Frequency of Pathogenic Germline Variants in CDH1, BRCA2, CHEK2, PALB2, BRCA1, and TP53 in Sporadic Lobular Breast Cancer

Invasive lobular breast cancer (ILC) accounts for ∼15% of invasive breast cancer. Its incidence has increased, in line with the greater use of combined hormone replacement therapy until the late 1990s (1, 2). ILC is commonly associated with lobular carcinoma in situ (LCIS), which is considered both a precursor lesion and a risk factor for invasive breast cancer. The risk of invasive cancer after LCIS is 2 to 11 times greater than the risk in general population resulting in a cumulative long-term rate of invasive cancer of 11% to 26% at 15 years, in contrast to ductal carcinoma in situ (DCIS), which is 20 times greater (3, 4). Not all invasive disease post LCIS presents as ILC, although there is an excess of ILC. Molecular studies of coexisting invasive ductal carcinoma (IDC)/DCIS and LCIS identified similar genomic alterations, suggesting that in some cases they may have a common clonal origin (5, 6). Unlike DCIS, LCIS is also a risk factor for developing invasive cancer in the contralateral breast (7). The underlying biological cause for this is not clear.

Both ILC and LCIS have been shown to have higher familial risks than the more common ductal/no special type cancers and are more likely to be bilateral (8–10). We have previously shown that many of the low-risk breast cancer predisposition loci also predispose to ILC and LCIS, with some differential effects between ILC and IDC (11). We have also identified a novel lobular-specific predisposition SNP at 7q34. However, there are little data on the prevalence of the known high and moderate penetrance breast cancer predisposition genes in lobular breast cancer, with the exception of CDH1 as lobular carcinomas are characterized by loss of E-cadherin expression (the protein encoded by CDH1) through somatic alterations. Germline CDH1 variants were initially reported in hereditary diffuse gastric cancer (HDGC; ref. 12) and about 30% of HDGC families with a CDH1 germline variant typically have at least one individual with ILC (13–16). However, germline CDH1 variants in women with ILC and no family history of HDGC are not common (17–20), although there is some evidence that bilateral cases of LCIS ± ILC have a higher incidence of germline CDH1 pathogenic variants (21).

Lobular cancers have been shown to be more frequent among BRCA2 carriers (8.4%) than BRCA1 carriers (2.2%; ref. 22), and there is scanty evidence that CHEK2 and PALB2 variants may be associated with ILC (23, 24). BRCA1 and TP53 are not well described in ILC.

In this study, we report the frequency of rare variants in six known breast cancer predisposition genes (CDH1, BRCA2, BRCA1, TP53, CHEK2, and PALB2) in 1,802 sporadic UK lobular cancers diagnosed in the UK in women ages 60 years or less.

All patients and controls gave written informed consent, and the studies were conducted in accordance with the Declaration of Helsinki and were approved by the following institutional review boards: GLACIER study, MREC 06/Q1702/64; King's Health Partners breast tissue bank, NHS REC ref. 12-EE-0493.

A total of 1,434 cases of ILC (with or without synchronous LCIS) and 368 cases of pure LCIS with no invasive disease diagnosed in women ages 60 or under were included in this study, together with 1,611 controls, between June 6, 2007 and August 28, 2012 (prevalent cases). Bilateral cases were eligible including those with ILC in one breast and IDC in the contralateral breast. The majority of cases (1,380 ILC, 364 LCIS) were recruited through the GLACIER study from 95 hospitals throughout the UK. This study was set up with the specific aim of investigating genetic predisposition to lobular cancer in the UK. These samples were also part of our study of low-risk breast cancer loci in lobular breast cancer (11). A further 58 cases, including four pure LCIS, were recruited through the King's Health Partners breast tissue bank. Cases ages 60 or under were collected in order to enrich for cases likely to have a genetic component to their disease.

Controls were identified by asking cases at the time of recruitment to identify female non-blood relatives or friends who were willing to donate a blood sample. These healthy volunteers were eligible only if they had no personal or family (up to second degree) history of breast cancer, LCIS, DCIS, or benign breast disease. Controls could be of any age, although older individuals were preferred, as they had lived through many of their at-risk years.

