Prevalence of BRCA1 and BRCA2 germline mutations in young breast cancer patients: A population-based study

Authors


Abstract

Our aim was to estimate the prevalence of mutations in the BRCA1 and BRCA2 genes among unselected incident cases of breast cancer in young women. We identified 158 incident breast cancer cases diagnosed before age 46 years in predefined geographic areas in Girona and Tarragona, Spain, during 1995–1997. Of these, 136 (86%) provided information on family history of cancer and were screened for BRCA1 and BRCA2 mutations. Nine of the 136 (6.6%) were found to carry BRCA deleterious mutations (MUT) (1 BRCA1 and 8 BRCA2), and 20 were detected with rare BRCA variants of unknown significance (UV). Both MUT and US BRCA alterations were more frequent in younger patients: 7 (11.6%) MUT and 12 (19.3%) UV carriers were found in the group of 62 patients younger than 40 years, whereas 2 (2.7%) MUT and 9 (12%) US carriers were identified in the group of 74 patients aged 40–45. Family history of breast and ovarian cancers suggestive of hereditary condition (at least 2 first- or second-degree relatives affected with breast cancer or at least 1 relative affected with ovarian cancer or early-onset breast cancer) was absent for 5 of 9 MUT carriers. This suggests that BRCA screening policies based on family history of cancer would miss a considerable proportion of BRCA mutations. Mutations in the BRCA1 and BRCA2 genes explain at least 10% of breast cancer cases diagnosed before age 40 years. The contribution of these genes to early-onset breast cancer is likely to be even higher given that certain UV cases might be disease-associated. © 2003 Wiley-Liss, Inc.

Breast cancer is the most frequent cancer type among women in the world, affecting up to 12% of all women in Europe and North America.1 Germline mutations of the BRCA1 and BRCA2 genes substantially increase the lifetime risk of developing breast and ovarian cancers.2, 3 The estimated cumulative risk for breast cancer by age 70 in carriers of these mutations is about 80%.2, 3, 4, 5 Estimates of penetrance and contribution of BRCA1 and BRCA2 to cancer incidence have been mainly based on data from high-risk families and might not reflect the situation in all BRCA1/2 carriers. Several studies on founder mutations in populations like Ashkenazi Jews or Icelanders have provided penetrance estimates that were lower than those obtained in high-risk families.4, 5 The same trend was observed in Australian and UK studies on outbred population-based series of breast cancers, estimating breast cancer penetrance to be 40–50% by age 70.6, 7 Several reports estimating the prevalence of BRCA1 and BRCA2 mutations in large population-based samples of breast cancer cases have been published.6, 7, 8, 9, 10, 11 However, all but the study of Loman et al.11 confronted the problems of a relatively low proportion of incident cases included in the analysis and limited mutation screening protocols.

Our aim was to evaluate in the Spanish population the prevalence of mutations in the BRCA1 and BRCA2 genes among unselected incident cases of breast cancer diagnosed before the age of 46 years.

MATERIAL AND METHODS

Patients

We identified 158 incident invasive breast cancer cases diagnosed before the age of 46 years and histologically confirmed in 2 predefined geographic areas in Spain, Girona and Tarragona, during the period January 1995–June 1997. Both areas are covered by population-based cancer registries. Data in the Tarragona registry are regularly included in the IARC publication Cancer Incidence in Five Continents.1 Patients were identified at diagnosis through the main oncology departments of each geographic area. Additional incident cases were identified by a systematic search of pathologists' records in each geographic area. After obtaining informed consent, a personal interview was carried out, to collect information on reproductive and medical histories, lifestyle factors and a detailed family pedigree. Women provided a blood sample for genetic analysis. The median time interval between cancer diagnosis and blood sampling was 10 months. For each relative, information was collected on clinical status, current age and, if dead, cause of death. If a relative had been diagnosed with cancer, information on diagnosis and the center where he or she had been treated was requested. Pathology reports were, however, available for a minority of the affected relatives.

