Cancer cell line microarray as a novel screening method for identification of radioresistance biomarkers in head and neck squamous cell carcinoma

Currently, no clinically useful biomarkers for radioresistance are available in head and neck squamous cell carcinoma (HNSCC). This study assesses the usefulness of Cell Line Microarray (CMA) method to enhance immunohistochemical screening of potential immunohistochemical biomarkers for radioresistance in HNSCC cell lines. Twenty-nine HNSCC cell lines were cultured, cell pellets formalin-fixed, paraffin-embedded, and arrayed. Radioresistance features of the cell lines were combined to immunohistochemical stains for p53, NDFIP1, EGFR, stem cell marker Oct4, and PP2A inhibitor CIP2A. Expression of p53, EGFR or CIP2A did not indicate intrinsic radioresistance in vitro. Stem cell marker Oct4 nuclear positivity and NDFIP1 nuclear positivity was correlated with increased intrinsic radioresistance. The usefulness of CMA in analysis of HNSCC cell lines and discovery of biomarkers is demonstrated. CMA is very well adapted to both testing of antibodies in a large panel of cell lines as well as correlating staining results with other cell line characteristics. In addition, CMA-based antibody screening proved an efficient and relatively simple method to identify potential radioresistance biomarkers in HNSCC.

Behavior of different cell lines may vary according to the planned experiment e.g. due to differences in protein expression or genomic features. Thus, selection from the multitude of available cell lines happens often by trial and error. Thus, there is a high demand for high-throughput screening method for cell line characteristics.
Tissue microarray (TMA) methods have proven to be an excellent tool for profiling and screening tumor samples [4,5]. Recently, to answer the demand for high-throughput screening methods for cell line characteristics, cell-based microarrays have been introduced [6,7]. The cell microarray (CMA) offers a powerful tool in the evaluation of cell lines. The absence of interference of extracellular matrix facilitates the study of novel antibodies or genomic methods and allows for more reliable specificity analysis of antibodies. The validity of stainings can be assessed by comparing the results to previously acquired data. The included cell lines can be cultured in various conditions, e.g., to include shRNA-treated cells to aid in validation of analysis of proteins of interest. The function of CMA as a fixed, paraffin-embedded biobank of cell lines is of special importance, as it can be stored virtually without limits for later use after the relatively tedious original cell cultures.
In this study, we aimed to assess a carefully selected panel of five immunohistochemical stains of putative radiotherapy biomarkers -p53, EGFR, Oct4, NDFIP1 and CIP2Ain head and neck cancer cell lines. Mutations of the tumor suppressor p53 occur frequently in HNSCC tumors and cell lines [8,9]. While the exact type of p53 genetic alteration may predict HNSCC radioresistance, the role of p53 expression is unclear [10]. The role of epidermal growth factor receptor (EGFR) for therapy resistance of HNSCC has been an active field of investigation, since EGFR inhibitor cetuximab was introduced and approved for use in HNSCC [11,12]. The results of cetuximab-based chemo-or chemoradiotherapy have proven disappointing, despite that EGFR expression seems to associate with HNSCC clinical radioresistance [12]. Oct4 is a stem cell marker which has been linked to radioresistance through cancer stem cell phenomenon and epithelial-mesenchymal transition in several previous studies [13][14][15]. The prognostic relevance of CIP2A to HNSCC has previously been established [13,16]. CIP2A mediates radioresistance in HNSCC and colorectal cancer, and is linked to several oncogenic signaling mechanisms such as c-Myc, p53, EGFR, mTOR signaling and Oct4 [17,18]. Nedd4 family interacting protein NDFIP1 is previously linked to radioresistance of HNSCC through PTEN regulation [19]. Overexpression of NDFIP1 RNA is associated with an unfavorable prognosis in The Cancer Genome Atlas (TCGA) RNA-Seq data [20].

