Editor—A number of genes are now known to be involved in inherited susceptibility to breast cancer. In the context of multiple case families, the most well identified and characterised genes are BRCA1 on 17q1 2 and BRCA2 on 13q.3 4 Both linkage and mutation studies indicate that these genes account for the large majority of families segregating both early age of onset breast cancer and ovarian cancer, and those with both male and female breast cancer cases. However, other studies5-8 indicate that BRCA1and BRCA2 account for less than half of site specific breast cancer families, suggesting that other major susceptibility genes may be important in this group. We therefore initiated a collaborative genomic search, using a large panel of female breast cancer only families in which germline mutations in theBRCA1 and BRCA2coding sequence were not detected and in which there was no strong evidence of linkage to BRCA1 andBRCA2. In this letter, we will report the results of linkage analysis in the Cowden region previously implicated in susceptibility to breast cancer.

Cowden disease (CD) is a rare, autosomal dominant syndrome characterised by multiple hamartomatous lesions of the skin, mucous membranes, intestinal polyps, and an increased risk of breast and thyroid cancer. The risk of breast cancer in CD may be of the order of 30% by the age of 50.9 After the CD gene was mapped to chromosome 10q22-23 by Nelen et al,10 CD has recently been shown to be the result of germline mutations in the tumour suppressor genePTEN (or MMAC1 orTEP1) on chromosome 10q.11 12All these findings suggest PTEN as a possible candidate gene that may account for those families not resulting from BRCA1 orBRCA2.

Families were genotyped at two institutions, the International Agency for Research on Cancer (IARC) and the Institute of Cancer Research (ICR). The IARC data set contains 10 families, identified from four centres (University of Pennsylvania, Creighton University, Institute Curie, and Centre Jean Perrin in Clermont-Ferrand). Each family contained at least four cases of breast cancer, at least three of which were diagnosed before the age of 60 and had a DNA sample available for analysis. The UK data set contains 46 families with at least three affected cases, contributed by ICR, University of Leiden, Deutsches Krebsforschungszentrum (DKFZ), and University of Southampton. All families were genotyped for several markers flanking theBRCA1 and BRCA2loci and all families had at least one affected subject screened for germline BRCA1 andBRCA2 mutations. Screening for mutations was carried out using either direct sequencing or SSCP/heteroduplex.7 The sensitivity of mutation testing for detecting mutations within the coding regions is thought to be over 85%. The overall sensitivity, taking into account non-coding alterations, is estimated to be 60-70%.8 For all families without a BRCA1/2 mutation, we calculated the posterior probability of the family being the result ofBRCA1/2, based upon the pedigree structure, the breast cancer phenotypes, the sensitivity of the mutation screening, and the linkage data. Prior probabilities used were those reported in Ford et al.8 The vast majority (95%) of these families had posterior probabilities of being the result of BRCA1 orBRCA2 of less than 30% (data not shown), and most had posterior probabilities less than 15%. The average number of affected per family was 4.9. The number of pedigrees typed for different markers is listed in table 1.

Using all 56 pedigrees, we carried out a power calculation based on computer simulation.13 Results indicate that this data set has reasonable power to detect linkage to a locus which can explain at least 75% of the families (expected lod score 5.04 for a marker with 10 equally frequent alleles 1 cM from the disease locus).

Both parametric and non-parametric multipoint linkage analysis was carried out using the GENEHUNTER program,14 with the exception of one large family for which the multipoint point score was calculated using the FASTLINK program.15 16 For the parametric analyses we assumed a susceptibility allele with frequency 0.003 and age specific risks to susceptible subjects given by the model of Claus et al, as modified by Eastonet al.17 Marker allele frequencies were estimated from the married in spouses in the pedigrees. The order and genetic distances between markers were obtained from the Genethon map18 and are as follows: D10S219 (6 cM) D10S573 (1.8 cM) D10S215 (1.1 cM)PTEN (1.1 cM) D10S541 (2.6 cM) D10S564 (4 cM) D10S583 (1 cM) D10S185 (1 cM) D10S574.

The multipoint analysis results using all markers flanking thePTEN locus failed to support the hypothesis of linkage (table 2). The 56 families included in this analysis gave an overall multipoint lod score at the PTENlocus of −8.25. The maximum likelihood estimate of the proportion of families linked to the PTEN locus was 0.0 with a 95% confidence interval of 0.00-0.32. Partitioning the families by average age of onset gave a small positive heterogeneity score (hlod=0.21) in the families with earlier age of onset with an estimate of 23% linked (95% CI 0.00-0.72). Similarly, when the families were stratified by number of breast cancers in the pedigrees, a small positive heterogeneity lod score (0.34) was also obtained in the 20 families with six or more cases, with an estimated proportion of 34% linked (95% CI 0.00-0.87). To guard against false exclusion ofPTEN owing to inaccuracies in the assumed genetic model, non-parametric analysis was also performed. The non-parametric analyses also gave no hint of linkage to thePTEN locus. In light of this study, we conclude that the gene responsible for Cowden disease is unlikely to be a major determinant of non-BRCA1/BRCA2familial breast cancer, at least in these 56 families, and that other major susceptibility loci are likely to exist.