Gene Panel Testing for Breast Cancer Reveals Differential Effect of Prior BRCA1/2 Probability

Simple Summary Previous studies reporting large gene panels in breast cancer have mainly reported on the contribution of additional genes compared to BRCA1/2. We have shown a very large differential additional effect from non-BRCA genes dependent on a priori likelihood of BRCA1 and BRCA2 combined. We have undertaken panel testing beyond BRCA1/2 in 1398 patients with breast cancer and identified 95 (6.3%) with actionable breast cancer genes. The highest rate was found for grade 3 ER+ Her2− breast cancers. Those with the lowest likelihood of BRCA1/2 by Manchester score had a 3-fold higher rate of non-BRCA genes whereas those with predicted rates of ~80% had 5-fold higher rate of BRCA1/2. Unless those referring patients with breast cancer for extended panel testing are certain there is no loss of sensitivity for BRCA1/2 they should opt of a bespoke BRCA1/2 test first in those with high prior likelihoods of BRCA1/2 pathogenic variants. Abstract Whilst panel testing of an extended group of genes including BRCA1/2 is commonplace, these studies have not been subdivided by histiotype or by a priori BRCA1/2 probability. Patients with a breast cancer diagnosis undergoing extended panel testing were assessed for frequency of actionable variants in breast cancer genes other than BRCA1/2 by histiotype and Manchester score (MS) to reflect a priori BRCA1/2 likelihood. Rates were adjusted by prior testing for BRCA1/2 in an extended series. 95/1398 (6.3%) who underwent panel testing were found to be positive for actionable non-BRCA1/2 breast/ovarian cancer genes (ATM, BARD1, CDH1, CHEK2, PALB2, PTEN, RAD51C, RAD51D, TP53). As expected, PALB2, CHEK2 and ATM were predominant with 80-(5.3%). The highest rate occurred in Grade-3 ER+/HER2− breast cancers-(9.6%). Rates of non-BRCA actionable genes was fairly constant over all likelihoods of BRCA1/2 but adjusted rates were three times higher with MS < 9 (BRCA1/2 = 1.5%, other = 4.7%), but was only 1.6% compared to 79.3% with MS ≥ 40. Although rates of detection of non-BRCA actionable genes are relatively constant across BRCA1/2 likelihoods this disguises an overall adjusted low frequency in high-likelihood families which have been heavily pre-tested for BRCA1/2. Any loss of detection sensitivity for BRCA1/2 actionable variants in breast cancer panels should lead to bespoke BRCA1/2 testing being conducted first.


Introduction
In recent years there has been a massive expansion in the use of multi-gene panels to test for breast cancer predisposition. These results have been used as a type of case control study to assess genes for breast cancer associations and the increase in actionable gene identification other than BRCA1 and BRCA2 [1][2][3][4][5][6][7][8][9][10]. These studies have generally shown an almost equal rate of pathogenic variants (PVs) identified in BRCA1/2 compared to all of the additional potentially actionable genes combined. In addition, many of the additional variants have been identified in the lower penetrance moderate risk genes [11], ATM and CHEK2, that occur with a higher population prevalence. The only other actionable breast cancer gene consistently identified at a substantial rate is PALB2, which is now considered to be a high-risk gene [12]. Many studies have also included a range of organ sites, as well as including unaffected individuals [2,[7][8][9].
In general, these studies have concentrated on the headline rates of pathogenic variants (PVs) identified in the additional genes and not on the differential effect of the original likelihood of a BRCA1/2 pathogenic variant. Most studies that have used a 'control' frequency for PVs in the relevant genes have used population databases such as gnomAD [13], which may not be ideal controls for the population tested as they may not represent the rates in certain populations especially those with founder effects. We have assessed the frequency of potentially actionable breast cancer gene variants in a large series of women with breast cancer comparing rates of detection of non-BRCA PVs with BRCA1/2 at different BRCA1/2 likelihood scores. We have also assessed the variant frequencies in different breast pathology subtypes.

