CRC tissues and normal colorectal mucosa were obtained from patients with sporadic CRC, who were operated at the AOU Federico II and Istituto Nazionale dei Tumori (Naples, Italy) and primary cell cultures were established from these tissues. Data regarding tumour stage were recovered from the medical records of each patient, in accordance with the TNM classifications and tumour budding grades.

Samples from all subjects who participated in this study were collected after obtaining authorisation from the Comitato etico per le attività Biomediche - Carlo Romano of the University of Naples Federico II (protocol no. 432/17). Authorisation was granted only once the study had received ethical approval and written informed consent had been obtained from all participants. All methods were performed in accordance with the relevant guidelines and regulations.

The T88 and T93 cells were previously isolated and stabilized in vitro (24). The HM110 colon cells were isolated and stabilized during this study from the healthy colon mucosa (HM) of a patient with sporadic colon cancer, as previously described (24). Briefly, samples were washed overnight at 4°C in PBS containing antibiotics, finely minced and digested in collagenase II in DMEM/FBS-10% for 1 h at 37°C, 5% CO₂. The obtained cell suspension was then collected by centrifugation at 1,000 × g, at room temperature, washed twice and subsequently cultured in DMEM/F12-10% FBS medium (1:1), 100 U/ml penicillin, 100 μg/ml streptomycin, and 2.5 μg/ml amphotericin B. The HT29 and RKO cells were obtained from ATCC (Manassas, VA, USA). To inhibit GSK-3β activity, cells were incubated in medium containing 30 mM LiCl for 10 days. The T88 and T93 primary colon cancer cells were then cultured as spheres in serum-free stem cell medium on low-adhesion plates as previously described (24). Cancer spheroids were also obtained by the hanging drop assay. Briefly, the cells were diluted to a final concentration of 3.7×10⁴ cells/ml in DMEM/F12 (1:1) containing 2% foetal bovine serum (FBS), 100 U/ml penicillin, 100 μg/ml streptomycin, 2.5 μg/ml amphotericin B, 10 μg/ml basic fibroblast growth factor (bFGF) and 20 μg/ml epidermal growth factor (EGF), with or without 35 mM LiCl, (all reagents were purchased from Sigma-Aldrich Merk KGaA, Darmstadt, Germany). The drops (27 μl) were seeded on the cover of a plate and incubated at 37°C at 5% CO₂ for 72 h. Under these conditions, the cells aggregated to form 3D spheroids. Cancer spheroids obtained in presence or absence of LiCl were finally disaggregated into single cells, which were again grown in adhesion.

Cell migration was evaluated using in vitro wound healing assays and the Boyden chamber assay. In vitro wound healing assays were performed as previously described by Liang et al (25). Briefly, the cells were seeded at 1×10⁴ cells/well in 24-well plates. After the cells formed a monolayer, a scratch wound was made with the tip of a 1,000-μl pipette tip, and the scratch was photographed under a light microscope (Leica DM IL LED inverted microscope, type 11090137002, serial no. 11521258/335209; Meyer Instruments, Inc. Houston, TX, USA) at ×20 magnification. Subsequently, cells were incubated with or without (untreated control) 35 mM LiCl. After 24 h, cells were rinsed, and a second set of images were acquired.

Boyden chamber assays were performed using the Corning® Transwell® polycarbonate membrane cell culture inserts (pore size, 8.0 μm; Sigma-Aldrich, Merk KGaA). A 300 μl of aliquot of a 10,000 cells/ml suspension was resuspended in serum-free medium and seeded into the upper chamber of the polycarbonate membrane. Subsequently, 500 μl of medium containing 10% FBS (chemoattractant) were added to the lower well of the migration plate. Finally, the cells were removed from the top of the membrane and the migrated cells were stained with crystal violet solution for 15 min at room temperature (Sigma-Aldrich, Merk KGaA). Alternatively, 24-well Transwell plates (pore size: 8.0 μm; Corning Inc., Corning, New York, NY, USA) were used. Each membrane was released from the apparatus using a scalpel. The permeable membranes were then washed twice with PBS/1% FBS at room temperature. The cells on the membranes were fixed, permeabilized and stained with DAPI (Sigma-Aldrich, Merk KGaA) for 15 min at room temperature. Finally, the membranes were mounted on glass slides, covered with coverslips, and fluorescence was visualized and imaged under a fluorescence microscope (Leica DM IL LED inverted microscope, type 11090137002, no. 11521258/335209; Meyer Instruments, Inc.).

