Somatic mutations and copy number variations in breast cancers with heterogeneous HER2 amplification

Intratumour heterogeneity fuels carcinogenesis and allows circumventing specific targeted therapies. HER2 gene amplification is associated with poor outcome in invasive breast cancer. Heterogeneous HER2 amplification has been described in 5–41% of breast cancers. Here, we investigated the genetic differences between HER2‐positive and HER2‐negative admixed breast cancer components. We performed an in‐depth analysis to explore the potential heterogeneity in the somatic mutational landscape of each individual tumour component. Formalin‐fixed, paraffin‐embedded breast cancer tissue of ten patients with at least one HER2‐negative and at least one HER2‐positive component was microdissected. Targeted next‐generation sequencing was performed using a customized 53‐gene panel. Somatic mutations and copy number variations were analysed. Overall, the tumours showed a heterogeneous distribution of 12 deletions, 9 insertions, 32 missense variants and 7 nonsense variants in 26 different genes, which are (likely) pathogenic. Three splice site alterations were identified. One patient had an EGFR copy number gain restricted to a HER2‐negative in situ component, resulting in EGFR protein overexpression. Two patients had FGFR1 copy number gains in at least one tumour component. Two patients had an 8q24 gain in at least one tumour component, resulting in a copy number increase in MYC and PVT1. One patient had a CCND1 copy number gain restricted to a HER2‐negative tumour component. No common alternative drivers were identified in the HER2‐negative tumour components. This series of 10 breast cancers with heterogeneous HER2 gene amplification illustrates that HER2 positivity is not an unconditional prerequisite for the maintenance of tumour growth. Many other molecular aberrations are likely to act as alternative or collaborative drivers. This study demonstrates that breast carcinogenesis is a dynamically evolving process characterized by a versatile somatic mutational profile, of which some genetic aberrations will be crucial for cancer progression, and others will be mere ‘passenger’ molecular anomalies.

Intratumour heterogeneity fuels carcinogenesis and allows circumventing specific targeted therapies. HER2 gene amplification is associated with poor outcome in invasive breast cancer. Heterogeneous HER2 amplification has been described in 5-41% of breast cancers. Here, we investigated the genetic differences between HER2-positive and HER2-negative admixed breast cancer components. We performed an in-depth analysis to explore the potential heterogeneity in the somatic mutational landscape of each individual tumour component. Formalin-fixed, paraffin-embedded breast cancer tissue of ten patients with at least one HER2-negative and at least one HER2-positive component was microdissected. Targeted next-generation sequencing was performed using a customized 53-gene panel. Somatic mutations and copy number variations were analysed. Overall, the tumours showed a heterogeneous distribution of 12 deletions, 9 insertions, 32 missense variants and 7 nonsense variants in 26 different genes, which are (likely) pathogenic. Three splice site alterations were identified. One patient had an EGFR copy number gain restricted to a HER2-negative in situ component, resulting in EGFR protein overexpression. Two patients had FGFR1 copy number gains in at least one tumour component. Two patients had an 8q24 gain in at least one tumour component, resulting in a copy number increase in MYC and PVT1. One patient had a CCND1 copy number gain restricted to a HER2negative tumour component. No common alternative drivers were identified in the HER2-negative tumour components. This series of 10 breast cancers with heterogeneous HER2 gene amplification illustrates that HER2 positivity is not an unconditional prerequisite for the maintenance of tumour growth. Many other molecular aberrations are likely to act as alternative or collaborative drivers. This study demonstrates that breast carcinogenesis is a dynamically evolving process characterized by a versatile somatic mutational profile, of which some genetic aberrations will be crucial for cancer progression, and others will be mere 'passenger' molecular anomalies.

