Estrogen receptor alpha drives mTORC1 inhibitor-induced feedback activation of PI3K/AKT in ER+ breast cancer

The mTORC1 inhibitor RAD001 (everolimus) is approved for treatment of recurrent/metastatic estrogen receptor (ER)-positive breast cancer in combination with the aromatase inhibitor (AI) exemestane. The benefits of A) continued anti-estrogen therapy for anti-estrogen-resistant disease in the context of mTORC1 inhibition, and B) adjuvant everolimus in combination with anti-estrogen therapy for early-stage disease are being tested clinically, but molecular rationale remains unclear. We hypothesized that mTORC1 inhibition activates the IGF-1R/InsR/IRS-1/2 axis in an ER-dependent manner to drive PI3K/AKT and promote cancer cell survival, implicating ER in survival signaling induced by mTORC1 inhibition. Anti-estrogen treatment synergized with RAD001 to inhibit ER+ breast cancer cell growth. Inhibition of ER, IGF-1R/InsR, or IRS-1/2 suppressed AKT activation induced by mTORC1 inhibition. RAD001 primed IGF-1R/InsR for activation, which was enhanced by ER signaling. Post-menopausal patients with early-stage ER+ breast cancer were treated presurgically +/− the AI letrozole. Viable tumor fragments from surgical specimens were treated with RAD001 and/or OSI-906 ex vivo; RAD001 increased AKT activation, which was abrogated by presurgical letrozole. Letrozole decreased IGF-1R and IRS-1/2 tumor levels. These data suggest that ER drives PI3K/AKT activation in response to mTORC1 inhibition, providing molecular rationale for therapeutic combinations of anti-estrogens and mTORC1 inhibitors in endocrine-sensitive disease.


INTRODUCTION
Approximately 70% of primary breast tumors express estrogen receptor α (ER) and/or progesterone receptor (PR). Patients with such hormone receptorpositive breast cancers are typically treated with adjuvant anti-estrogen therapies such as tamoxifen or aromatase inhibitors (AIs) for 5-10 yrs after surgical removal of the primary tumor [1]. While adjuvant anti-estrogen therapies have been used to successfully treat hormone-dependent breast cancer, approximately one-third of patients develop recurrent advanced/metastatic disease that is rarely cured
Mechanistic target of rapamycin (mTOR) complex I (mTORC1) is a critical signaling hub that regulates cellular and organismal homeostasis by balancing anabolic and catabolic processes with nutrient, energy, oxygen availability, and growth factor input signaling [4]. mTORC1 inhibitors continue to be investigated as therapeutics for many cancer subtypes. Hyperactivation of the phosphatidylinositol 3-kinase (PI3K)/AKT/mTORC1 pathway is associated with anti-estrogen resistance in ER+ breast cancer [5]. In patients with recurrent/metastatic ER+ breast cancer resistant to a non-steroidal AI (e.g., letrozole, anastrozole), treatment with the steroidal AI exemestane plus the mTORC1 inhibitor everolimus (RAD001; afinitor) significantly improved median progression-free survival (but not overall survival) compared to exemestane/ placebo (11 vs. 4.1 months) in the BOLERO-2 study [6][7][8]. The TAMRAD study similarly evaluated the use of tamoxifen +/-everolimus for the treatment of advanced, anti-estrogen-resistant ER+ breast cancer; patients in the combination arm experienced a prolonged time-toprogression compared to tamoxifen/placebo (8.6 vs. 4.5 months), and a 55% reduction in risk of death [9]. Such findings supported the U.S. FDA approval of everolimus for use in combination with exemestane for patients with ER+ breast cancer resistant to a non-steroidal AI. However, the benefit of continued anti-estrogen therapy (with exemestane) in the context of mTORC1 inhibition (with everolimus) remains unproven (being tested in the ongoing BOLERO-6 trial [10]); indeed, single-agent everolimus elicited clinical benefit in 6/22 (27%) patients with advanced ER+ breast cancer [11]. Positive findings with exemestane/everolimus in the setting of advanced disease prompted the ongoing clinical testing of adjuvant everolimus in combination with anti-estrogen therapy for patients with high-risk early-stage ER+/HER2-breast cancer [12]; many such cases will have anti-estrogensensitive micrometastatic disease.