Study Population
A total of 5060 women with breast cancer and 1443 control samples were included in the study, male cases were excluded. Of the patients with a breast cancer diagnosis, 4536 were referred to St Mary's, Wythenshawe and the Christie hospitals in Manchester, consented and had blood samples taken for DNA extraction testing of breast cancer genes through the Manchester Centre for Genomic Medicine (MCGM) (Figure 1). MCGM participants agreed to return of results on genes predisposing to breast cancer. The remaining participants were recruited through the Predicting the Risk Of Cancer At Screening (PRO-CAS) study. This study recruited women aged 46-73 years of age from a population screening programme and included 187 breast cancers diagnosed before study entry and 337 after study entry (total n = 524). In addition, 1443 control samples without breast cancer from PROCAS [14] were tested as part of Breast Cancer Risk after Diagnostic Gene Sequencing (BRIDGES). PROCAS women only initially provided consent for return of a polygenic risk score, but also consented to further research genetic testing. We are applying for ethical approval to approach women with panel testing to assess whether they wish to receive results.
PVs were annotated according to the ACMG guidelines [17] and verified using Sanger sequencing. Additionally, the PV rate of all 5060 women with breast cancer tested for BRCA1/2 at the MCGM (1997-2021) and in PROCAS was also assessed, regardless of whether they have had (additional) panel testing. The samples clinically tested at MCGM were annotated by a clinically approved national laboratory hub. Annotation of research results from BRIDGES were initially confirmed by the BRIDGES team and confirmed by our in-house research group (EvV).

Calculated Likelihood of a BRCA1/1 Pathogenic Variant
The pathology adjusted Manchester Scoring System (MS) was used to assess likelihoods of a pathogenic variant in BRCA1/2 [18]. Briefly, each breast cancer in a direct lineage to the index case is scored based on age at diagnosis for each gene (1-6 points) as well PVs were annotated according to the ACMG guidelines [17] and verified using Sanger sequencing. Additionally, the PV rate of all 5060 women with breast cancer tested for BRCA1/2 at the MCGM (1997-2021) and in PROCAS was also assessed, regardless of whether they have had (additional) panel testing. The samples clinically tested at MCGM were annotated by a clinically approved national laboratory hub. Annotation of research results from BRIDGES were initially confirmed by the BRIDGES team and confirmed by our in-house research group (EvV).

Calculated Likelihood of a BRCA1/1 Pathogenic Variant
The pathology adjusted Manchester Scoring System (MS) was used to assess likelihoods of a pathogenic variant in BRCA1/2 [18]. Briefly, each breast cancer in a direct lineage to the index case is scored based on age at diagnosis for each gene (1-6 points) as well as each non mucinous epithelial ovarian cancer (5-8 points). All breast and ovarian cancers in the index individual are scored with an adjustment for pathology. Her2+, low grade and lobular cancers receive minus scores for BRCA1 whilst triple negative and high-grade cancers receive positive scores. Pathology adjusted scores of 15-19 are equivalent to a 10% probability of a BRCA1/2 pathogenic variant with a 20-24-point score equivalent to a 20% probability [18]. The proportion of positive tests from panel testing other than BRCA1/2 was adjusted to reflect previous testing of BRCA1/2 and the proportion of positive tests for BRCA1/2 at each MS score range. Thus a 10% detection rate in BRCA1/2 negative samples where the BRCA1/2 detection was 50% translates to a 5% overall panel detection rate beyond BRCA. An actionable breast cancer gene was defined as a gene confirmed in the BRIDGES study as having a 2-fold relative risk for breast cancer [16].

Participants
A total of 1398 people with breast cancer were included in this study. There were 874 women tested through the MCGM as part of the clinical service or research projects. Of these, 740 women had a family history of only breast cancer, with 134 having an additional personal or family history of ovarian cancer. Additionally, 524 affected women and 1443 female controls that took part in the PROCAS study were included (age range 29-75; median 58.8 years) and were tested as part of the BRIDGES study ( Figure 1).
Overall, 480 women with a family history (34.3%) and MS of −1 (some score below 0 with pathology adjustment) up to 56 (median 18) had been negative for BRCA1/2 on a pre-screen.
Additionally, the PV status of all 5060 women with breast cancer tested for BRCA1/2 was included in order to determine the effect of extended panel testing in women stratified by MS.
A total of 95 of 1398 (6.3%) women who underwent panel testing were found to be positive for actionable non-BRCA1/2 breast/ovarian cancer genes (ATM, BARD1, CDH1, CHEK2, PALB2, PTEN, RAD51C, RAD51D, TP53). As expected, PALB2, CHEK2 and ATM were the most frequent with 80 (5.3%) PVs accounting for 84.2% non BRCA1/2 actionable results. Of the other high-risk genes there were 10 TP53, one CDH1 and one PTEN PVs, but none in STK11. All the expected breast cancer associated genes (ATM, CHEK2, PALB2, TP53) had significantly increased odds ratios of above 2-fold compared to the 1443 PRO-CAS control samples (Table 1). Other PVs were identified in genes that have a less clear association with breast cancer such as: BRIP1, NBN, RAD50 and RECQL, but none of these showed significant associations.

