We previously found that in CRC cell lines, DLD1 and SW620, SR141716 induced G2/M and S-G2/M arrest, respectively, without induction of apoptosis or necrosis¹¹. Here we found that SR141716 inhibited HCT116 cell growth (Fig. 1a), induced a significant increase of Sub-G0/G1 cell phase, persistent until 48 hours (Fig. 1b) and raised the percentage of PI/Annexin V-FITC double stained cells (Fig. 1c). In both HCT116 and SW48 cell lines, the increase of Caspase 3- and PARP-cleaved protein levels starting from 24 hours (Fig. 1d and see Supplementary Fig. S1a) strongly suggested induction of apoptosis. This was also confirmed by results from human apoptosis antibody array, performed in HCT116 cells, revealing also an upregulation of Cytochrome C and of death receptors (TRAILR-1, -2 and -3) and downregulation of Bcl-2 and X-linked Inhibitor of Apoptosis Protein (XIAP) (see Supplementary Fig. S1b).Figure 1

Effect of SR141716 on cell viability. Cell viability (a), cell cycle analysis using propidium iodide staining (b) and apoptosis analysis (c) using AnnexinV-FITC/PI of HCT116 cells treated with SR141716 or vehicle alone. (d) Representative western blot analysis of Caspase 3 and PARP (total and cleaved forms) expression in total protein lysates from HCT116 and SW48 cells untreated (−) or treated (+) with the compound (mean ± SD; unpaired two tailed Student’s t-test *p < 0.05, **p < 0.01 and ***p < 0.005). Cropped blots from full-length gels are displayed in d.

In the absence of secreted Wnt ligands, the degradation complex destabilizes β-Catenin by phosphorylating it at Ser45 and Thr41/Ser37/Ser33. Ser33/Ser37 double phosphorylation (p-β-Catenin in the figures) marks β-Catenin for ubiquitin-mediated proteolysis^{19, 20}. In HCT116 and SW48 cell lines, SR141716 increased p-β-Catenin starting from 8 hours of treatment (Fig. 2). To assess whether the observed effects represent a common mechanism in CRC, the expression of total and p-β-Catenin was also evaluated in DLD1 and SW620 cell lines where the compound induced a transient increase of the p-β-Catenin not persistent at 48 hours (Fig. 2). In HCT116 cell line, where the SR141716-mediated β-Catenin phosphorylation was more effective and persistent, the analysis of nuclear and cytoplasmic extracts from treated cells revealed a reduced amount of β-Catenin in the cytoplasm at 24 hours and a nuclear localization lower than control cells at both 24 and 48 hours (Fig. 3a and b). The immunofluorescence staining of SR141716-treated cells, substantially confirmed the β-Catenin localization and the reduction of its nuclear translocation (Fig. 3c). Wnt3 triggers the canonical β-Catenin signalling through the bound with Fzd7 and LRP5 or LRP6 in several cancers, including CRC^{1, 21}. In HCT116, SR141716 was able to reduce Wnt3 protein levels and both Fzd7 and LRP6 co-receptor at the same time points. Furthermore, Dvl3 protein expression was inhibited by the treatment, suggesting that signal transduction across the plasma membrane and activation of Dvl3 not occurred. Despite with less extent, similar expression profiles were noticeable also in SW48 cell line, at least for Fzd7 and LRP6 (Fig. 4a and see Supplementary Fig. S2a). The β-Catenin phosphorylation- and degradation-complex also consists of GSK3 β whose activity can be inhibited by its Akt-mediated phosphorylation at Ser9²². We observed that SR141716 induced a precocious phosphorylation of GSK3 β in both cell lines, persistent until 48 hours of treatment in SW48 but not in HCT116 (Fig. 4a and see Supplementary Fig. S2b). Finally, a significant increase of APC levels was also observed in HCT116 cells (Fig. 4b and see Supplementary Fig. S2c). Wnt5 downregulation has been associated with higher tumor grade and poor prognosis^{1, 5, 23}, its overexpression inhibits canonical pathway and triggers the β-Catenin degradation or the inhibition of TCF/Lef-mediated transcription^{4, 24}. Wnt5 interacts with ROR2 (Receptor Tyrosine Kinase Like Orphan Receptor 2) tyrosine kinase receptor activating actin-binding protein, filamin A, and the JNK signalling pathway^{25, 26}. However, the non-canonical pathway triggers intracellular calcium flux, associated with CaMKII activation and canonical signalling inhibition. In HCT116 cells SR141716 increased protein levels of both Wnt5A and ROR2 at 8 hours of treatment and induced activation of CaMKII (Fig. 4c and see Supplementary Fig. S3). The results strongly support an SR141716-mediated inhibition of the canonical Wnt/β-Catenin signalling in both HCT116 and SW48 and a drug-induced activation of the non- canonical pathway triggered by Wnt5a in HCT116 cells.Figure 2