Cell lines
The included 26 UT-SCC cell lines are summarized in Table 1. UT-SCC cell lines were established as previously reported from various HNSCC tumors according to ethical approval by the Ethics Committee of the Hospital District of Southwest Finland as well as patient informed consent [21]. All experiments were carried out according to institutional guidelines and the Declaration of Helsinki. The clinical course of every donor patient's disease was monitored and clinical data from patient was gathered [22,23]. Cell lines were cultured at 37°C and 5% CO 2 in Dulbecco's Modified Eagle's Medium (DMEM), with 10% FCS, glutamine and antibiotics (penicillin and streptomycin). The intrinsic radioresistance of UT-SCC cell lines was previously determined using the 96-well clonogenic assay for radioresistance [8,9,22]. In short, the cells were cultured, plated on 96well plates, exposed to photon irradiation doses ranging from 0.75 to 7.5 Gy the number of dividing cells calculated, and area under the curve (AUC) of cell survival was calculated. For analysis, the mean inactivation dose defined as the AUC of survival curve is used.

shRNA cell lines
Three previously established stable shRNA-transfected UT-SCC cell lines were included, two shCIP2Asilenced variants of UT-SCC-24A and one shCIP2Asilenced variant of UT-SCC-14A. shRNA cell lines were generated using pGIPZ lentiviral vectors consisting of a GFP tag and puromycin resistance (Open Biosystems). Cells transfected with lentiviral vector pGIPZ.NS shRNA containing non-silencing shRNA served as control cells expressing high CIP2A. Two stable cell lines for low CIP2A expression were generated using pGIPZ.shRNA (#556) and pGIPZ.shRNA (#557) containing targeting CIP2A antisense sequences TACATCAGCAGCAAGTTTG and TACTCA ATGTCTTTATGTG, respectively. Lentivirus production followed the standard protocol. At the time of transduction, cell line confluency was approximately 40-50%. After infection, the cells were selected for puromycin resistance, and if the number of GFPpositive cells was low, the cells were sorted with FACS for further experiments. Finally, all cell lines were tested negative for replication competent viruses (RCV test) as well as for Mycoplasma, Acholeplasma, Entomoplasma and Spiroplasma (The MycoAlert™ Mycoplasma Detection kit, Lonza). Western blot using anti-CIP2A antibody (dilution 1:1000, 2G10-3B5, sc-80,659, SantaCruz) was used to confirm the successful silencing of CIP2A expression (Fig. 4A).

CMA construction
Cultured cells were harvested by trypsinization before pelleting the cells. Approximately 40 × 10 6 cells were used for each cell pellet. Pellets were washed with PBS and resuspended in 120-160 μl of 10% neutralbuffered formalin. Cells were then added into a microfuge tube containing a conical fill made of 2% agarose in PBS. After the spin (1000 rpm × 5 min) the supernatant was removed and 10 ml of buffered formalin was added for 48 h, after which the pellets were stored in PBS (+ 4°C). Microfuge tubes were cut open, the pellet transferred into a tissue cassette, and submitted for paraffin embedding (Fig. 1). A microarray was assembled by using the services of Auria Biobank (Auria Biobank, Turku, Finland). The formalin-fixed, paraffin-embedded cell pellets were cut for 6 μm sections and haematoxylin-eosin stained. The slides were scanned and annotated for 0.6 mm cores using Pannoramic Viewer software. The annotated cores were combined into duplicate receiver blocks using TMA Grand Master (3D Histech). Samples from normal human liver were included for orientation.

Scoring of immunohistochemistry
All stainings were analyzed independently by at least two authors (JR, KS, IL, SV), and contradictory cases were discussed until consensus was reached. For microscopic photographs, slides were scanned using Pannoramic 250 Flash slide scanner. Using the CaseViewer software, photographs of either 10-fold or 20-fold magnification were exported in 300 dpi quality. Images were cropped in image editing software, while no color adjustments were performed. p53 staining was analyzed using a previously established 3-tier system consisting of deletion-like absence of p53 staining, wild-type pattern of staining, and mutation/amplification-associated aberrant-type overexpression. EGFR staining was scored in a 4-tier system (0, +, ++, +++) for cytoplasic/membraneous staining pattern. For CIP2A, the intensity of the cytoplasmic/membraneous staining was scored on a scale of 1 to 3 as weak/ negative, intermediate or strong, taking into account the number of positive cells. In patient samples, CIP2A was uniformly present and was scored similarly based on cytoplasmic/membraneous staining intensity. Oct4 immunostaining was analyzed for the presence of positive nuclei. Oct4 positivity was present only in a subpopulation of cells, and thus the presence of strong nuclear immunoreactivity of individual cells was regarded positive. NDFIP1 staining was scored positive, when strong, uniform nuclear staining pattern was present, whereas cytoplasmic staining was not taken into account. A three-tier scoring system was used for uniform nuclear PME-1 and SET staining [24].

