LDB1 overexpression is a negative prognostic factor in colorectal cancer

Background Colorectal cancer (CRC) is the third most common cancer in western countries and is driven by the Wnt signaling pathway. LIM-domain-binding protein 1 (LDB1) interacts with the Wnt signaling pathway and has been connected to malignant diseases. We therefore aimed to evaluate the role of LDB1 in CRC. Results Overexpression of LDB1 in CRC is associated with strikingly reduced overall and metastasis free survival in all three independent patient cohorts. The expression of LDB1 positively correlates with genes involved in the Wnt signaling pathway (CTNNB1, AXIN2, MYC and CCND1). Overexpression of LDB1 in CRC cell lines induced Wnt pathway upregulation as well as increased invasivity and proliferation. Upon separate analysis, the role of LDB1 proved to be more prominent in proximal CRC, whereas distal CRC seems to be less influenced by LDB1. Materials and Methods The expression of LDB1 was measured via RT-qPCR in 59 clinical tumor and normal mucosa samples and correlated to clinical end-points. The role of LDB1 was examined in two additional large patient cohorts from publicly available microarray and RNAseq datasets. Functional characterization was done by lentiviral overexpression of LDB1 in CRC cell lines and TOP/FOP, proliferation and scratch assays. Conclusions LDB1 has a strong role in CRC progression, confirmed in three large, independent patient cohorts. The in vitro data confirm an influence of LDB1 on the Wnt signaling pathway and tumor cell proliferation. LDB1 seems to have a more prominent role in proximal CRC, which confirms the different biology of proximal and distal CRC.


INTRODUCTION
Colorectal cancer (CRC) is one of the most common cancer entities worldwide [1]. Survival rates are highly dependent on the occurrence of distant metastases [2]. Several signal pathways are known play a role in CRC development and metastasis, most prominently the Wnt signaling pathway, which is altered in 93% of all CRC tumors [3]. During the activation of the Wnt pathway, Wnt ligands bind to Frizzled receptors to induce the phosphorylation of LRP coreceptors [4]. A complex formed by Axin, APC, CK1α (serine/threonine kinases casein kinase 1 alpha) and GSK3 (glycogen synthase kinase 3) is recruited to the membrane and hampers the
LIM domain binding protein 1 (LDB1, also known as CLIM2 and NLI) is an ubiquitous nuclear adaptor protein working as a transcriptional modulator [10]. In mice, intrauterine knockout of Ldb1 causes severe anteriorposterior patterning defects, including anterior truncation and posterior duplication, which can be partially explained by a downregulation of Wnt pathway antagonists [11][12][13].
In adult mice, tamoxifen-induced knockout of Ldb1 results in drastic changes in the small intestine including a loss of Lgr5 + intestinal stem cells and significant activation of the Wnt pathway [14]. In a chemically induced mouse model of hepatocellular carcinoma, hepatocyte-directed Ldb1 knockout lead to larger and more frequent tumors, which also displayed increased proliferation and resistance to apoptosis. Microarray and RT-PCRs assays from tumor cDNA confirmed Wnt activation in Ldb1-deficient tumors [15]. Taken together, the currently available data indicates a prominent role of Ldb1 in Wnt signaling and murine intestinal homeostasis as well as inhibitory effects of Ldb1 on (hepatocellular) tumorigenesis.
Despite the role of LDB1 in intestinal homeostasis and Wnt signaling, there is no data available on the role of LDB1 in CRC. We therefore aimed to evaluate the role of LDB1 in human CRC and investigate its molecular effects on colorectal tumorigenesis. To this end, we evaluated the expression of LDB1 in CRC patient samples and correlated the LDB1 expression to clinical parameters. Moreover, functional assays in cell lines overexpressing LDB1 were used to investigate the molecular effects of LDB1 in CRC.

