Highlights
- •Three polygenic risk scores for breast cancer assessed in 3 groups of French women.
- •All of them associated with breast cancer but effects varied depending on group.
- •Predictive ability is higher in women with no BRCA1 or BRCA2 pathogenic variant.
- •Opportunity to refine risk stratification in carriers of a moderate-risk variant.
- •Effects of SNPs in polygenic risk scores vary depending on country and population.
Abstract
Background
Three partially overlapping breast cancer polygenic risk scores (PRS) comprising 77, 179 and 313 SNPs have been proposed for European-ancestry women by the Breast Cancer Association Consortium (BCAC) for improving risk prediction in the general population. However, the effect of these SNPs may vary from one country to another and within a country because of other factors.
Objective
To assess their associated risk and predictive performance in French women from (1) the CECILE population-based case-control study, (2) BRCA1 or BRCA2 (BRCA1/2) pathogenic variant (PV) carriers from the GEMO study, and (3) familial breast cancer cases with no BRCA1/2 PV and unrelated controls from the GENESIS study.
Results
All three PRS were associated with breast cancer in all studies, with odds ratios per standard deviation varying from 1.7 to 2.0 in CECILE and GENESIS, and hazard ratios varying from 1.1 to 1.4 in GEMO. The predictive performance of PRS313 in CECILE was similar to that reported in BCAC but lower than that in GENESIS (area under the receiver operating characteristic curve (AUC) = 0.67 and 0.75, respectively). PRS were less performant in BRCA2 and BRCA1 PV carriers (AUC = 0.58 and 0.54 respectively).
Conclusion
Our results are in line with previous validation studies in the general population and in BRCA1/2 PV carriers. Additionally, we showed that PRS may be of clinical utility for women with a strong family history of breast cancer and no BRCA1/2 PV, and for those carrying a predicted PV in a moderate-risk gene like ATM, CHEK2 or PALB2.
Keywords
1. Introduction
Prediction of cancer risk is an essential part of preventative medicine that can help guiding clinical management, in particular in hereditary breast and ovarian cancer (HBOC) families. Breast cancer risk prediction classically includes risk factors such as age, sex, family history of disease, lifestyle, hormonal and clinical features, breast density and genotype for pathogenic variants (PV) in the predisposition genes BRCA1 and BRCA2 (BRCA1/2) [
[1]
]. In recent years, some risk models have been updated to include additional genomic information, typically the effects of rare PV in other genes (ATM, CHEK2 and PALB2) [[2]
,[3]
] and the joined effect of single nucleotide polymorphisms (SNPs) summarized in polygenic risk scores (PRS) [4
, 5
, 6
]. Indeed, in the general population, some studies suggested that stratification of women according to their risk of breast cancer based on their PRS could personalize screening and prevention strategies [7
, 8
, 9
, 10
]. Several PRS for breast cancer have been defined and validated in women of European ancestry from the large multi-centric and multi-country studies conducted by the Breast Cancer Association Consortium (BCAC). The establishment of the most recent PRS, which includes 313 SNPs, and its validation in independent prospective sample sets required almost all resources available before 2019 [[11]
]. Therefore, few studies have independently benchmarked its performance. It was recently assessed in Asians [[12]
,[13]
] and African Americans [[14]
,[15]
] as well as in BRCA1/2 PV carriers in international studies conducted by the Consortium of Investigators of Modifiers of BRCA1/2 (CIMBA) [[16]
]. However, even in European populations, the effect of some of these SNPs may vary from one country to another or within one country because of other factors, which may modify predictive performance of PRS. Here, we assembled a large dataset of women of European descent residing in France and representing three populations of women at different level of breast cancer risk to investigate in each group the performance of three partially overlapping PRS for breast cancer comprising 77 [[10]
], 179 [[17]
,[18]
] and 313 SNPs [[11]
]. Our goal was to assess whether these PRS modify breast cancer risk prediction in order to further assess if, when incorporated in predictive models, they could improve ability to predict breast cancer in the French population, and particularly in women attending family cancer clinics because of their personal or family history of breast cancer.2. Material and methods
2.1 Study participants, genotyping and genotype imputation
The dataset comprised 7173 women of European ancestry from three studies on breast cancer conducted in France, namely CECILE, GEMO and GENESIS. The study designs are outlined in Supplementary methods and were described in detail previously [
19
, 20
, 21
].CECILE is a population-based case-control study involving incident breast cancer cases and cancer-free controls that were frequency-matched to the cases by 10-year age group and study area [
[19]
,[22]
]. Genotyping of samples was performed using the iCOGS beadchip (Illumina Inc. USA) in the context of genome-wide association studies (GWAS) conducted by BCAC [[17]
]. iCOGS genotyping data were available for 1019 cases (of which 900 cases had invasive tumours and 119 had in situ tumour) and 999 controls [[19]
]. After quality control (QC) procedures, we analysed data from 1015 cases and 996 controls.GEMO is a resource to study genetic factors modifying cancer risk in HBOC families segregating BRCA1/2 PV carriers. Participants are tested positive for a confirmed PV in BRCA1/2 and are enrolled through the national network of cancer genetics clinics [
[20]
]. Genotyping of samples was performed using iCOGS or OncoArray beadchips in the context of GWAS conducted by CIMBA [[23]
,[24]
