The role of TNF-α in chordoma progression and inflammatory pathways

Cell culture and treatment

The U-CH1 [30] and MUG-Chor1 cell lines [31], available from the ATCC Bioresource Center, were kindly provided by the Chordoma Foundation (Durham, NC, USA). The cells were cultured on 0.1% gelatin (Cat. No: G1890-100G, Sigma-Aldrich, USA)-coated tissue culture plates and maintained in IMDM (Cat. No: 12440053, ThermoFisher, USA)/RPMI (Cat. No: 61870–010, ThermoFisher) medium at a ratio of 4:1 containing 10% fetal bovine serum and 1% antibiotics. This medium will be called culture medium (CM) after this point throughout the text. Routine flow cytometry and gene expression analyses for pertinent molecular markers, polymerase chain reaction (PCR)-based mycoplasma screening and Sanger sequencing for short tandem repeats to verify the characteristics of the original cell lines were routinely carried out. Chordoma cell characteristics were retained in both the control cells and long-term treated samples (Supplementary Tables 1 and 2). For short-term experiments, 10 ng/ml human recombinant TNF-α (Cat. No: T0157, Sigma-Aldrich) was administered to cells for 1, 4 and/or 7 days, depending on the experiment. To obtain long-term TNF-α-treated cell lines, TNF-α was administered to U-CH1 and MUG-Chor1 cells for 12 months by gradually increasing the TNF-α concentration from 2.5 and 5 to 10 ng/ml every 2 months. Thereafter, the cell lines were grown in CM containing 10 ng/ml TNF-α. The NF-kB inhibitor QNZ (N4-[2-(4-phenoxyphenyl)ethyl]-4,6-quinazolinediamine) (Cat. No: SML0579, Sigma-Aldrich, USA) was administered to cells with TNF-α as a TNF + QNZ control for long-term treatment. To determine the permanent effects of prolonged TNF-α treatment, the U-CH1 and MUG-Chor1 cells were deprived of TNF-α after 6 months of TNF-α exposure. By doing so, another control group was established and labeled “TNF-α deprivation group”. The relationship and interaction between LIF and TNF-α in chordoma cells were assessed by treatment of the cells with 100 ng/ml recombinant human leukemia inhibitory factor (Cat. No: LIF 1050, Millipore, USA) in CM twice per week during 3 weeks.

Immunocytochemistry

Trypsinized cells were washed and resuspended in phosphate-buffered saline (PBS) (Ca⁺⁺/Mg⁺⁺-free) containing human FcR blocking reagent (Cat. No:120–000-442, Miltenyi Biotech, Germany). The cells were initially incubated on ice for 5 min and later incubated with an anti-CD338 antibody conjugated with PerCP-Cy5.5 (Cat. No: 561460, BD Biosciences, USA) for 30 min before being washed with PBS (Ca⁺⁺/Mg⁺⁺-free) to eliminate unbound antibodies. Both groups were stained with DAPI (Cat. No: 422801, Biolegend, USA) to exclude dead cells. To visualize the behavior of NF-kB molecules, cells grown on gelatin-coated coverslips were fixed with the 4% paraformaldehyde, blocked with 1% BSA and incubated overnight with a primary anti-NF-kB p65 antibody (Cat. No: ab16502, Abcam, UK). All washes were performed with PBS containing 1.5% Tween. Alexa Fluor 488-labeled goat anti-rabbit IgG (Cat. No: A-11034, Thermo Fisher, USA) was used as a secondary antibody and incubated for 1 h. Nuclear staining was carried out using DAPI (Cat. No: 422801, Biolegend, USA), and the cells were visualized under a fluorescence microscope.

Cell migration and invasion assays

Transwell membrane inserts with an 8.0 μm pore size (PIEP12R48, Millipore, USA) were used to assess changes in the invasion and migration of TNF-α-treated cells. To assess invasion, the assay chambers were coated with 250 μg/ml basement Matrigel (Cat. No: 354234, BD Biosciences, USA) and incubated overnight at 37 °C. Next, 4 × 10⁴ cells were transferred to the upper chambers in serum-free CM containing TNF-α (10 ng/ml). The lower chambers were supplied with 1 ml CM containing 10% fetal bovine serum. To analyze short-term effects of TNF-α exposure, U-CH1 and MUG-Chor1 cells were seeded in serum-free complete medium, serum-free complete medium + TNFα, and serum-free complete medium + TNFα + QNZ, and incubated for 24 h. The inserts were fixed with 3.7% formaldehyde, permeabilized with methanol and stained with crystal violet. The cells were visualized and counted in four different fields at 400-fold magnification under an inverted light microscope.

