A Wnt-mediated phenotype switch along the epithelial–mesenchymal axis defines resistance and invasion downstream of ionising radiation in oral squamous cell carcinoma

Abstract

Background

Dynamic transitions of tumour cells along the epithelial–mesenchymal axis are important in tumorigenesis, metastasis and therapy resistance.

Methods

In this study, we have used cell lines, 3D spheroids and tumour samples in a variety of cell biological and transcriptome analyses to highlight the cellular and molecular dynamics of OSCC response to ionising radiation.

Results

Our study demonstrates a prominent hybrid epithelial–mesenchymal state in oral squamous cell carcinoma cells and tumour samples. We have further identified a key role for levels of E-cadherin in stratifying the hybrid cells to compartments with varying levels of radiation response and radiation-induced epithelial–mesenchymal transition. The response to radiation further entailed the generation of a new cell population with low expression levels of E-cadherin, and positive for Vimentin (ECAD^Low/Neg-VIM^Pos), a phenotypic signature that showed an enhanced capacity for radiation resistance and invasion. At the molecular level, transcriptome analysis of spheroids in response to radiation showed an initial burst of misregulation within the first 30 min that further declined, although still highlighting key alterations in gene signatures. Among others, pathway analysis showed an over-representation for the Wnt signalling pathway that was further confirmed to be functionally involved in the generation of ECAD^Low/Neg-VIM^Pos population, acting upstream of radiation resistance and tumour cell invasion.

Conclusion

This study highlights the functional significance and complexity of tumour cell remodelling in response to ionising radiation with links to resistance and invasive capacity. An area of less focus in conventional radiotherapy, with the potential to improve treatment outcomes and relapse-free survival.

Background

Intratumoral heterogeneity, or the presence of functionally distinct subpopulations within tumours, is a major confounding factor in tumour prognosis and response to therapy.¹ Studies have alluded to different levels of heterogeneity ranging from genetic composition to phenotypic variations, each with functionally unique capacities in tumour propagation and post-therapy remodelling.^2,3

Adoption of distinct but often plastic states along the epithelial–mesenchymal (E–M) axis is a major aspect of tumour heterogeneity with key functions in tumour metastasis and therapy resistance.^4–7 Epithelial-to-mesenchymal transition (EMT) is an evolutionarily conserved process through which epithelial cells lose specialised properties like top–bottom cell polarity and intercellular adhesions and gain mesenchymal phenotypes like front–back cell polarity, enhanced motility and invasiveness.^8,9 Studies in multiple cancer models have established that EMT in tumours is not simply a binary process and instead occurs through distinct intermediate states resulting in the formation of multiple tumour cell subpopulations phenotypically distributed along the E–M spectrum.¹⁰ The prominent intermediary state, also known as partial or hybrid EMT, mediates cellular plasticity, resistance to therapy and metastasis in tumours.^11–16 Although the presence of a hybrid E–M phenotype has been established, its dynamics post-therapy is yet to be fully elucidated. In this study, we have used oral squamous cell carcinoma (OSCC) as a model to further study the cellular and molecular dynamics of the hybrid E–M phenotype in response to ionising radiation (IR).

OSCC is the most common type of oral cancer, accounting for nearly 95% of all diagnosed cases.^17,18 The disease is commonly treated by surgical resection of the tumour mass followed by adjuvant radiotherapy and/or chemotherapy.¹⁹ However, frequent local recurrence and metastatic relapse together with the development of radiation resistance have resulted in a low survival rate of 57% in patients diagnosed with OSCC.²⁰

IR, commonly used in radiation therapy, generates high levels of free radicals that upon accumulation will lead to cell damage and ultimate cell death. It is known that IR can also activate secondary signalling events that cause evasive responses leading to downstream metastasis and resistance.^21,22 The signalling pathway mediated through Wingless/Int-1 (Wnt) is one such signalling cascade and is the focus of the current study. Canonical Wnt signalling is activated by ligand-receptor binding, through stabilisation of cytoplasmic β-catenin that, once translocated into the nucleus, acts as a transcriptional regulator.^23,24 Studies have established that signals from particular Wnt ligands can also be transmitted through two major non-canonical mediator pathways, known as planar cell polarity (PCP) and Ca²⁺-mediated cascades.²⁵ Misregulation of Wnt signalling has been heavily implicated in cancer progression and invasive tumour behaviours.^26–28 However, the role of Wnt signalling in cancer cell responses to radiation and their possible functional feed into tumour invasion and radioresistance is yet to be fully elucidated. Here, using OSCC cells and 3D spheroids, we have delineated the phenotypic dynamics of tumour cells along the E–M axis in parallel to changes in their gene signature, in response to IR. Further, we have uncovered a key role for radiation-induced Wnt signalling pathway in the generation of a radio-resistant population that involves a phenotype switch along the E–M axis.

Methods

OSCC cell lines and general tissue culture

SCC25 and BICR22 were the two established tongue cancer cell lines used in this study. SCC25 and BICR22 cell lines were purchased from American Type Culture Collection (ATCC) and European Collection of Authenticated Cell Cultures, respectively. Both cell lines were confirmed to be authentic using short tandem repeat profiling and were tested to be free of mycoplasma. The cells were kept in culture for no >2 months. Both cell lines were expanded in Dulbecco’s Modified Eagle’s Medium/Nutrient Mixture F-12, GlutaMAX (Gibco) supplemented with 10% foetal bovine serum (FBS) (Gibco) at 37 °C in a humidified incubator with 5% CO₂. All the sorted subpopulations were also grown in the same conditions. As commercially available material, both cell lines were exempt from ethical clearance.

Generation of tumour spheroids

OSCC cells were grown to ~80% confluency in 10 cm culture dishes. The cells were then harvested using the TrypLE enzyme solution (Invitrogen) and resuspended in the complete culture medium. Afterwards, 30 µl drops each containing 100–110 K cells were deposited on an inverted lid of a 10 cm culture dish in order to form hanging drops. The lid carrying the drops was then placed back onto the plate containing Dulbecco’s phosphate-buffered saline (PBS) (Invitrogen) to prevent the drops from evaporating. After overnight incubation of the hanging drops, cell sheets formed at the bottom of each drop were individually transferred into the wells of 6-well plates. The plate was subsequently kept for another 24 h on top of an orbital shaker placed in the incubator to allow the cell sheet to turn into sphere-shaped structures of 500–700 µM in diameter.

