14
Genetic Epidemiology 1 :7-20 (1984) Genetic Epidemiology of Breast Cancer: Segregation Analysis of 200 Danish Pedigrees Wick R. Williams and David E. Anderson Section of Human Genetics, The University of Texas M. D. Anderson Hospital and Tumor Institute, Houston An investigation of the genetic epidemiology of breast cancer involving complex segregation analysis of 200 breast cancer pedigrees of Danish extraction is pre- sented. The observed distribution of breast cancer is compatible with transmission of an autosomai-dominant gene with no evidence for residual family resemblance. The gene frequency of the abnormal allele is 0.00756, and the displacement between the homozygous genotype means is 1.695. The gene frequency accounts for a significant proportion of breast cancer in young women, whereas by an advanced age a majority (87%) of affected women are phenocopies. Genetic modeling of other breast cancer families and results of linkage studies are reviewed. Key words: breast cancer, segregation analysis, mixed model INTRODUCTION The familial occurrence of breast cancer has a long history. Roman physicians were aware of it in AD 100, pedigrees of breast cancer-prone families have been reported for over a century, and increased risks stemming from a family history of the disease have been recognized for the past 50 years [Lynch, 1981al. In spite of this long history and an impressive number of studies devoted to the subject, the etiologic significance of familial breast cancer is still being debated. Some contend that familial clustering represents a chance phenomenon because breast cancer is common in the general population. Others contend that familial clustering is the consequence of relatives sharing common customs, dietary habits, or exposure to a common environ- mental hazard. Still others contend that it is the consequence of an inherited suscepti- bility to the disease. Early family studies provided evidence of two- to fourfold higher risks for the disease among relatives of patients than among controls, and that these risks applied Received for publication October 5, 1983; revision received December 21, 1983. Address reprint requests to Wick R. Williams, PhD, Section of Human Genetics, M.D. Anderson Hospital, 6723 Bertner Avenue, Houston, TX 77030. 0 1984 Alan R. Liss, Inc.

Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

Embed Size (px)

Citation preview

Page 1: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

Genetic Epidemiology 1 :7-20 (1984)

Genetic Epidemiology of Breast Cancer: Segregation Analysis of 200 Danish Pedigrees

Wick R. Williams and David E. Anderson

Section of Human Genetics, The University of Texas M. D. Anderson Hospital and Tumor Institute, Houston

An investigation of the genetic epidemiology of breast cancer involving complex segregation analysis of 200 breast cancer pedigrees of Danish extraction is pre- sented. The observed distribution of breast cancer is compatible with transmission of an autosomai-dominant gene with no evidence for residual family resemblance. The gene frequency of the abnormal allele is 0.00756, and the displacement between the homozygous genotype means is 1.695. The gene frequency accounts for a significant proportion of breast cancer in young women, whereas by an advanced age a majority (87%) of affected women are phenocopies. Genetic modeling of other breast cancer families and results of linkage studies are reviewed.

Key words: breast cancer, segregation analysis, mixed model

INTRODUCTION

The familial occurrence of breast cancer has a long history. Roman physicians were aware of it in AD 100, pedigrees of breast cancer-prone families have been reported for over a century, and increased risks stemming from a family history of the disease have been recognized for the past 50 years [Lynch, 1981al. In spite of this long history and an impressive number of studies devoted to the subject, the etiologic significance of familial breast cancer is still being debated. Some contend that familial clustering represents a chance phenomenon because breast cancer is common in the general population. Others contend that familial clustering is the consequence of relatives sharing common customs, dietary habits, or exposure to a common environ- mental hazard. Still others contend that it is the consequence of an inherited suscepti- bility to the disease.

Early family studies provided evidence of two- to fourfold higher risks for the disease among relatives of patients than among controls, and that these risks applied

Received for publication October 5 , 1983; revision received December 21, 1983.

Address reprint requests to Wick R. Williams, PhD, Section of Human Genetics, M.D. Anderson Hospital, 6723 Bertner Avenue, Houston, TX 77030.

0 1984 Alan R. Liss, Inc.

Page 2: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

8 Williams and Anderson

only to breast cancer and not to cancer in general [see reviews by Clemmesen, 1965; Post, 19661. Although the early studies were criticized on the grounds that the study and control groups were not comparable or the analytical procedures were inadequate, or both, later studies such as those of Woolf [1955], V.E. Anderson et a1 [1958], Tokuhata [1969], and Macklin [1959], using improved sampling schemes and more critical analytic techniques, revealed risks similar to those found in early studies and additional evidence of a site-specific effect.

The environmental hypothesis has support from evidence indicating the two- to fourfold higher risks associated with a positive family history of breast cancer are not very different from those associated with environmentally and physiologically me- diated effects, such as parity, socioeconomic class, age at first delivery, age at menarche, or age at menopause [Kelsey and Hildreth, 19831. Second, findings emanating from population comparisons of migrants to native-born women also point to an environmental basis, since the breast cancer rates in migrants approach the rates of the locale or country to which the women migrate [Petrakis et al, 19821. One possibility not generally considered by the environmentalists, however, is that genetic factors could determine the response of the host to specific geographical carcinogens absent in the original country but present in the new locale, because for any geneti- cally determined tumor the tumor itself is not inherited; only the predisposition is inherited and some further step(s) mediated through the environment are required for tumor development [Knudson et al, 19731.

