Risk of Breast Cancer Among Carriers of Pathogenic Variants in Breast Cancer Predisposition Genes Varies by Polygenic Risk Score

Abstract

PURPOSE

This study assessed the joint association of pathogenic variants (PVs) in breast cancer (BC) predisposition genes and polygenic risk scores (PRS) with BC in the general population.

METHODS

A total of 26,798 non-Hispanic white BC cases and 26,127 controls from predominately population-based studies in the Cancer Risk Estimates Related to Susceptibility consortium were evaluated for PVs in BRCA1, BRCA2, ATM, CHEK2, PALB2, BARD1, BRIP1, CDH1, and NF1. PRS based on 105 common variants were created using effect estimates from BC genome-wide association studies; the performance of an overall BC PRS and estrogen receptor–specific PRS were evaluated. The odds of BC based on the PVs and PRS were estimated using penalized logistic regression. The results were combined with age-specific incidence rates to estimate 5-year and lifetime absolute risks of BC across percentiles of PRS by PV status and first-degree family history of BC.

RESULTS

The estimated lifetime risks of BC among general-population noncarriers, based on 10th and 90th percentiles of PRS, were 9.1%-23.9% and 6.7%-18.2% for women with or without first-degree relatives with BC, respectively. Taking PRS into account, more than 95% of BRCA1, BRCA2, and PALB2 carriers had > 20% lifetime risks of BC, whereas, respectively, 52.5% and 69.7% of ATM and CHEK2 carriers without first-degree relatives with BC, and 78.8% and 89.9% of those with a first-degree relative with BC had > 20% risk.

CONCLUSION

PRS facilitates personalization of BC risk among carriers of PVs in predisposition genes. Incorporating PRS into BC risk estimation may help identify > 30% of CHEK2 and nearly half of ATM carriers below the 20% lifetime risk threshold, suggesting the addition of PRS may prevent overscreening and enable more personalized risk management approaches.

INTRODUCTION

Breast cancer (BC) is the most common cancer among women in the United States.¹ Screening of BC improves survival rates but also may result in false positives and overtreatment of indolent disease.² Hence, better risk stratification is needed to provide a more personalized approach for women to benefit from modified screening strategies including age of screening initiation.^3,4

CONTEXT

• Key Objective

• Does polygenic risk score (PRS) refine breast cancer (BC) risk in women from the general population with and without pathogenic variants (PVs) in BC predisposition genes?

• Knowledge Generated

• PRS showed particular importance of risk stratification among carriers of PVs in moderate penetrance genes such as CHEK2 and ATM. The risks associated with change in PRS were smaller among carriers of BRCA1 or BRCA2 but no other effect modification by PRS was observed for carriers of PVs in the other genes tested.

• Relevance

• PRS yielded a meaningful risk gradient among both carriers and noncarriers of PVs in BC predisposition genes and may be particularly important for women with PVs in ATM and CHEK2. The incorporation of PRS into risk prediction models may help to determine the potential benefit of breast magnetic resonance imaging and the age of BC screening initiation.

Pathogenic variants (PVs) detected in clinical multigene cancer predisposition panels are increasingly used to counsel women regarding their risk for developing BC, particularly for women with family history of cancer. Germline genetic testing has also evolved substantially, enabling rapid and simultaneous detection of genetic variation in many genes.⁵ However, our understanding of how to translate the genetic information into evidence-based clinical decisions needs improvement. PVs in high-penetrance genes such as BRCA1 and BRCA2 are well studied, but the added benefit of tailoring BC prevention and surveillance strategies based on carrier status for moderate penetrance genes (eg, CHEK2 and ATM) is less clear.