]. After QC, we analysed iCOGS data from 1883 BRCA1/2 female carriers (1181 of whom had breast cancer) and OncoArray data from 756 BRCA1/2 female carriers (428 of whom had breast cancer).GENESIS is a case-control study [
[21]
] involving familial breast cancer cases tested negative for BRCA1/2 and with at least one sister affected with breast cancer and unrelated controls (friends or colleagues) aged-matched (±3 years) to cases at interview. Genotyping of samples was performed using the iCOGS beadchip. After QC, we analysed data from 1257 index cases and 1266 unrelated controls.Detailed information about the design, genotyping and initial QC for iCOGS [
[25]
] and for OncoArray [[17]
,[18]
,[26]
] can be found within the original publication of the consortia. We performed genotype calling, QC and genotype imputation of missing SNPs for iCOGS and OncoArray datasets separately. The different steps of the process are summarized on Supplementary Fig. S1. The number of SNPs in each PRS that are not present on the iCOGS and OncoArray chips are also provided on Supplementary Fig. S1. We performed imputation of these missing SNPs in each study separately using a two-stage procedure, with SHAPEIT2 [[27]
] to derive phased genotypes and IMPUTE2 [[28]
] to perform imputation, using 1000 Genomes Project (Phase 3) data as the reference panel [[29]
].The modifier effect of PRS313 in carriers of a predicted PV in ATM, CHEK2 and PALB2 was assessed in 974 cases and 1135 controls from GENESIS, for whom sequencing data were available [
[30]
] (Supplementary methods).2.2 Statistical analyses
PRS is the weighted combined effect of uncorrelated SNPs calculated under the hypothesis of additivity of SNP effect. We used SNP effect size estimated by the BCAC as weight, as described in the Supplementary data. We examined association with breast cancer and performance of the three following PRS: PRS77 which includes 77 independent SNPs identified in early GWAS on breast cancer conducted by the BCAC [
[10]
], PRS179 which includes 179 independent SNPs identified afterward in successive GWAS [[17]
,[18]
,[31]
], and PRS313 which was most recently developed and validated in women of European ancestry enrolled in 10 prospective cohorts and in the UK biobank [[11]
] (Supplementary Table S1).We assessed the association of the standardized PRS in CECILE and GENESIS using logistic regression models, adjusted for age (age at diagnosis of breast cancer for cases, age at interview for controls). In GEMO, to account for nonrandom sampling with respect to breast cancer status, we assessed association with risk of breast cancer using a weighted Cox proportional hazards regression model [
[32]
] (see Supplementary methods).3. Results
3.1 Association and performance of PRS313 in the three French studies
Demographic and clinical characteristics of women included in the three groups of the population with presumably different levels of breast cancer risk are presented in Table 1. As anticipated, we found that PRS313 was associated with breast cancer in all three groups, but the associated risk varied substantially from one group to another. In CECILE, the OR per SD was 1.71 (95% CI 1.57–1.86) and AUC was 0.67 (95% CI 0.64–0.69) (Table 2). These results were similar to those reported for overall breast cancer in BCAC (OR per SD, 1.65; 95% CI 1.59–1.72; AUC = 0.64). In GENESIS, the associated risk was higher (OR per SD, 1.84; 95% CI 1.70–1.99) and this PRS performed even better (AUC, 0.75; 95% CI 0.73–0.77) than in CECILE women (Table 2) [
[11]
].Table 1Characteristics of women included in the analyses.
| Study | Population | SNP array | Cases | Controls | ||||
|---|---|---|---|---|---|---|---|---|
| N | Mean age at diagnosis (SE) | Mean year at birth (range) | N | Mean age at interview/censure (SE) | Mean year at birth (range) | |||
| CECILE | Cases selected from cancer registries and frequency-matched population controls (for age and area of residence) | iCOGS | 1015 | 54.4 (0.3) | 1951 (1930–1980) | 996 | 54.7 (0.3) | 1951 (1931–1981) |
| GENESIS | Index cases from HBOC families and unrelated controls (friends and colleagues) | iCOGS | 1257 | 50.5 (0.3) | 1951 (1918–1977) | 1266 | 56.4 (0.3) | 1953 (1926–1986) |
| GEMO | BRCA1 carriers | iCOGS | 697 | 41.0 (0.3) | 1960 (1907–1986) | 491 | 43.3 (0.6) | 1964 (1913–1994) |
| OncoArray | 184 | 40.6 (0.7) | 150 | 41.1 (1.1) | ||||
| GEMO | BRCA2 carriers | iCOGS | 484 | 43.8 (0.5) | 1955 (1914–1985) | 211 | 44.2 (0.9) | 1963 (1913–1992) |
| OncoArray | 244 | 43.0 (0.6) | 178 | 45.0 (1.1) | ||||
SE, standard error.
Table 2Association of PRS313 with breast cancer.
| Study | Cases | Controls | PRS313 | OR (95% CI) | P-value | AUC (95% CI) |
|---|---|---|---|---|---|---|
| CECILE | 1015 | 996 | BC | 1.71 (1.57–1.86) | 1.9 × 10−36 | 0.67 (0.64–0.69) |
| ER+ | 1.82 (1.66–2.00) | 1.5 × 10−37 | 0.67 (0.65–0.69) | |||
| ER− | 1.53 (1.39–1.70) | 5.2 × 10−17 | 0.61 (0.59–0.64) | |||
| GENESIS | 1257 | 1266 | BC | 1.84 (1.70–1.99) | 1.1 × 10−53 | 0.75 (0.73–0.77) |
| ER+ | 2.00 (1.83–2.18) | 8.7 × 10−56 | 0.76 (0.74–0.77) | |||
| ER− | 1.83 (1.66–2.02) | 2.9 × 10−34 | 0.72 (0.70–0.74) | |||
| HR (95% CI) | P-value | AUC (95% CI) | ||||
| GEMO – BRCA1 | 881 | 641 | BC | 1.15 (1.05–1.26) | 0.01 | 0.54 (0.52–0.57) |
| ER+ | 1.14 (1.03–1.26) | 0.03 | 0.54 (0.51–0.57) | |||
| ER− | 1.26 (1.13–1.41) | 5.0 × 10−4 | 0.57 (0.54–0.60) | |||
| GEMO – BRCA2 | 728 | 389 | BC | 1.28 (1.15–1.42) | 2.0 × 10−4 | 0.58 (0.55–0.62) |
| ER+ | 1.30 (1.16–1.46) | 2.0 × 10−4 | 0.58 (0.54–0.62) | |||
| ER− | 1.20 (1.04–1.37) | 0.03 | 0.56 (0.52–0.59) |
a Three versions of PRS313 were derived using per-allele log ORs from the BCAC: one for overall breast cancer (BC), one for ER-positive breast cancer (ER+), and one for ER-negative breast cancer (ER−). PRS were standardized using the mean PRS and standard deviation in BCAC controls.
b Odds ratio per 1 unit of standard deviation.
c Hazard ratio per 1 unit of standard deviation. CI, confidence interval. AUC, area under the receiver operating characteristic curve.