An assumption implicit in the early family studies was that breast cancer was a single disease; ie, all patients in a study regardless of their pathologic diagnosis or clinical characteristics were assumed to have the same disease. This misconception still occurs in studies of the etiology of breast cancer. That breast cancer is not a single disease has long been demonstrated by its different clinical and pathologic types (eg , lobular, intraductal, Paget’s, inflammatory, tubular carcinoma, etc), and their different natural histories [Haagensen, 19711. The concept of a heterogeneous disease is now being used to explain the variability in response to different therapeutic modalities [Canellos et a!, 19821. If the disease is, in fact, heterogeneous, and not homogeneous as assumed in the early studies, then this erroneous assumption has the effect of diluting or obscuring evidence of a genetic effect.

A family study of breast cancer that one of us (D.E.A.) initiated in 1969 [Anderson, 19711 was used to demonstrate heterogeneity in familial risk. The study differed from previous family studies in that it was based on breast cancer patients who had family histories of the disease. Rather than uniform risks, variable risks were found for first-degree relatives according to whether the pedigree was classified by age at diagnosis in the proband (premenopausal vs postmenopausal), laterality of the disease in the proband (unilateral vs bilateral), and type of family history (whether the secondary proband was a mother, a sister, or a second-degree relative of the initial proband). Premenopausal onset of the disease was associated with a threefold higher risk of breast cancer in first-degree relatives than in similar-aged controls, and bilateral breast cancer was associated with a fivefold increase in risk among relatives. When the primary and secondary probands were first-degree relatives (mother- daughter or sister-sister) with premenopausal andfor bilateral disease, the lifetime risk to other sisters and daughters approached 50%, which is highly suggestive of a dominantly inherited form of breast cancer. In contrast, probabilities associated with unilateral and/or postmenopausal onset were not far removed from the 7 % for women

Page 3: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

Genetic Epidemiology of Breast Cancer 9

in the general population [Anderson, 19821. The finding that premenopausal and/or bilateral disease is associated with a family history of breast cancer has now been confirmed by a number of workers [see review by Anderson, 1982, as well as Brinton et al, 1979, 1982; Bain et al, 1980; Lubin et al, 19821. Tulinius et al [1982] and Adami et a1 [1980, 19811 have also reported higher but nonsignificant risks for pre- menopausal patients than for postmenopausal patients. In addition, several entities involving premenopausal and bilateral breast cancer and segregation ratios approach- ing 50% have been described [Lynch, 1981b1.

Recent developments in genetic epidemiology offer a new approach for resolu- tion of genetic and environmental causes of familial aggregation of a trait. The primary purpose of complex segregation analysis is discrimination of major loci in the presence of other sources of familial correlation. Here we apply complex segre- gation analysis under a mixed model (major locus + multifactorial component) to investigate the cause of family resemblance of breast cancer in a sample of 200 pedigrees of Danish extraction by 1) testing whether the observed distribution of breast cancer is the consequence of a major locus, multifactorial inheritance, etc; 2) testing for etiological heterogeneity; and 3) determining estimates of parameters of the model of inheritance that characterize the observed distribution of breast cancer in the general population.

MATERIALS

The pedigree material upon which the present investigation is based was col- lected by Oluf Jacobsen and published in his monograph, “Heredity in Breast Cancer” [Jacobsen, 19461. Probands were identified from a series of over 300 breast cancer patients who were listed in the files of the Danish Cancer Registry, a population- based resource, living in the area of Greater Copenhagen, and treated at Copenhagen hospitals. Of these patients, 262 were contacted between October 1942 and November 1943. Information was not collected on 62 patients for the following reasons: born out of wedlock (22); had little knowledge of their relatives (29); refused to participate (6); and provided too little information (5) . The remaining sample of 200 probands consisted of 197 females and three males.

The pedigrees were developed by obtaining information about a defined set of relatives, namely the probands’ parents, siblings, children, aunts, uncles, and grand- parents (Fig. 1). Information from the proband was supplemented by inquiries to other relatives and by keepers of census and parish registers. Diagnoses of cancer were confirmed by inquiries to physicians, provincial hospitals, and parish register offices; examination of cancer records and necropsy journals of hospitals in Copen- hagen; and examination of death certificates. The frequency of reported cancers that could be confirmed by some type of medical documentation was high, 95 % . Prior to 1920, no ordered system of records existed in the rural areas of Denmark; thus, a diagnosis of cancer was accepted for a person who died prior to 1920 only when concordant information was provided by at least two of his or her relatives.

METHODS

Pedigrees were analyzed using the pointer strategy which was developed by Lalouel and Morton [1981] to provide a realistic approach to segregation analysis of

Page 4: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

10 Williams and Anderson

multigenerational pedigrees under the mixed model, with provision for ascertainment and the manner in which the pedigrees were extended. In this approach, pedigrees are partitioned into their component nuclear families and, possibly, an outside relative who led to ascertainment of the family, for analysis. This outside relative is said to be pointing to the family and consequently has been referred to as a “pointer.” Given the fixed sampling rule used by Jacobsen [1946] in developing the pedigrees, the pointer method was implemented as follows.

