Genetic and immunohistochemical profiling of small cell and large cell neuroendocrine carcinomas of the breast

Neuroendocrine carcinomas (NEC) of the breast are exceedingly rare tumors, which are classified in the WHO system as small cell (SCNEC) and large cell (LCNEC) carcinoma based on indistinguishable features from their lung counterparts. In contrast to lung and enteropancreatic NEC, the genomics of breast NEC have not been well-characterized. In this study, we examined the clinicopathologic, immunohistochemical, and genetic features of 13 breast NEC (7 SCNEC, 4 LCNEC, 2 NEC with ambiguous small versus large cell morphology [ANEC]). Co-alterations of TP53 and RB1 were identified in 86% (6/7) SCNEC, 100% (2/2) ANEC, and 50% (2/4) LCNEC. The one SCNEC without TP53/RB1 alteration had other p53 pathway aberrations (MDM2 and MDM4 amplification) and was immunohistochemically RB negative. PIK3CA/PTEN pathway alterations and ZNF703 amplifications were each identified in 46% (6/13) NEC. Two tumors (1 SCNEC, 1 LCNEC) were CDH1 mutated. By immunohistochemistry, 100% SCNEC (6/6) and ANEC (2/2) and 50% (2/4) LCNEC (83% NEC) showed RB loss, compared to 0% (0/8) grade 3 neuroendocrine tumors (NET) (p < 0.001) and 38% (36/95) grade 3 invasive ductal carcinomas of no special type (IDC-NST) (p = 0.004). NEC were also more often p53 aberrant (60% vs 0%, p = 0.013), ER negative (69% vs 0%, p = 0.005), and GATA3 negative (67% vs 0%, p = 0.013) than grade 3 NET. Two mixed NEC had IDC-NST components, and 69% (9/13) of tumors were associated with carcinoma in situ (6 neuroendocrine DCIS, 2 non-neuroendocrine DCIS, 1 non-neuroendocrine LCIS). NEC and IDC-NST components of mixed tumors were clonally related and immunophenotypically distinct, lacking ER and GATA3 expression in NEC relative to IDC-NST, with RB loss only in NEC of one ANEC. The findings provide insight into the pathogenesis of breast NEC, underscore their classification as a distinct tumor type, and highlight genetic similarities to extramammary NEC, including highly prevalent p53/RB pathway aberrations in SCNEC.


INTRODUCTION
Neuroendocrine carcinomas (NEC) of the breast are rare high-grade malignancies that are poorly understood biologically and clinically. Although neuroendocrine differentiation of breast tumors has long been recognized, classification has been problematic and has continued to shift over the years, with the most recent World Health Organization (WHO) classification (5 th edition) based on a consensus that terminology should be more uniform across anatomic sites [1][2][3][4] . Accordingly, the WHO has defined NEC of the breast as small cell neuroendocrine carcinoma (SCNEC) and large cell neuroendocrine carcinoma (LCNEC), emphasizing that these tumors are histologically and immunohistochemically indistinguishable from their respective lung counterparts 2,5 . SCNEC has been recognized as a distinct and clinically aggressive breast cancer subtype for many years, although most published literature is based on case reports and small series [6][7][8][9] . The classification of LCNEC as a separate entity in the breast is again acknowledged in the most recent WHO edition, yet debated 10 , and clinical and pathologic features of these rare tumors remain largely uncharacterized.
NEC of the breast should be distinguished from neuroendocrine tumors (NET), which are morphologically distinct from NEC and are usually Nottingham grade 1 or 2 5,11 . However, it is recognized that some Nottingham grade 3 neuroendocrine neoplasms (NEN) of the breast do not resemble SCNEC or LCNEC morphologically 1 . These tumors lack high-grade nuclei, necrosis, and other nucleocytologic features of SCNEC or LCNEC despite areas of high mitotic activity and diffuse neuroendocrine marker expression. It is unclear if such tumors should be classified as NEC or grade 3 NET in the current WHO classification. In the enteropancreatic system, it has been well-established that high-grade (G3) NET can be challenging to differentiate from NEC histopathologically, yet these neoplasms are biologically and clinically distinct from one another [12][13][14][15] . Whether a similar paradigm exists for NEC and these other grade 3 NEN in the breast is unknown.