Cases and controls donated a blood sample and were asked to complete a self-administered paper-based questionnaire on their family history at the time of recruitment.

After DNA extraction from peripheral blood, Access Array technology (Fluidigm) was used to amplify all exons of CDH1, BRCA2, BRCA1, TP53, CHEK2, and PALB2 using a custom-made targeted sequencing panel consisting of 321 amplicons (Supplementary Table S1). Purified libraries were quantified using Qubit High Sensitivity Assay Kit and sequenced on a HiSeq2500 (Illumina).

Primer sites from the amplicons were trimmed using Btrim and then sequences were aligned to the reference genome (http://www.novocraft.com, GRCh37 version) using Novoalign (Gap opening penalty = 65 and gap extension penalty = 7 thresholds were applied). Picard tools (v1.74 https://github.com/broadinstitute/picard) and Bedtools (v2.17.0) were used to assess coverage. Variant calling was performed using Samtools and annotated using the Annovar tool (25). We optimized the calling based on a set of variants that were positive controls, and Samtools was the optimal caller compared with HaplotypeCaller from GATK. The transcript that was used for each gene is reported in Supplementary Table S1. The frequency of variant alleles from European reference populations was obtained from three sources (1000 genomes, ESP, and ExAC).

Variants were further filtered based on read depth, quality score, and genotypic quality. All variants with a read depth < 10, quality score < 20, or genotypic score < 20 were excluded from the analysis. In addition, variants with an allelic ratio < 0.2 were excluded irrespective of read depth and variants with an allelic ratio < 0.3 and read depth < 50 were also removed.

Variants that had been previously clinically evaluated and deposited in the ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar/) were assigned labels of benign, variants of unknown significance (VUS), conflicting or pathogenic as per ClinVar.

Variants not present in the ClinVar database were considered pathogenic if they were predicted to lead to protein truncation (frameshift indels, stop-gain, stop-loss and splicing variants within 2 base pairs of the splicing junction; Supplementary Table S2) and VUS if they were nonsynonymous substitutions or in-frame indels.

Fisher exact test was used for gene based rare variant analysis for both case–control and case-only analyses. One-sided test was selected because the expectation was enrichment rather than deficit of variants in cases over controls. No adjustments have been made to account for multiple testing. With the current sample size of ILC and controls, we have more than 80% power (alpha = 0.05) to detect variants of combined allele frequency = 0.001 and an effect size of OR = 5.

All putative pathogenic alleles and CHEK2 1100delC variants, identified by the above methods were confirmed by Sanger sequencing (Supplementary Fig. S1).

A total of 1,434 cases of ILC (with or without synchronous LCIS), 368 cases of pure LCIS (with no invasive disease) together with 1,611 controls were analyzed. The median age of cases was 52 years (interquartile range, 9) and of controls was 52 (interquartile range 12; Supplementary Table S3). The mean coverage of our target region was 800 reads across all samples, with an average of at least 40 reads for more than 98% of the target region per sample. Of the 321 amplicons analyzed, seven failed to amplify consistently across the majority of the samples; however, even for these seven the majority of samples had at least 10 reads for 90% of the amplicon (Supplementary Table S4). There was one novel pathogenic variant (PALB2:exon5:c.2487delG) that did not validate by Sanger sequencing as there was no further DNA available.

Case–control analysis of the 1,434 cases of ILC and 1,611 controls revealed an association with putative pathogenic variants in BRCA2, CDH1, CHEK2, and PALB2. No association of putative pathogenic variants was observed in BRCA1 or TP53 (Table 1; Supplementary Table S5).