Histology, haematoxylin and eosin staining, staining of paraffin-embedded tumour samples and image analysis

Tumour spheroids were prepared for immunohistochemistry using a standard protocol. In brief, they were washed with ice-cold PBS for 5 min, and fixed at room temperature in 4% paraformaldehyde (PFA)/PBS for an additional 10 min. After two washes in PBS for 5 min each, the fixed spheroids were treated with 25% sucrose for ~15 min or until settled to the bottom of the well, and then transferred to cryomolds containing O.C.T. (Emgrid) where they were frozen initially on dry ice and then transferred to −80 °C. Frozen blocks containing the spheroids were cut as 6–8 μm sections on a cryostat (Leica Biosystems) and processed for downstream staining. For haematoxylin and eosin staining, sections of spheroids on glass slides were air-dried and stained by dipping the slide in haematoxylin, followed by rinsing the slide with water, dipping in eosin and finally a quick rinse with water. Afterwards, the slides were dipped sequentially in a range of rising concentrations of ethanol from 70 to 100% and then quickly dried on a hot plate before submerging the sections in the mounting medium to preserve the stain. For immunocytochemistry, OSCC cells cultured as monolayers and the sphere sections were fixed using 4% PFA in PBS. Following an hour of blocking and permeabilisation with 0.1% Triton X-100, 3% donkey serum (Sigma, Cat. # D9663) and 1% bovine serum albumin (BSA) in PBS, samples were incubated with primary antibodies including anti-E-cadherin (ECAD) (R&D Systems Cat. # AF648), anti-Vimentin (VIM) (Abcam, Cat. # 92547 and Cat. # ab8069), anti-γH2AX (Abcam, Cat. # 11174) and anti-pan-Keratin (Abcam, Cat. # ab8068) at 4 °C overnight. The next day, samples were washed with 0.1% Tween-20 in PBS prior to incubating with either Alexa Fluor 555-, Alexa Fluor 488- or Alexa Fluor 405-conjugated donkey anti-rabbit (Abcam, Cat. # 150073 and Cat. # 175649), donkey anti-mouse (Abcam Cat. # 150105 and Life Technologies Cat. # A-31570) and donkey anti-goat (Abcam, Cat. # 150130) secondary antibodies for 1 h at room temperature. Afterwards, the samples were washed, and the cell nuclei were stained with Hoechst 33342 (Invitrogen, Cat. # H3570). Paraffin-embedded OSCC tumour samples were first cut into 4–5 µm sections using a microtome (Leica Biosystems, RM2245). Wax removal was then performed by putting the slides in a sequence of chemicals including Hystalin (5 min, 2 times), 100% v/v ethanol (2 min, 2 times), 95% v/v ethanol (2 min), 70% v/v ethanol (2 min), followed by rinsing with water (1 min). To perform antigen retrieval, the slides were put in citrate buffer pH 6.0 and incubated in the microwave at high power for 15 min, followed by 20 min incubation at room temperature for cooling, and finally 3 min of rinsing with water. For immunostaining, the same protocol was applied as the one used for the sphere sections with a slight difference in the concentration of ECAD and VIM primary antibodies. The final concentrations of each of the used primary antibodies were 1/50 for goat anti-ECAD (R&D Systems, Cat. # AF648) and 1/500 for rabbit anti-VIM (Abcam, Cat. # 92547).

Fluorescent images were acquired using an Olympus U-RFL-T fluorescent microscope and Leica Application Suite. Microscopic images taken of immunostained cells, or sections of spheroids or tumour samples, were analysed using the the ImageJ software.²⁹ For each quantification, at least five fields in each of three biological replicates were quantified and averages were calculated. Likewise, in OSCC tumour samples, five fields were quantified to calculate average percentages. Percentages were calculated based on the number of cells positive for each signal relative to the total number of nuclei.

Extraction of total protein and Western blot analysis

Samples of tumour spheres or SCC25 cells were lysed in RIPA buffer (Thermo Fisher Scientific, Cat. # 89900) supplemented with protease/phosphatase inhibitors (Thermo Fisher Scientific, Cat. # 78440), using sonication pulses created by VEVOR FS-450N Ultrasonic Homogeniser Sonicator, while kept on ice. Cell lysates were then centrifuged at 13,200 r.p.m. for 15 min at 4 °C. Afterwards, the lysates were mixed with Laemmli buffer (Bio-Rad, Cat. # 1610747) loaded onto 4–15% Tris-Glycine Gradient Gel (Bio-Rad), subject to sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto 0.2 μm nitrocellulose membranes. Membranes were then blocked using 5% milk powder in TBS-T (1 M Tris-HCl pH 7.5, 0.8% NaCl, and 0.1% Tween-20) for 15 min prior to incubating with appropriate antibodies. The following primary antibodies were used to detect the protein bands: anti-ECAD (R&D Systems Cat. # AF648), anti-γH2AX (Abcam, Cat. # AB11174), anti-active β-catenin (Sigma-Aldrich, Cat. # 05665), anti-Vinculin (Sigma-Aldrich, Cat. # V9131) and anti-α-Tubulin (Santa Cruz, Cat. # sc-5286). Membranes were incubated at 4 °C overnight with primary antibodies diluted in 5% BSA in TBS-T and subsequently incubated for an hour at room temperature with fluorescent secondary antibodies IRDye 680RD (LI-COR Biosciences, Cat. # 925-68072) and IRDye 800CW (LI-COR Biosciences, Cat. # 925-32213) diluted in the similar buffer used for the primary antibody dilution. Fluorescent signals were detected using the Odyssey Fc Imaging System (LI-COR Biosciences). The images were acquired and analysed with Image Studio software (LI-COR Biosciences).

Total RNA isolation and real-time reverse transcription (RT-PCR) analysis

Total RNA was extracted from either OSCC cells or tumour spheres using Isolate II RNA Mini Kit (Bioline, Cat. # BIO-52073) according to the manufacturer’s instructions. RNA samples were then converted to complementary DNA (cDNA) using SensiFAST cDNA Synthesis Kit (Bioline, Cat. # BIO-65054). Quantitative RT-PCR (qRT-PCR) was subsequently performed using SensiFAST SYBR Lo-ROX Kit (Bioline, Cat. # BIO-94020). In order to study the EMT state of the cells, seven primer pairs were used to amplify ECAD as the epithelial marker along with four EMT transcription factors (TFs), including SNAIL, SLUG, TWIST and ZEB1, as well as VIM and N-cadherin (NCAD) as mesenchymal markers. Expression levels of several key genes involved in the Wnt signalling pathway were also assessed. All gene expression tests were performed at least in duplicates and in three biological replicates. GUSB was amplified as the housekeeping gene in each sample after initial confirmation for consistent levels across tested conditions using a published methodology.³⁰ In all cases, the delta-delta Ct method (ΔΔCt) was used to analyse the relative changes in the transcript levels. In brief, ΔCt was first calculated based on the transcript Ct vs GusB Ct, and this value for both control and treatment samples was used to further calculate the ΔΔCt. Where magnitude of expression for particular markers was to be assessed (i.e. Fig. S1B), ΔΔCt was calculated relative to the levels of normalised transcript expression (ΔCt) in human dermal fibroblasts (mesenchymal reference) or human-induced pluripotent stem cells (epithelial reference) to reflect the degree of epithelial and mesenchymal marker expressions, respectively. The statistical significance was measured through t test and analysis of variance (ANOVA) methods as explained later. For all primer sequences refer to Table S1.

