A three-dimensional spheroid-specific role for Wnt–β-catenin and Eph–ephrin signaling in nasopharyngeal carcinoma cells

Nasopharyngeal carcinoma (NPC) derived from the epithelium of the nasopharynx is a major public health problem in many parts of the world (Yoshizaki et al., 2012). Although rare in the USA and Europe, it is highly prevalent in Southern China, Southeast Asia, the Middle East, North Africa and the Arctic. It is also a cancer that most commonly affects working-age populations, which poses significant socio-economic challenges. Current chemo- and/or radio-therapy have achieved favorable clinical outcomes for NPC patients with early-stage disease. However, the majority of NPC patients are diagnosed at advanced stages and have a 5-year overall survival that ranges from 26% to 42% and a median overall survival of only 20 months (Zhang et al., 2013). This highlights the urgent need for effective therapeutic interventions. Although targeted therapies, including VEGF, AKT and mammalian target of rapamycin (mTOR) inhibitors, have been evaluated in numerous phase II trials (Zhang et al., 2013), the discovery of NPC therapeutics is in part hindered by the lack of representative physiological and cost-effective model systems.

It is increasingly recognized that conventional two-dimensional (2D) cultures are not appropriate preclinical cancer models for in vitro studies (Ravi et al., 2017; Riedl et al., 2017). Three-dimensional (3D) cell cultures possess several in vivo tumor features, such as cell adhesion, cell–matrix interaction, drug penetration, hypoxia and nutrient gradients, which can bridge the gap between 2D and animal models. To date, there has been a growing number of comprehensive and systematic studies comparing the various characteristics of malignant tumors in different 2D and 3D culture models (Amann et al., 2014; Gangadhara et al., 2016; Riedl et al., 2017). However, no previous studies have compared NPC in 2D and 3D contexts, and the important phenotypic changes that could result are still unknown. It is also unclear whether changes in the critical signaling pathways of NPC cells occur under 3D culture. Moreover, despite Epstein–Barr virus (EBV) infection being consistently present in nearly all undifferentiated NPCs, the most common histological subtype of NPC, and being successfully identified as a major independent risk factor for NPC occurrence in endemic regions, most studies have mainly focused on studying EBV-negative NPC cells (Tsao et al., 2017).

In our study, we first analyzed the differences between 2D monolayers and 3D spheroids of EBV-positive and -negative NPC cells (Lin et al., 2018). We then investigated differences between monoculture and co-culture models. Under 3D spheroid culture conditions, cell cycle progression, migratory capacity and activation of cancer stem-like cells (CSCs) were observed. Moreover, 3D spheroids displayed hypoxia and decreased anti-cancer drug responses. RNA sequencing (RNA-seq) identified the upregulation of VEGF, AKT and mTOR pathways, as well as several CSC-related signaling pathways in NPC spheroids, suggesting a reprogramming of tumor signaling in 3D spheroids. Drugs that block these specific signaling pathways were effective in monocultures but faced higher resistance in tumor–endothelial co-culture. Our findings provide the first proof-of-concept for a 3D spheroid platform to study EBV-positive and -negative NPC for the purposes of drug discovery and development.

Two undifferentiated EBV-positive NPC cell lines that we established previously were used in this study: C666-1, which has been propagated and used intensively in studies of NPC tumor biology (Cheung et al., 1999), and the recently established NPC43 (Lin et al., 2018). The well-differentiated EBV-negative HK1 cell line and its corresponding HK1+EBV (stably transfected with EBV) were also included. These cell lines formed spheroids within 72 h after seeding on a dish coated with 0.5% agarose. Spheroid morphology appeared to be compact (Fig. 1A). Propidium iodide staining was employed to determine the cell cycle distribution of cells grown in 2D monolayers or 3D spheroids. A significant decrease in the proportion of S-phase and G2/M-phase cells was observed for 3D spheroid cultures compared to 2D monolayers across all four cell lines (Fig. 1B), which implies increased cell cycle arrest at the G0/G1 checkpoint in 3D spheroids compared to 2D monolayer cultures.

