Analysis of cancer risk and BRCA1 and BRCA2 mutation prevalence in the kConFab familial breast cancer resource

Introduction The Kathleen Cuningham Foundation Consortium for Research into Familial Breast Cancer (kConFab) is a multidisciplinary, collaborative framework for the investigation of familial breast cancer. Based in Australia, the primary aim of kConFab is to facilitate high-quality research by amassing a large and comprehensive resource of epidemiological and clinical data with biospecimens from individuals at high risk of breast and/or ovarian cancer, and from their close relatives. Methods Epidemiological, family history and lifestyle data, as well as biospecimens, are collected from multiple-case breast cancer families ascertained through family cancer clinics in Australia and New Zealand. We used the Tyrer-Cuzick algorithms to assess the prospective risk of breast cancer in women in the kConFab cohort who were unaffected with breast cancer at the time of enrolment in the study. Results Of kConFab's first 822 families, 518 families had multiple cases of female breast cancer alone, 239 had cases of female breast and ovarian cancer, 37 had cases of female and male breast cancer, and 14 had both ovarian cancer as well as male and female breast cancer. Data are currently held for 11,422 people and germline DNAs for 7,389. Among the 812 families with at least one germline sample collected, the mean number of germline DNA samples collected per family is nine. Of the 747 families that have undergone some form of mutation screening, 229 (31%) carry a pathogenic or splice-site mutation in BRCA1 or BRCA2. Germline DNAs and data are stored from 773 proven carriers of BRCA1 or BRCA1 mutations. kConFab's fresh tissue bank includes 253 specimens of breast or ovarian tissue – both normal and malignant – including 126 from carriers of BRCA1 or BRCA2 mutations. Conclusion These kConFab resources are available to researchers anywhere in the world, who may apply to kConFab for biospecimens and data for use in ethically approved, peer-reviewed projects. A high calculated risk from the Tyrer-Cuzick algorithms correlated closely with the subsequent occurrence of breast cancer in BRCA1 and BRCA2 mutation positive families, but this was less evident in families in which no pathogenic BRCA1 or BRCA2 mutation has been detected.


Introduction
The Kathleen Cuningham Foundation Consortium for Research into Familial Breast Cancer (kConFab) is a multidisciplinary, collaborative framework for the investigation of familial breast cancer. Based in Australia, the primary aim of kConFab is to facilitate high-quality research by amassing a large and comprehensive resource of epidemiological and clinical data with biospecimens from individuals at high risk of breast and/or ovarian cancer, and from their close relatives. Methods Epidemiological, family history and lifestyle data, as well as biospecimens, are collected from multiple-case breast cancer families ascertained through family cancer clinics in Australia and New Zealand. We used the Tyrer-Cuzick algorithms to assess the prospective risk of breast cancer in women in the kConFab cohort who were unaffected with breast cancer at the time of enrolment in the study. Results Of kConFab's first 822 families, 518 families had multiple cases of female breast cancer alone, 239 had cases of female breast and ovarian cancer, 37 had cases of female and male breast cancer, and 14 had both ovarian cancer as well as male and female breast cancer. Data are currently held for 11,422 people and germline DNAs for 7,389. Among the 812 families with at least one germline sample collected, the mean number of germline DNA samples collected per family is nine. Of the 747 families that have undergone some form of mutation screening, 229 (31%) carry a pathogenic or splice-site mutation in BRCA1 or BRCA2. Germline DNAs and data are stored from 773 proven carriers of BRCA1 or BRCA1 mutations. kConFab's fresh tissue bank includes 253 specimens of breast or ovarian tissue -both normal and malignant -including 126 from carriers of BRCA1 or BRCA2 mutations. Conclusion These kConFab resources are available to researchers anywhere in the world, who may apply to kConFab for biospecimens and data for use in ethically approved, peerreviewed projects. A high calculated risk from the Tyrer-Cuzick algorithms correlated closely with the subsequent occurrence of breast cancer in BRCA1 and BRCA2 mutation positive families, but this was less evident in families in which no pathogenic BRCA1 or BRCA2 mutation has been detected.

