Targeting STAT3 signalling using stabilised sulforaphane (SFX-01) inhibits endocrine resistant stem-like cells in ER-positive breast cancer

PURPOSE Estrogen receptor (ER) positive breast cancer is frequently sensitive to endocrine therapy. Multiple mechanisms of endocrine therapy resistance have been identified, including cancer stem-like cell (CSC) activity. Sulforaphane (SFN) has previously been shown to target CSCs but its mechanism of action is unclear. Here we investigate SFX-01, a stabilised formulation of SFN, for its effects on breast CSC activity in ER+ preclinical models and to study its mechanism. EXPERIMENTAL DESIGN CSC activity was measured by mammosphere formation efficiency (MFE), aldehyde dehydrogenase (ALDH) activity, and tumor formation using patient samples and patient-derived xenograft (PDX) tumors treated with SFX-01 alone or in combination with tamoxifen or fulvestrant. Gene expression and SFN target proteins in treated samples were assessed. RESULTS SFX-01 reduced MFE of both ER+ primary and metastatic patient samples. Both tamoxifen and fulvestrant increased MFE and ALDH activity of PDX tumors, which was reversed by combination with SFX-01. SFX-01 significantly reduced tumor initiating cell frequency in secondary transplants at limiting dilution and reduced the formation of spontaneous lung micrometastases by PDX tumors in mice. Mechanistically, we establish that both tamoxifen and fulvestrant induce STAT3 phosphorylation. SFX-01 suppressed phospho-STAT3 and SFN directly bound STAT3 in patient and PDX samples. Analysis of ALDH+ cells from endocrine-resistant patient samples revealed activation of STAT3 target genes MUC1 and OSMR, which were inhibited by SFX-01 in patient samples. Increased expression of these genes after 3 months’ endocrine treatment of ER+ patients (n=68) predicted poor prognosis. CONCLUSIONS Our data establish the importance of STAT3 signaling in CSC-mediated resistance to endocrine therapy and the potential of SFX-01 for improving clinical outcomes in ER+ breast cancer.


INTRODUCTION
Three out of four cases of breast cancer express the estrogen receptor alpha (ER) and are treated with endocrine therapies, such as selective ER modulators (SERMs, e.g. tamoxifen), aromatase inhibitors (AIs, e.g. letrozole) and selective ER downregulators (SERDs, e.g. fulvestrant) (1). However, despite the undoubted success of endocrine treatments, distant breast cancer recurrences and death occur at a steady rate for at least 15 years after the 5-10 year treatment period is completed (2). This finding stresses the need for new approaches that can provide long-term diseasefree survival.
Endocrine resistance hampers the cure of ER+ breast cancer, and therefore major efforts have been employed to address the mechanisms. Acquired resistance can be mediated by modulation of ER activity through its mutation, by up-regulation of ER co-activators (e.g. FOXA1), by activation of mitogenic signalling pathways (e.g. MAPK, PI3K/AKT) induced by receptor tyrosine kinase activity (e.g. EGFR, ERBB2) or by overexpression of substrates for cyclin-dependent kinases (CDKs) (3,4).
Several drugs targeting these pathways have been tested in clinical studies, and recently CDK4/6 kinase inhibitors have been shown to increase overall survival (OS) of patients with advanced ER+ breast cancer and three (palbociclib, ribociclib and abemaciclib) are now FDA approved (5)(6)(7).
Cancer stem cells (CSCs) can be responsible for tumor initiation and growth and are more resistant than non-CSCs to cancer therapies, such as chemo-and radiotherapy (8). We and others have shown that breast CSCs (measured by the percentage of ALDH+, or mammosphere-forming cells) are not targeted by endocrine therapies in ER+ breast cancer (9)(10)(11). This leads to enrichment in cells with breast CSC activity which are dependent upon developmental signalling pathways, such as Notch and Wnt (12,13). Eradication of endocrine-resistant CSCs is likely to provide long-term disease-free survival but so far none of the approved drugs for patients with ER+ tumors has been shown to target CSCs. Sulforaphane (SFN), an isothiocyanate found in cruciferous vegetables, has demonstrated activity against breast CSCs (14), although clinical application has been hampered due to its inherent instability. In order to improve its stability, SFX-01, SFN conjugated within an alpha-cyclodextrin complex, has been developed (15).