Total RNA was extracted from the cells using QIAzol reagent (Qiagen, Hilden, Germany) according to the manufacturer's instructions. Following DNase incubation, cDNA was synthesized from 1 μg total RNA, 500 ng random hexamers, and 1 μl Superscript III reverse transcriptase in the presence of 4 μl 5X RT buffer, 1 μl DTT (0.1 M) and 1 mM dNTPs (Invitrogen/Thermo Fisher Scientific, Waltham, MA, USA) as previously described (26). Quantitative (real-time) PCR (qPCR) was performed on 0.5 μl cDNA, Snail and Twist1 primer pairs as previously described (24,27). Relative expression was calculated with the 2^−ΔΔCq method (28) and normalized against glucuronidase (GUS) mRNA. qPCR was performed on a Bio-Rad iCycler iQ Real-Time PCR Detection System (Bio-Rad Laboratories, Hercules, CA, USA). A non-template control was run for each assay, all determinations were performed at least in duplicate to ensure reproducibility and each experiment was performed two times. The synthesis of the expected PCR product was confirmed by melting curve analysis.

Fractionated nuclear and cytosolic protein extracts were isolated from the T88 and T93 cells using the Qproteome Nuclear Protein kit (Qiagen Hilden, Germany), and quantified by using Traditional Bradford kit (Bio-Rad Laboratories Srl, Segrate, Italy), adopting bovine serum albumin standards. A total of 50 μg of proteins were separated by 10% SDS-polyacrylamide gel electrophoresis. Blots were prepared on Amersham Hybond-ECL nitrocellulose membranes (Amersham Pharmacia Biotech, Cologno Monzese, Italy), and proteins were blocked in non-fat dried milk diluted in TBST to reduce the background, as previously described (29,30). Primary antibodies against N-cadherin (rabbit monoclonal anti-human; #4061; 1:1,000) and CD133 (rabbit monoclonal anti-human; #5860; 1:1,000) were obtained from Cell Signaling Technology (Danvers, MA, USA). The primary antibody against CD44v6 (mouse monoclonal anti-human; #5640; 1:1,000) was from R&D Systems (Minneapolis, MN, USA), the anti-actin (goat polyclonal anti-human; sc-1615; 1:8,000) and anti-Histone H1 (monoclonal anti-mouse; sc-393358; 1:1,000) antibodies were from Santa Cruz Biotechnology (Santa Cruz, CA, USA). The membranes were probed with peroxidase-conjugated secondary antibodies against rabbit (#7074), mouse (#7076; both from Cell Signaling Technology or goat IgG (#ab97110; from Abcam, Cambridge, UK), all used at 1:3,000 dilution, and immunoreactive bands were detected using the enhanced chemiluminescence HRP Substrate Immobilon Western (Millipore, Billerica, MA, USA).

Primary CRC cells were seeded and grown in 12-well cultivation chambers with removable microscopy glass slides (ibidi, Martinsried, Germany); cancer spheroids obtained by the hanging drop assay were also sedi-mented in the same chamber. Immunofluorescence analyses were performed as previously described by Di Maio et al (31). Briefly, following fixation in 4% paraformaldehyde in PBS for 10 min, the cells were permeabilized in 0.1% Triton X-100 in PBS for 30-120 min, and then blocked in 10% bovine serum albumin for 30 min. The cells were incubated with primary antibodies (Table I) overnight, and then with secondary antibodies (Alexa Fluor 546 donkey anti-rabbit, A10040; Alexa Fluor 488 donkey anti-mouse, A21202; Thermo Fisher Scientific) for 1 h, and then with DAPI (Sigma-Aldrich) for 30 min at room temperature to label the nuclei. Negative controls without primary antibodies were also included, and these exhibited no staining. Following the indicated treatments, coverslips were mounted on glass slides and examined under a fluorescence confocal microscope (Zeiss LSM 700, Carl Zeiss, Oberkochen, Germany).

