Study participants and tumour material

GENESIS is a national study on HBOC families identified through French family cancer clinics [20]. Index cases were women diagnosed with invasive mammary carcinoma or in situ ductal carcinoma, having at least one sister affected with BC, and with a negative test result for a pathogenic variant in BRCA1 and BRCA2. ATM carriers of a TV or of a rare likely deleterious missense substitution (MS) were identified during the course of a large-scale case-control mutation-screening study (F. Lesueur, PhD, unpublished data, March 2018). Tumour material from 11 of them was assessed in the present study (Table 1). In addition, we investigated tumours from seven HetAT subjects enrolled in the Australian kConFab study [21] (Table 1).

Selection of ATM variant carriers

Individuals included in the study were either homozygous, compound heterozygous or heterozygous carriers of a variant considered pathogenic for A-T disorder. We also selected HetAT BC participants from HBOC families carrying a TV that had not necessarily been reported in A-T families, as well as carriers of a rare likely deleterious MS classified as C65, C55 or C45 according to the Align-GVGD tool as previously described [10, 22]. Carriers of the p.Val2424Gly variant identified in kConFab were not included in this study, because tumour characteristics of carriers of this variant had been already investigated [13, 15, 16]. In total, 41 tumour tissues were available from 3 A-T and 36 HetAT subjects for histopathological review (Table 1).

Pathology review

The Hematoxylin and Eosin Stained (HES) breast tumour tissue was reviewed and scored for morphology features and graded by two pathologists (AVS and GB) using the modified system of Elston et al. [23]. The World Health Organisation classification of tumours of the breast was used to determine histological subtype of ATM-associated tumours, and TNM stage according to tumour size, nodal infiltration and metastasis status [24]. Oestrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) status, as well as the expression of proliferating marker Ki-67, was obtained from histopathology reports held by diagnostic laboratories. When incomplete, hormonal status was determined by immunohistochemistry (IHC) staining at Institut Curie. Tumours were considered HER2+ if they were scored 3+ by IHC or for tumours scored 2+ by IHC if fluorescence in situ hybridisation showed an HER2 gene amplification. Tumours were classified using IHC data according to the St. Gallen molecular subtypes as follows: triple-negative (ER−, PR− and HER2−), HER2-overexpressing (ER−, PR− and HER2+), luminal A (ER+, PR+/−, HER2− and Ki-67 < 20%), luminal B (ER+, PR+/−, HER2− and Ki-67 ≥ 20%), and luminal B/HER2+ (ER+, PR+/−, HER2+ and Ki-67 ≥ 20%) [25].

Morphological features of ATM-associated breast tumours were compared with the series of breast tumours from patients who had surgery at Institut Curie between 2005 and 2006, named the PICBIM series (from the programme incitatif et collaboratif - Cancer du sein: invasion et motilité). None of the PICBIM patients received neoadjuvant treatment. Patients who had developed a previous cancer at any site were excluded, as were known BRCA1 or BRCA2 mutation carriers. In this series, ATM mutation status of participants had not been determined. In total, 516 patients diagnosed with invasive carcinoma and 83 patients diagnosed with in situ carcinoma served as control subjects.

DNA preparation and confirmation of the familial ATM deleterious variant

Tumour DNA was extracted from tumour-enriched areas (with ≥ 50% tumour content when possible) delimited from the most representative HES-stained slides for the 35 tumours for which formalin-fixed, paraffin-embedded (FFPE) material was available (Table 1). The relevant areas were macrodissected from four 10-μm sections, and DNA was purified using the NucleoSpin Tissue protocol according to the manufacturer’s instructions (Macherey-Nagel, Düren, Germany). DNA quantity and quality were assessed using a Qubit fluorometer (Life Technologies/Thermo Fisher Scientific, Carlsbad, CA, USA) and SYBR Green-based qPCR assay (Promega, Madison, WI, USA). Of the 35 available FFPE tumour DNA samples, sufficient quantity and quality to perform subsequent molecular analyses were obtained for 23 of them. Matched blood DNA was extracted with the QuickGene-610L automated system (AutoGen, Holliston, MA, USA) according to the manufacturer’s instructions. The presence of the familial ATM deleterious variant was confirmed in all analysable blood and tumour DNA samples by Sanger targeted sequencing on the Applied Biosystems ABI 3500xL DNA analyser (Thermo Fisher Scientific, Forest City, CA, USA).