As seen in Figure 1, each pedigree could be partitioned into four different sibships, as distinguished by relationship to the proband and, consequently, the type of ascertainment correction necessary. Here we will refer to sibships wherein the proband is a child as group 1 ; group 2 will refer to sibships wherein the proband is a parent; and groups 3 and 4 will refer to the proband’s father’s and mother’s sibships, respectively. An ascertainment correction has previously been specified for families in groups 1 and 2, which represent examples of incomplete and complete selection, respectively [Morton and MacLean, 19741. Families in groups 3 and 4 represent a two-stage level of sampling. They were not ascertained directly through the proband (as a parent or a child), but indirectly through a line of descent from the proband, who is therefore considered to be a pointer to these families. Since Jacobsen [1946] sampled relatives in these two sibships regardless of whether any individual was affected, they may be treated as examples of complete selection with the additional provision that a correction be made for the original sampling event, selection through a pointer. Details of this ascertainment correction are given in Lalouel and Morton [1981]. At present, when the relationship of the pointer to the family is as a child, an approximation has been introduced by treating the pointer to the parents through an unknown child.

Knowledge of the ascertainment probability, 7r, representing the probability that an affected person is a proband, was required to provide an ascertainment correction for families under incomplete selection, group 1. In view of the restrictive sampling frame and the relatively high frequency of breast cancer, T must be quite small, corresponding to single selection. Further support for this comes from the fact that no pedigree contained more than one proband. Therefore, for the present analysis this probability was taken as 0.01, the precise value chosen having no notable effect upon the resulting analysis.

THE MODEL

Segregation analysis was carried out under a unified version of the mixed model of Morton and MacLean [ 19741, incorporating modifications introduced by Lalouel and Morton [ 19811 and the transmission frequencies of Elston and Stewart [ 19711.

As shown in Figure 2, the mixed model assumes an underlying liability scale to which a major locus, multifactorial component, and random environment contribute independently. The major locus has two alleles, and genotype frequencies are distrib- uted in Hardy-Weinberg proportions. The frequency of the abnormal allele associated with breast cancer is denoted by q. The distance between the two homozygous genotype class means is represented by t, referred to as displacement. The position of the heterozygote genotype mean relative to the means of the two homozygous genotype means is represented by d, the degree of dominance. When the heterozygote mean is near the mean of the lower homozygote, d = 0, the abnormal gene is

Page 5: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

Genetic Epidemiology of Breast Cancer 11

21 **2 f f *% Fig. 1. unknown phenotype).

Partition of a pedigree for analysis by the POINTER method (u refers to an individual of

bAJt 4 Fig. 2. The mixed model of inheritance.

recessive; when near the mean of the higher homozygote, d = 1, the abnormal gene is dominant; and when it is in the middle, d = 0.5, the effect of the abnormal allele is additive. Variation around each major genotype mean is assumed to be normally distributed, with common variance, C + E, where C is the variance due to multifac- torial transmissible effects, attributable in theory to a large number of genetic and environmental effects acting additively which are transmitted from parents to their children. E is the residual environmental variance component reflecting the proportion of the phenotypic variance (V) that cannot be explained by the major locus or by the multifactorial contribution. Multifactorial transmission is defined through H, the heritability, which reflects the proportion of the total phenotypic variance due to multifactorial effects: H = C/V.

Affection is defined by a threshold (2) on the liability scale, which is determined from the morbid risk to affection. Variation in morbid risk is treated by a shift in the liability scale. For breast cancer the morbid risk varies with age and sex; accordingly, we defined ten liability classes by partitioning the overall morbid risk into distinct

Page 6: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

12 Williams and Anderson

age- and sex-specific classes, as shown in Table I. For each class the corresponding morbid risk represents the sex-specific cumulative incidence from birth to the mid- point of the respective age interval, as determined from age- and sex-specific inci- dences reported by Clemmesen [1976]. Assignment of the proper liability class to each individual according to age at last observation and sex provides a correction for variation in morbid risk for a dichotomous phenotype, which, unlike a quantitative measurement, is not amenable to covariance adjustment.

Parameters of the model were estimated by maximizing the likelihood of the phenotypes of the sibship conditional upon the phenotypes of the parents (groups 1 and 2) or the phenotypes of the parents and the pointer (groups 3 and 4). To test hypotheses, minus twice the log likelihood (-2 In L), as calculated under a full model, is subtracted from -2 In L obtained when one or more of the parameters have been held constant. The difference is distributed asymptotically as a chi-square with n-k degrees of freedom, where n refers to the number of parameters being fit in the full model and k to the number being fit in the reduced model.

Complex segregation analysis was performed on the pedigree material using the computer program POINTER, written by J.-M. Lalouel and S. Yee [Morton et al, 19831.

RESULTS

The distribution of families by mating type and mode of ascertainment is shown in Table 11. It is evident how complete Jacobsen’s sampling efforts were. Of the expected 200 families in each group, only group 2 is significantly less (128), a deficit attributable to the young average age of these females and their low frequency of marriage and having children.