Here, we report that SFX-01 used in combination with endocrine therapies prevents breast CSC enrichment in patient samples and PDX tumors in vivo, and mechanistically it directly targets STAT3 to inhibit its activity in endocrine resistance.
STAT3 activation in patient primary tumors measured by expression of STAT3induced genes predicts resistance to endocrine treatments.

SFX-01 reduces breast CSC activity in primary and metastatic ER+ breast cancer patient derived-samples
Mammosphere forming efficiency (MFE) and ALDH1 enzymatic activity have previously been shown to be a characteristic of breast CSCs (16,17), which can be targeted by SFN in breast cancer cell lines (14). SFX-01 (SFN stabilised in alphacyclodextrin complex, Figure 1A) reduced MFE in thirteen of sixteen patient-derived ER+ tumor samples (Figure 1B and Tables S1 and S2 for patient and tumor characteristics). Next, we assessed CSC activity following SFX-01 pre-treatment of cells from patients with endocrine-resistant breast cancer. SFX-01 reduced the percentage of ALDH+ cells in all six samples (and by >60% in five; Figure 1C) and also significantly decreased MFE in four of these samples ( Figure 1D). These data suggest that short term SFX-01 treatment targets breast CSC activity in patientderived and endocrine-resistant cells.
We then investigated the effect of SFX-01 on MFE of patient-derived ER+ tumor cells co-treated with tamoxifen. SFX-01 remained effective and significantly reduced MFE in combination with tamoxifen in eight out of ten samples tested ( Figure 1E). In addition, we tested 3-day pre-treatment of ER+ cell lines (MCF-7, T47D and ZR-75-1) with tamoxifen or fulvestrant in combination with SFX-01. Both ALDH activity and MFE were reduced by SFX-01 in all cell lines confirming reversal of the breast CSC activity that is induced by short-term treatment with anti-estrogen drugs (Supplementary Figure 1).

SFX-01 prevents tamoxifen enrichment for cells with cancer stem cell properties in patient-derived xenograft tumors
To better model the clinical scenario of our treatments, two ER+ patient-derived xenograft (PDX) models were grown subcutaneously in NSG mice, one from an early (HBCx34, (18)) and another from a metastatic tumor (BB3RC31, (19)). We assayed proliferation and breast CSC activity in these PDX models after 14-day in vivo treatment with either SFX-01, tamoxifen or combination of both drugs (Figure 2A).
We observed that tamoxifen treatment decreases the proliferation marker Ki67 and tumor growth in both PDX models and SFX-01 reduced tumor growth in the HBCx34 model, but had no significant impact on BB3RC31 ( Figure 2B, Supplementary   Figure 2A). However, breast CSC activity measured by ALDH enzymatic activity or MFE was increased by tamoxifen and decreased by SFX-01, both alone and in combination with tamoxifen (Figures 2C and 2D). We then performed in vivo limiting dilution transplantation of HBCx34 PDX cells derived from tumors previously treated with tamoxifen and/or SFX-01. Cells from tumors treated with SFX-01 could not form any tumors on reimplantation of 4,000 cells, in contrast with cells from tumors treated with vehicle control or tamoxifen alone ( Figure 2E). SFX-01 inhibits PDX tumor growth, prevents CSC activity and prevents spontaneous metastasis to the lungs To further validate the in vivo effects of combining SFX-01 with anti-estrogens on CSC activity we treated HBCx34 PDX cells for 56 days with either tamoxifen or fulvestrant ( Figure 3A). As expected, both anti-estrogens strongly reduced the number of proliferative cells measured by Ki67 expression (Figure 3B) whilst increasing both ALDH activity and MFE (Figure 3C and 3D). SFX-01 had no effect on proliferation ( Figure 3B) but significantly inhibited anti-estrogen stimulated breast CSC activity (Figures 3C and 3D). 56-day treatment of the second PDX model (BB3RC31) also demonstrated that the anti-estrogen enrichment of breast CSCs can be reduced by SFX-01 (Supplementary Figure 3A and 3B). Over the 8-week treatment period, tamoxifen plus SFX-01 significantly suppressed tumor growth versus tamoxifen alone in the HBCx34 (Figure 3E) but not the BB3RC31 PDX models (data not shown). On the other hand, fulvestrant treatment maintained tumor growth suppression at 8 weeks and no additive effect was seen with SFX-01 in either HBCx34 ( Figure 3F) or BB3RC31 (data not shown) PDX models.