As 85% of CECILE cases and 84% of GENESIS cases had developed ER-positive breast cancer, we also examined association and performance of the ER-positive breast cancer PRS313 developed by BCAC in these two studies. Although the performance of this subtype-specific PRS was similar to that of the PRS for overall breast cancer in CECILE (AUC, 0.67; 95% CI 0.65–0.69) and in GENESIS (AUC, 0.76; 95% CI 0.74–0.77), point estimates of ORs were higher in both studies (OR per SD was 1.82 in CECILE and 2.00 in GENESIS) (Table 2).
In GEMO, risks associated with overall breast cancer or with breast cancer subtypes were similar and much lower than those observed in CECILE and GENESIS: for BRCA1 PV carriers, HR per SD ranged from 1.15 (95% CI 1.05–1.26) for the overall breast cancer PRS313 to 1.26 (95% CI 1.13–1.41) for the ER-negative PRS313; for BRCA2 PV carriers, HR per SD ranged from 1.28 (95% CI 1.15–1.42) for the overall breast cancer PRS313 to 1.30 (95% CI 1.16–1.46) for the ER-positive PRS313 (Table 2). Moreover, the discriminatory ability of PRS was at most 0.57 for BRCA1 PV carriers and at most 0.58 for BRCA2 PV carriers (Table 2).
3.2 Comparison of performance of PRS77, PRS179 and PRS313
Although most genomic regions containing the SNPs included in the three PRS overlap, only a limited number of SNPs are identical or correlated (Supplementary Table S2). We therefore compared the effect size and performance of the three PRS in the different datasets. As PRS313, PRS77 and PRS179 were associated with breast cancer risk in CECILE and GENESIS, with OR per SD varying from 1.67 to 2.01. However, we found that PRS77 may confer a lower risk of breast cancer than PRS179 and PRS313 in GENESIS (Table 3).
Table 3Comparison of effect of PRS313, PRS179 and PRS77 on breast cancer risk.
| Study | PRS313 | PRS179 | PRS77 | ||||||
|---|---|---|---|---|---|---|---|---|---|
| OR (95% CI) | P-value | AUC (95% CI) | OR (95% CI) | P-value | AUC (95% CI) | OR (95% CI) | P-value | AUC (95% CI) | |
| CECILE | 1.84 (1.68–2.03) | 1.9 × 10−36 | 0.67 (0.64–0.69) | 1.81 (1.65–1.99) | 3.3 × 10−35 | 0.66 (0.64–0.69) | 1.67 (1.53–1.84) | 9.5 × 10−28 | 0.64 (0.62–0.67) |
| GENESIS | 2.01 (1.84–2.19) | 1.1 × 10−53 | 0.75 (0.73–0.77) | 2.01 (1.83–2.20) | 1.3 × 10−51 | 0.75 (0.73–0.77) | 1.71 (1.57–1.86) | 2.8 × 10−36 | 0.73 (0.71–0.74) |
| HR (95% CI) | P-value | AUC (95% CI) | HR (95% CI) | P-value | AUC (95% CI) | HR (95% CI) | P-value | AUC (95% CI) | |
| GEMO – BRCA1 | 1.17 (1.06–1.30) | 0.01 | 0.54 (0.52–0.57) | 1.14 (1.04–1.26) | 0.03 | 0.54 (0.51–0.56) | 1.14 (1.03–1.26) | 0.03 | 0.53 (0.50–0.56) |
| GEMO – BRCA2 | 1.32 (1.18–1.50) | 2.0 × 10−4 | 0.58 (0.55–0.62) | 1.35 (1.20–1.52) | 2.0 × 10−5 | 0.59 (0.55–0.62) | 1.28 (1.15–1.44) | 3.0 × 10−4 | 0.58 (0.54–0.61) |
AUC, area under the receiver operating characteristic curve.
a PRS were standardized using the mean PRS and standard deviation in CECILE controls.
b Odds ratio per 1 unit of standard deviation.
c Hazard ratio per 1 unit of standard deviation.
In GEMO, the effect size was unchanged whatever the PRS, with HR per SD varying from 1.14 to 1.17 for BRCA1 PV carriers and from 1.28 to 1.35 for BRCA2 PV carriers (Table 3). Regarding their discriminative ability, all three PRS performed equally in each dataset.
3.3 Breast cancer risk modification by PRS313 in carriers of a predicted pathogenic variant in a moderate-risk breast cancer gene
We next investigated the performance of PRS313 in carriers of a predicted PV in ATM, CHEK2 or PALB2. In a previous work, we confirmed that carriers of such variants in GENESIS were at increased risk of developing breast cancer [
[30]
]. In GENESIS participants with both sequencing and iCOGS genotyping data, women carrying an ATM, CHEK2 or PALB2 variant had a two-fold increased risk of breast cancer per unit of SD of PRS313 (Table 4). Risks associated with PRS313 in carriers were higher than those for BRCA1/2 PV carriers. However, we did not observe significant variation in the effect of the PRS according to the altered gene or the variant type (loss-of function or missense variant) (Table 4). We also observed a significant 8.7-fold increased risk of breast cancer for carriers of a predicted PV in the highest tertile of PRS313 when compared with noncarriers with a PRS313 in the middle tertile (Table 5). This associated risk was different from the risk of noncarriers with a PRS313 in the highest tertile (PHet <0.001).Table 4Effect of predicted pathogenic variants in ATM, CHEK2 or PALB2, and effect of PRS313 on breast cancer risk.
