Sustained conditional knockdown reveals intracellular bone sialoprotein as essential for breast cancer skeletal metastasis.

Increased bone sialoprotein (BSP) serum levels are related to breast cancer skeletal metastasis, but their relevance is unknown. We elucidated novel intracellular BSP functions by a conditional knockdown of BSP. Conditional MDA-MB-231 subclones were equipped with a novel gene expression cassette containing a tet-reg-ulated miRNA providing knockdown of BSP production. These clones were used to assess the effect of BSP on morphology, proliferation, migration, colony formation and gene expression in vitro, and on soft tissue and osteolytic le-sions in a xenograft model by three imaging methods. BSP knockdown caused significant anti-proliferative, anti-migratory and anti-clonogenic effects in vitro (p<0.001). In vivo, significant de-creases of soft tissue and osteolytic lesions (p<0.03) were recorded after 3 weeks of miRNA treatment, leading to complete remission within 6 weeks. Microarray data revealed that 0.3% of genes were modulated in response to BSP knockdown. Upregulated genes included the endoplasmic reticulum stress genes ATF3 and DDIT3, the tumor suppressor gene EGR1, ID2 (related to breast epithelial differentiation), c-FOS and SERPINB2, whereas the metastasis associated genes CD44 and IL11 were downregulated. Also, activation of apoptotic pathways was demonstrated. These results implicate that intracellular BSP is essential for breast cancer skeletal metastasis and a target for treating these lesions.


INTRODUCTION
Bone metastasis is an unfavorable event occurring in up to 70% of patients with advanced breast cancer. Severe bone pain, pathologic fractures, spinal cord compression, alteration of hematopoiesis by bone marrow infiltration and hypercalcemia are skeletal-related events caused by this malignant complication [1,2]. Altogether, they have a decisive impact on the quality of life, morbidity and mortality of respective cancer patients [3]. Until today, metastasis renders breast cancer incura ble, but deciphering the molecular mechanism of this process will help to improve the prognosis of affected patients.
Bone sialoprotein (BSP) has been associated with prognosis of breast cancer pa tients, since elevated serum levels were marker of subsequent bone metastasis and highly related to poor survival [4,5]. BSP is a sialic acidrich, phosphorylated glycopro tein, which is secreted and part of the non-collagenous extracellular organic matrix in human bone [4,6]. It has an apparent molecular mass of 70-80 kDa including three poly glutamic acid domains, which confer hydroxyapatite-binding abilities [2,5] and belongs to the small integrin-binding ligand N-linked glycoprotein family of proteins, which are clustered on human chromosome 4. Members of this family undergo a high degree of posttranslational modifications, which vary for a given protein in time (cellular differ entiation) and space (tissue), and directly affect their biological functions [7]. Physiologi cally, BSP is synthesized by osteoblasts, osteoclasts, osteocytes and chondrocytes [2] and contains an Arg-Gly-Asp (RGD) sequence, which is a common recognition site for integrins [8] such as alphav-beta3 and alpha-v-beta5. The interaction of integrin alpha-v-beta3 with the RGD motif of BSP is related to cell attachment and plays a fundamental role in allowing BSP to engage with endothelial cells, osteoclasts and tumor cells [2]. BSP was found in primary malignancies such as breast [5,9], prostate [10] and thyroid cancers [11]. Specifically, the BSP-mediated interaction between tumor cells and bone tissue was suspected to participate in the pathogenesis of bone metastasis [2,5,12,13]. It is assumed that the secretion of BSP from tumor cells provides a selective advantage for their survival via binding to alpha-v-beta3 integrin and factor H, which protects them from complement -mediated lysis [13].
Based on these considerations, we hypothesized that BSP could serve as valuable target for anti-metastatic therapy [14,15]. To study the effects resulting from the inhibition of BSP synthesis and secretion, we generated MDA -MB -231 breast cancer cell clones containing a tet-regulated expression cassette with a specific miRNA against BSP and used Flp-recombinase-mediated cassette exchange (RMCE) for introducing the gene cassette into a well-defined ge nomic locus. Through this combination of the tetracycline-dependent gene regulation system with RNA interference (RNAi), tem poral control of BSP production is estab lished, avoiding compensatory changes on gene expression provoked by long last ing, persistent BSP inactivation. In addition, targeted insertion of the respective con ditional expression cassette into a predetermined locus warrants the integrity and persistence of MDA-MB-231 clonal properties. Respective cell lines were used to investigate the influence of BSP knockdown on cellular functions in vitro and on tumor growth in vivo. In addition, microarray data for ex pression profiling were yielded to identify signaling cascades, which are modulated by conditional, long lasting knock down of BSP and contribute to its mechanism of action.

