Genetic Testing in an Ethnically Diverse Cohort of High-Risk Women:  A Comparative Analysis of BRCA1 and BRCA2 Mutations in American Families of European and African Ancestry FREE

Mammography beginning at age 40 years has been the mainstay of screening for breast cancer in the United States for many years. Given the recent advances in our understanding of breast cancer risk factors and the promise of prevention, women from high-risk families are encouraged to consider genetic testing to quantify their risk. High-risk women are advised to begin intensive surveillance at either 10 years younger than the earliest case of breast cancer in the family or at age 25 years, consider risk-reducing prophylactic surgery, and undergo screening magnetic resonance imaging of the breast.^1- 4

An estimated 5% to 10% of breast cancer cases arise in individuals with inherited mutations in autosomal dominant, highly penetrant breast cancer susceptibility genes.⁵ Although the proportion of cases due to a mutation in each of these genes remains to be determined, 7 genes—BRCA1, BRCA2, TP53, PTEN, CHK2, ATM, and STK11—are known to contribute to breast cancer susceptibility.^6- 15 Germline mutations in BRCA1 and BRCA2 are by far the most common and account for 80% to 90% of families containing multiple cases of breast and ovarian cancer.¹⁶

The proportion of breast cancer attributed to mutations in BRCA1 or BRCA2 has varied widely among different studies and different ethnic groups. Mutations in BRCA1 are the most common in one report of non-Jewish Russian families,¹⁷ while BRCA2 mutations are the most common in Iceland, where a single common mutation explains the majority of inherited cancers of the breast and ovary.¹⁸ The proportion of high-risk US families attributable to BRCA1 mutations has been reported to vary from 16% to 39%, while BRCA2 mutations account for 19% to 25% of families.^19- 23 It is not known, however, if the proportions vary among blacks and whites and if the spectrum of mutations reflects those of founder ancestors, as US families are almost exclusively immigrants. Most studies in the United States and Canada have included a large percentage of women of European and Ashkenazi Jewish ancestry. Little information exists about genetic testing in other ethnic minorities. Of note, one of the largest ethnic minorities in the United States, the African American population, remains understudied, despite having a proportionately higher incidence of early-onset breast cancer.

As genetic testing for breast cancer susceptibility has moved from the research setting into the clinical arena, and with direct marketing to consumers leading to increased awareness, more individuals are demanding genetic testing. Despite the drawbacks of genetic testing, including high expense and inability to detect all mutations, several professional groups have endorsed genetic counseling and testing for high-risk women because of the potential benefits of risk-reducing prophylactic surgery and intense surveillance.²⁴ Unfortunately, minority populations remain underrepresented in genetic studies. A recent study has identified large racial disparities in the use of genetic counseling and BRCA1 and BRCA2 testing.^25- 26 Many of the risk-assessment tools used in cancer risk clinics, such as the BRCAPRO statistical model, were developed based on mutation rates observed primarily in Ashkenazi Jewish and other white women of European descent. Thus, the performance of BRCAPRO and other models used in cancer risk clinics needs to be validated for use in African American and other minority groups. Using a unique, ethnically diverse cohort of high-risk families, we sought to characterize the clinical predictors of BRCA1 and BRCA2 mutations among high-risk individuals of European and African ancestry, highlighting the similarities and differences. To our knowledge, this is the first such comprehensive comparative analysis of families evaluated in hospital-based cancer risk clinics.

Families were primarily identified through individuals who presented to the Cancer Risk Clinic at the University of Chicago between February 1992 and May 2003. Families who presented to high-risk clinics at the Mayo Clinic, Rush University Medical Center, and the University of California-San Francisco and who participated in Myriad Genetics Laboratory beta testing for BRCA1 and BRCA2 mutation conducted through the University of Chicago were also included in this analysis; recruitment from these sites occurred between October 1996 and March 1997. All participants were informed that their DNA samples would be used for mutation analysis, were offered genetic counseling under protocols approved by the institutional review board at each institution, and provided written informed consent.

Families were selected for this analysis if they fulfilled the inclusion criteria of having 2 or more cases of breast cancer, ovarian cancer, or both among first- and second-degree relatives. Individuals who had been tested for a BRCA1 or BRCA2 mutation before, or those with a family member who had previously been tested, were excluded from this study to limit bias for mutation positivity.