Common variants (single-nucleotide polymorphisms [SNPs]) found through genome-wide association studies (GWAS) are also associated with elevated BC risk.⁶ The risk conferred by each individual SNP is small but the combined effect of multiple SNPs as a polygenic risk score (PRS) can have an important impact on risk prediction.^3,4,7,8 In a recent PRS study from the Breast Cancer Association Consortium, one standard deviation (SD) change in a PRS was associated with a 61% increase in BC risk (odds ratio [OR] = 1.61; 95% CI, 1.57 to 1.65) and the lifetime risk for women in the top percentile of the PRS was 32.6% compared with 9% in the lowest 20th percentile.⁹

Jointly modeling PVs in BC predisposition genes and established BC SNPs in the same population allows us to assess whether the effect of PRS differs between carriers and noncarriers of PVs, and whether PRS can further stratify individual BC risk among carriers of PVs. Previous work found that PRS are predictive of BC risk among carriers of PVs recruited from cancer genetics clinics or referred for clinical genetic testing.^10-13 However, there are few studies evaluating the effect of a PRS among carriers of PVs in genes other than BRCA1/BRCA2^10,12 and CHEK2,¹¹ or in samples drawn from population-based studies.

In this study, we evaluated the combined effect of PRS, PVs in nine established BC predisposition genes, and family history of BC (among first-degree relatives) using 26,798 cases and 26,127 controls in the Cancer Risk Estimates Related to Susceptibility consortium.¹⁴ We assessed the performance of an overall BC PRS as well as estrogen receptor (ER) subtype-specific PRS. We also estimated 5-year and lifetime absolute risks of developing BC across percentiles of PRS for carriers and noncarriers of PVs, respectively, and for women with and without family history of BC.

METHODS

Study Population

The study consists of non-Hispanic white women from nine prospective cohorts, two population-based case-control studies, and one clinic-based case-control study. The nine cohort studies are Cancer Prevention Study (CPS) II,¹⁵ CPS 3,¹⁶ California Teachers Study,¹⁷ Multiethnic Cohort,¹⁸ Mayo Mammography Health Study,¹⁹ Nurses' Health Study (NHS),²⁰ NHSII,²¹ Women's Health Initiative,²² and the Sister Study.²³ The two population-based case-control studies are the Women's Circle of Health Study²⁴ and the Wisconsin Women's Health Study.²⁵ The Mayo Clinic Breast Cancer Study²⁶ is a clinic-based case-control study (cases were diagnosed at the Mayo Clinic, and controls were recruited from women attending the Department of Internal Medicine at the Mayo Clinic).

Cases (women with BC) were either identified by self-report and confirmed by medical record or identified through cancer registry linkage. Controls (women without BC) from the CPSII, California Teachers Study, Multiethnic Cohort, Mayo Clinic Breast Cancer Study, NHS, NHSII, Women's Circle of Health Study, Women's Health Initiative, and Wisconsin Women's Health Study were matched to cases by age. CPS3, Sister Study, and Mayo Mammography Health Study used a case-cohort design; members of the reference subcohort who were free of breast cancer at baseline and not diagnosed with BC during follow-up were used as controls. The number of cases and controls by age group and family history of first-degree relatives is shown in Table 1. All participants provided informed consent for research, and detailed description of each participating study is in Data Supplement 1 (online only). This study was approved by the Mayo Clinic Institutional Review Board.

TABLE 1.

Number of Cases and Controls and Distribution by Participating Studies, Age Groups, and Family History of Breast Cancer (First-Degree Relatives) in Cancer Risk Estimates Related to Susceptibility Consortium

Breast Cancer Predisposition Genes and PRS

Nine validated BC predisposition genes were evaluated: BRCA1, BRCA2, ATM, CHEK2, PALB2, BARD1, BRIP1, CDH1, and NF1.^27-30 A total of 105 independent (r² < 0.2) common variants were used to construct the PRS (Data Supplement 2, online only). The overall BC PRS was calculated as a weighted sum of risk alleles, using effect estimates (per allele log ORs) from the largest published BC GWAS as weights.³¹ To construct a PRS specific for both ER-negative and ER-positive BC, we used a hybrid method to obtain the weights, in which effect estimates from a GWAS of ER-negative or ER-positive BC were used to weight an SNP if the P value from a case-only heterogeneity test (ER-positive v ER-negative disease) was < .05 and overall BC effect estimates were used as weights otherwise. Both the overall BC PRS and ER-subtype specific PRS were standardized to a mean of 0 and an SD of 1. Detailed description of the gene sequencing and PRS calculation method can be found in Data Supplement 1.