Generation of cell clones
To characterize the role of BSP in breast cancer, the cell line MDA -MB -231 was used. Our parental cell clone (Figure 1Aa) constitutively expresses the tetracycline-controlled transactivator tTA (Tet-Off system) and contains an expression cassette, in which the bidirectional tet-regulated promoter P tet-bi effi ciently controls production of the reporter genes mStrawberry and firefly luciferase. Moreover, the tet-responsive gene expression cassette is situated in a genomic locus conferring silent but activat able tet-regulatory properties; it is flanked by Flp recom binase consensusse quences, allowing site-directed exchange of the gene cassette, thus conserving all beneficial properties of the genomic locus and avoiding disadvantages associated with random genomic integration. Conditional BSP cell clones were generated in a two-step procedure of recombinase-mediated cassette ex change (RMCE; Figure 1A) according to Weidenfeld et al. [16]. Resulting double-trans genic MDA-MB-231 cell lines contain a tet-regulatory gene expression cassette in which P tet-bi regulates mCherry and firefly luciferase gene expression. Within cell clones interfering with BSP produc tion an artificial miRNA targeting BSP (designed according to the miRNA3 design described in Berger et al. [17]), is located within the artificial intron preceding mCherry. In absence of doxycycline, tTA drives the expres sion of reporter and the miRNA genes, whereas, when doxycycline is added, the drug will attach to the tTA protein, inducing a conformational change that prevents its binding to tet-regulated promot ers. Genomic control and miRNA containing cell clones were generated as described in Figure 1A. The cell clones with conditional (doxycycline dependent) expression of miRNA targeting BSP were used for elucidat ing BSP functions and assessing whether the selected protein is an appropriate tar get for antimetastatic therapy.

Characterization of cell clones
Doxycycline-dependent regulation of reporter gene mCherry was evaluated in two cell clones (designated B3 and B7) by flow cy tometry analysis. As shown in Figure  1B, the cell clones display no mCherry produc tion when tet-regulated gene expression is switched off (+dox), whereas mCherry signals increase about two or ders of magnitude when tet-regulated gene expression is switched on (-dox). Next, the BSP expression was analyzed at protein level for control and miRNA containing cell clones by western blot following 6 days of cultiva tion in media with or without doxycycline (Figure 1 C). Here, a highly significant reduction of BSP production, ranging from 50 to 73%, was detected upon activation of tet-regulated expression (-dox) of the miRNA targeting BSP (