Introduction
Cancer is a genetic disease, resulting from an accumulation of successive somatic gene mutations that drive cancer cell proliferation (Tomasetti et al., 2017). Invasive breast cancer is heterogeneous and comprises different molecular subtypes (Perou et al., 2000). Around 12-20% of invasive breast cancers have a HER2 gene amplification, which generally results in overexpression of the HER2 protein (Kraus et al., 1987;Ross, 2010;Venter et al., 1987). The HER2 gene, located at 17q12-21, encodes a 185 kDa transmembrane tyrosine kinase receptor. The HER2 receptor has no known ligand of its own but activates other receptors of the HER family by heterodimerization (Barros et al., 2010). HER2 amplification is associated with shorter disease-free and overall survival in patients with node-negative and node-positive invasive breast cancer treated with adjuvant chemotherapy and/or local radiation (Slamon et al., 1987(Slamon et al., , 1989. With the advent of the humanized monoclonal anti-HER2 antibody trastuzumab (Herceptin; Genentech, San Francisco, CA, USA), HER2 has evolved from a mere prognostic marker to a predictive marker and a target for therapy (Ross and Fletcher, 1999). Since then, the anti-HER2 treatment arsenal has substantially expanded, and current therapeutic options include trastuzumab, pertuzumab (Perjeta; Genentech), trastuzumab emtansine or T-DM1 (Kadcyla; Genentech) and lapatinib (Tykerb; GlaxoSmithKline, Brentford, UK).
Most HER2-positive carcinomas, both in situ and invasive, present with homogeneous HER2 overexpression and amplification, implying that it is a key molecular event that propels cancer cell proliferation. Such genetic events occur early in the process of carcinogenesis and are designated 'truncal' somatic events (McGranahan et al., 2015). However, an intratumoral heterogeneous pattern of HER2 amplification is not uncommon. Heterogeneity has been described in 5-41% of HER2-positive breast cancers, depending on its definition (Cottu et al., 2008;Ng et al., 2015;Ohlschlegel et al., 2011). The latest ASCO/CAP guidelines do not define intratumour heterogeneity (Wolff et al., 2018), but previous studies discerned regional from genetic heterogeneity (Bartlett et al., 2011;Hanna et al., 2014;Seol et al., 2012;Vance et al., 2009). Genetic HER2 heterogeneity is defined as > 5% and < 50% of infiltrating tumour cells presenting with a HER2 copy number ≥ 6 (Pekar et al., 2019;Vance et al., 2009). Regional heterogeneity comprises an amplified tumour component admixed with a negative and/or equivocal tumour component based on immunohistochemistry and ISH studies (Bartlett et al., 2011;Cottu et al., 2008;Seol et al., 2012). The observed heterogeneity suggests that in some tumours, not all cancer cells are depending on the HER2 oncogene. Other genomic aberrations might act as potent alternative drivers of cancer cell proliferation and invasion in HER2-negative subclones, such as the previously identified BRF2 and DSN1 gene amplification and the HER2 p.I767M somatic mutation (Ng et al., 2015).
In the current study, we aimed to further explore the landscape of somatic mutations and copy number variations (CNVs) in HER2-heterogeneous breast cancers. We performed an in-depth analysis of ten breast cancers containing at least two distinct components with different HER2 expression and copy number profiles, designated regional HER2 heterogeneity. We investigated whether these immunohistochemically distinct components were clonally related and whether the HER2-negative components were associated with specific molecular aberrations that might act as alternative drivers of carcinogenesis.

Patient samples
This retrospective study collected formalin-fixed, paraffin-embedded (FFPE) tissue samples from 10 breast cancer patients who were treated between 2010 and 2018 at the Erasmus Medical Center Cancer Institute (Rotterdam, the Netherlands). Coded leftover patient material was used in accordance with the Code of Conduct of the Federation of Medical Scientific Societies in the Netherlands (FEDERA, 2011), as previously described (Agahozo et al., 2019). The study methodologies conformed to the standards set by the Declaration of Helsinki. The study methodologies were approved by the local ethics committee. Both core biopsies and resection specimens were eligible. Any histological type of breast cancer was included, provided that the tumour presented with regional heterogeneous HER2 amplification and corresponding heterogeneous HER2 protein overexpression, as previously described (Seol et al., 2012). Heterogeneous HER2 status was defined as the presence of at least one HER2-positive in situ and/or invasive component and at least one HER2-negative in situ and/or invasive component, as demonstrated by immunohistochemical and in situ hybridization (ISH) analysis. These different components had to be in close proximity of one another: all components were present in a single tissue block (with the exception of axillary lymph node metastases, if present). One representative tissue block was selected for all subsequent analyses.