ECAD staining for flow cytometry and cell sorting

Cultured cells were first harvested using StemPro® Accutase® Cell Dissociation Reagent (Gibco) and then washed once using ice-cold 2% FBS in PBS. Afterwards, the cells were stained with anti-ECAD (R&D Systems, Cat. # AF648) and donkey anti-Goat IgG H&L Alexa Fluor® 488 (Abcam, Cat. # ab150129). In brief, the cells were incubated with appropriate concentration of either antibody for half an hour on ice, followed by three washes using ice-cold 2% FBS in PBS. Cells were later resuspended in 200 µl of ice-cold 2% FBS in PBS. DAPI (4′,6-diamidino-2-phenylindole) was also added to distinguish between the live and dead cells. Flow cytometry analysis was then performed on BD FACS Canto-II Flow cytometer.

Cell sorting to isolate cells expressing (different levels of) ECAD was performed on FACS Aria III (BD) and data were analysed by FACS Diva software (BD). Ten percent of the cells that expressed ECAD at the highest levels (on the right shoulder of the histogram for ECAD-positive cells) were sorted as ECAD^High cells, whereas the 10% expressing ECAD at the lowest levels (on the left shoulder of the histogram for ECAD-positive cells) were sorted as ECAD^Low/Neg population. For all experiments, DAPI^neg cells were sorted as controls (parent).

Apoptosis assay

Annexin V staining was performed using the eBioscience™ APC-Annexin V Apoptosis Detection Kit (Invitrogen, Cat. # 88-8007-72). Briefly, SCC25 cells were irradiated with 2 Gy of IR, and afterwards, the incidence of apoptosis in exposed cells was assessed at two post-IR time points, 24 and 48 h. According to the manufacturer’s instructions, cell pellets were collected by centrifugation after washing the cells with PBS and trypsinisation. The cell pellets were subsequently washed once with PBS and then with 1× binding buffer. Afterwards, the cells were resuspended in 1× binding buffer at a concentration of 10⁶/ml prior to adding 5 μl of APC-conjugated Annexin to 100 μl of the cell suspension. After 15 min of incubation at 15–25 °C, the cells were once washed with 1× binding buffer and 5 μl of propidium iodide was added with analysis conducted by flow cytometry right after this step using a BD FACS Canto-II flow cytometer. The experiments were performed in three biological replicates.

Growth curve analysis and clonogenic assays

For growth curve analysis, 40,000 cells of each cell population were seeded in triplicates in 6-well plates. The cells grown in each well of every triplicate were counted every day for 8 consecutive days afterwards. The growth curves were then created and the doubling time for each group was calculated according to the following formula based on ATCC instructions: doubling time = t*ln²/ln(N(t)/N(0)), where N(t) is the number of cells at time t, N(0) is the number of cells at time 0 and t the incubation time.

For clonogenic assays, 400 cells were seeded in each well of 6-well plates 4 h prior to being irradiated. In parallel, triplicates of 200 cells were seeded and grown as the non-irradiated controls for calculating the PE. All the cells were let to grow into colonies for 10 days before fixing in methanol for 10 min, followed by staining with 0.5% crystal violet. Only colonies with a minimum number of 50 cells were counted. Plating efficiency (PE) and the surviving fraction (SF) were calculated for each experiment based on the following formulas:³¹

$$\mathrm{PE}=\frac{\mathrm{no}\mathrm{.}\mathrm{of}\mathrm{colonies}\mathrm{formed}}{\mathrm{no}\mathrm{.}\mathrm{of}\mathrm{cells}\mathrm{seeded}}\times \mathrm{100}\%,\mathrm{SF}=\frac{\mathrm{no}\mathrm{.}\mathrm{of}\mathrm{colonies}\mathrm{formed}\mathrm{after}\mathrm{treatment}}{\mathrm{no}\mathrm{.}\mathrm{of}\mathrm{cells}\mathrm{seeded}\times \mathrm{PE}}$$

Library preparation, RNA-sequencing and data analysis

Using Isolate II RNA Mini Kit (Bioline, Cat. # BIO-52073), total RNA was isolated from irradiated spheres at 30 min, 2 h and 24 h post 2 Gy of IR. RNA was also isolated from two control samples one as the control for the first two time points, half an hour and 2 h and the other one grown for an extra 24 h as the control for the 24 h post-irradiation time point. The whole messenger RNA content of each sample was converted to cDNA using TruSeq RNA Library Prep Kit (Illumina, Cat. # 20020595) to prepare the cDNA libraries, which were then sequenced on an Illumina Hiseq2500 System on 100 bp single-end mode by the Australian Genome Research Facility.

Reads from the sequencing were aligned to the hg38 genome with STAR v2.5.3a,³² discarding multi-mapping reads. Aligned reads were then summarised to gene counts using the gencode v26 gene annotation. Gene counts were normalised using tumour node metastasis³³ to account for library size differences. Tests for differential expression between groups were performed using voom³⁴ with a robust estimation of the prior variances. Adjusted p values that account for multiple comparisons were calculated using a Bonferroni correction with a threshold of 0.05 used to call statistical significance. Pathway analysis was performed using Wilcoxon’s rank-sum tests on the Gene Ontology biological processes downloaded on 3 May 2016.³⁵ A copy of raw data containing fastq files has been deposited on SRA under BioProject PRJNA611666 and is accessible through this link: https://dataview.ncbi.nlm.nih.gov/object/PRJNA611666?reviewer=bl6v8sfovsg5ob4dkm3vphutg6.

OSCC tumour samples and human ethics protocol

Paraffin-embedded samples from human primary squamous cell carcinoma of the lower right alveolus (pre-radiotherapy) and secondary/relapsed squamous carcinoma of the upper right alveolus (post-radiotherapy) were obtained from Professor Camile S. Farah’s laboratory at the Australian Centre for Oral Oncology Research and Education (Nedlands, WA, Australia). The ethics protocol was approved by the University of Western Australia human research ethics committee (Protocol #RA/4/1/8562) and written consent from the patient was obtained in accordance with the Declaration of Helsinki.