Copy number variation analysis

Copy number variation (CNV) analysis using the Affymetrix OncoScan SNP array (Thermo Fisher Scientific, Santa Clara, CA, USA) was performed on the 23 FFPE ATM-associated tumours, including 2 tumours from 2 A-T participants (Table 1). Data were analysed with the Genome Alteration Print (GAP) method, which takes into account both ploidy and large-scale genomic rearrangements [18, 26]. Copy number ranged from zero to eight copies, and all segments exceeding eight copies were ascribed eight-copy status. Chromosome number was estimated by the sum of the copy numbers detected at the peri-centric regions. Output processing files derived from the GAP tool were analysed using VAMP in-house software [27] to define the boundaries of regions recurrently altered in ATM-mutated tumours. Copy loss and gain for near-diploid tumours were called for the segments with zero or one copy and four or more copies, respectively. Copy loss and gain for near-tetraploid tumours were called for the segments with less than or equal to two and six or more copies, respectively. LOH status was ascribed to regions having monoallelic content, regardless of copy number. LOH associated with copy loss was referred as LOH/loss. Breakpoints (changes in the copy number or major allele counts within chromosomes) in each genomic profile were characterised on the basis of resulting absolute copy number profile and after filtering for regions with < 50 SNP variations. Recurrent alterations (CNV, LOH) among the cohort were obtained using the CNTools R package (version 1.24.0; R Foundation for Statistical Computing, Vienna, Austria) and a homemade script. In order to find inherent grouping structure, hierarchical clustering was performed using the alteration status (absence/presence of an alteration) per segment using the Jaccard distance and the Ward linkage function, available in the vegan R package (version 2.3-3).

Validation of LOH status at ATM locus using microsatellites

Tumour and matched blood DNA were evaluated on a subset of participants by using a PCR-based LOH assay with four fluorescence-labelled microsatellite markers (namely D11S1113, D11S1819, D11S2179 and D11S1778) spanning a 14.4-Mbp region encompassing the MRE11A and ATM genes. Capillary electrophoresis was performed on the ABI 3500xL DNA analyser. Raw electrophoretic data were analysed with GeneMarker software version 1.3 (SoftGenetics, LLC, State College, PA, USA) to assess allele ratios. We considered LOH at the ATM locus when the allele ratio fell below 50% in the tumour DNA sample.

Whole-genome sequencing

WGS was performed on four tumour-normal DNA pairs from three participants for whom frozen tumour tissue was available (Table 1). Paired-end libraries were prepared from 2 μg of DNA using the TruSeq DNA PCR-Free Low-Throughput Sample Preparation Kit (Illumina, San Diego, CA, USA) and were sequenced on the HiSeq 2500 instrument (Illumina). Tumour DNA was sequenced at a higher depth of coverage (100×) than the germline counterpart (30×). Sequencing reads were mapped to the reference genome (assembly hg19) using Burroughs-Wheeler Aligner (version 0.7.5a) [28]. Regions of CNV and LOH were identified using the FACETS algorithm (version 0.5.6) [29], and single-nucleotide variations (SNVs) and indels were called using VarScan 2 [30]. Somatic variants were filtered and annotated using an in-house pipeline.

Statistical analysis

Statistical analyses were performed using STATA version 14.1 software (StataCorp, College Station, TX, USA). Two-tailed tests with a 5% significance level were used throughout. Logistic regressions were used to assess the level of association between the presence of an ATM variant and various features of interest when comparing the ATM series with the PICBIM series. Fisher’s exact test (FET) was used to assess molecular subtype differences between ATM-associated tumours and breast tumours from sporadic cases from The Cancer Genome Atlas (TCGA) [31] and from the Norwegian series [32].

Histopathological features associated with ATM variant status

Among the 41 reviewed ATM-associated breast tumours, 36 were invasive carcinomas and 5 were in situ carcinomas. Overall, subjects with invasive carcinoma and subjects with in situ carcinoma from the ATM series tend to be diagnosed at a younger age than subjects from the PICBIM series (mean age 52.4 vs. 56.2 years, P = 0.08 for invasive carcinomas; 45.5 vs. 54.1 years, P = 0.07 for in situ carcinomas). This can be explained by the fact that women related to an A-T child or belonging to an HBOC family are more likely to benefit from early detection of the disease owing to their higher risk of developing BC than the general population. We also compared mean age at diagnosis in participants who developed invasive breast carcinoma between the studies, and we observed no difference (CoF-AT/Retro-AT 51.9, GENESIS 55.9, kConFab 50.6, P = 0.90).

Invasive breast carcinomas developed by HetAT and A-T participants were mostly ductal carcinomas (86%) with an intermediate to high grade (II–III), which was the same as the distribution of histological types and grades found in the invasive tumours from the PICBIM series. With respect to the IHC of tumours arising in ATM variant carriers, 97% of ATM-associated tumours were ER+, which was significantly higher than the proportion of ER+ tumours in the PICBIM series (59%, P = 0.004). Low to moderate lymphocytic infiltration was observed in ATM-associated tumours (data not shown), but this information was not available in the PICBIM, so no comparison could be performed.