| GENESIS participants | Cases | Controls | Effect of rare variants | Effect of PRS313 | ||
|---|---|---|---|---|---|---|
| OR (95% CI) | P-value | OR (95% CI) | P-value | |||
| All | 974 | 1135 | n/a | n/a | 1.84 (1.69–2.00) | 1.7 × 10−45 |
| Noncarriers | 847 | 1067 | n/a | n/a | 1.84 (1.68–2.01) | 1.7 × 10−41 |
| Carriers of a variant (any type) in: | ||||||
| ATM or CHEK2 or PALB2 | 127 c Five cases carried two variants: two cases carried an ATM loss-of-function variant and a PALB2 missense variant, one case carried a CHEK2 loss-of-function variant and a PALB2 missense variant, one case carried an ATM missense variant and a CHEK2 missense variant, and one case carried an ATM missense variant and a PALB2 missense variant. These cases were included in each per gene analysis. | 68 | 2.18 (1.59–2.99) | 1.5 × 10−6 | 2.05 (1.48–2.83) | 1.6 × 10−5 |
| ATM | 58 | 38 | 1.79 (1.16–2.76) | 0.008 | 2.15 (1.37–3.37) | 8.5 × 10−4 |
| CHEK2 | 48 | 21 | 2.43 (1.43–4.15) | 0.001 | 1.62 (0.92–2.85) | 0.10 |
| PALB2 | 26 | 9 | 3.08 (1.40–6.76) | 0.005 | 2.07 (0.93–4.62) | 0.08 |
| Carriers of a loss-of-function variant in: | ||||||
| ATM or CHEK2 or PALB2 | 38 | 8 | 4.64 (2.12–10.15) | 1.3 × 10−4 | 2.69 (0.97–7.50) | 0.06 |
| Carriers of a missense variant in: | ||||||
| ATM or CHEK2 or PALB2 | 92 | 60 | 1.82 (1.29–2.58) | 7.2 × 10−4 | 1.94 (1.38–2.73) | 1.5 × 10−4 |
a Overall breast cancer PRS313 standardized using the mean PRS313 and standard deviation in BCAC controls.
b GENESIS participants with available multigene panel sequencing data and iCOGS genotyping data.
c Five cases carried two variants: two cases carried an ATM loss-of-function variant and a PALB2 missense variant, one case carried a CHEK2 loss-of-function variant and a PALB2 missense variant, one case carried an ATM missense variant and a CHEK2 missense variant, and one case carried an ATM missense variant and a PALB2 missense variant. These cases were included in each per gene analysis.
Table 5Joint effect of PRS313 and predicted pathogenic variants in ATM, CHEK2 and PALB2.
| Cases | Controls | OR (95% CI) | P-value | |
|---|---|---|---|---|
| Noncarrier & PRS313 in the middle tertile | 213 | 349 | Ref. | |
| Noncarrier & PRS313 in the lowest tertile | 99 | 352 | 0.49 (0.37–0.66) | 1.3 × 10−6 |
| Noncarrier & PRS313 in the highest tertile | 535 | 366 | 2.43 (1.95–3.04) | 4.3 × 10−15 |
| Carrier & PRS313 in the middle tertile | 36 | 29 | 1.92 (1.13–3.28) | 0.02 |
| Carrier & PRS313 in the lowest tertile | 14 | 26 | 0.97 (0.49–1.93) | 0.94 |
| Carrier & PRS313 in the highest tertile | 77 | 13 | 8.66 (4.65–16.13) | 9.9 × 10−12 |
a Odds ratio per 1 unit of standard deviation.
b Noncarrier of a predicted pathogenic variant in ATM, CHEK2 or PALB2.
c Carrier of a predicted pathogenic variant in ATM, CHEK2 or PALB2.
Finally, when women were binned by tertile of the PRS, we observed in CECILE and GENESIS significant risk modification for women in the lowest and highest tertiles of PRS313 as compared to women in the middle tertile (Table 6). In GENESIS, point estimates for ATM and PALB2 variant carriers in the highest tertile were substantially increased (OR, 5.65; 95% CI 1.82–17.5 for ATM and OR, 10.2; 95% CI 0.98–106 for PALB2). However, these observations should be interpreted with caution due to the limited number of variant carriers in GENESIS and consequently the large confidence intervals. In GEMO, we observed a significant increased risk of breast cancer for BRCA1 PV carriers in the highest tertile of PRS313 as compared to those in the middle tertile (OR, 1.39; 95% CI 1.11–1.75), and a significant reduced risk for BRCA2 PV carriers in the lowest tertile as compared to those in the middle tertile (OR, 0.67; 95% CI 0.48–0.95).
Table 6Odds ratios for developing breast cancer by PRS313 tertile.
| PRS313 tertile | ≤33 | >33 to 66 | >66 | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Cases | Controls | OR (95% CI) | P-value | Cases | Controls | OR (95% CI) | Cases | Controls | OR (95% CI) | P-value | |
| CECILE | 143 | 332 | 0.46 (0.36–0.59) | 1.5 × 10−9 | 309 | 332 | Ref. | 563 | 332 | 1.82 (1.48–2.24) | 1.1 × 10−8 |
| GENESIS | |||||||||||
| All | 113 | 378 | 0.49 (0.37–0.64) | 2.3 × 10−7 | 249 | 378 | Ref | 612 | 379 | 2.48 (2.01–3.06) | 3.3 × 10−17 |
| ATM | 7 | 15 | 0.62 (0.19–2.02) | 0.43 | 16 | 17 | Ref. | 35 | 6 | 5.65 (1.82–17.5) | 2.8 × 10−3 |
| CHEK2 | 4 | 8 | 0.33 (0.07–1.62) | 0.17 | 14 | 7 | Ref. | 30 | 6 | 2.62 (0.69–9.96) | 0.16 |
| PALB2 | 4 | 3 | 1.01 (0.15–6.96) | 0.99 | 7 | 5 | Ref. | 15 | 1 | 10.2 (0.98–106) | 0.05 |
| GEMO | |||||||||||
| BRCA1 | 256 | 214 | 1.02 (0.80–1.30) | 0.84 | 263 | 213 | Ref. | 362 | 214 | 1.39 (1.11–1.75) | 4.4 × 10−3 |
| BRCA2 | 167 | 130 | 0.67 (0.48–0.95) | 0.02 | 257 | 129 | Ref. | 304 | 130 | 1.32 (0.97–1.82) | 0.08 |
a GENESIS participants with available multigene panel sequencing data and iCOGS genotyping data.
