Evolution of Cancer Stem-like Cells in Endocrine-Resistant Metastatic Breast Cancers Is Mediated by Stromal Microvesicles Free

Tumor heterogeneity and resistance to therapy may occur from microvesicle-mediated transfer of genetic material between cells (1–3). Thus, the characterization of this phenomenon could have important clinical ramifications most notably in the development of new therapeutically relevant compounds.

Although adjuvant hormonal therapy (HT) improves disease-free survival in luminal breast cancer patients, HT-resistant (HTR) metastatic disease commonly develops in the bones of these patients. This observation suggests that the bone microenvironment may foster estrogen receptor (ER)-independent growth of luminal breast cancer leading to HTR metastases.

The interaction of stromal cells (cancer-associated fibroblasts; CAF) with tumor cells has been shown to mediate and modulate estrogen receptor–dependent (e.g., fibronectin, collagen) and independent proliferation (e.g., laminin) of luminal breast cancer cells, suggesting that stroma–tumor communication may play a pivotal role in the ER-independent self-renewal of breast cancers (4). In the metastatic microenvironment, we hypothesize that chronic inflammation incurred by anti-estrogen therapy and the effects of disseminated tumor cells on the local microenvironment will lead to the activation of resident stromal cells or circulating mesenchymal stem cells to become CAFs. Once activated, the CAFs may sustain a feed-forward circuit of self-renewal, proliferation, and differentiation of CSCs, resulting in metastasis.

As tumors become more metastatic and resistant to targeted therapies, the number and types of CSCs increases, suggesting that CSCs evolve from non-CSC cells in a given tumor niche (5, 6). The role of stroma microvesicles (MV) in the generation of therapy-resistant cancer and the regulation of self-renewal remains poorly investigated.

Here, we investigated the hypothesis that HT- and CAF-derived microvesicles converge to promote HT resistance and ER-independent self-renewal in luminal breast cancer. By employing patient-derived xenografts from breast cancer bone metastases and experimental models of luminal breast cancer, we uncovered a unique process of CAF-mediated resistance to HT. Our data demonstrate the formation of therapy-resistant stromal-tumor niches via an IL6/Stat3-driven expansion of CAFs, CAF-MV–mediated oncomiR-221 transfer to cancer cells leading to the expansion of Notch3^hi/ER^lo/CD133^hi CSCs. These data reinforce the concept of targeting the stromal niche to prevent both HT resistance and metastatic progression (7–9).

Plasma (10 mL) was collected and processed within 4 hours from patients with metastatic disease (Supplementary Table S1) and in healthy controls who were consented to an MSKCC biospecimen protocol #12-137. The plasma and conditioned media (CM) from cancer and CAF cultures was collected from 10⁷ cells grown in 5 × 10 cm dishes and centrifuged for 20 minutes at 3,000 × g at 4°C. The supernatant was subsequently centrifuged for 30 minutes at 12,500 × g at 4°C. The supernatant was transferred and centrifuged at 100,000 × g for 90 minutes at 4°C. The supernatant was discarded while the pellet, containing microvesicles, was resuspended in 25 mL of PBS and loaded onto a 5 mL 30% sucrose cushion to deplete microvesicles from extracellular proteins (300 g/L sucrose, 24 g/L Tris base, pH 7.4). Samples were centrifuged at 100,000 × g for 90 minutes 4°C. Cushion (3.5 mL), containing microvesicles, was diluted with 10 mL of PBS and centrifuged at 100,000 × g for 90 minutes at 4°C. The supernatant was discarded and the pellet resuspended in 25 μL of PBS. Microvesicles were treated with 0.1 mg/mL of DNase I solution (Epicentre) to eliminate contaminating DNA bound to the microvesicles surface or present in solution. Nanosight (Lyden laboratory, Cornell Medical Center) and electron microscopy (MSKCC Electron Microscopy Core) were used to characterize the physical structures of these microvesicles (size and distribution). Confocal microscopy (MSKCC Microscopy Core) of cancer cells educated with prelabeled (PKH67-Green Fluorescent Cell Linker Kit, Sigma) CAF-MVs was performed to ensure transfer and uptake. The in vivo role of CAF-MVs in the promotion of HT-resistant luminal breast cancer was determined by injecting CAF-MVs (Mu-CAFs, isolated from HT-resistant xenografts and cultured in vitro) and control microvesicles (from MCF7 cells) into the arterial circulation (retro-orbital injection, 3 × 10⁹ particles/mouse/weekly) of tumor-bearing mice (MCF7 cells). Once mammary fat pad (MFP) xenografts were established (after 4 months), mice were treated with HT (fulvestrant; 100 μg/injection/once a week for 3 months).

Patient-derived xenografts (PDX) were established from n = 2 of 5 HT-resistant and 1 de novo stage IV breast cancer bone metastatic tissue isolated at MSKCC (Supplementary Table S2). Patients who developed metastatic breast cancer in the bone were enrolled in the study (IRB protocol #97-094). Following surgery, tissue was processed by the pathologist, 60% of the specimen was used for confirmation of diagnosis and molecular analyses (IHC, IMPACT analyses), while 40% was used for further analysis. Tissues were placed in sterile Epicult (Voden Medical), minced with sterile razor blades and incubated at 37°C for 8 to 12 hours in the presence of Collagenase/Hyaluronidase enzyme mix (1,000 Units, Voden Medical). To grow tumor cells devoid of its cognate stroma, we performed serial centrifugations to separate epithelial cells as mammosphere cultures (MS). Secondary and tertiary MS potential (II or III-MS) was performed as follows: 7-day primary MS started to form after 4–6 days, then they were disaggregated in 1× Trypsin-EDTA (StemCell Technologies), washed in complete MEGM, filtered through a 40-μm nylon mesh, and seeded to form second generation MS. Number of MS was assessed by counting the total number of spheres (size > 100 μm) from cells seeded in low-attachment plates (from 100 to 1,000). To establish primary CAF cultures from patient-derived tissue, MS-depleted supernatant was centrifuged at 450 × g for 10 minutes; this pellet was enriched with stromal cells was plated onto 10-cm plates supplemented with DMEM 10% serum media. CAF primary cultures were expanded in vitro for n = 10 passages. III-MS primary cultures were used to establish PDXs: 50–100 MS (size ∼100 μm) were injected in the MFP of NOD/SCID mice and tumor growth was determined over a period of 5 months. At the endpoint of the experiment, xenograft tissues were collected and primary PDX cultures were established. Multiple passaged PDX were generated following repeated orthotopic injection of PDX-derived EpCAM^pos cancer cells (recognizes only human cells) in the MFPs of NOD/SCID mice (from 1st to 4th generation). Luminal breast cancer cells expressing a vector for GFP/Luciferase were generated and used for all the in vivo experiments. Tumor growth was determined using in vivo bioluminescence technology (BLI: Xenogen, IVIS System). Luminal cancer xenografts from the coinjection of human CAFs and MCF7 cells were also generated to determine the effect of the stroma on the generation of de novo–resistant endocrine tumors