Clinically, EBV-positive NPC is highly metastatic (Nakanishi et al., 2017). In tissues, NPC tumors grow within a 3D microenvironment rich in stromal cells and extracellular matrix (ECM) (Song et al., 2020). To this end, we adopted a previously described 3D spheroid-based migration assay (Vinci et al., 2013) in collagen I, fibronectin, gelatin and Matrigel, which more accurately simulates the tumor microenvironment. 2D monolayer cultures in the form of cell droplets were seeded onto ECM and observed in parallel with the 3D spheroid migration assay for comparison (Fig. S1). As shown in Fig. 2 and Fig. S2, the cells of 3D spheroids overlaid on ECM migrated from the compact spheroids and formed a circular rim surrounding the spheroids. 3D spheroids showed increased migration, with more migrating cells of a mesenchymal cell-like shape at the front edge of the rim over 6 days on all types of ECM, whereas cells in 2D monolayer culture did not migrate.

Tumor angiogenesis is necessary for NPC tumor metastasis and is associated with poor survival in NPC patients (Foote et al., 2005). As shown in Fig. 3A, Human umbilical vein endothelial cells (HUVECs) drove significantly more aggressive morphological and migratory behavior of NPC spheroids in a direct cell–cell contact co-culture model. Images throughout 6 days of direct contact co-culture illustrate enhanced cell dispersion from the NPC spheroids and outgrowth of cells towards surrounding HUVECs. In contrast, cell dispersion was significantly slower in monoculture. The enhanced dispersal of cells from spheroids facilitated by HUVECs was cell-type specific, with a profound dispersal effect observed for C666-1 and HK1 spheroids, whereas a lower degree of radial outgrowth was observed for NPC43 and HK1+EBV spheroids (Fig. 3A). To determine whether this effect of co-culture with HUVECs was mediated by soluble factors in paracrine secretions, an indirect co-culture experiment was performed where 2D monolayers or 3D spheroids were grown in conditioned medium from HUVEC cultures. This conditioned medium supported robust outgrowth of cells from 3D spheroids but not from 2D monolayers. Consistent with our direct cell–cell contact co-culture model findings, outgrowth of cells was more significant for C666-1 and HK1 spheroids than for NPC43 and HK1+EBV spheroids (Fig. 3B; Figs S3,S4).

Stemness is one of the hallmarks of an aggressive phenotype in NPCs. We tested whether expression of stemness markers in 3D cultures was increased and observed statistically significant transcriptional upregulation of Oct4 (also known as POU5F1), Nanog and Bmi1 in 3D spheroid cultures (Fig. 4A). Furthermore, increased mRNA (Fig. 4B) and protein (Fig. 4C) expression of the hypoxia markers HIF-1α (HIF1A) and HIF-1β (ARNT), master regulators that are known to play an important role in tumor progression and angiogenesis in solid tumors, was observed in 3D spheroids compared to levels in 2D monolayer cultures.

Chemotherapeutic drugs, particularly cisplatin, doxorubicin and paclitaxel, are the first line treatments for advanced NPC patients whose tumor has spread to distant organs, either used alone or in addition to radiation (Peng et al., 2010). The majority of advanced-stage NPC patients die because of chemoresistance (Tang et al., 2012). The relative viability rate in the presence or absence of chemotherapeutic drugs used to treat NPC (i.e. cisplatin, doxorubicin and paclitaxel) was tested in 2D and 3D cultures. The range of drug concentrations used was set based on our preliminary drug dose-response curves (data not shown). Both NPC43 and C666-1 tended to show relative resistance to cisplatin, doxorubicin and paclitaxel in the 3D spheroids as compared to 2D cultures after normalization to the 3D:2D proliferation ratio (Fig. 5A). Consistent with these findings, treatment with cisplatin, doxorubicin or paclitaxel resulted in significant downregulation of cleaved caspase 3 (Fig. 5B) as well as of cleaved caspase 7 and cleaved PARP (Fig. S5) in 3D spheroid cultures compared to the levels in 2D cultures.

