IDH2 Mutations Define a Unique Subtype of Breast Cancer with Altered Nuclear Polarity

Breast cancer is a heterogeneous disease at the clinical, morphologic, and genetic level (1–3). The current World Health Organization (WHO) classification recognizes 21 histologic types (1), but few have been found to harbor specific genetic alterations, such as CDH1 alterations in invasive lobular carcinoma (4) and recurrent ETV6–NTRK3 and MYB–NFIB fusion genes in secretory and adenoid cystic carcinomas, respectively (5–7). Additional rare breast cancers with distinctive morphologic features also exist, but have not yet been included in the WHO classification. One such malignancy has been described as “breast tumor resembling the tall cell variant of papillary thyroid carcinoma (PTC)” (8, 9), which has a unique histologic appearance consisting of solid, circumscribed nodules of epithelial cells, many of which have fibrovascular cores, resulting in a solid papillary growth pattern. Thirteen such tumors have been reported to date (8, 9), of which approximately half lack estrogen (ER) and progesterone receptor (PR) expression, and all lack HER2 overexpression (8, 9). These tumors are generally associated with a favorable prognosis (8, 9); however, two of 13 patients have been reported to develop metastatic disease, one to an intra-mammary lymph node and another to the bone (8, 9).

Although these tumors have some histologic similarities to PTC, they consistently lack expression of thyroid-specific markers, including TTF-1 and thyroglobulin (8, 9). RET rearrangements and BRAF exon 15 mutations commonly found in PTC have also not been detected in these tumors (9). Given the morphologic overlap with other papillary lesions of the breast and lack of immunohistochemical and genetic evidence of an association with PTC, it has recently been suggested that these neoplasms should be considered morphologic variants of papillary breast carcinoma (9, 10).

In this study, we sought to characterize the morphologic and genetic landscape of this rare and morphologically unusual breast tumor and determine whether it represents a distinct subtype of breast cancer underpinned by disease-specific genetic alterations. To this end, we performed an extensive immunohistochemical characterization and whole exome (WES), targeted, and Sanger sequencing of 13 previously unreported tumors. We found that 10 of 13 (77%) tumors harbored R172 IDH2 mutations, of which eight had a concurrent pathogenic mutation affecting PI3K pathway canonical genes (i.e., six PIK3CA hotspot mutations and two PIK3R1 likely pathogenic mutations). In addition, a PIK3CA-mutant but IDH2-wild-type tumor was found to harbor a TET2 Q548* truncating mutation. Functional studies using nonmalignant breast epithelial cells demonstrated that IDH2 and PIK3CA mutations constitute likely drivers of this tumor and contribute to its characteristic phenotype. Because of their unique histologic and genetic properties, we redefine these tumors here as a discrete subtype of breast carcinoma: solid papillary carcinoma with reverse polarity (SPCRP).

Between 2005 and 2014, 13 SPCRPs were identified by two of the authors (S.J. Schnitt and E. Brogi) on the Breast Pathology Consultation (11 cases) or Breast Pathology (1 case) Services at Beth Israel Deaconess Medical Center (Boston, MA), and the Breast Pathology Consultation Service (1 case) at Memorial Sloan Kettering Cancer Center (MSKCC; New York, NY). Hematoxylin-and-eosin–stained sections were reviewed to delineate the morphologic features of these lesions. Available clinical data were reviewed for demographics, presentation, clinical and family history, treatment, and outcome. The study was approved by the Institutional Review Boards from the respective authors' institutions, and all samples were anonymized prior to analysis.

Primary antibodies for IHC were as follows: calponin, smooth muscle myosin-heavy chain (SMMHC), p63, AE1/AE3, cytokeratin (CK) 7, CK5/6, CK34βE12, ER, PR, androgen receptor (AR), HER2, mammaglobin, H3K27me3, gross cystic disease fluid protein-15 (GCDFP-15), TTF-1, thyroglobulin, E-cadherin, and MUC1 (Supplementary Table S1). For H3K27me3, we semiquantitatively assessed its nuclear expression using the H-score system as previously described (11); 500 nuclei were counted per case.

