Evaluation of linkage of breast cancer to the putative BRCA3 locus on chromosome 13q21 in 128 multiple case families from the Breast Cancer Linkage Consortium

Methods

Statistical Analysis.

We performed standard parametric linkage analyses, essentially identical to our previous analyses of linkage in breast cancer families (e.g., refs. 1, 5, and 8) and to the analysis conducted by Kainu et al. (10). These analyses assume the model of susceptibility to breast cancer based on the segregation analysis of Claus et al. (9). Under this model, susceptibility to breast cancer is conferred by a dominant allele with population frequency of 0.003. The risk of breast cancer by age 80 is assumed to be 0.80 in carriers and 0.08 in noncarriers. Risks are modeled in seven age categories (<30, 30–39, 40–49, 50–59, 60–69, 70–79, and 80+) as described in Easton et al. (12).

Multipoint linkage analyses were carried out using the programs GENEHUNTER (V. 2.0-B; ref. 13), VITESSE (14), and FASTLINK (15). GENEHUNTER was used where possible because it can analyze large numbers of polymorphic loci simultaneously and hence all of the markers we used could be incorporated into a single analysis. However, 33 families were too large to be accommodated by GENEHUNTER without discarding informative individuals. For these families we computed multipoint LOD scores by using either VITESSE (29 families) or FASTLINK (four families with multiple founders). The analyses assumed the intermarker distances as shown in Table 2.

We used the multipoint LOD scores for each family to compute heterogeneity LOD scores, using the standard admixture model, and hence estimated the proportion of families (α) linked to the putative “BRCA3” locus by maximizing the heterogeneity LOD score. A 95% confidence interval for α was derived by computing the values of the heterogeneity LOD score that were within 0.83 (corresponding to a Z value of 1.96) of its maximum value. Ninety-nine percent confidence intervals were also computed.

Because the putative “BRCA3” locus on 13q21 is linked to BRCA2, we performed a further analysis to allow for the possibility that preferential selection for families unlinked to BRCA2 may have biased the results against linkage at “BRCA3.” In this analysis, we computed multipoint heterogeneity LOD scores at the candidate “BRCA3” locus, conditional on the LOD scores at BRCA1 and BRCA2, according to the formula:

In this formula α₁, α₂, and α₃ are the proportions of families meeting the eligibility criteria that are linked to BRCA1, BRCA2 and “BRCA3,” respectively, and μ is the sensitivity of BRCA1/2 mutation screening. For the purposes of these analyses, α₁ and α₂ were set to 0.15 and μ to 0.7. LOD₁(θ₁) and LOD₂(θ₂) are the LOD scores at BRCA1 and BRCA2, respectively, whereas LOD_2;3(θ₂) and LOD_2;3(θ₃) are the LOD scores at BRCA2 and “BRCA3,” respectively, based on markers typed at both loci; LOD₂(θ₃) is the LOD score for “BRCA3” calculated using only markers at BRCA2. This calculated LOD score is the likelihood for the linkage data at “BRCA3” conditional on the existing linkage and mutation evidence at BRCA1 and BRCA2, and hence corrects (albeit conservatively) for any bias in the “BRCA3” evidence produced by exclusion of families linked to BRCA2.

Results

Total LOD scores were strongly negative throughout the 8-cM interval between D13S153 and D13S1291 (Table 2 and Fig. 1). At the location of “BRCA3” estimated by Kainu et al. (10), D13S1308, the total LOD score was −38.00. Based on the admixture model, the estimated proportion of linked families (α) was 0, with an upper 95% confidence limit of 0.13. The estimated α was also zero for all possible positions in the interval D13S153-D13S1291. Of the 128 families, only four had a multipoint LOD score of greater than 0.5 at D13S1308, the highest of which was 0.67 (one additional family achieved a LOD score of 1.55 at a more distal marker, D13S800). Twelve families achieved LOD scores less than −1 at D13S1308.

We reanalyzed the data conditioning on the genotyping data at BRCA1 and BRCA2. In this analysis the total LOD score was −25.08. In the heterogeneity analysis based on these conditional LOD scores, the estimated proportion of families linked to “BRCA3” was again 0, with an upper 95% confidence limit of 0.18.

In the 95 families that could be analyzed with GENEHUNTER, we also analyzed the data by using the nonparametric method (13) to evaluate haplotype sharing among affected women. Again, no significant evidence of linkage was found (data not shown).