RNA-seq of all four cell lines used in this study was performed in order to delineate the transcriptional programs that may underlie the different cellular phenotypes. Differential expression analysis was based on a fold-change cut-off value and false discovery rate calculated for each P-value and was used to identify genes that were differentially expressed in 3D spheroids versus 2D monolayers (Fig. S6). An expression heatmap of the genes that were identified as differentially expressed in 2D monolayers and 3D spheroids revealed differential gene expression signatures between the two groups (Fig. S6A). DESeq2 analysis (cutoff of log₂ fold change ≥1, adjusted P≤0.05; Love et al., 2014) comparing 2D monolayer versus 3D spheroid cells revealed that 671 out of 17680 genes were upregulated in 3D spheroids, 645 of 17680 genes were downregulated in 3D spheroids and there was difference in the remaining 16364 genes (Fig. S6B). Moreover, KEGG pathway enrichment analyses (Kanehisa et al., 2017) showed upregulation of pathways involved in cell growth and death, cell mobility, signaling molecules and interactions, cancer and drug resistance (Fig. S6C), and particularly the Wnt, phosphoinositide 3-kinase (PI3K)–AKT and mTOR signaling pathways (Fig. S6D), which are known to be activated in NPCs.

RNA-seq results were validated using western blotting, in which higher protein levels of VEGF (encoded by VEGFA), p110α (encoded by PIK3CA), phosphorylated AKT (p-AKT; referring to AKT1, AKT2 and AKT3), phosphorylated GSK3B (p-GSK-3β), Raptor (also known as RPTOR), phosphorylated mTOR (p-mTOR) and mTOR were detected in 3D spheroids of both cell lines (Fig. 6A). The activation of VEGF and AKT–mTOR axes in 3D spheroid cultures prompted us to investigate three FDA-approved specific inhibitors that target these pathways. Pazopanib (NCT00454142) and MK2206 (NCT01349933) are VEGF and AKT inhibitors that have undergone phase II clinical trials to study treatment of advanced and recurrent NPC patients. Upon treatment with pazopanib and MK2206, there was a reduction of cancer cell survival in 3D spheroid cultures, but after normalization to the proliferation ratio, 3D spheroid cultures were found to have higher drug resistance compared to 2D monolayers (Fig. 6B). PF-04691502 is a dual PI3K and mTOR inhibitor that has been evaluated as a novel therapeutic drug in preclinical models of NPC (Mehta et al., 2012). Similarly, in accordance with the activation of mTOR in these cells, treatment with PF-04691502 suppressed tumor cell survival in 3D spheroid cultures, although higher drug resistance was observed compared to that of 2D monolayers, suggesting that novel pharmacological agents, including PF-04691502, can play a role in the control of NPC growth. In addition, higher protein levels of β-catenin (CTNNB1) were detected in 3D spheroids as compared to levels in 2D cultures (Fig. 6C). In concordance with the activation of Wnt–β-catenin signaling in these cells, our data showed that tumor cell survival was markedly suppressed by treatment with the inhibitor of Wnt–β-catenin signaling, ICG-001, and 3D spheroids showed higher resistance to ICG-001 than 2D monolayers (Fig. 6D).

In searching for novel signaling pathways, we identified Eph–ephrin signaling to be significantly upregulated in 3D spheroids as compared to levels in 2D cultures. Higher protein levels of ephrin (specifically Efna1) were detected in 3D spheroids (Fig. 6C). Consistent with this, tumor cell survival was found to be markedly suppressed upon treatment with dasatinib, which inhibits Eph–ephrin signaling, with higher drug resistance found in 3D spheroids compared to 2D monolayers (Fig. 6D). These results provide proof-of-principle that 3D spheroids can be used as a tool to predict drug sensitivities that are clinically relevant.

Finally, we assessed the capacity of co-culture with HUVECs to enhance the dispersion of NPC cells in 2D monolayers and 3D spheroids in the presence of the specific kinase inhibitors. We observed that tumor–endothelium co-culture overcame the sensitivity of tumor spheroid dispersion to treatment with pazopanib, ICG-001 or dasatinib (Fig. 7). Similar effects were observed when using conditioned medium from HUVEC cultures (Fig. 8), suggesting that HUVECs are capable of inducing drug resistance via soluble factors and in an essentially contact-independent manner. Notably, amongst these specific kinase inhibitors, dasatinib exerted the greatest effect in suppressing dispersal of cells from 3D spheroids (Figs 7, 8).

