Pleomorphic adenomas and mucoepidermoid carcinomas of the breast are underpinned by fusion genes

Primary pleomorphic adenomas (PAs) and mucoepidermoid carcinomas (MECs) of the breast are vanishingly rare. Here we sought to determine whether breast PAs and MECs would be underpinned by the fusion genes reported to occur in their salivary gland counterparts. Our study included three breast PAs and one breast MEC, which were subjected to RNA sequencing (PAs, n = 2; MEC, n = 1) or to Archer FusionPlex sequencing (PA, n = 1). Our analyses revealed the presence of the HMGA2-WIF1 fusion gene in breast PA3, the CTNNB1-PLAG1 fusion gene in breast PA2, and the CRTC1-MAML2 fusion gene in the breast MEC analyzed (1/1). No oncogenic fusion genes were detected in breast PA1, and no additional oncogenic fusion genes were detected in the cases studied. The presence of the fusion genes identified was validated by fluorescence in situ hybridization (n = 1), reverse transcription-PCR (n = 1), or by both methods (n = 1). Taken together, our findings indicate that PAs and MECs arising in the breast resemble their salivary gland counterparts not only phenotypically but also at the genetic level. Furthermore, our data suggest that the molecular analysis of breast PAs and MECs might constitute a useful tool to aid in their differential diagnosis.


INTRODUCTION
Pleomorphic adenomas (PAs) and mucoepidermoid carcinomas (MECs) of the breast are rare tumors that histologically resemble their salivary gland counterparts 1 . PAs are characterized by an admixture of epithelial and myoepithelial cells immersed in myxochondroid stroma 1,2 . Cartilaginous and osseous metaplasia may be present 1,2 . Breast PAs usually have a benign course, although local recurrences have been described 1 . PAs arising in the salivary glands, skin, and soft tissue harbor recurrent gene rearrangements involving PLAG1 or HMGA2 (refs. 3,4 ). MECs, although common in the salivary glands, are vanishingly rare in the breast 1,5 . MECs are characterized by the admixture of different cell types, including mucinous, intermediate, and squamous cells 1 . MECs arising in the salivary gland, lung, and cervix have been shown to harbor the t (11,19)(q14-21; p12-13) translocation that results in the CRTC1-MAML2 fusion gene [6][7][8] , or less frequently the CRTC3-MAML2 fusion gene 9 .
A subset of salivary gland-like tumors arising in the breast have been shown to be underpinned by the same genetic alterations as their salivary glands counterparts 1,10 . For instance, adenoid cystic carcinomas arising either in the breast or in the salivary glands harbor MYB or MYBL1 rearrangements, regardless of their anatomic origin [10][11][12] , and secretory carcinomas of the breast and mammary analog secretory carcinomas of the salivary glands are characterized by the ETV6-NTRK3 fusion gene 10,13,14 . The genetic underpinning of other breast salivary gland-like tumors, such as acinic cell carcinomas, however, is different than that of their salivary gland counterparts. Salivary gland acinic cell carcinomas are characterized by the t(4;9)(q13; q31) translocation, which results in a rearrangement involving the nuclear transcription factor NR4A3 (ref. 15 ), whereas breast acinic cell carcinomas seem not to harbor a pathognomonic genetic alteration and consistently display complex genomes and TP53 mutations 16,17 . Here we sought to determine whether breast PAs and MECs arising in the breast would be underpinned by fusion genes, in particular those reported in their salivary gland counterparts.