Colony formation assay

To assess the self-renewal of long-term TNF-α-treated cells, a colony formation assay was carried out. To this end, six-well plates were coated with 1 ml CM with 1% low melting point agarose (Cat. No: A9045, Sigma-Aldrich, USA) and incubated at 37 °C to solidify. Next, a top layer containing cells in complete medium containing 0.7% low-melting point agarose was added. Medium was replenished every 3 days. After 3 weeks, the plates were stained with 0.05% crystal violet, and colonies containing more than 20 cells were counted under an inverted light microscope.

Tumor sphere formation assay

Long-term TNF-α-treated cells (4 × 10⁵) were seeded in 25-cm² culture flasks coated with 2% agarose and 0.9% NaCl in medium containing CM (serum-free) with 1% antibiotics, 50 ng/ml human FGF-2 (Cat. No: PHG0021, Thermo Fisher, USA), 50 ng/ml EGF (Cat. No:PHG0311, Thermo Fisher, USA), 1× N2 supplement (Cat. No: 17502048, Thermo Fisher, USA) and 1× B27 supplement (Cat. No: 17504044, Thermo Fisher, USA). Cell clusters were routinely dispersed with a serological pipette. After 4 weeks, spheroid bodies were transferred to 96-well plates, and the total number of tumor spheres was quantified under a light microscope.

Chemoresistance assay

To evaluate the effect of TNF-α on chemoresistance, cells were seeded in 24-well plates in duplicate for each experimental group. Several chemotherapeutic agents were examined to determine their cytotoxicity to chordoma cells. Paclitaxel and bortezomib were most cytotoxic (data not shown). The cells were treated with freshly prepared 0.016 μM paclitaxel and 0.016 μM bortezomib in complete CM, which was refreshed every 2 days. The viability assay was conducted over 7 days.

Microarray-based expression assay

For microarray-based expression analysis, total RNAs isolated using TRIzol (Cat. No: 15596026, Thermo Fisher, USA) from cells treated with TNF-α for 1 week (n = 2) or 12 months (n = 4), and untreated U-CH1 and MUG-Chor1 cells (n = 6) were used for whole transcriptome analysis using a Human Gene 2.1 ST Array (GeneAtlas System, Affymetrix, USA) according to the manufacturer’s protocol. The acquired data were analyzed using the Transcriptome Analysis Console 3 software tool (Affymetrix, USA). Additional analyses were performed using the internet-based DAVID and ingenuity pathway analysis tools (Qiagen, Germany).

Human chordoma samples

Twenty-four human chordoma samples were obtained during surgery at the Department of Neurosurgery, Yeditepe University Hospital. None of the patients had undergone chemotherapy. Samples used for gene expression quantification were frozen at −80 °C immediately after surgery. Additionally, paraffin-embedded chordoma tissues were preserved for immunohistochemical (IHC) staining. Clinicopathological data, including age, sex, tumor location, diagnosis, tumor origin, surgery margin, tumor volume, recurrence, metastasis and adjuvant therapy, were collected retrospectively (Supplementary Table 3). The neutrophil/lymphocyte ratio (N/L) in patient serum samples was determined before surgery. The follow-up time was defined as the time that elapsed between surgery and death or last contact with the patient. Only samples that were pathologically defined as chordoma were included in the study. Samples with insufficient patient information or a follow-up time < 3 months were excluded from the study.

Quantitative RT-PCR

Fresh-frozen primary clival chordoma tissue samples were homogenized in liquid nitrogen using a mortar and pestle, followed by immediate addition of TRIzol according to the manufacturer’s protocol. Total RNA was also isolated from in vitro TNF-α-treated cells, LIF-treated cells, and their respective controls. RNAs were quantified using a NanoPhotometer (Implen, Germany). mRNA levels were quantified by two-step real-time PCR using TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA, USA). Relative mRNA levels were determined using the 2^-∆∆Ct method with GAPDH as internal control and by normalizing the values to those of the control groups.