Immunohistochemistry
Four-µm-thick FFPE tissue sections were mounted on Superfrost plus slides (Menzel-Gl€ aser, Braunschweig, Germany). Immunohistochemical stainings for oestrogen receptor (ER), progesterone receptor (PR), the myoepithelial cell marker p63, E-cadherin, HER2, FGFR1 and EGFR were performed using an automatic immunostainer (Benchmark XT; Ventana Medical Systems, Tucson, AZ, USA), according to the manufacturer's instructions (Table S1). HER2 expression was assessed according to the ASCO/CAP guidelines (Wolff et al., 2018). ER expression and PR expression were scored as percentages, regardless of the intensity. Hormone receptor status was determined according to the ASCO/CAP guidelines (Hammond et al., 2010). Surrogate molecular intrinsic subtyping was based on the combined ER/PR/HER2 status.

HER2 in situ hybridization analysis
Automated HER2 ISH analysis was performed on all cases using the BenchMark ULTRA (Ventana Medical Systems). Four-µm-thick FFPE tissue sections were deparaffinized and incubated with cell conditioning 2 (CC2) buffer at 86°C for 28 min. Tissue sections were treated with ISH-Protease-3 at 36°C for 12 min, followed by HER2 probe denaturation at 96°C for 8 min and hybridization at 80°C for 6 min. UltraView SISH was used for detection, and haematoxylin II was used as counterstain. Tumour components were considered HER2-amplified when a mean HER2 copy number of ≥ 6 per cell was observed, in accordance with the ASCO/CAP guidelines (Wolff et al., 2018).

DNA extraction
All tissue sections were first reviewed by two breast pathologists (MRVB and CHMVD) who selected tumour areas with an estimated minimum tumour cell percentage of 30%. Ten consecutive FFPE 5-µm-thick tissue sections were deparaffinized and haematoxylinstained prior to microdissection. Selected tumour areas and normal tissue areas were microdissected manually into 5% Chelex 100 Resin (Bio-Rad, Hercules, CA, USA) Cell lysis solution (Promega, Madison, WI, USA), using a sterile scalpel. DNA was extracted by proteinase K (Roche, Mannheim, Germany) digestion by overnight incubation at 56°C. Proteinase K was inactivated at 95°C for 10 min. Finally, the samples were centrifuged for 5 min at 20 000 g to remove remaining cell debris and Chelex resins. The DNA was collected into new tubes and stored at À80°C until further use. DNA concentrations were measured by a Qubit 2.0 fluorometer (Thermo Fisher Scientific, Waltham, MA, USA).

Targeted next-generation sequencing
For targeted next-generation sequencing (NGS), a custom-made amplicon panel was applied. This panel comprised 2778 amplicons covering 53 genes (Table S2), including single nucleotide polymorphisms (SNPs) and hotspot mutation regions. Gene selection for this panel was based on two large tumour profiling studies (ICGC/TCGA and METABRIC), as well as frequently found driver mutations in breast cancer (Koboldt et al., 2012;Nik-Zainal et al., 2016). The Ion AmpliSeq Designer tool was used to design amplicons for the multiplex PCR assay, thereby aiming for 150bp amplicons and allowing efficient amplification of fragmented DNA isolated from FFPE tissue. Full sequence coverage of large exons required amplification and sequencing of overlapping amplicons. Therefore, the multiplexed PCR was split into two reactions, using 10 ng of DNA for each reaction. The Ion AmpliSeq Library Kit Plus (Thermo Fisher Scientific) protocol was used to process the samples analysed by the Ion AmpliSeq custom 53-gene panel, according to the manufacturer's instructions. Each sample was barcoded using IonXpress barcode adapters, allowing multiplexed sequencing. A total of 18 PCR cycles were performed. Ten samples were multiplexed on an Ion 540 Chip and sequenced on the Ion S5XL Semiconductor Sequencer (Thermo Fisher Scientific, Waltham, MA, USA).

