Prognostic factor identification by screening changes in differentially expressed genes in oral squamous cell carcinoma

Objective: This study was designed to identify changes Differences between survival curves by the log- rank test. investigate the association between the expression of different marker proteins and the occur rence of LNM in OSCC patients, Spearman's rank correlation was performed using the respective grouped H- score values (above or below the median) and the pN status (LNM present or not). All of the tests were performed at a significance level of α = 5% using Wizard Pro statistical software (version 1.9.41, by Evan Miller, Chicago, IL, USA) and SPSS (version 24, IBM). Graphs were created with GraphPad Prism software (version 8, GraphPad Software).

microenvironment (TME) (Massagué, 2008). To detach from their surroundings, tumor cells must lose their cell polarity and acquire migratory abilities in an epithelial-mesenchymal transition (EMT) process (Stone et al., 2016). Tumor cells with EMT characteristics have been observed on the invasive front-a region rich in stromal transforming growth factorβ (TGFβ) and other cytokines (e.g., TNFα, EGF, IGF-1) that cooperate in EMT induction (Katz et al., 2013). The TGFβ signaling pathway is the major pathway known to promote tumor cell invasiveness and metastasis (Liao et al., 2019). Little is known about OSCC-specific biomarkers relevant for tumor progression. Their identification might allow more accurate patient's overall estimated prognosis and foster novel targeted tumor therapy (Marcu et al., 2018).
To identify new genes involved in OSCC progression, we treated the human OSCC cell line UPCI-SCC-040 in vitro with TGF-β1 and performed transcriptome analysis. Compared with untreated controls, many DEGs were found that could play a crucial role in OSCC progression. To validate the impact of some selected DEGs on patient prognosis, we evaluated their corresponding protein levels by immunohistochemistry using a tissue microarray (TMA) generated from the tissue samples of 39 OSCC patients, and we correlated the expression profiles with disease-free survival (DFS) as the primary clinical endpoint. We identified stearoyl-CoA desaturase-1 and sclerostin as independent prognostic factors in OSCC.

| Cell culture
The human OSCC cell line UPCI-SCC-040 was obtained from DSMZ.
After 2 hr of starvation, control cells (no TGF-β1 treatment) were collected and frozen. Other cells were treated with TGF-β1 in culture medium without FBS and penicillin/streptomycin for 48 or 72 hr.

| Sandwich ELISA
Concentrations of MMP-9 in the supernatants were measured at dilution 1:100 using commercial DuoSet ELISA kits (R&D systems) according to the manufacturer's instructions.

| Migration assay
UPCI-SCC-040 cells (5x10 4 ) were seeded in 96-well plates, and when confluency was reached, a scratch was made using a toothpick. After detached cells were washed away, the wounded cells were incubated with or without TGF-β1, and images were acquired at immediately (0 hr) or after 15 hr. The distance between the two sides of the scratch was measured at multiple points along the scratch using the ImagePro plus 4.5 software (Media Cybernetics). The average length after 15 hr was subtracted from the average length at 0 hr to determine the length to which the cells migrated.

| Proliferation assay
Tumor cells (5x10 4 cells) were seeded in 96-well plates, and after overnight incubation, the medium was replaced with 0.5% FCSsupplemented medium and incubated with or without TGF-β1 (10 ng/ml) for 48 hr. The CCK-8 solution (Sigma-Aldrich) was then added to the wells according to the manufacturer's instructions, and after 2 hr, the absorbance at 450 nm with 540 nm reference was read.

| Western Blot analysis (WB)
UPCI-SCC-040 cells lysates were prepared in RIPA buffer, and equal amounts of total protein (15 μg/lane) were loaded onto 12% SDS-PAGE and separated by electrophoresis. Proteins were transferred onto a nitrocellulose membrane and after blocking in AdvanBlock-Chemi (R-03726-E10, Advansta), and the membrane was incubated with anti-E-cadherin, anti-vimentin, or antiβ-actin primary antibodies (Abcam) diluted 1:1,000 for 1 hr, then washed three time in 1X TBS with 0.05% Tween-20 and incubated with HRP-conjugated secondary antibody or 1 hr. After additional three washes, the membrane was incubated with Western Bright ECL-HRP substrate (K-12045, Advansta), and protein bands were visualized using the Omega Lum G Imaging System (Aplegen).
The membrane was stripped and re-probed with the next antibody.