In recent years, it has become apparent that cancer cells cultured in conventional 2D in vitro systems have shortcomings when used as a preclinical model to faithfully reflect clinical situations. Clinical trial failures indicate that the efficacy of anti-cancer drugs may be overestimated when tested using a 2D platform. Therefore, a more reliable preclinical model is urgently needed to more accurately mimic the real situation of tumor growth in human body. In view of the complexity, high cost and inconvenience of in vivo animal experiments, in vitro 3D spheroid culture models serve as a good bridge between in vitro and in vivo assays. Although 3D spheroid culture models are increasingly being used to investigate many types of malignancies (Nakanishi et al., 2017), we are the first to study both EBV-positive and EBV-negative NPC, which is a highly malignant tumor, in a 3D spheroid culture system and compare it with traditional 2D monolayer culture.

In our study, 3D spheroids were more resistant than 2D monolayers to a wide dosage range of chemotherapeutic drugs. This is likely due to an inhibition of S-phase progression (Fig. 1B), which is the mechanism of action of many anti-cancer drugs that selectively target the fast-dividing and proliferating cells (Mehta et al., 2012). In line with the chemoresistance shown in 3D spheroid cultures, we observed a significant increase in tumor stemness properties and hypoxic development compared to NPC cells in 2D monolayers. Recent studies have also suggested that CSCs contribute to the metastatic potential of NPCs (Qi et al., 2018). We showed that 3D spheroids exhibited enhanced migratory capacity in response to various ECM, as compared to cells in 2D culture. There is increasing evidence suggesting that ECM not only provides structural support but also modulates self-renewal and differentiation of CSCs (Nallanthighal et al., 2019). The motile cells from NPC spheroids migrated with a round morphology (amoeboid migration mode) on collagen and fibronectin, and with elongated morphology (mesenchymal migration mode) on gelatin and Matrigel. Their migration behaviors were closely related to the protrusion of lamellipodia from the spheroids, which has recently been shown to be regulated by the PI3K–AKT pathway (Chen et al., 2018). Whether the 3D spheroids secrete their own ECM during spheroid formation, mimicking the in vivo ECM in solid tissues, and whether the trypsin-based replating method applied to the 2D monolayers has an impact in modifying cell-surface proteins, and thus altering the cellular signaling, warrant further investigations.

Despite rapidly evolving chemotherapy methods, treatment of recurrent and metastatic NPCs remains largely dissatisfying and challenging (Lee et al., 2017). In recent years, the management of loco-regional or distant recurrences of NPC has been geared toward plausible molecular targeted therapy. RNA-seq analysis of NPC spheroids in our study unveiled novel driver genes and distinct pathobiology signatures, which might drive NPC resistance to therapy. Besides the known oncogenic signaling pathways (i.e. VEGF, AKT and mTOR) in NPCs (Almobarak et al., 2019), we also identified altered novel signaling pathways such as Wnt–β-catenin and Eph–ephrin. VEGF is one of the most investigated molecular targets in NPCs. High expression of VEGF is correlated with adverse prognosis and metastatic relapse of NPC in relation to EBV (Hui et al., 2002; Krishna et al., 2006). In a phase II clinical trial involving platinum-pretreated recurrent or metastatic NPC, pazopanib has been shown to improve the progression-free and overall survival rates as well as tolerability in patients (Lim et al., 2011). There is a positive association between the activation of the AKT–mTOR pathway and poor prognosis of NPC (Wang et al., 2014). In addition to pazopanib, our results also demonstrate the ability of a new dual PI3K and mTOR inhibitor, PF-04691502, and the AKT inhibitor MK2206 to suppress NPC tumor cell survival. It is interesting to note that dual targeting of both AKT and its downstream effector will potentially be a superior option in terms of treatment efficacy.