In this study, we comprehensively characterized a cohort of breast NEC (SCNEC, LCNEC, and NEC with features ambiguous between small and large cell morphology) by capture-based nextgeneration sequencing (NGS) and immunohistochemistry in order to identify molecular drivers and to determine if these rare aggressive tumors share pathogenetic features with NEC arising in other sites. NEC and invasive ductal carcinoma of no special type (IDC-NST) components of mixed tumors were separately analyzed to assess shared clonality and immunophenotypic divergence between the components. The immunoprofiles of NEC were additionally compared to other grade 3 NEN that morphologically were not considered to meet NEC criteria (including high-grade NET), and to grade 3 IDC-NST. Our results provide novel insights into the pathogenesis of breast SCNEC and LCNEC and highlight genetic similarities to extramammary NEC, including highly prevalent p53/RB pathway aberrations in SCNEC. Overall the findings support the separate classification of breast NEC, especially SCNEC, and suggest that LCNEC may be a more heterogeneous group.

MATERIALS AND METHODS Study population and tumor classification
With institutional review board approval, the pathology archives of University of California San Francisco (UCSF), Stanford University, and Kaiser Permanente (San Francisco, CA) were searched for cases of Nottingham grade 3 NEN of the breast. This was supplemented by the consultation service of one of the authors (S.J.S.); three SCNEC were described in part in a prior report 6 . NEC (n = 11) and NEC components of mixed NEC/IDC-NST tumors (n = 2) comprising the study population were classified as SCNEC or LCNEC based on independent review by two pulmonary pathologists experienced in the diagnosis of neuroendocrine carcinoma (A.U. and K.D.J.), who classified them based on morphologic criteria used in the lung. Two tumors with discordant classification as SCNEC or LCNEC were classified as NEC with features ambiguous for small cell versus large cell morphology (ANEC1 and ANEC2). All NEC expressed at least one neuroendocrine marker, and diffuse (>90%) staining with synaptophysin and/or chromogranin was required for LCNEC. Of these 13 tumors, eleven (6 SCNEC, 2 ANEC, and 3 LCNEC) were analyzed by DNA sequencing using the UCSF500 assay, and two (1 SCNEC, 1 LCNEC) were submitted for FoundationOne tumor-only sequencing for clinical purposes (Foundation Medicine, Cambridge, MA).
Eight Nottingham grade 3 NEN with diffuse neuroendocrine morphology and extensive (>90%) synaptophysin and/or chromogranin expression that did not show characteristic cytomorphologic features of either SCNEC or LCNEC of the lung were identified. We opted to classify these tumors as grade 3 NET for purposes of comparison to NEC (see Results for additional histologic description of these tumors). Nottingham grade 3 invasive carcinomas with less than diffuse neuroendocrine differentiation, including <90% synaptophysin and/or chromogranin expression, were classified as invasive ductal carcinomas with neuroendocrine differentiation (IDC-NED) (n = 2) or invasive lobular carcinoma with neuroendocrine differentiation, solid pattern (ILC-NED) (n = 1) 16 .
Clinical information was obtained from electronic medical records when available. All tumors were confirmed to be mammary in origin based on clinical history, imaging, and pathologic findings.