Mutation analysis
The VARIANT CALLER v5.6.0.4 (Thermo Fisher Scientific) was used for variant calling. Filtering was performed by the 'somatic low stringency' default of the Torrent Variant Caller. Variants were annotated in a local Galaxy pipeline (www.galaxyproject.org) using ANNOVAR (Wang et al., 2010). Exonic and splice site variations were selected for analysis. Synonymous point mutations, as well as variants identified as common polymorphisms in the 1000 Genomes database (with a frequency of > 1%), were removed from the dataset. Variants were kept in the dataset if they had a minimum read depth of 100 reads and if they were present within a tumour component with a frequency higher than 10%. Variants were excluded if a strand artefact was suspected (forward/reverse or reverse/forward ratio of < 1/10). For each case, a patientmatched normal tissue sample was analysed to verify whether the identified variants were somatic or germline. Pathogenic and likely pathogenic variants were considered germline if their variant allele frequency ranged within 45-55% in the normal tissue sample.
Four prediction algorithms, MUTATIONTASTER (http:// www.mutationtaster.org/), PROVEAN (http://provean.jc vi.org/index.php), UMD-PREDICTOR (http://umd-predic tor.eu/) and SIFT (https://sift.bii.a-star.edu.sg/), were used to predict the effects of coding nonsynonymous variants. The Catalogue of Somatic Mutations in Cancer (COSMIC; https://cancer.sanger.ac.uk/cosmic) was interrogated to assess for previous reports on the selected variants. Variants were selected when at least three of the four prediction algorithms indicated that the variant was pathogenic or probably pathogenic. If this criterion was not met, the variant was retained only if the COSMIC database indicated it was a known pathogenic or likely pathogenic variant. All variants were reported at the cDNA level (c. annotation) and the protein level (p. annotation) according to the Human Genome Variation Society (HGVS) nomenclature (Richards et al., 2015).

Copy number variation analysis
The presence of high-level gene copy number gains was investigated by using the relative coverage, as previously described (Eijkelenboom et al., 2019). Sample normalization was performed to correct for differences in the number of total reads. The normal tissue samples of all patients constituted the reference series. The normalized coverage of the reference series was calculated by dividing the number of reads for each amplicon by the total number of reads for each normal tissue sample. The arithmetic mean was calculated for each amplicon, based on all samples in the reference series. The coverage of each amplicon from the tumour tissue samples was normalized by dividing the number of reads by the total number of reads per tumour tissue sample. The relative coverage for each amplicon of the tumour tissue samples was calculated by dividing the normalized coverage of the sample by the mean normalized coverage of the reference series (Eijkelenboom et al., 2019). Copy number gains were suspected when at least five amplicons clustered together, provided that the log2 scale of the relative coverage amounted > 1.5. Copy number losses were not investigated as the presence of background (due to the use of FFPE tissue samples) hampered reliable interpretation of the presence of copy number losses. Visualization of CNVs was achieved by the construction of scatter plots in MS Office Excel (Windows, Washington, WA, USA).

Patient population
Ten patients with a breast cancer with spatially heterogeneous HER2 amplification were included in this study. This series included eight patients with invasive carcinoma of no special type (NST) and associated ductal carcinoma in situ (DCIS), one patient with invasive lobular carcinoma (ILC) and associated lobular carcinoma in situ (LCIS), and one patient with metaplastic carcinoma (MC; spindle cell type) and associated DCIS. Figure 1 illustrates the presence of a HER2-positive and a HER2-negative DCIS component, associated with a HER2-positive invasive MC (patient #1). Figure 2 demonstrates the presence of a HER2-positive and HER2-negative LCIS component associated with a HER2-negative invasive component (patient #3). In some patients, heterogeneous HER2 amplification was associated with heterogeneous hormone receptor status as well ( Fig. S1; Table 1). All patients underwent nodal staging. Six patients had no sentinel lymph node metastases. Patient #10 had seven axillary macrometastases, with sufficient tissue available for targeted NGS. Patients #4, #9 and #5 had a sentinel lymph node with isolated tumour cells, a single micrometastasis and a single macrometastasis, respectively. These metastases were not analysed due to insufficient amounts of available tumour tissue.