Model Fitting

We fit a baseline model using logistic regression, with overall BC as the outcome, and the following explanatory variables: study; indicator variables denoting carriage of a PV for each gene; PRS as a continuous variable; age in years in five categories (age ≤ 40, 41-50, 51-60, 61-70, and > 70 years); an indicator variable for family history of BC in first-degree relatives (yes or no); and product interaction terms between ordinally coded age categories and PV carrier status of BRCA1, BRCA2, and PRS. Because of established modification of the PRS effect by both BRCA1 and BRCA2 genes in recent publications,^12,13 we also included an interaction term between PRS and carriers of PV in either BRCA1 or BRCA2 in our model. Age was defined as the age at diagnosis for cases, age at baseline for controls in case-cohort studies, and age at matching for controls in case-control studies. To account for differences in age-specific case:control sampling fraction between case-control and case-cohort studies, we included interaction terms between age categories and study design (case-cohort v nested case-control v case-control studies). Family history was defined as having at least one first-degree relative (including mother, sisters, daughters, and father) diagnosed with BC. Missing values for age (0.9% missing) and family history (3.3% missing) were replaced using conditional draw imputation as implemented in the MICE R package.³² We included both invasive (n = 21,975) and in situ (n = 4,125) BC cases (analysis restricting to only invasive cases found no appreciable changes to results). We also performed a sensitivity analysis restricting to only prospective studies. In addition, we tested whether the relative risk associated with PRS differed between noncarriers and carriers of PVs in any genes considered individually or in aggregate (Data Supplement 2).

To assess whether allowing the effect of the PRS to change by PV status and family history improved the discrimination ability of our model, we performed L₁ penalized logistic regression using the glmnet R package.³³ All covariates in the baseline model were preselected for inclusion. Additional covariates included all the other possible interactions between carrier status in each predisposition gene, age, family history, and PRS. The final model was chosen by 10-fold cross-validation maximizing the area under the curve as a function of the L₁ penalty. We used the same modeling process to fit an ER-subtype–specific PRS, using ER-subtype–specific BC versus women without BC as outcome.

The above models were combined with average age-specific incidence rates from SEER to calculate 5-year and lifetime absolute risks (by age 80 years) of developing BC for carriers and noncarriers of PVs. Pointwise absolute risk confidence intervals were calculated using 10,000 parametric bootstrap samples. Detailed methodologic descriptions can be found in Data Supplement 1.

RESULTS

According to our best fitting model, a 1-SD difference in the PRS was associated with a 1.63-fold change (95% CI, 1.55 to 1.71) in the odds of overall BC among noncarriers before age 40 years (Tables 2 and 3). This effect declined with increasing age: the OR associated with a 1-SD change in PRS decreased to 1.46 (95% CI, 1.42 to 1.51) after age 70 years. Consistent with previous findings,^12,13 the OR associated with 1-SD change of PRS decreased among BRCA1 or BRCA2 carriers compared with noncarriers. Multiplicative PRS-by-PV interactions for other genes were not statistically significant after accounting for multiple testing (Data Supplement 2).

TABLE 2.

Adjusted ORs and 95% CIs for Risk of Breast Cancer by Age Group

TABLE 3.

Lifetime Absolute BC Risk (by age 80 years) of BC for Different Pathogenic Variant Carriers With Respect to Different PRS Percentile and BC Family History Status

The OR associated with PRS in the best fitting model was the same for noncarriers and carriers of PVs in genes other than BRCA1 or BRCA2, but it is important to note that the absence of multiplicative interactions implies that the absolute risk difference associated with a difference in PRS is larger in carriers of PVs in genes other than BRCA1 or BRCA2 than in noncarriers. For example, among women between the age of 41 and 50 years, the excess risk associated with a 1-SD increase in PRS was 2.4 times greater among carriers of PVs in CHEK2 than among noncarriers (Table 2).