In search of potentially novel therapeutic targets, we identified Wnt–β-catenin and Efna1–EphA10 signaling. Deregulated expression of Wnt signaling components is frequently observed in NPCs and is associated with tumor aggressiveness (Chan et al., 2015; Chen et al., 2018; Zeng et al., 2007) and poor prognosis (Xu et al., 2013). Recent studies have identified several Wnt signaling regulators as solid tumor therapies (Jung and Park, 2020), yet their potential for treatment of NPCs remains unknown. Our findings reinforce other studies showing that Wnt–β-catenin signaling may be a clinically druggable vulnerability in NPCs. Ephrin A1 (Efna1) and its receptor Eph receptor A10 (EphA10), which are highly associated with many invasive and metastatic solid tumors including breast (Shiuan et al., 2020) and lung cancer (Ieguchi et al., 2014), have never been reported in NPCs. However, Efna1 is interrelated to many aberrant signaling pathways in NPCs, including Wnt–β-catenin and HIF-1α signaling (Tanabe et al., 2018). Data from The Cancer Genome Atlas (TCGA) also reveals an association between higher expression of Efna1 and poor survival of the head and neck cancer cohort (Fig. S7). There is evidence that EBV could infect target cells through Eph receptors and activation of downstream signaling pathways (Wang et al., 2020). Thus, this understudied candidate is worthy of further exploration as a novel therapeutic target in NPCs.

The important effect of the tumor microenvironment on tumor behavior is becoming increasingly clear. Multiple lines of evidence suggest that blood vessels play an important role in NPC progression: (1) NPCs lie close to major carotid blood vessels in the neck; (2) blood vessels are an important scaffold for NPC migration; and (3) microvessel density is a negative prognostic indicator in NPC (Li et al., 2018). In tumor–endothelial co-culture of NPC cells in 3D spheroids in contact with human vascular endothelial cells or their conditioned medium, we have shown significant NPC tumor spheroid dispersion that closely mimics tumor behavior in vivo. Our study has further shown that this dispersion can be inhibited by selective inhibition of VEGF, Wnt–β-catenin or Eph–ephrin signaling. At present, there are many drugs that interfere with the advanced stages of the disease. Our results suggest that targeting these pathways may offer novel anti-cancer regimens to treat metastatic NPCs, a possibility that is worth further investigation.

The physiologically relevant 3D spheroid model established in our study successfully recaptures a wide range of structural and clinical features of solid NPCs and is capable of mimicking tumor evasion mechanisms in response to anti-cancer treatment approaches, including chemotherapy and FDA-approved targeted therapies. It has advantages over 2D monolayers for the study of tumor behaviors and discovery of new drug candidates. It is comparable to other 3D systems in terms of simplicity, time and cost effectiveness, and is highly reproducible. A 3D preclinical model is especially useful for NPC – a rare malignancy in most of the world – because the low incidence limits the use of large clinical trials and there is a lack of suitable animal models.

We have undertaken, for the first time, a systematic comparison of EBV-positive and EBV-negative NPC cells in 3D spheroids and 2D monolayer cultures. We have demonstrated that key malignant phenotypes are different in the two culture systems and have elucidated the underlying molecular mechanisms. Our research not only shows the great potential value of 3D spheroids as a drug screening platform in cancer research, but also provides a basis for further study of both EBV-positive and EBV-negative NPCs.

Undifferentiated EBV-positive NPC cell lines that we established previously – C666-1 (Cheung et al., 1999) and NPC43 (Lin et al., 2018) – were cultured in RPMI 1640 medium supplemented with 10% (v/v) fetal bovine serum (FBS; Life Technologies, Darmstadt, Germany), 4 μM ROCK inhibitor (SCM075; Sigma-Aldrich, St. Louis, MO, USA), 100 U/ml penicillin and 100 μg/ml streptomycin and in DMEM medium supplemented with 10% (v/v) FBS, 100 U/ml penicillin and 100 μg/ml streptomycin, respectively. Well-differentiated EBV-negative HK1 and its corresponding HK1+EBV (stably transfected with EBV; Lo et al., 2006) cell lines were cultured in RPMI 1640 medium supplemented with 10% (v/v) FBS, 100 U/ml penicillin and 100 µg/ml streptomycin. Stably transfected cells were selected using G418 (Life Technologies, Darmstadt, Germany). All the cell lines were authenticated and tested. To form 3D spheroids, cells were grown on plate coated with 0.5% agarose. After 72 h, spheroid morphology was evaluated microscopically, and the spheroids were used for subsequent assays. HUVECs were obtained from Lonza (CC-2519) and were cultured in F12K (Life Technologies) supplemented with 10% (v/v) FBS, 100 μg/ml streptomycin, 100 U/ml penicillin, 1% (v/v) 0.1 g/ml Heparin (Merck Millipore) and 10% (v/v) ECGS (Merck Millipore). All cell lines were cultured at 37°C with 5% CO₂.