The strongest evidence of association was with BRCA2, which contained 27 putative pathogenic variants in 1,434 cases and two in 1,611 controls (OR = 15.44; 95% CI, 3.66–65.04, P = 1.7.0 × 10⁻⁷). Variants were spread throughout the gene (Fig. 1A). Five were novel (Supplementary Table S6), two located in exon 11, one in exon 22, and two in exon 27. The novel frameshifts in exon 27 although unlikely to result in loss of function through nonsense mediated decay may have substantial impact on the protein product and were included as putative pathogenic variants. The two controls with pathogenic BRCA2 variants were recruited at the age of 41 and 43 years; both were in exon 11 and have been described previously: rs80359550: c.5946delT: p.S1982fs and rs397507634: c.C2612A: p.S871X.

Case-only analysis showed that BRCA2 variants were more common in younger patients; 7% of women with ILC ≤40 years of age were carriers and 3.4% of women ≤50 years (case-only analysis: ≤50 years vs. > 50 years: OR = 4.83; 95% CI, 1.95–12.01, P = 0.0003; Table 2). The majority (16/27; 59%) of carriers presented with ILC between 40 and 50 years of age. There was a borderline association with first-degree family history of breast cancer (case-only analysis OR = 2.36; 95% CI, 1.05–5.31, P = 0.043) and any family history of breast cancer (OR = 2.25; 95% CI, 1.02–4.98, P = 0.046; Table 3). One case of pure LCIS had a BRCA2 stop-gain variant in exon 25, c.C9294G:p.Y3098X. Case-only analysis showed a significant association of BRCA2 pathogenic variants with ILC compared with LCIS (P = 0.03; Table 1).

Nineteen cases of ILC and five controls (OR = 4.31; 95% CI, 1.61–11.58, P = 0.0017; Supplementary Table S7) had pathogenic variants in CHEK2. Seventeen cases had known pathogenic variants, two with the recently described Norwegian variant: c.C283T:p.R95X (both bilateral); two at c.349A>G (p.Arg117Gly) and the majority (N = 13) being the well-described 1100delC variant. A novel frameshift variant in exon 12 (c.1262delT:p.L421fs) was found in two cases. Four controls also had the 1100delC variant and one a splicing variant in exon15:(c.1462–2A>G). No pathogenic variants were found in women under the age of 40, and there was no association with age (case-only analysis: ≤50 years vs. > 50 years: OR = 0.78; 95% CI, 0.31–1.99, P = 0.65) or first-degree family history of breast cancer (case-only analysis OR = 2.59; 95% CI, 1.02–6.60, P = 0.067; Table 3). However, a significant excess of variants was observed in individuals with any family history of breast cancer (OR = 3.95; 95% CI, 1.42–11.01, P = 0.008).

There was also an association with pure LCIS (OR = 9.90; 95% CI, 3.42–28.66, P = 1.4 × 10⁻⁵; Table 1). Eleven cases had CHEK2 variants, of which eight were the 1100delC variant, one was the c.C283T:p.R95X variant, and there were two novel variants: a frameshift in exon 2 (c.188_189insC:p.L63fs) and a stop gain in exon 6 (c.G697T:p.E233X). Case-only analysis revealed that CHEK2 variants were more strongly associated with LCIS than with ILC (ILC vs. LCIS, P = 0.037; Table 1). Assessment of the pathology reports revealed that 17 of 19 ILC cases with pathogenic CHEK2 variants had associated LCIS identified by the pathologists. Combining the ILC and LCIS cases gives an overall OR of 5.42 (95% CI, 2.10–13.99, P = 6 × 10⁻⁵) for lobular breast cancer.

Eleven cases of ILC had pathogenic PALB2 variants, of which seven were novel (OR = 12.45; 95% CI, 1.60–95.52, P = 0.002; Supplementary Table S8). One control also carried a novel frameshift variant (exon5:c.2050delC:p.P684fs). Like CHEK2, there was no evidence of association with age (all cases were over 40 years old at diagnosis and six were over 50 (case-only analysis: ≤50 years vs. > 50 years OR = 1.12; 95% CI, 0.34–3.68, P = 0.99), or with a family history of breast cancer (Table 3). There was one case of pure LCIS with a germline PALB2 variant in exon 10, c.G3113A:p.W1038X.

