Small molecules targeted to the microtubule–Hec1 interaction inhibit cancer cell growth through microtubule stabilization

Highly expressed in cancer protein 1 (Hec1) is a subunit of the kinetochore (KT)-associated Ndc80 complex, which ensures proper segregation of sister chromatids at mitosis by mediating the interaction between KTs and microtubules (MTs). HEC1 mRNA and protein are highly expressed in many malignancies as part of a signature of chromosome instability. These properties render Hec1 a promising molecular target for developing therapeutic drugs that exert their anticancer activities by producing massive chromosome aneuploidy. A virtual screening study aimed at identifying small molecules able to bind at the Hec1–MT interaction domain identified one positive hit compound and two analogs of the hit with high cytotoxic, pro-apoptotic and anti-mitotic activities. The most cytotoxic analog (SM15) was shown to produce chromosome segregation defects in cancer cells by inhibiting the correction of erroneous KT–MT interactions. Live cell imaging of treated cells demonstrated that mitotic arrest and segregation abnormalities lead to cell death through mitotic catastrophe and that cell death occurred also from interphase. Importantly, SM15 was shown to be more effective in inducing apoptotic cell death in cancer cells as compared to normal ones and effectively reduced tumor growth in a mouse xenograft model. Mechanistically, cold-induced MT depolymerization experiments demonstrated a hyper-stabilization of both mitotic and interphase MTs. Molecular dynamics simulations corroborate this finding by showing that SM15 can bind the MT surface independently from Hec1 and acts as a stabilizer of both MTs and KT–MT interactions. Overall, our studies represent a clear proof of principle that MT-Hec1-interacting compounds may represent novel powerful anticancer agents.


INTRODUCTION
Aneuploidy and chromosome instability are strongly involved in tumorigenesis. However, studies on the cancer susceptibility of mouse models heterozygous for genes controlling chromosome segregation suggest that aneuploidy may also act as a tumor suppressive mechanism, depending on the tissue analyzed and its intrinsic chromosome instability. 1,2 This evidence suggests that excessive chromosome mis-segregation may compromise genome stability and be incompatible with cell viability. Concordantly, partial downregulation of mitotic checkpoint proteins in tumor cell lines produced cancer cell death associated with severe chromosome mis-segregation. 3 Moreover, overexpression of cyclin E inducing multipolar divisions and numerous chromosome segregation errors was found to have a significant antitumorigenic effect in a mouse model for lung cancer. 4 Consequently, the idea of promoting cell death by inducing massive chromosome mis-segregation at mitotic division was proposed as a therapeutic strategy to selectively eliminate actively proliferating tumor cells. 5 Highly expressed in cancer protein 1 (Hec1) is a subunit of the kinetochore (KT)-associated Ndc80 complex, which mediates the attachment of microtubules (MTs) to the KT during mitotic division and ensures proper alignment and segregation of sister chromatids. 6,7 Moreover, HEC1 mRNA is highly expressed in many malignancies and elevated Hec1 expression is associated with negative prognosis in multiple cancer types. [8][9][10] Therefore, Hec1 represents a promising molecular target for developing new therapeutic drugs that exert their anticancer property by producing massive chromosome aneuploidy and cell death in cancer cells. We previously showed that expression of a Hec1 protein modified at its N-terminus, the region of interaction with MTs, massively kills cancer cells both in vitro and in tumor xenografts. Live cell imaging of N-terminus modified Hec1expressing cells demonstrated that cancer cell death is triggered by a prolonged mitotic arrest owing to an attempted chromosome segregation within multipolar spindles followed by cell death at mitosis. 11,12 These studies provided a clear proof of concept that the KT protein Hec1 can be a good target for developing anticancer therapies.
X-ray crystallographic studies have revealed that the N-terminal globular head of the Hec1 protein directly binds the MT lattice through a calponin homology domain. 13,14 Structural reconstructions of Ndc80 complexes bound to MTs have further restricted the region of interaction by identifying a region within the calponin homology domain of Hec1 named 'toe' region that binds a pocket on the MT present at both intra-dimer and inter-dimer tubulin interfaces. 15,16 In the 'toe' region positively charged amino acid residues interact with negatively charged residues on the 'toe print' MT interface, and residues important for MT binding within the 'toe' have also been identified. 14,17,18 The 'toe'-'toe print' interaction interface represents an ideal target for developing new compounds that specifically inhibit/modify the MT-Hec1 interaction and can potentially produce massive chromosome missegregation and cancer cell death.
In this study, the SPECS library of commercially available compounds was screened in silico aiming to find small molecules (SMs) able to bind at the interaction domain between MTs and Hec1. The adopted approach has led to the identification of a first hit compound and an analog of the initial hit with promising anticancer properties.