Cases
Three PAs and one MEC arising in the breast were included in this study. The breast PAs (n = 3) included in our series were wellcircumscribed lesions (Fig. 1a), composed of epithelial cells and myoepithelial cells arranged in cords, clusters, and forming glands immersed in a myxochondroid stroma (Fig. 1b, c). Myoepithelial cells floating in the stroma as single stellate cells (Fig. 1b) and focal areas of solid growth were frequently observed (Fig. 1d). No cellular atypia or increased mitotic index were detected (Fig. 1d) The breast MEC studied here corresponded to a low-grade MEC with well-circumscribed borders (Fig. 1e), composed of mucinous, intermediate, and squamous cells arranged in cords, sheets, and clusters (Fig. 1e). No cytologic atypia or necrosis were observed (Fig. 1e, f).
To determine whether the breast PAs (n = 3) and the MEC (n = 1) included in our series would harbor fusion genes, akin to their counterparts arising in the salivary glands, we subjected two breast PAs (BPA1 and BPA3) and one MEC (BMEC2) to RNA sequencing (Fig. 1g, Supplementary Table 1). RNA-sequencing data were analyzed using a validated bioinformatics pipeline for de novo discovery of fusion genes as previously described 18,19 . RNA derived from one breast PA (BPA2) with insufficient   encodes for a transcriptional regulator of the TGF-β signaling pathway 22 , and WIF1 for a tumor suppressor modulating signaling via the Wnt pathway 20 . The HMGA2-WIF1 fusion gene detected in BPA3 resulted in a chimeric transcript composed of exons 1-3 of HMGA2 and exon 10 of WIF1, and was predicted to be translated in the initial segment of HMGA2 fused to the C′-terminus of WIF1, with the loss of its active WIF domain 20 (Fig. 2a). The HMGA2-WIF1 fusion gene identified in BPA3 was validated by reverse transcription (RT)-PCR, followed by Sanger sequencing (Fig. 2b).
As predicted, the HMGA2-WIF1 fusion resulted in retained expression of HMGA2 exons 1-3 and decreased expression of exons 4 and 5 (Fig. 2c). We identified the presence of a CTNNB1-PLAG1 fusion gene in BPA2 by Archer FusionPlex targeted RNA sequencing (Figs. 1g and 2d, Supplementary Table 1). The CTNNB1-PLAG1 fusion gene was predicted to result in a chimeric transcript encompassing exon 1 of CTNNB1 and exons 4 and 5 of PLAG1 (Fig. 2d) and has been described in PAs of the salivary glands and of the lacrimal glands 23,24 . The CTNNB1-PLAG1 fusion identified in BPA2 was validated by fluorescence in situ hybridization (FISH) using PLAG1 dual-color break-apart probes (Fig. 2e).
No additional likely pathogenic fusion genes were identified in the breast PAs analyzed (Supplementary Tables 1 and 2). Given that HMGA2 and PLAG1 rearrangements are the hallmark genetic alterations of PAs, regardless of their anatomic location 3,4 , we subjected the breast PA included in this study in which no fusion genes were identified by RNA sequencing (BPA1) to FISH analysis using HMGA2 and PLAG1 dual-color break-apart probes, which confirmed the absence of HMGA2 or PLAG1 rearrangements (Supplementary Table 1). Notably, BPA1 did not differ histologically from the PAs found to harbor fusion genes.
Breast adenomyoepitheliomas and PAs may show histologic and genetic overlap 25 . Indeed, we have reported on a bona fide breast adenomyoepithelioma displaying focal myxoid matrix and harboring an HMGA2-WIF1 fusion gene 26 . Hence, we sought to determine whether the breast PAs of this series subjected to RNA sequencing would harbor hotspot mutations 27 affecting HRAS, PIK3CA, and AKT1 as in breast adenomyoepitheliomas 28 . Upon manual inspection of the HRAS Q61 hotspot locus, PIK3CA H1047, E545, E542, G118, G1050, and G106 hotspot loci and AKT1 E17 hotspot locus in the RNA sequencing BAM files of BPA1 and BPA3 using the Integrative Genomics Viewer (IGV) 29 , no hotspot mutations affecting these genes were identified.
CRTC1-MAML2 fusion gene in a breast mucoepidermoid carcinoma RNA sequencing analysis of BMEC2, a MEC arising in the breast, revealed the presence of a CRTC1-MAML2 fusion gene (Figs. 1g  and 3a), the hallmark genetic alteration of MECs arising in other anatomic origins [6][7][8] . These findings are in agreement with the study by Bean et al. 30 , who reported on two breast MECs harboring the CRTC1-MAML2 fusion gene. The fusion gene identified in BMEC2 was predicted to result in a chimeric transcript encompassing exon 1 of CRTC1 and exons 2-5 of MAML2 (Fig. 3a,  b), akin to MECs in other anatomic locations and to the two breast MECs previously reported [6][7][8] . As predicted, the CRTC1-MAML2 fusion resulted in increased expression of exons 2-5 and decreased expression of exon 1 of MAML2 (Fig. 3c), and its presence was confirmed by RT-PCR and FISH using a three-color break-apart probe (Fig. 3b, d).

