Genetic Epidemiology of Breast Cancer and Associated Cancers in High-Risk Families. I. Segregation Analysis 1,2,3
Rodney C. P. Go, 4 Mary-Claire King, 5,6 Joan Bailey-Wilson, 7 Robert C. Elston,7 and Henry T. Lynch 8,9
ABSTRACT-Genetic and environmental hypotheses that might explain the patterns of occurrence of breast cancer and associated cancers in 18 large families at high risk of the disease were tested with the use of segregation analysis. For 16 pedigrees, results were consistent with the hypothesis that breast cancer has a genetic etiology. In 2 other families, breast cancer appeared more likely to have an environmental origin. Breast cancer susceptibility is best explained by hypotheses that pos- tulate autosomal dominant susceptibility alleles in 10 families with pri- marily premenopausal breast cancer and ovarian cancer, in 4 families with primarily postmenopausal breast cancer, and in 2 families with breast cancer, brain tumor, sarcoma, leukemia, and adrenocortical car- cinoma in children and young adults. In an accompanying paper, genetic susceptibility in the first 2 groups of families is further explored with the use of linkage analysis .- JNCI 1983; 71:455-461.
For more than a century, families with many cases of breast cancer have intrigued and disturbed clinicians. It is now well estabished that the presence of breast cancer in a woman’s mother or sister increases her risk of the disease (1). If more than one relative has had breast cancer, especially if the relatives were affected at young ages or in both breasts, a woman’s risk may be dramati- cally increased. Consistent epidemiologic evidence also indicates that breast cancer patients in high-risk families are more likely to have premenopausal, bilateral breast cancer than are breast cancer patients with no family history of the disease. In such high-risk families, other cancers may appear in association with breast cancer (2, 3). Pedigree analysis methods have been used to estimate susceptibility to breast and other cancers for members of one large family at very high risk of breast cancer (4).
However, none of the studies cited above has tested specific modes of genetic or cultural inheritance of breast cancer susceptibility, although they provided considera- ble evidence for the familial nature of breast cancer. In this paper, we test the hypothesis that single genes that increase breast cancer susceptibility may segregate ac- cording to Mendelian laws in some families. In the fol- lowing companion paper we test for linkage of a hypo- thetic gene that increases susceptibility to breast cancer by known polymorphic genetic markers.
METHODS
Subjects .- The present sample comprised 18 families at high risk of breast cancer. In the families, 122 women had breast cancer, 11 had ovarian cancer, and 9 had endometrial cancer (table 1). All the families are Cauca- sian; most now live in the midwestern United States. All were contacted between 1967 and 1975 and asked to participate in the breast cancer family resource at Creighton University, Omaha, Nebr. More than 80% of
breast and ovarian cancer cases were verified by pathol- ogy review. Medullary carcinoma was slightly more fre- quent among these familial breast cancer patients than in most series of patients. Unfortunately, histologic data were not available by family at the time of this analysis (5).
A difficult ascertainment problem was introduced by the varied manner in which families came to be part of this study. Several of the families were ascertained from hospital records when a breast cancer patient had at least 1 first-degree relative with breast cancer. Several other families volunteered for the study because individuals were concerned about the extensive history of breast cancer in their own or their spouses’ families. Each family in the present sample includes at least 3 first-degree relatives with breast cancer. Because it is impossible to define either the population from which the families were drawn or a uniform sampling scheme by which they were selected, we cannot use this sample of families to estimate either the proportion of breast cancer that may be genetically influenced or the frequency of a breast cancer susceptibility allele in the U.S. population (6). Instead, we have assumed that each family was in- cluded in the study because of a cluster of breast cancer cases, and we also assumed that families were not selected
ABBREVIATIONS USED: df=degree(s) of freedom; SEER=Surveillance, Epidemiology, and End Results; TNCS=Third National Cancer Sur- vey.
’ Received October 19, 1982; accepted April 12, 1983.
2 Supported by Public Health Service (PHS) grants CA-13556, CA- 28198, and CA-27632 from the National Cancer Institute; by PHS contracts CB-44003 and CB-84274 from the Division of Cancer Biol- ogy and Diagnosis of the National Cancer Institute; and by PHS grant GM-28356 from the National Institute of General Medical Sciences.
3 Research procedures were in accord with the ethical standards of the Committee for the Protection of Human Subjects, University of California, Berkeley, Calif.
4 Department of Public Health and Diabetes Research and Training Center, University of Alabama Medical Center, Birmingham, Ala. 35294.