Standard Western blotting was performed as previously described (12) using primary antibodies against IDH2 and H3K9me3 (Abcam); phospho-Rb (S807/811), Rb (4H1), histone H3 (D1H2), and α-tubulin (DM1A; Cell Signaling Technology); H3K27me3 (EMD Millipore), and E-cadherin (BD Biosciences). Secondary conjugated IRDye 680RD/800CW antibodies (LI-COR Biosciences) were used. Detection and quantification were performed using the Image Studio Software from LI-COR (LI-COR Biosciences; refs. 13, 14).

Eight micron–thick representative formalin-fixed paraffin-embedded (FFPE) tumor and normal tissue sections from two SPCRPs (cases 10 and 13), and tumor sections from two SPCRPs (cases 5 and 11) were stained with nuclear fast red and microdissected using a sterile needle under a stereomicroscope (Olympus SZ61), to ensure >80% tumor cell content and that normal tissue was devoid of neoplastic cells as previously described (5, 14). DNA was extracted from microdissected tumor and normal samples using the DNeasy Blood and Tissue Kit (Qiagen) following the manufacturer's guidelines. Tumor and matched germline DNA from SPCRP10 and SPCRP13 were subjected to WES on a Illumina HiSeq2000 at the Integrated Genomics Operation (IGO) at MSKCC (14), and tumor DNA from SPCRP5 and SPCRP11 to Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) sequencing assay (15) targeting all exons and selected regulatory regions and introns of 410 key cancer genes on an Illumina HiSeq2500 (16). Massively parallel sequencing analysis was performed as previously described (Supplementary Methods; refs. 16, 17). WES and MSK-IMPACT data have been submitted to SRA under the accessions SRP066429 and SRP066430, respectively.

Representative FFPE sections of each case were reviewed to select the most appropriate tumor-enriched area for analysis. Tumor tissue was manually microdissected from serial 5-μm–thick unstained sections. Total nucleic acid was extracted from FFPE tumor tissue using a modified FormaPure system (Agencourt Bioscience Corporation) as previously described (18). Genotyping analysis of specific mutations in 22 clinically relevant cancer genes (Supplementary Table S2) was performed on six SPCRPs (cases 2, 4, 7–9, and 12) for which insufficient material for massively parallel sequencing was available using a custom Applied Biosystem (ABI) Prism SNaPshot Multiplex platform, as previously reported (18).

For three cases (SPCRP1, 3, and 6) with very limited tissue availability, DNA was subjected to PCR amplification using primer sets encompassing the IDH2 R140 and R172 hotspot residues and the PIK3CA E542/545 and H1047 hotspot residues (Supplementary Table S3), followed by standard Sanger sequencing as previously described (14). All analyses were performed in duplicate. Sequences of the forward and reverse strands were analyzed using MacVector software (MacVector, Inc; ref. 14).

Methylation profiling was performed using the Infinium MethylationEPIC Kit (Illumina). DNA samples from five IDH2-mutant SPCRPs (cases 3–5, 7, and 9), one TET2-mutant SPCRP (case 11), and two invasive ductal carcinomas of no special type (IDCs, one ER-positive/HER2-negative and one ER-negative/HER2-negative) were bisulfite-converted using the EZ-96 DNA Methylation Kit (Zymo Research), restored using the Illumina Infinium HD FFPE DNA Restore Kit, and whole-genome amplified (Supplementary Methods). The BeadChips were scanned, and the raw data files containing the fluorescence intensity data for each probe were generated. Data analysis was performed using RnBeads (Supplementary Methods; ref. 19).

The human IDH2 (NM_002168) cDNA ORF clone pCMV6-IDH2::Myc-DDK was purchased from Origene (RC201152), and the R172S mutation introduced using the Q5 Site-Directed Mutagenesis Kit (New England Biolabs) following the manufacturer's recommendations. The presence of the correct mutation was confirmed by Sanger sequencing (for primers, see Supplementary Table S3). IDH2 wild-type and mutant (R172S) open reading frames were cloned into the pCMV6-TagRFP or pCMV6-TagGFP vectors to generate pCMV6-IDH2::TagRFP/GFP and pCMV6-IDH2 (R172S)::TagRFP/GFP plasmids, respectively, as previously described (13, 14). Transfections of plasmids were performed using Lipofectamine 3000 reagent according to the manufacturer's guidelines (13). IDH2 and IDH2 R172S expression was verified by immunofluorescence-based mitochondrial colocalization using MitoTracker staining (Thermo Fisher Scientific), quantitative RT-PCR, and whole-cell lysate protein detection.

