Tumor sequencing is useful to refine the analysis of germline variants in unexplained high-risk breast cancer families

Patients were eligible for inclusion if they had a personal history of breast and/or ovarian cancer, met the criteria for clinical genetic counseling and testing based on the guidelines of the Belgian Society of Human Genetics and were negative for BRCA1, BRCA2, TP53, and CHEK2 pathogenic mutations. Matching tumor material had to be available. All patients signed an informed consent approved by the Ethics Committee of the hospital. Demographic, familial, and clinical data were recorded to calculate the breast cancer (BC) lifetime residual risk and BRCA mutation carrier pre-test probability for each patient using the BOADICEA algorithm [14]. Ten milliliters of blood was drawn from each patient for DNA extraction using the Wizard genomic DNA purification kit (Promega).

Briefly, 1 μg of genomic DNA was processed. Genomic DNA was captured using Agilent in-solution enrichment methodology with their biotinylated oligonucleotide probes library (SureSelect V6 Exome, Agilent Technologies), followed by paired-end 150 bases massively parallel sequencing on Illumina HiSeq4000 to at least 60× average coverage. Sequence capture, enrichment, and elution were performed according to the manufacturer’s instructions and protocols. Image analysis and base calling were performed using Illumina Real-Time Analysis (2.7.7) with default parameters.

Five consecutive 10-μm sections were obtained from the most tumor-representative formalin-fixed and paraffin-embedded (FFPE) sample. A matched hematoxylin and eosin-stained section was used to macrodissect the tumor area. DNA was extracted using the QIAamp DNA FFPE Tissue Kit (Qiagen) and quantified using the Qubit dsDNA high-sensitivity Assay kit (Thermo Fisher Scientific). Tumor WES was performed by Integragen (Ivry, France) with similar capture kit and sequencer as the germline WES. Specifically, a minimum amount of 50 ng of DNA was needed to create the libraries, followed by paired-end 75 bases massively parallel sequencing to at least 120× average coverage.

Both germline and tumor reads were aligned to the reference human genome sequence GRCh37 using Burrows-Wheeler Aligner 0.7.15 (Wellcome Trust Sanger Institute). Duplicate reads were marked and removed using Picard 1.107 (Broad Institute). Local realignment around indels and base quality score recalibration were performed using the Genome Analysis Toolkit 3.3 (Broad Institute). Germline single-nucleotide variants (SNV) and small indels were identified using GATK Haplotype Caller 3.3 whereas somatic SNV and small indels were identified using the Mutect2 algorithm based on GATK Haplotype Caller 3.7 (Broad Institute). Called variants were annotated, filtered, and visualized using Highlander (http://​sites.​uclouvain.​be/​highlander/​), an in-house bioinformatics framework.

We used a list of 735 candidate genes, including 565 genes selected for germline mutation analysis in a previous landmark study of cancer predisposition [15], supplemented with genes implicated in DNA repair or related to BC by literature mining (Table S1). Variants were retained if passing quality criteria (Phred score for quality of mapping > 30, no more than two different haplotypes at the considered position, variant called outside the 3′ end of the supporting reads, absence of strand bias) had an allele frequency in the ExAC database of < 0.015, were considered pathogenic by at least two prediction softwares (among SIFT, CADD, Fathmm, LRT, DEOGEN2, Mutation Assessor, Mutation Taster, and Polyphen2), or affected splicing (estimated by 2 ensemble learning methods [16]).

Germline variants in well-established BC predisposing genes (BRCA1, BRCA2, TP53, PALB2, ATM, CHEK2, CDH1, PTEN, STK11) were manually classified according to ACMG guidelines [17]. Variants classified as VUS, pathogenic, or likely pathogenic were retained, as were the variants with conflicting interpretation in ClinVar [18]. Germline variants in the remaining genes (thus without known association with the phenotype) were retained if meeting the aforementioned sequencing filtering criteria.

We assessed each tumor for somatic variations (point mutations; INDELs; CNAs, i.e., amplifications and deletions; loss of heterozygosity (LOH); and copy-neutral loss of heterozygosity (CN-LOH)) in the genes containing a germline variant. The presence (both positive as negative) or absence of selection pressure of the germline variant in the tumor sample was assessed by the difference in allele balance between tumor and normal (DAB) analysis (chi-square test, p value threshold of 0.05 for significance) derived from the allelic depths in both samples. DAB of the considered genomic region was further considered by analysis of the behavior in the tumor of each germline heterozygous SNV on the given chromosome. The Benjamini-Hochberg procedure was used to correct for multiple testing, and p values were plotted as in Manhattan plots from genome-wide association studies. True DAB was retained only if the region surrounding the germline variant depicted significant p values for DAB. CNAs were assessed by the FACETS algorithm [19]. Regarding the locus of the germline variant, loss of heterozygosity (LOH) was defined as the loss of the normal allele in the tumor. Copy-neutral LOH (CN-LOH) was defined as a diploid status with DAB in the tumor. Homozygous deletion (HZ-DEL) was defined as the loss of both alleles. Amplification was defined as a copy number status ≥ 6, similar to the threshold used when considering clinically meaningful ERBB2 amplification [20]. The validity of the allele calls was cross-checked with the DAB analysis. Somatic mutations in a gene carrying a considered germline variant were called using Mutect2, as described above, and retained only if the germline variant did not display negative pressure selection in the tumor.