5 Department of Biomedical and Environmental Health Sciences, School of Public Health, University of California, Earl Warren Hall, Berkeley, Calif. 94720.
6 Address reprint requests to Dr. King.
7 Department of Biometry, Louisiana State University Medical Cen- ter, New Orleans, La. 70112.
8 Department of Preventive Medicine and Public Health, Creighton University, Omaha, Nebr. 68178.
9 We thank J. Lynch, K. Maloney, L. Rankin, and E. J. Rie for family interviews and E. B. Kaplan, R. Leung, and R. Harris for programming and statistical assistance.
| Family code | No. of members in family | No. of women with cancer in: | Mean age at breast cancer diagnosis ± SD, yr | ||
|---|---|---|---|---|---|
| Breast | Ovary | Endo- metrium | |||
| Group I: Mean age at breast cancer diagnosis, 44-52 yr | |||||
| B016 | 16 | 6 | 0 | 0 | 45.2±7.4 |
| B021 | 129 | 9 | 1 | 0 | 46.3±10.5 |
| B052 | 124 | 6 | 0 | 0 | 44.7±13.8 |
| B073 | 26 | 3 | 0 | 0 | 49.3±15.0 |
| B085 | 206 | 3 | 6 | 4 | 48.3±9.5 |
| B093 | 208 | 6 | 0 | 0 | 52.0±7.8 |
| B095 | 212 | 8 | 0 | 0 | 45.0±8.3 |
| B102 | 187 | 7 | 2 | 0 | 49.2±11.0 |
| B103 | 270 | 19 | 0 | 0 | 44.9±13.4 |
| B110 | 244 | 10 | 1 | 2 | 48.6±13.4 |
| B113 | 66 | 5 | 0 | 0 | 49.8±8.8 |
| B126 | 50 | 6 | 1 | 0 | 50.4±16.4 |
| Group II: Mean age at breast cancer diagnosis, 54-68 yr | |||||
| B025 | 98 | 7 | 0 | 3 | 63.0±14.2 |
| B058 | 209 | 5 | 0 | 0 | 68.4±15.4 |
| B086 | 80 | 7 | 0 | 0 | 54.4±13.6 |
| B108 | 76 | 5 | 0 | 0 | 55.5±16.4 |
| Group III: Mean age at breast cancer diagnosis, 31-32 yr | |||||
| B083ª | 112 | 5 | 0 | 0 | 32.0±7.0 |
| B4736 | 21 | 5 | 0 | 0 | 31.3±8.4 |
“B083 also includes 6 relatives with brain tumors, 5 with sarcomas, 4 with leukemia, 2 with adrenocortical carcinoma, and 10 with cancers at other sites.
bB473 also includes 2 relatives with brain tumors, 1 with sarcoma, and 1 with choriocarcinoma.
because some clusters of cases were more likely to indi- cate genetic transmission than were others. We under- took the genetic studies with the understanding that the principal purpose of segregation analysis would be to determine the best model, whether genetic or environ- mental, for breast cancer susceptibility within each fam- ily. The conclusive demonstration of the existence of an allele that increases susceptibility to breast cancer in any of the families will rely on the detection of genetic linkage between the hypothetic allele and a marker locus (7).
Statistical methods .- To determine whether Mendelian segregation at a single locus can account for the observed clustering of cases in these pedigrees, we used the meth- ods developed by Elston and associates (8-10). The underlying assumptions for the analyses were a) that sufficient family data were appropriately collected, b) that the log10 of the age of breast cancer diagnosis is approximately normally distributed, c) that only females are susceptible to breast cancer, and d) that there is no assortative mating in the pedigrees. The hypothesis of Mendelian segregation was tested under a general model that assumes 3 types of individuals, AA, Aa, and aa, who may have different age distributions at the onset of breast cancer. Under the genetic hypothesis these 3 types are the three possible genotypes for a locus with two segre- gating alleles, but the model also allows for the possibility that differences in the age of onset distributions are
environmentally determined. These 3 types can be 2 types under a genetic hypothesis with dominance, i.e., AA and Aa may give rise to the same age of onset distributions.