The results of complex segregation analysis as applied to the entire sample of pedigrees are shown in Table 111. In fitting the general model (major locus + multifactorial component) the multifactorial component went to zero, implying that a major locus alone was sufficient to account for the observed distribution of breast cancer in these pedigrees. In comparison with the general solution, a model not providing for family resemblance (q = H = 0) can be rigorously excluded ( x 2 4 df = 44.7, P < 0.01). This model assumes that familial occurrence of breast cancer is attributable to chance alone, and it is referred to as a sporadic model. Likewise, the distribution of breast cancer is not compatible with a multifactorial model of disease transmission (x23 df = 10.2; P < 0.02). Of the various models that incorporate a major locus (q > 0), a recessive model (d = 0.0) does not explain the observed segregation pattern as well as a dominant model (d = 1.0); an additive model (d = 0.5), however, yielded an equivalent likelihood to that of the dominant model, owing to the low frequency of the susceptibility allele (G’) and consequently the low probability that any of the affected individuals are homozygous (G’G’).

To confirm the major locus model from segregation analysis, an auxiliary test involving transmission frequencies and a test of heterogeneity based upon a partition of the material were performed.

The test procedure involving transmission frequencies was used to detect depar- tures from Mendelian segregation and, thereby, evidence against a major locus segregating for breast cancer in some of the pedigrees. The test consisted of allowing the heterozygote transmission frequency, TG,G:G defined as the probability that a

Page 7: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

Genetic Epidemiology of Breast Cancer 13

TABLE I. Liability Indicator for Segregation Analysis

Liability Cumulative class Sex Age incidence”

1 F 0-30 0 .000 I 3 2 F 30-40 0.00 162 3 F 40-50 0.00810 4 F 50-60 0.01951

0.03417 5 F 60-70 6 F 70-80 0.05441 7 F 80 + 0.06622’ 8 M 0-50 0.00002 9 M 50-70 0.0002 10 M 70 + 0 .OW5 1 ‘ aAs defined up to midpoint of class interval. bRepresented by the cumulative incidence up to age 80.

TABLE 11. Distribution of 719 Families by Mating Type and Ascertainment

Sibship Type of Mating type” Number of group selection N X N N X A N X U U X U children

1 Incomplete 178 22 960 2 Complete 119 9 283 3 Complete + 136 6 32 18 649

4 Complete + 161 4 21 13 753 pointer

pointer Total 594 41 53 31 2.645

aN = normal, A = affected, U = unknown status.

TABLE 111. Results of Segregation Analysis for the Total Sample

Model d t 4 H -2 In L+c

Sporadic, q = H = 0 1,128.21 Multifactorial, q = 0 0.285 1,093.67 Major locus, H = 0

Recessive, d = 0 1.999 0.13070 1,093.36 Intermediate, d = 0.5 (0.5) 3.311 0.00767 1,084.36 Dominant, d = 1 (1) 1.695 0.00756 1,083.47

Mixed (general model) lb 1.695 0.00756 Ob 1,083.47

aParameter fixed, not iterated. bParameter went to this bound.

parent of genotype G’G transmits the G gene to his or her offspring, to deviate from its Mendelian expectation of 0.5.

Beginning with the general solution, t, q, and 7G’G:G were iterated simultane- ously. Although TG’G:G decreased from 0.5 to 0.23, the change was not significant (x21 df = 2.91; P > 0.05).

Page 8: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

14 Williams and Anderson

The second test procedure was one of internal consistency and thereby a measure of confidence in the resultant dominant model. It consisted of partitioning the families according to some objective criteria (eg, mode of ascertainment, mating type, etc), solving for maximium likelihood estimates of parameters of various models within each subset, and comparing the sum of -2 In L as defined for the subsets with -2 In L as determined from analysis of the pooled material. Given the sampling scheme employed by Jacobsen [ 19461 to construct the pedigrees, we chose to partition the material according to the four groups of sibships that make up each pedigree (Table 11). Evidence of heterogeneity resulting from this partition would reflect incomplete sampling, an incorrect overall model, etiologic heterogeneity, improper ascertainment correction, etc. Owing to the young ages of offspring, group 2 con- tained very little information and it was therefore pooled with group 1 for the present test procedure.

Two models were fit to the three subsets of data: a multifactorial model (H alone), and a general model (iterating d, t, q, and H simultaneously). Fitting a multifactorial model to each subset yielded an estimate of the heritability, which may be considered an index of the degree of familial aggregation, whereas the comparison involving the general solution provides evidence as to the credibility of the dominant major locus model. Results are presented in Table IV.

Estimates of heritability in the three groups were not significantly different (x22 df = 0.13, P > 0.05), and, in fact, their similarity for group 3 and group 4 provides support for the contention that breast cancer is transmitted equally through paternal as well as maternal sides of a family, as previously documented [Macklin, 1959; V.E. Anderson, 1958; D.E. Anderson, 1974; Tulinius et al, 1980, 19821.

In fitting the general model to each subset of families, the multifactorial com- ponent and degree of dominance went to bounds of zero and one, respectively, and the resulting maximum likelihood estimates o f t and q were not significantly different (x24 df = 6.03; P > 0.05).

Results from both auxiliary tests thus provide support for the model obtained from segregation analysis of the total material, namely that a dominant major gene is responsible for familial aggregation of breast cancer in some of the pedigrees.

The characteristics of the major locus model are presented in Table V. Penetr- ance of the abnormal gene (G’) increases with age. By age 80 a female heterozygote has a 57% risk of developing breast cancer. Conversely, the frequency with which affection can be attributed to the abnormal allele decreases with age. Another way of stating this is that the proportion of phenocopies increases with age. Prior to age 30, 88% of affected females are heterozygous carriers, whereas by age 80 only 13 % of affected females are carriers of the abnormal gene; the remaining 87% are pheno- copies with a normal genotype. The results agree with epidemiologic evidence demonstrating a higher risk of breast cancer in relatives of premenopausal as com- pared to postmenopausal patients [Anderson, 19821.