4. Discussion
The three partially overlapping PRS investigated here had been defined using datasets composed of women of European ancestry from collaborative analyses involving heterogeneous populations from multiple countries. In these datasets, the vast majority of cases were unselected for their personal or family history of breast cancer. Because effect of SNPs on cancer risk may vary from one country to another due to other factors, there is some uncertainty on how the proposed PRS influence breast cancer risk in a given population. We thus sought to confirm their association with breast cancer in three groups of French women at different level of breast cancer risk: cases from the general population, cases with a family history of breast cancer in siblings and tested negative for BRCA1/2, and BRCA1/2 PV carriers. Within each group, the effect of the three PRS on risk was similar with the exception of PRS77 which was associated with a lower OR per SD than PRS313 and PRS179 in GENESIS. Since these PRS were constructed at different time using different approaches to select the SNPs, only a limited number of SNPs are shared or strongly correlated between the three PRS. Therefore, the choice of the PRS to use in risk prediction models may be driven by the efficiency of the genotyping technologies in order to minimize the number of missing SNP genotypes and imputations. Notably, using the OncoArray data, 71 out of 179 SNPs (39.7%) had to be imputed for PRS179 and 204 out of 313 SNPs (65.2%) had to be imputed for PRS313, and no significant improvement in either OR or AUC was observed for PRS313.
Remarkably, we found the predictive value of PRS313 for overall breast cancer in women with a familial predisposition and tested negative for BRCA1/2 was higher than in women from the general population and women from BCAC (AUC = 0.75 in GENESIS versus 0.67 in CECILE versus 0.64 in BCAC [
[11]
]). Because GENESIS controls were not selected based on family history of breast cancer, it is possible that a residual familial effect correlated to the PRS would explain partially the higher performance. However, using the PRS to discriminate women at higher risk of breast cancer among those with a family history of breast cancer and no BRCA1/2 PV may still be useful. In addition, although GENESIS cases were diagnosed at a younger age than CECILE cases, it is unlikely that a birth cohort effect could explain the difference in the performance of the PRS in the two studies (Supplementary Table S3). By contrast, PRS313 was much less performant in BRCA2 PV carriers (AUC = 0.58) and not predictive in BRCA1 PV carriers (AUC = 0.54). In terms of associated risks, HR per SD obtained in GEMO are in line with those reported in the CIMBA retrospective cohort for overall breast cancer (HR per SD = 1.20 (1.17–1.23) for BRCA1 PV carriers and 1.31 (1.27–1.36) for BRCA2 PV carriers in CIMBA when no family adjustment was made) [[16]
].It should be mentioned that CECILE was one of the 69 BCAC studies contributing to the development of PRS313 [
[11]
] and that GEMO contributed to the CIMBA study that assessed this PRS in BRCA1/2 PV carriers [[16]
]. However, CECILE samples only accounted for 1.2% of the BCAC dataset and GEMO samples accounted for 8% of BRCA1 PV carriers and 9% of BRCA2 PV carriers investigated by CIMBA. In absence of other available datasets, our study represents a unique opportunity to assess the relevance of the published PRS to improve breast cancer prediction in the French population.Specific PRS have been described to predict risk of breast cancer subtypes, which could be useful to stratify women according to prognosis or for more beneficial treatments [
[11]
,[33]
]. We confirmed the strongest association of ER-positive PRS in the French general population and in the high-risk population with no BRCA1/2 PV, with OR per SD and AUC close to those reported by Mavaddat et al. (OR per SD, 1.74 (95% CI 1.66–1.82) and AUC = 0.65 in BCAC) [[11]
]. On the other hand, the ER-negative PRS was more strongly associated in BRCA1 PV carriers. However, the performance of this latter PRS was limited, which agrees with CIMBA results (AUC = 0.57 in GEMO and 0.58 in CIMBA) [[16]
]. In BRCA2 PV carriers, we confirmed the strongest associations for the overall and the ER-positive breast cancer PRS and their similar performance (AUC = 0.58 in GEMO and 0.60 in CIMBA) [[16]
].Interestingly, in women with a familial predisposition involving a moderate-risk gene, OR associated with PRS313 are of the same magnitude that OR associated with rare predicted PV in ATM, CHEK2 and PALB2. This indicates that adding PRS as a risk factor in risk prediction models may significantly improve management of women who receive genetic counselling.
Due to small numbers, we could not assess the effect of age or family history on the performance of PRS, nor assess PRS that have been recently developed in Asian [
[12]
,[13]
] and African populations [[14]
,[15]
], and had to restrict our analyses to women of European descent according to genetic markers to assess PRS that had been constructed using per allele log OR obtained in the BCAC European population. Hence, further efforts should be conducted to assess and probably improve the discriminatory power of PRS in under-studied ethnic groups, as this could limit PRS adoption and applicability and exacerbate health disparities [[34]
]. PRS have the potential to enhance disease risk prediction and diagnostic refinement, to predict progression and recurrence of disease and to improve the efficiency of population-level screening, however PRS cannot be used as a standalone tool, since as for other risk factors, they are estimates with a level of uncertainty that could affect risk stratification owing to statistical imprecision and the use of discrete cut-offs. Therefore, communication of PRS result to patients by healthcare professionals trained in genetics requires careful consideration as they may be incorrectly conflated with return of diagnostic test for high penetrance variants [[34]
].In conclusion, efforts should be pursued to develop and validate more specific PRS, especially in high-risk women with a known PV in a moderate-to high-risk gene in order to integrate the effects of the common SNPs with family history, lifestyle/hormonal and other risk factors like mammographic density in risk prediction models that are applicable for each country. Indeed, the identification of groups of women with sufficiently different cancer risks will be informative in the genetic counselling process to allow female PV carriers to make more informed choices about the type and timing of cancer screening, prevention and possible risk reduction treatments.