In vitro invasion assay, inhibition of Wnt signalling and irradiation

Cells resuspended in serum-free medium were seeded in 6.5 mm Transwell® with 8.0 µm pore size polyester membrane inserts (Corning, Cat. # 3464) with a density of 50,000 cells per each well. Prior to this step, inserts were coated with Matrigel (BD Biosciences, Cat. # 356330). The lower chambers were filled with a medium containing 10% serum as the chemoattractant. After incubating at 37 °C for 20 h, the non-invaded cells were wiped out of the inner surface of the inserts’ membrane, while the cells invaded through the membrane were fixed with methanol and stained with 0.5% crystal violet. Images of at least five different fields were then taken using an inverted microscope (Leica CTR 6000) and the average cell numbers were calculated and normalised to the control.

To inactivate the Wnt signalling pathway, 0.5 μM of small-molecule IWP2 (Sigma, Cat. # 10536) was added to the cell culture medium immediately after irradiating SCC25 or BICR22 cells or spheres.

Cells/spheres were exposed to 2 Gy of X-ray in the chamber of X-RAD 320 (Precision X-Ray) irradiator at a dose rate of 1 Gy/min. Irradiated cells/spheres were then incubated at 37 °C in a humidified 5% CO₂ incubator for the defined period before any of the further post-irradiation analyses.

Statistical analysis

All obtained data are reported as average numbers and standard deviation (SD) of at least three independent biological replicates is displayed as error bars. Statistical analyses were performed using ANOVA (GraphPad Prism version 7.00; GraphPad Software, San Diego, CA, USA; www.graphpad.com) and Student’s t tests (Excel, Microsoft) to calculate the significance of the results. Comparisons with p values <5% (p < 0.05) were regarded significant.

Results

OSCC cells and tumour spheroids manifest a partial EMT state

Several studies have demonstrated that EMT presents at multiple stages along the E–M axis with prominent semi-stable states of the so-called partial EMT.^14,36 To gauge the E–M status of oral squamous carcinoma cells, we assessed the expression patterns of ECAD and VIM, canonical markers of epithelial and mesenchymal states, respectively,^37,38 in two human OSCC cell lines, SCC25 and BICR22 (Fig. 1a). Immunofluorescent labelling of cells grown in monolayer cultures demonstrated heterogeneity in the expression of ECAD and VIM in both cell lines. The majority of cells expressed ECAD, although at variable levels, and cytoplasmic expression was evident in minority populations (Fig. S1A). Further analysis of VIM expression confirmed a significant subpopulation of cells expressing both ECAD and VIM (SCC25 = 80–90%; BICR22 = 30–40%), suggesting a hybrid E–M state present in both cell lines. Likewise, real-time RT-PCR detected transcripts for ECAD and four key EMT master regulators SNAIL, SLUG, TWIST and ZEB1 in both cell lines, further confirming the co-occurrence of both epithelial and mesenchymal programmes (Fig. S1B).

Fig. 1. The partial EMT state in OSCC cells and spheroids.

a I and II: bright-field images of SCC25 and BICR22 cells (scale bar: 50 μm). III and IV: Immunocytochemistry of SCC25 and BICR22 cells with antibodies against E-cadherin (red) and Vimentin (green). Nuclei are counterstained with Hoechst (blue) (scale bar: 100 μm). b Diagram demonstrating the protocol for the generation of tumour spheroids. c Haematoxylin (blue) and eosin (red) staining of cryopreserved sections from SCC25 tumour spheroids (scale bar: 100 μm). d Immunohistochemistry of cryopreserved sections from SCC25 spheroids with antibodies against E-cadherin, Vimentin and pan-Keratin. ECAD: E-cadherin, VIM: Vimentin, pan-K: pan-Keratin.

To further explore the patterns of E–M gene expression in a 3D model, we generated tumour spheroids³⁹ from SCC25 cells and re-assessed expression patterns of epithelial and mesenchymal markers (Fig. 1b–d). Histological analysis of the spheroids showed a compact outer layer of cells recapitulating flattened squamous epithelium with an inner morphologically heterogeneous core more akin to the spinous layer of stratified squamous epithelium (Fig. 1c). Similar to cells grown in monolayers, the majority of the cells grown in spheroids expressed epithelial markers (ECAD and pan-Keratin) (Fig. 1d). In addition, there was a minor subpopulation with low or negative levels of ECAD or pan-Keratin that expressed VIM (Fig. S1C). The majority of cells expressing VIM, however, co-expressed either ECAD or pan-Keratin (Fig. 1d). These observations support the presence of a prominent subpopulation in OSCC cells that co-express epithelial and mesenchymal markers, suggesting a hybrid E–M character and/or a state of partial EMT.

Levels of ECAD correlated with the degrees of response to IR and radiation-induced EMT remodelling

Levels of ECAD correlate with poor prognosis in patients with head and neck carcinoma and a range of other cancers.^40,41 Although the majority of cells in both BICR22 and SCC25 cell lines expressed ECAD, the expression levels in these cells varied (Figs. 1 and 2a). We, therefore, asked if subpopulations expressing different levels of ECAD had different functional properties. To assess this, we isolated the cells with different levels of ECAD from the SCC25 cell line by introducing an arbitrary 10% cut-off in the ECAD flow cytometry signal and termed the cells isolated from the 10% lower and higher ends of the graph ECAD^Low/Neg and ECAD^High, respectively (Fig. 2a). Live (DAPI^Neg) parent cells were used as controls. Expression analysis of EMT TFs did not show a significant enrichment for these transcripts in ECAD^Low/Neg subpopulation with only ZEB1 expressed at lower levels in ECAD^High cells compared with both ECAD^Low/Neg and parent cells (Fig. S2B). As expected, ECAD^Low/Neg cells expressed significantly lower levels of the ECAD transcript (Fig. S2A) and both subpopulations retained signature levels of ECAD^Low/Neg and ECAD^High cells after several rounds of division across from 2 to 3 passages (Fig. S2B, C). No significant difference in cell growth or doubling times was seen between these cell populations (Fig. S2D).

Fig. 2. E-cadherin expression levels determine radiation response and EMT remodelling post-IR.

a Top: Flow cytometry histograms depicting the isolation strategy for ECAD^Low/Neg and ECAD^High cells from SCC25. (ECAD^pos: E-cadherin positive). Bottom: Experimental design for (b, c). b Left: Representative images of the colony assay depicting survival in control vs irradiated (2 Gy radiation; IR) in different subpopulations. Right: quantification results of triplicate experiments (*p < 0.05). c Real-time qRT-PCR analysis of total RNA from spheroids derived from different subpopulations, for the expression of transcripts for EMT transcription factors SNAIL, SLUG, TWIST and ZEB1 and mesenchymal markers N-cadherin (NCAD) and Vimentin (VIM) (*p < 0.05, **p < 0.01).