Preclinical data show that combined therapeutic targeting of ER and mTORC1 is more effective than single agents to inhibit growth and induce apoptosis of breast cancer cells and xenografts [13,14]. However, the synergistic mechanism of anti-cancer action of such drug combinations is incompletely understood. These signaling pathways exhibit crosstalk at several levels. mTORC1 and its effector ribosomal protein S6 kinase β-1 (p70S6K) catalyze serine phosphorylation of insulin receptor substrate-1 (IRS-1), inducing IRS-1 downregulation [15]. ER drives the transcription of genes encoding growth factor receptor tyrosine kinases (RTKs), ligands, and signaling adaptors, including IRS-1 and its activator, insulin-like growth factor-1 receptor (IGF-1R). Estrogen can also stimulate non-genomic ERmediated activation of IGF-1R/insulin receptor (InsR) complexes, epidermal growth factor receptor (EGFR), Src, PI3K, and mitogen-activated protein kinase kinase (MEK) [16]. Estrogen stimulation induces ER binding to the mTORC1 component regulatory-associated protein of mTOR (Raptor), which drives nuclear translocation of Raptor. mTORC1 also phosphorylates ER on S104/106 to promote transcriptional activity [17]. We and others have shown that inhibition of mTORC1 relieves negative feedback on upstream activators, including IRS-1, IGF-1R, human epidermal growth factor receptor 3 (HER3), and PI3K, which may subsequently promote cell survival and antagonize the anti-cancer effects of mTORC1 inhibitors [18][19][20][21]. Herein, we demonstrate a role for ER in such feedback activation of survival pathways in breast cancer cell lines and human tumors, providing supportive rationale for using anti-estrogens to enhance the effects of mTORC1 inhibitors in ER+ breast cancer.

Loss of ER activity abrogates mTORC1 inhibitorinduced feedback activation of PI3K/AKT
Prior reports showed that ER+ breast cancer cells are growth-suppressed by treatment with anti-estrogens, estrogen deprivation, or mTORC1 inhibition [13,14]. In line with these findings, we observed that treatment of 4/4 ER+ breast cancer cell lines with fulvestrant (fulv) or RAD001 suppressed growth (Supplementary Figure 1). Combined targeting of ER and mTORC1 with the combination of fulv/RAD001 synergistically inhibited growth (CI≤1 in 4/4 cell lines; Figure 1A and Supplementary Figure 1).
To explore the mechanism underlying the synergistic effects of anti-estrogens and RAD001 on growth, we pre-treated ER+ breast cancer cells +/-fulv for 24 h (to allow time for ER transcriptional effects to subside), and then co-treated +/-RAD001 for 1 h or 24 h. Treatment with RAD001 inhibited mTORC1 activity, as indicated by decreased levels of phospho-p70S6K (an mTORC1 substrate) and phospho-S6 (a p70S6K substrate) ( Figure  1B). As we and others observed previously [19,22], shortterm (1 h) and longer-term (24 h) mTORC1 inhibition induced increased phosphorylation of AKT at T308 (substrate of PDK1; marker of PI3K activity) and S473 (substrate of mTORC2), implying increased activation of the PI3K/AKT pathway. Co-treatment with fulv mitigated the RAD001-induced increases in P-AKT T308 and P-AKT S473 , suggesting that ER is required for mTORC1 inhibitor-induced PI3K/AKT activation in endocrinesensitive ER+ breast cancer cells ( Figure 1B and Supplementary Figure 2A/2B). The ER-specific effects of fulv were confirmed by RNAi-mediated knockdown of ER, which similarly suppressed RAD001-induced AKT phosphorylation ( Figure 1C). In contrast, fulv less effectively mitigated RAD001-induced AKT activation Oncotarget 8812 www.impactjournals.com/oncotarget in long-term estrogen-deprived MCF-7/LTED cells, and RAD001 increased AKT activation in fulv-resistant MCF-7/FR and T47D/FR cells despite ER inhibition with fulv (Supplementary Figure 3). Thus, combined anti-estrogen/ mTORC1 inhibitor therapy may be most effective in the setting of endocrine-sensitive disease.