Mutation screening

Genomic DNA was extracted from blood samples using the Qiagen (Chatsworth, CA) DNA extraction kit. The entire coding sequence and intron–exon junctions of the BRCA1 and BRCA2 genes were screened with heteroduplex or denaturing HPLC (DHPLC) analysis. For heteroduplex analysis 33P-dATP-labeled PCR products were denatured at 95°C for 5 min and cooled to room temperature over at least 30 min to induce heteroduplex formation. DNA was electrophoresed through 1XMDE gel (FMC Bioproducts, Rockland, ME) at 300–600 V for 14 hr using a vertical gel electrophoresis apparatus. Gels were dried and exposed to Kodak (Rochester, NY) BioMax MR film. DHPLC was carried out on an automated instrumentation WAVE DNA fragment analysis system (Transgenomic, Omaha, NE). PCR product (3–10 μl), containing about 50 ng of amplified DNA, was denatured for 3 min at 95°C and then gradually reannealed by decreasing the temperature to 65°C over 30 min. PCR products were then eluted with a linear acetonitrile gradient adjusted to the size of the PCR fragment, at a flow rate of 0.9 ml/min and at a denaturing temperature determined with the DHPLC melting algorithm (available at http://insertion.stanford.edu/cgi-bin/melt.html). Under conditions of partial heat denaturation and acetonitrile gradient, heteroduplexes formed in PCR samples carrying sequence variations displayed reduced retention times compared to their homoduplex counterparts. PCR fragments showing abnormalities were directly sequenced using the 310 DNA sequencer (Applied Biosystems, Foster City, CA). All detected mutations were confirmed independently.

Statistical analysis

Clinical characteristics of mutation carriers vs. noncarriers were compared by means of the t-test for continuous variables and the χ2 heterogeneity test or Fischer's test for categorical variables.

The odds ratio (OR) and its 95% confidence interval (CI) were calculated for several breast cancer risk–related characteristics in mutation carriers and noncarriers.

Index cases (n = 158) reported 2,328 first- and second-degree relatives, of whom 417 were siblings. Among these, 964 were female relatives, 58 of whom were affected with breast cancer and 6 with ovarian cancer. The cancer incidence in the relatives of index cases was compared to the incidence estimates in the general population based on the Cancer Registry of Tarragona. The age-standardized incidence ratio (SIR) was used to compare the observed number of cases with the expected number in the general population. The 95% CIs for the SIR were estimated using the approximation of Breslow and Day.12

RESULTS

We identified 158 incident breast cancer cases diagnosed before the age of 46 years in Girona and Tarragona, Spain, during January 1995–June 1997. Here, we report on 136 women (86.1%) for whom both personal interview and BRCA1/BRCA2 analysis were completed. The remaining patients were excluded for the following reasons: 3 died before we could contact them, 1 could not be contacted, 7 provided blood but no interview and 11 provided an interview and blood sample but the small amount of DNA extracted did not allow complete screening of the BRCA1 and BRCA2 genes.

The entire coding regions and exon–intron junctions of BRCA1 and BRCA2 were screened in the136 early-onset breast cancer patients. One BRCA1 (0.7%) and 8 BRCA2 (5.9%) frameshift truncating mutations (MUT) were identified (Table I). All mutations but one (4164del19-ter1328) were previously reported in the Breast Cancer Information Core (BIC) database (http://www.nhgri.nih.gov/Intramural_research/Lab_transfer/Bic/). These mutations have been identified previously in families of west European origin. Some of them (BRCA1, 243delA-ter49; BRCA2, 3036del4-ter958, 3492insT–ter1098, 9254del5-ter3015) have been detected multiple times in Spain, thus probably representing founder mutations in this population.13, 14