Introduction
Ten to fifteen percent of women with breast cancer also have close relatives affected, and in many multiple-case families susceptibility to the disease appears to manifest as a dominantly inherited Mendelian trait. The BRCA1 locus on chromosome 17 co-segregates with breast cancer and is characteristic of families in which both early-onset breast and ovarian cancers occur [1]. A similar proportion of breast cancer families, especially those that include one or more cases of male breast cancer, is linked to a second locus (BRCA2) on chromosome 13 [2,3].
In the ten years since BRCA1 and BRCA2 were cloned and sequenced, thousands of pathogenic mutations have been identified. The original expectation was that carriers of these mutations would have very high (>80%) lifetime risks of breast cancer. Such dramatic effects seem to be confined to families with six or more cases of breast cancer or with a combination of breast, ovarian and male breast cancers. Population-based studies show that the average risk to carriers (40% to 65%) may be somewhat less than in multiple-case families and may vary between populations [1,4,5]. The prevalence of BRCA1 and BRCA2 mutations (approximately 1:500, but higher in Ashkenazi Jewish individuals) indicates that they are unlikely to account for more than 3% to 5% of all breast cancers. However, mutation carriers from multiple-case families constitute cohorts in which hypotheses about the role of potential modifying genes and non-genetic risk factors can be tested. Such analyses need to be complemented by population-based studies so that estimates of the prevalence, attributable fraction, and risk of breast cancer can be made for individuals who have not been selected on the basis of a strong family history [4].
Mutations in BRCA1 and BRCA2 are present in only 35% to 40% of concentrated family clusters of breast/ovarian cancer; hence it is likely that other 'high-penetrance' breast cancer susceptibility genes remain to be identified. Multiple-case families also have the potential, given sufficient numbers, to identify genes, such as CHEK2, that confer lower levels of risk than BRCA1 and BRCA2, and that are more prevalent in the population [6].
For these and other reasons, a series of discussions within the Australian cancer genetics and genetic epidemiology community in 1995 led to the formation of a national consortium. Its underlying goal was to foster research into the causes and impact of familial breast cancer. This would be achieved through construction of a research resource of genetic, epidemiological and clinical data, with appropriate biospecimens, to be available to researchers from anywhere in the world. From its inception the consortium developed important links with the network of family cancer clinics (FCCs) around Australia, facil-itating recruitment into The Kathleen Cuningham Consortium for Research into Familial Breast Cancer (kConFab) and transfer of research findings to participants and their families. kCon-Fab received its first funding in 1997 from the Kathleen Cuningham Foundation [7,8]. We report here on the progress to date of the consortium, and the characteristics of its first 822 multiple-case families. We demonstrate that the Tyrer-Cuzick algorithm effectively predicted breast cancer occurrence in women unaffected at the time of recruitment if they were in a family carrying a BRCA1 or BRCA2 mutation, though not otherwise. In addition, for these unaffected women, calculated probabilities of a pathogenic mutation in BRCA1 or BRCA2 were moderately predictive of such a mutation being found.

kConFab: structure and funding
The aims of kConFab are to collect relevant data and biospecimens from families with multiple cases of breast and/or ovarian cancer and to make these widely available, not only to its members, but to researchers from anywhere in the world for use in peer-reviewed, ethically approved research projects. To further enrich the resource, data from these research projects are returned to kConFab after publication. kConFab's policies are set by an executive committee, whose membership includes a consumer representative. Detailed policies and procedures and reports of progress are available at the kConFab website [9].

Ascertainment of probands and family members
Since 1997, kConFab has been recruiting multiple individuals (affected and unaffected by cancer) from families that have presented to FCCs with evidence of high familial risk of breast cancer. The eligibility criteria for recruitment of families into kConFab have evolved over time but are intended to maximize the number of living potentially high-risk individuals, including carriers of high-penetrance alleles, whether affected by breast cancer or not. Families are recruited by research nurses located in all the largest FCCs in Australia and New Zealand into five categories described in Table 1.
The families are identified and initially evaluated as part of clinical care in an FCC. The de-identified pedigree is submitted to kConFab for review and, if judged eligible, the research nurse approaches a clinic-nominated Family Contact Person. Additional family members are invited to participate only with the permission of this person, or another relative subsequently recruited to kConFab. Participants can be identified from other sources, for example, by surgeons treating breast cancer patients from families who appear to fit the 'potentially high risk' category, although recruitment is again via the kConFab research nurse.