DISCUSSION

Genetic modeling of other pedigrees presenting familial aggregation of breast cancer has been published. One of the most well known of these is Kindred 107, which was ascertained by Dr Eldon Gardner in 1947 through a student who had two aunts who had died of breast cancer [Gardner and Stephens, 1950; Stephens et al,

Page 9: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

Genetic Epidemiology of Breast Cancer 15

TABLE IV. Heterogeneity According to Type of Family

Sibs h i p Multifactorial model General model" group H -2 In L f c

1 + 2 0.271 656.98 1.288 0.014 655.14 3 0.261 116.39 2.308 0.002 107.81 4 0.304 320.17 1.909 0.008 314.49 Total 1,093.54 1,077.44

t -2 In L+c

aFor each sibship group, H and d went to bounds of zero and one, respectively.

TABLE V. Characteristics of the Major Locus for Each Liability Class

Liability class

1 2 3 4 5 6 7 8 9 10

P(affection/genotype) GG G'G or G'G'

0.0 0.01 0.0 0.07 0.0 0.20 0.02 0.33 0.03 0.43 0.05 0.52 0.06 0.57 0.0 0.00 0.0 0.01 0.0 0.03

P(genotype/affection)"

GG G'G

0.12 0.88 0.35 0.62 0.62 0.38 0.14 0.26 0.81 0.19 0.85 0.15 0.87 0.13 0.05 0.95 0.15 0.85 0.23 0.77

"P(G'G'/affection) = 0 owing to rarity of the G' allele

19581. In another, more recent review of this lundred there were 29 individuals with breast cancer, 16 with benign breast tumors, 19 with cancers of nonbreast nonurogen- ital sites, and five with benign tumors of nonbreast, nonurogenital sites out of 1,650 descendants of an English couple who immigrated to the United States in 1857 [Gardner, 19801. Gardner concluded that the distribution of breast cancer in Kindred 107 was consistent with transmission of an autosomal-dominant gene with incomplete penetrance. Hill et a1 [I9781 fit a model to Kindred 107 that incorporated an autoso- mal-dominant gene and sporadics, with allowance for different age- and sex-specific penetrances according to whether onset of breast cancer occurred at a premenopausal or a postmenopausal age. This was compared with a model that assumed the familial aggregation was due to chance alone. The likelihood for the genetic model was lo2' greater than the likelihood favoring the sporadic model when individuals with breast cancer were considered affected and greater when individuals with cancers of any site were considered affected.

Segregation analysis was also performed on Kindred 107 by Bishop and Gardner [I9801 using the computer programs POINTER [Lalouel and Morton, 19811 and PAP [Hasstedt and Cartwright, 19791. The POINTER analysis indicated the most likely model was that of an autosomal-dominant gene with no evidence for residual herita- bility, similar to results of our analysis of Danish pedigrees. The PAP analysis indicated a dominant gene model was much more likely than a sporadic model. Analytic results by Hill et a1 119781 and Bishop and Gardner [1980] support the original contention of Gardner. Owing to the high degree of selection of the family, it

Page 10: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

16 Williams and Anderson

is unlikely that an adequate correction was provided for ascertainment and the manner in which subsequent relatives were sampled; therefore, the numerical results should be accepted with caution.

Other than the present analysis of 200 Danish pedigrees, the largest sample of breast cancer pedigrees to which segregation analysis has been applied consists of 18 midwestern pedigrees that are part of the breast cancer family resource at Creighton University [Go et al, 19831. Pedigrees were selected for analysis because of the presence of breast cancer in three first-degree relatives, although the varied manner by which they were attracted to the resource provided an additional selection bias that could not be corrected for. The primary purpose of the analysis was to determine the most likely model of inheritance for each family, which was then used in a linkage study involving the same kindreds [King et al, 19831. Initially, the 12 largest pedigrees were analyzed individually. The information content of each pedigree was not suffi- cient to discriminate between the different models, with the exception of Pedigree B052, which favored an environmental model; subsequently, the 18 pedigrees were subdivided into three groups and segregation analysis was applied to each group. Pedigrees in group I were defined primarily by premenopausal onset of breast cancer and secondly by the type of nonbreast cancers observed in relatives. Of 12 pedigrees in this group, five contained one or more relatives who had ovarian cancer. The four pedigrees that constitute group I1 were defined by postmenopausal onset of breast cancer. The constellation of tumors in two pedigrees, B083 and B473, was suggestive of the Li-Fraumeni or SBLA syndrome [Bottomley and Condit, 1968; Li and Frau- meni, 1969; Lynch, 1981a1, a postulated autosomal-dominant condition characterized by the associated occurrence of sarcoma, breast cancer, brain tumors, leukemia, and adrenal and thyroid neoplasms within a pedigree; these two pedigrees were considered a third group.

The results of segregation analysis were equivocal and depended on which nonbreast cancers were included as part of the phenotype. For pedigrees in group I, an autosomal-dominant model provided the best explanation when ovarian cancer was also used to denote affection. The environmental model could be ruled out; however, an autosomal-recessive model provided a likelihood similar to that of the autosomal- dominant model. The four pedigrees in group I1 were compatible with all genetic models tested, but not the environmental model. In this analysis, endometrial cancer as well as breast cancer was used to denote affection. For the two SBLA pedigrees, the likelihood was highest for the autosomal-dominant model, but other models of inheritance could not be ruled out.