β-Catenin modulation in CRC cell lines. Representative western blot analysis of β-Catenin (total and phosphorylated form) expression in CRC cells treated with SR141716 (SR, 10 μM). The histograms represent the densitometric analyses of phospho-β-Catenin expressed as fold change of the total β-Catenin amount and normalized versus GAPDH (mean ± SD; unpaired two tailed Student’s t-test *p < 0.05, **p < 0.01 and ***p < 0.005). Cropped blots from full-length gels are displayed in left panel.

Aimed to confirm that SR141716 makes β-Catenin unable, or at least feeble, to activate the transcription of Wnt target genes, we performed luciferase assays in HCT116 cell line. In our model, the efficiency of the transient transfection was higher than 70% (see Supplementary Fig. S4a). In HCT116 cells, transiently transfected with the reporter containing tandem repeats of a specific Transcriptional Response Element (TRE) for TCF/Lef and treated with SR141716, the luciferase activity was significantly lowered of approximately 50% compared to untreated cells (Fig. 5a). Moreover, SR141716 was able to significantly reduce protein levels of some well known Wnt/β-Catenin target genes, such as Cyclin D1, c-Myc (Avian myelocytomatosis virus oncogene cellular homolog) and COX-2 (Cyclooxygenase-2), involved in CRC progression (Fig. 5b and c and see Supplementary Fig. S5).Figure 5

SR141716 downregulates Wnt/β-Catenin target gene expression. (a) SR141716-mediated effects on luciferase activity controlled by TRE for TCF/Lef-binding element in HCT116 cell line. Histograms represent luciferase activity measured at 18 hours from transfection in whole cell extracts from HCT116 cells transfected with reporter construct containing the TRE for TCF/Lef and treated with SR141716 (SR, 10 μM) or vehicle. Firefly luciferase was normalised to Renilla luciferase reading and the data were plotted as fold change (mean ± SD; unpaired two tailed Student’s t-test ***p < 0.005) compared to control (vehicle treated) cells. Western blot analysis of Wnt/β-Catenin targets in HCT116 (b) and in CRC (c) cell lines untreated (−) or treated (+) with SR141716. Cropped blots from full-length gels are displayed in b and c.

We tested SR141716 efficacy in vivo in a subcutaneous (s.c.) HCT116 xenograft model. Tumor cell suspension was injected s.c. into 20 female SCID mice and when the tumor reached approximately the size of 50–70 mm³, 10 mice in the treated group received the peri-tumoral injection of SR141716, while 10 mice in the control group received vehicle alone, three times a week for 6 weeks. The tumor sizes have been recorded on the first day of SR141716 treatment (day 0) and bi- or three-weekly at the indicated time points. Mice in the control group developed tumors beyond 2,0 cm³ on average by day 42. In contrast, the mice in SR141716 group developed much smaller tumors (Fig. 6a). In particular, starting from the thirtieth day of treatment, ANOVA analysis indicates a significant smaller tumor size in treated group compared with animals in the control group (p < 0.001) (Fig. 6b). Excised tumor sections were analyzed for β-Catenin localization in cellular compartments through immunofluorescence staining with specific antibodies for β-Catenin and for Lamin A/C (green and red fluorescence, respectively in Fig. 6c). The confocal microscopy demonstrated that in tissue sections from treated mice β-Catenin localized mainly in the cytoplasm whereas nuclear staining was almost devoid of specific β-Catenin signal (Fig. 6c and see Supplementary Fig. S6a). Finally, western blot analysis of total extracts from tissue specimens demonstrated that, despite the awaited tumor samples heterogeneity, the amount of p-β-Catenin, clearly detectable in the treated xenografts, was lost in control tumors. In contrast, the immunoreactivity for both Cyclin D1 and c-Myc, distinguishable in control samples, were just barely apparent in whole extracts from treated tumors (Fig. 6d). Quantification of western blot analysis, performed in samples from six tumors per condition, seems to sustain this feature even if it not reach statistical significance at least in our xenograft groups (see Supplementary Fig. S6b).Figure 6