(page number not for citation purposes)
After completion of recruitment and data collection, including verification of the family history of cancer, the pedigree is again reviewed and the final category recorded. The apparent strength of family history may have altered as a consequence of verification of details, and so a proportion of families collected do not fit any category on completion (category 0), most often because insufficient eligible individuals are alive. These families are not discarded as later follow-up may provide further information that may change their category. Breast, ovarian and other cancers are regarded as 'sporadic' if they affect individuals in mutation-negative families who are unrelated by descent to other cases. Most 'sporadic' cancers occur in spouses who do not themselves appear to have a maternal or paternal history of breast or ovarian cancer. 'Sporadic' cases are excluded from the data presented in the tables. The young-est participating individual in the family affected by breast cancer is designated as the index case for the purposes of mutation testing.

Recruitment and follow-up
Each eligible subject is approached by the research nurse via the Family Contact Person or another family member and their informed consent is obtained for participation in kConFab. The research nurse invites participation by all family members reported to have been affected by breast or ovarian cancer, their first-degree relatives, all individuals in ancestral line between affected individuals and both parents of any of these eligible individuals. Each subject is then interviewed to establish their family structure and any reported diagnoses of can- Two or three cases of ovarian cancer among first degree relatives, one or more of whom is alive and Zero or one case of breast cancer and Four or more living first or second degree unaffected female relatives of affected cases a All criteria refer exclusively to individuals over the age of 18. b Families with unclassified variants in BRCA1 and BRCA2 were included in category cer for themselves or eligible relatives. In 2003, the research nurses began annual follow-up of families via the Family Contact Person. The main aim is to offer participation to individuals who have become eligible for kConFab since the family collection was completed, including newly affected family members, children who have attained age of majority since the family ascertainment was completed and first degree relatives of participants who have recently become affected with breast or ovarian cancer. The nurses also approach first-degree relatives of men and women who have learned (through attendance at a FCC) that they carry a BRCA1 or BRCA2 mutation and notify the kConFab research nurse of this. The approach to all new participants is always mediated through an alreadyconsented individual.
Epidemiological risk factor questionnaire Questionnaires [10][11][12][13] covering a broad spectrum of risk factors are completed by telephone, and appropriate elements are collected by proxy for deceased eligible individuals.

Biospecimens
Blood collection and processing A 20 ml sample of anticoagulated blood is collected from all participants and returned to the central core laboratory by courier. Participants who are unable to give blood are mailed a mouthwash kit [14] and 73% of those from whom mouthwash samples have been requested have returned samples. The blood processing protocol [15] generates a nucleated cell product for DNA extraction, Ficoll-hypaque separated lymphocytes for Epstein Barr virus transformation, non-lymphocytic leukocytes for further DNA or RNA extraction, Guthrie cards of blood spots, and plasma aliquots. DNA is extracted as required (QIAamp DNA blood kit, Qiagen GmbH, Hilden, Germany). Lymphoblastoid cell lines are established by Epstein Barr virus transformation as required for research projects or to replace DNA stocks as they are depleted.

Collection of fresh normal and tumor specimens
Subjects are encouraged to report if and when they are to have breast or ovarian surgery for prophylaxis or cancer treatment. Chilled specimens are grossly dissected into 3 mm sections by the clinical pathologist, placed in a histocassette and snap frozen in liquid nitrogen and transferred on dry ice by courier to the kConFab tissue bank. In the case of prophylactic mastectomy, portions containing connective tissue are separated from fat as much as possible and marked as such by the pathologist. In remote locations, or where dry ice is not readily obtainable, specimens are collected and transported in RNAlater (Ambion, Austin, TX, USA). A pro forma questionnaire about menstrual status is completed by women who donate tissues.
All material received in the bank undergoes standardized evaluation and processing, including a report on transfer conditions and state of the shipment on arrival, an assessment of the nature of the specimen and a record of number and size of pieces. After the final histopathology report is obtained, the frozen sections of the sample are examined after haematoxylin and eosin staining (5 µm sections) and five thick (10 µm) sections are then stored in a cryovial in liquid nitrogen. A further 'bookend' thin section is taken for haematoxylin and eosin staining to document the tumor content of the next level of the specimen.
Cancer verification kConFab uses several methods to verify all reports of cancer in the family: through medical records or state-based cancer registries and by systematic searches by the Database Manager of Australian cancer registry records of all eligible consented living participants and deceased family members. The state-based cancer registries have been collecting data since 1982 (and some back to 1970), but most have only recorded ductal carcinoma in situ since 1982. The final level of verification that is reached is noted and, where possible, a copy of the final pathology report is obtained from which the location of archival, diagnostic tumor specimens is recorded so that paraffin blocks and slides can be requested as necessary.