Targeting extracellular BSP by anti-BSP antibody
To investigate the possibility whether a lack of secreted extracellular BSP would be responsible for the observed effect on cell proliferation, MDA-MB-231 cells were exposed to an antibody against BSP (80 to 320 µg/ ml) for up to 4 days. As shown in Supplementary Figure  1A exposure to this anti-BSP antibody had no effect on their proliferation. parental cell clone (a) was transfected with a plasmid containing the gene cassette for hygromycin resistance and thymidine kinase (Hyg-TK), then a hygromycin resistant cell line (b) was selected for further use. Upon gancyclovir treatment, the Hyg-TK cassette of (b) was exchanged by a cassette, in which the tet-dependent bidirectional promotor P tet-bi controls the simultaneous ex pression of reporter genes firefly luciferase and red fluorescence protein mCherry as well as the miRNA targeting BSP (c). A control cell clone without miRNA (d, termed Ic1) was generated using the same strategy. Abbreviations used: hPGK -human phosphoglycerate kinase promoter; tTAtetracycline-controlled transactivator; IRES -internal ribosome entry site; eGFP -enhanced green fluorescent protein; luciferase -firefly luciferase; P tet-bi -bidirectional tet-regulated promoter; strawberry -red fluo rescent protein strawberry; HygTK -hygromycin/ thymidine kinase selection marker gene conferring resistance to hygromycin and sensitivity to ganciclovir; pCMV -cy tomegalovirus promoter; FRT and F3 -wild type and mutant Flp -recombinase target sites, B. and C: Regulation of transfected genes by doxycycline. B. Flow cytometry analysis of two double transgenic cell clones; A distinct increase in mCherry ex pression is observed in response to cultivation of cells in media without doxycycline; Q1-Q4 indicate quadrants 1-4; eGFP -enhanced green fluorescent protein; mCherry -red fluorescent protein mCherry; C. Western blot analysis of BSP expression in control (parent and Ic 1 clones) and two specific cell clones following six days of cultivation in media with or without doxycycline; distinct inhibition of BSP production is observed in response to cultivation of cells in media without doxycycline; numbers correspond to the following experi mental conditions: 1-parent cell clone in medium with doxycycline (+dox control), 2-Ic1 cell clone in medium with dox (+dox control), 3-B3 cell clone in medium with dox (+dox specific control), 4-B7 cell clone in medium with dox (+dox specific control), 5 -parent cell clone in me dium without dox (-dox control), 6-Ic1 cell clone in medium without dox (-dox control), 7-B3 (miBSP) cell clone in medium without dox (-dox anti-BSP effects), 8-B7 (miBSP) cell clone in medium without dox (-dox anti-BSP effects); D. Densitometric analysis of the respective bands from Figure 1C, based on the loading control (βactin) and normalization of the data.

Modulation of gene expression following BSP knockdown
To identify altered regula tion of genes in response to BSP production inhibition, an expression profiling analysis was performed with cell clones B3 and B7, displaying 73 % and 50% re duction in BSP protein concentration upon expression of its targeting miRNA. Based on a minimum of a 2-fold modulation as cutoff for alterations in gene expression, three days of BSP knockdown in the B3 cell clone led to a change in expression of 41 genes, of which the majority (87.8%) was down-regulated. However, after 6 days of BSP knockdown, 121 genes were modulated and the majority (75.2%) was now upregulated. In the B7 cell clone the gene modulation was examined only after 6 days of miRNA-mediated BSP inhibition and from 35 genes, 71.4% were upregulated. The total number of genes displaying changes in expression was very low compared to all an alyzed mRNAs. For categorizing altered expression levels, we differentiated between genes with significant early (day 3) modulation, which returned to control concentrations by day 6. This change in gene expression was termed "compensated short term" (Supplementary Table 1). Correspondingly, if genes showed no early modulation, but a change in expression on day 6, they were considered as "long term modulated" genes. Finally, there were genes with significant modulation observed at days 3 and 6, and this was considered as "biphasic" for those with changes from decreased to increased levels or vice versa, or as "persistent" for those with continuously increased or decreased levels. These categories were used to differentiate between short lived compensated-from persistent long term modulations under the assumption that the former will have no persistent contribution to the cells fate, as opposed to the latter. For the two time points of expression profiling, the percentage of genes displaying either long term changes (71.2%) or persistent  Table 1). To delineate the most prominent molecular processes influenced by reduced BSP production, a 3-fold change in gene expression was used as a cutoff, which identified 27 misregulated genes in cells of the B3 cell clone and 8 in cells of the B7 clone (Table 1). There was an overlap of 5 genes, which showed 3 fold modulation in both clones, the remaining 25 were more intensively modulated in the B3 clone (n=22) or in the B7 clone (n=3). From those, the majority (18 genes) is involved in transcription pro cesses (ontol ogy classes transcriptional regulators, chromosomes, nuclear RNAs) implicating a major role of BSP on these intracellular events. Identified transcriptional regulators primarily act on either ER stress Western blot confirmed the changes observed for a selection of these genes (DDIT3 (CHOP), ATF3, FOS, ID2 and CD44). In addition, we observed the induction of apoptosis because of BSP inactivation, demonstrated by cleavage of caspases 8, 9, 3, 7 and of PARP ( Figure  3). Collectively, these data indicate that BSP knockdown is associated with specifically altered expression of genes related to breast cancer, ER stress and apoptosis.