Coverage and mutation analysis
Sufficient DNA for sequencing was extracted from all but two microdissected tissue samples (Table 1). The mean percentage of amplicons with at least 100 and 500 reads was 94.4% and 78.5%, respectively, with an average base coverage depth of 2216 (Table S3). No pathogenic or likely pathogenic somatic variants were detected in ARID1B, BRCA2, CCND3, CHECK2, ERBB2, ERBB3, MAP2K4, MLL, NCOR1, NOTCH1, PBRM1 and PDGFRA. Overall, germline pathogenic variants were not observed. We identified 63 pathogenic or probably pathogenic variants in 26 different genes (Table S4), based on four prediction tools and the COSMIC database. These variants included 12 deletions, 9 insertions, 32 missense variants, 7 nonsense variants (with introduction of a stop codon) and 3 splice site alterations. These somatic aberrations were commonly found in ARID1A, MLL3, NF1, PIK3CA and TP53 (Fig. 3). The tumour suppressor gene TP53 was mutated in at least one component in 7 out of 10 breast cancers (70%). The TP53 aberrations included five missense variants, two deletions and one splice site change. The presence of a TP53 mutation was homogeneously present in all components of the breast cancers of patients #4, #5 and #8. Patients #1, #2, #7 and #10 each presented with a tumour with heterogeneous presence of a TP53 mutation (Table 2) Immunohistochemistry for HER2 (B) demonstrates a positive 3+ score in the HER2-amplified LCIS component (orange squares) and an equivocal 2+ score in the HER2 nonamplified LCIS component (blue diamond) and the ILC of classic type (grey triangle; original magnification 50xscale bar size = 500 µm). The scatter plot confirms the HER2 copy number gain in the HER2-positive LCIS (indicated by orange squares) and its absence in the HER2-negative components (C). Additionally, the presence of an FGFR1 copy number gain in the ILC is noted (indicated by grey triangles).  (Table 2). Patient #3 showed a p.G1049R missense mutation in the HER2-negative invasive component, which was not detected in the DCIS components, irrespective of their HER2 status. Patient #6 showed a p.Q546E PIK3CA mutation which was present in the HER2-negative invasive component and absent in the admixed HER2-positive invasive component. Patient #4 had a p.H1047R PIK3CA mutation in each tumour component, but the HER2-negative invasive component harboured an additional p.W1057X mutation, which was not detected in the other tumour components. Somatic ARID1A, MLL3 and NF1 mutations were found in at least one component in three, six and four breast tumours, respectively, and the presence of these mutations was unrelated to the HER2 status (Fig. S2, Table 2). For instance, patient #1 showed a p.C327F MLL3 mutation in the HER2-positive carcinoma component, which was absent in both DCIS  (Table S5). Somatic mutations in these genes were often heterogeneously present throughout the different tumour components, and their presence seemed unrelated to the HER2 amplification status, except for GATA3 mutations. Somatic GATA3 mutations occurred in HER2-positive tumour components. Patient #2 had a GATA3 mutation in the HER2-positive in situ and invasive components, which was absent in the HER2-negative invasive carcinoma component. Patient #3 had a GATA3 splice site mutation in the HER2-positive LCIS component, which   was not observed in the HER2-negative in situ and invasive components.

Copy number variation analysis
The presence of high-level CNVs was investigated, and confirmed the presence of HER2 amplification in all HER2-positive carcinoma samples, which served as an internal quality control (Figs 1 and 2; Figs S1 and S2). Additionally, we observed an EGFR copy number gain in the HER2-negative DCIS component of patient #1 (Fig. 1). Patients #3 and #5 had an FGFR1 copy number gain in at least one tumour component. In patient #3, the HER2-negative ILC harboured this FGFR1 amplification (Fig. 2), which was absent in the HER2negative and HER2-positive LCIS components. In patient #5, all tumour components displayed the FGFR1 copy number gain. Patients #9 and #10 both had a HER2-negative tumour component with a gain of 8q24, which comprised a copy number gain of both MYC and the adjacent long noncoding RNA (lncRNA) plasmacytoma variant translocation 1 (PVT1). Patient #9 also had a CCND1 copy number gain in the HER2-negative tumour component. The potential presence of copy number losses was difficult to interpret with certainty, as some amplicons showed a consistently lower coverage throughout this series. The use of FFPE samples caused a relatively high background, which further hampered the assessment of any potentially relevant deletions.