Estimated lifetime risks (by age 80 years) of BC increased with increasing PRS and were higher among carriers of PVs and women with family history of BC (Table 3). The lifetime risk of BC for noncarriers in the 10th and 90th percentiles of PRS were 6.7% and 18.3%, respectively, for women without family history of BC (first-degree relatives) and 9.1% and 23.9%, respectively, for women with family history. Looking at the 10th percentile and the 90th percentile of PRS, the estimated lifetime risk for women without family history of BC ranged from 12.8% to 32.3% for ATM carriers, 15.2% to 37.3% for CHEK2 carriers, and 21.5% to 49.2% for PALB2 carriers.

In Table 4, the percentage of women who have > 20% of lifetime risk based on their PRS are shown, stratified by carrier status and first-degree family history of BC. Most (> 90%) carriers of PVs in BRCA1, BRCA2, and PALB2 have > 20% lifetime risk. By contrast, 52.5% and 68.7% of ATM and CHEK2 carriers, respectively, are above the threshold without a family history of BC and 78.8% and 89.9%, respectively, with a family history of BC.

TABLE 4.

The Percentage of the US Non-Hispanic White Women Identified by Polygenic Risk Score With > 20% Estimated Lifetime Risk of BC, Given Pathogenic Variant Status and Family History of BC

We also estimated 5-year absolute risks (Fig 1) of developing BC across different percentiles of PRS for women at age 40 years (Appendix Fig A1, online only). For 40-year-old women, the estimated 5-year risks of BC for BRCA1 and BRCA2 carriers were substantially larger than that of CHEK2/ATM/PALB2 carriers and noncarriers, regardless of family history status. Many women with PVs in CHEK2 and ATM, particularly those in the low PRS percentiles with no first-degree relatives of BC, had a 5-year risk < 1% at age 40 years.

FIG 1.

Five-year absolute risk of BC across 10-90 percentiles of PRS for different pathogenic variants carriers at age 40 years, (B) with and (A) without family history of BC (first-degree relatives). PRS is standardized with a mean of 0 and a standard deviation of 1. BC, breast cancer; PRS, polygenic risk score.

The age-specific ORs for overall BC with respect to the PRS (10th percentile, median, and 90th percentile) and variant carrier status are shown in Data Supplement 2. For noncarriers < 40 years old, the OR of BC at the 90th percentile of PRS was 3.4 times higher than that at the 10th percentile of PRS, and for BRCA1 carriers, the OR at the 90th percentile of PRS was just 1.6 times higher than that at the 10th percentile of PRS.

The OR for ER-negative BC was 1.51 (95% CI, 1.18 to 1.93) for 1-SD change in the ER-negative PRS and 1.22 (95% CI, 1.08 to 1.37) for 1-SD change in the overall BC PRS among women age ≤ 40 years (Data Supplement 2). Similarly, the OR for ER-positive BC was 1.95 (95% CI, 1.70 to 2.24) for 1-SD change in the ER-positive PRS and 1.73 (95% CI, 1.62 to 1.85) for 1-SD change in the overall BC PRS (Data Supplement 2). Sensitivity analysis restricting to prospective studies showed similar results as our main analysis (Data Supplement 2).

DISCUSSION

In our large case-control study, we found that the PRS created meaningful risk gradients among both carriers and noncarriers of PVs in nine BC predisposition genes and that the risks associated with the PRS were smaller among carriers of PVs in BRCA1 or BRCA2. In particular, we showed that PRS may help differentiate BC risk among carriers of PVs in moderate-penetrance genes such as CHEK2 and ATM, enabling more informed decisions about screening practices and BC risk management.

The US National Comprehensive Cancer Network^34,35 recommends magnetic resonance imaging (MRI) screening for women with a lifetime BC risk > 20%. Our results suggest that PRS information is unlikely to change clinical recommendations for BRCA1 and BRCA2 PV carriers: these women are above the 20% threshold based solely on their carriers status (average lifetime risks of 41.4% and 49.5%, respectively, in Table 3) and very few (< 1%) have estimated lifetime risks below 20% after accounting for PRS (Table 4 and Appendix Fig A1). By contrast, in the absence of PRS information, women who carry PVs in CHEK2 or ATM are above the lifetime risk threshold (average lifetime risks of 25.5% and 21.9%, respectively, in Table 3), but 30.3% of CHEK2 carriers and 47.5% of ATM carriers would be reclassified as below the risk threshold based on their PRS (Table 4 and Appendix Fig A1). ATM carriers at the 10th percentile of PRS have an estimated lifetime risk of BC of 12.8%, which is similar to the population average³⁶ (Table 3).