Expression profiling following exposure of cell clones to an anti-BSP antibody
Both cell clones were cultivated for 6 days in media containing doxycycline and a rat monoclonal antibody against BSP (100 µg/ml) in comparison with respective cells exposed to doxycycline only. Subsequent microarray analyses of mRNA (both clones) and miRNA (B3 clone) showed that the expression of genes was identical in both clones, regardless of the antibody, when using a 2fold modulation as basis for a significant difference.
Similarly, the miRNA profiles were identical in the B3 clone cells cultivated in doxycycline containing medium with or without the antibody (Supplementary Figure 1B)

BSP knockdown inhibits skeletal metastasis
Having demonstrated that miRNA mediated inhibition of BSP expression leads to a decreased proliferation, migration and colony formation in MDA -MB -231 cells in vitro, we tested the effect of BSP knockdown in an in vivo model of osteolytic metas tasis. Parental control cells showed a steady increase in light emission and presence of osteolytic lesions within 30 days after tumor cell inoculation (Figure 4 A). Animals, which survived longer than 40 days tended to show a plateau of light emission as indication of a balance between proliferation and necrosis of tumor cells. Subsequent to these observations, controls were terminated for ethical reasons when their tumor growth reached a plateau. Here, cell clones B3 and B7 were examined in three groups of rats, respec tively (Table 2). In this experiment, the expression of miRNA targeting BSP for 3 weeks led to a significant reduction of the soft tissue tumors and the osteolytic le sions in B3 and B7 cell clones, respectively, compared to the control groups (p<0.03).
In the second experiment, the B3 cell clone was inoculated in a group of four rats to access the efficacy of a longer persistence of BSP inactivation. The tumors were monitored up to 8 weeks after tumor cell inoculation by BLI, as well as after 5 and 7 weeks by MRI and VCT. In all rats, full remissions of soft tissue and osteolytic lesions were detected (Figure 4 B and C).
Collectively, these data show a high efficiency to treat bone metastasis by inactivat ing BSP production in vivo.