Complementary immunohistochemical analysis
Immunohistochemistry for EGFR was performed on all tumour tissue samples. In patient #1, the identified EGFR amplification in the HER2-negative DCIS component was associated with EGFR protein overexpression ( Fig. 1F-H). No EGFR protein overexpression was noted in the other tumours (data not shown). Immunohistochemistry for FGFR1 was performed on tumour tissue samples of patients #3 and #5, which revealed no apparent positivity in either of the tumour components (data not shown).

Discussion
Carcinogenesis is an evolutionary process governed by the principles of Darwinian dynamics (Gillies et al., 2012). Tumours are clonal proliferations, originating from a single cell that acquired genomic instability through an accumulation of somatic mutations. Early genomic anomalies, including crucial oncogenic drivers, will therefore be present in all tumour cells and constitute clonal molecular aberrations. Acquisition of additional oncogenic drivers and passenger mutations will result in subpopulations of cancer cells with different genotypes and phenotypes, and these subclonal aberrations contribute to intratumour heterogeneity (McDonald et al., 2019). This heterogeneity is caused by somatic mutations and CNVs, as well as differences in epigenetics (Assenov et al., 2018;Easwaran et al., 2014). Somatic evolution is driven by a combination of genetic instability and a selective tumour microenvironment, including acidosis, hypoxia and cytotoxic stress imposed by chemotherapy, hormonal therapy and/or targeted therapies (Gillies et al., 2012). HER2-targeted therapies impose an evolutionary selection pressure on HER2-positive cancer cells.
Those cancer cell populations that are not exclusively dependent on the overexpression of the HER2 oncogene will be able to constitute an anti-HER2 therapyresistant subclone, regardless of their HER2 status. These subclones harbour alternative and/or collaborative drivers of carcinogenesis, which circumvent the blockade of the HER2-driven pathways. The high prevalence of both intrinsic and acquired resistance to single-agent treatment regimens already caused a shift towards dual HER2-targeted therapy, such as pertuzumab or T-DM1 (Konecny, 2013;Pernas et al., 2018). Interestingly, 5-41% of HER2-positive breast cancers present with regional heterogeneous HER2 amplification (Cottu et al., 2008;Ng et al., 2015), although this percentage depends on the applied definition. In this study, we subjected ten breast cancers with spatially heterogeneous HER2 amplification and corresponding HER2 overexpression to targeted NGS. We investigated the potential heterogeneity in the somatic mutational landscape of each individual tumour component. Some mutations were, if present, homogeneously found in each component. For instance, four of seven tumours with a PIK3CA mutation presented this mutation in each component. Somatic TP53 mutations seemed more often heterogeneously distributed, and their presence seemed generally unrelated to the HER2 amplification status.
Two breast cancers in this series harboured a gain of the 8q24 region, comprising both MYC and the adjacent lncRNA PVT1, which stabilizes the MYC protein and enhances its activity (Tseng and Bagchi, 2015). Co-amplified MYC and PVT1 genes have been identified as candidate oncogenes in ER-positive, HER2-positive breast cancers (Sircoulomb et al., 2010). A recent meta-analysis concluded that increased PVT1 expression was associated with lower overall survival in a wide variety of solid tumours, including breast cancer (Zou et al., 2019). High PVT1 expression was associated with clinicopathological markers of poor prognosis, such as larger tumour size, higher TNM stage and the presence of both lymph node and distant metastases (Zou et al., 2019). In vitro studies demonstrated that PVT1 expression drives cancer cell proliferation through promotion of the KLF5/BAP1/ beta-catenin signalling pathway (Tang et al., 2018).
One patient had a CCND1 copy number gain in a HER2-negative invasive tumour component. CCND1 amplification is associated with a particular gene expression profile and decreased survival in ER-positive, HER2-negative node-negative breast cancer patients (Lundberg et al., 2019), indicating that CCND1 amplification might act as an alternative driver of carcinogenesis. Similar observations have been reported for FGFR1 amplification within breast cancer and other types of carcinoma (Helsten et al., 2016).