TaqMan quantitative RT-PCR (qRT-PCR; Life Technologies) was performed for IDH2 (Hs00158033_m1), Twist1 (Hs01675818_s1), SNAI1 (Hs00195591_m1), SNAI2 (Hs00161904_m1), FN1 (Hs01549976_m1), and using GAPDH (Hs99999905_m1) for expression assay normalization, as previously described (5, 20).

Ten serial 10 μm–thick representative tumor sections were obtained from SPCRP12 and five IDCs, the latter identified in the pathology database at Massachusetts General Hospital and diagnosis confirmed. Tumors were manually microdissected. 2-Hydroxyglutarate (2HG) was analyzed in lysates from these samples using gas chromatography/mass spectrometry, as previously described (21).

Parental MCF10A (MCF10A^P) and MCF10A PIK3CA H1047R (MCF10A^H1047R) knock-in cells were obtained from Horizon in December 2014. The identities of the cell lines were confirmed by short tandem repeat profiling using the GenePrint 10 System (Promega) in the IGO at MSKCC three months after receipt. In addition, cells were tested for mycoplasma infection using a PCR-based test (ATCC). Confluent cells (MCF10A^P and MCF10A^H1047R, expressing vector control, IDH2 wild-type, or IDH2 R172S plasmids) were collected from 6-well plates, and 2HG detection was performed in conditioned media and in cell lysates, as previously reported (22; Supplementary Methods). Fluorometric detection was carried out in triplicate with excitation at 540 nm and emission of 610 nm (Victor X4 Multimode Plate Reader; Supplementary Methods).

MCF10A^P and MCF10^H1047R cells were transfected with pCMV6-TagRFP empty vector, pCMV6-TagRFP-IDH2^WT or pCMV6-TagRFP-IDH2^R172S using Lipofectamine 3000 (Invitrogen) as per manufacturer's instructions. After 48 hours, cells were trypsinized, collected and RFP-, RFP-IDH2^WT-, and RFP-IDH2^R172S–expressing cells selected by flow cytometry sorting. For proliferation, the sorted cells were seeded in 96-well plates (1,000 cells/well) in triplicates. Cell growth was assessed every 24 hours by CellTiter-Blue assay (Promega) and fluorometric detection performed with 560 nm excitation and 590 nm emission using a VICTOR X4 Multilabel Plate Reader (PerkinElmer). For migration, cells were starved for 16 hours in 2% horse serum DMEM/F12 medium without EGF. Cells were seeded (0.5 × 10⁵) in the top compartment of a 24-well Transwell insert (8-μm pore membranes; Corning) and allowed to adhere and migrate at 37°C toward 10% horse serum and EGF-containing DMEM/F12 medium for 16 hours. Cells adherent to the top side of the Transwell membrane were removed, and cells on the inferior side of were fixed in 4% paraformaldehyde (PFA), permeabilized (0.5% Triton X-100 in PBS), and stained with crystal violet (0.3%). Transwell membranes were imaged (EVOS XL Imaging System, Life Technologies) and analyzed by ImageJ.

3D cell culture assays were carried out as previously described (13, 14, 23, 24). Phase-contrast images of acinar structures were acquired after six days (EVOS XL Imaging System). Size (mm³) of 3D acinar structures was determined using Fiji (ImageJ software) to define Feret diameter (D) and minimum Feret diameter (d) and applying the formula (D × d²)/2 = acini volume. Immunofluorescent staining of 3D acini (12 days) was performed as previously described (24). The structures were fixed in 4% PFA and permeabilized (0.5% Triton X-100 in PBS), stained using antibodies against GM130 (D6B1) and E-cadherin (Cell Signaling Technology), and Alexa Fluor–conjugated secondary antibodies and ProLong Gold antifade reagent with DAPI (Life Technologies), and imaged by confocal analyses [Leica SP5 DM, including an ultraviolet (UV) diode (405 nm), argon laser (458, 476, 488, and 514 nm), 543-nm HeNe laser, and 633-nm HeNe laser]. Polarity assessment was performed with the observer blinded to the genotype of the cell line being analyzed, with apical polarity defined as nuclei at the basal pole (near the periphery of the acini) and Golgi apparatus facing upward (toward the lumen), and reverse polarity as nuclei at the apical pole (toward to the lumen) and Golgi apparatus facing away from the lumen (near the periphery of the acini). A minimum of 100 abluminally located cells were counted per experimental condition in each cell line.