DISCUSSION
Primary breast cancer patients with elevated BSP serum concentrations were found at increased risk of developing skeletal lesions [4]. Therefore, we and others hypothe sized that BSP is a potential target for treating breast cancer induced osteolytic le sions. Here, we developed a cellular model to follow this hypothesis, namely MDA -MB -231 cell lines with the ability for a conditional knockdown of BSP, realized by combining tetracycline-controlled transcription activation ("Tet-Off system") with RNA interference. This approach    allowed for the first time full temporal control over BSP production. Persistent inhibition of endogenous BSP production resulted in a signifi cantly reduced proliferation, migration and colony formation of respective cell clones in vitro. In addition, we demonstrated that reduced BSP concentrations lead to the initiation of apoptosis. The distinct gene modulation observed following impeded BSP production was not paralleled by results obtained from exposing the cells to an antibody. Similarly this antibody had no influence on the proliferation of MDA-MB-231 breast cancer cells (Supplementary Figure  1A). Corresponding to this BSP knockdown effect in vitro, we observed complete remissions of soft tissue and osteolytic lesions in vivo. Thus, the first protracted conditional knock down of its kind revealed that BSP is a vital intracellular factor, which is necessary for tumor cell survival and growth in vivo. This assumption is strengthened by the observation that genes were more intensively modulated in cells with lower BSP concentrations, thus indicating a concentration dependent effect. Of course, only a large number of tumor cell lines will be able to mimic the properties of a certain cancer type as published for colorectal cancer cell lines recently [18]. Therefore future experiments should corroborate our results in other cell lines with affinity to the skeleton. The shift in paradigm from BSP being a secreted protein causally involved in the generation of osteolytic lesions to being an intracellular oncogenic driver of tumor growth was suggested by our technically advanced approach in the MDA-MB-231 clones used. In preceding studies decreased expression of BSP was caused by ASOs, siRNA or shRNA [14,15,19]. In general, the knockdown obtained in response to these types of gene therapy was short lived or less distinct than in this study. As a consequence, only certain cellular functions like migration and invasion were clearly inhibited [15,19]. In addition, there was some reduction in the incidence or size of osteo lytic lesions, but not in soft tissue tumor volume [15,19]. The intensive decrease in BSP expression in our study was achieved by a method, which allows conditional knockdown over a prolonged period. Furthermore, the model is applicable in cultured cells and in whole organisms [17,20]. The established cell clones with conditional miRNA expression were generated by RMCE from a parent cell clone without anti-BSP miRNA, which was used as a control. This procedure not only conserves the appro priate tet-regulatory properties of the parental cell line, but also warrants the genome integrity, since the integration locus of the artificial gene expression cassette is pre determined and not random as with other strategies to generate stable transgenic cell lines. In contrast to studies employing a constitutive reduction in BSP expression [19], our temporal control over BSP production avoids long term adaptation to BSP defi ciency, which is basis for potential survival strategies of the tumor cells.
Our model facilitated also to monitor for the first time the genome wide modulation of expression in response to inhibition of BSP production. These results are first steps in elucidating the mechanism of action of BSP knockdown leading to the induction of apoptosis, the inhibition of proliferation, migration and colony formation in vitro, and the anti-metastatic effects in vivo. From our microarray data, the exact trigger has not been determined yet. However, the increased expression of the tumor suppressor gene EGR1 (Table 1) is causal for the increase in the activating transcription factor 3 (ATF3), for EGR1 binds to the promoter region of ATF3 [21]. ATF3 