Susceptibility, y, is defined as the probability that an individual will eventually contract the disease, in this case breast cancer. Thus the proportion y of the popu- lation is susceptible and the proportion I-y is completely insusceptible to breast cancer; the age of onset distribu- tion is not relevant to the insusceptible group. Given that a woman is susceptible to breast cancer, she has a relevant age of onset density function f(a), with the corresponding cumulative distribution function F(a) = f& f(x)dx, the probability that she develops breast cancer by age a. This probability may be very small for one of the types of individuals, even for ages beyond the normal life-span, in which case the probability that these individuals would develop breast cancer is very small even though y may be very large. For this reason y is not made dependent on the type of individual. However, y is allowed to be dependent on sex, and so its presence in the model makes it unnecessary to make the type-dependent parameters in the model sex-specific. In many of the analyses, two different disease states are allowed (e.g., breast cancer and other cancers), and separate sex-specific susceptibil- Ity parameters are then estimated for each.
In any pedigree, there are those individuals who have neither parent in the pedigree and those who have at least one parent in the pedigree. We assumed that indi- viduals with neither parent in the pedigree are of types AA, Aa, or aa with probabilities q2, 2q(1 - q), and (1 - q)2, respectively, where q is an unknown parameter equal to the parental gene frequency of allele A under the genetic hypothesis. Individuals who have at least one parent in the pedigree have their types determined by their parents’ types according to a model that depends on the following three transmission probabilities: TAA = P (an AA individual transmits A to offspring); TAa = P (an Aa individual transmits A to offspring); Taa = P (an aa individual transmits A to offspring). Then each offspring is of type AA, Aa, or aa with the probability that both, one, or neither parent transmits A. Thus under this model, Mendelian segregation of two alleles at an auto- somal locus corresponds to the null hypothesis that TAA = 1, TAa = 0.5, and Taa = 0. It is also possible to subsume X-chromosome linkage under the same model by making the transmission probabilities sex-dependent (11).
The special case where the distribution of offspring types is independent of parental types corresponds to an environmental null hypothesis: TAA = TAa = Taa. Under this hypothesis, the 3 offspring types occur with frequen- cies 72, 27(1 - 7), and (1 - 7)2, where 7 is the common value for the transmission probabilities. If it were possi- ble to allow for the mode of ascertainment of the families, we should expect ₸ to be equal to q. Since this is not possible, we allowed 7 and q to be different. The model thus allows for increased risk of disease in offspring and hence a certain amount of familial clustering without genetic transmission. The assumption of “Hardy-Wein- berg” equilibrium proportions under this environmental
hypothesis is no real restriction when we assume that there are only 2 phenotypically distinguishable types.
The likelihood expressions for this model with variable age of onset and incomplete penetrance are given in full by Elston and Yelverton (9) and Elston (12). Under this general model, we used the likelihood ratio criterion to test whether in a given pedigree or set of pedigrees the observed clustering of breast cancer is consistent with Mendelian segregation. To accomplish this, we first max- imized the likelihood of the pedigree over all unknown parameters but restricted the transmission probabilities to 1.0, 0.5, and 0; then the likelihood of the pedigree was maximized again, but this time it was maximized over the transmission probabilities as well as the other parameters. Twice the difference in the natural loga- rithms of the two likelihoods can then be compared to a chi-square distribution, with 3 df, to determine the sig- nificance of departure from Mendelian segregation. However, the theoretic justification for expecting a chi- square distribution depends on large sample theory; for this reason only the larger pedigrees were analyzed in- dividually, and the significance levels are only approxi- mations. A nonsignificant result may, of course, occur because there are insufficient data. Therefore, the sec- ond null hypothesis, i.e., the environmental hypothesis TAA = TAa = Taa, was also tested, maximizing the likelihood over all unknown parameters but with the restriction that the transmission probabilities are equal to each other. Twice the difference between the logarithm of the likelihood thus obtained and the overall maximum logarithm of the likelihood was compared to the chi- square distribution with 2 df. If the result is significant, we can reject the hypothesis of the absence of vertical transmission, i.e., that the distribution of the disease among children is independent of the disease status of their parents. If the result of the test of this environmen- tal hypothesis is not significant, it is less likely that the trait segregates according to Mendelian laws, even though it may not be possible to reject the genetic hypothesis. We used the same model and method of analysis as that used by Elston et al. [(13), model I] and Tanna et al. (14), except that we analyzed multigenera- tional pedigrees with the assumption that they were randomly sampled from some population.
Bucher (15) has derived the appropriate theoretic expressions needed for a rigorous likelihood analysis of pedigrees ascertained through the presence of 2 affected individuals. Not only are these expressions cumbersome to compute, but also the pedigrees analyzed here are not a random sample of all pedigrees in which two first- degree relatives have breast cancer. For this reason the analyses performed here are approximations.