A cell viability assay for the assessment of oxidative stress was performed basically as previously described (25). Vector-, IDH2- or IDH2 R172S–expressing MCF10A^P and MCF10A^H1047R were grown overnight on a 96-well plate at a density of 1 × 10⁴ cells/well, treated with 1 mmol/L hydrogen peroxide (H₂O₂) in serum-free medium for 48 hours at 37°C, and cell viability was assessed using CellTiter-Blue as described above and reported as compared with absorbance measured in untreated control cells. Fluorometric detection was carried out in triplicate in two independent experiments.

SPCRP has a very distinctive histologic appearance that allowed us to identify a series of 13 such tumors from our consultation and institutional files. They consisted of solid, circumscribed nodules of columnar epithelial cells, some exhibiting a geographic, jigsaw-like growth pattern (Fig. 1A–C). Many nodules contained fibrovascular cores with foamy histiocytes and a double-layered epithelium without a clear lumen, resulting in a solid papillary appearance (Fig. 1B). The cells in many nodules appeared back-to-back and, in particular, the nuclei were present at the apical rather than basal pole of the cells, creating the impression of reverse polarization (Fig. 1C).

In all cases, all tumor nodules lacked a surrounding myoepithelial cell layer as assessed by immunohistochemical analysis for p63 (Fig. 1E), calponin, and SMMHC, supporting the invasive nature of these lesions. Tumor cells showed strong cytoplasmic expression of low molecular weight CK7, but also strongly expressed high-molecular weight/basal CK5/6 (Fig. 1D) and CK34βE12. GCDFP-15 and mammaglobin expression were detected in 62% and 58% of cases, respectively, supporting the breast origin of the tumors, whereas thyroglobulin and TTF-1 were negative in all tumors, excluding metastases of a thyroid carcinoma. In 62% of cases, tumor cells were entirely ER-negative (Fig. 1F); ER expression was seen in 1% to 10% of cells in the remaining tumors. PR expression was seen in 15% of cases, and all cases studied were negative for HER2 protein overexpression. AR was focally positive in only one case. In two cases studied, the Ki67 proliferation rate was low (<5%). E-cadherin staining revealed strong lateral membrane expression with absent apical or basal expression (Fig. 1G), suggesting that the epithelium is polarized. MUC1 staining, which highlights the apical membranes of columnar epithelial cells, was identified at the end of the cell closest to the nucleus, indicating that the nuclei were in an abnormal location (apical rather than basal) and creating the impression of reversed cell polarity (Fig. 1H). Clinical data are summarized in Table 1. 

Two SPCRPs (cases 10 and 13), for which sufficient tumor and normal DNA samples were available, and two SPCRPs (cases 5 and 11) with sufficient tumor DNA were subjected to WES and the MSK-IMPACT sequencing assay (15), respectively. In cases subjected to WES (cases 10 and 13), the median sequencing depth of coverage was 127× (range, 111–212×), and 45 and 24 nonsynonymous somatic mutations were identified (Supplementary Tables S4 and S5). Cases 5 and 11, subjected to MSK-IMPACT sequencing at a coverage if 734× and 573×, respectively, were found to harbor eight and 26 nonsynonymous somatic mutations, respectively (Supplementary Tables S4 and S5). Sequencing analysis revealed that three of these cases harbored somatic single nucleotide variants (SNV) affecting the R172 residue in the substrate-binding pocket of the isocitrate dehydrogenase 2 (NADP⁺), mitochondrial (IDH2) gene (Fig. 2A; Supplementary Table S5), the same IDH2 codon targeted by recurrent hotspot mutations in glioma, acute myeloid leukemia (AML), cholangiocarcinoma, and chondrosarcoma (26). In SPCRP11, which lacked an IDH2 somatic mutation, a TET2 Q548* truncating mutation was detected. We also identified mutations affecting canonical genes of the PI3K/AKT/mTOR pathway in all four SPCRPs, with two cases harboring PIK3R1 frameshift mutations and two cases harboring PIK3CA somatic mutations (Fig. 2A; Supplementary Table S5). No other recurrent somatic mutations were found. To validate these findings, we subjected DNA derived from an additional six SPCRPs (cases 2, 4, 7–9, and 12) to SNaPshot profiling assessing mutations in 22 cancer genes (Supplementary Table S2) and from three SPCRPs (cases 1, 3, and 6) to IDH2 and PIK3CA hotspot Sanger sequencing (Supplementary Fig. S1; Supplementary Table S3), which confirmed the presence of IDH2 R172 hotspot mutations in seven cases (70%), and PIK3CA H1047R and E542K hotspot mutations in six cases and one case (78%), respectively. No other hotspot mutations were identified in the SPCRPs subjected to SNaPshot profiling. In total, 10 of 13 (77%) SPCRPs analyzed here harbored an IDH2 R172 mutation, of which five had a concurrent PIK3CA H1047R hotspot mutation, one a concurrent PIK3CA C420R mutation, and two a concurrent PIK3R1 frameshift mutation (Fig. 2A). Given that one of the IDH2 wild-type SPCRPs harbored a truncating mutation in TET2, which encodes for an α-ketoglutarate (αKG)-dependent enzyme that catalyzes cytosine 5-hydroxymethylation resulting in demethylation of DNA (27), we sought to define whether the IDH2 wild-type cases SPCRP1 and SPCRP6 would also harbor mutations affecting this gene; however, both were found to be TET2 wild-type. No histologic differences were observed among the IDH2-mutant, TET2-mutant, and IDH2/TET2 wild-type SPCRPs.