Individuals were categorized based on self-reported race/ethnicity. Individuals were categorized as non-Hispanic, non-Jewish white (white); Ashkenazi Jewish (Jewish); African American; Hispanic; or Asian. Because of the unique spectrum and frequency of BRCA1 and BRCA2 mutations that occur in Ashkenazi Jewish individuals, these individuals were analyzed separately from other whites when making comparisons between ethnic groups.

Because the pedigrees of these families were consistent with a hereditary breast and ovarian cancer syndrome, individuals were tested for mutations in BRCA1 and BRCA2. Approximately 80% of the samples were analyzed at Myriad Genetics Laboratory using direct DNA sequencing as previously described.²⁷ The remaining 20% were screened by either single-strand conformation polymorphism (SSCP) or denaturing high-performance liquid chromatography (DHPLC) analysis, followed by sequencing of those with variant results as previously described.²⁸ Individuals who identified themselves as being of Ashkenazi Jewish heritage were initially screened for the 3 common founder mutations. Complete sequencing was performed only if the initial screening did not detect one of these founder mutations. Naming and interpretation of sequence analysis were performed as previously described.²⁷ All patients were classified as having a deleterious mutation, a variant of undetermined significance, or no mutation.

Family history information (including incidence of breast, ovarian, and other cancers; age at diagnosis; and relationship to the proband) was collected and recorded by a genetic counselor. The predicted likelihood of carrying either a BRCA1 or BRCA2 mutation was generated for each individual using BRCAPRO version 3.3.1.²⁹

The Fisher exact test was used to compare the distributions of whites and African Americans across the 3 mutation classifications (deleterious BRCA1 or BRCA2 mutation, sequence variations, or no mutation). Logistic regression models were used to examine the association between the likelihood of having a BRCA1 or BRCA2 mutation and several characteristics of family history.³⁰ A hierarchical approach was used in which we first modeled the likelihood of having either mutation and then modeled the likelihood of having a BRCA2 mutation conditional on having 1 of the 2 mutations.³¹ In both instances we started with models including the number of breast cancer cases and the number of ovarian cancer cases in first- and second-degree relatives, together with indicator variables distinguishing African Americans, Ashkenazi Jews, and whites. We then added additional variables one by one (mean age at diagnosis of breast cancer, number of cases of bilateral breast cancer, and number of family members with both breast and ovarian cancer), noting both the coefficient for the added variable and its effect on the coefficients and standard errors for variables already in the model. In the case of the model predicting the probability of either mutation, we also examined interaction terms between the indicator variable for African Americans and the family history variables (there were not sufficient data to do this for the model distinguishing between BRCA1 and BRCA2). For all models, the linear specification for the continuous covariates was checked by using smoothed plots of the partial residuals and by introducing quadratic terms. The results of these analyses are reported in terms of odds ratios (ie, the multiplicative change in the odds of the outcome associated with a 1-unit change in the covariate), together with 95% confidence intervals and P values (both based on Wald tests) for testing the null hypothesis that the true odds ratio is 1.

Predictions of the likelihood of either mutation generated using the BRCAPRO software were compared with the observed outcomes for African Americans and non–African Americans separately.^32- 33 Families were divided into quartiles, categories that each contained one fourth of the sample ordered by predicted carrier probability. For each quartile, we computed both the expected probability of a mutation (ie, the mean of the BRCAPRO predictions for that quartile) and the observed incidence. In addition, we computed the area under the receiver operating characteristic curve (AUROC) via the standard nonparametric method. Using each value of the BRCAPRO predictions as a classification cutpoint, we plotted the curve described by sensitivity vs 1 − specificity. The AUROC was then computed using the trapezoidal rule, and the binomial distribution was used to generate an exact 95% confidence interval.

All computations were performed using STATA release 8.2.³⁴ All P values reported are 2-sided.

One hundred sixty families who sought risk counseling at the University of Chicago between January 1992 and December 2003 met the eligibility criteria for this analysis. Five families were excluded because of prior testing, and 38 (11 African American and 27 non–African American) were not tested. A total of 117 of the 160 families (73%) from the University of Chicago who fulfilled inclusion criteria were included in this analysis. In addition, 16 families from the Mayo Clinic, 14 from Rush University Medical Center, and 8 from the University of California-San Francisco who participated in Myriad Genetics laboratory BRCA1/BRCA2 beta testing through the University of Chicago were included in this analysis. Of the 155 high-risk families, 78 (50.3%) were white; 43 (27.7%), African American; 29 (18.7%), Ashkenazi Jewish; 3 (1.9%), Hispanic; and 2 (1.3%), Asian. Complete pedigree analysis revealed no intermarriage among first- and second-degree relatives in any of the families. We did not observe deleterious mutations in the 5 families of Hispanic or Asian descent, and these families were excluded from further analysis because of the small numbers. Non–African American families were more likely to have ovarian cancer and breast cancer, while African American families were more likely to have only breast cancer. However, bilateral breast cancer was more common among non–African American families (Table 1).