All cancer cell lines were engineered to express a GFP-positive luciferase expression vector. Prior to in vivo inoculation, cancer cells were FACS purified (for GFP) and injected bilaterally in the MFPs of 5- to 7-week-old NOD/SCID) mice (obtained from NCI, Frederick, MD). For each in vivo experiment, cancer cells were mixed with an equal volume of Matrigel (BD Biosciences) in a total volume of 50 μL. Bioluminescence was used to monitor both tumour growth (weekly) and metastatic burden (at necropsy). Luminal cancer xenografts from the coinjection of human CAFs (HTR bone metastases) and MCF7 cells (10³ cells) were also generated to determine the effect of the stroma on the generation of de novo–resistant endocrine tumors. In addition, human bone marrow stromal cells HS27a, HS27shC, and HS27shIL6 (100 cells/injection) were coinjected with MCF7 cells (10³ cells/injection) into the MFP. For immunostaining assays, organs were collected and fixed overnight in 4% paraformaldehyde, washed, embedded in paraffin, and sectioned (Histo-Serve Core). Hematoxylin and eosin (H&E) staining was performed by standard methods. For the detection of metastases at secondary sites, we performed in vivo BLI as well as immunofluorescence/IHC staining for GFP and ER. All the surgical procedures and animal care followed the institutional guidelines and an approved protocol from our IACUC at MSKCC. For the preclinical studies, injectable fulvestrant (Faslodex, AstraZeneca) was given intramuscularly in the tibialis posterior/popliteal muscles (100 μg/injection/once a week) for 2 months. Tocilizumab (Actemra, Roche Pharmaceuticals) was diluted in PBS at a final concentration of 20 mg/mL. A dosage of 100 μg/g/mouse was administered intraperitoneally every week (this is ∼5-fold higher than the physiologic range, patients receive 8 mg/kg i.v.). Control mice received isotype control (placebo) or PBS injection.

Human cancer cell lines (Namalwa, lymphoma; HeLa, cervical carcinoma), human breast cancer cell lines (MCF7, ZR751, T47D, and BT474), human bone marrow stromal cell lines (HS5, HS27a), and human normal fibroblasts (MRC5, HMF) were purchased from the ATCC and authenticated by short tandem repeat (STR) DNA profiling (Genomic Core MSKCC). Murine CAFs (Mu-CAF) were isolated from HTR xenografts and PDXs by FACS purification (GFP⁻/EpCAM⁻). Cells were mycoplasma free and maintained in minimum essential medium and RPMI (ATCC and Media Core) supplemented with 5% FBS (Media Core), 2 mmol/L glutamine, 100 U mL⁻¹ penicillin, and 0.1 mg mL⁻¹ streptomycin (Media Core). Cancer cells from xenografts were isolated from primary and metastatic tissues by enzymatic digestion (Collagenase/Hyaluronidase, Sigma-Aldrich), sorted (GFP⁺/DAPI⁻), and cultured in vitro. The following reagents: 4-hydroxytamoxifen and fulvestrant were purchased from Sigma (Sigma-Aldrich). For FACS/flow analyses, tumors were digested in sterile Epicult media (StemCell Technology), minced with sterile razor blades, and incubated for 3 hours 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 40 μm nylon mesh (BD Biosciences). For the detection of CD44 and CD133, EpCAM antigens, cells were stained in a volume of 100 μL (PBS-BSA 1%) with each antibody CD44-APC (100 ng/10⁶ –10⁸ cells Clone IM7, eBiosciences), CD133/1-PE (100 ng/10⁶–10⁸ cells, clone AC133, Miltenyi Biotech) and EpCAM-FITC (250 ng/10⁶–10⁸ cells, Clone VU-1D9, StemCell Technologies). Cells were labeled on ice for 30 minutes and analyzed (BDFACS Aria III, Flow Core). Samples were analyzed for cell population distribution and sorted for GFP/viability (GFP⁺/DAPI⁻) and CD133/CD44 expression. For flow plot analyses, samples were run using FlowJo 7.5 software (Tree Star). shRNAs for Notch3 and IL6 were previously described (10, 11).

Normalized gene expression values were downloaded from the GEO under accession number GSE17705 and probes were aggregated to median gene level expression. A CAF gene set from Allinen and colleagues (12) was used in a single sample gene set enrichment analysis (ssGSEA). ssGSEA scores were z-scored and a "CAF score" was assigned to each patient. Patients were split at the median into CAF-high and CAF-low groups. PROM1-high and PROM1-low groups were split based on PROM1 (CD133) median expression. Statistical significance for differences in PROM1 expression and ESR1 expression were assessed with a Student t test. The heatmap for CAF signature gene and PROM1 was plotted in R with the heatmap.2 function. For real-time PCR (qPCR), we extracted RNA using TRIzol (Invitrogen). RNA concentration was determined with a NanoDrop 2000. For microarray analysis of published datasets, normalized gene expression data were downloaded from the Gene Expression Omnibus (GEO). Each gene was mean centered and scaled by SD. All analyses were conducted in R. Normalized gene expression data was downloaded from the NCBI for dataset GSE69280 (5). For qPCR, 1 μg of total RNA was reverse transcribed to cDNA using iScript Select cDNA Synthesis Kit (Bio-Rad) following the manufacturer's protocol. Reverse transcription PCR (RT-PCR) analysis was performed using the following primers: ERα; forward 5′-TGAAAGTGGGATACGAAAAGAC-3′, reverse 5′-CAGGATCTCTAGCCAGGCACAT-3′; β2μ forward 5′-ACCCCCACTGAAAAAGATGA-3′, reverse 5′-ATCTTCAAACCTCCATGA-3′. DNA was isolated using phenol/chloroform technique from PDX-derived EpCAM-positive/negative cells. The presence of murine and human cells was determined on 2 ng of DNA by PCR for GAPDH (murine: forward 5′-AGCAGCCGCATCTTCTTGTGCAGTG-3′, reverse 5′- GGCCTTGACTGTGCCGTTGAATTT-3′; Human: forward 5′-CTCTGCTCCTCCTGTTCGAC-3′, reverse 5′- ACGACCAAATCCGTTGACTC-3′). miRNA expression was analyzed as described previously (13). Briefly, miRNA were reverse transcribed using stem-loop RT-PCR technology (14) and amplified by real-time PCR using SYBR Select Master Mix (Applied Biosystems) and ViiA 7 Real-Time PCR System (Applied Biosystems) according to the manufacturer's instructions. The melting curve data were collected to check PCR specificity. miRNA expression was normalized against RNA U6 levels: (RT-miR-221) 5′-GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACGAAACCC-3′; (RT-miR-222) 5′-GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACACCCAGT-3′; (RT-miR-101) 5′-GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACTTCAGTT-3′; (forward-miR-221) 5′-AGCTACATTGTCTGCTGGGTTTC-3′; (forward miR-222) 5′-AGCTACATCTGGCTACTGGGT-3′; (forward miR-101) 5′-GCCGCTACAGTACTGTGA-3′; (forward U6) 5′-CTTCGGCAGCACATATACT-3′; (reverse U6) 5′-AAAATATGGAACGCTTCACG-3′ (reverse all miRs) 5′-TGCAGGGTCCGAGGTAT-3′. All primers were purchased from Eurofins MWG Operon. miRNA expression was analyzed as described elsewhere (13).