Virtual screening
The Hec1 'toe' region was demonstrated to bind the MT wall by interacting with intra-dimer and inter-dimer tubulin interfaces. Examining the model of the Ndc80 kinetochore complex (PDB code 3IZ0), an interesting and well-defined pocket was found at the interface between the two tubulin subunits (Figure 1a). Although this pocket seems not directly involved in the MT-Hec1 interaction, it is adjacent to the tubulin-Hec1 contact area and in proximity of Lys166, which is thought to be part of a MT conformation sensor. 16 The selected pocket, defined by Trp407 and Glu411 (α-tubulin chain) and Ile165, Asp163, Asp199 and Arg253 (β-tubulin chain), was used to screen the SPECS library of commercially available compounds (~350.000 SMs) aiming at identifying chemical inhibitors/modulators of the MT-Hec1 interaction. A high-throughput virtual screening identified 15 SMs as potential inhibitors/modulators of MT-Hec1 interactions. Among the 15 selected molecules, compound FB induced alterations in the mitotic process and reduced cell viability in HeLa cells (Supplementary Figure 1, IC 50 = 20.18 μM). To explore the structure-activity relationship and possibly increase the affinity for the target and the biological potency, several FB analogs were selected and their IC 50 was evaluated in dose-response experiments in HeLa cells (Supplementary Figure 2). Among the FB analogs, two compounds more cytotoxic than the lead compound (SM15 and SM17: IC 50 = 6.0 μM and 15.3 μM, respectively) and one inactive compound (SM16: IC 50 > 100 μM) were identified. The structures of FB and its active analogs are reported in Figure 1b. Time course experiments confirmed IC 50 studies by showing the high potency of SM15 and SM17 in inhibiting cell proliferation ( Figure 1c). Cancer specificity of the most cytotoxic molecule SM15 was then investigated by extending the analysis to a non Figure 1. A virtual screening targeted to the MT-Hec1 interaction identifies highly cytotoxic compounds. (a) Crystal structure of the α (sky blue) and β (pink) tubulin dimer in association with the Ndc80-SPC25 chimera protein (teal). The 'toe' region of Hec1 is in yellow. The zoom highlights the pocket (represented as gray molecular surface) at the interface between the two tubulin monomers where the virtual screening study has been performed. Important amino acid residues from α-tubulin chain and β-tubulin chain are reported. Hec1 Lys166 is in proximity of the selected pocket. The protein backbone is shown as ribbon. (  To support the assumption that the analogs directly interact with MTs, SM15 was subjected to surface plasmon resonance (SPR), which allows the evaluation of kinetics and affinity of the binding between proteins and SMs. SPR experiments showed that SM15 interacts directly and with high affinity with both assembled MTs ( Figure 1d) and dimeric tubulin ( Figure 1e). Scatchard plots of SPR experiments showed that SM15 binds to MTs with a higher affinity with respect to tubulin dimers (K D = 0.94 ± 0.1 μM and 1.80 ± 0.4 μM, respectively; Figure 1f).