Table 2 shows the prevalence of BRCA1 and BRCA2 mutations and sequence variations in our sample. African Americans had a higher rate of sequence variations but a lower rate of confirmed deleterious mutations in comparison with whites (44.2% vs 11.5% and 27.9% vs 46.2%, respectively; P<.001 for overall comparison). In contrast, Ashkenazi Jewish families had higher rates of deleterious mutations than other white families (41.4% vs 30.8% for BRCA1 and 27.6% vs 15.4% for BRCA2).

The spectrum of mutations observed in our cohort reflected the diverse ethnic origins of our study population. In the Ashkenazi Jewish families, 3 common founder mutations—185delAG, 5385insC, and 6174delT—accounted for 85% of the 20 mutations detected. One individual of Ashkenazi Jewish ancestry had a C61G mutation, a BRCA1 mutation that has been reported as a founder mutation in Poland.³⁵ Conversely, only 1 person of non-Jewish descent was found to have a mutation commonly seen in Ashkenazi Jewish individuals, the 185delAG mutation (see Table 3). However, the haplotype for this mutation was not the common Jewish haplotype. In the African American families, we observed 1 African founder mutation (943ins10),³⁶ and 2 recurrent mutations (1832del5 and 5296del4) that we have previously reported.³⁷ A 5950delCT mutation was observed in 3 families of German ancestry, and a C61G mutation was detected in 2 white families of German or Austrian origin. One family of Scottish ancestry had a 2800delAA mutation, which has been observed as a founder mutation in Scotland.³⁸ As shown in Table 4, 7 recurrent mutations account for 44.1% of the 68 deleterious mutations detected. A large percentage of the mutations represent founder mutations, reflecting the racial and ethnic ancestries of the families.

Among all families combined, the likelihood of having either mutation was strongly associated with several family history characteristics (Table 5). For example, each additional case of breast cancer in first- or second-degree relatives was associated with a 62% increase in the odds of a mutation (P = .002), while each additional case of ovarian cancer was associated with a 146% increase (P = .006). In addition, an increase of just 1 year in the mean age at breast cancer diagnosis was associated with a 10% reduction in the odds of having a mutation (P<.001). When added to the model, neither the number of family members with both breast and ovarian cancer nor the number with bilateral breast cancer yielded a statistically significant effect (P = .31 and P = .35, respectively), and the coefficients for the other variables remained largely unchanged.

Despite controlling for differences in family structure, African Americans had nearly half the odds of having a deleterious mutation as whites (odds ratio [OR], 0.52; 95% confidence interval [CI], 0.21-1.31), although this difference was not statistically significant (P = .17). Ashkenazi Jews had higher odds of a deleterious mutation than whites (OR, 5.09; 95% CI, 1.76-14.78; P = .003) (Table 5). When the model was estimated among only the African American families, both the number of breast cancer cases and the mean age at diagnosis showed effects similar to those shown for the whole sample, but the estimated effect of the number of ovarian cancer cases was much smaller and no longer statistically significant. Adding interaction terms between African American ethnicity and each of these covariates to the model of all families combined yielded values of P = .31 for number of breast cancer cases, P = .17 for number of ovarian cancer cases, and P = .45 for mean age at breast cancer diagnosis. Thus, the data provided no statistical evidence of a difference between the African American families and the other families in the effects of these covariates.

Table 6 shows results from a logistic regression predicting the likelihood of having a BRCA2 mutation conditional on having either mutation. Each additional case of ovarian cancer was associated with an 85% reduction in the odds of a BRCA2 mutation (OR, 0.15; 95% CI, 0.04-0.61; P = .008), while each additional year in the age at breast cancer diagnosis was associated with a 21% increase in the odds of having a BRCA2 mutation (OR, 1.21; 95% CI, 1.08-1.35; P = .001). No statistically significant differences were observed between African American, Ashkenazi Jewish, and white families in the likelihood of a BRCA2 mutation.