We then tested the effects of mTORC1 and ER inhibition on sensitivity to IGF-1 ligand in the absence of other exogenous growth factors. Serum-deprived cells were pretreated +/-fulv or RAD001, then stimulated +/-IGF-1 for 10 min. mTORC1 inhibition with RAD001 sensitized cells to IGF-1-induced activation of AKT without appreciably changing ligand-induced phospho-IGF-1Rβ/P-InsRβ levels ( Figure 2C and Supplementary Figure 4). Fulv treatment mitigated RAD001-induced AKT phosphorylation, which may have occurred by different mechanisms in different cell lines as suggested by differences between T308 and S473 phosphorylation patterns. Fulv hindered IGF-1-induced IGF-1R/InsR phosphorylation in MCF-7 cells, but had only modest effects in ZR75-1 and HCC-1428 cells ( Figure 2C and Supplementary Figure 4). Serum deprivation increased total IGF-1R, InsR, and IRS-1 levels in MCF-7 cells  Figure 2C), which may reflect a loss of negative feedback regulation of expression due to decreased AKT activation [24]. In contrast, fulv decreased IRS-1 and IGF-1R levels in serum-deprived ZR75-1 cells, and IRS-1 in HCC-1428 cells ( Figure 2C and Supplementary Figure 4). Thus, we postulate that fulv hindered RAD001-induced sensitization to IGF-1 in ZR75-1 and HCC-1428 cells by decreasing expression of signaling components (IRS-1, IGF-1R), and in MCF-7 cells by decreasing activation of IGF-1R/InsR; the mechanism underlying the latter is unclear.
To determine whether ER activation primes cells for response to IGF-1, cells were cultured in hormonedepleted conditions for 3 d, serum-deprived for 1 d in the presence or absence of RAD001 or E2, and then stimulated +/-IGF-1 for 10 min. Inhibition of mTORC1 or treatment with E2 each enhanced AKT activation in response to IGF-1, and combined RAD001/E2 was most effective ( Figure 2D). Taken together, these data implicate ER-modulated IGF-1R/InsR signaling in mTORC1 inhibitor-induced feedback activation of PI3K/AKT.

Requirement for IRS-1/2 in mTORC1 inhibitorinduced feedback activation of PI3K/AKT
IGF-1R and InsR activate PI3K by phosphorylating adaptor proteins such as IRS-1 and -2, and Tyr phosphorylated IRS-1/2 engage the p85 regulatory subunit of Class IA PI3K to promote PI3K activity [15,25]. IRS-1/2 levels increased upon RAD001 treatment in MCF-7 cells under growth conditions ( Figure 3A), which is likely due to loss of p70S6K phosphorylation that targets IRS-1/2 for proteasomal degradation [19,26]. Using RNAi against IRS-1 and -2, we observed that dual knockdown decreased AKT activation induced by RAD001 ( Figure 3A). RAD001 also increased association between the p85 regulatory subunit of PI3K and IRS-1 that was partly mitigated by fulv, possibly due to the fulvinduced decrease in IRS-1 expression ( Figure 3B and Supplementary Figure 5). Thus, ER-regulated IRS-1/2 expression and interaction with PI3K are implicated in feedback induced by mTORC1 inhibition.

Estrogen deprivation in patients with ER+ breast cancer prevents tumor AKT activation in response to mTORC1 inhibition ex vivo
Based on our preclinical findings, we hypothesized that inhibition of ER decreases IGF-1R/InsR/PI3K/AKT signaling induced by mTORC1 inhibition in ER+ breast cancer. To test this hypothesis in human tumors without exposing patients to an mTORC1 inhibitor, we conducted a presurgical study in which patients with Stage I-III ER+/ HER2-breast cancer were treated +/-the AI letrozole for 10-21 d to induce estrogen deprivation; the 10-21-day period allows time for maximal estrogen suppression and downstream changes in the levels of ER-regulated transcripts [27]. At the time of surgical resection of the tumor, viable tumor fragments were acquired, adapted Oncotarget 8815 www.impactjournals.com/oncotarget to serum-free medium ex vivo for 5 h, then treated +/-RAD001 or OSI-906 for 1 h. Tumor lysates were analyzed by immunoblot to assess AKT phosphorylation.
We first evaluated tumors from 10 patents in Arm A who did not receive presurgical anti-cancer treatment. Ex vivo treatment of tumor fragments with RAD001 increased the levels of AKT phosphorylation on both T308 and S473 ( Figure 4A). When quantifying (phospho-) protein levels by densitometry (control-treated tumors were set at "1"), we observed a 45% to 2,400% increase in P-AKT T308 compared to control (mean % change +/-SD = 388% +/-623%), and up to a 431% increase in P-AKT S473 (mean % change +/-SD = 157% +/-35%). Treatment with OSI-906 alone or in combination with RAD001 significantly decreased P-AKT levels compared to single-agent RAD001 ( Figure 4A), confirming our in vitro findings (Figure 2A). These results suggest that mTORC1 inhibition induces PI3K/AKT activation in an IGF-1R/InsR kinase-dependent manner in human ER+ breast tumors.