Capture-based next generation DNA sequencing
Matched tumor and normal tissue were selected from nine pure NEC (6 SCNEC, 1 ANEC, 2 LCNEC) and two mixed NEC/IDC-NST (ANEC2 and LCNEC1) for capture-based NGS (n = 11). For the two mixed NEC/IDC-NST tumors, DNA from NEC and IDC-NST areas was macrodissected and analyzed separately. For patients treated with chemotherapy, only pretreatment tumor was tested by NGS. Sequencing libraries were prepared from genomic DNA extracted from punch biopsies or macrodissected unstained sections from formalin-fixed paraffin-embedded tissue. Target enrichment was performed by hybrid capture using a custom oligonucleotide library. Capture-based NGS was performed at the UCSF Clinical Cancer Genomics Laboratory, using an assay (UCSF500 panel) that targets the coding regions of 480 cancer-related genes, select introns from~40 genes, and the TERT promoter with a total sequencing footprint of 2.8 Mb (Supplementary Table S1). Sequencing was performed on a HiSeq 2500 (Illumina, San Diego, CA). Duplicate sequencing reads were removed computationally to allow for accurate allele frequency determination and copy number calling. The analysis was based on the human reference sequence UCSC build hg19 (NCBI build 37), using the following software packages: BWA: 0.7.10-r789, Samtools: 1.1 (using htslib 1.1), Picard tools: 1.97 (1504), GATK: 2014.4-3.3.0-0-ga3711, CNVkit: 0.3.3, Pindel: 0.2.5a7, SATK: 2013.1-10-gd6fa6c3, Annovar: v2015Mar22, Freebayes: 0.9.20 and Delly: 0.5.9 [17][18][19][20][21][22][23][24][25] . Only insertions/deletions (indels) up to 100 bp in length were included in the mutational analysis. Somatic single nucleotide variants and indels were visualized and verified using Integrated Genome Viewer (Broad Institute, Cambridge, MA, USA). Tumor mutational burden was quantified, reflecting somatic synonymous and nonsynonymous single nucleotide variants and small indels in coding regions and splice sites. Genome-wide copy number analysis based on on-target and off-target reads was performed by CNVkit and Nexus Copy Number (Biodiscovery, Hawthorne, CA, USA). For ER, PR, and HER2, positive staining was defined according to ASCO/ CAP guidelines 26,27 . HER2 FISH testing was performed using the Abbott Vysis Pathvysion (Des Plaines, IL, USA) FDA-cleared kit, performed per manufacturer recommendations, and scored and interpreted according to ASCO/CAP guidelines 27 . For synaptophysin, chromogranin, INSM1, NSE, GATA3, and TTF1, positive staining was segregated as ≥90%, 50-89%, and 1-49%. For p16, diffuse (≥90%) strong nuclear and cytoplasmic staining (i.e., "block-positive") was considered overexpressed, patchy weak to strong cytoplasmic staining was considered normal/wild-type pattern, and lack of staining was considered negative 28 . For RB, patchy to diffuse nuclear staining was considered intact, and lack of nuclear staining was considered negative. For p53, diffuse (≥90%) moderate to strong nuclear staining was considered positive/aberrant, no nuclear staining was considered negative/null, and patchy heterogeneous nuclear staining was considered wild-type pattern.

Statistical analysis
Statistical analysis was performed using Fisher exact test, using a significance level of p < 0.05. The degree of interobserver agreement was quantified by kappa.