Overall, the mean number of breast cancer cases per family differed between the ethnic groups (African American families vs white and Jewish families), while the mean age at diagnosis did not. The mean number of breast cancer cases per family was 3.4 in African Americans and 4.4 in non–African Americans (P = .01), while the mean age at diagnosis of breast cancer was 46.2 years and 46.7 years, respectively (P = .79). To further characterize the 43 African American families, we subdivided them into 3 groups: those with deleterious mutations (n = 12), those with variants of undetermined significance (n = 19), and those with no identifiable mutation (n = 12) in BRCA1 or BRCA2. The mean number of cases of breast cancer in the family was 3.0, 3.2, and 4.2, respectively, while the mean age at breast cancer diagnosis in these groups was 43.6, 50.5, and 42.0 years, respectively.

The BRCAPRO statistical model overestimated the probability of a mutation among both the African American and non–African American families (Table 7). A closer inspection reveals that BRCAPRO underestimates risk at the lowest levels, while overestimating it at the higher levels. For example, BRCAPRO predicted that 38% of the African Americans would have a mutation compared with the 28% observed. However, among those in the lowest probability quartile, BRCAPRO predicted only a 2% mutation rate compared with the 18% observed; in the highest quartile, BRCAPRO predicted a 96% mutation rate as compared with the 80% observed. In terms of discriminating between those with a mutation and those without, BRCAPRO performed as well among the African American families as it did among the others. The AUROC (Figure) was 0.77 (95% CI, 0.61-0.88) for African Americans and 0.70 (95% CI, 0.60-0.79) for the white and Ashkenazi Jewish cohorts combined.

BRCA1 and BRCA2 mutations do occur with appreciable frequency in high-risk families of African ancestry, with 28% testing positive for a deleterious mutation in 1 of these genes, a rate consistent with other clinic-based studies in the United States. Shih et al³⁹ reported that 22% of families seeking genetic testing at the University of Michigan and the University of Pennsylvania between 1992 and 1995 had a deleterious BRCA1 or BRCA2 mutation; Frank et al²² found that 1720 (17.2%) of the first 10 000 individuals tested at Myriad Genetics Laboratory had a deleterious mutation in 1 of these 2 genes. While the frequency of confirmed deleterious mutations in the African American cohort was lower than that observed in whites and Ashkenazi Jews, these differences were not statistically significant when controlling for the covariates measuring family history. Because our study was not powered to detect significant differences in mutation frequency between these groups, larger studies are needed to further investigate this question. Nonetheless, all families in this ethnically diverse cohort benefited from the comprehensive risk assessment and counseling that accompanied genetic testing.

Our data underscore the need for larger studies among minority populations in the United States. There are documented racial/ethnic disparities in patterns of referral to cancer risk clinics, as demonstrated in the recent article by Armstrong et al.²⁵ In that study, African American women with a family history of breast or ovarian cancer were significantly less likely to undergo genetic counseling and BRCA1/BRCA2 testing than were white women with a family history of breast or ovarian cancer (odds ratio, 0.22; 95% CI, 0.12-0.40).

Differences in patterns of referral based on race/ethnicity might contribute to lower rates of deleterious mutations in the African Americans in our study. The African American families in this study had a lower incidence of ovarian cancer than the non–African American families. While this might suggest that the African American families referred for testing might be at lower risk for having a BRCA1 or BRCA2 mutation, even among the families with deleterious mutations, African American families had a lower mean number of ovarian cancer cases in first- and second-degree relatives compared with white and Jewish families (0.29 vs 0.72). In general, African Americans have a lower incidence of ovarian cancer compared with whites,⁴⁰ and our data suggest that even in the setting of BRCA1 or BRCA2 mutations, they have a lower rate as well. While we cannot exclude differences in patterns of referral as the reason for the lower rates of ovarian cancer seen in these high-risk African American women, differences in disease modifiers across ethnic groups could also explain differences in ovarian cancer rates. Further study of these populations matched for family history will best address this issue.