We next investigated whether treatments had an effect on PDX spontaneous lung metastasis. Using a human-specific mitochondrial antibody we could clearly observe micrometastases in the lungs from mice bearing HBCx34 PDX tumors treated with tamoxifen or fulvestrant (Figure 3G and 3H). In contrast, lungs from HBCx34 mice treated with tamoxifen or fulvestrant in combination with SFX-01 were free from micrometastases, although disseminated tumor cells persisted (Figure 3G and 3H).
In the BB3RC31 PDX model SFX-01 reduced micrometastases by 50% in combination with fulvestrant but no effect was seen in combination with tamoxifen (Supplementary Figure 3C and 3D), suggesting SFX-01 hinders colonisation of the lungs by disseminated tumor cells in sensitive tumors. SFX-01 targets STAT3 signalling, which is activated by anti-estrogen therapy We next sought to determine the mechanism of sensitivity to SFX-01. The direct binding targets of SFN have been assessed in breast cancer cell lines, identifying STAT3 and NF-κB subunits among the top high-affinity targets (20). STAT3 was a particularly interesting target since ALDH+ breast CSCs and tamoxifen resistant cells have previously been shown to express higher levels of phospho-STAT3 (active form), and inhibition of STAT3 reduces ALDH+ cell numbers and tumorigenicity (21,22). HBCx34 PDX tumors treated in vivo for 56 days with tamoxifen or fulvestrant showed increased phospho-STAT3 expression which was inhibited by co-treatment with SFX-01 ( Figure 4A). In the BB3RC31 PDX model we also observed that fulvestrant, but not tamoxifen, increased phospho-STAT3 expression and SFX-01 only reduced phospho-STAT3 levels in fulvestrant-treated tumors (Supplementary Figure 4B). We also assessed whether phospho-NF-κB p65 was modulated in a similar way to STAT3 but overall its expression was not changed in either HBCx34 or BB3RC31 PDX treated tumors (Supplementary Figure 4A and 4B). Next, we examined STAT3 activation in a HBCx34 PDX model that has been selected for tamoxifen resistance, which we previously demonstrated to display an enrichment of breast CSCs (12). Increased phospho-STAT3 was seen compared to the parental endocrine-sensitive HBCx34 PDX model and this was reversed by treatment with SFX-01 ( Figure 4B). Pull-down experiments with biotin-labelled SFN in HBCx34 PDX tumor cells confirmed direct interaction of SFN with STAT3 in both antiestrogen and vehicle-treated tumors (Figure 4C), establishing STAT3 as the likely target of SFX-01.
Next we tested whether SFX-01 inhibited STAT3 signalling in anti-estrogen resistant patient samples. Cells from metastatic sample BB3RC61 treated with SFX-01 showed a significant reduction in phospho-STAT3 expression compared to tamoxifen or fulvestrant treatments alone ( Figure 4D). Furthermore, we observed direct binding of SFN to STAT3 in cells from this same patient sample ( Figure 4E). Analysis of four additional anti-estrogen resistant metastatic samples confirmed that SFX-01 combined with either tamoxifen or fulvestrant inhibited STAT3 activity in three of these patient samples (BB3RC44, BB3RC66 and BB7103), though SFX-01 only inhibited fulvestrant-induced STAT3 activation in BB3RC66 sample (Supplementary Figure 4C). Interestingly, SFX-01 did not repress STAT3 activity in sample BB7121 that had previously been treated clinically with SFX-01 (STEM trial). This patient's disease had progressed in the presence of SFX-01 and was thus resistant to it.