To assess the inherent capacities of these subpopulations for a response to IR, all cell populations were exposed to 2 Gy of IR and the effect on colony formation was assessed 10 days post-IR through a standard colony assay.³¹ Quantification of surviving colonies (SF) confirmed a significantly higher capacity in the ECAD^Low/Neg population to resist the effects of IR compared to the other tested populations (Fig. 2b). It is known that irradiation-induced EMT has a major role in tumour cell remodelling post-IR.^42–44 Noting the absence of biased signatures with regard to the expression of EMT TFs among these cells, we asked if ECAD^Low/Neg, ECAD^High and parent subpopulations showed any difference in propensities for EMT induction after exposure to IR. For this, we generated tumour spheroids from all subpopulations and analysed the expression of EMT TFs in control spheroids and irradiated spheres at 2 and 24 h post-IR time points. Spheroids formed from ECAD^Low/Neg cells maintained significantly lower levels of ECAD transcripts (Fig. S2E). Our analysis confirmed major EMT induction by IR in ECAD^Low/Neg population, with significant induction in SLUG at 2 h and SNAIL, SLUG, TWIST1 and ZEB1 at 24 h post-IR (Fig. 2c). To further gauge the expression of mesenchymal markers, we also assessed the expression of the specialised mesenchymal cell junction protein, NCAD,⁴⁵ and the canonical mesenchymal marker, VIM.³⁸ Interestingly, expression of both markers was induced at the 24 h post-IR time point in spheroids from ECAD^Low/Neg and ECAD^High populations, despite an initial decline in the expression of NCAD in ECAD^Low/Neg spheroids (Fig. 2c). Altogether, our findings so far support a key role for levels of ECAD in assigning signatures for resistance and EMT-related remodelling in tumour cells in response to IR.

IR induces a phenotype switch to ECAD^Low/Neg-VIM^Pos in 3D spheroids and OSCC tumour samples

Our data support the presence of a hybrid E–M state in subpopulations of OSCC cells within the tumour spheroids. However, it has been reported that the semi-stable hybrid E–M state can change in response to stress or other environmental stimuli.⁶ To test the response of the hybrid E–M state of OSCC cells to IR, we exposed the SCC25-derived spheroids to 2 Gy of X-ray and analysed the expression of ECAD and VIM at 30 min, 2 h and 24 h post-IR (Fig. 3a and Fig. S3A). No significant induction of apoptosis was observed within the first 48 h of IR treatment, providing a safe, cell death-free window to track the possible phenotypic remodellings (Fig. S3B). Analysis of subpopulations positive for ECAD and VIM revealed a significant decline in the overall levels of ECAD, 24 h post-IR, with no change in VIM (Fig. 3a). No significant change was observed for either marker at 30 min or 2 h time points (Fig. S3A). Further, a detailed analysis of subpopulations expressing either ECAD or VIM confirmed the presence of a cell population with low to negative levels of ECAD that expressed VIM (ECAD^Low/Neg-VIM^Pos) (Fig. 3a). With initial observations supporting a time frame of dynamic cellular activities within the first 24 h post-IR, we checked the levels of γH2AX, the specialised histone protein that is commonly recruited to sites of double-strand breaks (DSBs),⁴⁶ at 30 min, 2 h and 24 h time points. The levels of γH2AX peaked at 2 h post-IR in spheroids derived from SCC25 cells, with a significant decline observed at 24 h (Fig. S3C, D), suggesting a temporal sequence of accumulation of DSBs, followed by possible resolution of the lesions at the 24 h post-IR time point. Of note, quantification of γH2AX signals in ECAD^Low/Neg, ECAD^high and parent populations did not show any significant difference between population-specific DSBs at the tested time points (Fig. S3E).

Fig. 3. The generation of ECAD^Low/Neg-VIM^Pos cells post-IR.

a Immunohistochemistry for E-cadherin (ECAD: red) and Vimentin (VIM: green) in spheroids 24 h after exposure to 2 Gy of radiation (24 h post-IR) vs control. Nuclei are counterstained by Hoechst (blue). Bar graphs represent the quantification of experimental triplicates (*p < 0.05). b Immunofluorescence on tumour samples from primary squamous cell carcinoma of the lower right alveolus (pre-IR) and secondary squamous cell carcinoma from the upper right alveolus (post-IR) with antibodies against E-cadherin (ECAD: red) and Vimentin (VIM: green) counterstained with Hoechst (blue). Arrows highlight the emergence of ECAD^Low/Neg/VIM^pos cells in the post-IR sample. Bar graphs demonstrate the average quantification of each cell population in 5–10 different fields (*p < 0.05, **p < 0.01; scale bar: 100 μm; SM: submucosa).

Our observations in SCC25-derived spheroids was indicative of a phenotype switch along the E–M axis in the OSCC cells. Given the elevated level of cellular and molecular complexity in tumours, we asked if a similar cellular transition was also present in OSCC tumours before and after a course of radiotherapy. We, therefore, obtained OSCC tumour samples from a primary OSCC in the lower right alveolus and a secondary OSCC in the upper right alveolus, both from an 80-year-old male patient who had first presented with a primary disease followed by a relapse after a complete course of radiotherapy. These samples were termed pre- and post-IR, respectively. The expression of ECAD and VIM in these samples showed the expected primary accumulation of ECAD in the neoplastic epithelium and VIM in the submucosal tissue (Fig. 3b). Further analysis, however, showed a scattered pattern of VIM expression in the ECAD^Pos epithelial layer. In the pre-IR samples, VIM-expressing epithelial cells accounted for ~20% of the entire ECAD^Pos population with a minority of cells (1–2%) presenting as ECAD^Low/Neg-VIM^Pos. In post-IR samples, however, a marked increase in the ECAD^Low/Neg-VIM^Pos was detectable in the epithelial layer (~8%) with a decline in ECAD^pos-VIM^Pos (Fig. 3b). Taken together, these observations confirm a phenotypic transition to ECAD^Low/Neg-VIM^Pos cells downstream of IR.

The phenotypic transition to ECAD^Low/Neg-VIM^Pos reflects altered radioresistance and invasive capacity

Given our earlier observations with a post-IR transition of otherwise hybrid cells to a ECAD^Low/Neg-VIM^Pos population, we asked if this phenotype switch leads to differences in response to IR or invasive capacities. For this, we isolated the cells with low and high levels of ECAD from SCC25 cells at the 24 h post-IR time point and control (ECAD^Low/Neg and ECAD^High) with parallel DAPI^Neg whole population cells used as sorting control (parent) (Fig. 4a). As expected, the number of cells with the ECAD^Low/Neg signature increased significantly after IR, confirming our earlier observations of an overall decline in ECAD levels (Fig. S4A). Further, real-time qRT-PCR showed an enriched expression for VIM in ECAD^Low/Neg cells, confirming that isolation of this population can efficiently capture the VIM^Pos population post-IR (Fig. S4B).