For immunoblotting assays, cells were lysed in buffer (50 mmol/L Tris at pH 7.5, 150 mmol/L NaCl, 5 μg/mL aprotinin, pepstatin, 1% NP-40, 1 mmol/L EDTA, 0.25% deoxycholate, and protease inhibitor cocktail tablet, Sigma). Proteins were separated by SDS-PAGE, transferred to polyvinylidene difluoride membranes and blotted with specific antibodies (Supplementary Table S3). For functional interference studies, anti-miR-221 and control RNA oligonucleotide were purchased from Applied Biosystems. MCF7 cells were seeded in a 6-well plate (8 × 10⁵ cells/well) at 60% confluence. After 24 hours, cells were transfected using Lipofectamine 2000 transfection reagent (Invitrogen) according to the manufacturer's instructions (RNA final concentration, 200 nmol/L). After 6 hours of incubation at 37°C, transfection medium was replaced with 2 mL of complete medium containing 10% FCS supplemented with/without CAF-MVs. For determination of cell viability, we seeded 2,500 cells per well in 96-well plates and treated them with fulvestrant (10 μmol/L). Viable cells were determined 7–14 days after treatment using Trypan blue and cell counting was done using bright-field microscopy or DAPI staining by flow cytometry (Dako Cytomation). Crystal violet assay was performed to obtain information of the relative cell density at the endpoint of proliferation potential experiments. IL6 ELISA assays were performed using the conditioned medium collected from 5-day cultures of CAF-derived cells seeded at 2 × 10⁶ cells/plate. Proliferation assay was carried out using CalceinAM technology (Invitrogen): cells were seeded in 96-well plates treated with the prefluorescent compound for 20 minutes and fluorescence was read using a plate reader (SpectraMax plate platform). To determine the selective growth potential of cancer cells over stroma cells, we analyzed the proliferation potential of luciferase-positive cancer cells by in vitro BLI: cells were seeded in 96-well plates in presence/absence of distinct CAFs/normal fibroblast (1:10 ratio of CAFs to tumor cells) and treated with fulvestrant (10 μmol/L/weekly for 3 weeks). Luciferase activity was measured weekly. Cytokine Arrays were performed on 10 μg of CM-derived proteins according to manufacturer's protocol (Antibody Array 3, RayBiotech. Inc.).

Serial sections of formalin-fixed paraffin-embedded samples were immunostained using monoclonal anti-CD133 diluted 1:70 (clone W6B3C1, Miltenyi Biotec), anti-ERα RTU (clone SP1, Ventana), anti-Pankeratin RTU (clone AE1/AE3/PCK26, Ventana), and polyclonal anti-Notch-3 diluted 1:400 (M-134, Santa Cruz Biotechnology). CD133 and Notch3 immunostaining was performed as follows: sections were dewaxed, rehydrated, and subjected to antigen retrieval treatment. Antigens were unmasked with a Tris-EDTA pH 9.0 buffer at 98°C for 20 minutes in a waterbath. Endogenous peroxidase activity was inhibited using a 0.5% H₂O₂ solution in methanol for 20 minutes at room temperature. Sections were processed using a non-biotin–amplified method (Novolink, Novocastra) according to the manufacturer's protocols. When mouse tissue was used, a short treatment (30 minutes at room temperature) with MOM blocking solution (Vector Laboratories Inc.) was conducted prior to primary antibody overnight incubation at 4°C. The reaction was visualized using the UltraView DAB Detection System. The immunologic reaction was developed using a 3,3′-diaminobenzidine (DAB)/H₂O₂ PBS pH 7.2–7.4 solution for 10 minutes. Sections were then washed in distilled water and counterstained in Harris hematoxylin. Anti-ERα (ER) and Pankeratin (CK) immunostaining was performed on an automated immunostainer (Benchmark Ultra, Ventana) using the UltraView DAB Detection kit according to the manufacturer's protocol. Antigen retrieval was performed onboard with UltraCC1 buffer (pH 8.2–8.5) at 95°C for 52 minutes (ER) or 20 minutes (CK). Primary antibodies were incubated 28 minutes at 37°C (ER) or 8 minutes at room temperature (CK). For CD133 and Notch3 evaluation, each section was examined at 400×. In each microscopic field, the neoplastic cells were classified according to both positive percentage and staining intensity: [percentage = 0 if <1%, 1 if >1% < 25%, 2 if >25% < 50%, 3 if >50% <75%, 4 if >75%; intensity = 1 (weak), 2 (moderate), and 3 (strong)]. A final classification was obtained by multiplying the two mean values (percentage and intensity, IRS score). As for ER evaluation, the neoplastic population was scanned using Image Cytometry and reported as percentage of positive cells (%) (IMAGE-Pro Plus V5.0.1, Media Cybernetics Inc.). A detailed histologic examination of xenograft tissues was performed at the collaborating institution (University of Bologna, Bologna, Italy). Xenograft tissue was stained with hematoxylin and eosin and examined by three independent pathologists (C. Ceccarelli, Donatella Santini, and M. Bonafe, from the University Hospital of Bologna, Bologna, Italy). For each microscope field (200×), the area occupied from cancer cells, stromal cells, and necrotic components was evaluated and represented as percentage.

Serial sections (5 μm) of paraformaldehyde-fixed paraffin-embedded samples underwent antigen retrieval using Leica Bond ER2 Buffer (Leica Biosystems) for 20 minutes at 100°C before staining with 1 μg/mL Desmin rabbit polyclonal antibody (Abcam catalog no. ab8592) and 1 μg/mL pStat3 (clone D3A7, Cell Signaling Technology) for 1 hour using Leica Protocol F (Molecular Imaging Core facility, MSKCC, New York, NY). Quantification of Desmin/pStat3 staining was performed using ImageJ/FIJI (NIH, Bethesda, MD). At least 19 fields at 400× were randomly selected and evaluated. The results were expressed as percentage of immunostained cells/over total area of tissue. To discriminate between cancer and stromal cells, fortified H&E staining was also performed (HistoServ Inc). A color deconvolution algorithm was then used, with RGB vectors for the stromal component and counterstain/background stain created from regions of interest drawn from example images (Molecular Imaging Core facility, MSKCC, New York, NY). Appropriate thresholds were then set for each cell type of interest and area measurements were taken for all images. To rule out possible non-CAFs/noncancer cells component, specific staining for CAFs (desmin-murine CAFs) and cancer cells (human pankeratin) was also performed in serial section slides. Stroma–tumor niches were evaluated as area of tissue slide with the copresence of pankeratin-positive cells and stroma cells.

ALDEFLUOR analysis was performed using the ALDEFLUOR Kit (StemCell Technologies) according to the manufacturer's protocol. Cancer cells from PDX primary cultures were washed with 5 mL 10% PBS supplemented with Accumax (Innovative Cell Technologies), and single-cell suspensions were first stained with anti-CD133-PE–conjugated antibody for 20 minutes, washed twice with PBS-BSA (5%), and then incubated with ALDEFLUOR reagent.

CM was isolated from CAFs and cancer cell lines (10⁸ cells), concentrated using Amicon Ultra-15 centrifuge tubes (Millipore), and protein levels were measured by the Lowry technique; 10 μg of total extracellular protein was loaded for zymographic/protein (MMP-2, MMP-9) and in vitro studies (invasion capacity). Cell growth of cocultured cancer cells with CAFs was determined with and without anti-Jagged1/Notch3 blocking antibody (AF1277, R&D Systems 500 ng every 72 hours). Briefly, luciferase-positive breast cancer cells (MCF7) grown with CAFs (1:50) were seeded in 96-well plate and treated with fulvestrant (10 μmol/L/weekly) in the presence of mouse anti-Jagged1 blocking antibody. BLI signals were measured every 48 hours and growth curves were generated accordingly.

Cells were plated in 6-well plates at a density of 2 × 10⁵ cells per well. Cells were transfected with 0.3 μg of promoter luciferase (CD133; ref. 15) and the activated form of Notch3 (pNICD3 2 μg; ref. 10). To normalize transfection efficiency, cells were also cotransfected with 0.1 μg of the pRL-CMV (Renilla luciferase, Promega). Forty-eight hours after transfection, luciferase activity was measured using the Dual-Luciferase Assay Kit (Promega). Three independent experiments were performed, and the calculated means and SDs are presented.

qPCR was performed on 100 ng of cDNA using TaqMan precustom probes (Applied Biosystems, ERα Hs00174860 62 pb, GATA3 Hs00231122 80 bp, FOXA1 Hs0418755 59 bp, GREB1 Hs00536409 67 bp, EGR3 Hs00231780 91 bp, CCL5 Hs00174575 63 bp, PGR Hs01556702 77 bp) and SYBR Green technique (α-sma forward 5′- CAGGGCTGTTTTCCCATCCAT-3′, reverse 5′-GCCATGTTCTATCGGGTACTTC-3′; SDF-1α forward 5′-CCATGAACGCCAAGGTCGTG-3′, reverse 5′- CCAGGTACTCCTGAATCCAC-3′; Vimentin forward 5′-TGGCACGTCTTGACCTTGAAA-3′, reverse 5′- GGTCATCGTGATGCTGAGAA -3′; Slug 5′-AGATGCATATTC GGACCCACA-3′, reverse 5′- CCTCATGTTTGTGCAGGAGA-3′; CD44 forward 5′-CAGCAACCCTACTGATGATGACG-3′, reverse 5′- GCCAAGAGGGATGCCAAGATGA -3′). ViiATM 7 Real-Time PCR System was used (Applied Biosystems) in accordance with the manufacturer's instructions. For analysis, ΔC_t method was applied and fold change was calculated (2^−ΔΔC_t). All values were normalized to GAPDH expression (TaqMan, Hs02758991). RT-PCR for Notch3 (forward 5′-TCAGGCTCTCACCCTTGG-3′, reverse 5′-AGTCACTGGCACGGTTGTAG-3′), Jagged1 (forward 5′-TCGCTGTATCTGTCCACCTG-3′, reverse 5′-AGTCACTGGCACGGTTGTAG-3′) and β2μ as internal control 5′-ACCCCCACTGAAAAAGATGA-3′, reverse 5′-ATCTTCAAACCTCCATGA-3′ was performed in MCF7 cells control and shNotch3 and mCAFs/fibroblast cell lines.

Statistical analysis was performed by SPSS (SPSS Inc.). Continuous variables were analyzed by unequal variance t test, paired t test (samples, n = 2), general linear model (GLM) ANOVA, or GLM for repeated measures (samples, n > 2). Mann–Whitney and Wilcoxon tests were used to analyze ordinal variables. P values were adjusted for multiple comparisons according to Bonferroni correction. Association among quantitative variables was quantified by Pearson correlation coefficient. Categorical variables were analyzed by Monte Carlo χ² test. All the tests were two-sided. P < 0.05 was considered significant. Elda software was used to measure the statistics of limiting dilution experiments (bioinf.wehi.edu.au/<http://bioinf.wehi.edu.au/>software/elda/).

The presence of CAFs have been assessed as prognosticators in breast cancer and an “active stromal signature” in normal fibroblasts exhibits a tumor-promoting phenotype (16). Many stromal-secreted factors including IL6, SFD-1α, and HFG participate in the communication between CAFs and tumor cells within the tumor microenvironment.

Stromal microvesicles have also been implicated in tumor progression in glioblastomas and ovarian cancers (17, 18). However, the molecular and pathologic relevance of CAF-derived microvesicles in luminal breast cancer remains unclear.

To study tumor progression in luminal breast cancer, we established long-term xenografts of highly tumorigenic MCF7 and ZR751 cells (5). Following tumor establishment (1 cm), all mice received HT (fulvestrant a selective estrogen receptor degrader commonly given to patients with ER⁺ metastatic disease, 10 μmol/L) for 3 months. Although the majority of xenografts displayed sensitivity to HT (HTS, stable disease or remission), approximately 10% of the xenografts (data not shown) grew in the presence of therapy (Fig. 1A, HTR resistance to HT). Interestingly, the histologic analysis of these tissues revealed the enrichment of CAFs in the HTR xenografts (Fig. 1B; Supplementary Fig. S1A). Furthermore, we could isolate and in vitro passage CAFs from HTR-derived xenograft tissues. These CAF cell lines displayed the upregulation of CAF markers by Western blot analysis and the capability of growth for multi passages (more than 20 passages; Fig. 1B and data not shown). Although we were able to isolate CAFs from HTS lesions in a small fraction of xenografts (5%, n = 3 out of 60), we could not propagate them in culture for more than 2 passages (2 weeks). Therefore, no CAF cell lines (0%) were established from HTS xenografts.

To further characterize these tumor-associated stromal cells, we cultured HTR tumors and isolated stromal cells by FACS (negative selection with EpCAM, which recognizes epithelial cells) and determined that EpCAM^neg cells were morphologically spindle shaped, were murine in origin (expressed murine genomic DNA, data not shown) and expressed markers of activated CAFs including Fap, vimentin, fibronectin, and activated Stat3 (Fig. 1B, phospho tyrosine 705 Stat3 pStat3).

Next, we asked whether these CAFs could promote de novo HTR disease. We cocultured murine CAFs and human-HS27a “CAF” like cells (bone marrow–derived immortalized mesenchymal cells) with HT (Luciferase^pos) naïve cancer cells in the presence/absence of HT (fulvestrant, 10 μmol/L/weekly) and cancer cell growth was analyzed by in vitro bioluminescence after 2 weeks (Fig. 1C). We found that CAFs promoted tumor cell growth following HT, whereas no difference was found in the absence of therapy (Fig. 1D; Supplementary Fig. S1B). In contrast to HS27a cells and murine CAFs, normal fibroblasts (mammary and lung) did not confer resistance to HT in cocultures (Supplementary Fig. S1C).

In addition to growth factors, stromal cells have been shown to secrete microvesicles, which can horizontally transfer numerous prosurvival factors and confer resistance to radiotherapy (19). Here we determined that the number of microvesicles produced by murine CAFs and HS27a cells was much greater than MCF7 tumor cells (Fig. 1E). To determine whether these microvesicles could confer a protumorigenic advantage, we set up an in vivo model (Fig. 1F). MFP xenografts from HT-naïve cells (MCF7) were established; CAF-MVs (3 × 10⁹) and control microvesicles (tumor-derived microvesicles, MCF7) were injected retro-orbitally weekly for 7 months. Once tumors were established (after 4 months) all mice received HT (fulvestrant weekly). Although there was no difference in tumor growth before HT, those animals treated with CAF-MVs had tumors resistant to HT (HTR) while MCF7-MVs provided no benefit as tumors regressed with HT (Fig. 1G). Overall, these data demonstrated that circulating stromal microvesicles can induce resistance to HT in vivo.

Distinct genes and pathways have been associated with resistance to HT including the activation of mutations in the ESR1 gene (20), increased Her2 expression (21), decreased ER levels, and ER transcriptional signatures (22, 23), increased expression of oncomiRs including the ER repressor miR-221/222 (24) and, more recently, increased Notch signaling in CSCs (5, 25).

As a reduction in ER expression is associated with resistance to HT (22), and CD133^hi CSCs have lower ER levels (mRNA and protein) as compared with CD133^lo/CD44^lo cells, we reasoned that the suppression of ER signaling could be a mechanism of stroma-mediated expansion of therapy-resistant CSCs (CD133^hi/Notch3^hi).

To test our hypothesis, we first demonstrated that the CM from CAFs (murine and human), but not normal fibroblasts, led to a decrease in ER protein expression and ER-dependent transcripts (e.g., GATA3, FOXA1, GREB1, EGR3, CCL5, PGR) in MCF7 cells (Fig. 2A; Supplementary Fig. S2A). As the suppression of ER protein occurred with CM from both murine and human CAFs, we hypothesized that rather than soluble factors (which can be typically species-specific; refs. 26, 27), CAF-derived miRNAs might be mediating the downregulation of ER expression. Microvesicles have been suggested to be mediators of nucleic acid transfer including miRNAs (27). Among different miRNAs, the forced overexpression of oncomiR-221/222 in luminal breast cancer cells has been found to reduce ER expression and promote HT resistance (24). In addition, increased plasma levels of miR-221 were found in ER-negative breast cancer patients (28). The administration of 10⁸ microvesicles from mCAFs to ER⁺ cancer cells (MCF7) reduced ER levels after 48 hours (Fig. 2B). We showed that oncomiR-221/222 sequences are conserved between human and mouse species, suggesting possible functional cross-species interactions (Fig. 2C). Importantly, miR-221 expression was found in circulating microvesicles from patients with HTR metastatic disease (independent of tumor burden) as compared with healthy controls (Fig. 2D, n = 11; Supplementary Table S1).

We determined that CAF-derived microvesicles were enriched for miR-221 compared with normal fibroblasts and distinct cancer cells lines (breast, cervical, and lymphoma), suggesting that stroma-derived microvesicle could account for the increased level of miR-221 expression in microvesicles (Fig. 2E). Accordingly, miR-221, but not a control miRNA (miR-101), was highly expressed (100- to 200-fold enrichment) in microvesicles from CAFs as compared with microvesicles from normal fibroblasts, CAFs and cancer-derived cell lines (Fig. 2F; Supplementary Fig. S2B). Administration of murine CAF-MVs compared with normal fibroblast-MVs (MRC5) to MCF7 cells (oncomiR-221 negative) led to the transfer of oncomiR-221 (Fig. 2G, 20-fold increase in expression in educated cancer cells).

To address the consequences of microvesicle-derived oncomiR-221 transfer in mediating HT resistance, anti-miR-221 was transfected in cancer cells before microvesicle administration. We demonstrated that anti-miR-221 transfection abrogated mCAF-MV–dependent downregulation of ER protein (Fig. 2H) and HT resistance following chronic mCAF-MV education (Fig. 2I). Restored sensitivity to HT in anti-miR-221–transfected cells was associated with an increase in ER expression/activity (Supplementary Fig. S2C and S2D). These data suggest that CAF–microvesicle–mediated HT resistance occurs via the transfer of onco-miR-221 promoting an ER^lo phenotype.

Increased expression of Notch and downstream signaling events as well as a higher number of CSCs are found in HT-resistant breast cancer (29, 30). Among Notch proteins, Notch3 and Notch4 are crucial mediators of resistance to HT in distinct models of luminal breast cancers (5, 25, 31, 32).

Given the pivotal role of CAF-derived microvesicle in promoting the switch from sensitive to resistant disease (HTS to HTR), we asked whether these stromal microvesicles could also modulate Notch3 expression. We cultured MCF7 cells with fulvestrant (10 μmol/L/weekly) for 2 weeks (see schematic Fig. 1C) with microvesicles (10⁸ particles/weekly) from either MRC5 or mCAFs. Although no effect was observed in the absence of HT (no treatment), mCAF-MVs restored Notch3 expression and activity (Hes1, Hey 1 mRNAs) following fulvestrant, which associated with increased growth (Fig. 3A; Supplementary Fig. S3A, HTR cells see Fig. 2I). Conversely, MRC5-MVs administered cells did not overcome ER-dependent downregulation of Notch3 expression and activity (Hes1, Hey 1 mRNAs) as well as the suppression of growth (Fig. 3A, for HTS cells, see Fig. 2I). The transfection of anti-miR-221 led to decreased Notch3 protein levels and restored sensitivity to HT (fulvestrant) of CAF-MV–treated cancer cells (Figs. 2I and 3B). These data demonstrate that miR-221 in CAF-MVs can block HT-mediated downregulation of Notch3 expression.

We recently described the enrichment of CD133^hi/ER^lo/Notch3^hi CSCs in HT-resistant tumors, which also expressed high levels of Notch-regulated genes such as Hey1 and Hes1 (GSE69280; ref. 5).

Given the role of CAF or stromal microvesicles in promoting a miR-221^hi/ER^lo/Notch3^hi phenotype, we tested the hypothesis that CAFs could promote the in vivo expansion of CD133^hi cells via Notch3 upregulation. The selective reduction of Notch3 expression in cancer cells (shNotch3) and activity (using an anti-Jagged1 blocking antibody) abrogated CAF-mediated HT resistance and the expansion of CD133^hi cancer cells (Fig. 3C and D; Supplementary Fig. S3B). In agreement with the knockdown experiment, overexpression of the activated form of Notch3 (pNICD3) in MCF7 cells led to an increase in CD133 promoter luciferase activity (pCD133) with HT (fulvestrant) in association with a reduction in ER protein levels (Supplementary Fig. S3C and S3D). These data suggest that higher Notch3 activation could promote a feed-forward ER^lo/CD133^hi loop necessary for the generation of CD133^hi CSCs (Supplementary Fig. S3C and S3D). Overall, our data describe a novel mechanism of HT resistance: CAF-MV–mediated transfer of the onco-miR-221 leading to reduced ER expression and Notch3 upregulation.

As the biogenesis of oncomiR-221/222^hi microvesicles occurs preferentially in CAFs (not normal fibroblasts), we hypothesized that the abrogation of a CAF phenotype would interfere with the generation of HT-resistant CSCs. To investigate this hypothesis, we examined possible candidates responsible for CAF growth. Compared with breast cancer cells, the CM of CAFs (HS27a cells) expressed higher levels of chemokines (e.g., IL8, MIP-1δ, CCL5) and cytokines, including IL6 an activator of Stat3 (Fig. 4A). These findings were further supported by evidence of high pStat3 levels in murine CAFs from HTR-derived xenografts (Fig. 4B). Differently from other signaling pathways (HER, PI3K, ER), pStat3 activity was required for CAF proliferation as well as the generation of oncomiR-221^hi microvesicles (Fig. 4C; Supplementary Fig. S4A). In concordance with these data, reduced IL6 expression in HS27a cells (using an IL6-shRNA) lowered secreted IL8/IL6R/CCL5 levels as well as the expression of CAF markers including pStat3, vimentin, and CD44 (Fig. 4D and E). In addition, compared with HS27shCT cells, HS27shIL6 cells had reduced proliferative and invasive potential (Supplementary Fig. S4B and S4C) as well as lowered MMP2/9 expression and activity, indicating a loss of characteristic CAF features (Supplementary Fig. S4D). Moreover, microvesicles from HS27shIL6 cells had essentially no expression of oncomiR-221 compared with HS27shCT cells with no change in microvesicle production (Fig. 4F, protein content as a surrogate marker of microvesicle yield). These data suggest that IL6/pStat3 signaling is crucial for the proliferation of CAFs and the production of oncomiR-221⁺ microvesicles.

To address the phenotypic consequences of decreasing IL6 signaling in CAFs, we coinjected MCF7 cells with HS27shCT and HS27shIL6 CAFs into the MFP of mice. Compared with controls (MCF7/HS27a), the coinjection of MCF7/HS27shIL6 cells resulted in impaired tumor growth, lower metastatic burden, and decreased expression of CD133^hi/Notch3^hi/ERα^lo CSCs (Fig. 4G–I; Supplementary Fig. S4E). In agreement with the loss of CD133^hi/Notch3^hi CSCs, MCF7/HS27shIL6–derived tumors had fewer stroma–tumor niches (Fig. 4J and K) and decreased pStat3 expression (Supplementary Fig. S4F). Overall, our data suggest that IL6-mediated generation of stromal niches is required for the expansion of CD133^hi CSCs.

To extend our results to clinical specimens, we established primary cultures of stromal cells from patient-derived bone metastases (Supplementary Table S2, BM-CAFs). IL6 is a well-known pleiotropic cytokine, secreted at high levels from the bone marrow microenvironment and CAFs (33, 34). We isolated and cultured CAFs from bone metastases (Supplementary Fig. S5A and data not shown) and detected high levels of IL6 protein from the CM of these primary cultures (Fig. 5A). These levels are similar to those found from the CM of HS27a cells. Consistent with the HS27 model, the abrogation of IL6 signaling, using the anti-IL6R-IL6 antibody (tocilizumab), abrogated the growth in 70% of CAF primary cultures (Fig. 5B), reduced IL6 secretion and the expression of CAF markers (Fig. 5C; Supplementary Fig. S5B), this finding suggests that autocrine IL6 maintains the CAF phenotype of these cells.

Although miR-221/222 expression is very low in luminal breast cancer tissues and cells (Fig. 2), high levels of miR-221 was found in circulating microvesicles from HTR patients (Fig. 2D). Next, we reasoned that, in agreement with other investigators, CAFs would be the major source of miR-221 (35). We then isolated microvesicles from CAF primary cultures derived from bone metastases and found increased levels of oncomiR-221 as compared with tumor microvesicles (Fig. 5D); these results were similar to those found in HS27a cells and mCAFs (Fig. 4). We subsequently demonstrated that the coinjection of BM-CAFs with MCF7 cells promoted HTR tumor growth (treated with fulvestrant) in an IL6-dependent manner, as combination fulvestrant/tocilizumab abrogated tumor growth (Fig. 5E).

We and others have recently demonstrated the enrichment of CSCs expressing CD133 or ALDH^hi activity in experimental luminal breast cancers following HT (5, 25). In addition, expression of Prominin1 (CD133) was identified in tumors from patients who progressed on adjuvant HT (23). In agreement with other investigators, CD133^hi and CD44^hi cells are functionally CSCs as they were capable of engrafting with low cell numbers (<1,000, Supplementary Fig. S6A). We previously demonstrated that differently from CD44^hi cells, CD133^hi/CD44^lo cells expressed embryonic stem cell signatures and increased the expression of ABCG2, a CSC gene associated with therapy resistance (GSE69280; refs. 5, 36). Here we determined that CD133^hi/CD44^lo cells expressed normal stem cell signatures by GSEA (Supplementary Fig. S6B, GSE69280; ref. 5). In addition, compared with CD133^lo/CD44^lo and CD44^hi/CD133^lo cells, the injection of CD133^hi/CD44^lo cells gave rise to slow-growing tumors (Supplementary Fig. S6C) with an increased capacity to disseminate to the bone marrow (Supplementary Fig. S6D). These data suggest that the CD133^hi phenotype is a distinct CSC population (37).

Next, we investigated the hypothesis that extrinsic (stromal heterogeneity) factors could regulate the evolution of therapy-resistant niches leading to metastatic progression (38, 39). We demonstrated high PROM1 (encoding CD133) expression by microarray (GSE17705) was associated with increased levels of a mammary CAF gene signature (12) in the setting of human HT-resistant primary tumors (Fig. 6A, P < 1.82 × 10⁻⁴; Supplementary Fig. S7A).

To examine the importance of CD133^hi cells in clinically relevant models of metastatic breast cancer, we generated PDXs from HT-resistant (HTR) luminal breast cancer bone metastases (Fig. 6B; Supplementary Table S2). Cultured tumor cells were serially transplanted into the MFP of immunocompromised mice in the presence of HT (fulvestrant: a selective estrogen receptor degrader). After three sequential passages, the enhanced tumorigenic capacity (% tumor take) of PDXs was associated with increased Notch3 expression (Supplementary Fig. S7B), enrichment of CD133^hi (∼40-fold) cancer cells and murine stromal cells (∼30 fold), which had infiltrated the tumor (Fig. 6B and C; Supplementary Fig. S7C and S7D). We further determined that the PDX-derived CD133^hi cells had low ALDH activity, suggesting different and unique stem cell populations arising through a stroma-mediated metastatic transition within ER⁺ breast cancer (Supplementary Fig. S7E).

Given the coenrichment of CD133^hi cells with CAFs, we hypothesized that CAFs could directly promote the expansion of HT-resistant CD133^hi cells. To address this hypothesis, we depleted HT-resistant PDX primary cultures from CAFs (by FACS for EpCAM^pos) and maintained these cancer cells in culture for several months. When long-term depleted for murine CAFs, EpCAM^pos cells from PDX primary cultures were growth inhibited by HT as compared with tumor cells freshly isolated from cocultures (sorting for EpCAM^pos, Fig. 6D and E). Moreover, acquired sensitivity to HT resulted in decreased numbers of self-renewing CD133^hi cells (Fig. 6E and F).

To determine whether CAFs could confer de novo HT resistance and promote the biogenesis of CD133^hi cells, we cultured HT-sensitive MCF7 cells with murine CAFs, isolated from xenografts, with/without HT (fulvestrant) for 4 weeks (Fig. 6G). When cultured with CAFs, MCF7 cancer cells became resistant to HT (fulvestrant), as compared with MCF7 cells alone (Fig. 6G). Concomitant with HTR cancer cell growth, by FACS we observed a marked (30-fold) increase in CD133^hi CSCs as compared with MCF7 cells alone (Fig. 6H). Further characterization of the CD133^hi cells revealed that ERα mRNA levels were lower in these cells than in CD133^lo/CD44^lo cells (Fig. 6I). In agreement with these data, higher CD133/PROM1 expression was associated with decreased ESR1 mRNA expression and activity in HT-resistant primary tumors (Fig. 6J, P < 0.0001). Overall, these findings demonstrate that CAFs, in the presence of HT, can promote the de novo formation of CD133^hi CSCs and HTR disease.

In summary, our data led us to propose the following model: IL6 upregulation drives a CAF phenotype leading to the generation of stroma–tumor niches in vivo. The presence of active CAFs promotes a skewing of the cancer cell population toward a CD133^hi/Notch3^hi/ER^lo CSC phenotype. This occurs via the production/action of CAF-derived microvesicle, which reduces ER activity leading to HTR disease (Fig. 7; ref. 40).

Significant progress has been made in identifying tumor cell–specific factors (e.g., tumor secretome and gene signatures) that promote cancer cell survival and proliferation in the bone microenvironment (27, 41). In addition, cross-talk between bone marrow–derived myeloid cells, osteoclasts, and osteoblasts with tumor cells has identified many growth factors, chemokines, and miRNAs as critical regulators of bone metastasis (42). A recent manuscript by Luo and colleagues has also demonstrated that increased osteoblast-derived IL6 promotes tumor cell seeding and bone metastases in a model of breast cancer cells injected in the arterial circulation (43). IL6 is a pleiotropic cytokine, which is secreted from several cell types, including CAFs and plays a crucial role in the expansion of cancer stem cells (44, 45) as well as in the proliferation potential of CAFs (46, 47).