Cell cycle arrest and apoptosis
To characterize the cell death pathway activated by the SMs we evaluated cell cycle distribution in HeLa cells by flow cytometry. SM15 treatment arrested HeLa cells in the G2/M phases of the cell cycle more efficiently than FB (Figures 2a and b) and significantly enhanced the percentage of hypodiploid apoptotic cells (Figures 2c and d). Analysis of mitosis Next, we sought to further investigate the possible mechanisms responsible for the observed cytotoxic effects. As cells defective for Hec1 function show chromosome segregation and spindle abnormalities, 6,[17][18][19] we analyzed the mitotic phenotypes of SMtreated HeLa cells. Mitotic progression from metaphase to anaphase was clearly inhibited after 3 h exposure to SM15, SM17 or FB (Figure 3a), suggesting activation of the spindle assembly checkpoint. Unattached or faulty attached KTs activate spindle assembly checkpoint that delays mitotic progression into anaphase until sister KTs attach properly to spindle MTs and chromosomes congress to the metaphase plate. 20, 21 We then analyzed chromosome congression by immunostaining with antibodies that visualize KTs and mitotic spindles ( Figure 3b). Treatment with SM15 and SM17 significantly increased the frequencies of unaligned, polar KTs, indicative of unattached or syntelically attached chromosomes (Figure 3b, SM15, Figure 3c). 22 These findings suggest that SMs delay mitotic progression by interfering with KT-MT attachment, thereby activating spindle assembly checkpoint. Consistently, KTs of unaligned chromosomes in SM15-treated cells accumulated the checkpoint protein BubR1 (Supplementary Figure 6). To test the nature of KT-MT interactions in the presence of MT-Hec1-interacting SM15, we took advantage of a calcium buffer that selectively depolymerizes both non-KT MTs and unstable KT-MT interactions. 23 Incubation with calcium buffer before fixation demonstrated that interactions between Hec1 and MTs, visualized by antibody staining, were maintained in the presence of SM15 (Figure 3d, arrowheads), indicating that the compound does not act as classical MT depolymerizing agents, such as nocodazole or colchicine, that prevent the formation of stable KT-MT interactions. The occurrence of polar chromosomes in the presence of stable KT-MT interactions led to the hypothesis that SM15 could inhibit the correction of syntelic or monotelic KT-MT attachments that maintain chromosomes close to the spindle poles. 22 To test this hypothesis, we used monastrol (MON), a chemical inhibitor of the kinesin-5 protein Eg5. As Eg5 activity is required for spindle pole . Consistent with our immunofluorescence results, chromosomes did not properly congress to the metaphase plate in treated cells. Furthermore, prometaphase arrest was followed by plasma membrane movements, DNA collapse and cell death (Figure 4e). The recorded death phenotypes recapitulate the morphological signs of mitotic catastrophe, a specific type of apoptosis elicited by mitotic disruption. 25,26 Hence, MT-Hec1-interacting compounds can promote cell death via mitotic catastrophe. Surprisingly, we also detected massive cell death starting from interphase in both HeLa (Figures 4f and h) and U2OS cells (data not shown). This occurred through extensive membrane blebbing (Figure 4f

Stabilization o f MTs
The unexpected finding that SM15 caused a strong induction of cell death from interphase prompted us to investigate the consequences of SM15 treatment on the MT network as a whole. To this aim, we induced MT depolymerization by incubating cells on ice (Figure 5a) and assessed the resistance of SM-treated cells to this procedure. Our data show that spindle MTs in control cells progressively depolymerized with increasing ice incubation times (Figure 5a) so that 480% of dimethylsulfoxide (DMSO)treated mitoses were fully depolymerized after 30 min on ice (Figure 5b). At opposite, spindle MTs in SM15-treated cells were resistant to MT depolymerization so that only half of the cells were fully depolymerized at the same time point (Figure 5b). A longer cold exposure (3 h) was then performed to depolymerize interphase MTs that are characterized by a lower dynamic instability. 27 In line with data on mitotic cells, the depolymerization of the cytoplasmic MT network caused by the low temperature in DMSO-treated cells was recovered in cells treated with SM15 (Figures 5c and d).
Altogether, these findings demonstrate that both mitotic and interphase MTs are stabilized by SM15.
Studies on tumor growth in vivo Next, we investigated the in vivo efficacy of SM15. To this purpose, established HeLa xenografts in nude mice were treated with SM15 starting from tumor palpability. As reported in Figure 6a, in vivo experiments confirmed in vitro results on HeLa cells. SM15-treated animals showed a significant inhibition of tumor growth starting from day 21 up to the sacrifice of the mice at day 25. Interestingly, apoptosis was confirmed as the death pathway activated by SM15, as determined by the enhanced poly ADP-ribose polymerase and caspase 3 cleavage in tumor protein lysates (Figures 6b and c) and by TUNEL assay (Figure 6d). Indeed, an increase of about two fold in the number of TUNEL positive cells/field was observed in treated xenografts (6.6 ± 1.2 vs 10.2 ± 1.4). Finally, SM15 did not produce any adverse health effects on mice, as monitored by diet consumption, body-weight loss, postural and behavioral changes. Altogether, these findings represent a clear proof of principle that MT-Hec1-interacting SM15 can exert anti-tumor activity.

Molecular dynamics simulation
To gain insights into the molecular mechanism mediating the interaction of SM15 with its targets, we investigated the modification exerted by SM15 on MTs and KT-MT attachments at the molecular level by carrying out a molecular dynamics (MD) simulation of the tubulin dimer-Hec1 complex in the presence of SM15. The predicted binding mode of SM15 is illustrated in Figure 7a and shows that the molecule occupies the selected binding area between the two tubulin subunits, interacting with the surrounding residues (for example, Asp163, Trp407 and Glu411) and placing the indole-bromophenyl-propenone group deep in the pocket. This best binding pose obtained for SM15 after docking at the isolated tubulin dimer interface was used as starting point for the simulation. After 35 ns of simulation, part of the compound gradually moved from the initial position on the pocket between the αand β-tubulin interface toward the Hec1 protein. After 45 ns, SM15 reached a binding conformation, inserted between the two tubulin subunits and Hec1, that was maintained until the end of the simulation. Particularly, the piperidine ring moved toward Hec1 inserting itself in a small hydrophobic pocket formed by Val122, Leu126, Phe147 and Tyr170. The rest of the molecule, especially the bromophenylpropenone part, remained between the two tubulin subunits, maintaining some interactions with the tubulin dimer ( Figure 7b). Altogether, these data indicate that the molecule can bind the MT surface independently from Hec1 and acts as a stabilizer of both the MT and the MT-KT interaction.