RESULTS
Diagnostic histologic and immunophenotypic features and classification of neuroendocrine carcinomas All breast NEC and NEC components of mixed NEC/IDC-NST showed morphologic features typical of poorly-differentiated NEC in the lung and were Nottingham grade 3, with high nuclear grade, high mitotic index, and poor glandular differentiation ( Fig. 1 and Table 1). Individual cytomorphologic features including cell size, nuclear, nucleolar, and cytoplasmic features, tumor growth pattern, and the presence of geographic necrosis are listed in Supplementary  Table S2. The tumors were independently classified as SCNEC versus LCNEC on H&E slides by two pulmonary pathologists experienced in diagnosis of neuroendocrine carcinomas. Distinction between small and large cell morphology was based primarily on nucleolar features and nuclear:cytoplasmic ratio. Morphologic assessment showed substantial agreement (κ = 0.675): both classified seven cases as SCNEC (SCNEC1-7) and four cases as LCNEC (LCNEC1-4). Disagreement was noted for two cases (ANEC1 and ANEC2), each of which showed high nuclear:cytoplasmic ratio yet variably prominent nucleoli. Although distinction between SCNEC and LCNEC for these two cases was not clear based on morphology, all pathologists agreed that each was NEC based on other histopathologic features (organoid architecture, nuclear features, geographic necrosis, and neuroendocrine marker expression). Two mixed tumors (ANEC2 and LCNEC1) demonstrated a component of high-grade IDC-NST in addition to the NEC. Two tumors (SCNEC5 and LCNEC4) exhibited a single-file growth pattern and aberrant E-cadherin expression, consistent with lobular differentiation ( Supplementary Fig. S1).
Among patients with available treatment information (12/13), all received chemotherapy (Supplementary Table S4); six of seven patients with SCNEC and one of two patients with ANEC received etoposide, an agent standardly used for small cell lung cancer but not included in guidelines for the management of breast cancer 29 .
No statistically significant associations were identified between SCNEC and LCNEC with respect to patient age, tumor size, lymphovascular invasion, lymph node metastasis, biomarker status, or outcomes.

Genetics of neuroendocrine carcinomas
Results of targeted DNA sequencing are shown in Fig. 4 and Supplementary Tables S5 and S6. The mean target sequencing coverage was 506 (±219) unique reads per target interval (Supplementary Table S7). Tumor mutational burden ranged from <1 to 18 mutations/megabase (median 4). No pathogenic or likely pathogenic germline alterations were identified in any cases.
The most frequent pathogenic aberrations were in TP53 and RB1, which were co-altered in 77% (10/13) of NEC. Of the seven SCNEC, six (86%) exhibited co-alteration of TP53 and RB1. The only SCNEC that lacked TP53 and RB1 aberrations (SCNEC7) showed amplifications of MDM2 and MDM4, which are well-known negative regulators of p53 [30][31][32] . Despite the absence of identified RB1 alteration, this tumor was RB negative by immunohistochemistry (see below). TP53 and RB1 co-alteration was also identified in both NEC with ambiguous small versus large cell morphology (ANEC1 and 2) and in 50% (2/4) LCNEC. TP53/RB1 co-alteration was significantly more common in NEC as a group and in SCNEC than in a group of matched grade 3 IDC-NST profiled by UCSF500 assay (7/45, 16%) (p < 0.001 and p = 0.001, respectively). Similar results were found when all NEC or SCNEC were compared to grade 3 IDC-NST in the large publicly available METABRIC dataset, in which TP53/RB1 co-alteration was reported in only 3% (35/1009) tumors (p < 0.001, p < 0.001, and p = 0.007, respectively) ( Supplementary  Fig. S3).