Fig. 4. Functional properties of the ECAD^Low/Neg-VIM^pos population.

a Experimental design for b and c (a). b Representative images from colony assays in control vs irradiated subpopulations. Bar graph depicts quantification of survival rates between experimental triplicates (*p < 0.05, **p < 0.01). c Representative images from Matrigel Transwell invasion assay in control and irradiated (IR) subpopulations. Bar graph depicts invasive capacity normalised to parent, averaged between experimental triplicates (***p < 0.001; scale bar: 100 μm).

The radiation sensitivity and invasiveness of ECAD^Low/Neg, ECAD^High and parent cells were investigated in downstream colony formation and invasion assays. Assessment of colonies formed 10 days post-plating from subpopulations confirmed a higher survival in colonies derived from ECAD^Low/Neg cells compared to the other populations tested (Fig. 4b). Both parent and ECAD^High cells formed similarly low number of colonies post-IR. However, the significantly higher capacity for colony formation in control ECAD^High cells confirmed an enhanced response to irradiation in this subpopulation. Using Matrigel-based Transwell invasion assay,⁴⁷ we then assessed the invasive capacities of these subpopulations. ECAD^Low/Neg cells showed higher levels of invasion in the absence of irradiation, compared to parent cells (Fig. 4c). Likewise, post-IR, higher levels of invasion was observed in ECAD^Low/Neg cells compared to irradiated parent cells (Fig. 4c). ECAD^High cells did not survive the invasion assay post-IR due to significant IR-induced cell death. Overall, our findings support an IR-induced phenotype switch to ECAD^Low/Neg-VIM^Pos with higher resistance to IR and retained invasive capacity.

Analysis of transcriptional dynamics in irradiated tumour spheroids highlights Wnt signalling pathway as a major player

The changes in cell phenotypes and variations in the levels of DNA lesions observed, implied highly dynamic molecular remodelling in SCC25 tumour spheroids within the first 24 h post-IR. We, therefore, performed RNA-sequencing on total RNA derived from tumour spheroids irradiated with 2 Gy of IR at 30 min, 2 h and 24 h time points, and control non-irradiated spheroids (Fig. 5a). Analysis of gene transcripts differentially expressed (DE) between irradiated and control samples confirmed a total of 3418 DE genes at 30 min post-IR, highlighting a drastic degree of transcriptional change at this early time point in response to radiation (Fig. 5b). The total number of DE genes further declined to 1763 and 468 at the 2 and 24 h time points, respectively, suggesting gradual resolution of the initial transcriptional shock. Hierarchical clustering of DE genes was used to identify six distinct clusters (Fig. 5c and Fig. S5A). Genes in cluster 1 showed an overall increasing expression trajectory in control samples within 24 h of spheroid growth and showed further enhancement in response to radiation (Fig. 5c and Fig. S5A). In clusters 2, 3 and 6, a significant increase in expression levels was observed at the early 30 min time point, which was further followed by a decline in expression at 2 h, to almost comparable levels between control and IR samples at the 24 h time point (Fig. 5c and Fig. S5A). In contrast, genes in clusters 4 and 5 showed an initial high expression level with a significant decline at 30 min post-IR that further resolved to comparably low expression levels in both control and IR samples at the 24 h time point (Fig. 5c and Fig. S5A). We performed further pathway enrichment analysis on the DE genes taking the direction of change (up-regulation vs down-regulation) into account (Fig. 5d, Fig. S5B and Table S2). A major decline in expression was observed in pathways associated with ribosome biogenesis, metabolism and RNA synthesis, at all three time points, highlighting a general decline in cellular metabolic activities in response to radiation-induced stress (Fig. 5d and Fig. S5B). Despite the decline in RNA biogenesis, however, significant up-regulation of epithelial pro-differentiative programmes associated with skin and epidermis development, and differentiation of epidermal cell and keratinocytes was observed at all three time points (Fig. 5d and Fig. S5B). In addition, a dynamic pattern of expression for human leukocyte antigens (HLAs) and related immune-modulatory factors like tumour growth factor-β (TGFβ) and C-X-C motif chemokine ligand 8 (CXCL8) (interleukin-8 (IL-8)) was also observed (Fig. S5C). Also, outstanding was the enrichment for genes associated with Wingless/Int-1 (Wnt) signalling pathway, with over-representation among the first 10 up-regulated pathways at the 30 min time point, which further declined, although still significantly over-represented at 2 h and 24 h post-IR (Fig. 5d–f). Signalling initiated by Wnt ligands is transmitted through both canonical and non-canonical downstream mediators.⁴⁸ Detailed analysis of the Wnt-associated DE genes and subsequent targeted analysis by real-time qRT-PCR confirmed the involvement of canonical/β-catenin, non-canonical/PCP and non-canonical/Ca²⁺-mediated pathways (Fig. 5f, g). Analysis of selected Wnt ligands by real-time qRT-PCR further demonstrated a tentative pattern of activation for Wnt ligands associated with the canonical pathway Wnt7A and Wnt4 at early time points, followed by activation of Wnt11 and Wnt5A, ligands that commonly signal through non-canonical PCP and Ca²⁺-mediated pathways, respectively^19,48 (Fig. 5g). Altogether, these findings support major transcriptional dynamics in 3D tumour spheroids within the first 24 h post-radiation, with significant involvement of canonical and non-canonical Wnt signalling pathways.

Fig. 5. Transcriptional dynamics of irradiated SCC25 tumour spheroids and involvement of Wnt signalling pathway.

a The experimental design for RNA-sequencing analysis of tumour spheroids. b Number of differentially expressed (DE) genes in irradiated spheroids vs respective controls. c Heatmap and hierarchical clustering of the top 100 differentially expressed transcripts among different samples, tested in triplicates. Six different clusters are marked based on patterns of change. d Pathway enrichment analysis at 30 min post-IR vs control. The top 10 up- and down-regulated pathways are demonstrated. e Statistical significance of Wnt pathway activation at different time points post-IR. f Heatmap and hierarchical clustering for DE transcripts associated with the Wnt pathway. g Real-time qRT-PCR analysis of total RNA derived from irradiated and control spheroids for different transcripts associated with canonical and non-canonical (PCP and Ca²⁺-mediated) Wnt pathways (*p < 0.05, **p < 0.01, ***p < 0.001).