Global patterns of somatic variants were analyzed using complete WES data. We analyzed mutational signatures and quantified the contribution of the known COSMIC signatures (http://​cancer.​sanger.​ac.​uk/​cosmic/​signatures) to the observed somatic mutational processes using the R package MutationalPatterns [21]. Homologous recombination deficiency (HRD) was determined in each tumor sample. We processed the data derived from FACETS to calculate three different HRD scores (telomeric DAB, large-scale state transition, and genomic LOH) combined to a global mean HRD score, using the R scripts kindly made available by Nathanson and Pluta et al., described elsewhere [22]. We used MutSigCV to identify significantly mutated genes [23]. Tumor mutation burden (TMB) was defined as the ratio of the number of somatic variants detected (after the exclusion of germline variants) and the size of the capture kit (60 Mb).

All analyses downstream of the variant calling were performed in R (version 3.5.1, http://​www.​R-project.​com). Data visualization was obtained with in-house developed scripts, with the Gviz and GenVisR packages [24, 25], or with ProteinPaint [26].

RNAs were extracted from lymphocytes with TriPure (Roche) and retro-transcribed using RevertAid H-Minus First Strand cDNA Synthesis Kit (Fermentas), with random hexamers. PCR amplification was done using specific primers, available upon request. Amplicons were cloned into pCRII-TOPO Vector (Invitrogen). Plasmids were purified with PureYieldTM Plasmid Miniprep System (Promega) and Sanger sequenced.

MSCV-human Erbb2-IRES-GFP was a gift from Martine Roussel (Addgene plasmid # 91888; http://​n2t.​net/​addgene:​91888; RRID:Addgene_91,888) [27] and served as a template for mutagenesis (the considered variant and the positive control V777L described in Bose et al. [28]). Primers for the mutagenesis (available upon request) were designed using QuikChange Primer Design (Agilent). The entire coding sequence was verified using Sanger sequencing before and after the insertion in a lentiviral vector. HEK293T cells were grown in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin. As the overexpression of wild-type ERBB2 has an oncogenic effect which could prevent us from seeing the effect of the mutations, we artificially reduced the number of ERBB2 proteins expressed in each cell, by transfecting a mix (5% ERBB2 plasmid and 95% of empty lentiviral vector) of plasmids into the HEK293T cells using jetPEI® (Polyplus, France) according to the manufacturer’s instructions. Protein lysates were homogenized using a 21-gauge needle and resolved on precast polyacrylamide gels (Bio-Rad). Primary antibodies were purchased from Cell Signaling Technologies: ERBB2, phospho-ERBB2(Y1248), EGFR, phosphor-EGFR(Y1068), phospholipase C gamma (PLCγ), phospho-PLCγ(Y783), and alpha-actinin and used at recommended dilutions. Separate membranes were used for total and phospho-antibodies. Visualization was performed with an anti-rabbit secondary antibody (BioSource) at 1:10,000 dilution with a femto-sensitive ECL detection system (Pierce).

Immunohistochemistry was performed on 4-μm paraffin sections. Heat-induced antigen retrieval was performed in a PT-link pre-treatment module (DAKO, Agilent Technologies). After endogenous peroxidase blocking, sections were incubated overnight at 4 °C with a PMS1 rabbit polyclonal primary antibody (1:100 dilution, 10859-1-AP, Proteintech). After three washes with TBS-Tween, sections were incubated with HRP-conjugated anti-rabbit polymer (Envision, DAKO) for 30 min at room temperature, and immunoreactivity was revealed using 3′3′-diaminobenzidine.

We could collect germline and tumor DNA for 70 unrelated BC patients. Patient characteristics are summarized in Table 1, depicting a population at high risk for HBC. Only one patient was male.

We identified 329 rare variants in 139 genes across the 70 patients (mean ± SD, 4.7 ± 2.0; range, 0–12): 293 (89%) missense SNVs or in-frame INDELs, 12 (4%) nonsense SNVs or frameshift INDELs, and 24 (7%) splice region variants (Table S2). Missense variants were predicted pathogenic by 4.5 ± 1.8 algorithms (mean ± SD). In two patients (including the sole male), no variant satisfied our filtering criteria. There were 18 variants in a well-established BC-predisposing gene. Of these, only one was a known pathogenic variant (PALB2 NM_024675 c.509_510delGA).

We detected a possible DAB of the variant (i.e., significantly different ratio of variant to reference allele, in tumor compared to blood) for 57 (17%) variants. This was confirmed by chromosome-wide SNV allele ratio analysis for 36 variants (11%), in 22 patients (31%) (Figure S1), including the PALB2 pathogenic mutation.