Detection Rate by Manchester Score
The detection rates of PVs in non-BRCA1/2 genes in the full extended panel cohort did not vary substantially by MS (ranging between 5.3-11.4%). However, when controlling for those with prior BRCA1/2 testing, the added value of the extended panel was actually lower in those with a higher MS (i.e., those individuals with a higher probability of a BRCA1/2 pathogenic variant). Only 2.97% of those with MS ≥ 30 tested positive for a non-BRCA1/2 PV when allowing for the higher likelihood of a BRCA1/2 positive result of 63.2% (251/397) in the group of all individuals tested for BRCA PVs.
Of the 918 women in the study cohort without pre-screening, 68 (7.4%) tested positive for PVs in BRCA1 (n = 25), BRCA2 (n = 42) or both (n = 1). There was an increase in BRCA1/2 PV detection rate with increasing MS, similar to the 5060 women tested only for BRCA1/2 PVs.
Of the 871 women tested for variants in the MMR genes, very few PVs were identified. There were only two cases with an MSH6 PV (Supplementary Table S1). No PVs in MLH1 or MSH2 were identified. Similarly, in the control group there was only one MSH6 and one MSH2 PV identified.
A total of 134 women with breast cancer had either a personal (n = 25) or family history (n = 109) of ovarian cancer. Detection rates were similar to those with breast cancer only with 8/134 (6%) having a PV beyond BRCA1/2 (PALB2 = 2) and 2/25 (8%) with breast and ovarian double primaries (Supplementary Tables S2 and S3).

Discussion
The present study has confirmed some utility for breast cancer gene panels extended beyond BRCA1/2 in women with breast cancer. Nonetheless, the increased diagnostic yield is dominated by just three additional genes (PALB2, ATM and CHEK2) accounting for over 80% of non-BRCA1/2 variants found. Additionally, although BRCA1/2 pre-testing increases the rate of pathogenic variant detection in these genes, with higher familial scores represented by the MS, the added benefit from a panel is lower when adjusting for a higher likelihood of BRCA1 and BRCA2 PVs. For instance, an index case with a MS ≥ 40 would have an~80% chance of a BRCA1/2 PV, but only a 1-2% chance of a PV in another actionable gene. In contrast, with a MS ≤ 8 there is only around a 1% chance of a BRCA1/2 PV, but a 5% chance of finding another PV (mainly in ATM or CHEK2). Previous studies have not addressed this large differential effect. For those with a very high likelihood of a BRCA1/2 PV by an accepted method (MS, BOADICEA/CANRISK, BRCAPRO), any loss in sensitivity of identification of a BRCA1/2 PV would mitigate any additional benefit from an extended panel vs. a targeted BRCA1/2 approach. Some loss of sensitivity is accepted in most panel tests but even a drop by 2% for a family with a MS ≥ 40 would be sufficient to warrant bespoke BRCA1/2 testing as the first test. Although most NGS panels claim to identify large single or multiple exon copy number variations (CNVs), the proven sensitivity of these tests is less clear. In Manchester, 20% of non-Jewish BRCA1 PVs are CNVs that in the past have been detected by multiple ligation-dependent probe amplification (MLPA). Given that any additional gene identified is likely to have lower penetrance than BRCA1 or BRCA2 e.g., in a large family with multiple breast/ovarian cancer cases (MS ≥ 40), it is unlikely that this will be the entire explanation for the pattern observed. For instance we observed a RAD51C PV in a large breast/ovary family which probably only fully explains the ovarian cancer predisposition [19], and for the ATM and CHEK2 variants in six families with MS of 30-39 and one CHEK2 in a MS of >40, these are unlikely to provide a full explanation for the pattern of disease in these families (Table 1) [11]. We have previously shown that a pathogenic RAD51D variant in a family with three women with ovarian cancer and four with breast cancer(MS = 55) fully segregated with disease and that BRCA1/2 testing, including RNA analysis was negative [19]. Given that only a two-fold increased likelihood of breast cancer is associated with RAD51D PVs, it is still unlikely that RAD51D fully explains the breast cancer risk in this family, although polygenic risk could add somewhat to the effects of RAD51D. Although results from BRIDGES [16] and other studies clearly differentiate PALB2 as a high-risk gene and CHEK2 as moderate-risk, this is not reflected in the similar ORs for CHEK2 and PALB2 of around eight-fold, but these risk estimates have large confidence intervals due to the small number of controls with identified variants and are based on testing many familial families with pre-screens of BRCA1/2.
Case control analysis is arguably the most informative method to identify breast cancer gene associations as it also provides confirmation of the level of any increased risk. Many recent large-scale studies have published regarding this, including recent evidence that variants in the MMR genes MSH6 and PMS2 may be associated with an increased risk of breast cancer [2,20]. Although the first of these studies was a true case control study, there has been concern that one of the variant calls in MSH6 (c.2945delC; p.(Pro982Leufs)) that was very frequent in cases, may be spurious due to a sequencing artifact [21]. This study also used gnomAD controls [13] rather than a matched, local population control series. The second study was not a true case control study and assessed risk in carriers identified on panel testing without adjusting for the proportion of referred cases with breast cancer [20]. The very high population frequencies of variants in MSH6 and PMS2 without typical Lynch syndrome associated cancers may ironically have falsely elevated the odds ratio [22]. Furthermore, no evidence for any association between these variants in MMR genes and breast cancer has been seen in larger case control studies [1], segregation analysis [23], nor in prospective analysis of PV carriers [24], with the possible exception of MSH6 in BRIDGES, which was borderline significant at two-fold risk [16]. As such, patients with MMR PVs, with the possible exception of MSH6, should not be advised that they are at increased risk of breast cancer.
The lowest detection rate overall both for BRCA1/2 and additional panel genes was for grade 1 cancers with 4.1% and 4.5% respectively. The highest detection rate was for grade 3 ER+ HER2− breast cancers with a non BRCA1/2 panel detection rate of 10%. There has been very little in the literature on the effect of pathology on panel detection rates in non-BRCA1/2 genes although we have recently reported an association with PALB2 and grade 3 [25]. A Brazilian study recently also confirmed a higher detection rate in high grade tumours [26]. Although this was a small study (n = 224; 61 PVs-19 BRCA1/2) and the association was likely driven primarily by BRCA1/2. The BRIDGES study did not assess grade, but there was a clear association between RAD51C, RAD51D, PALB2 and BARD1, with ER− breast cancer and ATM and CHEK2 with ER+ breast cancer with no association of ATM with ER− [16].
The present study does have some limitations. The size of the control population was probably not large enough to generate very reliable odds ratio estimates. We had to use our previous population estimate for TP53 [27], but had no such estimate for other rare genes such as CDH1. The study was not large enough to statistically assert that BRIP1 is not an actionable breast cancer gene [28]. Nor was it large enough to address the variant detection rates for the ovarian susceptibility genes RAD51C and RAD51D that have previously been excluded as breast cancer susceptibility genes [29,30], but which were more recently confirmed as breast cancer genes just reaching two-fold relative risk [16]. There is also some previous evidence that variants in RAD51D may predispose specifically to triple negative breast cancer [16]. The BRIDGES study also only called definitive PVs using NGS and did not call CNVs. As already stated, this could have reduced PV detection sensitivity in the UK population to below 80% for BRCA1. The odds ratios for BRCA1 could therefore be exaggerated by the fact that in-house testing of over 4536 cases in Table 1 included MLPA, since CNVs were not reported in the control samples tested in the BRIDGES study. Despite this, our study also has a number of strengths. This is a large study with well documented family history, including first and second-degree relatives. A high proportion of the study patients had full pathology available, including HER2 status, which was not generally available before 2000. We have also used matching local controls known not to have breast cancer.