We first confirmed, by immunofluorescence analyses, that both the T88 and T93 cells (Fig. 1A) expressed epithelial and mesenchymal proteins, in accordance with our previous data (24). As shown in Fig. 1B, Vimentin and Cytokeratin co-localized at the cytoskeleton level in both cell lines. E-cadherin did not exhibit the classic membrane staining of epithelial cells, such as in commercial HT29 CRC cells (Fig. 1C), but rather a diffuse punctate signal, as observed in the commercial RKO CRC cells, was found (Fig. 1C). Commercially available CRC cell cultures are largely characterised; thus, it is well known that HT29 cells have a more epithelial phenotype than RKO cells, which primarily exhibit mesenchymal features (32). N-cadherin exhibited an unexpected nuclear localisation (Fig. 1B). We also analysed two cell surface glycoproteins involved in stemness in several types of cancer cells, CD133 and CD44v6. Of note, similar to N-cadherin, these markers were mainly localised to the nucleus. As shown in Fig. 1D, these findings were confirmed by western blot analysis of the nuclear and cytosolic protein fractions of the T88 and T93 cells.

Furthermore, to evaluate whether the observed nuclear localisation phenotypes were tumour-specific, we isolated, cultured and analysed cells from the healthy mucosae of a sporadic CRC patient (HM110). Through immunofluorescence staining, we demonstrated that these cultures also expressed both epithelial and mesenchymal proteins (N- and E-cadherin, respectively), but found that they did not exhibit the same nuclear localisation of N-cadherin, CD133 and CD44v6, as was observed in the tumour cells (Fig. 1E). This finding strongly suggests a tumour-specific nature of this feature.

It is known that primary colon cultures are able to generate in vitro organoids, which are propagated by intestinal stem cells, and thus, represent an organotypic culture system. Furthermore, it has been demonstrated that it is possible to generate intestinal organoids starting from a single sorted leucine-rich repeat-containing G-protein coupled receptor 5 (LGR5)-positive intestinal stem cell from dissociated intestinal crypts (33,34). As shown in Fig. 2A and B, the T88 and T93 cells, all LGR5-positive (please also see Fig. 6), generated organoid-like in vitro three-dimensional bodies. Immunofluorescence staining revealed that cells of these three-dimensional bodies expressed both CD133 and CD44v6 mainly at the plasma membrane (Fig. 2C and D; images on top panels), contrary to their mesenchymal counterparts in monolayer, which retained nuclear localisation (Fig. 2C and D; images on bottom panels).

As GSK-3β is overexpressed and functions as an oncogene in CRC, we examined the effects of GSK-3β inhibition on tumour cells using LiCl, a specific inhibitor. We found that LiCl treatment inhibited cell migration using wound healing and Transwell migration assays. As shown in Fig. 3A, the wounds were almost completely healed after 24 h in the untreated cells, while the cells incubated with LiCl had wound sizes similar to time 0 after 72 h. We confirmed this observation using Transwell migration assays; no cell migration was detected when the cells were incubated with LiCl using crystal violet or DAPI staining (Fig. 3B).

We confirmed our previous finding showing that LiCl induces MET in T88 and T93 cell cultures (24). Indeed, by confocal microscopy, we confirmed the finding of E-cadherin upregulation described in our previous study (24). We assessed the expression and localization of the epithelial markers, E-cadherin and pan-cytokeratin (Ck), and the expression of the mesenchymal markers, Vimentin and N-cadherin, in untreated and LiCl-treated cells. As shown in Fig. 4, Vimentin and Cytokeratin co-localised in both untreated and treated cells, the latter exhibiting a different cytoskeletal organization. Following LiCl treatment, E-cadherin maintained a discontinuous punctate pattern, although the treated cells exhibited a differentiated, more polarized shape than the untreated cells. Furthermore, following LiCl incubation, a nucleolar localization of N-cadherin appeared visible, whereas it was not detected in the untreated cells.