The Wnt signalling pathway acts upstream of the radiation-induced phenotype switch to ECAD^Low/Neg-VIM^Pos to affect invasion and radioresistance

The common temporal patterns in the emergence of the ECAD^Low/Neg-VIM^Pos population and the induction of Wnt pathway mediators implied a possible regulatory link between the two events. We, therefore, asked if the radiation-induced Wnt signalling pathway acted upstream of the phenotype switch to ECAD^Low/Neg-VIM^Pos. To assess this, SCC25 and BICR22 3D spheroids were exposed to 2 Gy of IR and treated with the small-molecule inhibitor, IWP2,⁴⁹ to inhibit Wnt signalling during 24 h post-IR. IWP2 is known to inhibit the porcupine enzyme, hence impairing the processing and secretion of a variety of Wnt ligands.⁴⁹ As anticipated, treatment with IR resulted in enhanced levels of active β-catenin protein, the marker for activated canonical Wnt pathway⁴⁸ starting from 2 h post-IR time point (Fig. S6A). Further, treatment of tumour spheroids with IWP2 resulted in diminished expression of active β-catenin at the 24 h post-IR time point (Fig. S6A). At the cellular level and as observed before, 24 h post-IR a significant increase in the abundance of ECAD^Low/Neg-VIM^Pos cells was evident in SCC25-derived, compared to non-irradiated control spheroids (Fig. 6a). Treatment of the irradiated spheroids with IWP2, however, significantly impaired the phenotype switch, with ECAD^Low/Neg-VIM^Pos present at levels comparable to non-irradiated control spheroids (Fig. 6a). In accordance, real-time qRT-PCR confirmed an increase in the levels of ECAD transcripts, concordant with increased ECAD area in spheroids treated with IWP2 post-IR (Fig. S6B, C). Treatment of control non-irradiated spheroids with IWP2 did not result in any significant change in the number of ECAD^Low/Neg-VIM^Pos cells, indicating a specialised role for the Wnt signalling pathway in the context of radiation-induced cellular remodelling (Fig. 6a). Wnt signalling is implied in therapy resistance and invasion.^25,50 We, therefore, asked if inhibition of radiation-induced Wnt pathway could affect either of these properties. Clonogenic assays performed on irradiated SCC25 and BICR22 cells treated with IWP2 demonstrated a significant decline in the SF starting from 2 days of treatment, compared to irradiated controls (Fig. 6b and Fig. S6D). Treatment with IWP2 had no significant impact on the PE of control non-irradiated cells (Fig. S6E and data not shown). Further, Transwell invasion assay confirmed a significant decline in the number of invasive SCC25 cells post-IR with IWP2 treatment (Fig. 6c). Treatment of control non-irradiated cells with IWP2 did not result in any significant change in invasive capacity (Fig. 6c and Fig. S6F). Altogether, our findings support a key role for the Wnt signalling pathway upstream of the ECAD^Low/Neg-VIM^Pos phenotype switch, impacting radiation response and cell invasion.

Fig. 6. Functional involvement of the Wnt signalling pathway in the generation of ECAD^Low/Neg-VIM^Pos cells in spheroids, and in radiation response and invasion.

a Immunohistochemistry for E-cadherin (ECAD; red) and Vimentin (VIM; green) on control and irradiated spheroids with and without IWP2. Nuclei are counterstained with Hoechst (blue). Arrows point to examples of ECAD^Low/Neg-VIM^Pos cells. Bar graph depicts image quantifications from experimental triplicates (***p < 0.001; scale bar: 100 μm). b Representative images from colony assay of control and irradiated (IR) SCC25 cells with varying times of exposure to IWP2. Bar graph depicts the surviving fraction in experimental triplicates (D: days; *p < 0.05). c Representative images of Matrigel Transwell invasion assay on control and irradiated (IR) cells with and without IWP2. Bar graph depicts quantifications from experimental triplicates (*p < 0.05, **p < 0.01; scale bar: 100 μm).

Discussion

Considered together, our findings support a key role for the Wnt signalling pathway upstream of IR-induced ECAD^Low/Neg-VIM^Pos phenotype switch and associated radioresistance and cell invasion. The presence of hybrid, semi-stable E–M states in tumours has been demonstrated in several cancer models.^{13,14,51–54} The hybrid state is known to define tumour cell plasticity,^15,55 cancer progression,^14,56 metastasis^15,54 and therapy resistance.¹² In our study, both OSCC monolayers and tumour spheroids manifested a prominent hybrid E–M phenotype, although with different abundance. The hybrid phenotype was as well present in a primary OSCC tumour sample, as a further indication for a common phenotypic feature in OSCC. Further, within the cells with the hybrid signature, populations with varying levels of ECAD expression were identified that, despite negligible differences in the baseline expression levels of EMT TFs, showed biased EMT remodelling in response to IR. Of note, expression of VIM and NCAD was similarly induced in both ECAD^Low/Neg and ECAD^High populations post-IR, whereas all other tested EMT TFs showed significantly higher induction in the ECAD^Low/Neg-derived spheroids. Although a complete image of the cellular and molecular events in these cell populations requires more detailed analysis, it is plausible that the co-existence of EMT TFs and the bona fide mesenchymal markers NCAD and VIM in ECAD^Low/Neg cells assigns these cells to a different position along the E–M axis compared to the ECAD^High cell population, a position with elevated plasticity that has been earlier defined by the expression of Slug. In fact, studies in mammary stem cells and cancer stem cells have secured a key role for Slug in the stemness programme⁵⁷ and many other examples have linked the EMT-induced cellular plasticity to therapy resistance, invasion and metastasis,⁵⁸ which ties well with our further observations in the ECAD^Low/Neg cells.

Our findings strongly suggest that levels of ECAD mark cell populations with differential response to radiation with ECAD^Low/Neg cells bearing significantly higher resistance. Elevated levels of ECAD correspond to better prognosis in adenocarcinoma,⁵⁹ gastric carcinoma⁶⁰ and breast cancer,^61,62 a paradigm that has been challenged by contrasting studies where no significant correlation has been observed between the expression of ECAD and disease outcome.^63,64 Loss of E- cadherin marks the transition towards more mesenchymal states. It is therefore plausible, that it is not the ECAD expression levels per se that foreshadow prognosis, but that is the E–M remodelling potential associated with, but not dependent on ECAD levels that impact prognosis.

