ERα-LBD is encoded by an ESR1 transcript variant identified in both cell lines and primary breast cancers

To understand the basis of ERα-LBD formation we carried out RNA capture-sequencing and PCR assays to analyze the usage of ESR1 exons and transcripts in different models, including both ERα-FL positive and negative cells. High-depth capture-sequencing data showed that BC ERα-FL(−) cells expressing exclusively ERα-LBD, such as MDA-MB-453/-231 and PDX-ERα, have higher abundance of ESR1 read counts in the C-terminus (CDS exons #4, 5, 6, 7 and 8), compared to the N-terminus (CDS exons #1–3). In contrast, ESR1 exon usage profile was significantly different in ERα-FL(+) cells, such as T-47D, MCF-7, and PDX-ERα(+), where the expression of exons #1–3 was higher than exons #4–8 (Fig. ​(Fig.2a2a and Supplementary Data 1). To identify an ESR1 transcript variant for ERα-LBD, we searched the FANTOM CAT (CAGE Associated Transcriptome) database, which displayed an additional ESR1 exon between exons #3 and #4, with a transcription start site (TSS) at the 5′ end (Supplementary Fig. 2a)^27,28. We annotated this region as 5′UTR of a putative ESR1-LBD transcript variant (exon E3a). In this region, a higher number of read counts was found in ERα(−) cells, compared to ERα(+) (Fig. ​(Fig.2a).2a). Differential exon usage was determined for ESR1-FL (encoding ERα-FL) and a predicted ESR1-LBD (encoding ERα-LBD) transcript by comparing ERα-FL(+) and ERα(−) groups, revealing a significantly lower usage of exons E1, E2, E3 and E8-3UTR, and higher usage of exons E3a, E5, E6, and E7 for the ERα(−) group (Fig. ​(Fig.2b2b).

Open in a separate window

Fig. 2

ERα-LBD is encoded by an ESR1 transcript variant.

a Heat map showing ESR1 exon expression in BC cell lines, calculated by using DEXSeq-count tool (log2, mean-centered). Exons encoding ERα-FL protein are indicated in black (CDS). BC cell lines were grouped as ERα-FL(+), including T-47D, MCF-7, PDX-ERα(+) (in purple), and ERα-FL(−), including MDA-MB-453, MDA-MB-231, PDX-ERα(−) (in teal). b Differential ESR1 exon expression between ERα-FL(+) and ERα-FL(−) groups was calculated using DEXSeq and plotted using plotDEXSeq tools. Exons associated with two different ESR1 transcript variants were taken under consideration for the analysis: ESR1-FL (main variant encoding ERα-FL) and ESR1-LBD (predicted, encoding ERα-LBD). Expression values for ERα-FL(+) and ERα-FL(−) groups are plotted in purple and in teal, respectively. BH adjusted P-values (FDR test): *P<0.05, **P<0.01 (n=6/group). c Heat map showing ESR1 transcript (ESR1-FL and ESR1-LBD) expression among BC cell lines, grouped as ERα-FL(+) (purple) and ERα-FL(−) (teal). Transcript expression was calculated by using Kallisto tool and quantified as counts/transcript length. d ESR1-LBD transcript expression plotted as log2 fold change, relative to ESR1-FL. ESR1-FL, empty circles; ESR1-LBD, solid circles. Mean±s.e.m. is shown (n=6); **P<0.01, ratio paired t test, two-sided. e ESR1-FL transcript expression was calculated in BC samples by using Kallisto and plotted as Transcripts Per Kilobase Million (TPM). ERα(+) BC primary tumors (n=42); uninvolved breast tissue adjacent to ERα(+) primary tumors (n=30); TNBC primary tumors (n=42); uninvolved breast tissue adjacent to TNBC primary tumors (n=21); metaplastic BC primary tumors (MpBC, n=17). f Heat map showing ESR1 exon expression in BC patient groups. Exon expression was calculated by using DEXSeq-count tool [# of read counts/exon length] and scaled by column. g Dispersion plot for ESR1 transcript variants expression (ESR1-FL vs. ESR1-LBD), calculated as TPM. Dashed lines corresponding to ESR1-LBD expression median and Q1-Q3 quartiles are shown. Color code as in e. h Box and whiskers plot representing the distribution of ESR1-LBD expression values (TPM) from BC samples. Median, Q1/Q3 and Min/Max values are shown. Color code as in e. Percentage of samples from each group is also shown, distributed into quartiles. Source data are provided as Supplementary Data 2 file.

Exon data were supported by the analysis of ESR1 transcripts expression, demonstrating higher levels of ESR1-LBD transcript in ERα-FL(−) cells, compared to ESR1-FL (Fig. 2c, d and Supplementary Data 1). Also, qPCR (exon ‘walking’) and RT-PCR assays confirmed the RNA-seq results. ERα-FL(+) and ERα-FL(−) cells express an ESR1 mRNA ranging from exon 1 to 8 and from exon 4 to exon 8, respectively (Supplementary Fig. 2b, c). These results are in accordance with the western blot data in Fig. ​Fig.1b,1b, showing that ERα-FL(−) cells express a truncated ERα protein (LBD) and not full-length ERα. As definitive proof, the stable cloning of the putative ERα-LBD CDS into BC cells (exon 4 to exon 8), led to ERα-LBD overexpression (oe) whereas targeted genomic deletions within exon 4 (by CRISPR technology) induced ERα-LBD knockdown (kd) (Supplementary Fig. 2d). Alignment studies performed on public databases (UCSC Genome Browser) showed that the genomic region surrounding exon E3a is characterized by high-density transcription binding sites and candidate cis-regulatory elements (Supplementary Fig. 2e). Thus, we collected preliminary data based on luciferase-reporter assays to support the hypothesis of a putative alternative ESR1 promoter within exon E3a genomic region. Our data demonstrated the presence of transcriptional activity within the ESR1-LBD 5′UTR region (E3a) in both ERα(+) and ERα(−) cell lines (Supplementary Fig. 2f).