Western blotting

Total protein was extracted using RIPA buffer (Cell Signaling, USA). Protein concentrations were quantified using a Pierce™ BCA Protein Assay kit (Cat. No: 23225, Thermo Fisher, USA). Protein samples were loaded onto 12% polyacrylamide gels (Cat. No:#1610174, Bio-Rad, USA) at a concentration of 20 μg/lane and after electrophoresis transferred to polyvinylidene difluoride (PVDF) membranes. The membranes were blocked with TBS/Tween containing 5% nonfat dry milk for 1 h and subsequently incubated at 4 °C overnight with an anti-T (Brachyury) primary antibody (Cat. No: ab209665, Abcam, UK). Next, the membranes were washed with TBS/Tween three times and incubated with HRP-conjugated secondary antibodies. GAPDH (Cat. No: 5174, Cell Signaling, USA) was used as an internal positive control. Images were taken using Fusion Pulse (Vilber, France).

Immunohistochemical analysis

IHC analysis of paraffin-embedded chordoma tissues was carried out in a blinded manner using anti-TNF-α (Cat. No: ab1793, Abcam, UK) and anti-TNF receptor 1 (Cat. No: ab19139, Abcam, UK) antibodies by a pathologist who was uninformed about the clinical characteristics of the patients. An independent review was also performed. Cells in the cancerous regions of the tissues were scored as follows: (0) no, (1) low, (2), moderate or (3), high expression. Only membrane staining intensity was taken into consideration for anti-TNF receptor 1 scoring, whereas both membrane and cytoplasmic staining intensities were taken into consideration for anti-TNF-α scoring. Average cell intensity scores were calculated for every patient by weighted averaging of the scored cells and the percent area they occupied in the tissue.

Statistical analysis

Microarray-based expression data were analyzed using one-way ANOVA. Clinicopathological features age, sex, tumor location, recurrence, metastasis and survival status of the patients were analyzed in correlation with TNF-α and TNFR1 expression levels, IHC score and N/L ratio. Correlations were evaluated using Spearman’s rank correlation test. An overall survival graph was prepared using the Mantel-Cox (log-rank) test, which is analogous to the Kaplan-Meier survival analysis method. All other p values were obtained using a two-tailed Student’s t test on experiments carried out at least in triplicate. A p value < 0.05 indicated statistical significance. Patients with high and low TNF-α and TNFR1 expression levels and N/L ratios were separated into groups of equal number according to the relative median. Each experiment was carried out in triplicate, unless otherwise stated in the text. Significant outliers were evaluated for each experiment and excluded from the analysis.

Chordoma cells respond to TNF-α treatment

We found that after TNF-α treatment both the U-CH1 and MUG-Chor1 cell lines exhibited changed morphologies towards a more mesenchymal state with increased spaces between the cells and the formation of spindle-like cell structures, which indicates the occurrence of EMT in these cells (Fig. 1a). We also found that short-term TNF-α treatment had no effect on the proliferation of these cells (data not shown). Fluorescence microscopy analysis revealed that the U-CH1 and MUG-Chor1 cells express NF-kB, and that its abundance dramatically increased after TNF-α exposure compared to that in control cells. In addition, we found that the NF-kB inhibitor QNZ decreased the effect of TNF-α exposure. In the TNF-α-treated samples, a clear localization of NF-kB in the nucleus was observed, both at day 1 and day 7 (Fig. 1b), whereas long-term TNF-α deprivation did not cause any change in NF-kB localization and/or abundance (Fig. 1c).

TNF-α treatment increases invasion and migration and induces EMT-related gene expression

Using Transwell assays, we found that TNF-α treatment significantly increased the migration and invasion of both chordoma cell lines following both short-term (Fig. 2a) and long-term treatment (Fig. 2b), as indicated in representative images from long-term-treated samples (Fig. 2c). We also found that QNZ decreased the invasion and migration of the cells (Fig. 2a, b) and that long-term TNF-α deprivation did not cause any significant changes in invasion and migration of the cells (Fig. 2b).