RESULTS
Epidemiologic Considerations
The ages at diagnosis of breast cancer in these families were strikingly different from those in the Caucasian female population of the TNCS. Approximately 70% of
familial breast cancers, but only 40% of breast cancers in the larger population, are diagnosed before meno- pause. The median age of breast cancer diagnosis in the families in our study was 43 years, whereas the median age of breast cancer diagnosis from the TNCS was about 56 years (16). These 13 years of difference were far too great to be the result only of increased surveillance and early detection. Survival rates for our familial breast cancer patients diagnosed between 1950 and 1973 were as good or better than survival rates of U.S. Caucasian women of the same ages diagnosd over the same period and followed by the cancer SEER program (17). Familial patients diagnosed at age 45 years or older experience 5-year survival rates comparable to the SEER rates ad- justed for age and year of diagnosis. Among women diagnosed before age 45, 73% of familial patients sur- vived at least 5 years. The range of 5-year survival rates was 63-67% in the SEER population for this age group between 1950 and 1973. This slight difference may reflect earlier detection and therefore a greater fre- quency of localized tumors among young breast cancer patients in the high-risk families.
Age of onset of breast cancer appeared to cluster in the families. Analysis of variance of the ages of breast cancer diagnosis within and among families indicated that affected women from the same high-risk family are more likely to have similar ages at breast cancer diagnosis than are 2 patients from different high-risk families (F=2.02 with 21 and 86 df; P =. 012). It is possible, therefore, to use the ages at breast cancer diagnosis of the affected relatives of a high-risk young woman to predict her risk at various ages. However, there appears to be no significant tendency for right- or left-sided tumors to cluster in these families (18).
To determine whether breast cancer patients in the families differ from their cancer-free female relatives with respect to parity or age at first pregnancy, we matched each breast cancer patient with her unaffected sister closest in year of birth who had survived at least to age 35 (or at least as long as her affected sister, in the case of young women alive at present). If no unaffected sisters were available for matching, an unaffected first cousin was chosen by the same criteria. Twenty-two percent of the women in each group were nulliparous. The average number of live births per woman was 2.66 among the breast cancer cases and 2.50 among their matched sisters or first cousins (P =. 66). Among matched pairs in which both women had completed a pregnancy, the age at first pregnancy was slightly greater among breast cancer patients than among unaffected women (average difference of 1.1 yr, P =. 31).
Analysis of Single Large Families
Separate segregation analyses were performed for each of the 12 larger pedigrees. The environmental hypothesis and the following four genetic hypotheses were tested: 1) An autosomal dominant allele increases susceptibility specifically to breast cancer in women. 2) An autosomal dominant allele increases susceptibility to
breast cancer in women, or to other cancers in men and women. 3) An autosomal recessive allele increases sus- ceptibility to breast cancer in women. 4) An X-linked dominant allele increases susceptibility to breast cancer in women.
Chi-square values to test autosomal dominant and environmental hypotheses are presented in table 2. A statistically significant chi-square suggests that the hy- pothesis under consideration can be rejected. For some of the families in group I, it was possible to distinguish fairly clearly among hypotheses. For families B021 and B110, hypotheses based on susceptibility to breast cancer only were most readily distinguished. Autosomal domi- nant hypotheses were consistent with patterns of breast cancer in each of these families, but environmental hy- potheses fit considerably less well. If all cancers were considered in the model, genetic and environmental hypotheses could not be distinguished.
For families B052 and B113, genetic hypotheses fit poorly, whereas environmental hypotheses could not be rejected. For family B113, we could not reject the au- tosomal dominant hypothesis at P <. 05. However, the environmental hypothesis was also consistent with the data, and for the best-fitting autosomal dominant model, the estimated frequency of the putative susceptibility gene A was 0, the predicted median age of breast cancer onset among genetically susceptible (AA and Aa) women was 293 years, and the predicted median age of breast cancer onset among women who do not carry the sus- ceptibility allele (aa women) was 51.4 years. We con- clude, therefore, that cancer in families B052 and B113 is probably not of genetic origin. However, family B113,
the smallest family to be analyzed separately, comprises only 66 individuals.