Mutation detection
Testing for BRCA1 and BRCA2 mutations has been performed in 747/822 (91%) of families. In most cases the enrolling FCC will already have undertaken diagnostic testing using a clinical DNA sample from an affected family member according to their local protocols, which were estimated to be about 80% sensitive prior to 2001. Since 2001 most diagnostic laboratories in Australia have adopted denaturing high performance liquid chromatography as the method of choice for BRCA1 and BRCA2 mutation analysis, with estimated 96% sensitivity for exonic and splice-junction variants [16]. Testing of index cases in kConFab families has also been carried out by complete sequencing through Myriad Genetics (Salt Lake City, UT, USA) with the support of a grant from the National Cancer Institute to the Australian Breast Cancer Family Registry [11]. Families prioritized for complete sequence analysis are those with no previous clinical mutation testing because, for example, no affected member of the family wanted a clinical mutation test result, or because the family fulfill the entry requirements of a genome-wide linkage study being undertaken in collaboration with the Breast Cancer Linkage Consortium [1]. In addition, all individuals who have provided a fresh frozen breast tumor (cancer) specimen are tested by full sequencing unless they are already known to be positive for a family-specific mutation in BRCA1 or BRCA2. Analysis for genomic deletions and rearrangements detectable by multiplex ligation-dependent probe amplification [17] is currently underway.
kConFab classifies all BRCA1 and BRCA2 variants reported by diagnostic or research laboratories into categories: patho-genic, splice-site variant, variant of unknown significance and polymorphism. The criteria for classification are posted on kConFab's website [18].
Once the family mutation has been identified, all pathogenic (including splice site) variants of BRCA1 and BRCA2 are genotyped by kConFab in all available family members' DNA.
Access to the unique family number linking a variant to a particular family is password protected and is only available to investigators carrying out kConFab-related research. De-identified, individual mutation results are available only to researchers, and not to the originating research nurse(s) or health professionals in the FCC(s) that the family attends.
All index cases of non-BRCA1/2 families are genotyped by kConFab for the breast-cancer related mutation in ATM, 7271T>G [19], and a small number of families with features of Li Fraumeni syndrome have been tested for mutations in p53 by the diagnostic laboratories.
Haplotyping at BRCA1 and BRCA2 loci Mutation-negative families with female breast cancer, in which sufficient quantities of DNA are available from at least three affected individuals, have been analyzed for haplotype sharing at short tandem repeat polymorphism (STRP) markers flanking and internal to BRCA1 (D17S800, D17S855, D17S951, D17S1322) and BRCA2 (D13S260, D13S1700, D13S171, D13S267) in order to select families suitable for genetic linkage analysis for novel susceptibility loci.

Informatics
The Central Register, located at the Peter MacCallum Cancer Centre, until recently received data from peripheral versions of a Family Based Information database, developed by the Cancer Council of Victoria, by periodic uploads from the recruiting clinic-based research nurses. During 2005, we implemented a TCP-IP based application to permit live access to the central data server from multiple remote locations, based on a Progeny Anywhere (Progeny Software LLC, Grand Rapids, MI, USA) XML-compliant front end. Data in the Central Register can be accessed only by procedures that are ethically approved and in accord with national guidelines on privacy of genetic information [20]. Information is stored both by individual and by family. Individual data include personal and demographic information, biological relationships, cancer diagnoses, cancer treatments, epidemiological information, biological specimens, genotyping and mutation test results, as well as a variety of process and management outcomes.
Periodic checks are run to determine if individuals have been registered through more than one family and, if confirmed and appropriate to the family structure, the families are merged within the Central Register. In this way, and through other consent-approved communication between family cancer clinics, 43 families with individuals in common have been merged into 21, ensuring data and specimens for each individual are only accessed once, and that the combined characteristics of the entire family structure can be analyzed by researchers.

