn, et al. (1962). "Long-term reproductive performance of female mice. I. Effect of removing one ovary." J Reprod Fertil 3: 303-12. Biggers, J. D., C. A. Finn, et al. (1962). "Long-term reproductive performance of female mice. II. Variation of litter size with parity." J Reprod Fertil 3: 313-30. Block, E. (1952). "Quantitative morphological investigations of the follicular system in women." Acta Anatomica XIV: 108. Block, E. (1953). "A quantitative morphological investigation of the follicular system in newborn female infants." Acta Anatomica XVII(3): 201-206. Bondestam, J., N. Horelli-Kuitunen, et al. (1999). "Assignment of ACVR2 and ACVR2B the human activin receptor type II and IIB genes to chromosome bands 2q22.2— >q23.3 and 3p22 and the human follistatin gene (FST) to chromosome 5ql 1.2 by FISH." Cytogenetics & Cell Genetics 87(3-4): 219-20. Brambilla, D. J. and S. M. McKinlay (1989). "A prospective study of factors affecting age at menopause, [erratum appears in J Clin Epidemiol 1990;43(5):537.]." Journal of Clinical Epidemiology 42( 1IV . 1031-9. 202 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Bromberger, J. T., K. A. Matthews, et al. (1997). "Prospective study of the determinants of age at menopause." American Journal of Epidemiology 145(2): 124-33. Burger, H. G., E. C. Dudley, et al. (1999). "Prospectively measured levels of serum follicle-stimulating hormone, estradiol, and the dimeric inhibins during the menopausal transition in a population-based cohort of women." Journal of Clinical Endocrinology & Metabolism 84(11): 4025-30. Cooper, G. S., D. D. Baird, et al. (2001). "Measures of menopausal status in relation to demographic, reproductive, and behavioral characteristics in a population-based study of women aged 35-49 years." American Journal of Epidemiology 153(12): 1159-65. Cooper, G. S., D. P. Sandler, et al. (1999). "Active and passive smoking and the occurrence of natural menopause." Epidemiology 10(6): 771-3. Cosgrave, M. P., J. Tyrrell, et al. (1999). "Age at onset of dementia and age of menopause in women with Down's syndrome." Journal of Intellectual Disability Research 43(Pt 6): 461-5. Cramer, D. W., H. Xu, et al. (1995). "Does "incessant" ovulation increase risk for early menopause?" American Journal of Obstetrics & Gynecology 172(2 Pt 1): 568-73. Cramer, D. W., H. Xu, et al. (1995). "Family history as a predictor of early menopause." Fertility & Sterility 64(4): 740-5. Crawford, S. (2000). Epidemiology: Methodologic Challenges in the Study of Menopause. Menopause: Biology and Pathobiology. R. A. Lobo, J. L. Kelsey and R. Marcus. San Diego, Academic Press: 159-174. Curtis, K. M., D. A. Savitz, et al. (1997). "Effects of cigarette smoking, caffeine consumption, and alcohol intake on fecundability." American Journal of Epidemiology 146(1): 32-41. Daly, M. J., J. D. Rioux, et al. (2001). "High-resolution haplotype structure in the human genome." Nature Genetics 29(2): 229-32. de Bruin, J. P., H. Bovenhuis, et al. (2001). "The role of genetic factors in age at natural menopause." Human Reproduction 16(9): 2014-8. De Jong, F. H. and R. M. Sharpe (1976). "Evidence for inhibin-like activity in bovine follicular fluid." Nature 263(5572): 71-2. 203 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Do, K. A., B. M. Broom, et al. (2000). "Genetic analysis of the age at menopause by using estimating equations and Bayesian random effects models." Statistics in Medicine 19(9): 1217-35. Driancourt, M. A., L. P. Cahill, et al. (1985). "Ovarian follicular populations and preovulatory enlargement in Booroola and control Merino ewes." Journal of Reproduction & Fertility 73(1): 93-107. Erickson, G. F. and A. J. Hsueh (1978). "Secretion of "inhibin" by rat granulosa cells in vitro." Endocrinology 103(5): 1960-3. Everson, R. B., D. P. Sandler, et al. (1986). "Effect of passive exposure to smoking on age at natural menopause." British Medical Journal Clinical Research Ed. 293(6550): 792. Excoffier, L. and M. Slatkin (1995). "Maximum-likelihood estimation of molecular haplotype frequencies in a diploid population." Molecular Biology & Evolution 12(5): 