To date, IDH2 mutations have not been described in breast cancer, and a reanalysis of breast cancers reported by The Cancer Genome Atlas (TCGA) revealed that only one of the 971 invasive ductal and invasive lobular carcinomas harbored an IDH2 mutation, however affecting a different codon (E345K, Fig. 2B; www.cBioPortal.org, accessed on November 30, 2015; Fig. 2B; refs. 4, 28). Taken together, our results indicate that SPCRPs are uniquely characterized by highly recurrent IDH2 R172 hotspot mutations often in combination with mutations affecting PI3K pathway canonical genes, in particular in the form of PIK3CA H1047R hotspot mutations, and suggest that at least in a subset of SPCRPs lacking IDH2 mutations, somatic genetic alteration affecting TET2 may be present. These observations suggest that SPCRPs may constitute an example of a convergent phenotype stemming from alterations of genes leading to similar epigenetic defects and potentially function genetically in the same pathway (26, 29), as IDH1/IDH2 mutations inhibit TET2 function and are mutually exclusive with TET2 mutations in hematologic malignancies (30).

IDH2 hotspot mutations are enzymatic gain-of-function alterations that lead to an increased conversion of αKG to 2HG. Increased levels of 2HG result in hypermethylation of epigenetic targets and a subsequent block in cellular differentiation (31, 32). We first sought to define whether IDH2 mutations result in accumulation of 2HG oncometabolite in SPCRP as reported in glioblastoma (33) and AML (34) harboring this mutation. We measured intratumoral 2HG by gas chromatography/mass spectrometry in one IDH2-mutant SPCRP (case 12), where adequately preserved tissue was available, and five IDH-wild-type IDCs with available material. The SPCRP harboring the IDH2 R172T mutation demonstrated an elevated 2HG level of 49.36 pmol/μg protein (Fig. 2C). In four IDH2-mutant SPCRPs, tissue samples were inadequate and rendered uninterpretable levels of 2HG (data not shown). All five IDH2-wild-type IDCs harbored undetectable levels of 2HG (Fig. 2C). We next sought to define levels of methylation in IDH2/TET2-mutant SPCRPs, for which DNA samples were available. Genome-wide DNA methylation analysis using the Illumina Infinium MethylationEPIC BeadChip revealed that IDH2/TET2-mutant SPCRPs showed a genome-wide hypermethylation profile as compared with IDH2 wild-type IDCs, consistent with the hypermethylation profile reported in IDH1/IDH2-mutant cancers (Fig. 2D and Supplementary Fig. S2; refs. 30, 35). In fact, hierarchical clustering revealed that IDH2/TET2–mutant SPCRPs clustered together based on their methylation profile, and separate from the two IDH2/TET2 wild-type IDCs analyzed (Fig. 2D). Finally, the protein expression levels of H3K27me3 were assessed by IHC in four IDH2-mutant SPCRPs for which adequate histologic sections were available, and in four IDCs (two ER-positive/HER2-negative and two ER-negative/HER2-negative). Compared with IDH2 wild-type IDCs, IDH2-mutant SPCRPs displayed significantly higher levels of trimethylation of H3K27 (H3K27me3, P = 0.029, Mann–Whitney U test; Fig. 2E), consistent with the results reported in IDH-mutant gliomas (31).