Data accessibility
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
Ethics statement
Participants from CECILE, GEMO and GENESIS provided written informed consent. All three studies were approved by the relevant Advisory Committees on the Treatment of Health Research Information (Comité Consultatif de Protection des Personnes dans la Recherche Biomédicale (CCPPRB) Kremlin-Bicêtre for CECILE, Comité Consultatif sur le Traitement de l'Information en matière de Recherche dans le domaine de la Santé (CCTIRS) for GEMO, and CCPPRB Ile-de-France III for GENESIS) and by the National Data Protection authority (CNIL). Followed procedures were in accordance with the ethical standards of these committees.
Funding
The project was funded by the French National Institute of Cancer (INCa) and Cancéropôle Ile-de-France (grant SHS-E-SP 18-015).
Financial support for GEMO was initially provided by INCa (Inca PHRC Ile de France, grant AOR 01 082, 2001–2003, grant 2013-1-BCB-01-ICH-1), the Association ‘Le cancer du sein, parlons-en !’ Award (2004), the Association for International Cancer Research (2008–2010), and the Fondation ARC pour la Recherche sur le Cancer (grant PJA 20151203365). It also received support from the Canadian Institute of Health Research for the ‘CIHR Team in Familial Risks of Breast Cancer’ program (2008–2013), and the European commission FP7, Project ‘Collaborative Ovarian, breast and prostate Gene-environment Study (COGS), Large-scale integrating project’ (2009–2013). GEMO is currently supported by the INCa grant SHS-E-SP 18-015.
Financial support for GENESIS, including genotyping with the iCOGS array, was provided by Ligue Nationale contre le Cancer (grants PRE05/DSL, PRE07/DSL, PRE11/NA), INCa (grant No b2008-029/LL-LC) and the comprehensive cancer center SiRIC (Site de Recherche Intégrée sur le Cancer, grant INCa-DGOS-4654).
For CECILE and GEMO, genotyping with the iCOGS array was funded by the European Union (HEALTH-F2-2009-223175), Cancer Research UK (C1287/A10710) and the Canadian Institutes of Health Research (CIHR) for the CIHR Team in Familial Risks of Breast Cancer. For GEMO, genotyping with the OncoArray was funded by the CIHR, Genome Québec, the Quebec Breast Cancer Foundation and the NCI Genetic Associations and Mechanisms in Oncology (GAME-ON) initiative.
Credit author statement
Fabienne Lesueur, Yue Jiao, Nadine Andrieu: Conceptualization, Formal analyses.
Séverine Eon-Marchais, Noura Mebirouk, Dorothée Le Gal, Juana Beauvallet: Investigation.
Thérèse Truong, Édith Le Floch, Claire Dandine-Roulland, Delphine Bacq-Daian, Robert Olaso, Juliette Albuisson, Séverine Audebert-Bellanger, Pascaline Berthet, Valérie Bonadona, Bruno Buecher, Olivier Caron, Mathias Cavaillé, Jean Chiesa, Chrystelle Colas, Marie-Agnès Collonge-Rame, Isabelle Coupier, Capucine Delnatte, Antoine De Pauw, Hélène Dreyfus, Sandra Fert-Ferrer, Marion Gauthier-Villars, Paul Gesta, Sophie Giraud, Laurence Gladieff, Lisa Golmard, Christine Lasset, Sophie Lejeune-Dumoulin, Mélanie Léoné, Jean-Marc Limacher, Alain Lortholary, Élisabeth Luporsi, Véronique Mari, Christine M. Maugard, Isabelle Mortemousque, Emmanuelle Mouret-Fourme, Sophie Nambot, Catherine Noguès, Cornel Popovici, Fabienne Prieur, Pascal Pujol, Nicolas Sevenet, Hagay Sobol, Christine Toulas, Nancy Uhrhammer, Dominique Vaur, Laurence Venat, Anne Boland-Augé, Pascal Guénel, Jean-François Deleuze, Dominique Stoppa-Lyonnet: Resources.
Yue Jiao, Noura Mebirouk, Sandrine M. Caputo, Marie-Gabrielle Dondon, Mojgan Karimi: Data preparation and Curation.
Yue Jiao, Fabienne Lesueur: Writing – original draft.
Nadine Andrieu, Thérèse Truong, Dominique Stoppa-Lyonnet, Yue Jiao, Fabienne Lesueur: Writing – review & editing. All authors reviewed the manuscript and approved its final version.
Fabienne Lesueur: Supervision.
Conflict of interest statement
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:
DS-L and LG coordinated the genotyping of SNPs included in the PRS of the MammoRisk® test commercialized by Predilife until December 2021. This genotyping was performed in the Department of Genetics of the Institut Curie.
All other authors declare no conflicts of interest.
Acknowledgements
We are most grateful to all subjects who so willingly participated in all three studies. We would like to thank Joe Dennis and Daniel Barnes (University of Cambridge, UK) and Juliette Coignard for helpful discussions regarding the analysis of the iCOGS and OncoArray data, and Malgorzata Leslie for her help in curating GEMO data contributing to the CIMBA dataset. We wish to pay a tribute to Olga M. Sinilnikova, who was one of the initiators and principal investigators of GEMO and GENESIS and who died prematurely on June 30, 2014. We thank all GEMO and GENESIS investigators without whom this research would not be possible.