Molecular subtypes could be determined for 28 of 36 ATM-associated invasive breast tumours. ATM-associated breast tumours were mostly luminal B (46%) and luminal A (36%), and the distribution of the molecular subtypes differed significantly from that of the PICBIM series. In particular the luminal B and luminal B/HER2+ subtypes were over-represented among tumours developed by HetAT and A-T participants (P = 0.009 and P = 0.005, respectively) (Table 2). Because the PICBIM series might not reflect the distribution of the molecular subtypes of invasive breast tumours in the general population, we also compared the ATM series with a series of 1423 primary breast tumours from a Norwegian population-based survey of women born between 1886 and 1977 [32], as well as with 501 invasive breast tumours characterised with the PAM50 test [33, 34] available in TCGA, after exclusion of carriers of a TV in BRCA1, BRCA2 and ATM [31]. We found that the proportion of luminal B tumours was also significantly higher in ATM-associated tumours than in the Norwegian study (P_FET = 0.03) and in the TCGA series (P_FET = 0.02), whereas the prevalence rates of luminal B/HER2+ tumours and of triple-negative breast tumours in the two latter series were similar to those observed in ATM-associated tumours (Fig. 1).

We next performed two sensitivity tests. First, analyses comparing clinical and histological features of ATM-associated tumours with those of the PICBIM series were repeated after exclusion of four ATM-associated invasive breast tumours, three of which were second primary tumours (T0016-L, T0173-R and T0179-L) whose morphology and histology might have been affected by treatment of the first primary BC. The fourth tumour (T0077-L) was from a patient with synchronous bilateral tumours (T0077-R and T0077-L); we randomly excluded one of the two tumours to take into account only one tumour per patient in the analysis. The results remained unchanged (Table 2).

Second, we repeated the analyses after exclusion of five invasive tumours developed by carriers of a missense variant not reported so far as pathogenic for A-T, namely c.1464G>T (p.Trp488Cys), c.5750G>C (p.Arg1917Thr), c.8158G>C (p.Asp2720His), c.8614C>A (p.His2872Asn) and c.9008A>T (p.Asn3003Ile), to avoid possible misclassification of the deleterious effect of the variant based on in silico prediction only. Again, the results remained unchanged (Table 2).

We also compared features of the five in situ carcinomas (two ER+, two ER− and one with undetermined ER status) of the ATM series with those of the 83 in situ carcinomas from the PICBIM series, and we observed no difference in nuclear grade, tumour size and hormonal status between the two groups of tumours. However, owing to the low number of in situ carcinomas observed in this series, it was not possible at this stage to draw any conclusions about the characteristics of the in situ tumours developed by HetAT participants.

Genome-wide copy number and LOH profiles of ATM breast tumours

High-quality SNP array data were obtained for 23 FFPE breast tumours; 2 tumours were from A-T subjects, and 21 tumours were from HetAT subjects. Tumour ploidy inferred from the absolute segmental copy number profiles and genotype status by the GAP method [26] identified 16 of 23 (70%) near-tetraploid tumours and 7 of 23 (30%) near-diploid tumours. CNVs and regions of LOH were subsequently determined by taking into account the ploidy of each tumour. No evidence of the homologous recombination deficiency (HRD) signature as measured by large-scale state transition genomic signature [18, 35] was observed among the ATM-associated tumours (Fig. 2a).

Because ‘two-hit’ inactivation of the causative gene is regarded as a principal feature of molecular pathogenesis of most hereditary tumours, we next examined the LOH status of ATM-associated tumours at 11q22–23 in the tumours from HetAT participants. LOH was found in 14 of the 21 tumours (67%) developed by HetAT participants. Microsatellite analysis performed on a subset of 14 tumours confirmed the OncoScan results, except for 1 tumour (T0005). The one exception may be due to differences in sensitivity of the two methods. In the subsequent analyses, OncoScan results were not considered for this tumour. Sanger sequencing of the tumour-blood DNA pairs suggested that the ATM wild-type allele was lost in all tumours that underwent LOH at this locus (data not shown).

When we restricted the analysis to the 16 tumours in which biallelic inactivation of ATM was demonstrated (i.e., breast tumours from A-T participants and breast tumours from HetAT participants showing LOH at the ATM locus), we found that copy number losses at 8p, 11q, 13q and 22q corresponded to longer chromosome segments than the ones described in the 23 ATM-associated tumours. The segment loss at 21p11 was the same as the one initially described (Fig. 2c and Table 3).