is known for its role in endoplasmic reticulum (ER) induced stress [22]. In this signal cascade, increased expression of ATF4 leads not only to increased expression of DDIT3 (CHOP, GADD153, see Table 1 and Figure 3), but also to increased expression of ATF3 (Table 1 and Figure 3). This factor in turn will decrease the expression of ID1, which exerts anti-apoptotic effects [23] and will cause increased expression of its paralog ID2, as confirmed by microarray and western blot (Table 1and Figure 3), which is essential for the differentiation of breast epithelial cells. The combination of increased EGR1 levels and ER stress is known to induce apoptosis [22,[24][25][26][27][28].
The reduced level of EGR1 before knockdown of BSP corresponds to the observation that cancer cells often become resistant to apoptosis. However, in cells with decreased BSP concentrations, we observed induction of apoptosis, as shown by cleavage of caspases 8, 9, 3, 7 and of PARP (Figure 3), which are well-known indicators for programmed cell death [29]. Furthermore, we detected upregulation of genes that are downregulated in the prima ries of breast cancer patients, including c-FOS, EGR1 and ID2 [30]. These proteins are transcription factors, the first two being potential tumor suppressor genes and the third promoting breast epithelial cell differentiation [24,31,32]. We detected a 11fold in crease in c-FOS mRNA concentration and even stronger stimulation of its expression at protein level. However, the increase in c-FOS expression is not supposed to lead to increased formation of AP-1 transcription factor, since transcription of c-JUN is elevated only 1.4fold upon BSP inactivation. Therefore, c-FOS may act as a regulator of cell growth that is dependent on cell types, stimulation signals, differentiation stages, and cell cycle states [33]. In addition, published data indicate that an elevated level of c-FOS expression could precede the initiation of apoptosis in various tissues, including the prostate [31,34]. EGR1 inhibits angiogenesis [35], suppresses breast cancer and other tumor cells' growth in vitro and in vivo [36,37] and was significantly reduced in some cancers including breast can cer [36][37][38][39][40][41]. A decrease in EGR1 expression may play an important role in the development of breast cancer and serve as biomarker [24].
Additionally, we detected increased ID2 (inhibitor of DNA binding 2, dominant nega tive helix-loop-helix protein) mRNA and protein concentrations. ID2 promotes differ entiation of breast epithelial cells and its reduced expression in breast cancer is as sociated with an unfavorable prognosis [32].
Furthermore, we found that IL11 and CD 44, genes that are strongly related to me tastasis formation, were decreased in response to BSP knockdown. Interleukin 11 is an osteoclast-mobilizing factor, which enables breast cancer cells to establish osteo lytic metastasis in bone [42]. CD 44 is a cell-surface glycoprotein, which participates in a variety of cellular functions including tumor metastasis and is involved in many can cers, including breast cancer [43].
Finally, the urokinase-type plasminogen activator inhibitor SERPINB2 was elevated in response to BSP knockdown, which is known to be a factor of good prognosis in breast cancer patients [44].
In conclusion, the MDA -MB -231 cell clones generated by RMCE have allowed us to investigate the effect of a conditional, prolonged knockdown of intracellular BSP. Besides the clear anti-proliferative, anti-migratory and anti-clonogenic effects ob served in vitro, a curative effect was seen in rats with osteolytic lesions. These effects are probably related to the induction of apoptosis caused by modulation of ATF3 and DDIT3 (CHOP) expression in response to lowered intracellular BSP levels. This assumption needs to be confirmed in future experiments. In addi tion, transcription factors known as tumor suppressors or genes related to breast cancer including c-FOS, EGR1, ID2 and SERPINB2, as well as suppression of me tastasis associated genes (CD44 and IL11) have emerged as molecular targets of BSP knockdown. These results show that a decrease in intracellular BSP levels is likely a key mechanism for suppression of breast cancer induced skeletal lesions.

