We investigated the genetic models for familial breast cancer in a combined dataset consisting of both high-risk families and a population based series of breast cancer cases. We found the most parsimonious model to be one with a polygenic component with an equal variance in BRCA1/2 carriers and in non-carriers. This corresponds to a model in which several polymorphisms each act multiplicatively on risk (by for example, increasing the rate at which key mutational events will occur) to the same extent on carriers and non-carriers. A common recessive allele with moderate penetrance had the best fit among the major gene models fitted for ‘BRCA3’. However, the evidence for a major gene is weakened by the fact that the contribution of the ‘B’ families to the overall log-likelihood is always smaller under the major gene models for BRCA3 than their contribution under the sporadic model. On the other hand, the contribution of the ‘B’ families is larger under the polygenic models (with modifiers) than under the sporadic model.
Under the best fitting model (
), the frequency of BRCA1 mutations in the general population was estimated to be 0.051% (95% CI: 0.021–0.125%) which corresponds to about one in 974 individuals being a BRCA1 carrier. This is very similar to the estimate of Ford et al (1995)
of one in 800. Higher estimates were obtained by Whittemore et al (1997)
(0.14%) and Antoniou et al (2000)
(0.13%). These studies were, however, based on segregation analysis of ovarian cancer and may be inflated by not allowing for other familial effects. The estimate from the ABC dataset alone was 0.023% (Antoniou et al, 2001
), lower than the estimate from our combined analysis but included in the confidence interval. The mutation frequency for BRCA2 mutations was estimated to be 0.068% (95% CI: 0.033–0.141%). This corresponds to one in 734 individuals being a BRCA2 carrier in the general population. Again this is somewhat lower than 0.17% estimated in the ovarian cancer segregation analysis of Antoniou et al (2000)
(0.17%). The frequency from the ABC analysis alone was again somewhat lower, 0.041% (Antoniou et al, 2001
The fact that the observed number of BRCA1 mutations in the ‘B’ families was higher than predicted might suggest that we have slightly underestimated the BRCA1 allele frequency as a consequence of finding (by chance) too few BRCA1 mutations in the ABC study. Alternatively it may be that the sensitivity of mutation testing was lower in the much larger ABC study, although it is not clear why this should be true only for BRCA1.
Under our best fitting model the breast cancer risk for BRCA1 carriers, based on the average incidence over all modifying effects, was estimated to be 35.3% by age 70 years. This is moderately lower than the estimates derived by the BCLC studies of high risk families, which are in the range 71–85% (Easton et al, 1993
) and also slightly lower than the estimates derived from risks to relatives of mutation carriers in population based series of breast cancer patients (Struewing et al, 1997
; Hopper et al, 1999
). These differences are broadly consistent with our best model, since cases in high risk families will be expected to have a much larger polygenic component than average, while relatives of early onset breast cancer cases will be expected to have only a slightly higher polygenic component than average. The ovarian cancer risk for BRCA1 mutation carriers is estimated to be 26% by age 70 years. This is somewhat higher than the estimate of Struewing et al (1997)
but lower than the BCLC estimates (50.6–63.3%) (Easton et al, 1993
). It is also lower than our previous estimate based on ovarian cancer families (66%) (Antoniou et al, 2000
). The higher estimates from the last two studies are based on families that were selected on the basis of either ovarian cancer cases or number of breast and ovarian cancer cases, and could be the result of modifying genes or other familial factors increasing ovarian cancer risks. In principle, our model could be extended to account for such modifying factors.
The cumulative risk of breast cancer for BRCA2 mutation carriers was estimated to be 50.3% by age 70 years. Again this is lower than the BCLC estimates (Ford et al, 1998
) but more similar to the estimates of the population based study of Hopper et al (1999)
. The corresponding ovarian cancer risk by age 70 years was estimated to be 9.1%. This is lower than the first BCLC estimate (27%) (Ford et al, 1998
) and the estimate based on ovarian cancer families (31%) (Antoniou et al, 2000
). It is also somewhat lower than the estimate from a recent BCLC study of genotype-phenotype correlations (14%) (Thompson and Easton, 2001
) and the estimate of Struewing et al (1997)
based on carriers of Ashkenazi Jewish founder mutations.