If we considered only breast cancer in family B093, the environmental hypothesis appeared more likely than the autosomal dominant hypothesis. If we considered all cancers, neither hypothesis could be rejected. For fami- lies B085 and B095, the genetic hypothesis is more promising if only breast cancer was included, but cancers other than breast cancer are clearly not genetically influ- enced. For families B102 and B103, the autosomal dom- inant hypothesis is more likely to apply because the distinction between hypotheses is greater if all cancers were included.
In summary, in 7 families in group I-B021, B085, B093, B095, B102, B103, and B110-patterns of cancer occurrence were consistent with an autosomal dominant hypothesis, but not with an environmental hypothesis. Breast cancer appears to be due to genetic susceptibility in families B021, B085, B095, and B110, but not all cancers in 3 families are due to genetic susceptibility. In families B093, B102, and B103, all cancers may be influenced by genetic susceptibility (text-fig. 1).
Two families in group II were analyzed independently. For family B025 (98 members), no hypothesis could be rejected. For family B058, the hypothesis of genetic influence on all cancers was most likely, and the corre- sponding environmental hypothesis can be rejected. Text-figure 2 indicates the pattern of cancer in this family.
For family B083 in group III, genetic susceptibility to all cancers was consistent with the data, and the corre- sponding environmental hypothesis was rejected. The
| Susceptibility to breast cancer only | Susceptibility to all cancers | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Autosomal dominant | Environmental | Autosomal dominant | Environmental | ||||||||
| x2 ☒ | (df) | P-value | x ☒ | (df) | P-value | 2 x" ☒ | (df) | P-value | x2 ☒ | (df) | P-value |
| Group I | |||||||||||
| 0.33 | (3) | >.9 | 3.51 | (2) | .17 | 2.38 | (3) | >.5 | 3.14 | (2) | >.2 |
| 6.86 | (1-3)ª | .01 -. 08₺ | 1.75 | (0-2)ª | ≥.4ª | 8.20 | (3) | .04 | 0.36 | (2) | >.8 |
| 1.79 | (3) | >.6 | 2.93 | (2) | >.2 | 10.53 | (3) | .01 | 1.46 | (2) | >.4 |
| 5.99 | (3) | .11 | 2.12 | (2) | >.3 | 2.81 | (3) | >.4 | 3.01 | (2) | >.2 |
| 0.15 | (3) | >.9 | 3.82 | (2) | .15 | 7.76 | (3) | .05 | 1.28 | (2) | >.5 |
| 1.18 | (3) | >.7 | 2.25 | (2) | >.3 | 1.97 | (3) | >.5 | 3.28 | (2) | .19 |
| 4.51 | (2-3)ª | .11 -. 21b | 4.24 | (1-2)ª | .04 -. 12b | 0.33 | (2-3)ª | >.8 | 6.96 | (1) | .01 |
| 1.29 | (2-3)ª | >.5 | 4.02 | (1-2)ª | .04 -. 13৳ | 7.96 | (2-3)ª | .02 -. 056 | 8.01 | (1) | .005 |
| 4.65 | (2-3)ª | .10 -. 20b | 2.42 | (1-2)ª | .12 -. 30b | ||||||
| Group II | |||||||||||
| 1.07 | (3) | >.7 | 0.01 | (2) | 1.0 | 2.20 | (3) | >.5 | 0.14 | (2) | >.9 |
| 6.20 | (3) | .10 | 0.12 | (2) | >.9 | 0.23 | (3) | >.9 | 10.64 | (2) | .005 |
| Group III | |||||||||||
| 2.47 | (3) | >.4 | 1.03 | (2) | >.6 | 0.94 | (3) | >.8 | 14.82 | (2) | .001 |
“In those instances where the largest value of the likelihood occurs at a boundary value of one or more parameters, and hence not at a local maximum, twice the difference in log likelihoods is bounded, under the hypothesis considered, by two x2-distributions with the indicated df; (0-2) indicates that only an upper bound (x2 with 2 df) can be given.
b P-values corresponding to the bounding x2 distributions.
“No cancers other than breast cancer occurred in this family.
3
2
2
2
G
W
0
~
3
W
BIO3
0
a
Sk
Unk
&
S
3
Unk
Unk
5
0
V
C
~
#
G
2
2
3
2
3
2
3
2
3
5
pattern of cancer occurrence in this family, shown in text-figure 3, illustrates this result.
For every family, the likelihood of the pedigree under either a recessive gene hypothesis or an X-chromosome- linked dominant gene hypothesis was smaller than the likelihood under an autosomal dominant gene hypothe- sis, but in no pedigree was the difference large enough to reject either autosomal recessive or X-linked domi- nant hypotheses.
