A probability model for predicting BRCA1 and BRCA2 mutations in breast and breast-ovarian cancer families

Germline mutations in BRCA1 and BRCA2 genes predispose to hereditary breast and ovarian cancer. Our aim was to find associations between the clinical characteristics and positive mutation status in 148 breast cancer families in order to predict the probability of finding a BRCA mutation in a family. Several factors were associated with mutations in univariate analysis, whereas in multivariate analysis (logistic regression with backward selection) only the age of the youngest breast cancer patient and the number of ovarian cancer cases in a family were independent predictors of BRCA mutations. A logistic model was devised to estimate the probability for a family of harbouring a mutation in either BRCA1 or BRCA2. Altogether, 63 out of 148 families (43%) and 28 out of 29 (97%) mutation carrier families obtained probabilities over 10%. The mean probability was 55% for mutation-positive families and 11% for mutation-negative families. The models by Couch et al (1997) and Shattuck-Eidens et al (1997) previously designed for BRCA1 were also tested for their applicability to distinguish carrier families with mutations in either gene. The probability model should be a useful tool in genetic counselling and focusing the mutation analyses, and thus increasing also the cost-effectiveness of the genetic screening. © 2001 Cancer Research Campaign http://www.bjcancer.com


PATIENTS AND METHODS
The cohort studied consisted of 148 families with 3 or more 1st or 2nd degree relatives affected with breast or ovarian cancer. The families were identified by patient interviews, and full pedigrees were constructed with the confirmation of all genealogy data through the Finnish population registration as well as diagnostic data through hospital records and/or Finnish Cancer Registry as previously described (Vehmanen et al, 1997a,b;Eerola et al, 2000). Additionally, 295 breast cancer cases with one 1st degree relative affected with breast or ovarian cancer and identified in the patient cohorts described in Eerola et al (2000) were also studied. In the following, these are called small families. The family history of these cases was based on information reported by the index patient. All patients participating in the study signed an informed consent before the blood sample for the genetic analysis was taken. This study has been approved by the Ethical Committees of Departments of Obstetrics and Gynaecology, and Oncology, HUCH, and appropriate permissions were obtained from the Ministry of Social Affairs and Health in Finland.
The mutations identified by a complete mutation analysis of the whole coding sequences and exon/intron boundaries of the genes in 95 of these families have been previously reported (Vehmanen et al, 1997a,b). For 53 other families, all previously reported 18 Finnish BRCA1 and BRCA2 mutations (Vehmanen et al, 1997a,b;Huusko et al, 1998;Sarantaus et al, 2000), and one recently discovered new BRCA1 mutation (3264 delT) were analysed by allele-specific oligonucleotide (ASO) (Friedman et al, 1995) hybridization or restriction fragment length polymorphism (RFLP). The RFLP analyses were designed such that incomplete digestion would lead to a false positive hence minimizing the possibility of a false negative result. Sequences of the PCR primers and ASO probes, as well as the enzymes used for digestions are available upon request. Protein truncation test (PTT) (Hogervorst et al, 1995;Håkansson et al, 1997) of BRCA1 exon 11 and BRCA2 exons 10 and 11 was also used to search for new mutations in 36 families with an ovarian cancer case or a breast cancer patient diagnosed below 50 years. All positive mutation detection results were confirmed by direct sequencing using an ABI PRISM 310 Genetic Analyser and Dye Terminator Cycle Sequencing Ready Reaction Kit according to the manufacturer's instructions (PE Applied Biosystems, Foster City, CA, USA).
For 295 small breast cancer families ASO and RFLP analyses were used to screen all known Finnish mutations, and direct sequencing was used to confirm the positive screening results. In previous studies, 11 recurrent founder mutations have been found to account for vast majority (84%) of all Finnish BRCA1 and BRCA2 families (Vehmanen et al, 1997a,b;Huusko et al, 1998). Therefore, screening of the known mutations was used to evaluate the feasibility of screening of the BRCA1 and BRCA2 genes in these families.

Statistical analysis
Associations between specific familial characteristics (presented in Table 1) and the presence of a BRCA1 or BRCA2 germline mutation were studied by univariate and multivariate analyses. For univariate analysis, Mann-Whitney and Fisher's exact tests (SPSS 8.0 for Windows) were used. Variables that were predictive of a mutation in a univariate analysis were used in a multivariate analysis (stepwise backward logistic regression, 99%), and based on that a logistic probability model for harbouring a deleterious mutation was devised.
The models by Couch et al (1997) and Shattuck-Eidens et al (1997), previously designed for estimating mutation probability in the BRCA1 gene, were also tested in the 148 families and compared to the model developed here for their applicability to distinguish carrier families with mutations in either gene.

Mutations identified
A total of 29 germline mutations was found in 148 families (19.6%), 16 in BRCA1 (10.8%) and 13 in BRCA2 (8.8%). In addition to previously known Finnish mutations, two new protein truncating mutations were identified (BRCA1, 1806 C → T and BRCA2, 5797 G → T). Both of these mutations were subsequently found also in other study cohorts Sarantaus L, personal communication) making the total number of recurrent mutations in Finland now 13. Altogether, 24 (86%) of the mutation-positive patients carried one of the recurrent mutations, and 5 patients unique mutations not found in other families so far in Finland.