Quality control
Identity and DNA quality are checked on a randomly selected 10% of the DNAs extracted from leukocyte pellets by PCR amplification of five unlinked STRP markers. These genotypes are compared with those from DNA from the matching Guthrie spot.
To date, all DNAs tested in this way have yielded PCR product, and none have shown mismatches with the Guthrie spot DNA.
After its initial submission, and whenever genotyping has been extended within a family as a result of a newly detected mutation, the pedigree is reviewed by a geneticist for internal consistency of the family structure, dates of birth, cancer diagnosis and death, and for Mendelian inheritance of known genotypes. Using DNA isolated from a Guthrie spot, we repeat testing for the family-specific mutation in any individuals with apparently anomalous results; for example, unaffected but mutation-positive individuals with no confirmed carrier or affected offspring, who were linked to the closest confirmed carrier by two or more unaffected individuals of unknown genotype. To date, these and other types of quality control have not revealed any genotyping errors.

Pathology review
A standardized review of original or duplicate diagnostic slides from all index cases in families in which no mutation in BRCA1/2 had been identified was begun in 2004 with the aim of identifying those cases with a phenotype suggestive of a BRCA1 mutation in order to direct additional mutation testing, and to determine if there are distinct pathological phenotypes by which these families could be stratified. The review, aided by a reference document, involves recording tumor size, lymph node and hormone receptor status from original pathology reports and examining slides to evaluate tumor type and grade as well as features of particular significance to familial breast cancer such as lymphocytic infiltrate and the extent of a pushing tumor margin [21].  Ethics kConFab maintains human ethics approval at all participating institutions through which subjects are recruited. All research projects making use of data and/or materials collected by kConFab are required to have independent ethical approval from their host institutions. All participants give informed consent and understand that as a result of participation, personal details will be recorded and stored in a coded format on a database. They consent to samples of genetic material, blood cells and tissue (if applicable), being stored in a central location and to de-identified information and samples being made available for scientifically and ethically approved research projects.

Access to data and biospecimens
Applicants wishing to use data and/or biospecimens first submit a brief expression of interest, which is circulated to the entire kConFab membership with a five-day opportunity given to highlight major issues, especially duplication of, or complementarity to, existing projects. A full application is then submitted and checked against criteria that include evidence of external peer review, sufficient resources to conduct the project, and ethics approval. Given that these conditions are met, approval is given for up to three years, during which annual progress reports are a requirement, and after which the project can be renewed if not complete. To further enrich the kConFab resource, the researchers are required to supply their research data to kConFab after publication, and/or 12 months after completion of their project [22].

Analysis of breast cancer risk
The Tyrer-Cuzick algorithm [23] was developed to model breast cancer risk in unaffected women by taking into account: their probability of carrying genetic risk factors, namely a rare, high penetrance mutation in BRCA1 or BRCA2 and a notional common, low-penetrance dominant susceptibility allele that stood for all other genetic risk factors; and a range of other individual clinical and epidemiological factors known to influence risk, such as menarche and parity. The genetic risk component of the calculations derives from a Bayesian segregation analysis that takes into account known BRCA1/2 genotyping data and family history of breast cancer in first-and second-degree relatives.
A batch program kindly provided by Dr Jack Cuzick [23] was used to estimate the risk of breast cancer in the next decade for all women in the kConFab cohort who had not previously had breast cancer at time of recruitment. The reference date for the subject was the date of epidemiological risk factor interview, but the family data used (family structure and cancer diagnoses in relatives) were those that had been reported in the family by the date of analysis, not only those reported at the time of interview. Two calculations were performed, one in which all reports of previous family breast cancer diagnoses were taken into account, and one including only the verified reports. The program was also used to calculate the probability of testing positive for a BRCA1 or BRCA2 mutation. It is recognized that the circumstances of genetic testing of individuals in the kConFab cohort were often far from those for which this aspect of the algorithm was designed, namely the initial screening of an index case in a previously untested family. Receiver operating characteristic (ROC) analysis of the sensitivity and specificity of these estimates was performed using STATA 7.0 (College Station, TX, USA).