Immunohistochemical expression of RB, p16, p53, and other markers in neuroendocrine carcinomas with comparison to grade 3 neuroendocrine tumors To further explore RB and p53 pathway inactivation in NEC, immunohistochemical stains for RB, p16, and p53 were performed on tumors with available tissue (Table 2, Figs. 2 and 5). All NEC or NEC components of mixed NEC/IDC-NST harboring RB1 alterations (including frameshift and splice site mutations, deep deletions, and genomic rearrangement) were RB negative by immunohistochemistry, including all SCNEC (6/6) and ANEC (2/2), and 2/4 LCNEC. Notably, the only SCNEC lacking RB1 alteration (SCNEC7) was RB negative by immunohistochemistry. Both LCNEC without RB1 alterations showed intact RB staining. Diffuse p16 overexpression was seen in 82% (9/11) NEC (5/5 SCNEC, 2/2 ANEC, 2/4 LCNEC) and was exclusively associated with RB loss. An aberrant (null type or diffuse positive) p53 staining pattern was seen in 6/10 (60%) NEC, including 3/ 5 SCNEC, one ANEC, and 2/4 LCNEC. Six of seven (86%) tumors with TP53 alterations demonstrated aberrant p53 expression, whereas none of the three tumors without TP53 alterations had aberrant p53 expression. Negative p53 staining was associated with large deletion, focal homozygous deletion, and splice site mutation in TP53 (three SCNEC), whereas diffuse p53 staining was associated with missense mutations in two tumors and with the p.R342* nonsense mutation (LCNEC1), the latter of which showed nuclear as well as cytoplasmic staining 33,34 . Of note, TP53 p.G266E mutation has been previously shown to be associated with a non-diffuse staining pattern, as was seen in SCNEC5 33 . For comparison to NEC, immunohistochemical expression of RB, p53, and p16 was also assessed in the group of grade 3 NET (n = 8), as well as in two grade 3 IDC-NED and one grade 3 ILC-NED ( Fig. 3 and Supplementary Tables S3 and S8). In contrast to the NEC, none of the other grade 3 cancers showed aberrant RB or p53 expression patterns (RB loss in 83% NEC vs 0% grade 3 NET, p = 0.001; aberrant p53 in 60% NEC vs 0% grade 3 NET, p = 0.013), and none were diffusely p16 positive (82% NEC vs 0% grade 3 NET, p = 0.002). The results were also statistically significant when only SCNEC were considered in the analysis (RB loss in 100% SCNEC vs 0% grade 3 NET, p < 0.001; aberrant p53 in 50% SCNEC vs 0% grade 3 NET, p = 0.035; diffuse p16 in 100% SCNEC vs 0% grade 3 NET, p = 0.001).

Genetic and immunophenotypic analysis of neuroendocrine and invasive ductal carcinoma components of mixed tumors
The neuroendocrine and IDC-NST components of two mixed NEC/ IDC-NST tumors (ANEC2 and LCNEC1) were separately analyzed by immunohistochemistry and targeted DNA sequencing ( Fig. 6 and Supplementary Fig. S4). In ANEC2, the NEC and IDC-NST components shared hotspot PIK3CA and TP53 mutations and numerous chromosomal copy number changes, consistent with shared clonality between the components. LOH of the TP53 allele and focal homozygous deletion of RB1 exon 1 were exclusive to the NEC component. Consistent with this, RB protein loss and diffuse p16 expression were also restricted to the NEC component. Whereas the IDC-NST component expressed ER and GATA3, these markers were negative in the NEC areas (Fig. 6).
The LCNEC and IDC-NST components of LCNEC1 shared a duplication involving exons 3-25 of RB1 and a TP53 nonsense mutation, as well as numerous chromosomal copy number changes, again consistent with shared clonality between the components. TP53 LOH was detected only in the NEC component. By immunohistochemistry, RB and p53 were aberrant in both components. The IDC-NST component was ER+ PR-and expressed GATA3, whereas the NEC component showed the converse immunoprofile (ER-PR+ and GATA3 negative) ( Fig. 6 and Supplementary Fig. S4).