Discussion

Our results clearly conflict with those reported by Kainu et al. (10). Using a set of multiple case female site-specific breast cancer families analyzed for a similar set of markers within the candidate region and subjected to comparable statistical analysis, we found no evidence of linkage to 13q21. The proportion of linked families (65%) reported by Kainu et al. (10) is excluded with a high degree of statistical significance (the heterogeneity LOD score at α = 0.65 was −11.03 in our dataset). This is true even after a conservative correction for possible bias due to potential exclusion of families linked at the BRCA2 locus (conditional LOD at α = 0.65 was −7.64). In addition, under both unconditional and conditional analyses, the estimated proportion of linked families was 0, with upper 95% confidence intervals of 13% and 18%, respectively, indicating that if there is a susceptibility locus on 13q, it is likely to account for only a minority of breast cancer families. The paper of Kainu et al. (10) did not provide confidence limits on their estimated proportion of linked families. However, based on their LOD scores given under homogeneity and 65% heterogeneity, and assuming confidence intervals that are symmetrical about the best estimate, we have estimate a lower 95% confidence limit for α of 0.31. Thus the 95% confidence limits for the two studies do not overlap. Moreover, even when using a more stringent criteria of 99%, the upper confidence limit for our estimated proportion of linked families is 0.19 for the unconditional analysis and 0.26 for the analysis conditioning on BRCA2 markers, further indicating a minor role, if any, for this locus.

There were some differences in selection criteria between the two studies. Our study was restricted to families in which at least three cases of breast cancer were diagnosed below age 60, whereas Kainu et al. (10) included families with three cases diagnosed at any age. Thus, our families may be more heavily selected for genes conferring high risk. It is perhaps noteworthy that the initial hypothesis-generating family analyzed by comparative genomic hybridization (CGH) in Kainu et al. (10) would not have qualified for our study because only two of the five cases were diagnosed under age 60. However, in the subset of 51 families with less than three cases diagnosed under age 50 (Table 1), there is also considerable evidence against linkage to this locus (multipoint LOD = −8.06; HLOD = 0; upper 95% CI for α = 24%; HLOD for α of 65% = −3.57). Thus it is unlikely that difference in age criteria can explain the differences in results between the two studies.

An additional difference in selection criteria was exclusion of families with any cases of ovarian cancer in our series, given the close association of this disease with BRCA1 and BRCA2. Although no BRCA2 mutations were identified in the family set of Kainu et al., the combination of detection methods applied to screening families have detection sensitivities of ≈0.70 (1, 11). Thus, although simulated linkage results allowing for up to 25% of the families in the dataset of Kainu et al. (10) being due to undetected BRCA2 mutations only exceeded the observed maximal lod score in 1 of 3,000 replicates, it is not known to what extent the seven families with ovarian cancer contributed to the observed overall LOD score.

The families in our study were drawn from Western Europe, or in descendent populations in North America and Australia, whereas the families studied by Kainu et al. (10) were from the Nordic countries. Although we have not specifically examined the ethnic origins of each family in our set, it is anticipated that the set of families from the United States and Canada (n = 43) are more ethnically heterogeneous, although most, if not all, are of Western European origin. Only a small minority of all of the families in our set are likely to be of Scandinavian origin, most notably the families ascertained in Minnesota, Seattle, and other parts of the Midwest, which have a high concentration of families descendent from emigrants of Sweden and Norway. One might speculate that the difference in the results observed is due to a population specific founder effect—i.e., an excess of some specific mutation in “BRCA3” in the Nordic populations.

We believe this to be unlikely. The different Nordic populations have different population histories and do not originate from a single small founder population. Although closely related, the Swedish, Icelandic, and (to a lesser extent) Finnish populations are also genetically similar to English and Dutch populations (17). If the observed linkage were due to a susceptibility allele that had reached a high frequency in the Swedish and Finnish populations, this allele would also be expected to occur at a detectable frequency in the British and Dutch families. On the other hand, if the linkage is the result of several different mutations in the candidate “BRCA3” gene, the expectation would be that (as in the case of BRCA1 and BRCA2) mutations would also occur in the British, Dutch, and other populations, albeit the set of mutations might be different. Under either model, we would have expected to observe similar evidence of linkage in our families. Indeed, even when the prevalence of a population specific founder mutation has led to a specific susceptibility gene accounting for the majority of families of a hereditary cancer syndrome [e.g., BRCA2 in the Icelandic population accounting for 61.4% of breast cancer families (18); >50% of hereditary non-polyposis colon cancer (HNPCC) families in the Finnish population attributable to two specific MLH1 mutations (19)], these same genes account for a substantial fraction of families with the same cancer syndrome in other populations [breast cancer reviewed in (20); HNPCC (19)].

We conclude therefore that any contribution of a locus at chromosome 13q21 to familial breast cancer is likely to be small in breast cancer families of European origin. Further linkage studies in large series of multiple case families, or targeted association studies in large series of breast cancer cases and controls, will be needed to identify remaining genes underlying familial aggregation of the disease.