Results and Discussion
Family characteristics and biospecimen collection The majority (518; 63%) of the 822 families in this cohort have cases of female breast cancer in successive generations, but no cases of ovarian or male breast cancer (  The median age at diagnosis of female breast cancer, male breast cancer and ovarian cancer in BRCA1 and BRCA2 carriers or pathogenic or unclassified variants and those without identified variants are given in Table 3. As expected, the median age at diagnosis of the first female breast cancer was lower in BRCA1 and BRCA2 carriers than in those women without mutations. Although testing is incomplete, only two families (Table 1, category 3) have identifiable mutations in genes other than BRCA1 or BRCA2 that are likely to have conferred a high risk of breast cancer. One family has a mutation in TP53 and another has a 7271T>G mutation in ATM [19] (Table 4). Twenty-five potentially high-risk families were enrolled into kConFab (Table 1) because a family member wanted to give fresh surgical tissue, but the family did not fit other kConFab ascertainment criteria.
Germline DNA samples have been collected from 7,389 individuals (Table 4). Epidemiological questionnaires have been collected from 7,039 individuals, with more limited data concerning a further 4,383 deceased relatives collected by proxy (Table 4). Participation has been greater by female than male family members, with females accounting for 65% of the biospecimens and 64% of the questionnaires provided.
The mean number of germline DNA samples collected per family is nine (median of eight; Table 5). From 122 families, germline DNA samples were collected from three or more affected individuals (Table 5). Lymphoblastoid cell lines are currently available for 462 subjects, and can be made from any consented individuals from whom at least 10 ml blood was collected (92% of all blood samples).  To date, 20 families carry mutations that are believed to have resulted from Alu recombination events resulting in large genomic deletions and rearrangements in BRCA1. The common and well-characterized mutation BRCA1 exon 13 duplication accounts for 8 of these families and 12 families carry variants of either single or multiple exon deletions and duplications. The five families whose genomic rearrangements have not yet been fully characterized are not included as mutationpositive in the tables.
Genotyping has been used to select families suitable for genetic linkage analysis for non-BRCA1, non-BRCA2 high penetrance susceptibility loci, namely those in which all affected individuals do not share a haplotype at BRCA1 or BRCA2 (Table 8). Haplotyping has also been useful in redirecting mutation analysis, for example if the majority of affected individuals in the family shared a BRCA1 or BRCA2 haplotype but the screened individual(s) did not, and to prioritize families for more intensive mutation analysis such as full sequencing or genomic rearrangement analysis. Where germline DNA samples were available from at least three affected individuals, haplotyping at STRP markers flanking BRCA1 and BRCA2 was carried out in BRCA1, BRCA2 mutation-negative families.

Research projects supported
Since 1998, 45/48 research project applications have been given approval to use kConFab biospecimens and data, one third of these in the last two years. The average time from receipt of an application to approval of the project has been four months, which usually includes some clarification of the protocol with the applicant. Three projects did not go ahead because of lack of funding or the unwillingness of the host institution to sign a materials transfer agreement. Active research projects are listed on the kConFab website [24]. The investigators involved come from multiple institutions in Australia, one in the United States, one in the United Kingdom and one in France. To date, these projects have resulted in 29 primary research publications [25], ranging from reports on the cumulative risk of breast cancer in families with BRCA1 and BRCA2 mutations to studies of psychological morbidity in these families [8,19,[26][27][28][29][30][31][32][33][34][35][36][37].

Quantification of breast cancer risk
The risk of breast cancer in the ten years after recruitment was calculated for all 4,815 unaffected female kConFab subjects in which epidemiological interview data, collected directly or by proxy, were available. This risk was expressed as either the absolute risk or as a ratio of population risk for that individual (relative risk).
In the cohort as a whole, the distribution of relative risk indicated that our ascertainment scheme and procedures had been reasonably efficient. The median relative risk was 2.47, with 91% of subjects having a risk higher than that of a woman in the population of the same age and 859 (18%) having a more than 5-fold elevation.
The graphs in Fig. 1 show the relationship between the probability of having developed breast cancer for the first time since interview and the cumulative risk over the period of follow-up as calculated using the Tyrer-Cuzick model. Over this time (median 4.2 years) there were 107 cases of breast cancer confirmed by medical records, 2.9% of the 3,657 women alive at the time of initial data collection. This was very close to the 110 cases expected. Women in the highest risk quintile (absolute risk > 0.14, relative risk > 10.1) accounted for 49 of those cancers (46%, data not shown). Overall, 55 cancers (51%) occurred in women in the highest quintile of cumulative risk, in which length of follow-up was taken into account (Fig.  1a). In families with a pathogenic BRCA1 or BRCA2 mutation, women in the highest quintile of cumulative risk experienced almost all, 92% (36/39), of the cancers (Fig. 1b). In contrast, families without a BRCA1 or BRCA2 mutation showed a weaker, but more linear relationship between calculated cumulative risk and the probability of subsequent breast cancer (Fig. 1c). Interestingly, there was no gain in predictive power when the cancer diagnoses in family members, used in the risk calculation, were restricted to those verified from medical records rather than using all reports (data not shown). For example, women in the highest quintile of cumulative risk experienced 55 of the 107 breast cancers when the verified data were used, exactly as above. The data for BRCA1 and BRCA2 mutation positive families, using verified data only, were also essentially identical to those in Fig. 1b.
High risks calculated by the Tyrer-Cuzick algorithm therefore correlated well with future incidence of breast cancer, but this effect was much stronger in BRCA1/2 mutation positive families than in BRCA1/2 mutation negative families. Table 6 Mutation status of individuals from whom fresh frozen normal and malignant ovarian and breast specimens have been obtained Type of surgery/tissue BRCA1 mutation positive