921-7. Faddy, M. J. and R. G. Gosden (1996). "A model conforming the decline in follicle numbers to the age of menopause in women." Human Reproduction 11(7): 1484-6. Faddy, M. J., R. G. Gosden, et al. (1992). "Accelerated disappearance of ovarian follicles in mid-life: implications for forecasting menopause." Human Reproduction 7(10): 1342-6. Finch, C. E. (1990). Longevity. Senescence, and the Genome. Chicago, The University of Chicago Press. Finn, C. A. (2001). "Reproductive ageing and the menopause." International Journal of Developmental Biology 45(3): 613-7. Franchimont, P., J. Verstraelen-Proyard, et al. (1979). "Inhibin: from concept to reality." Vitamins & Hormones 37: 243-302. Frere, G. (1971). "Mean age at menopause and menarche in South Africa." South African Journal of Medical Sciences 36(1): 21-4. Frisch, R. and M. JW (1974). "Menstrual cycles: Fatness as a determinant of minimum weight for height necessary for their maintenance or onset." Science 185: 949-951. 204 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Gabriel, S. B., S. F. Schaffner, et al. (2002). "The structure of haplotype blocks in the human genome." Science 296(5576): 2225-9. Garcia Vela, A., L. E. Nava, et al. (1987). "La edad de la menopausia en la poblacion urbana de la ciudad de Leon, Gto." Revista de Investigation Clinica 39(4): 329-32. Gauderman, W. J. (2002). "Sample size requirements for association studies of gene-gene interaction." Am J Epidemiol 155(5): 478-84. Gold, E. B., J. Bromberger, et al. (2001). "Factors associated with age at natural menopause in a multiethnic sample of midlife women: Clinical pathway for evaluating women with abnormal uterine bleeding." American Journal of Epidemiology 153(9): 865-74. Gorai, I., K. Tanaka, et al. (2003). "Estrogen-metabolizing gene polymorphisms, but not estrogen receptor-alpha gene polymorphisms, are associated with the onset of menarche in healthy postmenopausal Japanese women." Journal of Clinical Endocrinology & Metabolism 88(2): 799-803. Gosden, R. G. (1987). "Follicular status at the menopause." Human Reproduction 2(7): 617-21. Gosden, R. G., S. C. Laing, et al. (1983). "Imminent oocyte exhaustion and reduced follicular recruitment mark the transition to acyclicity in aging C57BL/6J mice." Biol. Reprod. 28: 255-260. Gougeon, A. (1984). Influence of cyclic variations in gonadotrophin and steroid hormones on follicular growth in the human ovary. Clinical Pathology of the Endocrine Ovary. J. de Brux and J. Gautray. Lancaster, MTP Press: 63-72. Gougeon, A. (1996). "Regulation of ovarian follicular development in primates: facts and hypotheses." Endocrine Reviews 17(2): 121-55. Gougeon, A. (1998). "Ovarian follicular growth in humans: ovarian ageing and population of growing follicles." Maturitas 30(2): 137-42. Gougeon, A., R. Ecochard, et al. (1994). "Age-related changes of the population of human ovarian follicles: increase in the disappearance rate of non growing and early-growing follicles in aging women." Biology of Reproduction 50(3): 653-63. 205 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Groome, N. P., P. J. Illingworth, et al. (1996). "Measurement of dimeric inhibin B throughout the human menstrual cycle." Journal of Clinical Endocrinology & Metabolism 81(4): 1401-5. Haiman, C. A., D. O. Stram, et al. (2003). "A comprehensive haplotype analysis of CYP19 and breast cancer risk: the Multiethnic Cohort." Human Molecular Genetics 12(20): 2679-92. Hankinson, S. E., G. A. Colditz, et al. (2004). "Towards an integrated model for breast cancer etiology: the lifelong interplay of genes, lifestyle, and hormones." Breast Cancer Research 6(5): 213-8. Hardy, R., D. Kuh, et al. (2000). "Smoking, body mass index, socioeconomic status and the menopausal transition in a British national cohort." International Journal of Epidemiology 29(5): 845-51. Hefler, L. A., C. Worda, et al. (2002). "A polymorphism of the Nos3 gene and age at natural menopause." Fertility & Sterility 78(6): 1184-6. Hirschhom, J., C. L. Pearce, et al. (2003). Genomic Approaches to the Genetics of Hormone-Responsive Cancer. Hormones. Genes, and Cancer. B. E. Henderson, B. A. Ponder