SR141716 reduces in vivo tumorigenicity of HCT116 through inhibition of Wnt/β -Catenin canonical pathway. (a) Representative growth of HCT116 xenografts in control (upper panel) and in treated groups (lower panel) at day 42. (b) Tumor volume growth curve after peri-tumoral injection of SR141716. Growth retardation by the compound was statistically significant for all time points labelled with * (one-way ANOVA p < 0.05) or with # (one-way ANOVA p < 0.001). (c) Immunofluorescence staining of HCT116 xenograft tumor sections (10 μm; 42 days from treatment beginning) performed for Lamin A/C (red fluorescence) and β-Catenin (green fluorescence) localization. The image shown represents 3D front view; data are representative of at least three sections from each control and treated tissue sample. (d) Western blot analysis of total and phosphorylated β-Catenin, Cyclin D1 and c-Myc in whole lysate from resected tumor tissues. Cropped blots from full-length gels are displayed in d.

Obtained data substantially supported a drug-mediated inhibition of the canonical Wnt/β-Catenin signalling in CRC cell lines carrying stabilizing mutation of β-Catenin, such as HCT116 and SW48. Moreover, even though we found a direct inhibition of the TCF/Lef-mediated transcriptional activation in HCT116 but not in DLD1 and SW620 cell lines, in several models of human CRC the compound was able to decrease the expression of well known β-Catenin target genes (Fig. 5c and see Supplementary Fig. S5). We hypothesized that a member of the β-Catenin-TCF/Lef complex, other than β-Catenin, could be involved in the observed antitumor effects. Aimed to identify factors potentially able to interact with SR141716, we performed an in silico Inverse Virtual Screening testing the case-study compound on a panel of 306 proteins involved in cancer and inflammation events (see Supplementary Table S1). Briefly, this computational tool allows the analysis of different binding hypotheses between a single ligand and a high number of targets through molecular docking experiments, determining the selection of the most promising ligand-receptor favourite complexes after a normalization of the predicted binding affinities, and successfully directing the subsequent biological assays^27–31. Concerning the first two identified targets (A_2A 1^st position, and ErbB4, 2^nd position in the ranking) (see Supplementary Table S2) we first of all considered their expression in our panel of CRC cell lines. A_2A (Adenosine A2a Receptor, ADORA2A) is a component of the Adenosine receptor family comprising four G-protein coupled receptors (A₁, A_2A, A_2B and A₃) linked to Ca²⁺ mobilization and Cyclic AMP increase. Low amount of A₁, A_2A, and A_2B receptors have been detected in colon cancer cell lines cultured in normoxic condition, as in our experimental procedures^{32, 33}, whereas both HCT116 and DLD1 cells expressed high levels of the A₃ receptor subtype^32–34. ErbB4/Her4 (Erb-B2 Receptor Tyrosine Kinase 4) is a member of the ErbB protein tyrosine kinase family, which also includes EGFR/ErbB1/Her1. Despite recently an over-expression of ErbB4 was found in human CRC tissues, in cultured colon cancer cell lines ErbB4 protein expression is difficult to detect and mainly unmistakable in poorly differentiated CRC cells such as HCT116 in our panel^{35, 36}. Therefore, among the obtained results, we were intrigued by p300/KAT3B target at the 3^rd position in the final ranking of predicted most affine proteins of SR141716³⁷. Specifically, the careful analysis of the sampled docking poses enforced this result, showing a good accommodation of SR141716 in the p300/KAT3B binding site and supporting the potential inhibition of the histone acetyltransferase (HAT) activity exerted by the investigated compound. We found two interesting binding modes in which SR141716 is placed in p300/KAT3B occupying the ligand binding site (LBD) and exerting both polar and hydrophobic interactions. The analysis of the first pose, associated to the best docking score (ΔG_bind = −11.2 kcal/mol), disclosed the arrangement of SR141716 in the deep part of the LBD supported by an edge-to-face π-π interaction between the pyrazole core and the indole moiety in the side chain of Trp1466, and an H-bond with the carbonyl oxygen in the backbone of Leu1398 (Fig. 7a). Further polar interactions were established with Ser1396, Asp1399, Ser1400, Arg1410, Gln1455, Lys1456, and hydrophobic contacts with Tyr1414, Leu1463, Trp1466, Tyr1467 (Fig. 7a and c). Another interesting binding mode (ΔG_bind = −10.8 kcal/mol) showed the placement of the molecule in a more external part of the binding site, supported by halogen bonds between the dichloro-phenyl part of SR141716 and Arg1410 (Fig. 7b), while the edge-to-face π-π interaction between the pyrazole core and Trp1466 was again detected (Fig. 7b and d). The direct binding of SR141716 to the HAT catalytic domain (aa 1284–1673) of human recombinant p300/KAT3B was corroborated by the results of a surface plasmon resonance (SPR) assay, performed according to a well-established protocol^{38, 39}. In fact, Fig. 7e clearly shows a direct interaction between SR141716 and p300/KAT3B, displaying a concentration dependent SPR signal not observed with the negative control (see Supplementary Fig. S7). Fluorometric in vitro assay suggested a dose-dependent inhibition of p300/KAT3B HAT activity by SR141716 (5 μM–60 μM) (Fig. 8a). Moreover, the compound decreased the amount of acetyl-Histone H3 and acetyl-Histone H4 in both purified histones and total lysates from SR141716-treated HCT116 cells (Fig. 8b). The decreased acetylation seems a reproducible effect triggered by Rimonabant in CRC cell lines irrespective of the mutational profile and Wnt/β-Catenin signalling and thus can be highlighted also in DLD1 and SW48 cell lines (Fig. 8c and d).Figure 7