RNA sequencing and qPCR data showed reduced expression of the ESR1 3′UTR sequence in ERα-FL(−) cells as compared to ERα-FL(+) ones (Fig. ​(Fig.2a,2a, b and Supplementary Fig. 2b). It has been observed that shortening of 3′UTRs by alternative cleavage and polyadenylation is associated with increased RNA stability, and that this phenomenon can promote oncogene mediated transformation by enhanced protein production²⁹. Thus, we hypothesized that the partial/total loss of the 3′UTR sequence in ERα-FL(−) cells may enhance ESR1 mRNA stability. To test this, we treated BC cells with actinomycin-D (inhibitor of RNA synthesis) and indeed the LBD portion of ESR1 transcript was significantly more stable than the FL-specific in MCF-7 FulvRes and TNBC cells (Supplementary Fig. 2g).

We note that ERα-LBD is not the only ERα transcript variant that has been observed in breast cancer. An ERα-36 variant, encoded by ESR1 exons #2–6 and an additional exon downstream of the ESR1 gene, encoding a protein expressing the DNA-binding domain, a truncated LBD and a partial dimerization domain has been described in a variety of cancer models including tamoxifen-resistant breast cancers^30,31. Importantly, ERα-LBD is distinct from ERα-36 as specific primers and an antibody for the ERα-36 isoform failed to detect ERα-LBD (Supplementary Fig. 2h, i).

In addition, we examined ESR1-FL /-LBD expression by analyzing RNA-seq samples from human BC specimens. ESR1-FL overall expression was assessed on the following collection of samples: ERα(+) BC primary tumors (n=42); uninvolved breast tissue adjacent to ERα(+) primary tumors (n=30); TNBC primary tumors (n=42); uninvolved breast tissue adjacent to TNBC primary tumors (n=21); metaplastic BC primary tumors (MpBC, n=17)^32,33. As expected, ESR1-FL transcript was found to be markedly reduced in TNBC and MpBC samples (Fig. ​(Fig.2e).2e). Notably, ERα expression is known to be present in normal breast epithelium³⁴. For each cancer sample, ESR1 exon expression was also calculated and, similarly to what we observed with BC cell lines (Fig. ​(Fig.2a),2a), TNBC and MpBC samples showed higher amounts of 5′UTR exon 3a and CDS exons #4–8, compared to #1–3 (Fig. ​(Fig.2f).2f). We then examined the levels of the ESR1-LBD transcript in BC groups, relative to ESR1-FL (Fig. ​(Fig.2g).2g). 76% of ERα(+), 50% of TNBC, and 81% of MpBC were characterized by high ESR1-LBD levels (Q3+Q4 quartiles, above the median calculated on all samples). Interestingly, only 17% of ERα(+)^ADJ and none of TNBC^ADJ samples were found to have high ESR1-LBD expression (Fig. ​(Fig.2h).2h). All expression data are summarized in Supplementary Data 2.

ERα-LBD localizes to the mitochondria and cytoplasm of BC cells

ERα-LBD lacks the N-terminal domain of the full-length receptor and we thus hypothesized it may have a distinctive cellular location. Using ERα immunohistochemistry (IHC), full-length expressing MCF-7 cells show robust nuclear staining that is markedly reduced 24h after fulvestrant treatment. By contrast, ERα-LBD expressing models MCF-7 FulvRes, MDA-MB-453/-231, and PDX-ERα(−) display punctate cytoplasmic foci (Fig. ​(Fig.3a).3a). Based on this particular pattern, we hypothesized that ERα-LBD might be localized to the mitochondria and performed confocal analyses examining cells for ERα protein (green), OXPHOS (mitochondrial marker, red) and DAPI (nuclear marker, blue) (Fig. ​(Fig.3b).3b). We examined images from both separate and merged channels, and by 3-D image processing to evaluate the co-localization volume between ERα, nuclei or mitochondria. The overall IF staining confirmed data from the IHC: MCF-7 cells have a strong ERα nuclear localization which is decreased after fulvestrant treatment; MCF-7 FulvRes chronically treated with fulvestrant have both nuclear and mitochondrial ERα staining. In ERα full-length negative cells, ERα-LBD was largely absent from the nucleus and found in the cytoplasm with a substantial fraction co-localized with mitochondria. Mitochondrial ERα-LBD was also observed by western blot analysis of BC cell fractions confirming the association of ERα-LBD with mitochondria (associated to outer membrane or residing inside mitochondria) (Supplementary Fig. 3a, b).

Open in a separate window

Fig. 3

ERα-LBD localizes in the cytoplasm and mitochondria of breast cancer cells.

a Representative images of ERα staining by immunohistochemistry (IHC) in six different BC cell lines. All histological sections were counterstained with hematoxylin. Scale bar representing 50μm is shown. Magnification: ×40. b Representative confocal images of BC cells stained for ERα (green), mitochondria/OXPHOS (red) and nuclei/DAPI (blue). For each cell line, merged images and colocalization images are also shown. Color code for colocalization: cyan (ERα-Nuclei, N); yellow (ERα-Mitochondria, M). A scale bar for each image representing 40μm is shown. Magnification: ×63.

ERα-LBD expression and localization were further supported by IHC analysis of a human breast cancer tissue microarray (TMA) which included primarily TNBC. To perform this work, we utilized two different antibodies: one targeting the N-terminus of ERα protein which is commonly used in the clinical setting and one targeting the C-terminus, which would detect both ERα-FL and ERα-LBD. We observed that cytosolic staining was much more apparent using the C-terminal antibody (43.7% of total cores; 36.8% of ERα nuclear null/low cores) as compared to the N-terminal antibody (9.4% of total cores; 2.1% of ERα nuclear null/low cores). Notably, levels of cytoplasmic staining were weak compared to nuclear staining of ERα(+) cores and heterogeneous in distribution (Supplementary Fig. 3c–e and Supplementary Data 3). This expression pattern is consistent with our observations by western blot and RNA analyses in ERα(−) models (Figs. ​(Figs.11 and ​and22).