and R. K. Ross. Oxford, Oxford University Press: 99-119. Hoel, D. G., T. Wakabayashi, et al. (1983). "Secular trends in the distributions of the breast cancer risk factors-menarche, first birth, menopause, and weight— in Hiroshima and Nagasaki, Japan." American Journal of Epidemiology 118(1): 78-89. Hsueh, A. J., H. Billig, et al. (1994). "Ovarian follicle atresia: a hormonally controlled apoptotic process." Endocrine Reviews 15(6): 707-24. Ismael, N. N. (1994). "A study on the menopause in Malaysia." Maturitas 19(3): 205-9. Jick, H., J. Porter, et al. (1977). "Relation between smoking and age of natural menopause." Lancet i: 1354-1355. Kato, I., P. Toniolo, et al. (1998). "Prospective Study of Factors Influencing the Onset of Natural Menopause." Journal of Clinical Epidemiology 51(12): 1271-6. 206 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Kaufman, D. W., D. Slone, et al. (1980). "Cigarette smoking and age at natural menopause: Estrogen receptors, breast cancer, and smoking." American Journal of Public Health 70(4): 420-2. Kogawa, K., T. Nakamura, et al. (1991). "Activin-binding protein is present in pituitary." Endocrinology 128(3): 1434-40. Kok, H. S., K. M. van Asselt, et al. (2004). "Age at natural menopause is not linked with the follicle-stimulating hormone receptor region: a sib-pair study." Fertility & Sterility 81(3): 611-6. Kolonel, L. N., B. E. Henderson, et al. (2000). "A multiethnic cohort in Hawaii and Los Angeles: baseline characteristics." American Journal of Epidemiology 151(4): 346-57. Krailo, M. D. and M. C. Pike (1983). "Estimation of the distribution of age at natural menopause from prevalence data." American Journal of Epidemiology 117(3): 356-61. Kwawukume, E. Y., T. S. Ghosh, et al. (1993). "Menopausal age of Ghanaian women." International Journal of Gynaecology & Obstetrics 40(2): 151-5. Lawlor, D. A., S. Ebrahim, et al. (2003). "The association of socio-economic position across the life course and age at menopause: the British Women's Heart and Health Study." BJOG: an International Journal of Obstetrics & Gynaecology 110(12): 1078-87. Lewontin, R. C. (1964). "The interaction of selection and linkage. General considerations; heterotic models." Genetics 49: 49-67. Lindquist, O. and C. Bengtsson (1979). "Menopausal age in relation to smoking." Acta Medica Scandinavica 205(1-2): 73-7. Ling, N., S. Y. Ying, et al. (1985). "Isolation and partial characterization of a Mr 32,000 protein with inhibin activity from porcine follicular fluid." Proceedings of the National Academy of Sciences of the United States of America 82(21): 7217-21. Ling, N., S. Y. Ying, et al. (1986). "Pituitary FSH is released by a heterodimer of the beta-subunits from the two forms of inhibin." Nature 321(6072): 779- 82. 207 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Luoto, R., J. Kaprio, et al. (1994). "Age at natural menopause and sociodemographic status in Finland." American Journal of Epidemiology 139(1): 64-76. MacMahon, B., D. Trichopoulos, et al. (1982). "Cigarette smoking and urinary estrogens." New England Journal of Medicine 307(17): 1062-5. MacMahon, B. and J. Worcester (1966). "Age at menopause. United States— 1960-1962." Vital Health Stat 1 11(191: 1-20. Mattison, D. R., M. I. Evans, et al. (1984). "Familial premature ovarian failure." American Journal of Human Genetics 36(6): 1341-8. McCullagh, D. R. (1932). "Dual endocrine activity of the testes." Science 76: 19- 20. McKinlay, S. M., D. J. Brambilla, et al. (1992). "The normal menopause transition.