As panel testing has become more common, an increasing number of women are being identified with ATM and CHEK2 PVs without a family history of BC and these women may benefit from using their PRS for an individualized risk assessment. The incorporation of PRS may also aid in determining the age of screening initiation. National Comprehensive Cancer Network recommends breast MRI starting at age 40 years for women with PVs in CHEK2 or ATM; however, many such women appear to have a 5-year risk of BC < 1% and a later initiation of breast MRI may be appropriate.

Consistent with previous studies,^7,10,31,37 we observed that the OR associated with a 1-SD change in PRS is smaller among BRCA1 or BRCA2 carriers than among noncarriers. Previous studies examining the combined effect of PRS and CHEK2 PVs found that the relative risk associated with PRS was similar in carriers versus noncarriers, consistent with our results.^11,38 Despite smaller numbers of BRCA1, BRCA2, and CHEK2 carriers compared with previous clinic-based studies (Table 2, Data Supplement 2), to our knowledge this study is the first to evaluate the joint effect of PRS and PVs in the general population in nine different BC predisposition genes.

We also showed that an ER-negative and ER-positive PRS worked better in predicting ER-negative and ER-positive BC than the overall PRS, suggesting that subtype-specific PRS could eventually be used for targeted screening or prevention strategies that are specific to BC subtypes.

Our study has certain limitations. First, the PRS was calculated based on 105 SNPs, whereas a recent PRS study included 313 SNPs^31,39; however, the relative risk associated with a one-SD difference in PRS was similar in both studies (1.63 in our sample v 1.61 in the previous study). Second, our study only investigated women of European ancestry. PRS constructed using exclusively European-ancestry subjects were found to be less predictive of BC risk in other ancestry groups^40,41 and future studies performed among women of diverse ancestry are needed. In addition, we had limited numbers of ER-negative BC cases (Data Supplement 2), which limited our statistical power in examining subtype-specific risk estimates. Statistical methods for developing subtype-specific PRS are also an area of active research; the relative utility of subtype-specific models may change as larger training data sets become available. Our model is also not able to distinguish BC case subtypes such as invasive versus in-situ BC and further work needs to be carried out. Although this study was one of the largest to evaluate the combined effect of PRS and PVs on BC risk, an even larger sample size is needed to detect nonmultiplicative interactions between PRS and specific PVs and assess the fit of an additive risk difference model.⁴² Finally, we only had information on first-degree family history of BC; more detailed family history information when available can further refine personalized risk estimates.⁴³

Our data suggest that BC risk prediction models can be modified to include both PRS and PVs to provide more personalized estimates of BC risk. The discrimination of our best-fitting risk model including PRS was statistically significantly greater than the discrimination of a model that did not include PRS (c-statistics of 0.66 and 0.60, respectively). Moreover, our model incorporating both PRS and PV can reclassify risks for some carriers of PVs in moderate-penetrance genes such as ATM or CHEK2, leading to potential changes in screening recommendations. Future work is needed to establish the clinical utility and uptake of stratified primary and secondary prevention strategies based on rare variants in cancer predisposition genes and PRS. Absolute risk models incorporating both rare and common variants (and potentially information on clinical factors, reproductive history, and mammographic density) should be calibrated across diverse populations.^44-46 Particular care should be paid to calibration in the tails of the risk distribution, where model assumptions such as an exponential relationship between risk and PRS can have a large impact.^42,47 In the absence of randomized clinical trial evidence, simulation models can also assess the benefits, risks, and costs associated with stratified prevention and screening strategies.^48-50 Such studies assessing the theoretical and empirical performance of BC PRS as well as their implementation will help inform necessary guidelines for a joint application of PRS and rare PVs in the future.

Appendix

FIG A1.

Distribution of the 5-year absolute risk and lifetime (by age 80 years) absolute risk for women at age 40 years (B and D) with family history (first-degree relatives) of BC and (A and C) without family history of BC. BC, breast cancer.