GEMO collaborating cancer clinics and diagnostic laboratories: Service de Génétique, Institut Curie, Paris: M. Belotti, O. Bertrand, B. Buecher, S.M. Caputo, C. Colas, E. Fourme, M. Gauthier-Villars, L. Golmard, M. Le Mentec, V. Moncoutier, A. de Pauw, C. Saule, D. Stoppa-Lyonnet; Unité Mixte de Génétique Constitutionnelle des Cancers Fréquents, Hospices Civils de Lyon - Centre Léon Bérard, Lyon: N. Boutry-Kryza, A. Calender, S. Giraud, M. Léone; Gustave Roussy, Villejuif: B. Bressac-de-Paillerets, O. Cabaret, O. Caron, E. Rouleau; Centre Jean Perrin, Clermont–Ferrand: Y-J. Bignon, N. Uhrhammer; Centre Léon Bérard, Lyon: V. Bonadona, S. Dussart, C. Lasset, P. Rochefort; Centre François Baclesse, Caen: P. Berthet, L. Castera, D. Vaur; Institut Paoli Calmettes, Marseille: V. Bourdon, C. Noguès, T. Noguchi, C. Popovici, A. Remenieras, H. Sobol; CHU Arnaud-de-Villeneuve, Montpellier: I. Coupier, P. Pujol; Centre Oscar Lambret, Lille: C. Adenis, A. Dumont, F. Révillion; Centre Paul Strauss, Strasbourg: D. Muller; Institut Bergonié, Bordeaux: E. Barouk-Simonet, F. Bonnet, V. Bubien, A. Dupré, A. Floquet, M. Longy, M. Louty, C. Maninna, N. Sevenet; Institut Claudius Regaud, Toulouse: L. Gladieff, R. Guimbaud, V. Feillel, C. Toulas; CHU Grenoble: H. Dreyfus, D. Leroux, C. Legrand, C. Rebischung; CHU Dijon: A. Baurand, G. Bertolone, F. Coron, L. Faivre, C. Jacquot, S. Lizard, S. Nambot; CHU St-Etienne: C. Kientz, M. Lebrun, F. Prieur; Hôtel Dieu Centre Hospitalier, Chambéry: S. Fert-Ferrer; Centre Antoine Lacassagne, Nice: V. Mari; CHU Limoges: L. Vénat-Bouvet; CHU Nantes: S. Bézieau, C. Delnatte; CHU Bretonneau, Tours and Centre Hospitalier de Bourges: I. Mortemousque; Groupe Hospitalier Pitié-Salpétrière, Paris: F. Coulet, M. Warcoin; CHU Vandoeuvre-les-Nancy: M. Bronner, J. Sokolowska; CHU Besançon: MA Collonge-Rame; CHU Poitiers, Centre Hospitalier d'Angoulême and Centre Hospitalier de Niort: P. Gesta, S. Chieze-Valero, B. Gilbert-Dussardier; Centre Hospitalier de La Rochelle: H. Lallaoui; CHU Nîmes Carémeau: J. Chiesa; CHI Poissy: D. Molina-Gomes; CHU Angers: O. Ingster; CHU de Martinique: O. Bera, M. Rose.
GENESIS collaborating cancer clinics: Clinique Sainte Catherine, Avignon: H. Dreyfus; Hôpital Saint Jacques, Besançon: M-A. Collonge-Rame; Institut Bergonié, Bordeaux: M. Longy, A. Floquet, E. Barouk-Simonet; CHU, Brest: S. Audebert; Centre François Baclesse, Caen: P. Berthet; Hôpital Dieu, Chambéry: S. Fert-Ferrer; Centre Jean Perrin, Clermont-Ferrand: Y-J. Bignon; Hôpital Pasteur, Colmar: J-M. Limacher; Hôpital d’Enfants CHU – Centre Georges François Leclerc, Dijon: L. Faivre-Olivier; CHU, Fort de France: O. Bera; CHU Albert Michallon, Grenoble: D. Leroux; Hôpital Flaubert, Le Havre: V. Layet; Centre Oscar Lambret, Lille: P. Vennin†, C. Adenis; Hôpital Jeanne de Flandre, Lille: S. Lejeune-Dumoulin, S. Manouvier-Hanu; CHRU Dupuytren, Limoges: L. Venat-Bouvet; Centre Léon Bérard, Lyon: C. Lasset, V. Bonadona; Hôpital Edouard Herriot, Lyon: S. Giraud; Institut Paoli-Calmettes, Marseille: F. Eisinger, L. Huiart; Centre Val d’Aurelle – Paul Lamarque, Montpellier: I. Coupier; CHU Arnaud de Villeneuve, Montpellier: I. Coupier, P. Pujol; Centre René Gauducheau, Nantes: C. Delnatte; Centre Catherine de Sienne, Nantes: A. Lortholary; Centre Antoine Lacassagne, Nice: M. Frénay, V. Mari; Hôpital Caremeau, Nîmes: J. Chiesa; Réseau Oncogénétique Poitou Charente, Niort: P. Gesta; Institut Curie, Paris: D. Stoppa-Lyonnet, M. Gauthier-Villars, B. Buecher, A. de Pauw, C. Abadie, M. Belotti; Hôpital Saint-Louis, Paris: O. Cohen-Haguenauer; Centre Viggo-Petersen, Paris: F. Cornélis; Hôpital Tenon, Paris: A. Fajac; GH Pitié Salpétrière et Hôpital Beaujon, Paris: C. Colas, F. Soubrier, P. Hammel, A. Fajac; Institut Jean Godinot, Reims: C. Penet, T. D. Nguyen; Polyclinique Courlancy, Reims: L. Demange†, C. Penet; Centre Eugène Marquis, Rennes: C. Dugast†; Centre Henri Becquerel, Rouen: A. Chevrier, T. Frebourg†, J. Tinat, I. Tennevet, A. Rossi; Hôpital René Huguenin/Institut Curie, Saint Cloud: C. Noguès, L. Demange†, E. Mouret-Fourme; CHU, Saint-Etienne: F. Prieur; Centre Paul Strauss, Strasbourg: J-P. Fricker, H. Schuster; Hôpital Civil, Strasbourg: O. Caron, C. Maugard; Institut Claudius Regaud, Toulouse: L. Gladieff, V. Feillel; Hôpital Bretonneau, Tours: I. Mortemousque; Centre Alexis Vautrin, Vandoeuvre-les-Nancy: E. Luporsi; Hôpital de Bravois, Vandoeuvre-les-Nancy: P. Jonveaux; Gustave Roussy, Villejuif: A. Chompret†, O. Caron.