We next tested the functional impact of the IDH2 R172S mutation on the growth and phenotype of nonmalignant breast epithelial cells and investigated potential epistatic interactions between IDH2 R172S and the most frequent PIK3CA (H1047R) mutations concurrently detected in SPCRPs. We used the MCF10A model system, including parental MCF10A cells and MCF10A cells harboring a stable knock-in of the PIK3CA H1047R mutation. As expected, forced expression of wild-type IDH2 (IDH2^WT) and R172S mutant IDH2 (IDH2^R172S) in parental MCF10A cells (MCF10A^P) and in PIK3CA H1047R–mutant MCF10A cells (MCF10A^H1047R) resulted in increased IDH2 mRNA expression (Supplementary Fig. S3A) and IDH2 protein mitochondrial localization (Supplementary Fig. S3B). When expressed at similar protein levels, IDH2^R172S resulted in significantly higher neomorphic enzymatic activity than IDH2^WT based on the analysis of the production and secretion of 2HG, regardless of the presence of the H1047R PIK3CA mutation (Fig. 3A).

IDH1 and IDH2 mutations have been shown to constitute drivers of glioma (36, 37), AML (34, 38), and spindle cell hemangioma (39, 40) among others, resulting in a differentiation block and tumorigenesis (31, 32). Although in our transient transfection system, no statistically significant differences in the expression levels of H3K27 and H3K9 trimethylation were observed upon forced expression of IDH2^WT or IDH2^R172S as compared with cells expressing empty vector (Supplementary Fig. S4), expression of IDH2^R172S resulted in changes in expression of cell-cycle and epithelial–mesenchymal transition (EMT) markers in the nonmalignant breast epithelial cells, MCF10A. Forced expression of IDH2^R172S in MCF10A^H1047R cells led to significantly increased phosphorylation of RB1 protein compared with empty vector, as detected by immunoblot (Fig. 3B). Consistent with these observations, we observed a significant increase in cell growth upon forced expression of IDH2^WT and IDH2^R172S in MCF10A^H1047R cells (Fig. 4A), as well as a significant increase in migration of MCF10A^WT and MCF10A^H1047R cells expressing R172S-mutant IDH2 (Fig. 4B). Furthermore, we observed that forced expression of IDH2^WT and IDH2^R172S led to a reduction of E-cadherin protein expression in MCF10A^H1047R cells (Fig. 3B) and resulted in significantly increased levels of EMT markers in MCF10A cells harboring wild-type or H1047R-mutant PIK3CA (Fig. 4C).

Given that cancer cells can avoid cell death by inhibiting the initial deleterious effects of oxidative stress (25, 41), we sought to define the impact of IDH2^R172S on the viability of nonmalignant breast epithelial cells. In MCF10A^P cells treated with exogenous H₂O₂, forced expression of wild-type, but not mutant IDH2, displayed a higher viability than empty vector expression (Fig. 4D). In H₂O₂-treated MCF10A^H1047R cells, however, forced expression of both wild-type and mutant IDH2 resulted in increased viability compared with the empty vector control (Fig. 4D). These results suggest that IDH2^R172S may protect against reactive oxygen species (ROS) in nonmalignant breast epithelial cells, consistent with observations described previously for IDH1 and IDH2 in other nonneoplastic tissue types (25, 41).

We next sought to define the impact of IDH2^R172S on the growth and glandular architecture of MCF10A cells grown in a 3D model system, which has been previously used to assess the oncogenic properties of somatic mutations (13, 23, 42). Forced expression of IDH2^R172S in MCF10A^P and MCF10A^H1047R cells resulted in acinar structures that were significantly larger than those observed in MCF10A cells transfected with empty vector or IDH2^WT (Fig. 5A), a phenotype previously reported to be elicited by the expression of bona fide oncoproteins in this model system (13, 23, 42). In fact, IDH2^R172S expression in MCF10A^H1047R cells led to the formation of anastomosing solid cell nests/cell nodules (Fig. 5A and Supplementary Fig. S5). Consistent with the reverse polarization observed in the primary tumors, IDH2^R172S expression in MCF10A cells grown in the 3D system also resulted in an increase in cells displaying a reversed nuclear polarity, as defined by expression of adhesion molecule E-cadherin (CDH1) and apical Golgi marker GM130. This reversed nuclear polarity phenotype was particularly evident in MCF10A^H1047R cells expressing IDH2^R172S (Fig. 5B). Taken together, our data provide evidence that IDH2^R172S and PIK3CA^H1047R hotspot mutations constitute likely drivers of SPCRP, and that together these somatic genetic alterations are likely sufficient, but not necessarily required, to cause its unusual reverse polarization phenotype.