DISCUSSION
Small cell lung carcinoma (SCLC), the prototypical poorlydifferentiated NEC, demonstrates near-universal biallelic inactivation of tumor suppressor genes TP53 and RB1 35 , and frequent  TP53/RB1 co-alteration has also been identified in small cell/ poorly-differentiated NEC of the pancreas, prostate, bladder, and colon/rectum, and in Merkel cell carcinoma [36][37][38][39][40][41][42] . In contrast to SCLC, LCNEC of the lung is genetically more heterogeneous, with TP53/RB1 co-inactivation only in a subset of tumors, while those lacking these alterations harbor mutations that are more frequently seen in pulmonary adenocarcinomas 43,44 . In contrast to NEC, TP53 and RB1 alterations are absent or exceedingly rare in extramammary NET (including G3 NET) 12,45-49 , which instead have mutations in chromatin remodeling genes MEN1 (lung and pancreas), DAXX/ATRX (pancreas), the mTOR pathway (pancreas), and CDNK1B (small intestine) [45][46][47][48][49][50][51] . In comparison to extramammary NEN, a paucity of data exists for breast tumors with neuroendocrine differentiation. SCNEC has been recognized as an aggressive type of breast cancer for many years, although most of the literature consists of case reports and small series due to the rarity of these tumors. Histochemical and ultrastructural features of SCNEC were first reported nearly two decades ago in a series of four cases, including the identification of in situ carcinoma supporting primary breast origin 52 . A subsequent series detailed the morphologic features and expanded immunohistochemical findings of nine breast SCNEC, including four that were mixed with non-NEC components 6,9 . The largest series to date comprised 19 SCNEC and identified TP53 somatic mutations in 6/8 and PIK3CA mutations in 3/9 cases using a 47-gene NGS panel 53 . No RB1 mutations were reported, and RB1 copy number and RB immunohistochemistry were not assessed. Although limited by the number of cases, our study offers a comprehensive characterization of the molecular landscape of SCNEC. Akin to SCNEC of the lung and other anatomic sites [35][36][37][38][39][40][41][42] , we identified for the first time near-universal TP53 and RB1 co-alterations in our cohort of breast SCNEC, with the sole SCNEC that lacked these alterations being RB negative by immunohistochemistry and showing p53 pathway aberrations (MDM2 and MDM4 amplifications) that are known negative regulators of p53 [30][31][32] . These findings suggest that co-inactivation of TP53 and RB1 may be important for the small cell phenotype in the breast, as at other sites [35][36][37][38][39][40][41][42] . We speculate that methodological differences, notably including the detection of deep deletions, splice site mutations, and a genomic rearrangement, rather than only coding mutations, in our series, may help explain the discrepancy between the mutation prevalence in our study and the prior study that used a small targeted gene panel 53 .
LCNEC of the breast as defined by the most recent WHO classification is thought to be exceedingly rare and remains largely uncharacterized in terms of molecular features and clinical behavior. Although limited by a small number of cases, we show here that strictly defined LCNEC appear to be genetically heterogeneous, with some but not all harboring TP53/RB1 coalteration. Two NEC with ambiguous or mixed small versus large cell features (ANEC) in our study also had TP53/RB1 co-alteration. The genetics of breast LCNEC thus appear to mirror the genetic heterogeneity described in pulmonary LCNEC, where some LCNEC are SCNEC-like and others are not 44 . Whether LCNEC of the breast with and without p53/Rb co-alterations are clinically distinct from one another and/or from other high-grade NEN of the breast will require larger follow-up studies of patients with these rare tumors.
SCNEC with mixed NEC and non-NEC components have been previously reported 6 . We describe here the genetic and immunohistochemical analysis of a mixed ANEC/IDC-NST and a mixed LCNEC/IDC-NST, both of which were also associated with nonneuroendocrine DCIS. Genetic analysis of separate invasive NEC and IDC-NST areas of these mixed tumors confirms a shared clonality between the components and raises the possibility that NEC could arise as a secondary event from ductal carcinoma in the breast rather than de novo from a committed neuroendocrine progenitor cell. Indeed, a native neuroendocrine progenitor cell has not been identified in the breast 11,54,55 . We also show that LOH of a shared TP53 mutation was restricted to the NEC components of both mixed tumors, while deep deletion of RB1 was exclusive to the NEC component in one, again suggesting a role for TP53/RB1 co-inactivation in the pathogenesis of the NEC phenotype in the breast, including at least a subset of LCNEC. On the other hand, TP53/RB1 co-inactivation is by no means exclusive to NEC in the breast and is also found in a subset of non-NEC basal-type triple negative and luminal B breast cancers [56][57][58][59] .