| Immunofluorescence (IF)
Tumor cells (2x10 5 cells) were seeded on sterile round cover slips, placed in a 24-well plate, and treated with or without TGF-β1 (10 ng/ ml) for 48 hr. The cells were then fixed with 300 μl of ice-cold methanol for 5 min. Cell were air dried, permeabilized with 0.25% Triton X-100 in PBS for 10 min, and incubated in blocking solution (2% Normal Donkey Serum, 0.1% Triton X-100 in PBS) for 1 hr at room temperature. Then, cells were incubated for another 1 hr at room temperature with the primary antibodies (rabbit anti-human vimentin or E-cadherin, Abcam) diluted 1:500 in blocking solution. After three washes in PBS, cells were incubated with the secondary antibody (Alexa fluor®-555conjugated donkey anti-rabbit IgG, Jackson Immuno-Research labs) diluted 1:1,000 in Blocking Solution for 1 hr. Then, cells were washed three time with PBS for 5 min, incubated in DAPI (300 nM) at room temperature for 5 min, and washed again. Cover slips were mounted on slides with Fluoromount-G, and images were acquired by upright fluorescent trinocular microscope (Olympus BX-60) using the MS60 camera and the MShot Image Analysis System V1 (MSHOT).

| RNA isolation
RNA was extracted using the RNeasy Mini Kit (Qiagen). RNA concentration was measured using a NanoDrop spectrophotometer.
Samples containing 1 µg of total RNA were subjected to quality control and sequencing.

| RNA-seq library preparation and transcriptome analysis
RNA-seq libraries were prepared using the NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs). Libraries were pooled and sequenced on a HiSeq 4,000 sequencer (Illumina) to generate 50 base pair single-end reads.

| Raw read and quality check
Sequenced images were transformed with Illumina BaseCaller to BCL files, which were demultiplexed to fastq files with bcl2fastq.

| Mapping and read quantification
Sequences were aligned to the reference genome Homo sapiens using STAR aligner (Dobin et al., 2013). Subsequent read counting was performed using featureCounts (Liao et al., 2014). Read counts were analyzed in the R/Bioconductor environment using the DESeq2 R package (Love et al., 2014, p. 2). Candidate genes were filtered using an absolute log2 fold change >1 and FDR-corrected p-value <.05. Only the corresponding proteins of the most relevant genes ( Figure 3b) with an absolute log2 fold change >3.55/<−3.55 and a p-value <.05 were considered for further evaluation. Gene annotation was performed using biomaRt R (Durinck et al., 2009). Data are available at GEO under ID GSE16324.  Table 1. For evaluation of clinical baseline characteristics, stages T1 and T2, as well as T3 and T4, were combined into one group. Moreover, AJCC stages 1 and 2, as well as AJCC stages 3 and 4, were combined into one group. Tissue samples were obtained immediately after tumor resection, fixed in neutrally buffered 4% formalin, and embedded in paraffin. TMA was constructed, using the Arraymold TMA kit (60 cores with a diameter of 2 mm). Tissue samples of 26 primary OSCCs, 24 LNM, and 10 OM (internal controls) were used in TMA preparation ( Figure S1). Immunohistochemical reactions (Table S1) were performed on 2µm sections of the TMA using a fully automated slide stainer (Agilent Technologies).

| Image data acquisition and semiautomated immunohistochemical biomarker evaluation
TMA slides were digitally scanned using an Motic EasyScan One whole slide scanner (Motic) at x20 magnification with a resolution of  to identify tissue cores, and the resulting TMA grids were verified manually and changed if necessary. Spot vector and background estimates were used to improve stain separation within the software using color deconvolution (Ruifrok & Johnston, 2001).

| Statistics
The in vitro experiments are presented as means ± SEM (standard error of means) and analyzed with two-tailed unpaired student's acteristics on the DFS was investigated using CR analysis (Gill, 1982).
The strength of the influence was described by the hazard ratio (with F I G U R E 2 TGF-ß1 enhances the EMT process and metastasis. UPCI-SCC-040 cells (5 × 10 4 or 2 × 10 5 cells) were incubated with or without TGF-ß1 for 48 hr, and the expression of the EMT marker E-cadherin and vimentin was evaluated with (a, b) immunofluorescence (n = 7) and (c) Western blot analysis. Additional properties that support the effect of TGF-ß1 on metastasis, such as (d, e) migration (n = 5,7), (f) secretion of MMP-9 (n = 5), and (g) proliferation (n = 9), were also assessed an additional 95% confidence interval). Analyses were first performed in a univariate manner, that is, separately for each factor. Significant factors were further combined in a multiple CR analysis model.

| Confirmation that TGF-β1 induces EMT to acquire metastatic features
To demonstrate that TGF-β1 promotes EMT in UPCI-SCC-040 cells, Collectively, these results confirm that TGF-β1 promotes EMT and enhances the metastatic potential of these cells.