Given the relative absence of ERα-LBD nuclear localization, we examined its role in canonical ERα signaling, and we observed none of the transcriptional activation functions of full-length ERα. For instance, overexpression of ERα-LBD had no effect on Estrogen Response Element (ERE) luciferase expression or ERα target gene transcription in both ERα-FL(+) and TNBC cells (Supplementary Fig. 3f–k)³⁵.

To further establish the functional significance of cytoplasmic and mitochondrial ERα-LBD, we performed co-IP and mass spectrometry experiments aimed at identifying ERα-LBD protein–protein interactions (PPIs). We designed the experiment comparing MCF-7 cells overexpressing ERα-LBD with parental (NC), in the absence or presence of fulvestrant and we identified a total of 52 peptides associated with ERα-LBD (Fig. ​(Fig.4a).4a). Pathway enrichment and neural networking analyses performed on this protein set identified processes involved in carbohydrate metabolism (glycolysis and gluconeogenesis), cell signaling (MYC, mTOR/mTORC1, PIK3C1/AKT, ERα), hypoxia and angiogenesis (HIF1α and VEGFR signaling) and mitochondrial metabolism (oxphos and respiration) (Fig. 4b, c; Supplementary Data 4 and Supplementary Fig. 4a). Comparable experiments in TNBC models overexpressing or knocking down ERα-LBD identified a total of 93 peptides involved in ERα-LBD PPIs, and 29 of them were shared with those found in MCF-7 cells (Fig. ​(Fig.4d).4d). Similarly to the MCF-7 model, enriched pathways included signaling, carbohydrate and mitochondrial metabolism (including also fatty acid β-oxidation) and angiogenesis (Fig. 4e, f; Supplementary Data 4 and Supplementary Fig. 4b).

Open in a separate window

Fig. 4

Predicted ERα-LBD network: protein–protein interactions (PPIs) and biological pathways.

a Heat map showing the enrichment level of proteins that preferentially bind ERα-LBD, based on ERα co-immunoprecipitation (Co-IP) experiments followed by LC/MS analysis. Samples: NC and NC+Fulv (negative controls); ERα-LBDoe, ERα-LBDoe +Fulv (overexpression). Fulv=1µM fulvestrant treatment, 24h. Different comparative criteria between samples were used to identify groups of ERα-LBD protein interactors, color-coded as follows: ERα (blue); proteins present only in ERα-LBD samples (green, cyan and yellow); proteins enriched more in ERα-LBD samples compared to NC, in the absence or presence of fulvestrant treatment (red and purple, respectively). Protein names labeled in dark red color are shared between MCF-7 and TNBC models (see below). b ‘Enrichr’ platform based on multiple libraries was used to discover cell pathways in which ERα-LBD PPIs are significantly involved (Adj-P (Q)<0.05, FDR test), then clustered and annotated by using Cytoscape software and MCL algorithm. c ERα-LBD PPIs contribution to clustered pathways. d Same as in a. Samples: MDA-MB-453/-231 ERα-LBDoe (overexpression); MDA-MB-453/-231 ERα-LBDkd (knockdown). ERα PPIs and protein groups, color-coded as follow: ERα (blue); proteins enriched in both TNBC cell lines overexpressing ERα-LBD, compared to ERα-LBD silenced cells (green and cyan); enriched in MDA-MB-453 ERα-LBDoe cells only (yellow); enriched in MDA-MB-231 ERα-LBDoe cells only (red). e, f Same as in b, c. Source data are provided as Supplementary Data 4 file.

ERα-LBD regulates cell metabolism

The localization and protein–protein interaction analyses point to a potential metabolic function for ERα-LBD. We speculated ERα-LBD may play a role in regulating glycolysis and mitochondrial respiration, and so assessed the effect of ERα-LBD overexpression or knockdown on key metabolic parameters. In ERα-FL expressing MCF-7, control (NC) and ERα-LBDoe cells showed similar OCR (oxygen consumption rate, index for respiration) and ECAR (extracellular acidification rate, index for glycolysis) under basal conditions. However, following fulvestrant-mediated depletion of ERα-FL and stabilization of ERα-LBD, we observed significantly higher respiratory parameters (ATP production, basal and maximal respiration, and spare respiratory capacity) and higher glycolytic parameters (glycolytic capacity and glycolytic reserve) (Fig. ​(Fig.5a5a and Supplementary Fig. 5a). In the TNBC models featuring knockdown of ERα-LBD, ERα-LBDkd cells were characterized by reduced levels of respiratory parameters (ATP production, basal and maximal respiration, spare respiratory capacity) and glycolysis (glycolytic capacity and reserve; basal glycolysis only in MDA-MB-231), compared to controls (NC) (Fig. 5b, c and Supplementary Fig. 5b, c). In TNBC models, OCR and ECAR were also tested in the presence of fulvestrant treatment and, despite the overall reduction of respiration (MDA-MB-453 and −231) and glycolysis (MDA-MD-231), metabolic parameters were significantly reduced in ERα-LBDkd cells (Fig. 5b, c and Supplementary Fig. 5b, c).

