The regulatory effect of hyaluronan on human mesenchymal stem cells’ fate modulates their interaction with cancer cells in vitro

Metastatic spread of cancer cells into a pre-metastatic niche is highly dependent on a supporting microenvironment. Human bone marrow-derived mesenchymal stem cells (bmMSCs) contribute to the tumor microenvironment and promote cancer metastasis by inducing epithelial-to-mesenchymal transition and immune evasion. The underlying mechanisms, however, are incompletely understood. The glycosaminoglycan hyaluronan (HA) is a central component of the extracellular matrix and has been shown to harbor pro-metastatic properties. In this study we investigated the highly disseminating breast cancer and glioblastoma multiforme cell lines MDA-MB-321 and U87-MG which strongly differ in their metastatic potential to evaluate the impact of HA on tumor promoting features of bmMSC and their interaction with tumor cells. We show that adipogenic differentiation of bmMSC is regulated by the HA-matrix. This study reveals that MDA-MB-231 cells inhibit this process by the induction of HA-synthesis in bmMSCs and thus preserve the pro-tumorigenic properties of bmMSC. Furthermore, we show that adhesion of MDA-MB-231 cells to bmMSC is facilitated by the tumor cell-induced HA-rich matrix and is mediated by the HA-receptor LAYN. We postulate that invasive breast cancer cells modulate the HA-matrix of bmMSC to adapt the pre-metastatic niche. Thus, the HA-matrix provides a potential novel therapeutic target to prevent cancer metastasis.


Results
A deeper understanding of the interaction between tumor cells and the tumor microenvironment in the metastatic niche is needed to identify key mechanisms explaining the differences in the adaptation to the 'foreign soil' . Since mesenchymal stem cells are an important factor in orchestrating the bone marrow niche and are at the same time a significant source of extracellular matrix molecules, we investigated the interplay between tumor cells of cancer entities which invariably display tumor cell dissemination but different metastatic behaviors: MDA-MB-321 and U87-MG as model organisms for metastasizing breast cancer and non-metastasizing glioblastoma multiforme, respectively.
The HA matrix is a negative regulator of adipogenic differentiation of mesenchymal stem cells. As a prerequisite for the subsequent experiments, we first determined the presence of HA and its receptor CD44 on the bone marrow-derived MSCs (bmMSCs) utilised in our experiments. As expected, cytochemical stainings revealed a pronounced MSC-derived pericellular HA deposition and a high expression of the HA receptor and stem cell marker CD44 (Fig. 1a).
Within the bone marrow, MSCs mainly differentiate into adipocytes and osteoblasts 17 . To investigate the influence of HA on bmMSC differentiation, bmMSCs were differentiated either into adipocytes or osteoblasts over a period of 28 days with or without simultaneous treatment with 4-methylumbelliferone (4-MU), an inhibitor of HA synthesis. Osteogenic differentiation of MSCs (osteoMSCs) was visualised by staining of Ca 2+ -phosphate depositions with Alizarin Red S (Fig. 1b) and quantified by measuring the Ca 2+ -concentration in the cell layer (Fig. 1c). Inhibition of hyaluronan synthesis did neither lead to spontaneous osteogenic differentiation nor did it affect osteogenic differentiation. For the detection of adipogenic differentiation, lipid vesicles of the adipogenic differentiated bmMSCs (adipoMSCs) were stained with Oil Red O (Fig. 1d), and the area fraction was quantified (Fig. 1e). Inhibition of HA synthesis by 4-MU did not cause spontaneous adipogenic differentiation, even if in some cultures small formations of lipid vesicles could be observed. However, it significantly increased the adipogenic differentiation potential of bmMSCs after chemical stimulation, suggesting a stemness-preserving effect of HA.
Of note, measurement of HA divided individual MSC preparations into two different subgroups (Fig. 2a). We proceeded to investigate whether this difference alters the adipogenic differentiation potential. Indeed, bmMSC with a low amount of HA differentiated more easily into adipocytes (Fig. 2b). Next, we investigated whether extrinsically added HA would reverse this phenotype. Whereas high molecular weight HA (HMW HA) did not change the increased adipogenic differentiation in MSCs with low HA synthetic capacity, low molecular weight HA (LMW-HA) preserved the native MSC phenotype and inhibited adipogenic differentiation (Fig. 2b).