The autosomal-dominant models proposed by Go et a1 [1983] to explain the distribution of breast and ovarian cancers in group I and breast and endometrial cancers in group I1 are substantially different from the model defined by the present analysis. For the breast-ovarian pedigrees the age-specific penetrances of the suscep- tibility gene are 0.12 at age 35, 0.5 at age 50, and 0.87 at age 80; for the breast- endometrial pedigrees the probabilities at these respective ages are 0.01, 0.17, and 0.80. The model proposed for the breast-ovarian pedigrees does not provide for phenocopies, implying that the cancers in all affected individuals are due to the major gene. The model for the breast-endometrial pedigrees does allow for a small nonsig- nificant proportion of phenocopies (16%). In contrast, from our analysis (Table V) estimates of age-sex-specific penetrances are much lower, having a maximum of 0.57 at age 80, and most affected persons are phenocopies. Estimates of gene frequency were not published by Go et a1 [1983]; however, results of separate analyses of three

Page 11: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

Genetic Epidemiology of Breast Cancer 17

of the 18 pedigrees (B058, B103, and B083) by Elston et a1 [1981] were published, and gene frequencies were reported to range from 0.134 to 0.172.

The failure to provide for an ascertainment correction in the analyses by Go et al [1983] and Elston et a1 [1981] undoubtedly resulted in overestimation of the effect of the gene, resulting in age-specific estimates of penetrance that are too high, underestimation of the background frequency of phenocopies, and gene frequency estimates that are biologically unrealistic. The use of such a model to investigate linkage is questionable.

King et a1 [ 19831 reported a lod score of 1.95 between a dominant susceptibility allele for breast cancer and glutamate pyruvate transaminase (GPT) in seven families in group 1 with premenopausal onset and breast-ovarian cancer, based on the autoso- mal-dominant model proposed from analysis of the group I pedigrees by Go et a1 [ 19831. The seven families providing evidence of linkage were the same ones included in a previous report by King et a1 [1980]; however, three families included in the earlier investigation and two others were excluded from the latest analysis either because breast cancer segregation in the families did not fit a genetic model or because too few informative relatives were sampled. The latter reason is surprising because the excluded families were of the same size or even larger compared with their size in the initial analysis. Three of the excluded families also had negative lod scores in the initial analysis, so their exclusion in the latest analysis had the effect of increasing the total lod score. Inclusion of only positive lod scores represents a misuse of the sequential test for linkage [Morton, 19551.

Results of other linkage investigations involving breast cancer have recently been published. Cleton [1983] ascertained 16 families with at least three cases of breast cancer and submitted the pedigree and marker data to Dr King for linkage analysis. No significant positive lod scores were found for any marker, including GPT; however, none of the lod scores were reported. From a linkage investigation involving 11 Utah pedigrees results have been published that contradict the findings by King et a1 [1983]. A lod score of -5.29 was reported between GPT anu a dominant susceptibility allele for breast cancer, based upon a genetic model similar to that used in the analysis by King et a1 [1983]-ie, no allowance for phenocopies. Although the present evidence does not support the original finding of King et a1 [1980] that a major gene for breast cancer is located near GPT on chromosome 16 [McKusick, 19821, the possibility of genetic heterogeneity exists. Most of the pedi- grees used to investigate linkage by King et a1 [1983] had individuals with ovarian cancer, which may be a distinct clinicogenetic form of breast cancer [Lynch, 1981al; therefore, other pedigrees in which both breast and ovarian cancer are segregating require further investigation. As a final caveat, lod scores are sensitive to the genetic model that characterizes the test locus (age-sex penetrance, proportion of pheno- copies, gene frequency), and greater concern must be given to segregation analysis of a well-defined sample of pedigrees that have been extended in a known manner prior to linkage analysis.

An additional point of contention is that in the genetic model used by Go et a1 [1983] and Elston et a1 [1981], only females were considered susceptible to breast cancer. Although this probably had no effect on the analysis, since there were probably no males affected, there is evidence that this assumption may be incorrect.

An interesting finding by Jacobsen [ 19461 was the observation of females with breast cancer in kindreds selected through two out of three male probands with breast cancer. In addition, a male with breast cancer was observed in three of 197 pedigrees

Page 12: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

18 Williams and Anderson

ascertained through female probands. These observations prompted Jacobsen [ 19461 to extend his investigation to other males with the disease. Of an additional sample of five males with breast cancer that were ascertained, two knew little about their relatives and were uninformative. Of the remaining three, two had a first-degree relative with breast cancer; the affected relative was a sister to the proband in one pedigree and a daughter to the proband in the other. Thus, in Jacobsen’s pedigree material, four out of six male probands had a first- or second-degree female relative with breast cancer. From this evidence Jacobsen [ 19461 postulated that affected males in these pedigrees had inherited a severe liability to breast cancer. Unfortunately, since Jacobsen’s monograph was published little if any information has been reported to confirm this finding.