Analysis of Families Pooled by Tumor Associations and Ages of Breast Cancer Diagnosis
In view of the tumor associations reported by Lynch and associates (2, 3), we analyzed together families hav- ing the same associated tumors and having similar esti- mated mean ages of breast cancer onset and susceptibil- ity. The smaller families were grouped with larger fam- ilies whose tumor associations were similar. The same method of analysis was used as that for single large families.
Five families in group I-B021, B085, B102, B110, and B126-include cases of ovarian cancer as well as breast cancer. When ovarian and breast cancer were both considered expressions of the susceptibility gene and these families were analyzed together, there was no loss of fit to the autosomal dominant gene hypothesis (x2=4.17; P =. 25), but the environmental hypothesis ☒ (x2=11.49; P =. 003) fit significantly less well. Inclusion ☒
of cancers other than ovarian and breast cancers as part of the gene expresion, however, led to rejection of the genetic hypothesis (x2=15.26; P =. 002). Similarly, the genetic hypothesis was rejected if women with mastitis, mastectomies, and oophorectomies without diagnosis of cancer were considered affected. When both breast and ovarian cancers were included, the likelihood of the autosomal recessive hypothesis was almost identical to that of the autosomal dominant model. The X-chromo- some-linked dominant hypotheses fit the data less well, although not significantly so.
Two large families, B095 and B103, include relatives with cancers of the gastrointestinal tract. Inclusion of gastrointestinal cancers as a possible expression of ge- netic susceptibility led to a better fit of the environmental hypothesis and a poorer fit of the autosomal dominant hypothesis, suggesting that the gastrointestinal cancers in these families are more environmentally than geneti- cally caused. The X-linked dominant and autosomal recessive gene hypotheses did not fit the data as well as the autosomal dominant gene hypothesis. We can con- clude that the gastrointestinal cancers should not be considered to have the same etiology as the breast can- cers in these pedigrees.
When we analyzed the 4 families in group II-B025, B058, B086, and B108-together and considered breast and endometrial cancers to be expressions of the same susceptibility gene, all genetic hypotheses fit almost equally well, and the environmental hypothesis was re- jected (x2=6.49; P <. 05). The fact that the environmen- tal hypothesis was rejected in this analysis but was not rejected for family B025 in the initial analyses (table 2)
Unk
X
2
2
Unk
S
0
X
S
X
N
Pa
0
X
N
ـلى
A
~
M
K
0
Unk
B58
N
3
&
O
Thr
e
2
C+
+
CX
6
4
3
%
3
X
2
En
M
2
2
3
E
na
Leu Leu
na-
AC
S
7
DS
Leu
0
2
BT
BT
B83
Lx
2
Lung
Leu
05
Q
8
Pr
C
.eu
BI
5º
~
G
2
Sa BT AC
2
BT
2
BT
suggests that endometrial cancer in family B025 may be an expression of the breast cancer gene.
Two families, B083 and B473, were at high risk of sarcomas, brain tumors, leukemia, and adrenocortical carcinoma, in addition to breast cancer. Most of these cancers were diagnosed in very young individuals. When all sites were included as part of the cancer phenotype, the environmental hypothesis was rejected and the au- tosomal recessive and X-chromosome linked dominant hypotheses gave distinctly lower likelihoods than did the autosomal dominant hypothesis.
Table 3 indicates the estimates of the susceptibility and age of onset parameters, obtained as appropriately weighted average of the individual pedigree estimates. Table 3 also indicates the range of parameter estimates found over the individual pedigrees in groups I and II.