Open in a separate window

Fig. 5

Effect of ERα-LBD ‘gain/loss-of-function’ on BC cell metabolism.

a Mitochondrial respiration and glycolysis levels in MCF-7 cell clones. Fulvestrant 1µM (24h pre-treatment) or vehicle (DMSO) was added to cells. ERα-LBD overexpressing (ERα-LBDoe; in dark/light red) cells were compared to controls (NC; in dark/light blue). Analyses were carried out using XF Cell Mito Stress Kit (Agilent). Respiration was evaluated as oxygen consumption rate (OCR, pmol/min), normalized on cell number. Glycolysis was evaluated as extracellular acidification rate (ECAR, mpH/min), normalized on cell number. The calculation of metabolic parameters was based on ΔOCR and ΔECAR values, following manufacturer’s guidelines. b, c Mitochondrial respiration and glycolysis levels in MDA-MB-453 and MDA-MB-231 cells. Treatments as in (a). Cells with ERα-LBD knockdown (kd; in dark/light red) were compared to controls (NC; in dark/light blue). Analyses were carried out as described above. FCCP p-trifluoromethoxy-phenylhydrazone, Rot rotenone, AtmA antimycin A. Data in the figures are presented as mean±s.e.m. (n=2 independent experiments). *P<0.05, **P<0.01; unpaired t test, one-sided. Source data are provided as a Source Data file.

ERα-LBD promotes growth and endocrine resistance, in vitro and in vivo

Given the impact of ERα-LBD on breast cancer cell metabolism, we investigated the effects of ERα-LBD on cell proliferation and response to endocrine therapy. In vitro, MCF-7 cells overexpressing ERα-LBD, compared to control cells (NC), showed no significant differences in growth. However, upon treatment with fulvestrant, NC cells were potently growth-inhibited (~80% inhibition, P>0.0001, vs. untreated, two-way ANOVA) while ERα-LBDoe cells were only partially growth-inhibited (~40% inhibition, P=0.003, vs. untreated, two-way ANOVA), establishing a role for ERα-LBD in mediating endocrine resistance (Fig. ​(Fig.6a)6a) that was confirmed also by treating MCF-7 NC and ERα-LBDoe cells with tamoxifen (Supplementary Fig. 6a). Moreover, growth of MCF-7 FulvRes cells was significantly impaired after ERα-LDB knockdown (especially under fulvestrant treatment; P>0.0001, two-way ANOVA) whereas no growth inhibition was observed in parental MCF7 FulvRes cells treated with fulvestrant, compared to untreated cells (Fig. ​(Fig.6b).6b). Beyond the effects in ERα-FL(+) cells, we also assessed the effects of ERα-LBD on the growth of TNBC cells. In both MDA-MB-453 and MDA-MB-231 we observed ERα-LBD knockdown to reduce cell proliferation (~50%) (Fig. 6c, d).

Open in a separate window

Fig. 6

ERα-LBD promotes in vitro and in vivo growth and fulvestrant resistance.

a–d In vitro proliferation of BC cell clones, using resazurin reagent and expressed as fluorescence intensity (absorbance at 590nm). ERα-LBD overexpression (a) or knockdown (b–d) was compared to controls (NC). In a and b, cell proliferation was also tested in the presence or absence of fulvestrant 1µM treatment (n=3 independent experiments). All P values describing statistical differences between samples are shown. e–h Xenografts of BC cells implanted into the mammary fat pads of NOD scid gamma (NSG) mice. Plots in e and f show mean tumor volumes (mm³) as a function of time (days). MCF-7 model: when tumors reached 100 mm³, mice were randomized (n=6/group) to weekly treatments with 200mg/kg fulvestrant injected intramuscularly, and compared to control groups (n=6/group). MDA-MB-231 model: n=5/group. f–h Images and size of tumors, isolated from animals after reaching endpoint. In a and e, NC is in blue and ERα-LBD overexpression in red. In b–d and g, NC is in blue and ERα-LBD knockdown in red. All data in the figure are presented as mean±s.e.m., ns not significant, *P<0.05, **P<0.01, two-way ANOVA. Source data are provided as a Source Data file.

To further verify the biologic significance of these effects in vivo, these models were grown orthotopically as xenograft models. As observed in vitro, MCF-7 ERα-LBDoe tumors continued to grow upon fulvestrant treatment while tumors from NC cells were expectedly growth inhibited with fulvestrant (Fig. 6e, f). Moreover, in vivo tumor growth for MDA-MB-231 ERα-LBDkd mice group was markedly reduced, when compared to NC group (Fig. 6g, h). MCF-7 ERα-LBDoe cells showed a growth advantage when cultured in low-attachment (3-D growth) either in the presence or absence of fulvestrant treatment (Supplementary Fig. 6b), whereas ERα-LBD knockdown led to reduced 3-D growth in both MDA-MB-453 and −231 cells (Supplementary Fig. 6c). Similarly, the migratory capacity of BC models was enhanced in ER LBD overexpression models and impaired in knockdown models (Supplementary Fig. 6d, e).