[see comment]." Maturitas 14(2): 103-15. Meunier, H., S. B. Cajander, et al. (1988). "Rapid changes in the expression of inhibin alpha-, beta A-, and beta B-subunits in ovarian cell types during the rat estrous cycle." Molecular Endocrinology 2(12): 1352-63. Midgette, A. S. and J. A. Baron (1990). "Cigarette smoking and the risk of natural menopause." Epidemiology 1(6): 474-80. Morabia, A. and M. C. Costanza (1998). "International variability in ages at menarche, first livebirth, and menopause. World Health Organization Collaborative Study of Neoplasia and Steroid Contraceptives, [erratum appears in Am J Epidemiol 1999 Sep 1;150(5):546]." American Journal of Epidemiology 148(12): 1195-205. Nagata, C., N. Takatsuka, et al. (2000). "Association of diet with the onset of menopause in Japanese women." American Journal of Epidemiology 152(9): 863-7. Nakamura, T., K. Takio, et al. (1990). "Activin-binding protein from rat ovary is follistatin." Science 247(4944): 836-8. NCHS (2005). National Center for Health Statistics: Life Expectancy at Birth by Sex (Female) and Race (All Races): United States, Selected Years 1900- 2000. Data Warehouse on Trends in Health and Aging. Supported by The National Institutes on Aging at the National Institutes of Health. 2005. 208 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Nilsson, P., L. Moller, et al. (1997). "Social and biological predictors of early menopause: a model for premature aging." Journal of Internal Medicine 242(4): 299-305. Peccei, J. S. (1995). "A hypothesis for the origin and evolution of menopause." Maturitas 21(2): 83-9. Pike, M. C., L. N. Kolonel, et al. (2002). "Breast cancer in a multiethnic cohort in Hawaii and Los Angeles: risk factor-adjusted incidence in Japanese equals and in Hawaiians exceeds that in whites." Cancer Epidemiol Biomarkers Prev 11(9): 795-800. Pike, M. C., M. D. Krailo, et al. (1983). "'Hormonal' risk factors, 'breast tissue age' and the age-incidence of breast cancer." Nature 303(5920): 767-70. Qin, Z. S., T. Niu, et al. (2002). "Partition-ligation-expectation-maximization algorithm for haplotype inference with single-nucleotide polymorphisms.[comment]." American Journal of Human Genetics 71(5): 1242-7. Reyes, F. I., J. S. Winter, et al. (1977). "Pituitary-ovarian relationships preceding the menopause. I. A cross-sectional study of serum follice-stimulating hormone, luteinizing hormone, prolactin, estradiol, and progesterone levels." American Journal of Obstetrics & Gynecology 129(5): 557-64. Richards, J. S. (1980). "Maturation of ovarian follicles: actions and interactions of pituitary and ovarian hormones on follicular cell differentiation." Physiological Reviews 60(1): 51-89. Richards, J. S., D. L. Russell, et al. (2002). "Novel Signaling Pathways That Control Ovarian Follicular Development, Ovulation, and Luteinization." Recent Prog Horm Res 57(1): 195-220. Richardson, S. J., V. Senikas, et al. (1987). "Follicular depletion during the menopausal transition: evidence for accelerated loss and ultimate exhaustion." Journal of Clinical Endocrinology & Metabolism 65(6): 1231-7. Robertson, D. L. and D. M. Robertson (1987). "The isolation of polypeptides with FSH suppressing activity from bovine follicular fluid which are structurally different to inhibin." Biochemical & Biophysical Research Communications 149(2): 362-8. 209 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Robertson, D. M., L. M. Foulds, et al. (1985). "Isolation of inhibin from bovine follicular fluid." Biochemical & Biophysical Research Communications 126(1): 220-6. Sandler, D. P., M. Bohlig, et al. (1999). "Active and passive smoking and the occurrence of natural menopause." Epidemiology 10(6): 771-3. Santoro, N. (2003). "Mechanisms of premature ovarian failure. [Review] [43 refs]." Annales d Endocrinologie. 64(2): 87-92. Seltzer, G. B., N. Schupf, et al. (2001). "A prospective study of menopause in women with Down's syndrome." Journal of Intellectual Disability Research 45(Pt 1): 1-7. Sherman, B. M., J. H. West, et al. (1976). "The menopausal transition: analysis of LH, FSH, estradiol, and progesterone concentrations during menstrual cycles of older women." Journal of Clinical Endocrinology & Metabolism 42(4): 629-36. Shimizu, H. (1996). The basic report on Takayama Study. Takayama, Gifu, Japan, Japan Department of Public Health, Gifu University School of Medicine. Shiromizu, K. and D. R. Mattison (1984). "The effect of intraovarian injection of benzo(a)pyrene on primordial oocyte number and ovarian aryl hydrocarbon [benzo(a)pyrene] hydroxylase activity." Toxicology & Applied Pharmacology 76(1): 18-25. Snieder, H., A. J. MacGregor, et al. (1998). "Genes control the cessation of a woman's reproductive