High LDB1 expression in CRC is associated with decreased overall and metastasis-free survival
In an initial screening attempt to identify a possible role of LDB1 in CRC, LDB1 transcripts were measured in primary tumor and normal mucosa samples from 59 CRC patients of all stages (Table 1), who underwent colorectal tumor resection at the Department of Surgery of University Hospital Heidelberg. Tumor Ct values were normalized to the corresponding mucosa and patients were separated into two groups according to a higher or lower expression of LDB1 in comparison with the mucosa. As cutoff for high LDB1 expression ΔΔCt > 0.5 was chosen, resulting in two cohorts of 46 and 13 patients, respectively. Patients were followed for a median of 30 (0-74) months.
In contrast to previous data [14,15], high LDB1 expression in the primary tumor was significantly associated with decreased overall survival (66.2 vs. 31.5 months, Hazard Ratio (HR) = 5.86, p = 0.003) ( Figure 1A) in our cohort of CRC patients. In order to validate these results, we analyzed the influence of LDB1 expression (measured via Affymetrix microarray) on overall survival in a publically available, independent cohort of 550 CRC patients of all stages. Overall survival was again shorter in tumors overexpressing LDB1 (HR = 1.5 (1.1-2.1), p = 0.021, Figure 1B). Further validation could be achieved by exploring the role of LDB1 overexpression in the TCGA cohort [3], in which LDB1 expression was measured via RNA sequencing. In this cohort (n = 267), LDB1 overexpression again lead to a significantly reduced overall survival in CRC patients of all stages (HR 2.1 (1.0 -4.1), p = 0.038, Supplementary Figure S1A).
To evaluate the influence of LDB1 expression on systemic tumor dissemination, we performed a subgroup analysis on patients with non-metastatic disease only. This analysis revealed a significantly reduced metastasisfree survival (62.8 vs. 19.7 months, HR = 7.22, p = 0.001, Figure 1C) in patients with high LDB1 expression in the primary tumor. In the Affymetrix cohort of non-metastatic patients (n = 218), recurrence-free survival was also reduced in patients with LDB1-overexpression tumors (HR = 2.37 (0.97-5.78), p = 0.051, Figure 1D). The TCGA clinical data does not include recurrence-free survival on non-metastatic patients; we therefore analyzed only overall survival in this cohort (n = 176), which again showed unfavorable prognosis in LDB1-overexpressing tumors (HR 4.0 (1.1-13.7), p = 0.019, Supplementary Figure S1B).
In conclusion, LDB1 overexpression proved to be an unfavorable prognostic factor in three large, independent patient cohorts. LDB1 overexpression also seems to be strongly associated with early recurrence in non-metastatic patients.

Tumor LDB1 expression positively correlates with Wnt pathway activation in CRC patient samples
As LDB1 surprisingly turned out to be a negative prognostic factor in CRC and as LDB1 has previously been associated with the Wnt signaling pathway [13][14][15], we next aimed to investigate its influence on the Wnt signaling pathway in CRC.
To this end, the expression of LDB1 was correlated with the expression of the Wnt signaling-associated genes CTNNB1, AXIN2, MYC and CCND1 in patient primary tumor samples (Figure 2A). Upon direct comparison in patients with high or low LDB1 expression, CTNNB1 showed a strong tendency towards overexpression (p = 0.11), wheres AXIN2, MYC and CCND1 were significantly overexpressed (p < 0.05 for all three genes) in LDB1-overexpressing tumors. In addition, we observed a highly significant positive correlation (p < 0.001 for all four genes) between the tumor LDB1 expression and the expression of Wnt signaling-associated genes ( Table 2). This indicates a coincidence of Wnt activation and LDB1 Oncotarget 84260 www.impactjournals.com/oncotarget

Upregulation of LDB1 induces changes in MYC and CCND1 expression
As the positive correlation between LDB1 expression and Wnt signaling in tumor samples was at least partly contradictory to previously published data [13][14][15], we aimed to investigate this interaction in detail. To this end, 4 human CRC cell lines (HCT116, DLD1, SW480, SW620) were transduced with a third generation lentiviral vector to generate cell lines stably overexpressing LDB1 ( Figure 2B, Figure 3A-3D). The expression of LDB1 was incremented 7 to 18 fold in comparison to wild-type cells. The 4 cell lines were chosen as they represent different stages, grading and sites of origin of CRC manifestations.