Besides the capability to differentiate into various cell types of the mesodermal lineage, bmMSC exhibit immunomodulatory properties 9 . By using a T-cell proliferation assay, we investigated whether this potential is altered upon adipogenic differentiation. In this experimental setup, αCD3/αCD28 stimulated, CFSE labelled T-cells were seeded on a layer of bmMSC that has been treated for 10 days under various conditions of adipogenic stimulation and/or HA synthesis inhibition. To investigate the maximum proliferative potential of the T-cells, stimulated T-cells were seeded in empty wells (Fig. 2c, 1st bar). The activation of the T-cells was validated by flow cytometry, which showed typical forward scatter/side scatter changes upon activation compared to unstimulated T-cells (Fig. S2). The maximum immunosuppressive potential of the bmMSC was determined by seeding the T-cells on a layer of untreated bmMSCs (Fig. 2c, 2nd bar). T-cells seeded on a layer of pre-adipocytic bmMSCs showed an increased proliferation (Fig. 2c, 3rd bar). This effect was even more pronounced under a simultaneous treatment The HA system is strongly expressed in bone marrow-derived mesenchymal stem cells and influences their differentiation potential. (a) Cytochemical staining of HA (red) and the stem cell marker and HA interacting receptor CD44 (green). Nuclei were stained with Hoechst 33,342 (blue). Scale bar = 100 µm. (b) and (c) BmMSCs were differentiated into osteoblasts with or without 4-MU (100 µM) over a period of 28 days. For osteogenic differentiation the cells were stimulated with dexamethasone (10 nM), l-ascorbic acid (50 µM) and β-glycerolphosphate (10 mM) and control treated cells were treated with DMSO for the same period. (b) Ca 2+ -phosphate depositions on the cell surface were stained with Alizarin S. Scale bar = 200 µm. (c) to quantify osteogenic differentiation, the Ca 2+ -concentration was measured and normalised to the protein concentration of the cell lysate. *p < 0.05 compared to control/DMSO. (d) and (e) BmMSCs were differentiated into adipocytes with or without 4-MU over a period of 28 days. For adipogenic differentiation the cells were stimulated with dexamethasone (1 µM), insulin (100 µg/ml) and indometacin (200 µM) and control treated cells were treated with DMSO for the same period. ( Breast cancer cell-derived factors increase HA matrix and impair adipogenic differentiation of mesenchymal stem cells. As a reduced HA matrix supports adipogenic differentiation, we investigated in a next step, whether cancer cell lines influence the HA matrix of bmMSCs. We chose to compare the effects of the invasive breast cancer cell line MDA-MB-231 and the glioblastoma multiforme cell line U87-MG since both originating tumor entities are known to release disseminated cancer cells into the bone marrow, whereas only breast cancer cells are able to grow efficiently as micrometastases 18,19 . Thus, we used the two cell lines as models to reveal differences in their interaction with bmMSCs in order to explain their variance in the metastatic potential. BmMSCs were treated with cancer cell line-derived supernatant (SN) for 72 h, and the HA matrix was investigated via affinity cytochemical stainings. Compared to control-treated bmMSCs, bmMSCs stimulated with MDA-MB-231-derived SN showed a significantly increased HA matrix with a large amount of HA depositions (Fig. 3a, b). This result was confirmed by an ELISA-like HA assay, which revealed a significant increase of secreted HA into the SN after stimulation with MDA-MB-231-derived SN, whereas the amount of secreted HA after treatment with U87-MG-derived SN remained unaltered (Fig. 3c). Hyaluronan is synthesised by three different, transmembranous HA synthase isoforms (HAS 1-3) which extrude the growing HA-strand into the extracellular space 20 . HA is subsequently catabolized mainly by two hyaluronidases (HYAL1 and 2) 21 . We analyzed how the mRNA expression of HAS and HYAL isoforms in bmMSCs are affected by MDA-MB-231 supernatant and found that HAS3, which has been described to synthesize low molecular weight HA, was significantly induced (Fig. 3d).