Cell culture
Parental MDA -MB -231 cells and subclones were cultured in RPMI 1640 medium (Invitrogen, Germany) supplemented with 10% fetal calf serum (FCS), 100ng/ ml doxycycline (added to suppress tet-regulated gene expression), 2mM L-glutamine, 100 U/ml penicillin and 100µg/ ml streptomycin (Invitrogen, Germany). The cell lines were maintained in standard cell culture flasks (TPP, Switzerland) in a humidified incubator at 37˚C and 5% CO 2 . They were authenticated using Multiplex Cell Authentification by Multiplexion (Heidelberg, Germany) as described recently [45]. The sin gle nucleotide polymorphism (SNP) profiles matched known profiles.

Recombinase mediated cassette exchange (RMCE) and generation of stably trans fected cell lines
The RMCE targetable MDA-MB 231 cell line, stably transfected with hPGK-tTA-IRES-eGFP and pCMV-Hyg.TK gene cassettes (Figure 1), was cultured for 5 days in RPMI 1640 with 700 µg/ ml hygromycin B and 100 ng/ml doxycycline. Then, the hy gromycin-resistant cells were further maintained for 3 days in doxycy clinecontaining medium, seeded into six-well plates at a density of 8 x 10 4 and cul tured for 48h be fore transfection.
For plasmid transfection, X-tremeGENE 9 DNA Transfection Reagent (Roche, Ger many) or Lipofectamine 2000 (Invitrogen, Germany) were used according to the manufacturers` recommendations. Transfected cells were maintained in RPMI 1640 medium for 18h, also containing puromycin (1µg/ml) and doxycycline (100ng/ml) to keep tet-regulated expression turned off. Subsequently, cells were detached by tryp sin/ EDTA and transferred into 10cm dishes with standard dox-containing medium. After 24h, fresh medium was added containing ganciclovir (65 µM) for negative se lection, which was maintained for 10 days. Following 4 and 6 days of cultivation in medium without doxycycline, the transfection efficiency of different pools was checked by FACS analysis. As shown in Supplementary Figure 4, the highest per centage (48.6%) of successfully transfected cells was obtained after using Xtreme GENE 9 transfection reagent and the vector expressing Flpo recombinase. Cells showing highest transfection efficiency were used for single cell deposits in 96-well plates (Sarstedt, Germany). When individual clones had reached sufficient cell num bers, they were characterized for optimal regulation of marker gene activity in re sponse to doxycycline.

Fluorescence microscopy and flow cytometry analysis
Selected cell clones were in dividually maintained for 6 days in two small culture flasks (25cm 2 ) in absence or presence of doxycycline. Pictures were taken using a fluores cence microscope (Nikon ECLIPSE TE 200, Japan) with a Nikon digital sight DS-Fi1 camera. For flow cytometry analysis, the cells were harvested by trypsin/ EDTA, re suspended by PBS (Dul beco´s Phosphate-Buffered Saline) and transferred into 5ml tubes with cell-strainer cap (Becton Dickinson Labware, USA). The FACS analyses were performed on a LSR II of the Flow Cytometry Core facility (DKFZ). They were used to ensure purity (> 98.5%) of successfully transfected cell clones and as bench mark for later com parisons.

Cell proliferation, migration and colony formation assays
Proliferation of tumor cells was quantified by MTT assay. Here, 2 -3 x 10 3 cells/ well were seeded into 6-well plates and cultivated for 6 days in media with or without dox. The assay was per formed as described [48]. For determining the influence of extracellular BSP on proliferation, MDA-MB-231 cells were exposed to a rat monoclonal antibody against BSP (Immundiagnostik, Bensheim) for 4 days, followed by MTT assay. For migration and colony formation as says, control and spe cific cell clones with the miRNA targeting BSP were seeded for 6 days in culture flasks with or without doxycycline containing media. For migration assays, the bottom layer of each well in 24 -well plates consisted of 700 µl RPMI 1640 medium with 10 % FCS. Then, the cells were harvested by trypsin/ EDTA and re-suspended in media without FCS and seeded (1 x 10 5 ) into hanging cell culture inserts with 8 µm pore size membranes (Millicell, Millipore, Switzerland), which were transferred onto the prepared wells. After 24h, the inserts with nonmigrated cells were transferred onto new wells, containing fresh medium with 10 % FCS. The mi grated cells were quanti fied using Cell Titer Blue Reagent (Promega, Mannheim, Germany), according to the manufacturer`s protocol. Following an incubation period of 4h at 37˚C, the fluores cence was measured by a fluorescence reader (Synergy 2, Biotek, Germany) with excitation (560/15) and emission (590/20) filters. The tumor cell migration was fol lowed and quantified for 2 subsequent days. For colony for mation experiments, semi-solid mixtures (0.8% methylcellulose, +/ -doxycycline, 40% FCS), containing cell suspensions (1.25 x 10 3 ) were carefully mixed and ali quots (400µl) were dispensed into four wells of 24 well plates. Then, they were incu bated under usual cell culture conditions (37˚C, 5% CO 2 in humidified air) and colo nies (> 60 cells) were counted under an inverted microscope after 7 days.