We identified amplification of ZNF703 in 46% of NEC (three SCNEC, one ANEC, and two LCNEC), half of which were triple negative and half of which were ER positive (one LCNEC was triple positive). FGFR1 was co-amplified in two of the triple negative tumors. ZNF703 encodes a zinc finger protein and estrogenresponsive transcriptional cofactor, which appears to be preferentially amplified in luminal B breast cancers and is associated with high proliferation and high histologic grade 60,61 . FGFR1 amplification is also common in ER-positive carcinomas and has been associated with increased grade and proliferation index 62,63 . The significance of ZNF703 amplifications in triple negative NEC is not known. However, we note that the IDC-NST components of both mixed NEC/IDC-NST tumors in our series were ER positive, whereas the clonally related NEC components were ER negative. Together, this raises speculation that at least some ER negative SCNEC and LCNEC may be intrinsic luminal B tumors that have lost ER expression during tumor progression, which could explain the enrichment of luminal-type alterations in these tumors. Intrinsic molecular subtyping would help to conclusively address this question in the future.
Inactivating CDH1 mutations were identified in one SCNEC and one LCNEC and corresponded to aberrant E-cadherin expression by immunohistochemistry, supportive of lobular differentiation of these tumors. Although neuroendocrine differentiation has been reported in rare lobular carcinomas, the identification of the lobular phenotype in SCNEC and LCNEC is, to the extent of our knowledge, a novel finding [64][65][66] .
Most previous studies of breast tumors with neuroendocrine differentiation report an association with luminal subtypes and predominantly include NET and IDC-NED using the current taxonomy; relatively few grade 3 cancers have been studied. Ang et al. identified recurrent PIK3CA (3/15 cases) and FGFR (2/ 15) alterations in their series, which included only three grade 3 cancers, two with no detected alterations and one which had a pathogenic HRAS mutation 64 . Marchiò et al. reported recurrent alterations in GATA3, FOXA1, TBX3, ARID1A, PIK3CA, AKT1, and CDH1 in a series of IDC-NED, mucinous, and solid papillary carcinomas, which included only three grade 3 cancers; FOXA1, CDH1, AKT1, and KMT2C alterations were seen in one of the latter 65 . Neither of these studies appeared to include SCNEC or LCNEC. A study by Lavigne et al. using 2012 (4th edition) WHO terminology included 15 grade 3 cancers, with TP53 mutations identified in two "poorly-differentiated neuroendocrine carcinomas" and one grade 2 NET 67 . PIK3CA mutations were identified in grade 2 and 3 NET. Wei et al. recently reported a genetic analysis of 11 NEN classified as NEC and found no TP53 or RB1 alterations in any of the tumors, although other genes in the p53 and RB pathways were mutated at 18% and 27%, respectively 68 . Neither of these latter two studies specified whether the analyzed tumors were SCNEC, LCNEC, or neither. Taken together, the genetics of NEC in our study are clearly distinct from those of the NET and IBC-NED reported in prior studies, supporting the separate classification of NEC. With respect to NEC included in the study by Wei et al., differences in tumor classification are likely to be at least partly responsible for apparent discrepancies from our results.