In group I pedigrees the modal age of onset was 48.2 years, whereas in group II pedigrees it was 64.3 years, under an autosomal dominant hypothesis. In group I families, breast and ovarian cancers are included as part of the “affected” phenotype, whereas in group II families, breast and endometrial cancers are included; this, how- ever, does not account for the difference in the ages of onset between the two sets of pedigrees. Susceptibility was incomplete in group I families, but there were vir- tually no sporadic cases. To estimate under a dominant hypothesis the proportion of AA and Aa women in these families who have breast or ovarian cancers by age 50, for example, we calculated z=(In 50-3.876)/ 0.283=0.127 and determined that the corresponding probability under a standard normal curve was .551; this was then multiplied by the susceptibility, 0.9, to arrive at a risk estimate of 50%. Although there is complete susceptibility in group II families, there is also a larger proportion of sporadic cases: About 16% of aa women
| Age, yr | Age-specific risk of breast cancer/1,000 women | Relative risk | |
|---|---|---|---|
| Genetically susceptible women pre- dicted by model | TNCS white women | ||
| 20-24 | 1.6 | 0.01 | 145 |
| 25-29 | 6.6 | 0.09 | 76 |
| 30-34 | 15 | 0.23 | 66 |
| 35-39 | 23 | 0.53 | 43 |
| 40-44 | 26 | 1.04 | 25 |
| 45-49 | 27 | 1.59 | 17 |
| 50-54 | 22 | 1.72 | 13 |
| 55-59 | 19 | 1.92 | 10 |
| 60-64 | 13 | 2.26 | 6 |
| 65-69 | 9.4 | 2.34 | 4 |
| 70-74 | 6.2 | 2.60 | 2.4 |
| 75-79 | 4.0 | 2.95 | 1.4 |
| 80-84 | 2.4 | 3.01 | 0.8 |
| ≥85 | 1.6 | 3.08 | 0.5 |
“For the 10 families in group I fitting an autosomal dominant model for inheritance of a gene that increases susceptibility to breast cancer, predicted age-specific risks of breast cancer for genetically susceptible women are compared to risks for U.S. white women in the TNCS. Lower risks of women in the TNCS would also apply to women in high-risk families who do not carry the susceptibility allele.
| Parameter | Pooled parameter estimates (range) under single-gene hypotheses | |||
|---|---|---|---|---|
| Group I families (excluding B052 and B113) | Group II families | |||
| Dominant hypothesis | Recessive hypothesis | Dominant hypothesis | Recessive hypothesis | |
| Susceptibility, Yª | 0.900 (0.607-1.000) | 0.833 (0.623-0.958) | 1.000 (1.000) | 1.000 (1.000) |
| Mean loge (age of onset) AA or Aa | ||||
| 3.876 (3.800-3.951) | 4.761 (4.626-4.909) | 4.163 (3.996-4.225) | 4.626 (4.468-4.796) | |
| aa | 5.179 (5.027-5.335) | 3.799 (3.723-3.874) | 4.636 (4.470-4.832) | 4.203 (4.036-4.265) |
| SD of loge (age of onset) | 0.283 (0.153-0.340) | 0.379 (0.249-0.436) | 0.259 (0.212-0.295) | 0.286 (0.239-0.323) |
“Only females are susceptible to breast cancer or ovarian cancer in group I families and to breast cancer or endometrial cancer in group II families.
had breast or endometrial cancer by age 80 under an autosomal dominant model. However, this estimate had a very large standard error, and it was not significantly different from 0.
For group I families whose pattern of breast cancer incidence was consistent with inheritance of an autoso- mal dominant susceptibility gene, the age-specific risks of breast cancer for genetically susceptible women were far higher than those for women in the same family who do not carry the susceptibility gene or for women in the general population (table 4). The increased risk was particularly great at younger ages. These risk estimates for susceptible women are based solely on the genetic model we have described, not on empirical results.
DISCUSSION
The results are consistent with the hypothesis that cancers in the sample families have a genetic etiology except for pedigrees B052 and B113. However, traits such as breast cancer, which have variable age of onset and hence incomplete penetrance, require very large samples to permit discrimination among the various ge- netic hypotheses. Even for the breast and ovarian cancer families, which comprise a total of over 800 individuals, autosomal dominant, autosomal recessive, and X-linked dominant modes of inheritance all fit the data. Because many relatives are young or male, the prevalence of breast cancer in most of these pedigrees is 7% or less and is therefore too small to allow discrimination among the different modes of inheritance, although it is suffi- cient at times for discrimination between the genetic and environmental hypotheses.
Results of the analysis of the breast and ovarian cancer families as a group indicate that some families are sus- ceptible to both breast and ovarian cancer, although the mechanism of transmission is equivocal. Families with breast and gastrointestinal cancer show evidence for single gene segregation for breast cancer only; gastroin- testinal cancers do not have the same etiology as breast cancers in these pedigrees. The postmenopausal breast cancer families also show evidence that a major gene exists for susceptibility to breast and endometrial can- cers.