Factors associated with positive mutation status
Several factors were associated with the presence of germline BRCA1 or BRCA2 mutations in the univariate analysis (Table 1). In the multivariate analysis, only two variables were still significant: the number of ovarian cancer cases in a family (P < 0.00005) and the age at diagnosis of the youngest breast cancer patient (P = 0.0007). The presence of breast and ovarian cancer in the same patient was not significant in multivariate analysis, probably because it is closely associated with ovarian cancer cases overall. Bilateral breast cancer, another factor that has been correlated with a positive mutation status by for example Shattuck-Eidens et al (1997) and Ligtenberg et al (1999), was not significant in univariate analysis and, therefore, not included in further analysis.
Families carrying a mutation in either BRCA1 or BRCA2 were also analysed separately (data not shown). The results were similar for both genes except for the number of breast cancer patients that was associated with a BRCA2 mutation status in the univariate analysis. In the multivariate analysis the same variables were significant for both genes and, therefore, one common model could be used for distinguishing all mutation carriers. Early age of breast cancer onset as well as the presence of ovarian cancer in a family are thus highly characteristic for Finnish BRCA2 families also. It is of interest to note that only one of the BRCA2 mutations in this study was in the OCCR region where a higher risk of ovarian cancer, relative to breast cancer, has been suggested (Gayther et al, 1997;Ford et al, 1998).

Probability of identification of a mutation in the family
Based on the results from the multivariate analysis, a probability model for harbouring a deleterious mutation was devised, and can be written in the form of: and L can be calculated from the equation L = 2.87 + (-0.14) × V 1 + 2.11 × V 2 where 2.87 is a constant and -0.14 and 2.11 are the coefficients received from the regression analysis, V 1 is the age of the youngest breast cancer patient in a family, and V 2 is the number of ovarian cancer cases in a family. Among the 148 study families, 97% (28/29) of the mutation carrier families obtained a probability greater than an arbitrary cut off value of 10%. The mean probability was 55% for mutationpositive families and 11% for mutation-negative families. Altogether, out of 148 families 63 (43%) obtained probabilities over 10% and among these, 28 (44%) were mutation carrier families. Thus by using this model, mutation screening could be directed to a significantly smaller proportion of families.
Similar results were obtained also with the models of Shattuck-Eidens et al (1997) and Couch et al (1997) originally designed for BRCA1 (Table 2). Thus these models distinguish also BRCA2 mutation carrier families very efficiently. The one mutationpositive family missed in all 3 models has 3 affected breast cancer patients all diagnosed at later age. The proportion of mutations found is higher in the model developed in this study since it has been designed particularly for this study cohort, and the determination of sensitivity as well as specificity of this model requires analysis of a separate test population. The model here was also designed to estimate the carrier probability of a family with 3 or more affected cases, and therefore it could not be extrapolated to cases with a less profound family history.

Mutation frequencies in families with defined family history of cancer
All families classified by the family history of breast and ovarian cancer as well as age of breast cancer onset (below 40 years) are presented in Table 3. By analysing mutation-positive and -negative families, initially chosen by the criterion of at least 3 breast or ovarian cancer patients among 1st or 2nd degree relatives, we noted that mutation carrier families could be identified by a simple criterion of a breast cancer case diagnosed before the age of 40 or an ovarian cancer case in the family. Altogether, 80/148 (54% of all) families fulfilled this criterion, and among these, 28/29 (97%) of the mutations could be found. This simple criterion alone could thus be used as a rough estimation of a high likelihood of carrying a mutation in such families.
No mutations were found in 21 families with 4 or more cases of breast but no ovarian cancer or young breast cancer patient (diagnosis below 40 years). This is in agreement with our results from   1035 unselected breast cancer patients, where all 15 cases with heavy breast cancer family history were also mutation negative . Other, yet unknown susceptibility genes remain to be identified and may account for a large proportion of breast cancer families (Rebbeck et al, 1996;Serova et al, 1997;Vehmanen et al, 1997b;Ford et al, 1998;Kainu et al, 2000). In 295 breast cancer cases with one affected 1st degree relative only one mutation (BRCA2, 7708 C → T) was found giving the mutation frequency of 0.3%. In this family the index patient was diagnosed at the age of 37, and her mother had died of breast cancer at the age of 40. Ovarian cancer or a young breast cancer patient diagnosed under 40 years was present in 39 families, but among these only this one mutation was found (2.6%). This suggests that mutation screening in families with only 2 affected cases is not feasible in Finland. In contrast, Goelen et al (1999) reported that BRCA1/2 mutation testing can be done with reasonable efficiency in the Belgian population when there are 2 symptomatic family members. Prevalent founder mutations account for a large fraction of breast cancer families in Belgium (Peelen et al, 1997;Goelen et al, 1999), while BRCA1 and BRCA2 mutations are more rare in the Finnish population (Vehmanen et al, 1997a,b;Huusko et al, 1998). Also in studies of patients with early onset breast cancer, only a small proportion of familial risk of breast cancer has been attributed to these two genes, and the majority appears to be due to other genes (Peto et al, 1999).

CONCLUDING REMARKS
As the screening of both BRCA1 and BRCA2 is very laborious and expensive, and genetic testing may be emotionally very stressful for the families, the potential mutation carrier families should be recognized as efficiently as possible to avoid unnecessary analyses of non-carriers. Studies of breast cancer patients have indicated that it may be difficult to define mutation screening criteria among women with minimal or no family history (Malone et al, 1998). Furthermore, the carrier risks associated with the mutations may be highly variable, and population-based risk estimates have indicated much lower cancer risks than those obtained from multiplecase families and, therefore, lower predictive value of cancer for a positive mutation test result (Struewing et al, 1997;Fodor et al, 1998;Thorlacius et al, 1998;Warner et al, 1999). Accordingly, genetic screening would be of greatest benefit in families with high cancer risk, i.e. strong family history (Fodor et al, 1998), and a high probability of harbouring a BRCA1 or BRCA2 mutation. For this study, we chose families with a defined family history, and developed a model by which likelihood of carrying a BRCA1 or BRCA2 mutation can be estimated for each family separately. It should be a useful tool in genetic counselling and focusing the mutation analyses, and increasing thus the costeffectiveness of the genetic screening.