In an attempt to confirm and further clarify the underlying cause of a correlation in genetic liability to breast cancer in males and females, we have initiated a family study of breast and other types of cancer in relatives of male probands with breast cancer seen at The University of Texas M.D. Anderson Hospital and Tumor Institute at Houston. Owing to the rarity of the condition in males, it was necessary to extend our sampling frame from 1960 to the present to achieve a reasonable sample size from which to draw scientific inference. The present resource consists of approxi- mately 120 males with breast cancer (approximately 10 new cases are being added per year).

A preliminary review of the family histories of approximately 66 male patients revealed a positive family history of breast cancer in 22 [Williams et al, 19841. In contrast, a positive family history of breast cancer is reported in 15.5% of females with breast cancer at M.D. Anderson Hospital [Anderson, D.E., personal communi- cation] In 17 of the 22 families the disease occurred in at least one first-degree female relative, whereas four families had two or more first-degree relatives with breast cancer. The heritability of liability in relatives of male probands was estimated to be 0.37 (SE = 0.084), which is higher than the estimate of the heritability obtained from the 200 Danish pedigrees, which were selected primarily through females. Thus, initial findings from our resource are in agreement with those of Jacobsen [1946] in suggesting that males with breast cancer may have inherited a high liability to the disease, and, consequently, their relatives could represent a genetically high-risk group. It is of interest to note that two of the 29 documented cases of breast cancer in Kindred 107 occurred in males.

ACKNOWLEDGMENTS

The authors extend their gratitude to Professor D.C. Rao, Director of the Division of Biostatistics, Washington University Medical School, St. Louis, Missouri, for provision of computer facilities used in the analysis of the data. Partial support of the Harris computer (Dr Rao’s laboratory) was provided by grant NIGMS GM 28719. Support for W.R.W. was provided by training grant CA 09299 from the National Institutes of Health.

REFERENCES

Adami HO, Hansen J , Jung B, Rimsten A (1980): Farniliality in breast cancer: A case-control study in a Swedish population. Br J Cancer 42:71-77.

Page 13: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

Genetic Epidemiology of Breast Cancer 19

Adami HO, Hansen J, Jung B, Rimsten A (1981): Characteristics of familial breast cancer in Swcdcn: Absence of relation to age and unilateral versus bilateral disease. Cancer 48: 1688-1695.

Anderson DE (1971): Genetic considerations in breast cancer. In “Breast Cancer: Early and Late.” Chicago: Year Book Medical Publishers, pp 27-36.

Anderson DE (1974): Genetic study of breast cancer: Identification of a high risk group. Cancer 34: 1090-1097.

Anderson DE (1982): Familial predisposition. In Schottenfeld D, Fraumeni JF Jr (eds): “Cancer Epidemiology and Prevention.” Philadelphia: W.B. Saunders, pp 483-493.

Anderson VE, Goodman HO, Reed SC (1958): “Variables Related to Human Breast Cancer.” Minne- apolis: University of Minnesota Press, pp 1-172.

Bain C, Speizer FE, Rosner B, et al (1980): Family history of breast cancer as a risk indicator for the disease. Am J Epidemiol 111:301-308.

Bishop DT, Gardner EJ (1980): Analysis of the genetic predisposition to cancer in individual pedigrees. In Cairns J, Lyons JL, Skolnick M (eds): “Cancer Incidence in Defined Populations, Banbury Report 4.” Cold Spring Harbor, New York: Cold Spring Harbor Laboratory, pp 389-408.

Bottomley RH, Condit PT (1968): Cancer families. Cancer Bull 20:22-24. Brinton LA, Williams RR, Hoover RN, et al(l979): Breast cancer risk factors among screening program

participants. J Natl Cancer Inst 62:37-44. Brinton LA, Hoover R, Fraumeni JF Jr (1982): Interaction of familial and hormonal risk factors for

breast cancer. J Natl Cancer Inst 69:817-822. Canellos GP, Hellman S, Veronesi V (1982): The management of early breast cancer. N Engl J Med

306: 1430-1432. Cannon LA, Bishop DT, McLellan T, Skolnick MH (1983): Pedigree and linkage analysis of breast

cancer in eleven Utah kindreds: Non linkage of breast cancer to GPT. In “Program and Abstracts, The American Society of Human Genetics 34th Annual Meeting.” Am Soc Hum Genet 35:60A.

Clemmesen J (1965): Statistical studies in malignant neoplasms. I. Review and results. Acta Pathol Microbiol Scand 174: 11-19, 262-266.

Clemmesen J (1976): Cancer incidence in Denmark 1963-1967. In Waterhouse J, Muir C, Correa P, Powell J (eds): “Cancer Incidence in Five Continents,” Vol 111. Lyon: IARC Sci Pub 15, pp 292- 295.

Cleton FJ (1983): Genetic markers of cancer susceptibility. In “Proceedings, 13th International Cancer Congress, Part C.” New York: Alan R. Liss, pp 383-389.

Elston RC, Go RCP, King MC, Lynch HT (1981): A statistical model for study of familial breast cancer. In Lynch HT (ed): “Genetics and Breast Cancer.” New York: Van Nostrand Reinhold, pp 49- 64.

Elston RC, Stewart J (1971): A general model for the genetic analysis of pedigree data. Hum Hered

Gardner EJ (1980): Prevention and cure for hereditary cancers. In Cairns J, Lyons JL, Skolnick M (eds): “Cancer Incidence in Defined Populations, Banbury Report 4 . ” Cold Spring Harbor, New York: Cold Spring Harbor Laboratory, pp 365-378.