Effects of LDB1 overexpression on Wnt signaling activity
To directly examine the stimulation of Wnt signaling activity by LDB1 in CRC, the previously chosen 4 CRC cell lines were transfected with a TCF-firefly reporter construct indicating Wnt signaling activity via bioluminescence activity. Renilla-luciferase plasmids were used to estimate cell viability [24]. Results are displayed in relative light units of luminescence ( Figure 3A). DLD1 LDB1-High and SW620 LDB1-High cells exhibited a significant increase of Wnt signaling activity demonstrated by a change from 39.77 ± 4.34 to 57.57 ± 5.81 (p = 0.003) and 170.11 ± 12.15 to 209.44 ± 2.66 (p = 0.005) relative light units respectively. The Wnt signaling activity did not significantly change in HCT116 LDB1-High and SW480 LDB1-High cells.
To investigate the effects of LDB1 overexpression on cell cycle regulation, the cell cycle phases of CRC cell lines were analyzed via flow cytometry ( Figure 3C). HCT116 LDB1-High and SW480 LDB1-High cells did not show major differences in comparison with wild-type cells. DLD1 LDB1-High cells displayed a significant increase of cells in G0/G1 phase from 64.04% to 69.72% (p = 0.025), indicating an increased number of cells entering proliferation and thus confirming the results of the proliferation assays. This change simultaneously reduced the number of cells in G2/M from 17.73% to 13.64% (p = 0.010). Apoptotic cells decreased from 3.43% to 1.83% (p = 0.144) and cells in S phase remained constant at 14% (p = 0.992). In SW620 LDB1-High cells, apoptosis augmented from 2.82% to 7.14% (p = 0.029), the G0/G1 phase was reduced from 57.35% to 12.45% (p < 0.001) and the S phase changed from 20.05% to 7.77% (p < 0.001). The G2/M phase increased from 19.78% to 72.64% (p < 0.001), indicating an arrest of these cells before mitosis. This observation again confirms the results of the proliferation assays.

LDB1 may have different roles in tumors of the proximal and distal colorectum
During embryologic development, the colorectum is formed by two primordial organs, the midgut and the hindgut. The point of fusion is the left colic flexure (Cannon-Böhm's point), where the blood supply switches from the superior mesenteric artery to the inferior mesenteric artery. Multiple studies are currently investigating molecular differences between tumors of the proximal colon (proximal to the left colic flexure) and distal colon (distal to the left colic flexure down to the anal verge). Little is known so far; however, the currently available data indicates that proximal and distal tumors are in fact entirely different entities [28][29][30]. Interestingly, recent data suggests that the proximal colon is more susceptible to Wnt associated tumorigenesis, which results in tumor formation upon moderate Wnt activation. This increased susceptibility distinguishes the proximal from the distal colon, which requires more pronounced Wnt activation in order to result in tumor formation [31,32]. As we have demonstrated the role of LDB1 in Wnt signaling and have also observed differential effects of LDB1 on CRC proliferation and invasiveness, we hypothesized that the role of LDB1 may differ between tumors of different locations within the colorectum.
We therefore analyzed the Affymetrix cohort according to tumor localization ( Figure 4). The effects of LDB1 were more pronounced in the subgroup with tumors in the proximal tumor than in the subgroup with distal CRC (OS: HR = 1.69, p = 0.14 (proximal) vs. HR = 1.27, p = 0.35 (distal); RFS: HR = 1.7, p = 0.0076 (proximal) vs. HR = 1.35, p = 0.064 (distal)). Stratified analyses based on tumor location were not possible in the Heidelberg patient cohort due to low case numbers. We also evaluate the influence of LDB1 on recurrencefree survival in non-metastatic patients according to tumor localization and again observed a striking effect of LDB1 in the proximal colon; in this cohort (n = 45) no recurrences were observed in patients with low LDB1 expression ( Figure 4E and 4F).
After splitting the patient samples due to localization, the association between tumor LDB1 expression and Wnt pathway gene expression remained significant for almost all genes analyzed in samples from the proximal colon and for all genes in samples derived from the distal colon (Table 3A). In samples from the proximal colon, the association between LDB1 expression on Wnt signaling-associated genes seemed generally to be even stronger than in distal tumors; however, due to the relatively low case number (due to the lower incidence of proximal CRC) significance levels were lower. We also hypothesized that as the rectum is the most distal segment of the colon, the effects of LDB1 in the rectum may be even less pronounced. We therefore also performed an analysis after splitting our cohort in patients with tumors in the colon vs. the rectum. In this analysis, association between tumor LDB1 expression and Wnt pathway gene expression remained significant for samples derived from the colon, but not for samples derived from the rectum (Table 3B), again stressing the increased effects of LDB1 in the proximal colon as opposed to the distal colorectum.