The effect of TNF-α on EMT marker gene expression levels was determined in U-CH1 and MUG-Chor1 cells after exposure to TNF-α. The expression of the EMT markers E-cadherin, CK19, EMA, Twist, Slug, Snail, Vimentin, Zeb1, Zeb2, and N-cadherin, was assessed using qRT-PCR. By doing so, a drastic suppression of the epithelial markers E-cadherin, CK19 and EMA was observed, whereas the late mesenchymal marker ZEB2 was upregulated. Subsequent QNZ treatment decreased the effects of TNF-α, but failed to neutralize levels to those of the control cells in most cases. We also found that the Brachyury (T) gene expression levels were suppressed by long-term TNF-α treatment, and that QNZ treatment significantly increased the T expression levels, although not to the point of complete rescue (Fig. 2d). In the short-term treatment samples, CK19 and E-cadherin were found to be suppressed, whereas the EMT markers Vimentin, Slug, Twist, and N-cadherin were already upregulated on day 1. No significant changes were observed in the T gene expression levels on day 1. We found that on day 4 the E-cadherin, CK19 and T expression levels decreased, whereas the ZEB2 expression level increased, compared to those in control cells. QNZ treatment resulted in an opposite or neutralizing effect with respect to TNF-α treatment (Fig. 2e). Using Western blotting, we found that the brachyury protein expression data were mostly consistent with the mRNA expression data. The T protein levels were found to be decreased in MUG-Chor1 cells compared to those in control cells. We also found that the T protein levels were decreased in U-CH1 cells after TNF-α treatment. Interestingly, we did not detect any T protein in TNF-α-deprived MUG-Chor1 cells (Fig. 2f).

TNF-α promotes anchorage-independent growth, tumor sphere formation and chemoresistance in chordoma cells

We found that long-term TNF-α treatment increased the colony forming capacities of U-CH1 and MUG-Chor1 cells in soft agar. QNZ administration to both long-term TNF-α-treated chordoma cells decreased their colony forming capacities (Fig. 3a). Under non-adherent conditions, we found that TNF-α treatment increased tumor sphere formation in the chordoma cells, and that tumor sphere formation in the long-term TNF-α-treated groups was significantly increased compared to that of the control groups (p < 0.05) (Fig. 3b). We also found that long-term TNF-α-treated cells exhibited significantly increased chemoresistance to paclitaxel and bortezomib compared to that in control cells. Subsequent QNZ treatment decreased the chemoresistance of these cells (Fig. 3c). The concomitant changes in the levels of the drug transporter CD338 (ABCG2) corroborated our findings, which indicates that long-term TNF-α treatment transforms chordoma cells towards a more chemo-resistant state (Fig. 3d). These assays were not conducted in the short-term treatment groups because of the potential early effects of TNF-α on viability.

TNF-α bolsters tumor inflammatory and metastatic pathways in chordomas

Microarray-based expression data revealed that 702 genes were upregulated and 539 genes were downregulated in long-term TNF-α-treated samples compared to the control group. In addition, we found that 351 genes were upregulated and 200 genes were downregulated in the samples treated with TNF-α for one week. Subsequent KEGG pathway enrichment analysis revealed an enrichment in pro-inflammatory pathways both in the long-term and short-term treatment groups. Gene networks and genes related to inflammation and metastasis, such as Toll-like receptor signaling, TNF-α, NF-kB and cell adhesion molecules (CAMs), were found to be affected, with several genes being upregulated after one week of TNF-α treatment (Table 1). In long term TNF-α treated samples, crucial pathways and factors such as Ras and PI3K-AKT signaling pathways, HIF-1 and CAMs were found to be deregulated (Table 2).

Ingenuity pathway analysis revealed that TNF-α acts on a wide range of genes important for tumor progression. In the group treated with TNF-α for one week canonical pathways, including the tumor necrosis factor receptor 2 (TNFR2) signaling, macrophage activation, NF-kB signaling, Toll-like receptor signaling and IL-10 signaling pathways, were found to be affected. TNF-α, IL1B, the NF-kB complex, RelA and TGFB1 were found to rate among the top upstream regulatory genes identified by the analysis. Notable disorders related to these inflammatory responses are cancer and inflammatory diseases. Cell survival, movement and proliferation also ranked among the top features identified by the analysis (Supplementary Fig. 1).

In the long-term treated samples, in addition to the TNF-α downstream and inflammatory pathways, pro-angiogenic pathways were also found to be affected. The top upstream regulators in these analyses were TGFB1, TNF, IL1B, ERBB2 and Wnt3a. Notably, also these pathways are related to cancer and inflammatory diseases. The top molecular functions affected in these samples were cellular movement, development, growth and proliferation, as well as cell-cell interaction and cellular assembly (Supplementary Fig. 2).