Five cases of ILC had pathogenic CDH1 variants, of which one was novel, and none were found in the controls (P = 0.02; Supplementary Table S9). There was an association with age (case-only analysis: ≤40 years vs. >40 years, OR = 13.14; 95% CI, 2.19–78.75, P = 0.02) but none with family history (case-only with a first-degree relative with breast cancer, OR = 2.95; 95% CI, 0.49–17.73, P = 0.24 or with any family history of breast cancer, OR = 2.11; 95% CI, 0.35–12.65, P = 0.4; Table 3). There was one variant found in a case of pure LCIS, and analysis of the pathology reports showed that all the cases except one had evidence of synchronous LCIS with the ILC and in four that the LCIS was bilateral, as described previously (ref. 21; Supplementary Table S9).

There was no evidence of an association with BRCA1, with just one frameshift deletion detected in exon 10 (c.2680_2681del:p.K894fs) in a patient age 38 with bilateral LCIS, and ILC one breast and IDC and DCIS in the other, all ER-positive.

There were no TP53 variants detected in cases or controls.

There was no evidence of an excess of VUS in any of the genes including rs35187787, a rare polymorphism in CDH1 previously associated with breast cancer (OR = 1.23; 95% CI, 0.67–2.27, P = 0.54 for ILC; Supplementary Table S10).

One variant in CHEK2 classified as having conflicting interpretations of pathogenicity, rs77130927 (exon4:c.C538T:p.R180C), was found in 7 ILC cases and 1 control and showed a borderline association with ILC (OR = 7.89; 95% CI, 0.97–64.27, P = 0.05).

The cohort included 61 cases with pure bilateral lobular cancer, of which 11.5% had pathogenic variants, four in CDH1, two in CHEK2, and one in BRCA2. There were 47 cases with ILC/LCIS in one breast and IDC/mixed invasive/DCIS in the contralateral breast, of which 8.5% had a pathogenic variant, two in BRCA2, one in CHEK2, and one in BRCA1.

This is the first analysis of these six known breast cancer genes in an unselected population of ILC and LCIS and unlike other studies we have evaluated the entire coding region of each gene rather than focusing on a selected number of population-specific variants.

Although germline CDH1 variants have been extensively investigated in ILC, our study shows that the majority of rare germline variants found in women with ILC occur in BRCA2, PALB2, and CHEK2.

A previous study has shown that ILC does occur in 8.4% of BRCA2 variant carriers (22), but none have assessed the frequency of BRCA2 variants in a large cohort of cases purely of this special type. Our study shows that BRCA2 variants are the most common pathogenic variants found in ILC, with 7% of ILC cases under the age of 40 carrying a BRCA2 variant and 3% under the age of 50. The variants were distributed throughout the gene with no evidence of a lobular predisposition locus within the gene. Of the previously described variants, only one was a known founder variant, 5946delT (found in two ILC cases), which has previously been associated with pancreatic and ovarian cancers (26). The known pathogenic missense variant D2723H was found in two women with ILC and has been shown to decrease RAD51 nuclear retention even when wild-type BRCA2 is present (27).

The most frequent CHEK2 variant in European populations is the truncating variant c.1100delC, and this has been shown to confer a 2-fold increase in the risk of breast cancer (28, 29). There is also a rare missense variant p.I157T (c.470 T > C, rs17879961) found in 0.005% of non-Finnish European populations, which is associated with a 1.4-fold elevation in breast cancer risk (30); this is classified as a risk allele rather than a pathogenic variant and thus is not included in the list of detected pathogenic CHEK2 variants in Supplementary Table S7. However, this variant showed no association with ILC in our data and was detected in three controls and two cases.

Two other truncating founder variants have been described in CHEK2, mainly in Polish populations: IVS2+1G>A and del5395 (29). The 5,395 bp founder deletion that removes exons 9 and 10 of the CHEK2 gene would not be detected using our technique; however, the IVS2+1G>A variant would be detected and was not present in this data set.