By using targeted NGS with a 53-gene panel, we identified a plethora of somatic mutations and CNVs within the HER2-negative components in this series of ten HER2 heterogeneous breast cancers. The genetic heterogeneity within both the HER2-negative and HER2-positive components of a single tumour suggests that a wide range of different somatic mutations and/or CNVs may act as potential alternative drivers. These genetic aberrations might counterbalance the absence of HER2 amplification in the HER2-negative components. Of note, this targeted NGS-driven study focussed only on a subset of 53 breast cancer-related genes in a limited series of ten breast cancer patients. Since we did not apply whole-genome sequencing on a large patient series, it is impossible to exclude the existence of a more commonly present alternative driver in HER2-negative tumour components. Due to the use of FFPE material, we were confronted with high levels of background in some tumour tissue samples, which precluded an in-depth analysis of potentially important copy number losses. Nevertheless, our findings are in accordance with the observations of the TCGA network, who described a high frequency of TP53 (55%) and PIK3CA (31%) mutations, and an low frequency of mutations in RUNX1 (1%), PTEN (0%), NCOR (0%) and CDH1 (3%) in 75 clinically HER2-positive breast cancers (Koboldt et al., 2012). The TCGA network identified a high frequency of TP53 mutations in ER-negative, HER2-positive breast cancers, whereas ER-positive, HER2-positive breast cancers displayed more often a GATA3 mutation (Koboldt et al., 2012). This hormone receptor-dependent duality was not observed in our series, which might be due to its small size.
The limited gene panel precludes strong statements regarding the clonal relationship of all components within a single tumour. However, the integration of histopathological and immunohistochemical features, together with the uniform presence of some well-defined pathogenic mutations (such as TP53 or PIK3CA mutations), suggests a common progenitor for most heterogeneous lesions in this series. Based on the frequent homogeneous presence of the observed variants, it was estimated that the tumours of patients #1, #3, #4, #5, #7 and #9 were likely to have a common progenitor. The tumours of patients #2, #6 and #10 were considered to be less likely related to one another (i.e., a collision tumour of two independent neoplastic lesions), or to have a common progenitor with very early divergence of the subclones. Despite its limited size, this series of ten breast cancers demonstrates that regional heterogeneity in HER2 status is associated with further heterogeneity at the molecular level, and sometimes also at the protein level, since some tumour components presented with different hormone receptor status and/or EGFR protein expression status. Although regional HER2 heterogeneity is uncommon, this series illustrates that not all cells within one tumour depend exclusively on HER2 amplification and overexpression.
Due to its relatively high prevalence in invasive breast cancer and its association with worse prognosis, HER2 overexpression is suspected to play a major role as a driver of mammary carcinogenesis. HER2-positive invasive breast cancer more often presents with an associated in situ component, and if present, this DCIS component is substantially larger than in HER2-negative tumours (Doebar et al., 2016). The prevalence of HER2 overexpression amounts 35% in pure DCIS, which is paradoxically higher than its prevalence in invasive breast cancer (Siziopikou et al., 2013). Overall, HER2 expression profiles are highly concordant between admixed in situ and invasive breast cancer, but overexpression/amplification is less common in the DCIS component of admixed lesions than in pure DCIS (Burkhardt et al., 2010;Lambein et al., 2017;Latta et al., 2002;Park et al., 2006). One in three women with a HER2-positive pure DCIS lesion develops a subsequent HER2-negative invasive breast cancer (Visser et al., 2019), although the clonal relationship between primary and recurrent lesions was not investigated in that study. Taken together, these observations indicate that HER2 overexpression is more likely to play a role as an instigator of tumour cell proliferation, rather than being a crucial driver of cancer cell invasion (Sanati, 2019). The series of pathogenic and likely pathogenic somatic variants that we describe here yields a wide range of potential alternative drivers of cancer cell proliferation and invasion. Moreover, some genetic anomalies (such as PIK3CA and GATA3 mutations, or FGFR1 copy number gain) might drive resistance to treatment (Pernas et al., 2018;Turner et al., 2010).