Conclusions
Although the trend towards panel testing beyond BRCA1/2 is likely to be irreversible, there are some lessons to be learnt from the current study. The majority of the increased diagnostic yield results from just three genes PALB2, CHEK2 and ATM. Only PALB2 can be considered a strong enough risk factor on its own and the actionability of ATM and CHEK2 pathogenic variants needs to be taken in the context of other risk factors, including a polygenic risk score, ideally through a validated model such as BOADICEA/CanRisk [31]. Women with certain tumours, and in particular grade 1 invasive breast cancers, should be informed that there is a very low likelihood of any meaningful result from panel testing. Finally, the low adjusted diagnostic uplift in those with very high a priori likelihood of BRCA1/2 means that clinicians should ensure that panels used have high sensitivity for BRCA1/2 PVs present in their population or opt for bespoke BRCA1/2 testing in the first instance.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/cancers13164154/s1, Table S1: Results from MSH6 testing of 871 women with breast cancer (524 from PROCAS) and 1443 controls; Table S2: Panel testing results for women with a personal (n = 25) or family (n = 109) history of ovarian cancer; Table S3: Panel testing results for women with a personal (n = 25) history of breast and ovarian cancer. Institutional Review Board Statement: Clinical or research consent was given for extended testing of breast cancer associated genes (approval from the North Manchester Research Ethics Committee, reference 09/H1008/81(PROCAS) and 08/H1006/77).

Informed Consent Statement:
Clinical or research consent was given for extended testing of breast cancer associated genes. No individual identifiable information is in the manuscript.

Data Availability Statement:
Data is available on request.