Here, we characterize the morphologic, immunohistochemical, and genomic profile of SPCRP, a rare and histologically distinct subtype of invasive breast carcinoma. Through whole exome and targeted massively parallel sequencing analysis, we have identified IDH2 hotspot and TET2 truncating mutations in 77% and 8% of these cases, respectively. In fact, SPCRPs may constitute an example of a convergent phenotype as IDH2 and TET2 mutations lead to similar epigenetic defects (26, 29). This is the first report of IDH2 hotspot mutations detected in breast cancer, expanding the spectrum of solid tumors and hematologic malignancies in which IDH mutations may drive tumorigenesis (43). The high frequency of IDH2 mutations found in SPCRP highlights the association between genotype and the unique tumor morphology (29). In addition, mutations affecting canonical genes of the PI3K pathway were found in 85% of SPCRPs, including 54% harboring PIK3CA H1047R mutations, a common genetic alteration in breast cancer that may be enriched in this subset of tumors (44).

A striking feature of SPCRP is the unique epithelial morphology of a double layer of columnar cells with apical nuclei. The absence of expression of E-cadherin, a protein known to play an important role in cell polarity, in apical and/or basal membranes combined with strong MUC1 staining of apical membranes indicates that the cells are indeed polarized. The presence of nuclei near the apical membrane indicates reverse nuclear polarization, and loss of apical polarity was also observed in our cell line models, in particular in cells expressing mutant forms of IDH2 and PIK3CA, suggesting that this phenotype may result from epistatic interactions between mutations affecting these genes. Although our findings provide a novel example of a genotypic–phenotypic correlation in breast cancer, further analyses are required to define the genetic basis of the reverse polarity in SPCRPs lacking IDH2, TET2, and/or PI3K pathway gene mutations.

The mechanism by which IDH1 or IDH2 mutations cause human cancer remains to be fully elucidated. It has been found, however, that these alterations lead to a gain-of-function enzymatic activity that allows NADPH-dependent reduction of αKG to 2HG within tumor cells (33, 34), which inhibits αKG-dependent dioxygenases and alters genome-wide histone and DNA methylation, cell differentiation and survival, and extracellular matrix maturation (26, 31, 45). High 2HG concentrations disrupt TET2 catalytic function and prevent hydroxylation of 5-methylcytosine, resulting in attenuated TET2-dependent demethylation of DNA (30, 45). Akin to IDH-mutant glioma, AML and common forms of breast carcinoma harboring IDH1 mutations (33, 34, 46), we found high concentrations of intratumoral 2HG in the IDH2-mutant SPCRP successfully tested; the use of archival material makes this analysis challenging as formalin fixation and paraffin embedding may lead to oncometabolite loss (21). In addition, we observed global DNA hypermethylation and H3K27 trimethylation in IDH2/TET2–mutant SPCRPs as compared with IDH2 wild-type IDCs, consistent with those reported in IDH1/IDH2–mutant cancers (26, 30, 31, 35). IDH1 and IDH2 have been described previously to protect cells from ROS (25, 41). We observed that IDH2^R172S provided ROS protection in MCF10A^H1047R but not in MCF10A^P cells. Further studies are warranted to assess the potential mechanistic interaction between mutant IDH2, mutant PI3K and ROS, in particular because intracellular ROS levels have been reported to affect the PI3K pathway (for a review, see ref. 47).

IDH2 mutations in SPCRP may serve as a novel target for therapeutic intervention in breast cancer. Acquired enzymatic activity resulting in 2HG accumulation is specific to tumor cells, making mutant-specific small-molecule inhibitors an attractive alternative to chemotherapy regimens with systemic toxic effects. IDH2 inhibitors have now entered phase I clinical trials for the treatment of AML with favorable safety profiles and durable clinical activity (48). Clinical benefit in the solid tumor patient population has yet to be demonstrated.