Cell proliferation and manipulation

Cells (1000/well) were seeded into 96-well plates and, accordingly to experimental design, vehicle (DMSO 0.01%), fulvestrant 1µM or tamoxifen 1µM was added to the medium. Media and drug were replaced every 3 days. At different time points (day 1, 3, 5, 7, and 9), 20µl of Resazurin (R&D Systems, USA) was added to 180µl of medium in each well and incubated for 4h. Fluorescence (absorbance 590nm) was measured for each well using a SpectraMax M5 microplate reader (Molecular Devices, USA) and correlated to cell growth. To generate fulvestrant-resistant MCF-7 (MCF-7 FulvRes) and T-47D (T-47D FulvRes) cells, cells were cultured in the presence of increasing concentrations of fulvestrant (0.1–1µM). Cells were deemed resistant when able to actively proliferate in 1µM fulvestrant. MCF-7 Y537S CRISPR knock-in cells were generated as described elsewhere²⁴. PDX models of ERα(+)/(−) metastatic BC were generated and maintained as described elsewhere¹⁷. BC cell clones overexpressing ERα-LBD (ERα-LBDoe) were generated using stable retroviral transduction. Briefly, ERα-LBD coding sequence was cloned into pBABE-Puro vector (Addgene, USA; Supplementary Data 7). To generate retroviral particles, three plasmids were co-transfected into 293T packaging cells using Lipofectamine 2000 (Invitrogen, USA): pBABE-Puro-ERαLBDoe, pCMV-VSV-G and pUMVC (Addgene, USA). Viral supernatants were collected 48h and 72h later, centrifuged to remove cell debris, filtered through 0.45-µm filters (Millipore, USA) then used with polybrene 8µg/ml polybrene (Santa Cruz Biotechnology, USA) to transduce BC cell lines. After 2 days, stable clones were selected with puromycin 2µg/ml. BC cell clones with ERα-LBD knockdown (ERα-LBDkd) were generated using stable lentiviral transduction and CRISPR/CAS9 technology for genome editing. Briefly, three different guide RNA sequences targeting ESR1 gene (exon #6) were cloned in the BsmBI site of lentiCRISPRv2 vector (Addgene, USA; Supplementary Data 7) to create a pool of pCR-ERαLBDkd lentivectors. To generate lentiviral particles, the following plasmids were co-transfected into 293T packaging cells: pCR-ERαLBDkd (pool), pCMV-VSV-G and psPAX2 (Addgene, USA). Infection and selection procedures: same as retroviral approach (see above).

Western blot

Cells were homogenized in RIPA lysis buffer for protein extraction, supplemented with protease inhibitors (Thermo Scientific, USA). Denatured proteins were separated in 12% SDS–PAGE and then transferred onto nitrocellulose papers (Pall, USA). After blotting, nitrocellulose papers were incubated with specific antibodies. Primary antibodies: anti-ERα antibody D8H8, 1:2000 (Cell Signaling Technology, USA, #8644); anti-ERα F-10, 1:1000 (Santa Cruz Biotechnology, USA, #sc-8002), anti-ERα 1D5, 1:1000 (Invitrogen, USA, #MA5-13191), anti-ERα-36, 1:200 (Alpha Diagnostic International, USA, #ERA361-A); anti-β-actin 13E5, 1:2000 (Cell Signaling Technology, USA, #8457); anti-HDAC2, 1:1000 (Cell Signaling Technology, USA, #2540); anti-VDAC1 N-18 (Santa Cruz Biotechnology, USA, #sc-8828). Secondary antibodies (HRP conjugated): anti-rabbit, 1:5000 (Cell Signaling Technology, USA, #7074); anti-mouse, 1:2000 (Cytiva, USA, #RPN4201). Immunolabelling was visualized using ECL procedure (PerkinElmer, USA). Uncropped and unprocessed scans of the most important blots are supplied in the Source Data file. All blots derive from the same experiment and they were processed in parallel.

Gel extraction, mass spectrometry, and protein structure analysis

Protein lysates from MCF-7 FR and MDA-MB-231 (both treated with fulvestrant 1µM for 24h) were obtained using IP buffer (NaCl 150mM, EDTA 0.5mM, NP-40 0.5% (v/v), Tris-HCl 10mMpH 7.4, PMSF 1mM) and immunoprecipitated using anti-ERα antibody (1:100, Cell Signaling Technology, USA, #8644). IP samples were run on two electrophoresis gels, one used for staining with SimplyBlue Safe Stain (Invitrogen, USA) and band extraction, the other for WB sample check. In-gel digestion was performed using the method by Shevchenko et al.⁵². Briefly, gel bands were excised, washed with acetonitrile and 100mM ammonium bicarbonate solution (1:1) for 30min, dehydrated with 100% acetonitrile for 10min and dried in a speed-vac for 10min without heat. Gel slices were reduced with 5mM DTT for 30min at 56°C in a thermo-mixer with gentle mixing, removed, allowed to cool to room temperature then alkylated with 11mM IAA for 30min in the dark. Gel slices were washed with 100mM ammonium bicarbonate and 100% acetonitrile for 10min each. Excess acetonitrile was removed and the slices dried in a speed-vac for 10min without heating. Gel slices were then rehydrated in a solution of 25ng/ml trypsin in 50mM ammonium bicarbonate on ice for 30min. Digestions were performed overnight at 37°C in a thermo-mixer with gentle mixing. Digested peptides were collected and further extracted from gel slices in extraction buffer (5% formic acid and 50% acetonitrile, 1:2 vol/vol) at high-speed mixing. Extractions were combined and dried down in a vacuum centrifuge. Peptides were desalted with C18 resin-packed stage-tips, lyophilized to dryness, then re-constituted in 3% acetonitrile/0.1% formic acid for LC-MS/MS analysis. LC-MS/MS Analysis LC-MS/MS was performed using a Waters NanoAcquity LC system (with a 100mm inner diameter×10cm length C18 column (1.7mm BEH130; Waters, USA) configured with a 180mm×2cm trap column coupled to a Thermo Q-Exactive Plus orbitrap mass spectrometer (Thermo Scientific, USA). Trapping was performed at 15ml/min 0.1% formic acid (Buffer A) for 1min. The LC gradient was 0.5 to 50% B (100% acetonitrile; 0.1% formic acid) over 90min at 300nl/min. MS data were collected in data-dependent acquisition (DDA) mode utilizing a top ten precursor ion selection for HCD fragmentation. Full MS scans were performed with the following parameters: Resolution: 70,000; AGC target: 1e6; Maximum IT: 50ms; Scan Range: 400 to 1600 m/z. DDA parameters were as follows: Resolution: 17,500; AGC target 5e4; Maximum IT: 50ms; Isolation window: 1.5m/z; NCE: 27; Minimum AGC target: 2e3; Intensity Threshold: 4e4; Dynamic Exclusion: 15s; Charge exclusion: unassigned, 1, 6–8, >8. All MS/MS samples were analyzed using Mascot (Matrix Science, UK). Mascot was set up to search the SwissProt_sprot_20170705_20180523 database (selected for Homo sapiens, unknown version, 20215 entries) assuming the digestion enzyme trypsin. Mascot was searched with a fragment ion mass tolerance of 0.080Da and a parent ion tolerance of 10.0 PPM. Carbamidomethyl of cysteine was specified in Mascot as a fixed modification. Deamidated of asparagine and glutamine, oxidation of methionine, acetyl of the N-terminus and phosphorylation of serine, threonine and tyrosine were specified in Mascot as variable modifications. Mass spectrometry data were further processed using Scaffold software (Proteome Software Inc., USA). Peptide identifications were accepted if they could be established at >90.0% probability to achieve an FDR <1.0% by the Peptide Prophet algorithm with Scaffold delta-mass correction. Protein identifications were accepted if they could be established at >89.0% probability to achieve an FDR <1.0% and contained at least 1 identified peptide. Protein probabilities were assigned by the Protein Prophet algorithm. Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. Proteins sharing significant peptide evidence were grouped into clusters. Protein sequence and features were analyzed using Sequence Analysis software (Informagen, USA). Three-dimensional protein models were generated by Phyre² web portal (www.sbg.bio.ic.ac.uk)⁵³.