In a next step, the impact of this differential HA inducing effect of mammary carcinoma vs. glioblastoma cells on the differentiation potential of the stromal cell compartment was employed. The induction of HA synthesis in bmMSCs by MDA-MB-231-derived SN did not influence the osteogenic differentiation of bmMSC, as there were no differences in the Alizarin Red S staining (Fig. S3a, upper panel) and no changes in the Ca 2+ -concentration of the cell layer (Fig. S3b). The simultaneous treatment with 4-MU and osteogenic stimuli did not affect the differentiation potential in both MDA-MB-231 and U87-MG-derived SN (Fig. S3a, lower panel and Fig. S3b). Likewise, the U87-MG derived SN did not affect the adipogenic differentiation of bmMSC (Fig. 3e, f, 5th bar). In contrast to this, MDA-MB-231-derived SN-in parallel to inducing HA synthesis in bmMSC-led to a significant reduction of the adipogenic differentiation after 28 days by quantitative measurements (Fig. 3e, f, 3rd bar). This effect was reversed by the simultaneous treatment with 4 MU, which restored the adipogenic differentiation potential back to control level (Fig. 3e, f, 4th bar), validating the crucial role of HA synthesis in bmMSCs mammary carcinoma supernatant-mediated suppression of adipogenic differentiation. As already shown in Fig. 1d, 4-MU alone did not lead to a spontaneous adipogenic differentiation without further stimuli. These observations were confirmed in direct co-cultures of MDA-MB-231 breast cancer cells with bmMSCs (Fig. S4). www.nature.com/scientificreports/ To analyze, which factors, secreted by MDA-MB-231 cells, may induce HA-synthesis of bmMSCs, LC-MS analysis of MDA-MB-231 and U87-MG derived supernatants were performed. As expected, PCA analysis revealed that the secretomes of the investigated cell lines were highly different (Fig. 3g). Expressed in numbers, 499 proteins were significantly higher secreted in MDA-MB-231 derived SN (Fig. 3h). Of note, TGFβ, which has been reported to inhibit adipogenic differentiation 22 and to be an MSC homing molecule 23 , was among those proteins detected significantly higher in the SN of MDA-MB-231 cells.
Indeed, treatment of bmMSCs with TGFβ3 induces HA synthesis in our experimental setup (Fig. 4a). Additionally, the adipogenic differentiation potential was examined in the presence of TGFβ3 and 4-MU simultaneously. Our experiments confirmed the inhibitory effect of TGFβ3 on adipogenic differentiation (Fig. 4b, c) and thereby provide a plausible link between the cancer cell secretome, MSC HA-matrix synthesis and MSC differentiation potential. However, the simultaneous application of 4-MU and TGFβ3 did not restore the adipogenic differentiation potential indicating that TGFβ3 may act downstream of HA induced signaling cascades.

Breast cancer cells show close interaction with mesenchymal stem cells in dependence of HA.
It was recently reported that MSCs might play a crucial role in the invasion of invasive breast cancer cells into the bone marrow 24 . Thus, we investigated the mutual interaction between bmMSCs and cancer cell lines in direct co-cultures. Cancer cells were stained with CFSE prior to seeding to allow differentiation between cancer cells and bmMSC. Co-cultures were incubated over a period of 48 h followed by staining for HA. While MDA-MB-231 breast cancer cells were located in close proximity to the bmMSC U87-MG glioblastoma cells were much more randomly distributed, thus exhibiting a significantly lower co-localisation with bmMSC ( Fig. 5a, b). In addition, bmMSC also revealed a decreased HA matrix when co-cultivated with U87-MG cells compared to co-cultivation with MDA-BM-231 cells (Fig. 5a).
Time-lapse microscopy and manual tracking of the cells revealed no differences in the accumulated distance between MDA-MB-231 and U87-MG cells over a period of 24 h when grown in monoculture (Fig. 5c). Next, both cancer cell lines were independently co-cultivated with bmMSCs for 24 h. To quantify the co-localisation between tumor cells and bmMSC, only tumor cells in juxtaposition to bmMSC were tracked and the time and migrated distance were recorded (Fig. 5d, e). Interestingly, the time and distance of MDA-MB-231 cells spent in direct juxtaposition of bmMSCs was significantly higher than of U87-MG cells, indicating that MDA-MB-231 breast cancer cells actively seek and migrate towards bmMSC cells in vitro (Fig. 5d, e).

Adipogenic differentiation of mesenchymal stem cells inhibits their adhesive potential.
Hypothesising that close proximity to MSCs might be the preferred and potentially beneficial location of cancer cells within the BM microenvironment we next investigated whether cancer cells show a preference for interaction with native or adipogenic differentiated MSCs (adipoMSCs).
For this purpose, CFSE labelled cancer cells were seeded on a confluent layer of bmMSCs with an aprroximately 50% grade of adipogenic differentiation. After 24 h the cells were fixed and stained with Oil Red O. Microscopic investigations were used to analyse whether the tumor cells show a preference in terms of their interaction partners and either co-localised with native bmMSCs (Oil Red O negative) or adipoMSCs (Oil Red O positive). Both MDA-MB-231 and U87-MG cells interacted to a greater extent with native bmMSC compared to adipoMSCs (Fig. 6a, b).