Prediction of BRCA1 and BRCA2 mutation status
The Tyrer-Cuzick batch program was also used to calculate the posterior probabilities that an unaffected woman was a carrier of a BRCA1 or BRCA2 mutation. These estimates were compared with the actual prevalence of BRCA1 and BRCA2 mutations in these individuals. No data on the presence of these mutations were used in the generation of the estimates. This aspect of the algorithm was developed for use in the context of an unaffected woman presenting from a family in which no mutation screening had been done. Its application to all unaffected women in the kConFab cohort is somewhat artificial because in many cases they were recruited into kCon-Fab after a family mutation had already been found. Nonetheless, the proportion of women testing positive for a BRCA1 mutation is seen to correlate well with the estimated probability of BRCA1 mutation (Fig. 2a). The quintile with the highest probability showed more than six times the prevalence of BRCA1 mutations in the lowest quintile; the area under the ROC curve was 0.67. There was also positive correlation between the BRCA1 estimate and the observed prevalence of BRCA2 mutations (Fig. 2a). Interestingly, the estimated probability of BRCA2 mutation was somewhat less effective than the estimated probability of BRCA1 mutation in predicting the presence of a BRCA2 mutation (Fig. 2b): the area under the ROC curve was 0.61.

Conclusion
Knowledge of the underlying determinants of breast and ovarian cancer risk is still very incomplete. kConFab has therefore focused on recruitment and comprehensive characterization of a cohort of women at significantly higher than average risk of breast and/or ovarian cancer, together with large numbers of their relatives. This approach has already been productive in yielding many carriers of known predisposing mutations, and it is expected that these numbers will increase as BRCA1 and BRCA2 testing is completed, and as new predisposing genes are discovered. Ascertainment to kConFab is not population based, but is dependent on clinical identification of individuals at high risk of breast and ovarian cancer. Data and analyses derived from cohorts of this type are shaped by the selection criteria used to refer families to FCCs, the specific kConFab ascertainment criteria, and by the dynamics of family participation in research. Such clinic-based studies cannot replace population-based approaches but are complementary to them, and this project has already provided several examples of productive collaboration [19]. However, we are only recruiting a small proportion of high-risk breast cancer families in Australia and New Zealand, and recruitment is not yet reaching a plateau. The number (1,159) of BRCA1/2 mutation carriers for whom data is currently held is approximately 5% of the number predicted from the plausible population frequency of these alleles and the current population size of 22 million. By the time recruitment of the kConFab cohort is complete, we estimate that we will have enrolled approximately 10% of multiple case families from Australia and New Zealand. Further analyses will be needed to test how representative the cohort is of the diverse populations of our two countries.
The establishment of an organization such as kConFab cannot be achieved without some difficulties. For example, large scale centralized pathology review of breast and other cancers occurring in kConFab participants is beyond the scope of current core activities. As a consequence, systematic review has focused on index cases of non-BRCA1/2 families since these data are likely to be of value to research projects. Additional pathology data are contributed to kConFab by investigators who undertake review in the context of specific research projects.
Follow-up contact with the families has also proved challenging. For example, recontacting subjects for projects that include the study of known mutation carriers requires very careful attention to ensure that the presence or absence of a family mutation is not revealed to individuals who do not know, or want to know, their mutation status. Projects that involve approaching both mutation-positive and mutation-negative members of families, and can be done blind to the mutation status, are easiest to accommodate within kConFab.