life: a twin study of hysterectomy and age at menopause." J Clin Endocrinol Metab 83(61: 1875-80. Soules, M. R., D. E. Battaglia, et al. (1998). "Inhibin and reproductive aging in women." Maturitas 30(2): 193-204. Stanford, J. L., P. Hartge, et al. (1987). "Factors influencing the age at natural menopause." Journal of Chronic Diseases 40(11): 995-1002. Stram, D. O. (2004). "Tag SNP selection for association studies." Genetic Epidemiology 27(4): 365-74. Stram, D. O., C. A. Haiman, et al. (2003). "Choosing haplotype-tagging SNPS based on unphased genotype data using a preliminary sample of unrelated subjects with an example from the Multiethnic Cohort Study." Hum Hered 55(1): 27-36. 210 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Tamada, T. and H. Iwasaki (1995). "[Age at natural menopause in Japanese women]." Nippon Sanka Fuiinka Gakkai Zasshi - Acta Obstetrica et Gvnaecologica Japonica 47(9): 947-52. te Velde, E. R. and P. L. Pearson (2002). "The variability of female reproductive ageing." Human Reproduction Update 8121: 141-54. The North American Menopause Society (2002). Menopause Core Curriculum Study Guide. Cleveland, OH, The North American Menopause Society. Thomas, D., R. Xie, et al. (2004). "Two-Stage sampling designs for gene association studies." Genetic Epidemiology 27(4): 401-14. Tong, S., E. M. Wallace, et al. (2003). "Inhibins and activins: clinical advances in reproductive medicine." Clinical Endocrinology 58(2): 115-27. Torgerson, D. J., A. Avenell, et al. (1994). "Factors associated with onset of menopause in women aged 45-49." Maturitas 19(2): 83-92. Torgerson, D. J., R. E. Thomas, et al. (1997). "Alcohol consumption and age of maternal menopause are associated with menopause onset." Maturitas 26(1): 21-5. Treloar, S. A., K. A. Do, et al. (1998). "Genetic influences on the age at menopause." Lancet 352(9134): 1084-5. Trichopoulos, D., B. MacMahon, et al. (1972). "Menopause and breast cancer risk." JNatl Cancer Inst 48(3): 605-13. Ueno, N., N. Ling, et al. (1987). "Isolation and partial characterization of follistatin: a single-chain Mr 35,000 monomeric protein that inhibits the release of follicle-stimulating hormone." Proceedings of the National Academy of Sciences of the United States of America 84(23): 8282-6. Vale, W., J. Rivier, et al. (1986). "Purification and characterization of an FSH releasing protein from porcine ovarian follicular fluid." Nature 321(6072): 776-9. van Asselt, K. M., H. S. Kok, et al. (2003). "Factor V Leiden mutation accelerates the onset of natural menopause." Menopause 10(5): 477-81. van Asselt, K. M., H. S. Kok, et al. (2004). "Linkage analysis of extremely discordant and concordant sibling pairs identifies quantitative trait loci 211 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. influencing variation in human menopausal age." American Journal of Human Genetics 74(3): 444-53. van Noord, P. A., J. S. Dubas, et al. (1997). "Age at natural menopause in a population-based screening cohort: the role of menarche, fecundity, and lifestyle factors." Fertility & Sterility 68(1): 95-102. Weel, A. E., A. G. Uitterlinden, et al. (1999). "Estrogen receptor polymorphism predicts the onset of natural and surgical menopause." Journal of Clinical Endocrinology & Metabolism 84(9): 3146-50. Whelan, E. A., D. P. Sandler, et al. (1990). "Menstrual and reproductive characteristics and age at natural menopause." American Journal of Epidemiology 131(4): 625-32. WHO (1981). Research on the Menopause: Report of a WHO Scientific Group. Geneva, World Health Organization. Willett, W., M. J. Stampfer, et al. (1983). "Cigarette smoking, relative weight, and menopause." American Journal of Epidemiology 117(6): 651-8. Worda, C., K. Walch, et al. (2004). "The influence of Nos3 polymorphisms on age at menarche and natural menopause." Maturitas 49(2): 157-62. Zenzes, M. T. (2000). "Smoking and reproduction: gene damage to human gametes and embryos." Human Reproduction Update 6(2): 122-31. 212 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Insulin-like growth factor 1 genotype, phenotype and breast cancer risk, by racial/ethnic group