We found that the breast cancer incidence rates in BRCA1 mutation carriers relative to the population incidence rates decrease sharply with age. This contrasts with the pattern in BRCA2 carriers, in whom the relative risk is approximately constant over age 40 years. Thus, BRCA2 mutations are estimated to confer a breast cancer risk which is similar to the population risk but 10–12 times greater. This difference in the pattern of incidence rates is mirrored in the very different histopathology of tumours in BRCA1 and BRCA2 carriers (Lakhani et al, 1998
), and reflects important mechanistic differences. A practical implication of these different age-incidence patterns, is that BRCA1 mutation carriers are more likely to develop breast cancer at a younger age than BRCA2 mutation carriers.
The polygenic model suggests that several genes, each having small but multiplicative effects on risk can account for the residual non-BRCA1/2 familial clustering of breast cancer. Genes other than BRCA1 and BRCA2 implicated in the aetiology of hereditary breast cancer include the TP53 gene (Malkin et al, 1990
), and the PTEN gene (Nelen et al, 1996
). Mutations in these genes, however, predispose to rare autosomal dominant syndromes and their contribution to familial aggregation is very small. Heterozygous carriers of the ataxia-talangiectasia gene ATM have been reported to be at increased risks of breast cancer (Swift et al, 1991
; Easton, 1994
; Olsen et al, 2001
) and there is also evidence that rare HRAS1 alleles may be associated with moderate risks for breast cancer (Krontiris et al, 1993
). It has been suggested that genes involved in steroid hormone metabolism and transport could act together defining a high-risk profile for breast cancer (Henderson and Feigelson, 2000
). Genes included in this pathway include the HSD17B1 gene, the CYP17 and CYP19 genes and the Estrogen Receptor (ER) gene (Henderson and Feigelson, 2000
). At present, however, there is no clear evidence that polymorphisms in these genes are associated with a significant risk (Dunning et al, 1998
). In a meta-analysis of all the studies of low risk polymorphisms, significant evidence was found for carriers of the (TTTA)10
Cyp19 polymorphism, the GSTP1 Ile105Val polymorphism and the TP53 Arg72Pro polymorphism (Dunning et al, 1998
) but the evidence was not conclusive.
Rebbeck et al (1999b)
found that BRCA1 carriers with long repeat lengths of the (AG)n
polymorphism at the Androgen Receptor gene may have earlier age at diagnosis of breast cancer. Similarly, the steroid hormone metabolism gene AIB1 has been reported to modify the breast cancer risk in BRCA1 mutation carriers (Rebbeck et al, 1999a
). Carriers of a variant AIB1 allele were found to have an age of diagnosis of breast cancer which is on average 3 years younger than non-carriers. However, neither of these results have been replicated.
The Hypergeometric Polygenic Model is equivalent to a fully additive polygenic trait, with no dominance or epistatic variance (Lange, 1997b
). In practice some degree of dominance may exist, given the slightly higher familial relative risks among sibs (Pharoah et al, 1997
). Among the major gene models fitted, the recessive model was the most parsimonious, and a recessive effect was also found by Cui et al (2001)
in BRCA1/2 negative families. However, some of the evidence for a recessive effect may be due to secular increase in incidence, which would give rise to a higher risk to siblings of probands, and to the protective effect of parity which slightly reduces the risk to mothers of probands (Antoniou et al, 2001
). It is also, of course, possible that some of the familial aggregation modelled here in terms of genetic susceptibility may be due to clustering of lifestyle risk factors (for example diet or reproductive factors) within families.
In conclusion, the present results suggest that a number of low penetrant genes may account for the familial clustering of breast cancer outside BRCA1 and BRCA2 families. The modifying effect on the breast cancer risks of BRCA1 and BRCA2 carriers may explain some of the differences between the risk estimates from population based studies and high risk families. The resulting model provides a framework for risk estimation to counsel women with a family history of breast cancer, allowing one to estimate carrier probabilities (separately for BRCA1 and BRCA2) and incidence rates in the same analysis. The model can also be extended straightforwardly to include risks of other cancers, second breast/ovarian cancers and tumour pathology. In principle it should also be possible to include lifestyle factors, although this would depend on assumptions about their effects in susceptible individuals.