In previous publications, p.I157T and c.1100delC carriers have been reported to be associated with phenotypically different types of breast cancer, with ILC being more common in p.I157T carriers (31–33), and IDC in c.1100delC carriers (34). In our data, p.I157T was not associated with lobular breast cancer, and the most common variant found in CHEK2 was the c.1100delC variant, found in 1.3% of ILC cases in this study, a rate similar to the Breast Cancer Association Consortium (BCAC) study of 1100delC carriers, 1.2% in 4,349 lobular cases (35). The association with p.I157T and lobular histology was originally identified from Polish series and subsequently in a large meta-analysis of p.I157T performed by BCAC where the majority (93%) of the cases again came from Germany/Poland/Finland/Sweden (31), suggesting this may be a population-specific finding.

Due to the small number of pure LCIS, we had limited power to detect associations with LCIS compared with ILC. CHEK2 was the only gene in our study where pathogenic variants were also associated with pure LCIS, particularly c.1100delC, with 3% of pure LCIS cases carrying a pathogenic variant. This has not been described in previous studies of the c.1100delC variant (32–34), although the 2016 BCAC study did show a greater association with in situ (mostly DCIS) tumors (35). The finding that CHEK2 variants were more strongly associated with pure LCIS than ILC suggests that CHEK2 variants may be predisposing to the in situ stage of lobular cancer with not all progressing to the invasive stage. Of the 368 pure LCIS cases in this study, only 37 have follow-up data (median, 81 months; range, 34–333) and 13 developed subsequent invasive breast cancer, 6 ipsilateral and 7 contralateral, as previously described (36). Two of the 13 pure LCIS cases that had developed subsequent invasive disease had pathogenic variants in CHEK2 and one a VUS in CHEK2 (rs564605612); however, a larger sample size with longer follow-up would be required to ascertain whether germline variants in CHEK2 increase the risk of subsequent invasive disease after a diagnosis of pure LCIS. Germline variants in CHEK2 could explain the finding that invasive disease following a diagnosis of pure LCIS occurs with equal frequency in either breast and that these subsequent cancers can be in the form of ILC or IDC (7).

PALB2 was initially thought to be an intermediate-risk cancer predisposition gene such as CHEK2, but more recently the risk of breast cancer has been shown to be similar to that associated with BRCA2. Antoniou and colleagues reported a relative risk of 9.47 (95% CI, 7.16–12.57), compared with the general population breast cancer incidence in the United Kingdom (37), similar to that found in the present study (OR 12.37). As 70% of breast cancers associated with germline PALB2 variants are ER-positive (a similar frequency to that found in BRCA2 variant carriers and in sporadic breast cancer), it is not surprising that we have detected PALB2 variants in ILC. Previously, Cybulski and colleagues showed that 0.5% of ILC (7/1306) carried a PALB2 variant; however, they only screened for the two variants common in Polish populations, 172_175delTTGT and c.509_510delGA (38).

A previous study of germline PALB2 carriers (mainly the c.3113G>A variant) with invasive disease noted a borderline association with synchronous LCIS (39). However, we identified only one individual with pure LCIS carrying a PALB2-truncating variant, rs180177132, in our cohort of 368 LCIS cases.

Our previous study of CDH1 in 50 bilateral LCIS/ILC showed that 8% of bilateral lobular cases had pathogenic CDH1 variants (21). In this current study, we have identified 61 cases of bilateral lobular cancer (the majority included in our previous study) and found that 11% carry pathogenic variants in CDH1, CHEK2, and BRCA2. Although we did not detect any new pathogenic CDH1 variants in the bilateral cases, we did identify two in unilateral cases (one in ILC + LCIS and one in pure LCIS), giving a frequency of 0.1% in unilateral disease, confirming the result of other studies that germline pathogenic variants in CDH1 do not make a large contribution to the familial risk of unilateral lobular breast cancer.