ESR1-LBD transcript variant prediction

Comprehensive ESR1 transcript variants annotation was obtained from Ensembl Genome Browser (www.ensembl.org). We used ZEMBU Genome Browser (https://fantom.gsc.riken.jp/zenbu/) for variant prediction study, showing data collected on ESR1 gene, based on FANTOM5 and FANTOM CAT analyses^27,28. For ESR1 gene and transcripts details please refer to this link: https://fantom.gsc.riken.jp/zenbu/gLyphs/#config=eljPd0OTPhIoifL81WStGC;loc=hg19::chr6:152118533..152432431+.

RNA capture-sequencing and ESR1 transcripts analysis

Total RNA was extracted as described elsewhere⁵⁴ and samples (100ng) were input in the RNA library construction using the KAPA RNA Hyper library prep kit (Roche, Switzerland). Customized adapters with unique molecular indexes (UMI) (Integrated DNA Technologies, USA) and Sample-specific dual-indexes primers (Integrated DNA Technologies, US) were added to each library. Each RNA library was pooled for hybridization capture with customized ESR1 Panel (Integrated DNA Technologies, USA) using a capture protocol modified from NimbleGen SeqCap Target Enrichment system (Roche, Switzerland). Libraries were then sequenced on an Illumina MiniSeq with paired-end reads (×150 cycles, 1.4 millions reads/sample). Raw sequencing data output was processed for expression analysis using STAR Aligner⁵⁵, DEXSeq-count⁵⁶, Cluster 3.0 and Java TreeView^57,58, DEXSeq and plotDEXSeq⁵⁶, Kallisto⁵⁹ (Supplementary Fig. 8a).

RNA extraction and real-time qPCR

Total RNA was extracted from cells using Trizol reagent (Invitrogen, USA), according to the manufacturer’s instructions. Extracted RNA samples were quantified and then treated with DNase I to remove any genomic DNA contamination, using Ambion DNase I kit (Invitrogen, USA). Reverse transcription was carried out using iScript™ Select cDNA Synthesis Kit (Bio-Rad, USA). cDNA levels were analyzed by real-time PCR using TaqMan Universal PCR Master Mix or SYBR Select Master Mix reagents and ViiA 7 Real-Time PCR system, according to the manufacturer’s instructions (Applied Biosystems, USA). Melting curve data were collected to check PCR specificity. Samples were run in triplicate and mRNA levels were normalized against those of RPLP0 or β-actin. Relative expressions were calculated using the formula 2^−2ΔCt values. Expression data were obtained by using QuantStudio™ Real-Time PCR Software (Applied Biosystem, USA). RT-PCR was carried out by using OneTaq Hot Start 2X Master Mix (New England Biolabs, USA). PCR samples were separated on a 1.5% agarose gel and results visualized using Gel Doc XR+Imaging System (Bio-Rad, USA). For RNA Decay Assay, BC cells were seeded in 6-well plates and treated with actinomycin-D 5µg/ml (Sigma-Aldrich, USA). RNA samples were collected at different time points (0, 4, 8, 24, 48h). All PCR primers are summarized in Supplementary Data 7.

Transfection and luciferase assay

Cells were seeded in 24-well plate (2×10⁵ cells/well) 1 day before transfection and treatments were added accordingly to the experimental design (not treated, vehicle/DMSO 0.01% or fulvestrant 1µM). For the putative promoter experiment, three different human genomic DNA regions representing putative ESR1-LBD promoter (pESR1-LBD-1/2/3) were cloned into pGL3.basic plasmid (Promega, USA; Supplementary Data 7). BC cells were co-transfected with pGL3.basic (NC) or p-ESR1-LBD reporter plasmids (750ng) and pRL-TK Renilla Luciferase plasmid (75ng) (Addgene, USA) by using Lipofectamine 2000 (Invitrogen, USA) and following reagent protocol. For ERα-driven transcriptional activity experiment, BC cells were co-transfected with 3x-ERE-TATA-Luciferase reporter plasmid (750ng) and pRL-TK Renilla Luciferase plasmid (75ng) (Addgene, USA) by using the same procedure described above. Cells were harvested 24h after transfection and cell lysates were used for Dual-Luciferase® Reporter Assay System analysis, according to the manufacturer’s instructions (Promega, USA). Luciferase bioluminescence measurements were performed with the Veritas™ Microplate Luminometer (Promega, USA). For each sample, Firefly luciferase activity was normalized against Renilla luciferase activity.