| Sequence annotation and identification of differentially expressed genes (DEGs)
Gene counts from individual samples were normalized, and two dis- For the 3 sample groups (controls, 48 hr and 72 hr of TGF-β1 treatment), differential expression analyses were run as pairwise tests, resulting in 3 distinct differential expression tests: (1)  As such, this analysis confirmed the assumption from the clustering plots that tests (1) and (2) (1) and (2), while only a few overlapped between test (3) and the other tests, indicating that of the most DEGs genes occurred between the controls and TGF-β1, regardless of the different time points (Figure 1a).
To reduce the number of DEGs genes and focus only on the most DEGs ones, differential expression analysis was repeated using more stringent thresholds of adjusted p-values <.0001 and an absolute log2 fold change >3.55. For this new threshold, only 9 genes were found to be DEGs, and their normalized expression levels were plotted in a heatmap (Figure 1b), showing the clear differences in expression levels between the controls and TGF-β1-treated samples regardless of their time points. Antibodies were available for only 6 of the 9 identified genes (SCD1, TNFSF9, SOST, ID3, HIST3H2BB, and MYOM3) so that only these genes could be examined in the further immunohistochemical evaluation.

| Stearoyl-CoA desaturase-1 (SCD1 gene)
Cytoplasmic expression (Figure 4)  higher stearoyl-CoA desaturase-1 expression in primary OSCC than in OM (p < .01) and significantly higher expression in LNM than in OM (p = .02). Stearoyl-CoA desaturase-1 expression in the OSCCs was slightly increased compared with that of LNM, but the difference was not statistically significant ( Figure 4).

| CD137L (TNFSF9 gene, 4-1BBL)
Both cytoplasmic and membranous CD137L expressions were detected by immunohistochemical staining in all tissue types (OM, OSCC, and LNM). In addition, strong membrane expression in the invasive front was observed in the cells adjacent to the TME (Figure 4) (Figure 4).

| Sclerostin (SOST gene)
Cytoplasmic sclerostin expression was found in all tissue types ( Figure 4). However, the overall expression was found at low levels,

| DNA-binding protein inhibitor ID-3 (ID3 gene)
We detected high cytoplasmic DNA-binding protein inhibitor ID-3 expression in all tissue types. In addition, the adjacent cells of the TME showed high expression of ID3 (Figure 4)  [mean H-score 1.08 ± 0.78 SD] was at the lower detection limit, but the cells adjacent to the TME showed a higher degree of expression ( Figure 4). We noted a slightly higher myomesin 3 expression in OSCC than in LNM (p = .04) (Figure 4).

| Association between protein expression, disease-free survival (DFS) and LNM
H-score values of the individual biomarkers were divided into low expression and high expression groups based on the median value (indicated in the graphs, Figure 5). These group values were

| D ISCUSS I ON
Oral carcinogenesis and progression are complex processes involving a variety of phenotypic and genotypic changes (Santosh et al., 2016). OSCC differs from other tumors of the head and neck region (Anderson & Slotkin, 1975;Vossen et al., 2018).
Metastatic spread accounts for the majority of cancer-related deaths despite treatment (Nagafuchi et al., 1987). To invasively migrate and metastasize, tumor cells must lose their cell polarity and break down the intercellular connections that are mainly established by E-cadherin. E-cadherin downregulation has been associated with a transition to the mesenchymal differentiation program (Roche, 2018;Vleminckx et al., 1991). The EMT process is crucial to increasing invasiveness and metastasis (Tracz-Gaszewska & Dobrzyn, 2019).
In this study, we identified changes in gene expression by treating the human OSCC cell line UPCI-SCC-040 with TGF-β1 in vitro, and then, we validated the most important genes in human OM, OSCC, and LNM tissue samples. Furthermore, we translated .57 Note: T stage: T describes the size of the tumor and any spread of cancer into nearby tissue; N stage: N describes spread of cancer to nearby lymph nodes; M stage: M tells whether the cancer has metastasized to distant parts of the body; AJCC stage: A system to describe the amount and spread of cancer in a patient's body, using TNM. the expression profiles for DFS as a primary clinical endpoint for patient prognosis.