We then analyzed tumors from 7 patents in Arm B who received 10-21 d of letrozole treatment prior to surgery. Ex vivo treatment of Arm B tumors with RAD001 did not significantly increase P-AKT levels: P-AKT T308 ranged from -51% to 163% compared to control (mean % change +/-SD = -10% +/-38%), and P-AKT S473 measured -51% to 281% compared to control (mean +/-SD = 135% +/-91%). Accordingly, OSI-906/RAD001 co-treatment did not significantly alter P-AKT levels compared to RAD001 alone ( Figure 4B). These data suggest that estrogen-induced ER activation is required for mTORC1 inhibitor-induced activation of PI3K/AKT in human ER+ breast tumors. Presurgical anti-estrogen treatment often suppresses cell proliferation in ER+ breast tumors [28]. To confirm the growth-suppressive effects of presurgical letrozole, we measured tumor cell proliferation by Ki67 IHC. Tumor Ki67 scores were not significantly different between baseline biopsies and surgical specimens from patients who did not receive presurgical treatment (Arm A). In contrast, presurgical letrozole significantly decreased Ki67 score in Arm B ( Figure 5A and Supplementary Figure   Figure 4: Presurgical estrogen deprivation in patients with ER+ breast cancer prevents RAD001-induced PI3K/AKT activation in tumors ex vivo. Patients with early-stage ER+/HER2-breast cancer received either A. no presurgical treatment (Arm A, n = 10), or B. presurgical treatment with letrozole for 10-21 d (Arm B, n = 7). Within 1 h after surgical resection, 1-mm 3 punch cores were taken from primary tumors ex vivo. Tumor cores were put in serum-free medium for 5 h, then treated +/-20 nM RAD001, 4 µM OSI-906, or the combination for 1 h. Tumor lysates were analyzed by immunoblot using the indicated antibodies. Immunoblot film signals were quantified by densitometry. Signal values from control-treated tumors were set at "1." Data are shown as log 2 P-AKT of inhibitor-treated sample relative to control. Horizontal bars indicate mean values. *p < 0.05 by Bonferroni multiple comparison-adjusted post-hoc test. C. Representative results are shown at right from 3 patients' tumors from Arms A and B; actin or vinculin was assessed to confirm equal loading.
We further evaluated surgical tumor specimens by immunoblot analysis of lysates. Post-letrozole tumors showed lower IGF-1R, IRS-1, and IRS-2 levels than tumors from untreated patients ( Figure 5C); similar differences were detected by IGF-1R IHC ( Supplementary  Figures 6-7). Letrozole-treated tumors also showed decreased phosphorylation of 4EBP1 and MAPK compared to untreated tumors ( Figure 5C), suggesting that estrogen deprivation suppresses mTORC1 and MEK activation. These findings from human tumors collectively suggest that 1) ER drives the expression of IGF-1R, IRS-1, and IRS-2, and 2) IGF-1R/InsR kinase activity is required for feedback activation of PI3K/AKT upon inhibition of mTORC1, providing mechanistic insight into the effects of mTORC1 and ER inhibition on signaling in ER+ breast cancer.

DISCUSSION
Herein, we provide evidence from ER+ breast cancer cell lines and human tumors implicating ER in signaling responses to mTORC1 inhibition. mTORC1 inhibitors generally have a cytostatic effect on ER+ breast cancer cells [13,14], which is thought to be due in part to a loss of negative feedback signaling to activators upstream of PI3K including RTKs and IRS-1 [18][19][20][21], driving activation of PI3K/AKT to promote cell survival. We demonstrated that mTORC1 inhibition with RAD001 (everolimus) induces IGF-1R/InsR/IRS-1/IRS-2dependent activation of PI3K/AKT signaling in ER+ breast cancer cells and human tumors treated ex vivo. RAD001 synergized with the anti-estrogen fulv to inhibit growth in 4/4 ER+ breast cancer cell lines. Inhibition of ER with fulv, RNAi (in vitro), or estrogen deprivation (AI treatment of patients) decreased mTORC1 inhibitor-induced PI3K/ AKT activation in endocrine-sensitive cells and tumors. In most cases, anti-estrogen treatments decreased expression of IGF-1R/IRS-1/IRS-2 signaling components, which may be the predominant mechanism by which ER drives such feedback signaling ( Figure 6). However, it is also possible Oncotarget 8817 www.impactjournals.com/oncotarget that ER-mediated non-genomic signaling, such as via IGF-1R/InsR [22,29], drives PI3K/AKT activation in response to mTORC1 inhibition.