While rates of deleterious mutations were lower among African Americans, polymorphisms and variants of unknown significance were much more common than in whites (44.2% vs 11.5%). A comparison with rates in Ashkenazi Jews could not be made, as those who were found to have 1 of the 3 Ashkenazi founder mutations did not have complete sequencing performed. The role that polymorphisms and unclassified variants of BRCA1 and BRCA2 play in breast cancer etiology remains ill-defined. Misclassification of deleterious mutations as unclassified variants due to paucity of sequence information in the public database⁴¹ could contribute to the lower rates of deleterious mutations identified among these high-risk African American families. As demonstrated in Table 8, these families have multiple cases of breast cancer. Among African Americans, those with no identifiable mutation in either gene had a mean age at breast cancer diagnosis comparable with that in those families with deleterious mutations (43.6 years and 42.0 years, respectively). There are several possible explanations for this observation. One is that while some ethnic groups have high rates of BRCA1 or BRCA2 mutations, African Americans have a higher rate of mutation in another, as-yet unidentified, breast cancer susceptibility gene. Another possibility is that African Americans have large rearrangements or deletions, such as those that have been described in other populations.^42- 44 These alterations would not have been detected by the screening methods used in our study. Furthermore, one half of the African Americans included in our study were screened for mutations by SSCP or DHPLC analysis. While DHPLC has a sensitivity comparable to that of direct DNA sequencing, SSCP has a sensitivity of 72%⁴⁵; thus, we cannot exclude the possibility that some mutations might have been missed on the few samples screened in this manner. For the 19 African American families with sequence variations in BRCA1 and BRCA2, the mean age at diagnosis of breast cancer is older (50.5 years) and closer to the mean age at diagnosis of sporadic breast cancer. While this might suggest that these variants are not pathogenic, it is possible that they represent hypomorphic alleles that contribute to increased risk but not to the same degree as protein-truncating mutations. In a recent study of Nigerian breast cancer patients younger than 40 years and unselected for family history, we observed BRCA1 or BRCA2 sequence variations in 29 of 39 individuals (74%), with 69% having sequence variations in BRCA2.⁴⁶ Unfortunately, there is a paucity of data about the genetic diversity in BRCA1 and BRCA2 in African populations or the functional consequences of these variants. While some of these variants might represent benign alterations, it is possible that they do contribute to disease, thus adding to the complexity of genetic counseling in populations of African ancestry.

The spectrum of mutations we observed in African Americans is vastly different from what we observed in individuals of European descent. The 943ins10 founder mutation we detected in an African American woman has been traced back to its ancient origin in the Ivory Coast in West Africa.³⁶ The 2 recurrent mutations that we detected in African American families are also unique to this ethnic group.³⁷ As expected, the 3 common Ashkenazi Jewish mutations—185delAG, 5385insC, and 6174delT—with the exception of 1 individual, were only detected in women of Ashkenazi Jewish heritage. Because of the small numbers of African American mutation carriers in our cohort and the significant genetic diversity, it was not possible to identify a panel of mutations that occur with a high frequency in African Americans. However, we have now evaluated more than 200 individuals of African ancestry with early-onset breast cancer and have yet to come across additional recurrent mutations.^28,37,46 Given the genetic diversity of the African American population, it is unlikely that screening for a panel of founder mutations will be as effective in this population as is the case for the Ashkenazim.

Because BRCAPRO is widely used to aid in the genetic counseling process, it is important to evaluate the performance of this statistical model in an ethnically diverse cohort of individuals. Interestingly, the performance in our study was similar to that reported in other recent studies. Euhus et al,⁴⁷ based on a sample of 148 predominantly white and Jewish families, estimated the AUROC for BRCAPRO to be 0.71, nearly identical to the value observed here. Furthermore, Berry et al⁴⁸ observed that BRCAPRO both underestimated the likelihood of mutation at the lowest levels of risk and overestimated the likelihood of mutation at the highest levels of risk; again, just as we have observed here. The BRCAPRO model was designed to predict the likelihood of carrying a BRCA1 or BRCA2 mutation, as opposed to the likelihood of such a mutation being detected, and it is conceivable that our mutation detection methods are not sensitive enough to detect all potentially deleterious mutations. Furthermore, the BRCAPRO model only includes first- and second-degree relatives in risk prediction, which frequently leads to an underestimation of risk, particularly in cases in which cancer predisposition is inherited paternally. While BRCAPRO has limitations, our data suggest that it performs as well among African American families as it does among white and Jewish families, making it a useful clinical risk assessment tool in African American families.

Ten years after BRCA1 and BRCA2 were first identified as major breast cancer susceptibility genes, the spectrum of mutations and modifiers of risk among many racial/ethnic minorities remain undefined. Our data support the use of personal and family history of breast cancer, ovarian cancer, or both in making clinical decisions and identifying individuals who are likely to benefit from genetic counseling. Certain family characteristics—most notably the number of breast cancer cases among first- and second-degree relatives and the mean age at diagnosis of breast cancer—are associated with the likelihood of carrying a deleterious mutation among African Americans, as has previously been observed in white and Ashkenazi Jewish families.

Our observations underscore the need for large, collaborative studies to systematically validate the role of genetic testing, the use of risk prediction models, and the role of risk-reducing strategies in improving health outcomes for individuals of African ancestry.