This study has important limitations. First, owing to the rarity of SPCRP, we studied 13 bona fide cases of this entity. Despite the small sample size, this is the largest collection of SPCRPs reported to date. Second, most samples were individual contributions from different institutions; hence, we were unable to perform a survival analysis to ascertain the impact of IDH2 mutations on the outcome of SPCRP. Importantly, given that none of the patients included in this study developed a distant relapse during the follow-up period available and that less than 25% of SPCRPs lack IDH2 hotspot mutations, it is unlikely that these mutations would be of prognostic significance. Third, we were unable to identify the driver genetic alterations in two SPCRPs lacking IDH2 hotspot and TET2 mutations by Sanger sequencing; however, due to the rarity of IDH2/TET2–wild-type SPCRPs and the limited material available from the cases analyzed, we were unable to sequence the entire coding region of TET2 in the remaining cases. Further studies of IDH2 wild-type SPCRPs are warranted.

Despite these limitations, here we have identified somatic IDH2 hotspot mutations or TET2 mutations in conjunction with mutations affecting PI3K pathway genes in SPCRP, validating this rare breast cancer as a unique clinicopathologic entity underpinned by a distinctive constellation of somatic mutations. Detection of IDH2 mutations may serve as an ancillary marker for the diagnosis of SPCRP. Broad-base genetic profiling in breast cancer patients who develop progressive disease with this tumor type may efficiently identify those eligible for clinical trials for IDH inhibition.

S. Pusch is a patent holder of 2 HG assay patent. A.J. Iafrate has ownership interest (including patents) in ArcherDX and is a consultant/advisory board member for Roche. No potential conflicts of interest were disclosed by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Conception and design: S. Chiang, B. Weigelt, H.-C. Wen, A.J. Iafrate, J.S. Reis-Filho, S.J. Schnitt

Development of methodology: H.-C. Wen, A. Raghavendra, S. Piscuoglio, A.A. Jungbluth, A.v. Deimling, A.J. Iafrate, S.J. Schnitt

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Chiang, H.-C. Wen, F. Pareja, L.G. Martelotto, T. Basili, A. Li, F.C. Geyer, S. Piscuoglio, A.A. Jungbluth, J. Balss, G.M. Baker, K.S. Cole, J.M. Batten, J.D. Marotti, J. Serrano, K.P. Siziopikou, S. Lu, X. Liu, T. Hammour, E. Brogi, M. Snuderl, A.J. Iafrate, J.S. Reis-Filho, S.J. Schnitt

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Chiang, B. Weigelt, H.-C. Wen, A. Raghavendra, L.G. Martelotto, K.A. Burke, S. Piscuoglio, C.K.Y. Ng, J. Balss, R.S. Lim, A.J. Iafrate, J.S. Reis-Filho, S.J. Schnitt

Writing, review, and/or revision of the manuscript: S. Chiang, B. Weigelt, H.-C. Wen, A. Raghavendra, L.G. Martelotto, K.A. Burke, S. Piscuoglio, A.v. Deimling, J.D. Marotti, K.P. Siziopikou, E. Brogi, M. Snuderl, A.J. Iafrate, J.S. Reis-Filho, S.J. Schnitt

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Pusch, A.v. Deimling, B.L. McCalip, J. Serrano, A.J. Iafrate, S.J. Schnitt

Study supervision: A.J. Iafrate, S.J. Schnitt

Other (provided one of the breast cases involved in this study): H.-C. Soh

S. Piscuoglio was funded in part by a Susan G. Komen Postdoctoral Fellowship Grant (PDF14298348). J.S. Reis-Filho is funded in part by BCRF, and M. Snuderl is supported by the Friedberg Charitable Foundation. S. Chiang, B. Weigelt, H.-C. Wen, F. Pareja, A. Raghavendra, L.G. Martelotto, K.A. Burke, T. Basili, A. Li, F.C. Geyer, S. Piscuoglio, C.K.Y. Ng, A.A. Jungbluth, R.S. Lim, E. Brogi, and J.S. Reis-Filho were funded in part by a NIH/NCI Cancer Center Support Grant (P30 CA008748).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.