Table I. BRCA1 and BRCA2 Germline Alterations and Family History of Cancer in Early-Onset Breast Cancer Cases
CaseAge at diagnosis (years)GeneGenetic alterationPredicted effectFamily history of cancer
First-degree relativesOther relatives
Mutations
 101839BRCA24164del19ter1328 Breast (grandmother)
 103333BRCA26630del5ter2137 Breast (aunt age 78)
 104445BRCA29254del5ter3015Ovary (mother age 72)Ovary (grandmother age 83)
     Prostate (father age 47)Breast (cousin age 28)
      Breast (cousin age 30)
 106135BRCA26503delTTter2098  
 203030BRCA29326insAter3043 Breast (grandmother age 58)
 204244BRCA25804del4ter1862 Ovary (aunt age 70)
 205138BRCA1243delAter49Ovary (mother age 71)Ovary (grandmother age 81)
      Breast (cousin age 48)
 206131BRCA23492insTter1098Breast (sister age 30) 
 206436BRCA23036del4ter958Prostate (father age 64) 
Rare variants of unknown significance
 101239BRCA2654-24G>AUnknown Leukemia (uncle age 68)
 101542BRCA28982+11A>CUnknownBreast (mother age 48) 
 101639BRCA23260C>GThr1011Arg  
 102142BRCA1561-34C>TUnknown Leukemia (grandmother age 35)
  BRCA22999A>TAsn924IIe  
 102843BRCA23283C>GLeu1019Val  
 104534BRCA24289C>TThr1354Met  
 105245BRCA15525+45C>AUnknownBreast (sister age 45)Breast (aunt age 50)
      Leukemia (uncle age 72)
      Breast (aunt age 78)
 106845BRCA26551G>AArg2108HisBreast (mother age 56)Colon (uncle)
     Colon (father age 56) 
 107241BRCA24289C>TThr1354Metcolon (mother age 65) 
 107843BRCA22020A>GThr598Ala  
 107939BRCA1101-21insATUnknown  
 200943BRCA1421-24deIAATUnknown Breast (grandmother age 92)
      Colon (grandfather age 65)
 201532BRCA12852A>GUnknown Colon (aunt age 36)
      Breast (cousin age 55)
 202725BRCA29083A>GMet2952Val Leukemia (uncle)
 203233BRCA1101-21insATUnknown Breast (grandmother age 60)
 204137BRCA14654G>TSer1512Ile Breast (grandmother age 50)
 204437BRCA15525+25G>AUnknown  
 204437BRCA2654-24G>AUnknown  
 204728BRCA1675T>GSer186AlaLeukemia (father age 63) 
 205336BRCA2654-24G>AUnknownBreast (sister age 47) 
  BRCA21012-26T>GUnknownProstate (father age 69) 
 205733BRCA2451G>CAla75ProBreast (mother age 43)Colon (uncle)

Besides these 9 mutations, 23 (9 BRCA1 and 14 BRCA2) rare, probably disease-associated variants but of so far unknown significance (UV) were detected in 20 patients with no clearly deleterious mutations identified: 11 missense changes, 11 intronic variants and 1 isosemantic single-base substitution. Of 23 variants identified, 19 were distinct, 11 of them being novel. A previously unreported BRCA2 alteration (654-24G>A) was detected in 3 patients, and the BRCA1 101-21insAT variant was observed twice in our series. The remaining US mutations have been previously reported a few times1, 2, 3, 4, 5, 6, 7 in the BIC database, with the exception of the BRCA1 Ser1512Ile and the BRCA2 Thr598Ala, which are registered 50 and 28 times, respectively.

Thus, in total, 29 (21.3%) of 136 patients tested were found to carry an anomaly in the BRCA1 or BRCA2 gene (9 MUT, 6.6%, and 20 UV, 14.7%).

The average age at diagnosis of MUT (36.6 years) and UV (37.9 years) carriers was lower than that in women with no BRCA1/2 alteration identified (39.3 years), though differences in mean values did not reach statistical significance. However, women whose breast cancer was diagnosed before age 40 had an almost 3-fold increased chance of being a MUT or UV carrier compared to women diagnosed in their 40s (OR = 2.8, p < 0.05) (Table II). Family history of breast cancer was registered for nearly half of the patients analyzed and was somewhat more present in BRCA mutation carriers but not highly indicative of BRCA carrier status (Table II). Family history of ovarian cancer was a major predictor of being a MUT carrier, with a highly significant OR of 40.8 (95% CI 4.9–336.2).

Table II. Prevalence of BRCA1 and BRCA2 Alterations in Breast Cancer Cases by Age at Diagnosis and by Family History of Breast and Ovarian Cancers
 NumberBRCA1/BRCA2 carrier statusOR1 (95% CI)OR2 (95% CI)
Noncarriers n (%)UV n (%)MUT n (%)
  • 1

    Calculated for MUT carriers compared to noncarriers.

  • 2

    Calculated for UV carriers compared to noncarriers.

  • 3

    Adjusted for age.

  • 4

    At least 1 first- or second-degree relative of index case affected with breast cancer.

  • 5

    At least 1 first- or second-degree relative of index case affected with ovarian cancer.