The hyaluronan system mediates the interaction between breast cancer cells and mesenchymal stem cells via the HA receptor layilin. Given the increased HA synthesis of MSC in response to
MDA-MB-231 breast cancer SN and that HA is able to connect cells via binding of the HA strands to its receptors (CD44, RHAMM, LAYN) we went on to investigate whether a transmembranous HA receptor on the MDA-MB-231 cells might be responsible for the communal migration of the two cell types.
The role of the HA system in the interaction between cancer cells and bmMSCs was evaluated in direct cocultures. For this reason, HA synthesis of bmMSCs was inhibited by 4-MU. The treatment of bmMSCs with 4-MU reduced the HA signal in affinity cytochemical stainings and removed the reticular structures observed in co-culture with MDA-MB-231 cells (Fig. 7a). As a consequence, the interaction of MDA-MB-231 cells with bmMSCs was significantly decreased in comparison to control treated bmMSCs (Fig. 7a, b). Furthermore, analysis of microscopic time-lapse recordings revealed a reduction of the distance travelled and time spent in juxtaposition of bmMSCs (Fig. 7c, d; Fig. S5a). Contrary to the findings in the MDA-MB-231 cell line, both the co-localisation and the motility in juxtaposition of bmMSCs after 4-MU treatment of U87-MG cells remained unaffected ( Fig. 7a-d, Fig. S5a).
To identify the HA receptor that mediates binding of tumor cells to bmMSCs in our experimental setup, we next silenced the HA receptor genes CD44, RHAMM and LAYN by siRNA. qRT-PCR analysis was used to validate the decrease of gene expression (Fig. S6a-c). The knockdown of these receptors neither influenced the cell number (Fig. S7d) nor the motility of MDA-MB-231 and U87-MG cells in monoculture (Fig. S7e, f). In the case of the cell line MDA-MB-231, siLAYN specifically reduced co-localisation with bmMSCs (Fig. 7e, f), accumulated distance (Fig. 7g, Fig. S5b), and time in juxtaposition in co-localisation with bmMSCs (Fig. 7h). In contrast to this observation, both the co-localisation and motility in juxtaposition of bmMSCs were not changed when the expression of CD44, RHAMM and LAYN was repressed in U87-MG cells (Fig. 7e-h).
Taken together, these data indicate that the adhesion of the breast cancer cell line MDA-MB-231 to bmMSC is mediated by the HA receptor layilin and in turn results in reduced motility of cancer cells in the presence of undifferentiated bmMSCs.

Discussion
Our experiments provide novel insights into tumor cell-stroma interactions that may explain the differences in metastatic potential of different tumor entities. We show that metastatic breast cancer cells but not glioblastoma cells induce the extracellular matrix component HA in human bmMSCs. This process may be utilized by certain tumor entities to maintain a supportive metastatic niche since HA inhibits adipogenic differentiation of human bmMSCs thus maintaining their immunosuppressive and pro-tumorigenic properties. Additionally, this mechanism fosters close cellular interaction between tumor cells and bmMSCs, which depends on the binding of the HA matrix to the HA receptor LAYN.
To the best of our knowledge, our study is among the first to analyze the impact of the HA system on differentiation potential of primary human bone marrow-derived MSCs. In subcutaneous tissue, HA exerts opposing effects on adipogenesis via its two main receptors CD44 and RHAMM 25 . In detail, CD44 is a prerequisite for subcutaneous adipogenesis, while RHAMM prevents this process 26 . Our results imply that induced HA synthesis impairs adipogenesis potentially via RHAMM activation in bmMSCs. This hypothesis was further corroborated by the finding that supernatant derived from MDA-MB-231 cells increase HA synthesis and subsequently inhibited the adipogenic differentiation of MSCs, while U87-MG supernatant did not influence either. This impact of HA on adipogenic differentiation was confirmed by the fact that the impaired adipogenesis was reverted by treatment with the HA synthesis inhibitor 4-MU. These findings are well in line with recent studies which showed a connection between HA and the maintenance of the stemness of placenta-derived stromal cells 27 .