For further analysis, networks showing highly consistent gene regulations were investigated. For the short-term TNF-α treated samples, three networks were generated. Supplementary Fig. 3A explains how the figures should be interpreted. Supplementary Fig. 3B includes genes related to elevated inflammatory responses with functions such as activation of antigen-presenting cells, myeloid cell recruitment of phagocytes and cellular immune response. Supplementary Fig. 3C focuses on elevated pathways involved in cell movement, migration and invasion, as well as leukocyte infiltration. Supplementary Fig. 3D focuses on macrophage recruitment. To investigate the effects of long-term TNF-α treatment, Supplementary Fig. 3E was designed, focusing on escalated gene interactions involved in the invasion of tumor cells and the branching of endothelial cells, which potentially mimics late metastatic processes in which intravasation and/or extravasation occurs. Supplementary Fig. 3F focuses on cellular movement in cancer, and the genes involved have previously been associated with pathways related to breast cancer cell movement. Supplementary Fig. 3G focuses on VEGFA downstream pathways, which lead to vascularization and fibroblast mobilization.

The microarray-based expression data discussed are deposited in the NCBI Gene Expression Omnibus [32] and are accessible through GEO Series accession number GSE101866 (ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE101866) for short-term TNF-α treatment and GSE101867 (ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE101867) for long-term TNF-α treatment.

TNF-α acts synergistically with LIF and PD-L1 in chordomas

To determine whether TNF-α treatment affects the expression of the cytokine LIF (and vice versa) in chordoma cells, we used qRT-PCR. In doing so, we found that U-CH1 and MUG-Chor1 cells treated with LIF for 3 weeks exhibited increased levels of TNF and TNFR1 compared to those without treatment (Fig. 4a). Interestingly, we found that long-term TNF-α treatment resulted in elevated levels of LIF and LIFR in both cell lines compared to those in control cells (Fig. 4b). In primary chordoma tumor samples, TNFR1 and LIFR levels were found to be positively correlated (p < 0.05) (Fig. 4c). Additionally, the effects of LIF and TNF-α on PD-L1 gene expression in chordoma cell lines and primary tumor samples were evaluated. We found that long-term TNF-α treatment increased the level of PD‐L1 in both cell lines compared to that in untreated cells (Fig. 4d). LIF treatment did not result in a significant change in PD-L1 expression (data not shown). In the primary tumor samples, both the TNF-α and LIF expression levels were found to be positively correlated with the PD-L1 expression levels (Fig. 4e, f).

TNF-α correlates with clinicopathological chordoma features

Patient-derived chordoma samples were examined for any relationship between TNF-α and TNFR1 gene expression levels and clinicopathological features using Spearman’s rank correlation. By doing so, the Ki67 expression levels in the tumor samples were found to be positively correlated with both the TNF-α and TNFR1 expression levels (Fig. 4g, h). Also, the TNF-α and TNFR1 expression levels were found to be positively correlated in the cohort (Fig. 4i) and the tumor volume was positively correlated with the TNFR1 level (Fig. 4j). In addition, we found that the TNFR1 expression levels were positively correlated with the ZEB2 and negatively with the E-cadherin expression levels (Fig. 4k, l). In our cohort, the TNF-α or TNFR1 expression levels did not correlate with age at onset, metastasis, survival or recurrence, but a high neutrophil to lymphocyte ratio (N/L ratio) was found to be positively correlated with recurrence (Supplementary Table 4) and a short survival time (Fig. 5a).

In addition, paraffin-embedded chordoma tissue samples were analyzed by IHC for TNF-α and TNFR1 expression (Supplementary Fig. 4). We found that TNF-α and TNFR1 expression were positively correlated with recurrence in our patient cohort (Supplementary Table 4) and also that tumor volume was correlated with TNFR1 expression (Fig. 5b). Again, a positive correlation was observed between the TNF-α and TNFR1 expression levels (Fig. 5c). Kaplan-Meier survival analysis revealed that both TNF-α and TNFR1 expression were associated with a decreased overall survival (Fig. 5d, e).

The study was approved by the institutional review board of Yeditepe University Hospital (IRB 98-3943B and 101-4621B), and written informed consent was obtained from all participants. The study was performed in accordance with the Declaration of Helsinki.

Conflict of interest

The authors declare no conflict of interest.