Despite the prevalence of cells with a hybrid E–M signature in spheroids and OSCC tumour samples, a major switch towards a mesenchymal phenotype was observed post-IR. The switch to ECAD^Low/Neg/VIM^Pos phenotype led to increased resistance to radiation, with little change to already enhanced invasive capacity in this population. Mesenchymal tumours are known to be among the most aggressive cancers of the head and neck.⁶⁵ Although studies suggest that EMT is dispensable for metastasis,⁶⁶ its link to therapy resistance is well established and extensively reviewed.¹² However, a question that is yet to be addressed is the molecular mechanisms underlying the transition from hybrid E–M to a clearly mesenchymal state in response to therapy. In breast cancer, a major role for the EMT TF ZEB1 has been defined in this phenotypic transition.⁶⁷ ZEB1 is known to drive cell plasticity and resistance to mitogen-activated protein kinase inhibitors in melanoma.⁶⁸ It also acts within the cancer mesenchymal programme to enhance invasion, metastasis and drug resistance.⁶⁹ Indeed, ZEB1 was the only EMT TF differentially regulated between ECAD^Low/Neg and ECAD^High populations in the absence of radiation, suggesting a role for this TF in the assignment of mesenchymal characters. A second and highly plausible mechanism both for the maintenance of the hybrid E–M phenotype and as a mediator of transition to the mesenchymal state are changes at the epigenetic level. Epigenetic feedback regulation plays a major role in the regulation of EMT⁷⁰ and in EMT state transitions as demonstrated by the role of GRHL2 in ovarian cancer.⁷¹ Our transcriptome analyses of tumour spheroids post-radiation highlight a number of possible epigenetic players that are the subject of further studies beyond the scope of the current report.

Using next-generation RNA-sequencing of SCC25-derived tumour spheroids, we aimed to study the transcriptional dynamics related to radiation-induced cellular changes. The choice of 24 h as the final time point was based on the absence of apoptotic death within this time frame, securing a cell death-free window to monitor cellular remodelling. It is, however, notable that radiation-induced cell death does not necessarily occur through the apoptotic cascade and could, in fact, employ multiple mediators including the senescence pathway, autophagy and mitotic catastrophe, as well as other yet uncovered mechanisms.^72–74 Although gene signature related to these cascades were not highlighted in RNA-sequencing analysis, their total exclusion requires further detailed investigations.

Analysis of the RNA-sequencing data demonstrated a dynamic pattern of gene regulation post-IR, with an initial spike in the number of misregulated genes at the 30 min time point that further declined by 24 h. Of note, the control spheroids (non-irradiated) showed a noticeable degree of transcriptional remodelling along the 24 h transition period. Majority of these patterns of transcriptional regulation, however, were enhanced upon exposure to radiation, suggesting that IR-induced remodelling employs and enhances existing pathways involved in normal tumour growth. In fact, activation of EMT and related pathways are demonstrated in 3D spheroids derived from head and neck and other cancers,^75,76 a phenomenon that is anticipated, given the increase in size and the presence of a hypoxic core in the spheroids proportional to their time in culture. It is therefore essential that the transcriptional dynamics in the 24 h IR samples are compared to equivalent controls at the 24 h time point to account for specific radiation-induced effects.

The impact of IR on the immune system is established and with the recent promising developments in cancer immunotherapy, has attracted additional attention.⁷⁷ It is known that cancer cells might mask themselves to evade the immune system through many mechanisms including the specific switches in the HLA molecules.⁷⁸ Although this aspect of our data is beyond the scope of the current study, we have observed a dynamic response to irradiation in the HLA and related immune-modulatory molecules including CXCL8 (IL-8) and TGFβ factors. In fact, a significant number of these genes were induced at the 30 min time point concordant with the onset of the IR-induced EMT remodelling, which might indicate common modes of function. Among the regulated genes, CXCL8 (IL-8) has already been described to have a role in the induction of EMT, which encourages further pursuit of our preliminary observations.⁷⁹

The continuum along the E–M axis is marked by discrete signalling networks^80,81 and involves regulatory checkpoints that maintain cellular states with varying degrees of partial EMT.⁸² Our study highlights a key role for the Wnt signalling pathway in radiation-induced remodelling of tumour spheroids and in the generation of radio-resistant ECAD^Low/Neg/VIM^Pos cells post-IR. Once Wnt signalling was inhibited by IWP2, OSCC cells manifested an enhanced radio-naive response and a decline in invasive capacity. Although significant, neither of these properties were entirely impaired, which suggests the presence of other parallel pathways with overlapping functions. In fact, many pathways are described to act downstream of IR with an impact on resistance and tumour invasion.⁸³ Among them, Wnt signalling has been described as a key modulator of both processes. In oesophageal squamous cell carcinoma, ectopic induction of Wnt signalling resulted in radioresistance.⁸⁴ The pathway was also shown to impact the radiation response in head and neck cancer.⁸⁵ Moreover, expression of LEF1, the downstream effector of Wnt signalling pathway, corresponds to poor prognosis.⁸⁶ Inhibition of Wnt signalling by pyrvinium pamoate caused a decline in self-renewal and metastasis of breast cancer stem cells⁸⁷ consistent with our observations in OSCCs. A thorough picture of the signalling interactome is, however, necessary to highlight potentially beneficial interventions as observed with the combined IWP2 and radiation in our study. It is only in light of such a comprehensive image that possible adverse effects from parallel signalling events can be foreseen and accounted for in the design of potential therapeutics.

Author contributions

F.Z. and N.S. have jointly conceived and executed the project plans and written the manuscript. N.T., D.L., G.J., S.M. and V.W. have contributed to experiments. J.G.L. and E.H. have assisted in the establishment of experimental techniques and in addition to M.X. and H.Z. have had a major intellectual contribution to the project and the resulting manuscript. C.S.F. has provided the human tumour samples and been consulted for subsequent discussions and analyses. E.P. has had major contributions to the bioinformatics analyses and has intellectually contributed to the manuscript. All authors have contributed to the final version of the manuscript.

Ethics approval and consent to participate

The ethics protocol for the human OSCC tumour samples was approved by the University of Western Australia human research ethics committee (Protocol #RA/4/1/8562) and written consent from the patient was obtained in accordance with the Declaration of Helsinki.

Data availability

A copy of raw data containing fastq files has been deposited on SRA under BioProject PRJNA611666.

Competing interests

The authors declare no competing interests.

Funding information

This study was supported by the University of Sydney COMPACT research seed grant, Sydney Dental School, University of Sydney, research support and the Dr. Poyner award from Australian Dental Research Foundation (ADRF). F.Z. is supported by the University of Sydney international scholarship. G.J. is funded via a Sydney West Translational Cancer Research Centre (SW-TCRC) Ph.D. Scholarship. Cancer Institute NSW funds SW-TCRC.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-021-01352-7.