All patients136107 (100)20 (100)9 (100)  
Age at diagnosis (years)      
 25–396243 (40.2)12 (60.0)7 (77.8)5.2 (1.0–26.3)2.2 (0.8–5.9)
 40–457464 (59.8)8 (40.0)2 (22.2)11
Family history of breast cancer4      
 No7863 (58.8)12 (60.0)3 (33.3)11
 Yes5844 (41.1)8 (40.0)6 (66.7)2.9 (0.7–12.1)31.0 (0.4–2.5)3
Family history of ovarian cancer5      
 No131105 (98.1)20 (100)6 (66.7)11
 Yes52 (1.9)0 (0)3 (33.3)40.8 (4.9–336.2)3

The expected numbers of cancer cases based on population incidence rates compared to the observed numbers of breast and ovarian cancers in first- and second-degree relatives of index cases are shown in Table III. Fifty-eight women affected with breast cancer compared to an expected number of 40 were identified (SIR = 1.37, 95% CI 1.02–1.72). Breast cancer SIRs were similar for the relatives of MUT and UV carriers and noncarriers. In contrast, 5 relatives of MUT carriers (2 BRCA1 and 3 BRCA2) developed ovarian cancer compared to an expected number of 0.29, while there was only 1 case of ovarian cancer compared to 4.3 cases expected in relatives of noncarriers. Relatives of MUT carriers were estimated to have a 17-fold increased risk of developing ovarian cancer compared to the general population (SIR= 16.9, 95% CI 2.09–31.8).

Table III. Observed and Expected Numbers of Breast and Ovarian Cancer Cases in First-and Second-Degree Female Relatives of Patients with Early-Onset Breast Cancer and SIR
 Observed casesExpected casesSIR (95% CI)
Breast cancer   
 MUT42.11.9 (0.04–3.7)
 US96.51.4 (0.48–2.29)
 Noncarriers4531.41.4 (1.02–1.85)
Ovarian cancer   
 MUT50.2916.9 (2.09–31.8)
 US00.89
 Noncarriers14.30.23 (0.0–0.68)

Leukemia, colon cancer and prostate cancer were the most frequent malignant diseases other than breast cancer and ovarian cancer observed in the relatives of index cases. Only 2 cases of prostate cancer and no leukemia or colon cancer cases were diagnosed in MUT-carrying families. No significant differences in prevalence of colon and prostate cancers were detected among relatives of UV carriers and noncarriers; however, leukemia cases were significantly more frequent among relatives of US carriers. There were 5 leukemia patients compared to 1.4 expected among first- and second-degree relatives of UV carriers, whereas 6 leukemia cases were expected and observed among noncarriers' relatives.

Table IV shows the clinical characteristics of breast cancer patients in relation to their carrier status. The majority of patients had a unilateral ductal breast carcinoma, diagnosed predominantly at stages I and II. BRCA carriers were more likely to have a bilateral cancer (p = 0.06) and to be diagnosed at later stages. None of the 9 MUT carriers was diagnosed at stage 0 or I. No association between mutation status and presence of estrogen receptors or number of affected lymph nodes was observed.

Table IV. Clinical Characteristics of Breast Cancer in Study Subjects
 BRCA1/BRCA2 carrier statusP1
Noncarriers n (%)UV n (%)MUT n (%)
  • 1

    Values correspond to the estimation of difference between MUT carriers and non-carriers.

All patients107 (100)20 (100)9 (100) 
Histologic diagnosis    
 Ductal carcinoma87 (81.3)18 (90.0)8 (88.9) 
 Lobular infiltrant7 (6.7)1 (5)0 (0) 
 Comedocarcinoma5 (4.8)0 (0)0 (0) 
 Inflammatory carcinoma1 (1.0)0 (0)1 (11.1) 
 Others7 (6.7)1 (5)0 (0)0.660
Bilateral tumour    
 No103 (96.3)20 (100)7 (77.8) 
 Yes4 (3.8)0 (0)2 (22.2)0.065
Stage    
 0 or I31 (28.7)5 (27.8)0 (0) 
 II61 (60.4)10 (55.5)6 (66.7) 
 III9 (8.9)2 (11.1)0 (0) 
 IV1 (1.0)1 (5.6)3 (33.3)0.008
 Data missing420 (0) 
Hormonal receptors    
 Yes57 (81.4)14 (87.5)3 (60.0) 
 No13 (18.6)2 (12.5)2 (40.0)0.38
 Data missing3744 
Lymph node status    
 All negative50 (47.2)9 (42.9)5 (55.6) 
 One positive19 (17.9)4 (19.0)2 (22.2) 
 More than one positive37 (34.9)8 (38.1)2 (22.2)0.94