On the whole, we have established, using PDX models in vivo and patient samples in vitro, that SFX-01 targets STAT3 and can inhibit its activity that is induced by treatment with anti-estrogens. demonstrated differential expression (fold change ≥ 2) of 28 STAT3-related genes that were all up-regulated in ALDH+ cells ( Figure 5A). Next, we hypothesised that this 28-gene STAT3 signature of ALDH+ cells could predict prognosis of patients diagnosed with ER+ breast cancers. In published gene expression microarray datasets from 762 ER+ tumors (KMplotter, (23)), we found that elevated expression of the 28 STAT3-related genes before treatment was significantly associated with breast cancer recurrence ( Figure 5B). Thus, activation of STAT3 signalling in ER+ tumors is associated with more aggressive disease.
We therefore tested whether SFX-01 could reduce expression of the top four STAT3 related genes in ER+ anti-estrogen resistant patient samples, and we found that two of the genes (OSMR and MUC1) were significantly down-regulated by SFX-01 treatment ( Figure 5C). Based on these findings, we investigated if failure to reduce expression of these two STAT3 genes with anti-estrogen treatments was associated with worse outcomes due to STAT3 activation in a 68-patient matched dataset (24) before and after 3 months treatment with an aromatase inhibitor. Remarkably, increase in the sum of these two genes (OSMR and MUC1) upon anti-estrogen treatment was significantly associated with decreased breast cancer specific survival ( Figure 5D). The change in expression of each of these genes individually was not significantly associated with outcome (data not shown).
Thus, persistent activation of the STAT3 signalling pathway after anti-estrogen treatment could potentially be used as a biomarker to predict endocrine resistance and, hence, to identify the patient group most likely to benefit from STAT3 inhibition with SFX-01 therapy.

DISCUSSION
In this study we establish that SFX-01 targets anti-estrogen resistant breast CSCs in patient samples and patient-derived xenografts (PDXs). Compared with antiestrogen treatment alone, SFX-01 combined with anti-estrogen drugs significantly reduced ALDH+ CSCs, their sphere-forming activity and tumor initiating cell frequency, as well as formation of lung micrometastases in PDX tumors grown in mice. Mechanistically, both tamoxifen and fulvestrant induced STAT3 phosphorylation although not always to the same degree in individual tumors. We show that SFX-01 suppresses this adaptive STAT3 phosphorylation and establish that SFN directly binds to and targets STAT3 in cells from patient samples and PDX tumors. Importantly, analysis of ALDH+ cells from endocrine-resistant patient samples revealed activation of STAT3 signalling and expression of STAT3 target genes MUC1 and OSMR, which were down-regulated by SFX-01 in patient samples.
Moreover, increased expression of these two genes after 3 months' endocrine treatment predicted poor prognosis in patients with ER+ tumors. We propose that inhibiting STAT3 signalling with SFX-01 may help to overcome endocrine therapy resistance and recurrence in ER+ breast cancer.
We previously reported that CSC activity is increased by the anti-estrogens tamoxifen and fulvestrant, both in vitro and in vivo (12), and others have demonstrated a role for SFN in targeting breast CSCs. Here, we show a stabilised form of SFN, SFX-01, is a potent inhibitor of endocrine resistant CSCs. In contradistinction to the previous report that used cell lines (14), we establish that STAT3, not Wnt signalling, is the primary target mechanistically responsible for endocrine resistance of breast CSCs in human breast tumors. STAT3 activation has been associated with ALDH expression in triple negative breast cancer cell lines (22), however, this is the first time this association has been reported for endocrine resistant breast cancer. We previously reported that Notch4 signalling is increased in anti-estrogen resistant ALDH+ breast CSCs (12), and using a tamoxifen-resistant MCF7 sub-line, an association was reported between Notch4 and STAT3 activation (21). However, we have not confirmed here that STAT3 phosphorylation is confined to the ALDH+ population. The relevance of our finding is strengthened by demonstrating that STAT3 is activated in both anti-estrogen treated PDX and patient-derived samples.
The recent finding that SFN bound directly to STAT3 with high-affinity in breast cancer cell lines (20) led us to investigate STAT3 as the target in anti-estrogen treated PDX and endocrine resistant patient-derived samples. We show convincingly that binding occurs and that SFX-01 reduces the phosphorylation of STAT3 in vitro and in vivo, in short-term anti-estrogen treatment and established resistance. This establishes its utility and provided a rationale for its use in clinical trials in patients.