These results, along with estimates of the mean loga- rithm age of cancer onset, suggest that the 16 pedigrees other than B052 and B113 can be classified into the 3 groups indicated in table 1. Group I consists of 10 families: B021, B085, B093, B095, B102, B103 (text- fig. 1), B110, B016, B073, and B126, with an interme- diate age of breast cancer onset and including all the pedigrees with ovarian cancers. Group II is composed of the pedigrees B025, B058 (text-fig. 2), B086, and B108, with a much later age of breast cancer onset and endo- metrial cancer in some families. In group III are the 2 pedigrees B083 (text-fig. 3) and B473 characterized by an excess of childhood cancers; breast cancer is not the
most important cancer in these families.
In a companion paper (7) we undertake linkage anal- yses on families in groups I and II to determine whether the putative gene for breast cancer susceptibility can be shown to cosegregate with a polymorphic marker gene. For this purpose we use the estimates of the susceptibility and age of onset parameters presented in table 3. The purpose of the linkage analysis is to confirm the genetic model, using the results of this paper to characterize the relationship between phenotype (age, sex, and disease status) and hypothetic genotype at the susceptibility lo- cus.
REFERENCES
(1) PETRAKIS NL, ERNSTER VL, KING MC. Breast cancer. In: Schot- tenfeld D, Fraumeni JF, eds. Cancer epidemiology and preven- tion. Philadelphia; Saunders, 1982:855-870.
(2) LYNCH HT, KRUSH AJ. Carcinoma of the breast and ovary in three families. Surg Gynecol Obstet 1971; 133:644-648.
(3) LYNCH HT, KRUSH AJ, GUIRGIS HA. Genetic factors in families with combined gastrointestinal and breast cancer. Am J Gas- troenterol 1973; 59:31-40.
(4) BISHOP DT, GARDNER EJ. Analysis of the genetic predisposition to cancer in individual pedigrees. In: Cancer incidence in defined populations. Banbury report 4. New York: Cold Spring Harbor Laboratory, 1980:389-406.
(5) MULCAHY GM, PLATT R. Pathologic aspects of familial carcinoma of breast. In: Lynch HT, ed. Genetics and breast cancer. New York: Van Nostrand, 1981:65-97.
(6) ELSTON RC, SOBEL E. Sampling considerations in the gathering and analysis of pedigree data. Am J Hum Genet 1978; 31:62- 69.
(7) KING MC, Go RC, ELSTON RC, LYNCH HT. Genetic epidemiology of breast cancer and associated cancers in high risk families. II. Linkage analysis. JNCI 1983; 71:463-467.
(8) ELSTON RC, STEWART J. A general model for the genetic analysis of pedigree data. Hum Hered 1971; 21:523-542.
(9) ELSTON RC, YELVERTON KC. General models for segregation analysis. Am J Hum Genet 1975; 27:31-45.
(10) ELSTON RC, RAO DC. Statistical modeling and analysis in human genetics. Annu Rev Biophys Bioeng 1978; 7:253-286.
(11) DEMENAIS FM, ELSTON RC. A general transmission probability model for pedigree data. Hum Hered 1981; 31:93-99.
(12) ELSTON RC. Segregation analysis. Adv Hum Genet 1981; 11:63- 120.
(13) ELSTON RC, NAMBOODIRI KK, SPENCE MA, RAINER JD. A genetic study of schizophrenia pedigrees. II. One-locus hypotheses. Neuropsychobiology 1978; 4:193-206.
(14) TANNA VL, GO RC, WINOKUR G, ELSTON RC. Possible linkage between T-haptoglobin (Hp) and depression spectrum disease. Neuropsychobiology 1979; 5:102-113.
(15) BUCHER KD. The genetics of manic depressive illness: A pedigree and linkage study. Ph.D. dissertation. Chapel Hill: University of North Carolina, 1977.
(16) CUTLER SJ, YOUNG JL, JR, eds. Third National Cancer Survey: Incidence data. Natl Cancer Inst Monogr 1975; 41:120.
(17) AXTELL LM, ASIRE AJ, MYERS MH. Cancer patient survival. Report No. 5. Washington, D.C .: U.S. Govt Print Off, 1976 [DHEW publication No. (NIH)77-992].
(18) KING MC, LYNCH HT, SELVIN S. Laterality of breast cancer in families. Am J Epidemiol 1979; 110:88-98.
(19) LYNCH HT, MULCAHY GM, HARRIS RE, GUIRGIS HA, LYNCH JF. Genetic and pathologic findings in a kindred with hereditary sarcoma, breast cancer, brain tumors, leukemia, lung, laryn- geal, and adrenal cortical carcinoma. Cancer 1978; 41:2055- 2064.