DISCUSSION
CRC is among the most commonly investigated tumor entities [2,17,33,34]. The most prominent signaling pathway in CRC is Wnt signaling, its activation (most frequently by inactivating APC mutations) usually initiates the adenoma-carcinoma sequence and thus colorectal tumorigenesis [35]. Previous data from mouse models demonstrated a prominent role of Ldb1 in hepatocellular carcinoma and intestinal homeostasis, indicated an inhibition of Wnt signaling by Ldb1 and Oncotarget 84264 www.impactjournals.com/oncotarget hence a tumorsuppressive role of Ldb1 overexpression [14,15]. As the majority of CRCs are heavily dependent on Wnt signaling, we hypothesized that LDB1 may have a role in human CRC as well. More precisely, we aimed to show the role of LDB1 as a tumor suppressor in CRC.
Unsurprisingly, the expression of LDB1 in our patients proved to be heterogeneous. Surprisingly, however, patients with lower LDB1 expression in the tumor experienced improved overall and metastasis-free survival, thus demonstrating a negative prognostic effect of LDB1 expression in human CRC. These results could be reproduced in two independent, large cohorts of patients (Affymetrix and TCGA samples), therefore strongly validating our results. In addition, qPCR expression profiling showed higher expression levels of CTNNB1, AXIN2, MYC and CCND1 in patients with LDB1