The extensive crosstalk between the ER and PI3K/ AKT/mTORC1 pathways provides rationale to target these pathways in ER+ breast cancer. However, there is a lack of understanding of the reciprocity of this crosstalk. There are numerous examples of PI3K/AKT/mTORC1 signaling regulating ER activity, including AKT-, mTOR-, and p70S6K-mediated phosphorylation and activation of ER, and AKT-mediated phosphorylation of c-Jun, which complexes with ER [30][31][32][33][34][35][36]. However, the mechanism(s) by which ER affects PI3K/AKT/mTORC1 signaling remain to be fully elucidated. In a positive feedback model, ER promotes the transcription of genes encoding RTKs, ligands, and signaling adaptors, which are predicted to increase PI3K pathway activation. Anti  Figure   7C/7D), although effects varied between biological systems. Accordingly, clinical evidence suggests that ER function sustains PI3K activation. In patients with ER+ breast cancer, 6 months of neoadjuvant treatment with letrozole +/-cyclophosphamide reduces P-AKT S473 and P-mTOR S2448 tumor levels, which correlate with improved response and disease outcome [37]. In a second study, 14 d of neoadjuvant letrozole modestly decreased P-S6 levels [38]. Similarly, we showed that presurgical letrozole decreased P-4EBP1 levels ( Figure 5C). Hence, antiestrogens may suppress ER+ breast cancer growth in part by decreasing PI3K/AKT/mTOR signaling. The findings presented herein closely align with a positive feedback model, and mechanistically show how blocking ER can abrogate activation of PI3K/AKT in response to mTORC1 inhibition.
Clinical and preclinical data also support a negative feedback model of ER-PI3K/AKT/mTORC1 crosstalk in which these pathways antagonize each other. ER and PR levels are inversely correlated with markers of PI3K/ AKT/mTOR activation in primary ER+ breast tumors [5, is the canonical signaling pathway in which ligand-activated IGF-1R/InsR homo-and hetero-dimers phosphorylate IRS-1/2 at Tyr sites, creating docking sites for p85/PI3K that derepress p110/PI3K. p110/PI3K converts PIP 2 to PIP 3 , and PIP 3 enables recruitment of PH domain-containing proteins (e.g., AKT, PDK1) to the cytoplasmic face of the plasma membrane. PDK1 and mTORC2 phosphorylate AKT, and activated AKT in turn signals through the TSC1/2 complex to activate mTORC1. mTORC1 and its phospho-activated substrate p70S6K phosphorylate IRS-1/2 at Ser residues to promote proteasomal degradation of IRS-1/2, creating a negative feedback loop. Inhibition of mTORC1 with RAD001 blocks such negative feedback, leading to increased IRS-1/2/PI3K/AKT activation. ER drives the expression of IGF-1R and IRS-1/2, which are downregulated by anti-estrogen treatment. Thus, anti-estrogens mitigate mTORC1 inhibitor-induced activation of PI3K/AKT in ER+ breast cancer cells.
Oncotarget 8818 www.impactjournals.com/oncotarget 39]. Adaptation to long-term ER inhibition (with fulv or hormone deprivation) increases PI3K/AKT/mTORC1 activity in ER+ breast cancer cells [5,[40][41][42]. PI3K inhibitor treatment of patients with ER+ breast cancer for ≥14 d induces upregulation of tumor ER levels and activity [43], which Toska et al. recently demonstrated is mediated by loss of AKT-induced epigenetic silencing of ER activity [44]. Furthermore, we previously showed that adaptation of ER+ breast cancer cells to long-term PI3K inhibition induces upregulation of ER and PR [45]. Thus, substantial data support seemingly conflicting models of positive and negative feedback between the ER and PI3K/ AKT/mTORC1 pathways, resolution of which will require further study.