Animals
Male nude rats (RNU strain, Charles River, Germany) were obtained at an age of 4 to 6 weeks and kept under specific pathogen free (SPF) conditions in Mac rolon-IV-cages of a ventilated rack (Ventirack, UNO Roestvaststaal B.V., Zevenaar, The Netherlands) providing a 50-fold exchange of filtered air per hour as well as pos itive air pressure inside the cages. Constant room temperature (22 ± 1˚C), air humid ity (50 ± 10%) and dark-light-rhythm (12h) were maintained throughout. The animals had free access to autoclaved water and standard laboratory diet. After an adaptation period of a week, experiments were started. All animal experiments were approved by the responsible governmental animal ethics committee (Regierungspräsidium Karlsruhe, Germany).

Ethics Statement
The investigation involving animals has been conducted in accordance with the ethical standards and according to the Declaration of Helsinki, and according to national and international guidelines and has been approved by the authors' institutional review board. www.impactjournals.com/oncotarget

Tumor cell inoculation
Three days prior to tumor cell inoculation, the rats received doxycycline via the drinking water (2µg/ml), which was supplemented with 240mg sodium cyclamate and 24mg saccharin-sodium per liter.
Cell clones with conditional expression of miRNA targeting BSP were cultivated in media with doxycycline. Subconfluent cells were harvested using trypsin/ EDTA, washed by PBS, subsequently counted and resuspended in PBS to a concentration of 1 x 10 5 cells per 100µl. For tumor cell implantation, rats were anaesthetized with isoflurane (1-1.5 vol.%) in air. The cells were injected into the saphenous artery ac cording to the procedure detailed previously with slight modification [49].

Set up of animal experiments and tumor size determination
Initially, the two clones B3 and B7 were tested in three groups of rats (n=3, respec tively), receiving continuously doxycycline (2µg/ml) for one, two or four weeks in au toclaved water. The ex periment was terminated 4 weeks after tumor cell inoculation. Then, to assess late therapeutic effects, the B3 cell clone was examined in up to 8 weeks in animals (n=4), which had ingested doxycycline via the drinking water for 1 week at 2µg/ ml. Bioluminescence imaging (BLI), magnetic resonance imaging (MRI) and volume computed tomography (VCT) were used to follow the tumor size in the experimental animals [49].

Microarray analysis
Microarray analysis was accomplished as described [48]. Microar ray scanning was done using an iScan array scanner. As test for significance, the student's t-test was used on the bead expression values of the two groups of interest. The average ex pression value was calculated as mean of the measured expressions of beads to gether with the standard deviation of the beads. Modulations in gene expression were considered significant, if the p-value corrected by the Benjamini-Hochberg procedure, was lower than 0.01. This was observed for all modulations exceeding an at least 2fold change.
Entry name/accession number: GSE55432 in the Gene Expression Ombnibus (GEO) database. miRNA Profiling miRNA profiling was performed on RNA extracted from cells of the B3 cell clone, which had been maintained for six days in media with doxycycline and an anti-BSP antibody (100µg/ ml) or in doxycycline containing media only. To that purpose, fluorescently-labeled miRNA was prepared according to the Agilent protocol "miRNA Complete Labeling and Hyb Kit". Labeled miRNA samples were hybridized for at least 20h at 55°C on respective chips (Agilent human miRNA Microarray Release 19.0, 8x60k). Gene Expression Microarrays were scanned using the Agilent Scanner G2505C. GeneView raw data from the scanner were analyzed with the program language R, which gives one aggregated expression value for each miRNA ID. Statistical tests (t-tests) were performed over all samples of a group in linear scale. Benjamini-Hochberg correction was applied over all p-values of the differential expression analysis.

Statistics
The results of multiple measurements from in vitro functional tests and in vivo ex periments were given as mean with corresponding standard deviation. The two-way analysis of variance (ANOVA) test, Bonferroni-posthoc test and Wilcoxon test (GraphPad-Prism5) were used to examine for independent occurrence of investi gated parameters (in vitro functional properties and growth in vivo). A p-level less than 0.05 was considered significant.