Figure 4: Comparison of survival and recurrence rates depending on tumor location proximal (graphs on left side) and distal to the left colic flexure (graphs on right side). CRC patients of all stages with tumors located in the proximal colon and high
expression of LDB1 showed a tendency to shorter overall survival (A, HR 1.69, p = 0.14) and significantly shorter recurrence-free survival (C, HR = 1.7, p = 0.0076). LDB1 low Patients with non-metastatic (M0) CRC in the proximal colon showed a strikingly improved recurrence free survival with no recorded recurrences in the study cohort (E, HR not calculable, p = 0.038). The effects of LDB1 seem to be much less pronounced in the distal colon; overall survival rates of distal CRC patients are similar for LDB1 high and LDB1 low patients (B, HR = 1.27, p = 0.35). The effects of LDB1 expression on recurrence-free survival are also less pronounced in patients with distal CRC (D, HR = 1.35, p = 0.064). In patients with non-metastatic distal CRC, LDB1 overexpression leads to reduced recurrence-free survival, albeit to a lesser extent than in the proximal colon (F, HR = 3.7, p = 0.016).
Oncotarget 84265 www.impactjournals.com/oncotarget overexpression, suggesting an increased Wnt signaling activity in these tumors and supporting the clinical data reported above.
These findings are in contradiction to previous data [13 -15], which indicated a suppressive effect of Ldb1 on the Wnt signaling pathway. However, these data were from mouse models only; it is conceivable that murine Ldb1 has different effects and targets than human LDB1. Another explanation of the contradictory results of many studies on Ldb1 may be organspecific effects of Ldb1: While inactivation of Ldb1 led to an increase in tumor growth in murine hepatocellular carcinoma [15], its overexpression in mammary glands promoted breast tumorigenesis [36,37]. In addition, widespread LDB1 expression has been demonstrated in embryonic epithelial tissues [38,39] and at the invasive front of squamous cell carcinoma of the head and neck [40], preventing differentiation of LDB1-overexpressing cells and again suggesting tumorpromoting effects of LDB1. These findings strongly suggest organ-specific effects of Ldb1 and further support our data identifying LDB1 as a negative prognostic factor in CRC.
To further validate LDB1 as a tumor-promoting factor in CRC, we initiated a series of in vitro experiments using lentiviral vectors to overexpress LDB1 in human CRC cell lines. We observed differential effects; however, most results again pointed towards an activation of the Wnt signaling pathway in human CRC, supporting the results seen in patient samples. Interestingly, the 4 genes we used as indicators of Wnt signaling activity were not uniformly regulated upon LDB1 overexpression, indicating incomplete activation of the Wnt signaling pathway, which has recently be described in other tumor entities and transcription factors as well [41].
There are currently no studies available investigating the correlation of LDB1 overexpression with metastatic activity; however, LDB1 overexpression at the invasive front of squamous cell carcinomas of the head and neck suggests a connection between LDB1 overexpression and metastasis [40]. In the present study, patients with LDB1 overexpression in the tumor tissue displayed worse overall survival and metastasis-free survival. These results indicate that CRC tumors with low expression of LDB1 metastasized later than tumors with a higher expression, indicating pro-metastatic features of LDB1 in CRC. This finding was also supported by the other two clinical cohorts and is well in line with above mentioned data from human SCC of the head and neck [40].
Colorectal cancer is usually classified as a single entity despite data showing distinct molecular, pathological and clinical features of proximal and distal colon cancers [28][29][30]. This differentiation is embryologically well justified as the colorectum develops from both the midgut (ascending and transverse colon) and the hindgut  sigmoid colon and rectum). Above mentioned studies demonstrated increased activation of the Wnt pathway and MYC expression in distal colorectal carcinomas in comparison with proximal tumors. Given this differential role of Wnt signaling in CRC and the effects of LDB1 overexpression on Wnt signaling, we hypothesized that the role of LDB1 may be different in proximal and distal CRC. We therefore analyzed the clinical data of our patient cohort as well as previously published patients according to the location of the tumors and were able to demonstrate a larger effect of LDB1 overexpression in proximal CRC compared to distal CRC. In addition, the association between LDB1 expression and Wnt signaling activity was stronger in tumors of the colon than rectal tumors. These differential roles of LDB1 according to tumor location may be explained by the varying levels of Wnt activity on proximal and distal CRC: Proximal carcinomas upregulate the Wnt pathway to a lesser extent than distal tumors; therefore, LDB1 may upregulate Wnt in proximal tumors more easily than in distal tumors, which already exhibit high Wnt signaling activity. The relatively increased expression of Wnt pathway genes in the LDB1-overexpressing, proximal (colon) cancer samples analyzed here may be a result of the higher LDB1 expression; this in turn can explain the increased clinical effects of LDB1 overexpression in proximal CRC samples.
To study the influence of LDB1 on malignant features such as proliferation, invasiveness and cell cycle dysregulation, HCT116, DLD1, SW480 and SW620 cells were infected with a third generation lentivirus to overexpress LDB1. Although the gene expression and phenotypic changes after the upregulation of LDB1 did not show a unique pattern, it was possible to recognize that HCT116 LDB1-High and DLD1 LDB1-High cells developed a similar phenotype with increased proliferation and augmented the expression of CCND1. SW480 LDB1-High and SW620 LDB1-High cells constituted a second phenotype, exhibited an upregulation of Wnt pathway genes including a marked overexpression of MYC, reduced proliferation rate and increased invasivity.
Generally, the Wnt signaling activity on proximal colon tumors is lower compared to the Wnt signaling activity in distal colon tumors [31,32]. The site of origin (proximal vs. distal colon) of the cell lines is unknown; however, based on the Wnt signaling activity they may correspond to proximal (group 1) and distal colon (group 2). The overexpression of LDB1 resulted in an increase in proliferation and Wnt signaling activity in group 1, while it reduced the proliferation in group 2. This supports our clinical findings, indicating a greater role of LDB1 overexpression in tumors of the proximal colon. On the other hand, LDB1 overexpression also has an effect in distal CRC, albeit weaker. In addition to decreased proliferation, the effects of LDB1 overexpression on the group 2 cell lines included increased invasiveness which may partly explain the clinical findings.
In conclusion, LDB1 overexpression is a negative prognostic factor in colorectal cancer. Its effects are stronger in tumors of the proximal colon, which may be due to comparatively low intrinsic Wnt signaling activity in these tumors which is stimulated by LDB1 overexpression. In vitro, LDB1 overexpression has differential effects on human CRC cell lines, which is again dependent on their intrinsic Wnt activity. Limitations of this study include its retrospective nature. However, the clear prognostic effects of LDB1 overexpression demonstrated in this work clearly indicate the need for further studies. Given the unclear role of adjuvant treatment in stage II CRC patients and the prognostic value of LDB1 regarding the prediction of metastasis, a prospective study in stage II CRC may establish LDB1 as a valuable biomarker in nonmetastatic CRC.