AKT phosphorylation at both T308 and S473 is required for maximal kinase activation [46]. Phosphorylation of T308 is mediated by the PI3K effector PDK1, while S473 phosphorylation is mediated by mTORC2 [47,48]. The effects of mTORC1 inhibition on AKT phosphorylation varied between cell lines: RAD001 drastically increased both P-AKT T308 and P-AKT S473 in ZR75-1 and MCF-7 cells, but only robustly increased P-AKT S473 in HCC-1428 cells ( Figure 1B and Supplementary Figure 2). These observations indicate cell line-specific differences in PI3K-PDK1-mTORC2-AKT programming, and suggest that inhibition of ER and mTORC1 affect different feedback circuits upstream of T308 and/or S473. AKT has 3 human isoforms (AKT1, AKT2, and AKT3) with overlapping and distinct biological functions [46], and similar amino acid motifs surrounding T308 and S473. Since phospho-AKT antibodies do not distinguish between isoforms, it is possible that isoformspecific differences in AKT phosphorylation contributed to the observed differences between cell lines.
In 2012, the U.S. FDA approved everolimus (RAD001) as the first PI3K/AKT/mTOR pathway inhibitor for the treatment of advanced ER+ breast cancer in combination with the steroidal AI exemestane following progression on a non-steroidal AI. The BOLERO-2 study showed that combined exemestane/everolimus therapy increased median PFS from 4.1 to 10.6 months compared to exemestane/placebo [6]. An exploratory subgroup analysis revealed that patients with cancer that had progressed during or within 12 months after the end of (neo)adjuvant AI therapy, rendering exemestane +/-everolimus the first-line therapy for their advanced disease, experienced even greater benefit from the addition of everolimus: exemestane/everolimus increased median PFS to 15.2 months from 4.2 months with exemestane/ placebo [49]. Lingering questions remain regarding 1) the optimal time in the disease course to introduce everolimus, 2) the need for continued anti-estrogen therapy for antiestrogen-resistant disease in the context of mTORC1 inhibition with everolimus, and 3) the clinical significance of mTORC1 inhibitor-induced feedback activation of upstream pathways (i.e., RTKs, PI3K, AKT). Herein, we provide evidence from preclinical models and functional human tumor studies demonstrating a role for ER in RAD001-induced PI3K/AKT activation in endocrinesensitive ER+ breast cancer. These findings provide rationale for anti-estrogen co-treatment in the setting of endocrine-sensitive ER+ disease being treated with an mTORC1 inhibitor; such drug combinations are currently being tested clinically in the adjuvant setting [12].

Sulforhodamine B (SRB) assay
Cells were seeded in triplicate at 3,000-5,000/well in 96-well plates. The following day, cells were treated as indicated for 5 d. Relative numbers of adherent cells were determined by SRB staining [50]. IC 50 ), and an interim analysis was performed. The next 7 patients were presurgically treated with the AI letrozole for 10-21 d (Arm B). Primary tumors were surgically resected as standard of care, and tumor fragments for research use were then dissected from the surgical specimen within 1.5 h of removal from the patient. Tumor fragments were transported in serum-free DMEM on ice. Using a 1-mm biopsy punch device (Miltex), ~40 viable tumor cores were dissected from each tumor specimen within 30 min. Tumor cores (~10 per treatment group) were cultured ex vivo in serum-free DMEM for 5 h, then treated +/-20 nM RAD001, 4 µM OSI-906, or the combination for 1 h. Tumor cores were then snap-frozen, lysed, and analyzed by immunoblot.

Immunohistochemistry (IHC)
Five-micron sections of formalin-fixed, paraffinembedded (FFPE) tumor tissues acquired from patients at baseline (diagnostic biopsy) and at the time of surgery (untreated or after presurgical letrozole) were used for IHC staining with antibodies against Ki67 (Biocare Medical), ER (Cell Marque, SP1), PR (Leica Biosystems, # NCLl-PGR-312), or IGF-1R (Cell Signaling, #3027). IHC evaluation was performed by a breast pathologist blinded to treatment arm; details are provided in Supplementary Information. In Arm A, two patients had bilateral tumors, and two patients had two ipsilateral tumors; all were analyzed by IHC.

Authors' Contributions
WY and TWM designed the preclinical studies. WY, VC, NAT and TWM conducted preclinical experiments, and collected, analyzed, and interpreted data. GNS, TWM, and JG designed the clinical study. GNS directed and recruited patients to the clinical study. JDM procured clinical tissue specimens, and scored IHC specimens. WY performed ex vivo analyses of fresh clinical specimens. WY and TWM wrote the manuscript. All authors read and approved the final manuscript.