One of the most difficult issues has been to ensure that the participants understand the differences between 'research' and 'clinical' mutation testing. Anecdotal evidence suggests that many participants still confuse the issues, and expect kConFab to give them individual mutation results. Most of the people enrolled in kConFab are not clients of FCCs, but relatives of people who are. Consequently, kConFab's notification letters do not tell individuals about their own carrier status, but state that a family specific mutation has been discovered and that if the individual wishes to know his/her carrier status, he/ she should contact an FCC. A second blood sample can then be taken and tested in an accredited laboratory for the presence or absence of the family specific mutation. The result of this test is then given to the individual, after counseling.
Even though kConFab's ascertainment strategy is targeted to individuals with a potentially high risk of breast or ovarian cancer and their relatives, the family pedigrees contain many family members who are not necessarily at high risk, for example, spouses with no blood relationship to the cases of breast or ovarian cancer in the pedigree. Studies based on the secondary recruitment of potentially high-risk individuals within the kConFab cohort, for example, examining the determinants of clinical outcome or of psychological adjustment to high risk of cancer, have had to take additional steps to remove such individuals from their ascertainment, including implementing a system of risk stratification within the cohort.
kConFab can be compared to the European-based Breast Cancer Linkage Consortium (BCLC), which focuses on families ascertained through family cancer clinics, and the NCIfunded Breast Cancer Family Registry (BCFR) that includes both clinic-based and population-based sites. There are enough similarities between these cohorts to allow collabora- tive projects to use data from each cohort. A major strength of kConFab over the BCFR is the large number of available blood samples per family, which allows for more accurate penetrance estimates, and a larger number of available samples from known mutation carriers. The BCLC families are large, but epidemiological data have not been systematically collected, nor is there a central repository of biological specimens for which researchers can apply.
The Tyrer-Cuzick algorithm was applied to the cohort to test its ability to predict future breast cancer and positive BRCA1 or BRCA2 mutation tests, with considerable success. In BRCA1 or BRCA2 mutation positive families, the highest risk quintile included almost all the women that developed breast cancer over the comparatively short period of four years of follow-up. It should be noted that in these families there was already considerable information on the presence of these mutations throughout the families, and that this clearly contributed to the high predictive power of the risk estimates. When there was no BRCA1/2 mutation known in the family, the algorithm stratified the women much less strongly. This could be interpreted to mean that in these mutation negative families individual nongenetic risk factors such as parity and menarche still contribute to risk in the context of a strong family history but rather more weakly, even in aggregate, than as yet undetected major genetic effects.
The future data and specimen accrual goals of kConFab include recruitment of an additional 400 families to reach our target of at least 1,200 multiple-case families, collection of annual serum and plasma samples from a subset of women at risk, more intensive mutation analysis of non-BRCA1/2 families, further pathology review, and the generation of tissue Relationship between estimated probability and predicted relative fre-quency or BRCA1 and BRCA2 mutations Relationship between estimated probability and predicted relative frequency or BRCA1 and BRCA2 mutations. Women unaffected at time of interview were ranked in quintiles according to their probability of carrying either a (a) BRCA1 or (b) BRCA2 mutation, as estimated by a batch program implementing the Tyrer-Cuzick algorithm, without input of family or individual mutation status. The proportion of women that tested positive for either a BRCA1 (open bars) or BRCA2 (closed bars) mutation is plotted on the vertical axis.

Figure 1
Relationship between estimated cumulative probability and actual rela-tive frequency of breast cancer since ascertainment Relationship between estimated cumulative probability and actual relative frequency of breast cancer since ascertainment. Women unaffected at time of interview were ranked in quintiles according to their cumulative risk of breast cancer since interview, calculated from the absolute risk estimated by a batch program implementing the Tyrer-Cuzick algorithm. The proportion of women in each risk quintile (horizontal axis: 1 lowest to 5 highest) developing breast cancer during a median 4.2 years follow-up is plotted on the vertical axis plus standard error. (a) All families; (b) BRCA1 or BRCA2 mutation positive families; (c) BRCA1 or BRCA2 mutation negative families.