†Deceased prematurely.
Appendix A. Supplementary data
The following are the Supplementary data to this article:
- Multimedia component 1
- Multimedia component 2
- Multimedia component 3
- Multimedia component 4
- Multimedia component 5
References
- Assessing risk of breast cancer: a review of risk prediction models.J Breast Imaging. 2021; 3: 144-155
- Gene-panel sequencing and the prediction of breast-cancer risk.N Engl J Med. 2015; 372: 2243-2257
- A population-based study of genes previously implicated in breast cancer.N Engl J Med. 2021; 384: 440-451
- CanRisk tool – a web interface for the prediction of breast and ovarian cancer risk and the likelihood of carrying genetic pathogenic variants.Cancer Epidemiol Biomark Prev. 2021; 30: 469-473
- Genetic clinicians’ confidence in BOADICEA comprehensive breast cancer risk estimates and counselees’ psychosocial outcomes: a prospective study.Clin Genet. 2022; 102: 30-39
- Clinical validity assessment of genes frequently tested on hereditary breast and ovarian cancer susceptibility sequencing panels.Genet Med. 2019; 21: 1497-1506
- A case-control evaluation of 143 single nucleotide polymorphisms for breast cancer risk stratification with classical factors and mammographic density.Int J Cancer. 2020; 146: 2122-2129
- Breast cancer risk prediction using clinical models and 77 independent risk-associated SNPs for women aged under 50 Years: Australian breast cancer family registry.Cancer Epidemiol Biomark Prev. 2016; 25: 359-365
- The contributions of breast density and common genetic variation to breast cancer risk.J Natl Cancer Inst. 2015; 107
- Prediction of breast cancer risk based on profiling with common genetic variants.J Natl Cancer Inst. 2015; 107
- Polygenic risk scores for prediction of breast cancer and breast cancer subtypes.Am J Hum Genet. 2019; 104: 21-34
- European polygenic risk score for prediction of breast cancer shows similar performance in Asian women.Nat Commun. 2020; 11: 3833
- Polygenic risk scores for prediction of breast cancer risk in Asian populations.Genet Med. 2022; 24: 586-600
- Evaluating polygenic risk scores for breast cancer in women of African ancestry.J Natl Cancer Inst. 2021; 113: 1168-1176
- Polygenic risk scores for prediction of breast cancer risk in women of African ancestry: a cross-ancestry approach.Hum Mol Genet. 2022; 31: 3133-3143
- Polygenic risk scores and breast and epithelial ovarian cancer risks for carriers of BRCA1 and BRCA2 pathogenic variants.Genet Med. 2020; 22: 1653-1666
- Identification of ten variants associated with risk of estrogen-receptor-negative breast cancer.Nat Genet. 2017; 49: 1767-1778
- Association analysis identifies 65 new breast cancer risk loci.Nature. 2017; 551: 92-94
- Breast cancer risk, nightwork, and circadian clock gene polymorphisms.Endocr Relat Cancer. 2014; 21: 629-638
- GEMO, a national resource to study genetic modifiers of breast and ovarian cancer risk in BRCA1 and BRCA2 pathogenic variant carriers.Front Oncol. 2018; 8: 490
- GENESIS: a French national resource to study the missing heritability of breast cancer.BMC Cancer. 2016; 16: 13
- Night work and breast cancer: a population-based case-control study in France (the CECILE study).Int J Cancer. 2013; 132: 924-931
- Genome-wide association study in BRCA1 mutation carriers identifies novel loci associated with breast and ovarian cancer risk.PLoS Genet. 2013; 9e1003212
- Identification of a BRCA2-specific modifier locus at 6p24 related to breast cancer risk.PLoS Genet. 2013; 9e1003173
- Large-scale genotyping identifies 41 new loci associated with breast cancer risk.Nat Genet. 2013; 45 (61e1-2): 353-361
- The OncoArray consortium: a network for understanding the genetic architecture of common cancers.Cancer Epidemiol Biomark Prev. 2017; 26: 126-135
- Haplotype estimation using sequencing reads.Am J Hum Genet. 2013; 93: 687-696
- Genotype imputation with thousands of genomes.G3. 2011; 1: 457-470
- The international genome sample resource (IGSR): a worldwide collection of genome variation incorporating the 1000 genomes project data.Nucleic Acids Res. 2017; 45: D854-D859
- Familial breast cancer and DNA repair genes: insights into known and novel susceptibility genes from the GENESIS study, and implications for multigene panel testing.Int J Int Cancer. 2019; 144: 1962-1974
- A case-only study to identify genetic modifiers of breast cancer risk for BRCA1/BRCA2 mutation carriers.Nat Commun. 2021; 12: 1078
- A weighted cohort approach for analysing factors modifying disease risks in carriers of high-risk susceptibility genes.Genet Epidemiol. 2005; 29: 1-11
- Genome-wide association study identifies 32 novel breast cancer susceptibility loci from overall and subtype-specific analyses.Nat Genet. 2020; 52: 572-581
- Responsible use of polygenic risk scores in the clinic: potential benefits, risks and gaps.Nat Med. 2021; 27: 1876-1884
Article info
Publication history
Published online: December 09, 2022
Accepted:
November 6,
2022
Received in revised form:
October 26,
2022
Received:
August 24,
2022
Identification
Copyright
© 2022 The Authors. Published by Elsevier Ltd.
User license
Creative Commons Attribution – NonCommercial – NoDerivs (CC BY-NC-ND 4.0) | How you can reuse 
Elsevier's open access license policy
Creative Commons Attribution – NonCommercial – NoDerivs (CC BY-NC-ND 4.0)
Permitted
For non-commercial purposes:
- Read, print & download
- Redistribute or republish the final article
- Text & data mine
- Translate the article (private use only, not for distribution)
- Reuse portions or extracts from the article in other works
Not Permitted
- Sell or re-use for commercial purposes
- Distribute translations or adaptations of the article
Elsevier's open access license policy