SR141716 interacts with p300/KAT3B. (a) 3D docking models of SR141716 (colored by atom types: C green, O red, N blue, polar H light gray, Cl dark green) in the binding site of p300/KAT3B (secondary structure depicted in orange ribbons) and (c) associated 2D interaction panel; (b) alternative 3D docking models of SR141716 and (d) associated 2D interaction panel. Residues in the active site are represented in sticks (colored by atom types: C grey, N blue, O red, S yellow, H light gray). H-bonds ligand/protein interactions are represented in green dotted lines, while π-π interactions are depicted with cyan dotted lines and halogen bonds are reported with violet dotted lines. (e) Sensorgrams obtained from different injections (0.62, 1.25, 2.5, 4 and 5 μM) of SR141716 to immobilized hKAT3B/p300 (catalytic domain, aa 1284–1673).

SR141716 (Rimonabant) and AVE1625 were kindly donated by Sanofi-Aventis (Montpellier, France). It was dissolved in DMSO and added to cells cultures at the indicated concentrations. Anti-β-Catenin, anti-Dvl3, anti-Fzd7, anti-APC, anti-Wnt5A, anti-ROR2, anti-phospho-CaMKII anti LRP5 and anti-Histone H3 were from Abcam. Anti-Cyclin D1 and anti-Lamin A/C were purchased from Becton Dickinson and Sigma-Aldrich, respectively. Anti-acetyl-Histone H4 and anti-acetyl-Histone H3 were from Santa Cruz Biotechnology and Merck Millipore, respectively. Anti-Annexin V FITC conjugated was purchased from Miltenyi Biotec. Primary antibodies not previously reported, secondary HRP-linked goat anti-mouse or goat anti-rabbit IgG, were all from Cell Signalling Technology. All the cell culture reagents were from Sigma–Aldrich, Inc.

Human CRC cells DLD1, SW620, HCT116 and SW48 were obtained from the Interlab Cell Line Collection (IST, Genoa, Italy). Cells were routinely grown in RPMI-1640, in Dulbecco’s modified Eagle medium (DMEM), in McCoy’s 5 A or in DMEM/F12 medium, respectively, at 37 °C in a 5% CO₂ atmosphere as described previously^{15, 18}. Cells were exposed to various concentrations of SR141716 for the times showed in the figures and to evaluate cell viability the MTT assay was used as described previously¹⁸.

Cells were plated in 100 mm dishes, serum starved for 18 hours to synchronize at the G₁/S interface and treated with SR141716 (10 μM) for the indicated time. Flow cytometry and FACS analysis were performed as described previously¹¹. The cell cycle analysis was performed with ModFit LT v3.2 (Verity Software House, Inc.); 10000 events, corrected for debris and aggregate populations, were collected. To evaluate the apoptosis, at least 20000 events were collected, and the data were analyzed with FlowJo^® software (BDIS).