SCD1 expression is significantly increased in various human
tumor cells (Holder et al., 2013;Makar et al., 1975;Ran et al., 2018;Wang et al., 2016), and in this investigation, we also demonstrated a significant SCD1 increase in OSCC and LNM compared with healthy OM controls. SCD1 is an important component of fatty acids (FA) de novo biosynthesis, which catalyzes the conversion of saturated fatty acids (SFAs) into △9-monounsaturated fatty acids (MUFAs), and higher MUFAs levels have been found to modulate tumorigenic pathways (Scott et al., 2012). It has been shown that SCD1 promotes migration and invasion through accumulation of MUFAs (Ma et al., 2017). Inhibition of de novo FA synthesis in Srctransformed NIH/3T3 mouse embryo fibroblasts suppressed the formation of invadopodia, which are the membrane structures needed in the process of cell migration. However, this effect was reduced by the addition of oleic acid, the main product of SCD1 activity (Angelucci et al., 2018). Blockade of SCD1 activity with A939572 suppressed the migration and invasiveness of HCC cells and poorly or highly invasive breast cancer cell lines (Mauvoisin et al., 2013), whereas oleic acid restored the original ability to migrate (Wheelock & Johnson, 2003a). SCD1 has been previously proposed as a therapeutic target in the treatment of metastatic breast cancer (Abraham et al., 2018).

One mechanism in which SCD1 might be involved in induc-
ing EMT is the dysregulation of the β-catenin function (Michiels et al., 2017), which is a component of the adherent junctions and interacts with E-cadherin (Wheelock & Johnson, 2003b). SCD1- rates with increased SCD1 expression in the OSCC tissue samples (Liao et al., 2014).
Similar to the study described above, we here identified SCD1 as an independent prognostic factor in OSCC. SCD1 showed the highest impact in transcriptome analysis and in immunohistochemical validation and was significantly correlated with the DFS of patients.
However, contrary to Abraham et al., patients with high SCD1 expression in OSCC and LNM tissue samples showed longer DFS rates than those with low expression, indicating better prognosis in these patients.
Local jaw bone invasion is a crucial step during OSCC progression, and it correlates with increased rates of recurrence and lower OS (Cannonier et al., 2016). Evidence suggests that tumor cells can communicate with osteocytes to create a microenvironment that promotes bone destruction and invasion (Atkinson & Delgado-Calle, 2019).
Sclerostin has been shown to promote the migration, invasion, and bone osteolysis of breast cancer cells (Zhu et al., 2017), and pharmacological inhibition could serve as an effective strategy against bone metastases (Hesse et al., 2019). In the present investigation, we found significant upregulation of the SOST gene after TGFβ treatment and a significant influence of sclerostin expression on DFS in the CR analysis. In an established model of Down syndrome-dependent bone deficiency, Tamplen and colleagues treated mice with an anti-sclerostin antibody and could show that this treatment led to an increase in mandibular bone mass (Tamplen et al., 2019).
Contrary to the results of the presented studies, in which sclerostin expression was associated with a tumor-promoting effect, patients with high sclerostin expression showed a longer DFS than those with low sclerostin expression in the current investigation.
However, sclerostin appears to be an interesting target during locally bone invasive growing OSCC and should be considered in more in-depth studies.
Interactions between CD137 and its ligand CD137L are mainly involved in cytotoxic T-cell activation (Wang et al., 2008) which Urelumab improved cetuximab-activated NK cell survival, dendritic cell maturation, and tumor antigen cross-presentation. The authors concluded that combined immunotherapy using cetuximab and a CD137 agonist has an additional benefit during the treatment of HNC (Srivastava et al., 2017).
In the present study, we describe strong CD137L expression in OSCC and the surrounding TME for the first time. The univariate CR analysis indicated a significant influence of CD137L on DFS; this influence did not reach statistical significance in the multiple analysis model. However, we believe that the CD137/CD137L axis is crucially involved in OSCC progression and could possibly be a biological tool for immunomodulation that requires further investigation.
In this study, we used DFS as the primary clinical endpoint for determining patient prognosis. A growing debate exists on the choice of the correct endpoints in clinical trials in oncology. Thus far, only a modest correlation between disease-free survival and overall survival has been demonstrated (Love et al., 2014). Since overall survival is influenced by many other tumor-independent factors and is also difficult to properly record, we believe that DFS is the better clinical endpoint for capturing the changing tumor biology and the processes of invasion and metastasis. As stated above, our data again suggest the diversity of head and neck tumors and particularly OSCC, rendering further in-depth investigations necessary.
In conclusion, we identified changes in differentially expressed genes during OSCC progression in vitro and associated the most relevant ones with DFS as a primary clinical endpoint. Stearoyl-CoA desaturase-1 and sclerostin acted as independent prognostic factors in OSCC and thus could be of interest as new candidates for the development of targeted tumor therapy.

E TH I C S A PPROVA L A N D CO N S E NT TO PA RTI CI PATE
Patients gave written informed consent before participating the trial.
The study was conducted in accordance with the Ethical Standards

ACK N OWLED G EM ENTS
We thank Marlena Pantakani and Olga Dschun for their scientific and practical assistance in the conduct of the trials. Open Access funding enabled and organized by Projekt DEAL.

CO N FLI C T O F I NTE R E S T
None to declare.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/odi.13879.