DISCUSSION
Given that various salivary gland-like tumors arising in the breast have been shown to harbor the same genetic alterations as their salivary gland counterparts 10 , we posited that breast PAs and MECs might be underpinned by the oncogenic fusion genes described in PAs and MECs arising in the salivary glands. Our analyses resulted in the identification of an HMGA2-WIF1 and a CTNNB1-PLAG1 fusion gene in breast PAs, and of a CRTC1-MAML2 fusion gene in a breast MEC. The HMGA2 and PLAG1 rearrangements, and the CRTC1-MAML2 fusion genes identified in the breast PAs and breast MECs of this study, have been reported in PAs and MECs arising in the salivary gland and in other anatomic locations [6][7][8]20,21,23,24 . Moreover, our findings on breast MECs are in agreement with the study by Bean et al. 30 , who reported on the presence of the CRTC1-MAML2 fusion gene in two breast MECs.
HMGA2 is an oncogene with diverse oncogenic mechanisms, as it has been shown to induce the activity of E2F1 and AP1, result in inactivation of p53-dependent apoptosis, and in the activation of the TGF-β signaling pathway 22,31 . WIF1 is a tumor suppressor that encodes for a secreted antagonist that binds Wnt proteins hampering ligand-receptor interactions 32,33 . The HMGA2-WIF1 fusion gene identified in BPA3 results in the loss of the 3′ UTR regulatory sites of HMGA2, resulting in its overexpression and in increased signaling via TGF-β 22,34 . It is also predicted to result in the loss of the WIF domain of WIF1, required for its tumor suppressor activities, with ensuing Wnt signaling activation 32,33 . We have recently reported the presence of an HMGA2-WIF1 fusion gene in an estrogen receptor-positive breast adenomyoepithelioma (AME) lacking HRAS mutations 26 . Although the HMGA2 and WIF1 genomic breakpoints in the HMGA2-WIF1 fusion gene identified in the HRAS-wild type AME were different than those identified in BPA3, they resulted in the loss of the 3′UTR of HMGA2 and of the WIF domain of WIF1 (ref. 26 ), akin to the fusion gene present in BPA3 reported here. Notably, the previously reported AME harboring the HMGA2-WIF1 fusion gene 26 , despite having the cardinal diagnostic features of AMEs, displayed focal myxochondroid stroma, akin to that found in PAs 26 . We identified a CTNNB1-PLAG1 fusion gene in BPA2, described in PAs of the salivary gland. This fusion gene results in promoter swapping between PLAG1 and CTNNB1, leading to increased PLAG1 expression and reduction of CTNNB1 expression 35 . Taken together, these findings suggest that the spectrum of histologic features of breast PAs and AMEs overlaps, despite the differences in their repertoire of somatic genetic alterations (i.e. HMGA2 and PLAG1 fusion genes in the majority of PAs, in contrast with mutations affecting HRAS, PIK3CA, AKT1, and PIK3R1 in breast AMEs) 3,4,28 .
MECs of the salivary glands and two breast MECs previously reported are characterized by the t(11:19) (q21;p13) translocation, which results in the CRTC1-MAML2 fusion gene 6,30 . CRTC1 encodes for the CREB-regulated transcriptional coactivator 1 (ref. 36 ) and MAML2 for the NOTCH/RBPJ mastermind-like 2. MAML2 forms DNA-binding complexes with Rbp-j and NotchIC and augments Notch signaling 37 . The CRTC1-MAML2 fusion gene identified in BMEC2 has been shown to result in activation of the Notch signaling pathway, be tumorigenic in vitro, and silencing of its fusion product reportedly inhibits tumor growth 38 .
Given their rarity, the differential diagnosis of PAs and MECs arising in the breast from other breast tumors, such as metaplastic carcinomas and squamous cell carcinomas, respectively, might be challenging 1 . The absence of peripheral myoepithelial cell layer between the epithelium and stroma in PAs together with their infiltrative appearances, at least focally, makes their distinction from metaplastic carcinoma of the matrix producing type challenging, particularly in core biopsies. Such distinction may have significant management implications. Similarly, the distinction between MECs and the more aggressive metaplastic squamous cell carcinoma or the benign papilloma with squamous metaplasia can be difficult and an additional genomic testing may help in their accurate diagnosis. Reanalysis of the RNA sequencing results by Piscuoglio et al. 39 and Weigelt et al. 40 revealed that none of the metaplastic breast cancers analyzed in those studies harbored any of the fusion genes identified in breast PAs and MECs (data not shown). Our findings suggest that the assessment of rearrangements involving PLAG1 or HMGA2 in breast PAs, and of MAML2 rearrangements in breast MECs, might be used as ancillary tools to aid in the diagnosis of these rare breast tumors.
Our study has important limitations, including the small sample size due to the rarity of PAs and MECs arising in the breast. In addition, we could only perform transcriptomic analyses of the samples included here due to limited available material. Finally, no fusion genes were identified in one of the breast PAs analyzed, and its driver has yet to be identified. Despite these limitations, our findings support the notion that PAs and MECs arising in the breast harbor genetic alterations akin to those described in their salivary gland counterparts, and constitute additional examples of genotypic-phenotypic correlations in breast tumors. These findings may also constitute the basis for the development of ancillary methods for the diagnosis of these tumors in the breast.