Statistical analysis
Cell line data, patient data, and results of immunohistochemistry were entered in SPSS 25 software. Dependence of cell line radioresistance to the staining results was analyzed using General Linear Model statistics. Both main effects and interactions were observed. Estimated marginal means were calculated and 95% confidence intervals determined using bootstrapping. Correlation was determined by Spearman's method. Throughout, p value < 0.05 was deemed significant.

CMA construction
For the CMA construction, UT-SCC cell lines with previous data on radiosensitivity were preferentially selected. Altogether 26 UT-SCC cell lines were included, the majority of which were derived from male patients with oral cavity cancer (Table 1). Six cell lines were derived from metastatic samples, and two cell lines from recurring cancers. In addition, three stable CIP2A shRNA-silenced cell lines were included. The cell lines were cultured until a sufficient number of cells was obtained, whereafter cells were pelleted, fixed in formalin and embedded in paraffin. The following formalin-fixed, paraffin-embedded cell blocks were annotated, and a CMA was arrayed. The CMA construction is demonstrated in Fig. 1A and an example of immunohistochemical assessment in Fig. 1B.

Association of biomarker staining intensities and intrinsic radioresistance
Immunohistochemical stains for p53, EGFR, Oct4, CIP2A, and NDFIP1 were analyzed ( Fig. 2A-L, Table 2). There were significant correlations between Oct4 and NDFIP1 stains (ρ 0.46, p = 0.020), but no correlation between the expression of other biomarkers. Patient age and gender or tumor site, T class or nodal positivity were not directly linked with the expression of any investigated biomarkers.
The associations between staining results and intrinsic radioresistance of the cell lines were analyzed. Expression of p53, EGFR, and CIP2A were not associated with radioresistance ( Fig. 3A-B and D). Interestingly, nuclear NDFIP1 expression was associated with a significant increase in radioresistance (Fig. 3C). In accordance with our previous findings, also Oct4 expression was associated with a significant increase in radioresistance (Fig.  3E). Interaction effects of the biomarkers were analyzed to further explore the possible links between different potential radioresistance mechanisms. There was a significant interaction effect between Oct4 and p53 (Fig. 3F). A trend for a comparable interaction effect between p53 and high EGFR expression was noted (Fig.  3G).

Analysis of CIP2A-shRNA-silenced cell lines
To test the functionality of CMA in the validation of antibody specificity, and to study protein interactions in genetically modified cancer cell lines, three CIP2A shRNA-silenced cell lines were included in the CMA. CIP2A shRNA-silencing was confirmed by Western blot (Fig. 4A) and is demonstrated by the minimal immunoreactivity of the silenced cell lines compared to the parental non-silenced lines (Fig. 4B-F). Parallel immunohistochemistry of the potential radioresistance biomarkers did not reveal significant correlations between CIP2A silencing and p53, EGFR, NDFIP1, or Oct4 expression (Table 3).
Since CIP2A failed to predict radioresistance in the CMA and since CIP2A silencing did not affect other putative biomarkers expression, we were interested, whether CIP2A would have an association with other PP2A inhibitors and performed immunohistochemistry of two well-established endogenous PP2A inhibitors, PME-1 and SET, which are not pronouncedly identified as radiotherapy biomarkers (Supplemental Figure 1). Surprisingly, in all three silenced cell lines the expression of both PME-1 and SET was reduced as well, suggesting a CIP2A-mediated circuitry leading to a more universal loss of PP2A inhibitor expression (Table 3). However, CIP2A, PME-1 and SET expression levels were not correlated across other UT-SCC cell lines of the CMA.
Neither PME-1 nor SET was associated with intrinsic radioresistance of the cell lines.