Breast cancer patients data

RNA-seq FASTQ files of 135 breast cancer were obtained from the European Nucleotide Archive, ENA (Study Accession: PRJNA251383; Supplementary Data 2)³². RNA-seq FASTQ files of 17 metaplastic breast cancer were kindly provided by Dr. Jorge S Reis-Filho (Supplementary Data 2)³³. Raw data were processed by using DEXSeq-count⁵³, Kallisto⁵⁴ and Morpheus (https://software.broadinstitute.org/morpheus).

Immunohistochemistry and confocal microscopy

ERα immunostaining (IHC) was performed on Benchmark Ultra using the ultraView DAB Detection kit (Ventana, USA). Antigen retrieval was performed onboard with UltraCC1 buffer (pH 8.2–8.5) at 95°C for 52min. Primary antibody (anti-ERα antibody D8H8, Cell Signaling Technology, USA, #8644) 1:100 for 28min at 37°C. Secondary antibody 1:100 for 1h. Images were obtained using a Zeiss Axiovert Widefield Microscope and Zeiss ZEN software (Carl Zeiss, Germany). Immunofluorescence was performed using Leica Bond RX stainer and the Bond Polymer Refine Detection kit (Leica, Germany). Antibodies and detection: OXPHOS (2µg/ml, Invitrogen, USA, #A-21347); ERα D8H8 (1µg/ml, Cell Signaling Technology, USA, #8644); Alexa Fluor Tyramide signal amplification reagents (Life Technologies, USA); 4′, 6-diamidino-2-phenylindole (DAPI, Sigma-Aldrich, USA). Slides were mounted in Mowiol 4–88 (Calbiochem, USA). Confocal imaging was performed on a Leica SP8 inverted microscope (Leica, Germany). Image processing and analysis (2-D and 3-D) was performed using Imaris software (Bitplane, CH).

Cell fractionation for western blot

Whole-cell lysate. Cells were pelleted and homogenized in lysis buffer (50mM Tris-HCl, pH 7.5, 2mM EDTA, 100mM NaCl, 1% Triton X-100, protease inhibitors). Cell lysates were incubated 1h on ice and centrifuged at 12,000g at 4°C for 20min to collect supernatants (sample W). Nuclear and cytosolic fractions. Pelleted cells were resuspended in nuclear lysis buffer (10mM HEPES; pH 7.5, 10mM KCl, 0.1mM EDTA, 0.5% Nonidet‐40, 0.5mM PMSF, protease inhibitors). Cells were allowed to swell on ice for 15‐20min with intermittent mixing to disrupt cell membranes and then centrifuged at 12,000×g at 4°C for 10min. Supernatant were transferred into a new tube and used as cytosolic fraction (sample C). The pelleted nuclei were washed with nuclear lysis buffer and resuspended in nuclear extraction buffer (20mM HEPES pH 7.5, 400mM NaCl, 1mM EDTA, 1mM DTT, 1mM PMSF, protease inhibitors) and incubated in ice for 30–60min. Nuclear extract (supernatant) was collected by centrifugation at 16,000×g for 15min at 4°C and transferred into a new tube (sample N). Mitochondrial fraction. Cells were collected by centrifugation at ~500×g for 10min then resuspended in 10 packed cell volumes of NKM buffer, x2 (Tris-HCl 1MpH 7.4, NaCl 2M, KCl 1M, MgCl₂ 0.5M). Cells were pelleted and resuspended in 6 packed cell volumes of homogenization buffer (Tris-HCl 1MpH 6.7, KCl 1M, MgCl₂ 0.5M PMSF 200mM). Cell suspension was transferred to a glass homogenizer, incubate for 10min on ice, then homogenized using a tight pestle. The level of cell breakage was checked under the microscope. Homogenate was poured into a conical centrifuge tube containing 2 packed cell volume of 2M sucrose solution and mixed gently. Unbroken cells, nuclei and large debris were pelleted at 1200g for 5min. The supernatant was transferred to another tube containing 1.5 packed cell volume of 2M sucrose and mixed gently. Mitochondrial fraction was collected by centrifuging at 12,000×g for 15min and resuspended in isotonic mitochondrial buffer (10mM Hepes buffer pH 8.0, 250mM sucrose, 0.5mM EGTA). The suspension was split 1:2. One half, centrifuged at 12,000×g for 15min and resuspended in lysis buffer. After incubation of 30–60min on ice, mitochondrial fraction (supernatant) was collected by centrifugation at 12,000×g for 15min (sample M). The other half was incubated with proteinase K (50µg/ml, Thermo Fisher) for 30min at 25°C. PMSF was added to a final concentration of 2mM to terminate the proteinase K activity, followed by 10min of incubation. PK-treated mitochondria were collected by centrifugation at 12,000×g for 10min (sample M*).

TMA immunohistochemistry

ERα immunostaining (IHC) was performed on Benchmark Ultra using the ultraView DAB Detection kit (Ventana, USA). Antigen retrieval was performed onboard with UltraCC1 buffer (pH 8.2–8.5) at 95°C for 52min. The primary anti-ERα antibody (D8H8, Cell Signaling Technology, USA, #8644) was incubated 1:100 for 28min at 37°C. Secondary antibody 1:100 for 1h. Hematoxylin/eosin counterstain was also carried out. Slides were scanned by Leica Aperio AT2 (Leica Biosystems, USA) whole slide scanners. Images from ERα staining by using 6F11 antibody were kindly obtained by US BioMax Inc., USA. For each sample/core, IHC grading was assessed considering the intensity of the staining and the number of stained cells.