Patient samples
Clinical data and samples are being collected continuously at the Department of Surgery of University Hospital Heidelberg; our management of patients with colorectal cancer has been published in detail before [16,17]. Tumor and mucosa samples were collected from 59 CRC patients (of all UICC stages), which underwent tumor resection at the Department of General, Visceral and Transplantation Surgery at the University Hospital in Heidelberg between October 2006 and July 2012. The study was approved by the independent ethics committee of the University of Heidelberg. All patients provided written informed consent prior to surgery. This patient cohort is henceforth referred to as the Heidelberg cohort.

Affymetrix CRC cohort
A database of publicly available colorectal cancer patient samples measured by Affymetrix gene chips was set up as described previously for gastric cancer [18]. In brief, gene chip datasets with transcriptome-wide gene expression data generated by Affymetrix gene arrays and available survival data were identified in the Gene Expression Omnibus repository (www.ncbi.nlm.nih.gov/geo/). Samples were MAS5 normalized. We selected the probe set 35160_at for LDB1. Survival analysis was performed by employing Cox proportional hazard regression in the R statistical environment (www.r-project.org) as described previously [19]. Kaplan-Meier survival plot, and the hazard ratio with 95% confidence intervals and log-rank P value were plotted using the library "survival".

TCGA CRC cohort
Colorectal cancer patients measured by RNA-seq were published in The Cancer Genome Atlas (TCGA) [3]. Pre-processed level 3 data generated using Illumina HiSeq www.impactjournals.com/oncotarget 2000 RNA Sequencing V2 was used. For each sample, the expression level was determined using a combination of MapSplice and RSEM. The individual sample files were merged in R using the plyr package [20]. Survival analysis was performed as for the Affymetrix cohort.

Cell culture
CRC cell lines were obtained from ATCC (HCT116, DLD1, SW480, SW620) and cultured at 37ºC with 5% CO 2 and 95% humidity. DMEM (Thermo Fisher, Waltham, USA) and RPMI (Thermo Fisher, Waltham, USA) media were supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher, Waltham, USA) and no antibiotics were used. Cell lines were regularly screened for Mycoplasma contamination; the cell line authenticity was confirmed regularly.

LDB1 overexpression
Cell lines constitutively overexpressing LDB1 were generated by transduction with a third generation lentiviral vector system [21]. For virus production, HEK293T cells were cultured in DMEM medium with 10 % heat inactivated FBS. A mixture of 3 mL OptiMEM ® (Thermo Fisher, Waltham, USA) medium, 28 μL Lipofectamin ® 2000 (Thermo Fisher, Waltham, USA), 4 μg pMDLg/ pRRE (was a gift from Didier Trono, Addgene plasmid # 12251), 2 μg pRSVRev (Addgene plasmid # 12253), 2 μg pMD2.G (Addgene plasmid # 12259) and 6 μg EX-Z6214-Lv205 (GeneCopoeia, Rockville, USA) was added dropwise to transfect HEK293T cells. After 24 hours, medium was replaced with fresh medium and incubated for another 24 hours. The virus-enriched medium was collected, filtered and mixed with polybrene (8 μg/mL final concentration). 1 × 10 5 cells from each CRC cell line were seeded per well on 6-well plates and incubated with the virus solution for seven hours. After 48 hours, transduced cells were selected with puromycin (final concentration 10 μg/mL). To select brightly GFP expressing cells, CRC cell lines were sorted with a FACS ARIA II (BD Biosciences, San Jose, USA). Sorted cells were grown in medium with 10% FBS and 1% penicillin/streptomycin. Control cell lines were generated similarly using a modified version of the EXZ6214Lv205 vector, lacking the LDB1 gene.