We also performed unsupervised hierarchical clustering analyses of CNV data. This analysis did not allow separation of ATM-associated tumours according to molecular subtype, LOH status at 11q22 (ATM locus), type of inherited variant (TV vs. MS) or origin of HetAT participants (A-T families or HBOC families) (Fig. 2b). Interestingly, the synchronous bilateral tumours from HetAT patient T0077 showed similar CNV profiles, whereas tumours from the two A-T participants showed quite distinctive features (Fig. 2b and c).

Finally, although the hierarchical clustering of the CNV data did not separate the tumours according to the variant type (TV vs. MS), we performed a sensitivity analysis excluding tumours of the four HetAT participants carrying an MS. This analysis confirmed that loci 8p21, 13q14, 16q13-q24, 17p13-p12 and 21p11 were sites of recurrent alterations found in ≥ 70% of ATM-associated tumours (Additional file 2: Figure S1). However, after exclusion of these 4 tumours, the boundaries of altered loci were extended, and locus 22q11 was lost in only 12 of the 19 analysed tumours.

Comparison with publicly available data

We used the publicly available data from TCGA [31] accessible through cBioportal [38] to investigate whether the genes listed in Table 3 were specifically lost in ATM-associated tumours. A total of 745 TCGA tumour samples with available CNV data were used. Those tumours were from individuals who developed invasive primary breast tumours and did not carry a deleterious variant in ATM, BRCA1 or BRCA2. In addition to LOH at the ATM locus, which was observed more frequently in ATM-associated tumours (67%) than in the TCGA ‘sporadic’ tumours (40.1%) (P = 0.02), several genes at other loci appeared more frequently lost in ATM-associated tumours, including TPTE (21p11.2-p11.1), GSTT1, GSTTP1 and GSTTP2 (22q11.23), as well as LCP1, RB1 (13q14), YWHAE, USP6, RABEP1 and MAP2K4 (17p13.3-p12) (Additional file 3: Table S2).

Deep whole-genome sequencing of ATM-associated tumours

Whole-genome massively parallel sequencing of four ATM-associated breast tumours (T0001-L, T0015-L, T0077-L and T0077-R) and their respective germline DNA was used to characterise the genetic landscape of ATM-associated tumours at base pair resolution. Tumour DNA was sequenced at a mean depth of coverage of 97× (range 82×–104×), and paired blood DNA was sequenced at a mean depth of coverage of 36× (range 35×–37×). CNV patterns obtained from WGS data for frozen tumours T0015-L, T0077-L and T0077-R were compared with CNV patterns obtained in the OncoScan analysis in the corresponding FFPE tumours. Loss/LOH was confirmed by whole-genome analysis for loci 8p21.3, 11q21-q24.2 (containing ATM), 13q13.3-q32.3, 16q22.1 and 17p13.3, whereas discordant results were obtained at locus 21p11.2-p11.1 for one tumour and at locus 22p11.23 for two tumours (Fig. 3a). Divergent ploidy estimations between the WGS analysis (ploidy 3.5) and the OncoScan analysis (ploidy 4) or the use of different tumour sections to prepare tumour DNA may explain these discrepancies. No OncoScan data were available for tumour T0001, but the CNV profile obtained from the WGS data showed LOH at 11q21-q24.2 (containing ATM) and also loss/LOH at 13q13.3-q32.3, 17p13.3 and 22q12.3-q13.31.

In addition we found that the four tumour genomes shared a region of copy number loss/LOH at 6q23.3-q27, which contains ESR1 encoding the ER, as well as a region of copy number loss at 19p13.3-p13.2 measuring 7.9 Mbp (Fig. 3a) and containing 256 genes (Table 3). Going back to the OncoScan data, we found that these two latter regions were indeed altered but in < 40% of the analysed FFPE tumour genomes.

On the basis of our analysis of high-confidence SNVs identified in each ATM-associated tumour genome, we next looked for potential driver mutations. Post-filtering, 51,161 SNVs were identified, 1004 of which were shared by 2 tumours and 29 of which were shared by 3 tumours (Fig. 3b). Only 794 SNVs were shared by the synchronous bilateral tumours T0077-L and T0077-R (Fig. 3b). When analyses were restricted to the coding part of the genome (exome), no genes were found to be altered either in all four tumours or in the two tumours from patient T0077 (Fig. 3c). Six genes were found to be altered in two tumours: MYO1A, DNAH11, SH2D5, ATM, MUC4 and ROS1 (Fig. 3c). However, only DNAH11 variants (c.7134+1G>A and c.9255_9257del) and the nonsense variant in MUC4 (c.11207C>G) are likely to have a deleterious impact on the gene product function and therefore might represent candidate driver mutations. The two ATM somatic variants identified in tumours T0015 and T0077-L were predicted as benign variants according to the Align-GVGD prediction tool (Fig. 3c).