Thereafter, we evaluated in both the untreated cells and cells incubated with LiCl, the expression of CD44, its splice isoform CD44v6, CD133 and β-catenin, all proteins playing pivotal roles in CRC development and progression. In accordance with our previous data (24), following 10 days of LiCl treatment, CD44 and β-catenin expression decreased and they both localised mainly at the plasma membrane, while CD44v6 and CD133 proteins both exhibited cytosolic localisation (Fig. 5).

We also analysed the expression of the EMT-associated transcription factor, Snail, and the expression of several stem cell-specific markers, including LGR5, Nanog, aldehyde dehydrogenase 1 (ALDH1), octamer-binding transcription factor 4 (Oct4) and sex determining region Y-box 2 (Sox2), in addition to the aforementioned CD133, CD44 and CD44v6, following LiCl treatment. As shown in Fig. 6, we observed that LiCl induced a downregulation in the expression of the EMT-associated transcription factor, Snail, and all stemness markers (LGR5, ALDH1, Nanog, Oct4 and Sox2), which in the untreated cells exhibited nuclear localisation. Furthermore, only in the untreated cells, Nanog exhibited a higher nuclear expression in the T88 cells than in the T93 cells (yellow arrows), while Oct4 exhibited an opposite trend (white arrows).

Furthermore, we generated tumourspheres from the T88 and T93 cells by the hanging drop assay, thus establishing a system of adherent primary mesenchymal colon cancer cells and paired tumourspheres, which are useful for studying stemness features, cell plasticity and drug responses. It has been reported that only stem cells and/or stem cell-like cells are able to survive and grow in suspension (35), as the loss of adhesion induces death through anoikis in non-malignant and cancer differentiated cells. Under these culture conditions, undifferentiated tumour cells proliferate and grow as floating clusters termed tumourspheres. As shown in Fig. 7A, the T88 and T93 cells formed spheroid aggregates in hanging drop assays, indicating the acquisition of a dedifferentiated state. However, following 72 h of incubation with LiCl, the T88 cells lost their spherical shape, consistent with a differentiated state, and the T93 cells completely lost their ability to form spheres, generating only cellular aggregates, in agreement with the more epithelial phenotype of T93 cells compared with the T88 cells (24).

When analysed by confocal microscopy, both the epithelial (Cks and E-cadherin) and mesenchymal (Vimentin and N-cadherin) markers were expressed in the untreated tumourspheres (Fig. 7B). However, these spheres exhibited an upregulation of epithelial markers and a downregulation of mesenchymal ones, together with the retention of the expression of many stemness markers, thus suggesting a switching from a mesenchymal stem-cell-like phenotype to a more epithelial stem-cell-like phenotype, compared with the adherent T88 and T93 cells (Fig. 1). Moreover, in the untreated tumourspheres, N-cadherin did not localise to the nucleus (Fig. 7B), as observed in the adherent cells (Fig. 1), exhibiting a prevalent cytoplasmic localization. In addition, N-cadherin was downregulated in the untreated T93 tumourspheres, while E-cadherin was upregulated in both the untreated and treated T88 and T93 tumourspheres (Fig. 7B) compared with the adherent cells (Fig. 1). The untreated tumourspheres also exhibited CD133, CD44v6, CD44 and β-catenin expression; the last two were mainly expressed at the plasma membrane (Fig. 8). Furthermore, as shown in Fig. 9, the cells in the untreated cancer spheroids retained the expression of mesen-chymal and stemness biomarkers, exhibiting a lower nuclear Snail expression compared with the adherent T88 and T93 cells (Fig. 6). ALDH1 expression was very low in the T88 cancer spheroids (Fig. 9) compared with both its paired adherent cells (Fig. 6) and T93 tumourspheres (Fig. 9). The T93 tumourspheres, exhibited a very low Oct4 expression (Fig. 9) compared with both the paired adherent cells (Fig. 6) and T88 tumourspheres (Fig. 9).