Whole-cell lysates were prepared as described previously¹⁵. Subcellular fractionation was obtained by using NE-PER^® Nuclear and Cytoplasmic Extraction Reagents (Thermo scientific, Pierce biotechnology). Frozen tumor pieces from xenografts were disrupted by gentle homogenization (Potter-Elvehjem Pestle) in cold RIPA buffer. 10–30 μg of proteins were loaded on SDS–polyacrylamide gels under reducing conditions and western blot analyses were performed as described previously¹⁵. Immunoreactive bands were quantified with Quantity One 1-D analysis software (Bio-Rad).

Cells were transiently cotransfected with the TCF/Lef firefly luciferase construct (100 ng) and the Renilla luciferase vector (10 ng) (Cignal^TM TCF/LEF Reporter (GFP) assay; QIAGEN). Luciferase activity was measured using a Dual-luciferase assay system (Promega) and an EnSpire-2300 luminometer (Perkin Elmer). Relative firefly luciferase activity was obtained by normalizing it to that of Renilla luciferase activity.

Cells were grown on slides in 12 well plates (3 × 10⁴ cells/well). After treatment, cells were fixed in parafolmadehyde (PFA, 3,7% v/v in PBS), permeabilized in Triton X-100 (0,1% v/v in PBS), blocked with 4% Bovine Serum Albumin (BSA) and incubated with anti-β-Catenin (Abcam) and anti-lamin A/C (Sigma-Aldrich) primary antibodies, at 4 °C overnight. Immunofluorescence staining was performed with Alexa Fluor^® 488 donkey anti-rabbit and Alexa Fluor^® 594 goat anti-mouse IgG (Molecular Probes^®) secondary antibodies. Samples were vertically scanned from the bottom of the coverslip with a total depth of 5 μm and a Plan-Apochromat oil-immersion objective (magnification 63X^*1.7; 1.40 NA). A total of 10 z-line scans with a step distance of 0.5 μm were collected and single planes or maximum-intensity z-projection of stacks and an orthogonal projection (=xy, xz, yz planes for z-stacks series) were generated.

Fixed tumor pieces from xenografts were OCT-embedded, sectioned (10 μm), stained with primary and secondary antibodies (diluted in PBS/1% BSA/0,05% Triton-X-100).

A Zeiss LSM510 Laser Scanning Microscope (Carl Zeiss MicroImaging GmbH) for data acquisition was used.

Total RNA extraction, cDNA synthesis and reverse-transcription PCR were performed as described¹⁵. Primer pairs specific to human β-Catenin (5′-GTCCGCATGGAAGAAATAGTTGA-3′ forward and 5′-AGCTGGTCAGCTCAACTGAAAG-3′ reverse) or to human actin (5′-ACTGGGACGACATGGAGAA-3′ forward and 5′-ATCTTCATGAGGTAGTCAGTCA-3′ reverse) were used. All reactions were performed at least in triplicate; the PCR products were quantified with Quantity One 1-D analysis software (Bio-Rad) and results were normalized to those obtained from actin B.

The chemical structures of investigated compounds (SR141716, plus 30 “blank” compounds) were built with Maestro (version 10.2) Build Panel (Maestro, version 10.2, Schrödinger, LLC, New York, NY, 2015). Prior to perform molecular docking calculations, optimizations (Conjugate Gradient, 0.05 Å convergence threshold) of the structures were applied to identify possible three-dimensional models. Then, all the structures were converted in the pdbqt format using OpenBabel software (version 2.3.2)⁶⁴, adding Gasteiger charges.

306 protein 3D structures were prepared downloading the pdb files from the Protein Data Bank database (www.rcsb.org, see Supplementary Table S1). For each structure, “non-structural” water molecules were removed, and the processed file was then converted in.pdbqt format, merging non polar hydrogens and adding Gasteiger charges. Information about the panel of proteins are reported in Table S1.