DISCUSSION

We report the prevalence of BRCA1 and BRCA2 germline mutations in a population-based series of women with early-onset invasive breast cancer. Our study includes almost all invasive breast cancer cases diagnosed before the age of 46 years in the Girona and Tarragona regions of northeast Spain during 3 consecutive years. Blood samples were obtained from 154 of 158 patients identified through a cancer registry. Collection of information on family history of cancer and personal risk factors and mutation screening of the entire BRCA1/2 coding regions were performed in a particularly high proportion of incident cases (86%). The inclusion rate and scale of mutation analysis are comparable with those of Loman et al.11 in Sweden. The other studies in this field are characterized by either a much lower inclusion rate, probably leading to distortion between screened and unscreened groups of patients in terms of survival and family history of cancer, or a limited mutation screening protocol, mainly restricted to the testing of several gene regions or certain founder mutations.6, 7, 8, 9, 10

In our set of 136 patients tested, 9 (6.6%) were found to carry deleterious BRCA1 or BRCA2 germline mutations. The relative contribution of BRCA appeared to be very unequal, with the majority (8 of 9) of mutations occurring in BRCA2. This peculiar distribution, unusual in the absence of founder mutations, might be explained by chance overrepresentation of BRCA2 anomalies given the relatively small sample size or may reflect the true distribution. However, previous studies of Spanish high-risk breast cancer families described fairly equal representation of mutations in each gene.13, 15 Our estimates of BRCA1/2 mutation prevalence are comparable with those obtained in other population-based studies for groups of breast cancer patients with similar age at diagnosis. The frequency of BRCA mutations in cases diagnosed before age 40 in our set (11.6%) was somewhat higher than those reported in 3 analogous studies, which varied between 4.6% and 9.9%.6, 10, 11 Compared to Peto et al.8 and the Anglian Breast Cancer Study Group,7 we found quite similar proportions of BRCA mutation carriers in the following age groups: in cases diagnosed before age 36, deleterious mutations were identified in 11% of patients compared to 5.9–12.4% in the British studies, and in cases aged 36–45, mutations were detected in 4.9% of patients compared to 2.4–4.1%.

It is not unlikely that the mutation frequencies reported here are underestimated given that the sensitivity of the heteroduplex and DHPLC mutation detection techniques is not 100% and that certain types of mutation, e.g., large rearrangements, were not sought. This specific class of mutations occupies a nonnegligible place in the BRCA1 mutation spectrum (approx. 10%)15, 16 and is reported to occur in the BRCA2 gene as well.17 In the Netherlands, 2 large founder deletions in the BRCA1 gene accounted for 36% of all BRCA1 mutations detected in breast cancer families.18 However, screening of these alterations, being particularly laborious using existing techniques, is hardly feasible in a large number of samples.

The underestimation of mutation prevalence could also be related to the difficulty of establishing if some of the rare BRCA missense and intronic variants detected are pathogenic. Indeed, in our set, 20 patients carried rare variants whose functional consequences are unknown. This type of genetic lesion was underrepresented in many publications due either to nonreporting because of problematic interpretation of these BCRA1/2 sequence changes or to choice of technique used for mutation screening, permitting detection only of mutations that create premature stop codons (e.g., protein truncation test). Therefore, it is difficult to estimate the fraction of UVs in the BRCA1/2 mutation spectrum, though, according to several large comprehensive studies, USs might be as numerous as truncating mutations.8, 19

Certain genetic/epidemiologic features could be evocative of noncausality of UVs, e.g., the absence of familial cosegregation of the variant and breast cancer, or an association with a clearly deleterious mutation. Several assays have been established in which some specific BRCA1/2 functions can be tested, but these tests are very laborious and to date have been used only to characterize missense substitutions selected on the basis of their location in the relevant functional domains.20, 21, 22, 23 Thus, there is a clear need to develop tests that permit discrimination between disease-associated mutations and neutral polymorphisms in the BRCA1 and BRCA2 genes and are relatively easy to perform and applicable to most of the BRCA1/2 USs. The approaches based on analysis of interspecific sequence conservation appear to be very promising to indicate functionally important amino acid sites and missense changes likely to disrupt gene function.24, 25, 26