Therefore, we recently recruited 47 patients with metastatic breast cancer who were progressing on endocrine therapy and showed that the addition of SFX-01 to the endocrine therapy resulted in clinical benefit in 25% of patients (25). Our in vivo data demonstrate where primary tumor phospho-STAT3 is reduced, there is a reduction in metastatic lung colonisation. This indicates that SFX-01 could reduce distant recurrence rates in early breast cancer.
The in vivo data presented, demonstrating reduction in metastatic lung colonisation when primary tumors phospho-STAT3 was reduced, suggest that SFX-01 could reduce distant recurrence rates in early breast cancer. Although this clearly requires additional preclinical and clinical testing.
In order to better target STAT3-driven endocrine resistant breast cancer, we analysed gene expression in samples from advanced breast cancer patient who had received prior endocrine therapy. We derived a STAT3 signature consisting of 28 genes that was associated with poor progression-free survival of ER+ breast cancer patients. Importantly, two of the most expressed genes were down-regulated by SFX-01 treatment of patient-derived samples. These two genes, OSMR and MUC1, were both STAT3-responsive and SFX-01-sensitive in patient samples, and their increased expression after 3 months' endocrine treatment accurately predicted survival over a 14 year period after surgery for breast cancer. These data suggest that this two gene signature could have potential to select tumors responsive to SFX-01, which would have to be confirmed in a larger group of patients.
OSMR has already been associated with worst prognosis although not specifically for ER+ tumors (26), and an association between OSMR signalling and reduced ER levels has been reported (26). In addition, there are reports showing that OSMR signalling increases invasive capacity of breast cancer cell lines, perhaps explaining increased metastasis (27). Moreover, a MUC1 signature is associated with high MUC1 expression and predicts failure to tamoxifen treatment (28).
Thus, overall we establish the potential of SFX-01 for clinically meaningful improvements to endocrine therapy in ER+ BC by inhibiting STAT3 signalling and reversing CSC-mediated resistance. The derivation of a STAT3 pathway signature that predicts poor prognosis and resistance to endocrine therapy indicates that a sub-population of breast tumors might be the rational targets for SFX-01 therapy in combination with current endocrine therapies.

EXPERIMENTAL PROCEDURES
Breast cancer patient-derived samples and metastatic BC samples were processed as described previously (12,19).
Supplementary Tables S1 (early BC) and S2 (metastatic BC) present the clinicopathological characteristics of the samples used in this study.
PDX tumors were treated when reached 200-300 mm 3 . Tumor size and animal weight were recorded twice weekly. SFX-01 (Evgen Pharma, 300 mg/kg/day) and tamoxifen citrate (Sigma-Aldrich, #T9262, 10 mg/kg/day) were administered by oral gavage (0.1 ml/dose) on a five day from seven day basis (weekends excluded) for 14 or 56 days; whereas Fulvestrant (200mg/kg/week, Astrazeneca) was injected subcutaneously (0.1 ml/dose) on a weekly basis for 56 days. SFX-01 and tamoxifen citrate were made in 1% carboxymethylcellulose (Sigma-Aldrich, #C9481) dissolved in distilled water. Upon termination, xenografts tumors were collected in ice-cold DMEM media and processed as described in (12) for further downstream analyses.
Mouse lungs were formalin-fixed paraffin-embedded for histological assessment of metastatic disease.
In vivo limiting dilution assays were performed to evaluate tumor initiation ability of

Gene expression analysis of ALDH+/-populations in ER+ metastatic breast cancer
Whole transcriptomic datasets for ALDH+ and ALDH-populations from ER+ metastatic patient-derived samples were available here (30). Briefly, ER+ advanced breast cancer patients were drained of metastatic fluids (ascites or pleural effusions) due to discomfort. Breast cancer cells were isolated following the methodology described elsewhere (19) and then stained using the ALDEFLUOR assay. After   MFE data is represented as mean percentage ± SEM. * p < 0.05; ** p < 0.01 ELDA of tumor-initiating cell frequency is shown.