21:523-542.

Gardner EJ, Stephens FE (1950): Breast cancer in one family group. Am J Hum Genet 2:30-40. Go RCP, King MC, Bailey-Wilson J, et al(l983): Genetic epidemiology of breast and associated cancers

Haagensen CD: (1971): “Diseases of the Breast,” Ed 2. Philadelphia: W.B. Saunders, pp 503-616. Hasstedt SJ, Cartwright P (1979): “Pedigree Analysis Package.” Technical Report 13. Department of

Medicine, Biophysics and Computing, University of Utah. Hill JR, Carmelli D, Gardner EJ, Skolnick M (1978): Likelihood analysis of breast cancer predisposition

in a Mormon pedigree. In Morton NE, Chung CS (eds): “Genetic Epidemiology.” New York: Academic Press, pp 304-3 10.

in high risk families. I Segregation analysis. J Natl Cancer Inst 71:455-462.

Jacobsen 0 (1946): “Heredity in Breast Cancer.” London: H.K. Lewis, pp 1-306. Kelsey JL, Hildreth NG (1983): “Breast and Gynecologic Cancer Epidemiology.” Boca Raton, Florida:

King MC, Go RCP, Elston RC, et al (1980): Allele increasing susceptibility to human breast cancer may

King MC, Go RCP, Lynch HT, et al (1983): Genetic epidemiology of breast and associated cancers in

Knudson AG Jr, Strong LC, Anderson DE: (1973): Heredity and cancer in man. Prog Med Genet 9:113-

CRC Press, pp 5-70.

be linked to the glutamate-pyruvate transaminase locus. Science 208:406-408.

high-risk families. 11. Linkage analysis. J. Natl. Cancer Inst 71 :463-468.

Page 14: Genetic epidemiology of breast cancer: Segregation analysis of 200 Danish pedigrees

20 Williams and Anderson

158. Lalouel JM, Morton NE (1981): Complex segregation analysis with pointers. Hum Hered 31:312-321. Li FP, Fraumeni JF Jr (1969): Soft tissue sarcomas, breast cancer, and neoplasms: A familial syndrome?

Lubin JH, Burns PE, Blot WJ, et al: (1982): Risk factors for breast cancer in women in northern

Lynch HT (1981a): “Genetics and Breast Cancer.” New York: Van Nostrand Reinhold, pp 1-253. Lynch HT (1981b): Genetic heterogeneity and breast cancer: Variable tumor spectra. In Lynch HT (ed):

“Genetics and Breast Cancer.” New York: Van Nostrand Reinhold, pp 134-173. Macklin MT (1959): Comparison of the number of breast cancer deaths observed in relatives of breast

cancer patients and the number expected on the basis of mortality rates. J Natl Cancer Inst

McClellan T, Cannon LA, Bishop DT, Skolnick MH (1983): The cumulative LOD score between a breast cancer susceptibility locus and GPT is -3.86. In “Progress and Abstracts, International Human Gene Mapping Workshop VII,” p 190.

Ann Intern Med 71:747-752.

Alberta, Canada, as related to age at diagnosis. J Natl Cancer Inst 68:211-217.

221927-951,

McKusick VA (1982): The human gene map. Clin Genet 22:360-391. Morton NE (1955): Sequential tests for the detection of linkage. Am J Hum Genet 7:277-318. Morton NE, MacLean CJ (1974): Analysis of family resemblance. 111. Complex segregation of quanti-

Morton NE, Rao DC, Lalouel JM (1983): “Methods in Genetic Epidemiology.” New York: S. Karger,

Petrakis NL, Ernster VL, King MC: (1982): Breast. In Schottenfeld D, Fraumeni JF Jr (eds): “Cancer

Post RH: (1966): Breast cancer, lactation, and genetics. Eugen Q 13: 1-29. Stephens FE, Gardner EJ, Woolf CM (1958): A recheck of Kindred 107 which has shown a high

frequency of breast cancer. Cancer 11 :967-972. Tokuhata GK: (1969): Morbidity and mortality among offspring of breast cancer mothers. Am J

Epidemiol 89: 139-153. Tulinius H, Day NE, Sigvaldason H, et al (1980): A population-based study of familial aggregation of

breast cancer in Iceland, taking some other risk factors into account. In Gelboin HV, MacMahon B, Matsushima T, et al (4 s ) : “Genetic and Environmental Factors in Experimental and Human Cancer.” Tokyo: Japan Science Society Press, pp 303-312.

Tulinius H, Day NE, Bjarnason 0, et al (1982): Familial breast cancer in Iceland. Int J Cancer 29:365- 371.

Williams WR, Badzioch MD, Anderson DE, (1984): Epidemiology of male breast cancer. In Blumen- schein GR, Montague E, Ames F (eds): “Current Controversies in Breast Cancer.” Austin: University of Texas Press (in press).

Woolf CM (1955): Investigations on genetic aspects of carcinoma of the stomach and breast. Univ Cal Pub1 Public Health 2:265-350.

tative traits. Am J Hum Genet 26:489-503.

pp 1-261.

Epidemiology and Prevention.” Philadelphia: W.B. Saunders, pp 855-870.

Edited by D.C. Rao