The 5th edition of the WHO classification of breast tumors defines NET as NEN with >90% neuroendocrine morphology and "extensive" neuroendocrine expression by immunohistochemistry. Defined as such, NET are usually Nottingham grade 1-2, and such tumors can be readily distinguished from NEC, which are high-grade tumors that resemble pulmonary NEC 2,5,11 . However, NEN with intermediate nuclear grade and high mitotic activity (Nottingham grade 3) that do not resemble pulmonary NEC morphologically are occasionally encountered in practice, although more specific diagnostic criteria to distinguish these tumors from NEC are lacking. Classification of such grade 3 NEN tumors using the WHO system is nebulous. Indeed, they are not excluded from the NET definition and can be best classified as grade 3 NET if adhering strictly to the diagnostic criteria 1,11 . On the other hand, it is unclear if some authors interpret the WHO classification such that all grade 3 NEN are NEC and all grade 1-2 NEN are NET 68,69 . In the enteropancreatic system, it has been established that NET with well-differentiated morphology but high mitotic activity and/or Ki-67 index (grade 3 NET) are genetically and clinically distinct from NEC. An analogous dichotomy of grade 3 NEN has not been established in the breast. In this study, we classified a group of uncommon Nottingham grade 3 NEN that did not morphologically resemble NEC of the lung as grade 3 NET and compared them to our cohort of NEC. None of the grade 3 NET showed aberrant RB or p53 staining patterns or diffuse p16 expression, in contrast to the high frequency seen in the NEC, including both SCNEC and LCNEC. We note that immunohistochemical expression of p53 and especially RB reliably correlated with underlying RB1 and TP53 genetic alterations in this study and others 33 . Strong diffuse overexpression of the cyclin-dependent kinase inhibitor p16 also correlated exclusively with RB alteration, presumably due to the known negative feedback regulation of p16 by RB, and was only seen in NEC 70,71 . The immunohistochemical findings thus suggest differences in the underlying genetics and biology of NEC compared to tumors classified as grade 3 NET in our study. Significantly higher frequencies of ER and GATA3 expression in grade 3 NET further support an alternate phenotype from NEC. The small number of cases in our series limits any meaningful comparative outcome analysis. However, we note that 60% of patients with NEC developed distant metastases and 50% died of disease (mean follow-up 21 months, range 3-67), whereas no such adverse events were found in patients with grade 3 NET (mean follow-up 32 months, range 3-88). Overall, additional molecular and clinical outcome studies with larger numbers of carefully classified tumors will be essential to determine whether tumors with features of grade 3 NET are biologically distinct from NEC, or whether they should be considered along the spectrum of a heterogeneous group of LCNEC without p53/RB alterations. In this context and given the difficulty in morphologically distinguishing some LCNEC from grade 3 NET, we can speculate that classification of grade 3 NEN based on p53/RB alteration rather than morphology may be clinically relevant.
Our findings raise important questions regarding whether SCNEC of the breast should be managed distinctly from other high-grade breast carcinomas, especially triple-negative carcinomas. We have shown that SCNEC of the breast is convergent genetically, as well as histologically, with SCNEC of the lung, with frequent concurrent loss of p53 and RB. However, we also uncovered clonal origin of mixed NEC/IDC-NST, suggesting that SCNEC may genetically descend from an earlier IDC clone, rather than a distinct progenitor cell. Numerous case reports in the literature reflect the current uncertainty around optimal management of mammary SCNEC, including frequent reports of etoposide-based regimens that are used for SCNEC of the lung and not standardly for breast cancer 29,[72][73][74][75][76][77] . In our series, the majority of patients with SCNEC were treated with etoposide (6/7, 86%), indicating that the diagnosis can influence medical oncologists' choice of therapy, even in the absence of specific guidelines.
In summary, we have provided a comprehensive genetic characterization of carefully classified NEC of the breast and show that p53 and RB pathway co-alterations are highly prevalent in SCNEC and a subset of LCNEC, similar to their counterparts in the lung 35,43,44 . Genetic and immunophenotypic analysis of paired NEC and IDC-NST components of mixed tumors confirms their shared clonality and suggests that the NEC phenotype could arise from ductal carcinoma in the breast. Classification of uncommon grade 3 NEN without classic features of NEC is problematic in the most recent WHO classification, and our data suggest that these tumors may be biologically distinct from NEC, or, alternatively, that they fall along a spectrum of heterogeneous LCNEC that lack p53 and RB alterations. The clinical significance of recognizing grade 3 NEN with and without p53/RB alterations awaits further study.

DATA AVAILABILITY
Data generated and analyzed during the current study are available from the corresponding author on reasonable request.