Cells were dissociated in trypsin–EDTA (Life Technologies), resuspended in phosphate-buffered saline (PBS) and fixed in 70% (v/v) ethanol at −20°C overnight. The next day, pellets were washed twice with PBS, then resuspended in 0.5 ml propidium iodide (50 µg/ml) with 0.1% (v/v) Triton-X100 and 0.1 mg/ml RNase A. The cells were incubated for 30 min at 37°C in the dark and passed through a cell strainer (70 µm) before analysis using a BD FACSAria III (BD Biosciences, San Jose, CA, USA).

A total of 5×10³ NPC cells were seeded in 96-well plate for 3D spheroid formation. Spheroids that had pre-formed for 72 h were incorporated with CellTracker Orange CMRA fluorescent dye (Invitrogen, Thermo Fisher Scientific) for 1 h then transferred to either a 12-well plate containing 1×10⁵ unlabelled HUVECs cells seeded 24 h in advance or a cell-free 12-well plate as monoculture control. 3D spheroids were grown adjacent to HUVECs under direct cell–cell contact or in the monoculture condition for 6 days. The dissemination state of 3D spheroids in the presence or absence of HUVECs was captured using a brightfield imaging system (Cytation 1 Cell Imaging Multi-mode Reader; BioTek Instruments, Inc) daily over 6 days. Representative images on day 1, 2, 4 and 6 of mono- and co-culture are shown. To generate conditioned medium, 1×10⁶ HUVEC cells were cultured with F-12 Ham medium (Sigma-Aldrich) in a 100 mm dish. The next day, cells were gently rinsed three times with PBS, and the complete medium was replaced with serum-free medium. After 72 h, the conditioned media were collected, pooled, centrifuged at 1000×g for 5 min and filtered through a 0.2 µm filter to remove cellular debris before the supernatants were used fresh for co-culture assays with tumor spheroids. The conditioned medium or serum-free control was added to 2D monolayer and 3D spheroid tumor cells. The dissemination state of cells was captured using a brightfield imaging system (Cytation 1 Cell Imaging Multi-mode Reader; BioTek Instruments, Inc) daily over 6 days. Representative images on days 1, 2, 4 and 6 of serum-free or conditioned medium-treated co-culture are shown and were quantified using Fiji software (ImageJ; https://imagej.net/software/fiji/). In the co-culture model for specific kinase inhibitor treatments, HUVECs (1×10⁵ cells/well) were seeded onto a gelatin-coated 12-well plate and cultured for 8 h to allow adherence. The pre-formed spheroids were transferred to each well and co-cultured for 48 h. The cells were treated with specific kinase inhibitors at their IC₅₀, as determined using an MTT assay, for 48 h. Dispersion diameters of spheroids were captured at the end of the experiment using a brightfield inverted microscope, and were measured and quantified using TCapture software (ISCapture, Tucsen Photonics).

Total RNA was extracted from spheroids pre-formed for 72 h and the 2D monolayer control using TRIzol (Ambiol; Thermo Scientific, Waltham, USA). Reverse transcription-quantitative PCR (RT-qPCR) was performed using M-MLV reverse transcriptase (Invitrogen, Carlsbad, CA, USA) and 2×AceQ qPCR SYBR Green Master Mix (Q141-AA, Vazyme) according to the manufacturer's instructions. The primers used were: OCT4 forward, 5′-AGGTATTCAGCCAAACGACCA-3′; OCT4 reverse, 5′-GCACGAGGGTTTCTGCTTTG-3′; NANOG forward, 5′-TCTGGACACTGGCTGAATCC-3′; NANOG reverse, 5′-TGACTGGATGGGCATCATGG-3′; BMI1 forward, 5′-GGTTTCTTGGAAAGCAGGCA-3′; BMI1 reverse, 5′-GGCCCAATGCTTATGTCCAC-3′; HIF-1α forward, 5′-TGATTGCATCTCCATCTCCTACC-3′; HIF-1α reverse, 5′-GACTCAAAGCGACAGATAACACG-3′; HIF-1β forward, 5′-CAGTGAAAAAGGAAGGTCAGCA-3′; HIF-1β reverse, 5′-CAAGTCCATTCCTGCATCTGTT-3′; β-actin forward, 5′-CATCCTCACCCTGAAGTACCC-3′; β-actin reverse, 5′-AGCCTGGATAGCAACGTACATG-3′. RT-qPCR was performed using a Real-time PCR, ABI StepOne Plus system (Applied Biosystems, Thermo Fisher Scientific).