PDF

Genetic risk factors in breast cancer susceptibility: The multiethnic cohort

PDF

Breast cancer in the multiethnic cohort study: Genetic (prolactin pathway genes) and environmental (hormone therapy) factors

PDF

Family history, hormone replacement therapy and breast cancer risk on Hispanic and non-Hispanic women, The New Mexico Women's Health Study

PDF

Risk factors for diabetic retinopathy in Latinos: Los Angeles Latino eye study

PDF

Development and evaluation of standardized stroke outcome measures in a population of stroke patients in rural China

PDF

BRCA1 mutations and polymorphisms in African American women with a family history of breast cancer identified through high throughput sequencing

PDF

A case/parental/sibling control study of Ewing's sarcoma/peripheral primitive neuroectodermal tumor (pPNET)

PDF

A case-control study of passive smoking and bladder cancer risk in Los Angeles

PDF

Determinants of mammographic density in African-American, non-Hispanic white and Hispanic white women before and after the diagnosis with breast cancer

PDF

Identifying predictors of cytogenetic relapse and death for patients undergoing allogeneic blood or marrow transplantation for treatment of chronic myeloid leukemia in chronic phase: The City of...

PDF

Evaluation of the accuracy and reliability of self-reported breast, cervical, and ovarian cancer incidence in a large population-based cohort of native California twins

PDF

A prospective cohort study of morbidity and mortality among middle-aged men in Shanghai, China

PDF

Effect of hormone therapy on the progression of carotid-artery atherosclerosis in postmenopausal women with and without established coronary artery disease

PDF

Physical activity and adenomatous polyps: Measures of association and impact

PDF

An intervention and program evaluation to determine the effectiveness of public health reforms on primary prevention practices by chiropractic interns

PDF

A descriptive analysis of medication use by asthmatics in the Children's Health Study, 1993

PDF

Dietary fiber intake and atherosclerosis progression: The Los Angeles Atherosclerosis Study

PDF

A large-scale genetic association study of prostate cancer in a multi-ethnic population

PDF

Androgens and breast cancer

Asset Metadata

Creator Henderson, Katherine DeLellis (author)

Core Title Determinants of the age at natural menopause: The multiethnic cohort

School Graduate School

Degree Doctor of Philosophy

Degree Program Epidemiology

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag biology, biostatistics,health sciences, public health,OAI-PMH Harvest,sociology, ethnic and racial studies,women's studies

Language English

Contributor Digitized by ProQuest (provenance)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-400849

Unique identifier UC11340375

Identifier 3196812.pdf (filename),usctheses-c16-400849 (legacy record id)

Legacy Identifier 3196812.pdf

Dmrecord 400849

Document Type Dissertation

Rights Henderson, Katherine DeLellis

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA