Autonomous Stimulation of Cancer Cell Plasticity by the Human NKG2D Lymphocyte Receptor Coexpressed with Its Ligands on Cancer Cells

The stimulatory NKG2D receptor on lymphocytes promotes tumor immune surveillance by targeting ligands selectively induced on cancer cells. Progressing tumors counteract by employing tactics to disable lymphocyte NKG2D. This negative dynamic is escalated as some human cancer cells co-opt expression of NKG2D, thereby complementing the presence of its ligands for autonomous stimulation of oncogenic signaling. Clinical association data imply relationships between cancer cell NKG2D and metastatic disease. Here we show that NKG2D promotes cancer cell plasticity by induction of phenotypic, molecular, and functional signatures diagnostic of the epithelial–mesenchymal transition, and of stem-like traits via induction of Sox9, a key transcriptional regulator of breast stem cell maintenance. These findings obtained with model breast tumor lines and xenotransplants were recapitulated by ex vivo cancer cells from primary invasive breast carcinomas. Thus, NKG2D may have the capacity to drive high malignancy traits underlying metastatic disease.


Introduction
Cancer cell plasticity entails the development of traits enabling cancer cells to dissociate from primary tumor, disseminate, and expand clonally at distant sites. This process is regulated by the epithelial-mesenchymal transition (EMT) and the interrelated acquisition of regenerative cancer stem cell (CSC) attributes [1,2]. Known drivers of cancer cell plasticity include heterotypic cues from tumor-associated stromal and/or immune system cells [1]. We previously identified an unconventional homotypic receptorligand interaction on cancer cells [3] and show here that resultant signaling induces reprogramming towards migratory and stem-like capacities.
The receptor involved, NKG2D (natural killer group 2 member D), is an activating lymphocyte receptor mainly on NK cells and CD8 T cells and is best known for mediating immune surveillance of virally infected and malignant cells [4]. Human NKG2D signals via the DAP10 (DNAX-activating protein 10) adaptor, which binds either PI3K (phosphoinositide 3-kinase) or Grb2 (growth factor receptor-bound protein 2), thus activating PKB/AKT (protein kinase B) or MAP (mitogen-activated protein) kinase cascades [5]. Ligands for NKG2D in humans include MICA and MICB (MHC class I-related chains A and B) and six members of the ULBP (UL-16 binding protein) family [6]. NKG2D ligands are largely absent from the surface of normal cells but can be induced by oncogenesis-associated stress responses in cancer cells [7]. This selective ligand expression enables NK cells and CD8 T cells to target cancer cells, at least at early tumor stages before immunosuppressive tactics of progressing tumors stifle this arm of the immune response [4,8].
In addition to counteracting immune responses, some cancer cells co-opt NKG2D for their own benefit, complementing the presence of its ligands for self-stimulation of tumorigenesis [3]. Variable proportions of breast, ovarian, prostate, and colon cancer cells express signaling proficient NKG2D-DAP10 complexes, which activate the PI3K-AKT-mTOR (mammalian target of rapamycin) signaling axis and downstream effectors. Moreover, as in lymphocytes, NKG2D-DAP10 stimulates phosphorylation of ERK (extracellular signal-regulated kinase) and JNK in MAP kinase cascades [3]. Pathophysiological significance of NKG2D-DAP10 signaling is supported by a clinical association study that established positive correlations between percentages of cancer cells with surface NKG2D and tumor size and spread [3]. These relationships were extended by significant associations with lymph node metastasis, and by trend correlations with grade and lymphovascular invasion, suggesting NKG2D-DAP10 effects in tumor cell dissemination and metastasis formation [3]. The present study addresses the capacity of NKG2D-DAP10 to promote cancer cell plasticity underlying metastatic disease.

Induction of EMT reprogramming by ligand stimulation of NKG2D
Epithelial tumor lines typically express NKG2D ligands but are either negative for their NKG2D-DAP10 receptor or, as with the MCF-7, BT-20, and MDA-MB-453 breast cancer lines, scarcely positive as reflected by minimal shifts of flow cytometry profiles and very low NKG2D and DAP10 mRNA and protein expression (Figure S1A and S1B; also refer to Figure 2B and Figure 2C in references 3 and 9, respectively). This paucity of the receptor in tumor lines is opposed to relative abundance, both by mRNA and protein expression, on positive ex vivo cancer cells (Figure S1C-E; also refer to Figure 1C-E in reference 3). To test in an in vitro model whether NKG2D induces EMT, we thus examined MCF-7 cells that were stably transfected with NKG2D-DAP10 (MCF-7-TF cells) resulting in surface expression at levels similar to ex vivo cancer cells (compare Figure S1C and Figure S1F). By phase contrast microscopy, MCF-7-TF cells displayed morphological transformations in comparison to mock-transfected control cells, which exhibited tightly clustered cobblestone-like shapes. MCF-7-TF cells, in contrast, displayed fibroblast-like morphologies ( Figure 1A). These changes were due to ectopic expression of NKG2D-DAP10 as the parental phenotype was restored by recombinant lentivirus-mediated RNAi targeting of NKG2D in MCF-7-TF-KO cells ( Figure 1A; for flow cytometry profiles of these model lines see Figure S1A and S1F). These observations suggested that ligand-mediated stimulation of NKG2D resulted in induction of EMT, which involves coordinated molecular and cellular changes leading to loss of epithelial cell-cell adhesion, polarity, and cytoskeletal integrity, concomitant with acquisition of mesenchymal protein signatures, spindle-cell shapes, and invasive and migratory abilities [10,11]. Diagnostic of EMT are reduced expression of the cell junction-associated E-cadherin, and induction of N-cadherin and the cytoskeletal intermediate filament Vimentin. By immunofluorescence microscopy, MCF-7-TF cells displayed those changes in epithelial and mesenchymal marker proteins, which were reversed by RNAi targeting of NKG2D ( Figure 1A). Corresponding results were obtained by immunoblot and RT-PCR profiling ( Figure 1B). Extension of data to additional marker proteins and their corresponding mRNAs, including the epithelial tight junctional zona occludens-1 (ZO-1) and occludin, cytokeratin 19 (CK19) and mucin 1 (MUC1), and fibroblastoid asmooth muscle actin (a-SMA), yielded conforming results ( Figure 1A and 1B) [12]. EMT is orchestrated by transcription factors that include Snail1/2, Twist1/2, Zeb1/2, and LEF-1 [1]. Induction of the mRNAs for all of those factors except for Zeb1 was recorded with MCF-7-TF but not MCF-7-TF-KO cells ( Figure 1B). Preincubation of MCF-7-TF cells with a cocktail of antibodies to relevant NKG2D ligands inhibited all recorded protein and mRNA changes, thus confirming ligand involvement ( Figure 1B) [3]. Our previous study established the requirement of cell-cell contact for ligand-mediated stimulation of NKG2D-DAP10 signaling in cancer cells [3]. Accordingly, as seen by flow cytometry, the proportions of surface E-cadherin 2 /N-cadherin + mesenchymal MCF-7-TF cells correlated with cell culture confluency ( Figure 1C).
EMT-associated remodeling of cell-cell and cell-matrix adhesion endows cancer cells with migratory and invasive capabilities [10]. We tested for increased motility of MCF-7-TF cells in standard assays scoring migration through microporous membrane or traversion of reconstituted basement membrane (Matrigel). By comparison to the MCF-7 mock-transfected control, these experiments revealed about six-and 12-fold increases in migratory and invasive activities, respectively, of MCF-7-TF cells, which were suppressed in the NKG2D-depleted MCF-7-TF-KO cells ( Figure 1D and 1E). For an alternative approach, scrape wound closure by confluent monolayers of MCF-7-TF and control cells was imaged by phase contrast microscopy. Whereas MCF-7-TF cells achieved complete wound closure within 72 h, little if any change was seen with the negative control cells ( Figure 1F). Altogether, these results completed the in vitro studies of MCF-7-TF cells suggesting that NKG2D-DAP10 has the capacity to activate cancer cell EMT. MCF-7 is a luminal breast cancer line with epithelial features as well as some constitutive epithelial to mesenchymal plasticity, which is typical of most breast tumor lines [13,14]. Hence, Snail1 and Snail2 mRNAs were detectable by high-cycle (36 rounds) RT-PCR in untransfected MCF-7 parent cells ( Figure 1G). Consistent with an involvement of endogenous NKG2D in Snail1/2 induction, NKG2D depletion in MCF-7-NKG2D RNAi cells led to reduction or loss of these transcripts as compared to scrambled RNAi control transductants (MCF-7-scrRNAi cells; Figure 1G; for flow cytometry profiles of these model lines see Figure S1A and S1G). Endogenous NKG2D may thus in principle have a capacity to promote differentiation towards mesenchymal signatures but these effects may not penetrate due to its scarce expression. Accordingly, presence or lack of endogenous NKG2D expression in breast cancer lines, such as the epithelial MCF-7 (NKG2D + ) or mesenchymal SUM149PT (NKG2D -), is unaligned with their mainly epithelial versus mesenchymal representations [3,9,14,15].
With all four tumor lines, ectopic NKG2D-DAP10 expression (constitutive or Dox-induced) imprinted the morphological (shown only for MCF-10AT-TF cells), marker protein, and transcriptional EMT signatures recorded with MCF-7-TF cells ( Figure S3A-D). Moreover, flow cytometry of Dox-induced MCF-10AT-TF cells for surface E-cadherin and N-cadherin confirmed that activation of EMT was cell contact-and hence presumably ligand-dependent ( Figure S3E). All tumor lines with constitutive or Dox-induced ectopic NKG2D-DAP10 expression exhibited markedly increased in vitro migratory and invasive activities ( Figure S3F).  NKG2D was expressed in MCF-10AT cells by transduction together with DAP10 mutated at either its PI3K/p85 (M88Q*) or Grb2 (N87Q*) binding site [20]. After confirmation of proper expression and function of the variant NKG2D-DAP10 complexes ( Figure 3A and 3B), testing for induction of EMT parameters ascertained their dependence on recruitment of PI3K/p85 but not of Grb2 ( Figure 3C-E).

Association of NKG2D-DAP10 with EMT signatures of ex vivo cancer cells
Signaling proficiency of the NKG2D-DAP10 receptor in breast cancer cells has been documented [3] and was confirmed here with two additional breast cancer specimens ( Figure S1H). Antibody-mediated receptor crosslinking induced AKT phosphorylation downstream of PI3K in breast cancer cells sorted for absence of CD45 (hematopoietic cell exclusion), expression of EpCAM (cancer cell inclusion), and NKG2D. Appearance of phospho-AKT (P-AKT) was sensitive to Ly294002 and thus dependent on PI3K. To ascertain biological relevance of NKG2D-DAP10 in EMT reprogramming, freshly isolated cell suspensions from 12 primary invasive breast cancer specimens (referred to as BT1 to BT12) were examined by multi-parameter surface flow cytometry for CD45, EpCAM, NKG2D, and the Ecadherin 2 /N-cadherin + EMT signature. EpCAM is a bona fide cancer cell marker although downregulation may occur during EMT [12,21]. Indeed, EpCAM expression among CD45cell populations was heterogeneous in all but one (BT8) of the 12 tumor cell suspensions ( Figure 4A). We thus examined CD45 -Ep-CAM high and CD45 -EpCAM low cells separately for surface NKG2D and E-cadherin/N-cadherin patterns. Consistent with our earlier findings [3], NKG2D + cancer cells were present in all 12 breast cancer specimens, ranging between 0.5 and 27.8% (mean 9.7 +/2 9.1%) of total CD45 -EpCAM high cells (Table S1). EMT represents a continuum with epithelial/mesenchymal hybrid and fully transitioned mesenchymal-like states [10,11,21]. Based on our findings with the model tumor lines, cancer cell NKG2D should thus be associated with both hybrid (E-cadherin + Ncadherin + ) and further differentiated (E-cadherin -N-cadherin + ) phenotypes. In 9 of the 12 tumor cell suspensions tested (BT4 to BT12), most NKG2D + among the CD45 -EpCAM high cells had Ecadherin/N-cadherin patterns consistent with either partial (Ecadherin + N-cadherin + ) or more progressed (E-cadherin -N-cadherin + ) EMT ( Figure 4A and 4B). Except for the BT8, BT11, and BT12 suspensions, skewing towards those phenotypes was not apparent among the matched NKG2Dcancer cells. In three cases (BT1, 2, and 3), cancer cells with mixed or mesenchymal signatures were confined to the NKG2D + subset but were not predominant ( Figure 4B).
NKG2D + cells were also noted among CD45 -EpCAM low tumor cell populations, comprising between 0.3 and 59.3% (mean 12.6 +/2 18.9%) (Table S1). In all but the BT1 sample, NKG2D + CD45 -EpCAM low populations also preferentially displayed EMT phenotypes. Notably, in select samples (BT3, BT5, BT6, and BT12), proportions of mesenchymal (E-cadherin -Ncadherin + ) cells were considerably larger among NKG2D + CD45 -EpCAM low compared to NKG2D + CD45 -Ep-CAM high cells, suggesting that EpCAM downregulation may occur late in the EMT process ( Figure 4B and 4C). Altogether,  these observations corroborated the findings made with the model tumor lines, supporting a prominent role of NKG2D as a natural activator of EMT in cancer. This is consistent with the fact that the NKG2D + cells constituted substantive proportions when recorded as percent of total hybrid and mesenchymal marker profilepositive cells (Table S1). All 12 tumor cell suspensions contained varying proportions of CD45 -EpCAMcells that lacked Ecadherin and N-cadherin and thus could not be assigned to a specific cell type. With the exception of four tumor specimens (BT2, BT3, BT11, and BT12), NKG2D + cells were absent among those populations.

NKG2D induces EMT in a tumor xenotransplant model
Complementary evidence for a role of NKG2D in EMT was obtained from tumors derived from SUM149PT-TF or negative control cells orthotopically xenografted into mammary fat pads of NOD/SCID mice, a model experiment that was instrumental in establishing NKG2D-driven tumorigenicity [9]. More direct in vivo evidence was not attainable since spontaneous or carcinogen-induced cancers in mice lack NKG2D expression [3]. The SUM149PT line is phenotypically heterogeneous, harboring sizeable subsets of cells with mesenchymal marker profiles [14]. Hence, immunohistochemistry may pose difficulties in assessments of EMT-associated changes. Examination of xenograft tumor-derived cell suspensions by flow cytometry exposed strongly increased representations of hybrid/mesenchymal phenotypes in all of five NKG2D + SUM149PT-TF as opposed to NKG2Dmock control tumors ( Figure 5). EMTassociated migratory activity of xenotransplanted SUM149PT-TF cells may have contributed to enhanced tumor cell dissemination previously observed in this animal model [9].

Instruction of stemness reprogramming by NKG2D
In addition to conferring migratory abilities, EMT transcription factors Snail1 and Twist regulate surface marker profiles that define breast cancer cell populations enriched for cells with stemlike attributes [22][23][24]. Consistent with the induction of Snail1 and Twist, ectopic NKG2D-DAP10 expression (constitutive or Doxinduced) was associated with breast cancer stem cell-like CD24 2 / CD44 + profile shifts in the MCF-7-TF, MCF-10AT-TF and SUM149PT-TF lines ( Figure 6A) [22,25]. However, only a small fraction of cells with CSC-like phenotypes qualify as functional CSCs [2]. More relevant to EMT-stemness interrelations may be the fact that EMT transcription factors, such as Snail2 or Zeb1, in cooperation with separate genetic circuitries, also regulate functional breast CSC states [2,[26][27][28]. By microarray gene expression profiling, we identified a prominent induction of the Sox9 transcription factor in MCF-7-TF compared to mocktransfected control cells (see microarray data at http://www.ncbi. nlm.nih/geo/under accession code GSE53961), which was independently confirmed by quantitative RT-PCR and immunoblot ( Figure 6B and 6C).
Sox9 acts cooperatively with Snail2 (Slug) and to a lesser extent Snail1 as a master regulator of the mammary stem cell state [28]. Thus, by positively regulating Sox9 in addition to Snail2 and Snail1 (see Figure 1B), NKG2D may have the capacity to promote functional stem cell-like attributes. This notion was supported by three-dimensional matrigel organoid cultures and classical mammosphere assays. In both types of experiments, MCF-7-TF cells formed numerous large (.100 mm) organoids whereas mocktransfected control or MCF-7-TF-KO cells generated only few cell clusters that were typically small ( Figure 6D). As with EMT reprogramming, organoid formation was dependent on PI3K-AKT with no apparent involvement of ERK or JNK ( Figure 6D). Corresponding evidence for the capacity of NKG2D to induce Sox9 and organoid formation was obtained with Dox-induced MCF10AT-TF cells ( Figure 6B-D).
Sox9 and Snail2 are both required for mammary epithelial cells to enter and stably maintain a functional stem cell state [28]. Accordingly, RNAi-mediated depletion of Sox9, and separately of Snail2, abrogated organoid formation in MCF-7-TF mammosphere cultures [ Figure 6E and 6F). Unlike Snail2, which is thought to contribute to stem cell induction via activation of an EMT, effects of Sox9 in EMT reprogramming are considered minor [28]. Consequently, EMT-associated morphological changes of MCF-7-TF cells were thus reversed by RNAi-mediated depletion of Snail2 but not of Sox9 ( Figure 6G).
The defining hallmark of highest plasticity cancer cells is their efficient tumor initiation upon xenografting in immunodeficient mice. Our earlier mouse model study demonstrated markedly reduced latencies and enhanced tumor incidences with orthotopically transplanted MCF-7-TF versus control lines [9]. Those outcomes are consistent with NKG2D-mediated induction of Sox9 and Snail2 and attendant acquisition of CSC capacities. This induction also provided an explanation for the observed enhanced tumor initiation by untransfected MCF-7 cells expressing minimal endogenous NKG2D ( Figure 1G; for flow cytometry profiles of the parental MCF-7, MCF-7-NKG2DRNAi, and MCF-7-scrRNAi lines see Figure S1A and S1G) [9]. To probe for relevance in human cancers, ex vivo tumor cells from six primary invasive breast cancer specimens (BT13 to BT18) were tested for associations between NKG2D and Sox9 and Snail2 by polychromatic flow cytometry. In all tumors, proportions of CD45 -Ep-CAM + cells co-expressing Sox9 and Snail2 were significantly larger among NKG2D + compared to NKG2Dcells ( Figure 6H). Altogether, these results support tumorigenic significance of NKG2D via Sox9-and Snail2-mediated induction of CSC traits.

Discussion
High plasticity cancer cells are considered main culprits of tumor dissemination and rebounding following conventional cancer therapy [2,29]. The present study indicates that co-opted expression of, and signaling by, NKG2D on cancer cells promotes cancer cell plasticity with differentiation towards mesenchymal phenotypes and dissemination-enabling and tumor-initiating capacities. Physiological significance is supported by the recapitulation of EMT and Sox9 signatures recorded with model tumor lines by primary invasive breast cancers. It is likely that the role of NKG2D as a driver of cancer cell plasticity is broadly applicable, extending to at least ovarian, colon, and prostate carcinoma cells [3].
The relevance of EMT, and by inference of stemness reprogramming, in human tumor development is not uncontroversial [30]. However, recent flow cytometry-based demonstrations of circulating tumor cells with epithelial, mesenchymal, or hybrid signatures, and associated metastasis-initiating capacities, constitute direct evidence for human cancer cell plasticity and its in percent. Quadrant gates are based on fluorescence-minus-one isotype Ig control stainings. (B, C) Graphic display of data derived from 12 breast cancer specimens (BT1-BT12) that were analyzed as in (A). E + N -, E + N + , and E -N + refer to E-cadherin and N-cadherin status. Colors correspond to dot plot quadrants in (A). See main text for further explanation. doi:10.1371/journal.pone.0108942.g004 pathophysiological significance [21,31]. Our multi-parametric single-cell analysis of ex vivo tumor cell suspensions extends this evidence to primary breast cancers and uncovers that mixed epithelial/mesenchymal and mesenchymal-like breast cancer cell phenotypes are more prevalent and presumably more complex than previously thought. Because most NKG2D + breast cancer cell populations examined were skewed towards hybrid and mesenchymal phenotypes and represented substantive proportions of all cells with those differentiation stages, NKG2D may quite possibly have a prominent role in EMT induction. A reverse scenario with NKG2D induction as a result of EMT is unlikely as considerable proportions of hybrid or mesenchymal cancer cells lacked NKG2D and varying subsets of NKG2D + cancer cells retained epithelial marker profiles.
The NKG2D-mediated induction of Sox9 alongside Snail2 (and Snail1) is of particular interest as these factors have been functionally linked in mouse mammary epithelial and human breast CSC biology [28]. Both Sox9 and Snail2 are considered central for induction and maintenance of stem cell traits, with each factor contributing distinct attributes but both being required for stem cell function [28]. Our findings with the MCF-7-TF line and its Sox9-or Snail2-depleted variants are consistent with this functional dichotomy and confirm with human cells the necessity of both factors for induction of functional stem cell-like abilities. NKG2D-mediated activation of major oncogenic signaling pathways is compatible with its ability to induce Sox9 although specific intermediates are unidentified [32,33].
EMT and the regulation of stem cell states are of broader significance due to their involvement in embryogenesis. NKG2D receptor expression in embryonic tissues has not been reported, however, and NKG2D-deficient mice have no overt developmental abnormalities. But at least in mice, some NKG2D ligands are expressed at distinct embryonic stages, mostly in the central nervous system, and are present on subventricular zone-derived neural stem/progenitor cells where they stimulate proliferation and survival [34,35]. Hence, it remains possible that nonlymphocytic NKG2D expression and function during development may have gone unnoticed thus far.
Our results add a provocative twist to current knowledge as cancer cells may co-opt NKG2D as an oncoprotein serving their own benefit. Immune surveillance of advanced tumors is progressively thwarted by NKG2D ligand-mediated negative imprints on lymphocyte effector functions [8,36]. The here described capacity of NKG2D to promote high malignancy traits  strongly implies the possibility that the receptor itself may represent the main factor underlying the typically poor clinical outcomes that have been associated with cancer cell expression of its ligands [37][38][39].
In conclusion, the evidence presented here supports a major role of NKG2D in tumorigenesis via promotion of cancer cell plasticity. Various efforts targeting lymphocyte NKG2D or its ligands for cancer therapy are underway [40][41][42][43]. Those approaches are solely based on knowledge of its immune related functions and may thus be misdirected as underlying assumptions are incomplete.

Materials and Methods
Tumor lines, breast cancer specimens, and cancer cell suspensions  Table S1. Fresh specimens were processed to single-cell suspensions using a Human Tumor Tissue Dissociation Kit and a gentleMACS Dissociator (both from Miltenyi Biotech). SUM149PT-derived xenograft tumors were harvested in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and approved under the active FHCRC Institutional Animal and Use Committee (IACUC) protocol #1870. Xenograft tumor processing was performed as described [9].

Gene expression microarrays
For RNA conversion and biotin-labeling see Materials and Methods S2 in File S1. Biotin-labeled cRNA was processed on a HumanRef8v3 Expression BeadChip (Illumina) and imaged using the Illumina iScan system. Microarray data were assessed for quality followed by quantile normalization using the Bioconductor package lumi [44]. The data set was initially filtered by flagging probes that fell below a signal noise floor, which was established using three standard deviations of the negative control probe signals within each array. Subsequent data set filtering employed a variance filter using the 'shorth' function of the Bioconductor package genefilter. Pair-wise statistical analyses were performed using the Bioconductor package limma [45]. Illumina BeadChips are constructed using probes design with a 39-UTR bias. As such, the probe for KLRK1 encoding NKG2D (chr12:10525415-10525464) targets its untranslated region and therefore fails to bind to the 39-UTR-truncated NKG2D cDNA expressed in the MCF-7-TF line.

Immunofluorescence microscopy
For immunocytochemistry of EMT markers, formaldehydefixed and permeabilized cells (5 min in 0.2% Triton X-100/ PBS) grown on Nunc Lab-Tek II CC2 chamber slides (Fisher) were treated for 1 h with blocking buffer (5% goat serum in SuperBlock; Pierce) and sequentially stained in humid chambers with titrated concentrations of mAb (clones and sources as above) to E-cadherin, N-cadherin, occludin, ZO-1, Vimentin, or MUC1, and secondary antibody (goat anti-mouse IgG-Alexa Fluor 594; Invitrogen). Washed, nuclear counterstained (DAPI; Invitrogen), and coverslipped slides were examined using a Nikon Eclipse E800 microscope. Phase contrast images were taken with a DeltaVision microscope (Applied Precision).
Tumor cell invasion/migration, matrigel organoid cultures, and mammosphere assays BD BioCoat Matrigel Invasion Chambers (Discovery Labware) were seeded and cells cultured, harvested, and counted as described in Materials and Methods S3 in File S1. Migration assays were identical except that Control Inserts (Discovery Labware) were used instead of the Matrigel Invasion Chambers. Cell migration in wound-healing assays was imaged in regular intervals by phase contrast microscopy. Matrigel organoid cultures were as described [29]. For mammosphere assays, cells were cultured in 96-well ultra low attachment plates (1610 3 cells/well) in serum-free MammoCult medium (Stemcell Technologies) with 0.9% methylcellulose (Sigma Aldrich). After 7 days, mammospheres with .100 mm diameters were microscopically counted in three sets of triplicate wells.

Statistical analyses
Numerical data are presented as mean +/2 standard deviation. Student's t test was used to calculate p values, with p,0.05 assigned significance.

Supporting Information
(DOC)