The presence of CAFs have been assessed as a poor prognostic feature in breast cancer (48) and an “active stromal signature” in normal fibroblasts exhibits a tumor-promoting phenotype (16). Moreover, PDX tumor tissue was shown to be enriched with host CAFs (49); but the molecular and pathologic relevance of any given CAF^hi phenotype in PDXs remains unclear.

The majority of breast cancers (∼70%) are of the ER-positive or luminal subtype. Although, the suppression of ER activity with HTs has led to improved survival, when these cancers recur they preferentially metastasize to the bone, and eventually acquire resistance to HT and are thus incurable (50, 51).

The upregulation of distinct pathways, including Her2, PI3K, and/or overactive estrogen signaling, are found in metastatic cancer cells escaping tumor dormancy from adjuvant HT. Although targeting these pathways in the metastatic setting leads to clinical responses (52), resistance to anti-Her2/estrogens/PI3K regimens invariably occurs (27, 53). We and others have reported the clinical relevance of high numbers of CD133^hi cells in “therapy”-resistant cancers including lung cancer and HT-resistant luminal breast cancer metastases (5, 54).

Recently a crucial role for stromal microvesicle in tumor progression has been suggested (18, 19). However, whether CAF-MVs could promote therapy resistant breast cancer and whether these microvesicles could recapitulate the phenotypic role of CAFs in the tumor microenvironment remain under debate.

Although HT itself suppresses ER signaling, HT alone cannot be a unique trigger of metastatic disease in luminal breast cancer. We propose that communication between CAFs and tumor cells promotes an ER-dependent to an ER-independent switch in metastatic disease. This can occur via the genetic transfer of miRNAs (221/222), leading to the posttranscriptional downregulation of ER and the expansion of HT-resistant tumors. Recently, the overexpression of miR-221/222 was demonstrated to promote mammosphere generation in T47D cells. In addition, a manuscript from Shah and colleagues has proved the presence of miR-221–high microvesicles from CAFs (35). However, the phenotypic relevance of these CAF-MVs in the context of HTR disease was not investigated.

In this article, we have demonstrated that the expansion of CD133^hi CSCs is functionally associated with the expansion of CAFs in experimental and patient-derived HTR disease. We developed the hypothesis that CAF-derived microvesicles could generate de novo HTR disease via a miR-221–mediated conversion of non-CSCs (ER^hi) into therapy-resistant CSCs (ER^lo). We generated PDX models of luminal breast cancer and isolated CAFs from HTR bone metastases (Supplementary Table S2) and through their analysis uncovered a step-wise process of CAF-MV–mediated HT resistance: the (i) IL6-pStat3–dependent activation of CAFs, (ii) the biogenesis of oncomiR-221/222–high microvesicles, (iii) the transfer of these oncomiRs from CAF-MVs to ER^hi cancer cells, (iv) the suppression of ER signaling, Notch3 activation, and the generation of CD133^hi/ER^lo/Notch^hi CSCs.

Given the pivotal role of IL6 in CAF cell growth and the generation of stromal–tumor niches, we identified combination IL6R–IL6 blockade and HT as a therapeutic intervention to abrogate the establishment of stromal–tumor niches and endocrine resistance in metastatic luminal breast cancer (Fig. 7). Taken together, our data suggest a novel pathologic role of stromal IL6 in luminal breast cancer: the secretion of oncomiR-221/222 high microvesicles leading to the evolution of therapy-resistant stromal–tumor niches. We characterized the cellular components of this stroma–tumor niche: “CD133^hi CSCs and CAFs” and determined the molecular machinery responsible for the niche generation: autocrine IL6 in CAFs and CAF-derived microvesicle-dependent downregulation of ER in cancer cells.

N. Fabbri is a consultant/advisory board member for Illuminoss Medical Inc. No potential conflicts of interest were disclosed for the other authors.

Conception and design: P. Sansone, D. Lyden, J. Bromberg

Development of methodology: P. Sansone, M. Berishaj, C. Savini

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Sansone, M. Berishaj, V.K. Rajasekhar, Q. Chang, A. Strillacci, L. Shapiro, A. Benito-Martin, N. Fabbri, J.H. Healey, J. Bromberg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Sansone, M. Berishaj, C. Ceccarelli, Q. Chang, A. Strillacci, R. Bowman, C. Mastroleo, F. Perna, E. Spisni, M. Cricca, D. Lyden, M. Bonafé

Writing, review, and/or revision of the manuscript: P. Sansone, V.K. Rajasekhar, L. Shapiro, L. Daly, D. Lyden, M. Bonafé, J. Bromberg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Sansone, M. Berishaj, S.D. Carolis

Study supervision: P. Sansone, D. Lyden, J. Bromberg

We are grateful to Mesruh Turkekul, Afsar Barlas, Sho Fujisawa, Romin Yevgeniy (Molecular Cytology Core), and Donatella Santini (Department of Experimental, Diagnostic and Specialty Medicine, University of Bologna, Italy) for advice and technical assistance.

This study is supported by grants from Department of Defense (W81XWH-10-1-1013 to P. Sansone) the NIH (R01: CA87637 to J. Bromberg), Charles and Marjorie Holloway Foundation (J. Bromberg), Sussman Family Fund (J. Bromberg), Lerner Foundation (J. Bromberg), The Beth C. Tortolani Foundation (J. Bromberg and D. Lyden), MSK Cancer Center Support Grant/Core Grant (P30 CA008748 to J. Bromberg), NIH (U01-CA169538 to D. Lyden), The Manning Foundation (D. Lyden), The Hartwell Foundation (D. Lyden), Fundacao para aCiencia e a Tecnologia (D. Lyden), The Nancy C and Daniel P Paduano Foundation (D. Lyden), The Mary Kay Foundation (D. Lyden), Pediatric Oncology Experimental Therapeutic Investigator Consortium (POETIC; D. Lyden), James Paduano Foundation (D. Lyden), Malcolm Hewitt Weiner Foundation (D. Lyden), Theodore A Rapp Foundation (D. Lyden), American Hellenic Educational Progressive Association 5th District Cancer Research Foundation (D. Lyden). C. Savini won a Marco Polo fellowship from the University of Bologna. M. Bonafè is supported by the Cornelia and Roberto Pallotti Legacy.

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.