METHODS Ethics
This study was approved by the Memorial Sloan Kettering Cancer Center Institutional Review Board (IRB) and local research ethics committees of the authors' institutions. Patient consent was obtained if required by the approved IRB protocols. Formalin-fixed paraffin-embedded (FFPE) tissue blocks of three breast PAs and one breast MEC were retrieved from the pathology archives of the authors' institutions. All samples were anonymized before analysis.

Cases and RNA extraction
All tumors included in this study were centrally reviewed by four pathologists (F.P., M.V., F.C.G., J.S.R.-F.) for diagnosis confirmation, following the diagnostic criteria put forward by the World Health Organization (WHO) 41 . The three breast PAs and one breast MEC were microdissected from eight-micron-thick histologic sections under a stereomicroscope (Olympus SZ61) to ensure a tumor cell content >80%, and RNA was extracted using the RNeasy FFPE Kit.
RNA sequencing and identification of fusion genes FFPE-derived RNA from two breast PA and one breast MEC were subjected to paired-end RNA sequencing following the standard protocols used at the Integrated Genomics Operation at MSKCC 42 , as previously described 19 to identify read pairs supporting chimeric transcripts, followed by exclusion of candidate fusion transcripts and read-through candidates identified in a set of 297 normal breast tissues from the TCGA 47 . The Bayesian probability to constitute drivers of the remaining candidate fusion genes supported by at least two spanning reads was assessed using OncoFuse (v1.0.9b2) 48 . Reads per kilobase million (RPKM) values were calculated from raw counts after normalization for sequencing depth and following adjustment for length of genes. Data are available as described on the Data availability statement 49,50 .

ARCHER FusionPlex
One breast PA with insufficient material for RNA sequencing was subjected to Archer FusionPlex assay, an RNA-based panel that uses Archer Anchored Multiplex PCT technology and next-generation sequencing for the identification of fusion genes 51 at the Integrated Genomics Operation at MSKCC, as previously described 52 . Archer FusionPlex data were analyzed using Archer Software (v4.0.10).

RT-PCR for fusion gene validation
RNA was reverse-transcribed using SuperScript VILO Master Mix (Thermo-Fisher Scientific), following the manufacturers' instructions. PCR amplification of 10 ng cDNA was conducted using fusion gene-specific primers sets designed based on the identified breakpoints (Supplementary Table 3). PCR fragments were purified with ExoSAP-IT (ThermoFisher Scientific) and subjected to Sanger Sequencing. Sequence electropherograms of the forward and reversed strands were manually inspected.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

DATA AVAILABILITY
The data generated and analyzed during this study are described in the following data record: https://doi.org/10.6084/m9.figshare.12199781 (ref. 49 ). The RNA sequencing data have been deposited in the NCBI Sequence Read Archive under the project accession number SRP257886 (https://identifiers.org/ncbi/insdc.sra:SRP257886) 50 . The histologic images, fluorescence in situ hybridization, and Sanger sequencing data will be provided upon reasonable request to Dr. Fresia Pareja. The Archer FusionPlex data will be provided upon reasonable request to Dr. Arnaud Da Cruz Paula.