Current UK guidelines for BRCA1 and BRCA2 screening are based on a minimum combined probability of variant identification in 10% of cases. All women with triple-negative breast cancer younger than 50 years are considered to meet this criterion; however, women of this age with ER⁺ breast cancer are eligible only if they have a strong family history of breast cancer or are younger than 30. Although ILC aged 40 years or less is rare (71/1,434, 5% of cases in this study), we have shown that 11% of these cases carry germline variants in known breast cancer susceptibility genes, particularly BRCA2 (7%) and CDH1 (3%). For women 50 years of age or less, there was a 6% chance of having a germline variant and for women 60 years of age or less 5%, as variants in CHEK2 and PALB2 were not associated with younger age. It is likely that this is an underestimate as our methods will not detect some of the large deletions that have been described in BRCA2 (40).

The lack of an association with age for PALB2 variants in our study is supported by the findings of Antoniou and colleagues who showed there was a constant relative risk, irrespective of age, for PALB2 (37). However, unlike our study, Schmidt and colleagues did suggest that there is a relationship with age for the CHEK2 c.1100delC variant (35). PALB2 also did not show an association with a family history of breast cancer and BRCA2 only a weak association unlike CHEK2, which showed a strong association with any family history of breast cancer but not with a history of a first-degree relative with breast cancer. This is similar to the reports in pancreatic cancer series where most germline variants are found in patients without a significant family history of cancer (41).

In conclusion, this study has shown that CHEK2 variants are more frequent in LCIS than in ILC. Although our study is too small to yield stable estimates of associations, it gives useful estimates of prevalence in the UK population, suggesting that variants in this gene predispose to the in situ stage of lobular breast cancer. Longer follow-up would be required to ascertain whether cases of pure LCIS with pathogenic CHEK2 variants develop invasive disease. We have also shown that variants in BRCA2, PALB2, and CHEK2 are more common in ILC than in CDH1 variants, although CDH1 is still the most common germline variant in bilateral lobular cancer. Finally, we have demonstrated that women ages 40 years or less with ILC have an 11% chance of having a germline pathogenic variant indicating that they should be offered genetic screening of BRCA2, CHEK2, PALB2, CDH1, and BRCA1 under current UK guidance. As the majority of the invasive cancers associated with pathogenic mutations in CHEK2, PALB2, or CDH1 are ER-positive, an alternative to risk-reducing surgery would be chemoprevention and yearly screening with MRI, particularly for carriers of pathogenic CHEK2 variants, which are associated with a lower risk of breast cancer than variants in the other genes. Our data do not support routine genetic testing for pure LCIS under 40 as the frequency of germline mutations is age independent (<60 years 3.8%, <50 years 3.3%, <40 years 4.5%), with the exception of bilateral LCIS. However, if pathogenic germline variants in CHEK2 were found to identify a subset of LCIS more likely to develop invasive disease, then genetic testing may be of value as it would identify a group of women with LCIS that would benefit most from chemoprevention.

M.A. Simpson is Head of Human Genetics at Genomics plc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Petridis, R. Roylance, I. Tomlinson, M.A. Simpson, E.J. Sawyer

Development of methodology: C. Petridis, E.J. Sawyer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Petridis, I. Arora, V. Shah, A. Clifford, C. Gillett, S.E. Pinder, R. Roylance, E.J. Sawyer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Petridis, I. Arora, A. Mera, M.A. Simpson, E.J. Sawyer

Writing, review, and/or revision of the manuscript: C. Petridis, I. Arora, S.E. Pinder, R. Roylance, I. Tomlinson, E.J. Sawyer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Petridis, V. Shah, C.L. Moss, A. Mera, A. Clifford, C. Gillett, R. Roylance

Study supervision: C. Petridis, R. Roylance

We thank all the individuals who took part in this study and all the staff who have enabled this work to be carried out, in particular, Maria Troy, Kelly Kohut, Pat Gorman, and Michele Caneppele. This work was supported by a King's Health Partner's Research and Development Challenge Fund (R150505, to E.J. Sawyer), a Breast Cancer Campaign Project Grant (2010NovPR61, to E.J. Sawyer), and the National Institute for Health Research (NIHR) Biomedical Research Centre based at Guy's and St. Thomas' NHS Foundation Trust and King's College London (to C. Petridis and E.J. Sawyer). The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health. Sample collection was funded by a CRUK Project grant (8146, to R. Roylance and E.J. Sawyer).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.