Plasmid amplification
One Shot ® TOP10 chemically competent E. coli cells (Thermo Fisher, Waltham, USA) were transformed following the manufacturer's instructions, adding 2 μL from the corresponding plasmids. Plasmid DNA was isolated with the Qiagen EndoFree ® Maxi Kit (Qiagen, Hilden, Germany), following the manufacturer's protocol. DNA was eluted with nuclease-free water and concentration was adjusted to reach 1 μg/μL.

RNA isolation
Tissue samples were fragmented using the TissueLyser II (Qiagen, Hilden, Germany). The NucleoSpin ® RNA Kit (Macherey-Nagel, Düren, Germany) was used for RNA isolation following the manufacturer's instructions. Tissue fragments were inserted into microcentrifuge tubes containing 700 µL buffer RA1 and one 5 mm stainless steel bead (Qiagen, Hilden, Germany). Samples were shaken twice at 30 Hz for 2 minutes. Cell lines pellets were lysed with 350 µL buffer RA1. RNA was eluted with 40 µL RNAase-free water and concentration was measured using a NanoVue™ Plus Spectrophotometer (GE Healthcare, Piscataway, USA).

cDNA synthesis
The High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher, Waltham, USA) was used for cDNA synthesis. For each reaction, 1 µg of RNA was mixed with 2 µL RT Buffer, 2 µL RT random primers, 0.8 µL dNTP Mix, 1 µL MultiScribe™ Reverse Transcriptase and RNAase-free water to complete 20 µL. Tubes were incubated in a thermocycler at 25ºC for 10 minutes, at 37ºC for 120 minutes and at 85ºC for 5 minutes. Tissue samples were diluted with Nuclease-free water in a 1:5 ratio and cell lines in a 1:10 ratio.

Cell proliferation
Cell lines were seeded onto 24-well plates and incubated for three days at 37ºC with 5% CO 2 and 95% humidity. Cells were then trypsinized and resuspended in 1 mL of medium. Ten microliters from the cell suspension were pipetted per pocket into counting slides. Cells were counted with an automatic cell-counter.

Cell cycle analysis
General flow cytometry methods of our group have been published in detail before [27]. Briefly, cells were seeded onto 24-well plates. After three days in culture, supernatant and cells were collected into flow cytometry tubes. Pellets were resuspended with 1 mL ice-cold 70% ethanol and incubated at 4°C overnight. Samples were centrifuged and cells were washed twice with 1 mL PBS. To digest RNA, each pellet was incubated with 100 µL of 0.2 mg/mL RNase A solution (Sigma-Aldrich, St Louis, USA) at 37°C for 1 hour. Cells were incubated with 10 µL 1 mg/mL Propidium Iodide (Sigma-Aldrich, St Louis, USA). Flow cytometry was performed with the BD FACSCalibur™ (BD Biosciences, San Jose, USA). In total 10,000 events were counted per cell sample. Cell counts were analyzed with the WinMDI 2.9 software (The Scripps Research Institute, La Jolla, USA) to determine the percentage of cells per cell cycle phase.

Scratch assay
One to two million cells were seeded per well onto 6-well plates. After 24 hours, scratches were made on each well with a 100 µl pipet tip. Perpendicular lines were marked bellow the bottom of the plate to determinate a specific place. Pictures were made after 24 hours with a Leica DMI3000B microscope (Leica Microsystems, Wetzlar, Germany) using the Leica Application Suite v3.6 software (Leica Microsystems, Wetzlar, Germany). The distance between the two edges was measured and shortening was calculated using the same software.

Statistical analysis
Survival/recurrence rates were depicted using Kaplan-Meier curves and compared using the log-rank test. Results from in vitro experiments were compared using the Student's independent samples t-test (to compare means) and bivariate correlation (Pearson correlation). P values < 0.05 were considered statistically significant. The following code was used in this article to display significance levels: (*) p < 0.05, (**) p < 0.01, (***) p < 0.001.