The size of HA has previously been reported to influence its signaling properties 28 . Presumably, HA needs a minimal molecular size to engage to its receptors, further size increase should, in theory, have little effect on receptor recognition. However, convincing evidence for size-dependent effects has not been provided in the past. A possible explanation for this fact might be that a difference in HA size mediates the complex and variable clustering of HA-receptors with other binding partners, e.g. cMET, PDGFR and integrins 29,30 . Interestingly, in our experimental setting only low molecular weight HA was able to maintain MSC stemness whereas high molecular weight HA had no effect. MDA-MB-231 cells increased the expression of HAS3 which has been described to synthesize smaller sized HA compared to the other HAS isoforms. In addition, upregulation of HYAL1 was detected which is capable of facilitating the production of low molecular weight HA from larger precursors. Based on our findings we hypothesize that directed modulation of the MSC HA-matrix by the tumor cells might lead to an increased stemness of MSCs in the metastatic niche.
An important step in the population of the pre-metastatic niche by tumor cells is the adhesion to the target site. The HA matrix derived from bmMSCs represents a possible conducive soil. The importance of HA and its interaction with the HA interacting receptor CD44 in the process of extravasation is known to recruit activated T-lymphocytes to the site of inflammation 31 . Our experiments translate these findings to the stromal cells compartment and indicate that HA-inducing MDA-MB-231 cells interact significantly stronger with bmMSC, which was abolished by the treatment of bmMSCs with 4-MU. By performing knockdown experiments, we identified that this interaction is mediated by LAYN. So far, little is known about the function of LAYN, however, a role of LAYN in cell adhesion and motility has been proposed 32 . Correlating with our data, inhibition of LAYN expression resulted in a significant reduction of lymphatic metastasis of A549 lung cancer cells in vivo 33 . Furthermore, LAYN was recently considered as a prognostic marker and a high LAYN expression correlated with increased lymphatic metastasis and a decreased survival of patients with gastric and colon cancer 34 . Interestingly, the interaction between U87-MG cells and bmMSCs remained unaltered after knockdown of LAYN. This observation may be explained by the generally low interaction between these two cell types and the fact that the HA synthesis of bmMSCs is not induced by U87-MG cells. A possible therapeutic approach derived from the results from our study could be based on the close interaction between bmMSCs and invasive breast cancer cells and on the tumor-trophic migratory properties of MSCs shown in previously published studies 28 . Hence, engineered MSCs may be considered as drug delivering vectors who specifically target cancer cells and provide anti-tumoral and/or anti-metastatic effects.  www.nature.com/scientificreports/ Another critical aspect for tumor cell survival is the escape from immunosurveillance. BmMSCs can modulate the immune system by reducing the proliferation, interferon-γ production and cytotoxicity of CD4 + and CD8 + cells, eventually leading to the conversion of these cells into regulatory, immunosuppressive T-lymphocytes 9, 35 . Thus, MSCs are able to suppress the immune response and have been shown to alleviate allogeneic graft-versushost (GvH) reactions and to provide pro-tumorigenic effects in murine in vivo models 36,37 . The differentiation of MSCs can change their influence on other tissues; osteoblasts derived from MSCs positively influence hematopoiesis 38 , while adipocytes are negative regulators of the hematopoietic microenvironment 39 . Adipogenic differentiation of MSCs was shown to alter the immunosuppressive properties in inflamed adipose tissue in the sense that suppression of neutrophil recruitment was abolished after adipogenic differentiation 40 . In the T-cell inhibition assay, we demonstrated a reduced immunosuppressive potential of the bmMSC on activated T-cells after adipogenic differentiation. An enhanced adipogenic differentiation by HA synthesis inhibition via 4-MU resulted in a further reduction of the immunosuppressive potential. This observation provides a new mechanism how inhibition of adipogenic differentiation supports escape from immunosurveillance. These findings may provide an explanation why the only reported cases of brain cancer patients, which show extracranial metastasis, are patients with profound immunosuppression due to radio-and chemotherapy 41, 42 as U87-MG are not able to maintain the immunosuppressive function of bmMSC. Our data suggest that a reduced adipogenic differentiation might preserve the pro-tumorigenic properties of bmMSC.
In conclusion, we propose that invasive breast cancer cells interfere with the adipogenic differentiation potential of bmMSCs via the induction of HA in bmMSC mediated in part by TGFβ3. Thus, MDA-MB231 cells create their own pro-metastatic microenvironment by preserving pro-tumorigenic properties of undifferentiated bmMSCs which potently support tumor cell proliferation, mediate intimate tumor cell-bmMSC co-localisation and interaction and provide immuno-suppressive effects. The interaction of tumor-associated MSCs with tumor cells has been recognised as a promising emerging target 43 . Our study indicates that targeting HA binding proteins such as LAYN on metastatic cancer cells or interference with the HA-modulating effect of tumor cells on MSC via antagonizing TGFβ3 might provide novel strategies to prevent the dissemination and outgrowth of metastatic cancer cells in the metastatic niche.