Molecular docking calculations were performed using the Autodock-Vina software⁶⁵. In the configuration files linked to 3D structures of the protein, coordinates and dimensions along x,y,z axes of the grid related to the site of presumable pharmacological interest, with spacing of 1.0 Å between the grid points. The exhaustiveness value was set to 64, saving 10 conformations as maximum number of binding modes. For all the investigated compounds, all open-chain bonds were treated as active torsional bonds.

A first set of promising interacting proteins of SR141716 was selected setting a predicted binding affinity cutoff = −7.5 kcal/mol. The identified proteins (166 items) were then also screened against “blank” molecules, the latter needed for the normalization^27–31 of the binding affinities of SR141716, as reported in equation 1:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V={V}_{0}/{V}_{{\rm{R}}}$$\end{document}V=V0/VRwhere, for each target investigated, V represents the normalized value of SR141716, V₀ is its predicted binding affinity from docking calculations (kcal/mol), V_R is the average value of binding energy calculated on all the “blanks” (kcal/mol). It is important to note that V is a dimensionless number, and then it can be used to predict the interacting targets of a case-study compound, rather than to have precise indications about the related binding affinities. After the normalization process a final ranking was obtained, from the most to the less promising target. Normalized values and predicted binding energies for SR141716 are collected in Table S2 (see Supplementary Table S2), respectively. The ligand/protein complexes were visually inspected with Maestro (version 10.2). Illustrations of the 3D models were generated using Maestro (version 10.2).

SPR analyses were performed on a Biacore 3000 optical biosensor equipped with research-grade CM5 sensor chips (Biacore AB). Recombinant p300/KAT3B (Enzo Life Sciences, catalogue number BML − SE451; GenBank accession number NM_001429) HAT domain was immobilized (30 μg/mL in 10 mM sodium acetate, pH 4.5) at a flow rate of 10 μL/min by using standard amine-coupling protocols to obtain a density of 15 kRU. Myoglobin was used as negative control, and one flow cell was left empty for background subtractions. Rimonabant, dissolved in DMSO (100%), was diluted in HBS-P (50 mM HEPES pH 7.4, 150 mM NaCl, 0.005% Tween 20) and injected at 0.62, 1.25, 2.5, 4 and 5 μM always maintaining a final 0.5% DMSO concentration. Binding experiments were performed at 25 °C by using a flow rate of 30 μL/min, with 120 s monitoring of association and 200 s monitoring of dissociation. Regeneration of the surfaces was performed, when necessary, by a 10 s injection of 1 mM NaOH. The simple 1:1 Langmuir binding fit model of the BIAevaluation software was used for determining equilibrium dissociation constant (K_D) from kinetic dissociation (k_d) and association (k_a) constants.

Total histones were extracted from cell lines (3 × 10⁶) treated with SR141716 (10 μM) or vehicle alone using Histone extraction kit (Abcam) following the manufacturer’s protocol and stored at −80 °C until use. The protein concentration of the eluted histones was estimated using a Bradford protein detection kit (Bio-Rad, Hercules, CA) as previously described¹⁵. The p300/KAT3B enzymatic activity was performed using a KAT3B/p300 Inhibitor fluorometric assay kit (Abcam) according to the manufacturer’s protocol. Briefly, assay was performed with recombinant p300/KAT3B in the presence of SR141716, AVE1625, anacardic acid (used as inhibitor control) or vehicle alone. Histone H3 peptide and Acetyl CoA were used as acetylation substrates. The amount of fluorescence in the solution, amplified by a thiol detecting probe, was measured at Ex/Em = 392/482 nm with an EnSpire 2300 (Perkin Elmer) multimode plate reader. The results were expressed as percentage of inhibition compared to the enzyme control. Quantification of total acetyl-Histone H3 and H4 was performed in histone extracts, purified as earlier described, through a colorimetric ELISA assay, and in total nuclear extracts, using western blot assay. In both analysis, extracts from cell lines treated with SR141716 (10 μM) or with the vehicle alone, for the times shown in the figures, were evaluated in at last three independent experiments.

Data obtained from multiple experiments were calculated as means ± SD, if not otherwise specified, and analyzed for statistical significance by using the two tailed Student t-test, 1- or 2-way ANOVA for independent groups, with the Tukey or Bonferroni correction for multiple comparisons. All data shown are representative of at least three independent experiments performed in triplicate. Values of p < 0.05 were considered statistically significant.

All data generated or analysed during this study are included in this published article (and its Supplementary Information files).

Supplementary Information