We further investigated the molecular basis of the GSK-3β-mediated inhibition of the ability of T88 and T93 cells to generate tumourspheres, by the immunofluorescence analysis of epithelial, mesenchymal and stemness markers, on spheroids obtained in medium containing LiCl. In accordance with above-mentioned findings on adherent cells, we observed a drastic downregulation of all stemness and mesenchymal markers analysed. As shown in Fig. 7, LiCl induced a down-regulation of Vimentin and N-cadherin. Moreover, the pan-Ck signal was lower in the LiCl-treated tumourspheres compared with the untreated tumourspheres, while E-cadherin (Fig. 7), CD44 and β-catenin (Fig. 8) exhibited a similar expression in the T88 and T93 spheroids obtained in medium containing or not containing LiCl. In accordance with the effect of LiCl on adherent cells, the expression of Snail, CD133, CD44v6 proteins (Fig. 8) and of all other stem-cell markers analysed (Fig. 9) strongly decreased in the tumourspheres following incubation with LiCl.

To further investigate the plasticity of the T88 and T93 cells, we disaggregated spheroids obtained by hanging drop assays, grown in medium with or without LiCl, (hereinafter referred to as treated and untreated tumourspheres) and cultivated them in adhesion, to examine the ability of the tumourspheres to re-adhere to a matrix and grow in adherent culture. As shown in Fig. 10A, these cells were able to adhere to the cell culture dishes and assume a mesenchymal phenotype. However, the numbers of surviving cells were significantly lower when they were derived from the treated spheroids compared to the cells derived from the untreated spheroids. Furthermore, the T93 cells, but not the T88 cells, derived from the treated spheroids, were able to proliferate and grow in adhesion (data not shown). By confocal microscopy, we observed that all cells obtained by disaggregating tumourspheres exhibited a similar phenotype with regards to the expression of mesenchymal and stemness markers (Figs. 10B and 11). However, cells derived from the treated tumourspheres exhibited a major nuclear localization of stem cells and mesenchymal markers (Fig. 10B) and an upregulation of the Snail transcription factor was observed only in the T93 cells, when analysed by RT-qPCR (Fig. 10C).

In both cell cultures, and under both experimental conditions, we observed the nuclear localisation of E-cadherin, N-cadherin, CD133 and CD44v6 (Fig. 10B), but not that of the canonical isoform of CD44 or β-catenin (Fig. 11). When analysed by RT-qPCR, Twist1 mRNA expression was downregulated in both cell cultures, while Snail mRNA was upregulated, mainly in the T93 cells, and even in cells disaggregated from the treated tumourspheres (Fig. 10C). Indeed, both cell cultures again expressed the mesenchymal protein, Vimentin, which was upregulated in the T88 cells derived from the treated tumourspheres (Fig. 11), in accordance with our previous observation showing that following 10 days of incubation with LiCl, Vimentin mRNA and protein expression levels strongly decreased in the T93 cells, while they increased in the T88 cells (24). Nuclear localisation was also observed for Snail, LGR5 and Sox2 (Fig. 12). Nanog exhibited cytoplasmic localisation in the cells derived from the untreated tumourspheres, particularly in the T88 cells, while a clear nuclear localisation in the cells derived from the treated tumourspheres was observed (Fig. 12). Oct4 also exhibited cytosolic localisation in the T88 cells derived from the untreated tumourspheres, but moved into the nucleus when the tumourspheres were treated with LiCl (Fig. 12). By contrast, Oct4 exhibited a nuclear expression in the T93 cells under both experimental conditions (Fig. 12). ALDH1 was undetectable under all the conditions analysed (Fig. 12).