STUDY GROUP

WHI. Program Office: (National Heart, Lung, and Blood Institute, Bethesda, MD) Jacques Rossouw, Shari Ludlam, Joan McGowan, Leslie Ford; Nancy Geller Clinical Coordinating Center: (Fred Hutchinson Cancer Research Center, Seattle, WA) Garnet Anderson, Ross Prentice, Andrea LaCroix, Charles Kooperberg; Investigators and Academic Centers: (Brigham and Women's Hospital, Harvard Medical School, Boston, MA) JoAnn E. Manson; (MedStar Health Research Institute/Howard University, Washington, DC) Barbara V. Howard; (Stanford Prevention Research Center, Stanford, CA) Marcia L. Stefanick; (The Ohio State University, Columbus, OH) Rebecca Jackson; (University of Arizona, Tucson/Phoenix, AZ) Cynthia A. Thomson; (University at Buffalo, Buffalo, NY) Jean Wactawski-Wende; (University of Florida, Gainesville/Jacksonville, FL) Marian Limacher; (University of Iowa, Iowa City/Davenport, IA) Jennifer Robinson; (University of Pittsburgh, Pittsburgh, PA) Lewis Kuller; (Wake Forest University School of Medicine, Winston-Salem, NC) Sally Shumaker; (University of Nevada, Reno, NV) Robert Brunner; Women's Health Initiative Memory Study: (Wake Forest University School of Medicine, Winston-Salem, NC) Mark Espeland.

AUTHOR CONTRIBUTIONS

Conception and design: Chi Gao, Eric C. Polley, David J. Hunter, David E. Goldgar, Katherine L. Nathanson, Fergus J. Couch, Peter Kraft

Financial support: A. Heather Eliassen, David J. Hunter, Polly A. Newcomb, Katherine L. Nathanson, Fergus J. Couch

Administrative support: Leslie Bernstein, Elizabeth S. Burnside, A. Heather Eliassen, Polly A. Newcomb, Katie M. O'Brien, Rulla Tamimi, Fergus J. Couch

Provision of study materials or patients: Christine B. Ambrosone, Leslie Bernstein, Elizabeth S. Burnside, A. Heather Eliassen, Christopher Haiman, David J. Hunter, Esther M. John, Susan L. Neuhausen, Polly A. Newcomb, Janet E. Olson, Dale P. Sandler, Jack A. Taylor, Amy Trentham-Dietz, Celine M. Vachon, Clarice R. Weinberg, Song Yao, Jeffrey N. Weitzel, Fergus J. Couch

Collection and assembly of data: Chi Gao, Eric C. Polley, Steven N. Hart, Hongyan Huang, Chunling Hu, Jenna Lilyquist, Christine B. Ambrosone, Paul L. Auer, Leslie Bernstein, Elizabeth S. Burnside, A. Heather Eliassen, Mia M. Gaudet, Christopher Haiman, David J. Hunter, Eric J. Jacobs, Esther M. John, Sara Lindström, Huiyan Ma, Susan L. Neuhausen, Polly A. Newcomb, Janet E. Olson, Irene M. Ong, Alpa V. Patel, Dale P. Sandler, Rulla Tamimi, Jack A. Taylor, Lauren R. Teras, Amy Trentham-Dietz, Celine M. Vachon, Clarice R. Weinberg, Jeffrey N. Weitzel, Katherine L. Nathanson, Fergus J. Couch, Peter Kraft

Data analysis and interpretation: Chi Gao, Eric C. Polley, Steven N. Hart, Chunling Hu, Rohan Gnanaolivu, Nicholas J. Boddicker, Jie Na, Mia M. Gaudet, Susan L. Neuhausen, Polly A. Newcomb, Katie M. O'Brien, Janet E. Olson, Julie R. Palmer, Dale P. Sandler, Rulla Tamimi, Amy Trentham-Dietz, Song Yao, Susan M. Domchek, Katherine L. Nathanson, Fergus J. Couch, Peter Kraft

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Risk of Breast Cancer Among Carriers of Pathogenic Variants in Breast Cancer Predisposition Genes Varies by Polygenic Risk Score

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).