5×10³ cells/well were seeded in 96-well plates for 24 h and treated with different concentrations of drugs prior to the addition of 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide (MTT; Sigma-Aldrich; 5 mg/ml). Following 4 h of incubation with MTT, dimethyl sulfoxide (Sigma-Aldrich) was added into the wells to dissolve the purple formazan. After 48 h or 72 h, cell viability was assessed by measuring the absorbance at 490 nm. Five replicated wells were measured for each drug concentration. In 3D spheroid culture, 2×10⁵ cells were cultured for 3 days to form spheroids and treated with the drugs of different concentrations, followed by MTT assays.

Cells were lysed in RIPA lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1 mM PMSF, 1 μM pepstatin, 0.5 mg/ml leupeptin and 0.3 μM aprotinin), and lysates (30 µg/lane) were separated on a 10% SDS–PAGE gel and transferred to nitrocellulose membrane (Bio-Rad, Hercules, CA, USA). The primary antibodies used in this study were as follows: anti-β-actin (1:5000; A5316; Sigma Life Science), anti-caspase-3 (1:1000; 9662S; Cell Signaling Technology), anti-cleaved caspase-3 (1:1000; 9664T; Cell Signaling Technology), anti-cleaved PARP (1:1000; 5625T; Cell Signaling Technology), anti-cleaved caspase-7 (1:1000; 8438T; Cell Signaling Technology), anti-HIF-1α (1:1000; 36169S; Cell Signaling Technology), anti-HIF-1β (1:1000; 5537S; Cell Signaling Technology), anti-Raptor (1:1000; 2280T; Cell Signaling Technology), anti-mTOR (1:1000; 2983T; Cell Signaling Technology), anti-p-mTOR (1:1000; 2971S; Cell Signaling Technology), anti-pan-AKT (1:1000; 4691T; Cell Signaling Technology), anti-p-AKT (1:1000; 9271S; Cell Signaling Technology), anti-p110α (1:1000; 4249T; Cell Signaling Technology), anti-p-GSK-3β (1:1000; 9336S; Cell Signaling Technology), anti-β-catenin (1:1000; 2698S; Cell Signaling Technology), anti-Efna1 (1:500; ab124911; Abcam) and anti-VEGF (1:1000; MAB293; R&D Systems). Bound antibodies were detected using peroxidase-conjugated secondary antibodies and enhanced chemiluminescence reagents (Amersham Biosciences, Little Chalfont, UK). Signal intensities were quantified using ImageJ software (Bethesda, MD, USA).

Total RNA was extracted using TRIzol (Ambion, Thermo Scientific, Waltham, USA) and the sample quality control was performed using an Agilent 2100 Bioanalyzer (Agilent RNA 6000 Nano Kit). The RNA samples were sequenced using DNBSEQ Eukaryotic Transcriptome sequencing by BGI Tech Solutions (Hong Kong) Co. Ltd. Following sequencing read filtering by internal software SOAPnuke (https://github.com/BGI-flexlab/SOAPnuke), genome mapping was performed using HISAT (Kim et al., 2015). Clean reads were mapped to reference (hg19 human) using Bowtie2 (Langdon, 2015), then gene expression level was calculated with RSEM (Li and Dewey, 2011) followed by gene expression cluster analysis, differentially expressed gene (DEG) detection with PossionDis (Byadgi et al., 2016), hierarchical clustering for DEGs using pheatmap (Cheng et al., 2017), gene ontology analysis of DEG and pathway analysis of DEG.

All assays were performed at least three times independently. Mean and s.d. values were calculated using SPSS 21 and GraphPad Prism (San Diego, CA, USA). Two-tailed, unpaired Student's t-test and one-way analysis of variance (ANOVA) followed by Dunnett's multiple comparison test were performed to determine the significance of differences among groups. Differences were considered statistically significant at P-values indicated as *P<0.05, **P<0.01, ***P<0.001 or ****P<0.0001.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.256461