Materials and methods
Reagents. If not other indicated all reagents were purchased from Sigma-Aldrich (St. Louis, MO, USA).
The ethics committee of the Medical Faculty of the Heinrich-Heine-University, Düsseldorf (No. 1830), approved the generation of MSCs from healthy volunteer bone marrow donors for research purposes.

LC-MS analysis.
For the LC-MS analysis a QExactive plus (Thermo Scientific, Bremen, Germany) connected with an Ultimate 3000 Rapid Separation liquid chromatography system (Dionex/Thermo Scientific, Idstein, Germany) equipped with an Acclaim PepMap 100 C18 column (75 µm inner diameter, 25 cm length, 2 mm particle size from Thermo Scientific, Bremen, Germany) was applied. The length of the isocratic LC gradient was 120 min. The mass spectrometer was operating in positive mode and coupled with a nano electrospray ionization source. Capillary temperature was set to 250 °C and source voltage to 1.4 kV. In the QExactive plus mass spectrometer for the survey scans a mass range from 200 to 2000 m/z at a resolution of 70,000 was used. The automatic gain control was set to 3,000,000 and the maximum fill time was 50 ms. The 10 most intensive peptide ions were isolated and fragmented by high-energy collision dissociation (HCD) 46 .
Computational mass spectrometric data analysis. Peptide and protein identification and quantification was done by using MaxQuant (version 1.6.2.10, MPI for Biochemistry, Planegg, Germany) applying standard parameters. As human samples were analyzed, searches were conducted using a specific proteome database (human Swissprot, downloaded 02/19/18) from UniProt. Methionine oxidation and acetylation at protein N-termini were set as variable modification and carbamidomethylations at cysteines were considered as fixed modification. Peptides and proteins were accepted with a false discovery rate set to 1%. Unique and razor peptides were used for label-free quantification and peptides with variable modifications were included in the quantification. The minimal ratio count was set to two and the matched between runs option was enabled. The normalized intensities as provided by MaxQuant were analyzed by using Perseus framework (version 1.5.0.15, MPI for Biochemistry, Planegg, Germany). Only proteins containing at least two unique peptides and a minimum of 4 valid values in each group were taken into consideration for protein quantification. Proteins which were identified only by site or marked as contaminant (from the MaxQuant contaminant list) were excluded from the analysis. For the calculation of enriched proteins in the two groups a Student's t-tests was applied (p ≤ 0.01) 46 .
Cytochemical HA staining. Cells were grown on coverslips. At the endpoint of the experiment, the medium was removed, the cells were washed with pre-warmed PBS and fixed with an acidic fixation solution (70% EtOH, 4% PFA and 5% acetic acid in ddH2O) for 15 min at RT. The cells were washed three times with PBS for 5 min and blocked with 5% bovine serum albumin (BSA) in PBS at RT on a rocking plate for at least 1 h. The samples were incubated with 0.4 µl HABP (Calbiochem, San Diego, CA, USA) per 100 µl 2.5% BSA in PBS and incubated in a moist chamber overnight. Streptavidin-Cy3 (Invitrogen, Karlsruhe, Germany) was used as a secondary in a dilution of 1:1000 in PBS. The nuclei were stained with Hoechst 33342 (Invitrogen, Karlsruhe, Germany). The coverslips were embedded in ProlongTM Gold (Invitrogen, Karlsruhe, Germany) and sealed with nail polish. Microscopic pictures were made with an Axio Observer.Z1 (Carl Zeiss, Oberkochen, Germany).

CFSE staining.
To assess the proliferative rate, cells were stained with CFSE. Therefore, 1 × 10 6 cells were resuspended in 1 ml PBS containing 0.1% FCS. 2 µl of a 5 mM CFSE stock solution was directly added to the cell suspension and incubated at 37 °C for 15 min. The reaction was stopped by adding 5 ml of DMEM + 10% FCS + 1% FCS + 100 U/ml, and centrifuged at 300 rcf for 5 min. Afterwards, the cells were washed twice with PBS and seeded in a density of 12,500 cells/cm 2 . The proliferative rate was determined via flow cytometry by measuring the mean signal intensity at a wavelength of 525/30 nm using an EasyCyte 5 (Millipore, Burlington, MA, USA).