The majority of rare missense changes identified in our samples affected residues identical in human and rodent gene products, some missense alterations resulting in significant changes of physicochemical properties of the amino acid concerned. One BRCA2 substitution (Thr1011Arg) was located in the core motif of the first BRC repeat, a region shown to contribute to the interaction between BRCA2 and RAD51.27

Since more than one-third of UVs identified in our series correspond to variants in noncoding regions, it is important to examine their consequences on the stability and splicing of variant transcripts. However, this approach would be expected to be also relevant in exonic base substitutions. Mutations within exonic splicing enhancers can lead to a variety of effects, including exon skipping, intron retention and activation of cryptic splice sites.28, 29, 30 Thus, certain nucleotide changes might be wrongly considered missense or synonymous variants, while, in reality, they affect transcript properties. This underscores the importance of carefully scrutinizing the effect of any base substitution before inferring its potential consequences or eliminating its pathogenic candidacy.

Interestingly, regarding age at diagnosis of breast cancer, UV carriers formed a group intermediate between that of MUT carriers and noncarriers. UV carriers together with MUT carriers were more likely to have early-onset breast cancer (< 40 years of age) compared to patients with no BRCA1/2 alteration identified (p < 0.05) (Table II). This observation suggests that certain proportion of rare variants detected in our study might correspond to the BRCA germline alterations associated with increased breast cancer risk.

Three deleterious mutations as well as 14 rare BRCA1 and BRCA2 variants have never been reported before or have been detected in Spanish samples only and may be country-specific. Similar to previous studies performed on Spanish families,13, 14, 31, 32, 33, 34 no strong mutation founder effect was detected, the majority of the alterations being distinct, with few observed 2 or 3 times.

Data on family history of cancer suggested that genetic factors might be significant in a proportion of early-onset breast cancers, with 37% of cases reporting at least one first- or second-degree relative affected with breast cancer. Indeed, this estimate is double the 18% reported in our study including breast cancer cases diagnosed at any age in the same geographic area.35 Fifty-eight first- and second-degree female relatives of cases in this series had breast cancer compared to an expected number of 40, based on regional cancer incidence rates. However, a small portion of the familial risk was attributable to mutations in the BRCA1 and BRCA2 genes.

Although family history of breast cancer alone was not highly indicative of BRCA carrier status, noticeable familial clustering, particularly containing ovarian cancer cases, was quite predictive of BRCA mutation. Four MUT and 2 US carriers were identified among 14 patients having at least 1 first- or second-degree relative affected with ovarian cancer diagnosed at any age or breast cancer diagnosed before age 46.

Even though in family-based studies BRCA2-associated ovarian cancer risk was reported to be much lower than that associated with BRCA1 mutations,36 in our series high ovarian cancer risk was observed in relatives of both BRCA1 and BRCA2 carriers. Of 4 patients who reported at least 1 case of ovarian cancer among first- and second-degree relatives, 2 carried a mutation in BRCA2 and 1 in BRCA1.

Family history of breast cancer and ovarian cancer was, however, absent or not suggestive of hereditary condition in one-fourth of MUT carriers and in more than half of UV carriers. This suggests that BRCA screening policies based on family history would miss a considerable proportion of mutation carriers.

Our results indicate that mutation carriers were more likely to be diagnosed at an advanced stage and to have a bilateral tumor. These observations are consistent with the data of the Breast Cancer Linkage Consortium study describing higher overall grade of breast cancers in carriers of BRCA1 and BRCA2 mutations compared to sporadic breast cancers.37 Although contradictory results concerning the association between the presence of bilateral cancer and detection of a BRCA mutation could be found in the literature, our data are in line with those of several studies reporting that BRCA carriers affected with early-onset breast cancer are prone to develop a second breast tumor.11, 38, 39, 40

In conclusion, we detected alterations in BRCA1 and BRCA2 in >20% of early-onset breast cancer cases; in one-third of these, alterations were clearly deleterious mutations. It is likely that some of the rare alterations of unknown significance detected may have a functional relationship with breast cancer development, but this remains to be further explored.

Acknowledgements

We thank Ms. J. Sola and Dr. K. Langorh for statistical analysis and Dr. D. Goldgar and Dr. G. Capella for helpful comments. S. de Sanjose was a visiting scientist at the National Cancer Institute, USA, during 2002–3.

Ancillary