Conclusion
The HER2-negative components of HER2 heterogeneous breast cancers display a variety of somatic mutations and CNVs within 53 breast cancer-related genes. Although these somatic mutations and CNVs were often present in the HER2-positive component as well, they might act as potential alternative drivers to counterbalance the absence of HER2 amplification. Since these potential alternative drivers may have the capacity to circumvent HER2 pathway blockade, their widespread presence throughout these HER2 heterogeneous cancers might explain the high level of innate and acquired resistance to HER2-targeted therapies in breast cancer. Our findings indirectly imply that a targeted monotherapy is unlikely to have high efficacy in the long term, since it causes cytotoxic distress and selection of those resistant clones that already harbour alternative drivers of carcinogenesis. Future translational breast cancer research should focus on how to handle this molecular heterogeneity in the clinical setting.

Conflict of interest
MRVB is supported by the Mathilde Horlait-Dapsens Foundation (Brussels, Belgium) and the non-for-profit organization Foundation Against Cancer (Grant 2019-089, Brussels, Belgium). The other authors report that they have no conflicts of interest to disclose.

Author contributions
MRVB and MCA performed the mutation analysis and copy number variation analysis, and designed the figures and tables of the manuscript. MRVB and CHMD performed the histopathological review and interpreted immunohistochemical and SISH analyses. RM, PNA and WNMD were responsible for performing tissue microdissection, DNA isolation, library preparation and targeted next-generation sequencing. HFBMS was responsible for HER2 ISH analysis. EHL, LLV and JW designed the breast-specific gene panel used for targeted next-generation sequencing. CHMD was responsible for the study design and patient selection. MRVB wrote the first draft of the manuscript. All contributors reviewed and edited the manuscript, and approved its final version.

Data availability
All data generated and/or analysed during the current study are available from the corresponding author on reasonable request.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. Copy number variations, HER2 status and hormone receptor status in the breast cancer of patient #6. The scatter plot confirms the presence of a HER2 copy number gain in the HER2-positive invasive carcinoma component (A; indicated by red squares). Immunohistochemistry for HER2 (B), oestrogen receptor (C) and progesterone receptor (D) illustrate opposite protein expression profiles in both invasive carcinoma components (original magnification 12,5xscale bar size = 2,5 mm). Fig. S2. Copy number variations in the breast cancers of patients #5, #9 and #10. The scatter plot of patient #5 (A) confirms a HER2 copy number gain in the HER2-positive invasive carcinoma component (indicated by green triangles), and demonstrates an FGFR1 copy number gain (cytogenetic location: 8p11.23) in each tumour component. The tumour of patient #9 harbours a neighbouring copy number gain located at 8q24 in all carcinoma components, which comprises both the MYC and PVT1 genes, as well as a CCND1 copy number gain in the HER2negative invasive carcinoma component (B). A similar 8q24 copy number gain was noted in the HER2-negative invasive carcinoma component (indicated by red triangles) of patient #10 (C). In patient #9, this coamplification was present in both the HER2-positive DCIS and the HER2-negative invasive carcinoma components, indicating that this genetic aberration can occur as an early event in carcinogenesis. However, this co-amplification was not present in the HER2-positive DCIS component and the HER2-negative axillary metastasis. Table S1. Materials and methods for immunohistochemistry. Table S2. List of genes included in the panel used for targeted next-generation sequencing. Table S3. Detailed information on base coverage and number of reads for targeted next-generation sequencing of normal and tumour tissue samples of ten breast cancers with regional HER2 heterogeneity. Table S4. Mutation analysis. Table S5. Detailed information on the presence of somatic mutations in the less frequently mutated genes per patient and per tumour component.