Genome-wide aneuploidy (cumulative size of amplifications, LOH, and deletions) in the tumors ranged from 0.1 to 56% (median 13.6%). Per tumor, aneuploidy covered 3–114 (median 32.5) of the 735 genes.

Somatic variation at the genomic coordinates of the 329 germline variants is depicted in Fig. 1 (lower panel). Fifty-five germline-mutated genes (16% of the retained variants) from 32 patients presented a somatic CNA or second mutation in the matched tumor. Of the 55-s hits, two were somatic second mutations: TSC2 NM_000548 c.774G>C in the tumor of CABR47, but presumably a passenger mutation (8% allele frequency (AF)); TP53 NM_001126112 c.365_366delTG in the tumor of CABR45, likely the driver mutation (20% AF). Of the 53 somatic CNAs, 28 would result in enrichment and 25 in depletion of the germline variant in the tumor. Genome-wide aneuploidy size did not differ between the samples with or without a somatic second hit (Mann-Whitney U test, p = 0.80).

We next broadened our analysis to the global patterns of somatic variation observed in tumor data and assessed for whether these patterns could also be exploited to prioritize additional variants. CABR95, carrying the germline pathogenic PALB2 mutation enriched in the tumor by somatic LOH, had a high mean homologous recombination deficiency (HRD) score, in keeping with the expected effects of loss-of-function of this gene (Figure S2A). CABR46 carried an interesting candidate germline variant in PMS2 (NM_000535 c.1937G>T): rare (minor allele frequency (MAF) 0.012% in the European non-Finnish population, with no homozygotes in the GnomAD database) and predicted damaging by 6 algorithms. No somatic second hit in PMS2 was detected in the matched tumor sample. Nevertheless, this tumor was hypermutated (5444 somatic variants, tumoral mutation burden (TMB) of 90 variants/Mb) (Fig. 1, upper panel) and had the highest number of indels in the series (Figure S3). C>T transitions predominated (44%), and signatures 6 and 20 (related to mismatch repair deficiency) accounted for 15% of the tumor mutational profile in this sample mainly characterized by signatures 1 and 5 related to aging and deamination, respectively (Fig. 2). Although no second hit was found, these data orient towards the probable existence of mismatch repair deficiency, supporting a causal role for the PMS2 variant and the presence of a second hit of a different type (e.g., an epigenetic event). The detection of microsatellite instability through the exome data, using MSIsensor [29], was also not definitely conclusive; no sample reached a score of 10 considered in one study to define microsatellite instability [30]. Nevertheless, CABR46 had an outlier score as compared to the other samples (1.3 vs 0.4, Figure S4), as did an in-house positive control case of breast cancer with proven microsatellite instability (3.2, Schröder et al., unpublished). The exhaustion of tumor material unfortunately prevented us from quantifying microsatellite instability using PCR amplification of validated microsatellites loci. In contrast, CABR19, who had bilateral BC before 38 years of age and carried an even stronger germline candidate variant on PMS2 (c.883C>T; one occurrence in the European non-Finnish population in GnomAD, predicted damaging by all algorithms) had low TMB (1.4/Mb) and somatic indel count [4], a very low MSIsensor score (0.01), and did not display signatures 6 and 20. This supports the absence of a somatic second hit inactivating PMS2.

CABR51 had the third most highly mutated tumor (395 somatic variants of which 47% were C>T transitions, TMB of 6.6 variants/Mb) with the second-highest number of indels (Figure S3). This could be related to the germline variant on NTHL1 (NM_002528 c.527 T>C; MAF 0.21% in GnomAD including only one homozygote, and predicted damaging by all algorithms), enriched in the tumor by CN-LOH. Signature 30 was the main contributor to this somatic mutational profile, a feature unique to this tumor sample. In contrast, CABR90 also had a germline NTHL1 variant (c.298 T>C; MAF 0.12% in GnomAD, predicted damaging by 7 algorithms), but with a balanced amplification of the region in the tumor. NTHL1 is expected to act as a tumor suppressor gene, arguing against amplification as a bona fide second hit. This was underscored by the low TMB (1 variant/Mb), low indel count (1 indel), and the absence of signature 30 in this tumor.

Despite the suboptimal conservation of the FFPE tumor samples, the validity of the WES data was strongly supported by expected observations. Mutational signatures related to aging (signature 1), activity of the APOBEC cytidine deaminases (signatures 2 and 13), and HRD (signature 3) predominated in tumor samples (Fig. 2) [31]. The contribution of signature 3 was significantly more pronounced in triple-negative breast cancer (TNBC) (Figure S2B) [32]. APOBEC deregulation was predominant in ERBB2-overexpressing cases (Figure S2C) [33]. The HRD score was higher in TNBC than in the other subgroups (Figure S2D). This score correlated with the relative contribution of signature 3 (R² = 0.13, p = 0.002, Figure S2A). Significantly, CABR95, carrying the germline pathogenic PALB2 mutation enriched by somatic LOH, had a high mean HRD score, in keeping with the expected effect of loss-of-function of this gene. MutSigCV identified significantly greater-than-expected somatic mutation rates of TP53, PIK3CA, and GATA3 (Table S3) [34].