Discussion
HNSCC consists of a genetically and behaviorally heterogeneous group of malignancies. The observed genomic instability is due to mutagenic insults on the mucosal lining of the upper aerodigestive tract such as tobacco and alcohol [25]. Accordingly, HNSCC-derived  [26][27][28][29]. The field cancerization phenomenon makes radiotherapy an especially inviting treatment option, and it is included in the treatment plan of nearly one half of head and neck cancer patients [30]. Since no biomarkers can currently be used to identify patients benefitting of different treatment strategies, or to explain the clinical diversity in radiotherapy outcomes, the question of radioresistance identification is of tremendous importance for treatment of HNSCC.
Prior studies presenting similar methods of cell line array construction have mostly included relatively small amounts of cell lines, and have been constructed with other than head and neck cancers [6,31,32]. To our knowledge, this is the first CMA constructed with primarily HNSCC cell lines to this large extent. CMA is a practical, cost-efficient and time-saving way to determine the expression of a protein of interest using a specific antibody in a large number of cell lines simultaneously. In this study, we demonstrate that formalin-fixed, paraffin-embedded cell lines may be handled with similar staining protocols used for clinical samples. Cells retained their morphology and overall cellular architecture, leading to good quality staining result, allowing for a reliable evaluation of IHC staining intensity. Loss of CMA spots is low and comparable to TMA methods.
In addition to efficiency, CMA provides other notable benefits as well. Use of exclusively cellular material allows strictly focusing the evaluation of IHC staining positivity on cancer cells and removing the majority of unspecific sources of unwanted IHC signaling positivity. This also provides a fixed, paraffin-embedded biobank of cell lines. It is possible to manipulate and modify the cell lines before constructing CMA, allowing e.g. molecular profiling of cell lines by inhibiting or enhancing the expression or effect of certain genes. This may also be applied to drug development and discovery studies, since and D CIP2A demonstrate no association with radioresistance, whereas C NDFIP1, and E Oct4 have a significant (*) association with the intrinsic radioresistance of the cell lines. F A highly significant interaction effect between p53 and Oct4 was revealed, implying a predictive role of Oct4 in the absence of p53 mutation/amplification type staining. G The interaction trend between p53 and EGFR did not reach significance cell lines may be exposed to drug of interest during culture. Moreover, in a CMA, all types of analysis available for FFPE patient samples can be used. As the CMA paraffin block is handled similarly to FFPE samples, cell line results can directly be compared with patient staining results. Our UT-SCC CMA included three CIP2A-shRNA-silenced cell lines, which demonstrate a highly successful silencing after lengthy cell line culture. CMA offers a powerful tool for the characterization of cellular effects of targeted silencing. A major weakness of cell line experiments in general and thus also of this study, is that cell line models are unable to take into account patient-related factors such as immunological responses, that may decisively affect the success of radiotherapy in HNSCC patients. However, the intrinsic radioresistance of the cancer cells is an important phenomenon in clinical radioresistance [1,9,22]. In the present study, expression of the radioresistance-associated Oct4 and NDFIP1 in patient-derived UT-SCC cell lines could not be explained by correlation with clinical factors, supporting the notion, that radiotherapy response cannot be sufficiently predicted by analysis of the patient clinical parameters alone. Interestingly, other putative radioresistance biomarkers -p53, EGFR, and CIP2Afailed to be associated with intrinsic radioresistance of the cell lines in this relatively large panel of HNSCC cell lines.
Oct4 is a transcription factor essential for pluripotency in embryonic stem cells and testicular maturation. Oct4related stemness has been found to have a role in radiotherapy resistance of various cancers, including HNSCC [13,14,[33][34][35]. Especially interesting, regarding HNSC C, is that Oct4 expression is universally acknowledged to be a favourable biomarker for cisplatin sensitivity in testicular cancer, but seems to predict poor cisplatin response in other cancers [36][37][38]. Since radioresistance is one of the most important issues in the treatment of HNSCC, further studies should investigate the potential  Table 3 Immunohistochemistry of UT-SCC-14 and UT-SCC-24A and corrresponding CIP2A-shRNA-silenced cell lines role of Oct4-related stemness in the clinical radiosensitivity and chemoradiosensitivity of HNSCC. Furthermore, since Oct4 is expressed by a minority of cells as predicted by the cancer stem cell hypothesis, especially Oct4 overexpression experiments would provide essential information on the relationship of therapy resistance and stemness-like characteristics.