ERα co-immunoprecipitation and LC/MS

Protein were extracted using Co-IP buffer (NaCl 150mM, EDTA 0.5mM, NP-40 0.5% (v/v), Tris-HCl 10mMpH 7.4, PMSF 1mM). Immunoprecipitation (IP): anti-ERα antibody D8H8 1:75 (Cell Signaling Technology, USA, #8644) and Dynabeads™ Protein A (Invitrogen, USA). IP samples were washed four times with 50mM ammonium bicarbonate buffer and collected by centrifugation. Samples were then digested overnight with 2µg trypsin in 80µl of 50mM ammonium bicarbonate at 37°C. Desalted and dried peptides were reconstituted in 10µl 0.1% (vol/vol) formic acid and analyzed (4µl) by microcapillary liquid chromatography with tandem mass spectrometry. Pathway enrichment analysis was performed by using Enrichr platform (https://maayanlab.cloud/Enrichr/)⁶⁰. Enriched pathways network, clustering, and protein–protein interaction (PPI) network analyses were carried out using Cytoscape, EnrichmentMap, AutoAnnotate and stringApp softwares⁶¹.

Metabolic assay

Mitochondrial respiration and glycolysis were assessed using Seahorse extracellular flux analyzer (XFe96) and Seahorse XF Cell Mito Stress Test Kit (Agilent Technologies, USA). Cells were seeded in 96-well plates (2×10⁴ cells/well), with or without Fulvestrant 1µM 24h pre-treatment. Raw data output was collected and analyzed using Wave Software (Agilent Technologies). Procedures and OCR/ECAR data interpretation were carried out accordingly to the manufacturer’s guidance.

Animal models

NOD scid gamma (NSG) female mice at age of 6–8 weeks were obtained from Jackson Laboratory (USA) and maintained in pressurized ventilated caging. To sustain tumor growth in MCF-7 models, 17β-estradiol pellets (0.18mg) were implanted subcutaneously 3 days before BC cells injection. For both MCF-7 and MDA-MB-231 models, cancer cells were injected in the mammary fat pads (MFPs) of mice. For each experimental sample, cell suspensions were mixed with an equal volume of Matrigel (BD Biosciences, USA). For the MCF-7 model, injectable fulvestrant (Faslodex®, AstraZeneca, UK) was given intramuscularly in the tibialis posterior/popliteal muscles (200mg/kg injection, once a week) for 20 days. Control mice received isotype control (placebo) or PBS injection. Tumor volumes were measured with vernier calipers starting from 14 days after cell implantation. Animals were sacrificed as they reached an experimental endpoint. All procedures and experiments were completed in accordance with the Guidelines for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committees at MSKCC (MSKCC#12-10-016).

Microscopy, 3-D growth, and wound healing assay

MCF-7 cells (WT and FulvRes) were seeded into six-well plate (1×10⁵ cells/well) and treated accordingly to the experimental design. After controls reached 100% confluence, images of cells were captured with Zeiss Axiovert Microscope and processed with Zeiss ZEN software (Carl Zeiss, Germany). For 3-D growth, BC cell clones were suspended sparsely and plated in 24-well ultra-low attachment plates (Corning, USA), 3×10³ cells/well. Treatments (DMSO 0.01% or fulvestrant 1µM) were added to medium accordingly to experimental design. After 5 days, representative images of cells were acquired with a Zeiss Axiovert Microscope. Quantification of cell 3-D proliferation was based on the optical density (OD 600nm) of cell suspensions and determined using a SpectraMax M5 microplate reader (Molecular Devices, USA). For the wound healing assay, about 1×10⁵ cells were seeded into each well of 12-well plates and cultured until 100% confluence. The ‘scratch’ was created with a p200 pipet tip on the cell monolayer through the center of the well. The debris was removed by washing the well with 1ml of culture medium and then 1ml of medium (with 1% serum) was added into each well. Treatments (DMSO 0.01% or fulvestrant 1µM) were also added accordingly to the experimental design. The plate was incubated at 37°C, and images of the ‘scratches’ were captured at various time points with a Zeiss Axiovert Microscope. The width of scratch was measured with FIJI software.

RNA sequencing

Total RNA was extracted from cell lines as described elsewhere⁵⁴ and sequenced on an Illumina HiSeq instrument (2×150 paired-end, 100 million reads/sample) by GENEWIZ, LLC. (USA). Raw sequencing data output was processed for downstream analyses using the following bioinformatic tools: STAR (reads alignment)⁵⁵, HTSeq (gene expression)⁶², Cluster 3.0 and Java TreeView (hierarchical clustering and heat map)^57,58, GSEA (enrichment of gene sets)⁶³ (Supplementary Fig. 8b and Supplementary Data 6).

Flow cytometry (FACS)

For FACS/flow analyses, tumors were digested in sterile Epicult media (StemCell Technology, Canada), minced with sterile razor blades and incubated for 3h in the presence of collagenase/hyaluronidase (1,000 Units/sample). Cells were washed with sterile filtered PBS supplemented with 1% BSA (PBS-BSA 1%) and filtered through a 40mm nylon mesh (BD Biosciences, USA). Cells were then stained in a volume of 100µL (PBS-BSA 1%) with CD44-APC antibody 100ng/10⁶–10⁸ cells (IM7, eBiosciences, USA, # 17-0441-82) on ice for 30min and analyzed by flow cytometry at the MSKCC’s flow core with BD FACS Aria I instrument (BD Biosciences, USA). Samples were analyzed for cell population distribution and sorted for viability (DAPI^neg) and CD44 expression. For flow plot analyses, samples were run using FlowJo 7.5 software (Tree Star, USA).

This work was supported by grants from the National Institute of Health to MSKCC’s Cancer Center (P30CA008748) and to S.C. (R01CA204999), and a grant from the University of Bologna to E.S. (RFO 2015). J.B. acknowledges Charles and Marjorie Holloway Foundation, Sussman Family Fund, Lerner Foundation, and Beth C. Tortolani Foundation. S.C. also acknowledges support by the Breast Cancer Research Foundation. F.P. is partially funded by a National Institutes of Health/ National Cancer Institute K12CA184746 grant.