DISCUSSION
KTs represent an interesting anticancer drug target owing to their fundamental role in mitosis. In an effort to find SMs acting at KTs we carried out a virtual screening study targeting the interaction surface area between Hec1 and MTs. A new interesting and welldefined pocket at the interface between the two tubulin subunits was identified and its vicinity to the tubulin-Hec1 contact area made this pocket a potential target for our in silico studies. The virtual screening process led to the identification of two new SMs (SM15 and SM17) with elevated anti-proliferative, pro-apoptotic and anti-mitotic activities. The molecular interaction of SM15 to MTs had been further clarified. Docking studies on SM15 showed that the compound could potentially bind at the tubulin dimerization interface and could interact with amino acids from both αand β-tubulin, acting as potential bridge (stabilizer) between the two tubulin subunits. These findings are in line with the SPR results, in which SM15 was able to bind directly and with high affinity to MTs and (to a lesser extent) to tubulin. Our results also show that SM15 promotes chromosome mis-segregation by inhibiting the correction of faulty KT-MT attachments through the stabilization of the KT-MT interaction. MD simulation studies showed how SM15, after potentially binding at the tubulin dimer interface, could interact with Hec1 forming a potential bridge between Hec1 and tubulin, thereby stabilizing the KT-MT interaction. Furthermore, the molecular modeling suggests that SM15 can interact with tubulin also in the absence of Hec1, stabilizing MTs and does not affect the independent binding of Hec1 to tubulin. Live cell imaging demonstrates that cell death intervenes both from mitosis and from interphase and that this phenotype is associated with hyper-stable mitotic and interphase MTs. Overall, these experimental data, supported by the molecular modeling results, suggest that SM15 acts specifically toward the MTs, stabilizing MT dynamics in mitosis and interphase. Thus, we can conclude that the virtual screening study designed to identify novel molecules able to inhibit/modify the MT-Hec1 interaction led to the identification of a novel tubulin stabilizer drug. The tested molecule has the unique characteristic of interacting with the external surface of the MT cylinder, at odds with known MT stabilizing agents that interact at specific locations within the MT lumen. 28 Apart from a recent work that evidenced a binding site for paclitaxel at the outer MT surface before the drug reaches the internal lumen site, 29 several interaction sites have been mapped for MT stabilizing agents, but all reside in the internal lumen. 28 The interaction of SM15 on external surface of the MTs could provide an explanation for its effects outside mitosis, since its binding to MTs could efficiently affect already assembled interphase MTs. This conclusion is supported by the modeling studies that delineate the binding site of SM15 on the external interface between the tubulin chains. In addition, SM15 could stabilize the KT-MT interaction by bridging the tubulin dimer and Hec1.
In conclusion, starting from a rational approach, we identified a new compound, SM15, which present a dual mechanism: MT stabilizer agent and stabilizer of KT-MT interactions. Both these properties give to SM15 the potentialities to emerge as a novel powerful anticancer agent, promoting chromosome missegregation by stabilizing faulty KT-MT attachments. In addition to this mechanism, stabilization of MT dynamics in interphase by SM15 may represent an important feature of its anticancer activity, by promoting massive apoptosis of malignant cells, independently of mitosis.

Molecular modeling
All molecular docking studies were performed on a Viglen Genie Intel Core i7-3770 vPro CPU@ 3.40 GHz × 8 running Ubuntu 14.04. Molecular Operating Environment (MOE) 2015.10 30 and Maestro (Schrödinger Release 2016-1) 31 were used as modeling softwares. The cryo-electron microscopy density map and the Ndc80 docking model coordinates were downloaded from the PDB data bank (PDB code 3IZ0). 16 The tubulin dimer (chains A and B) together with the Ndc80-SPC25 chimera protein (Hec1containing chain C) were extracted from the protein model. The complex was preprocessed using the Schrödinger Protein Preparation Wizard by assigning bond orders, adding hydrogens and performing a restrained energy minimization of the added hydrogens using the OPLS_2005 force field. The library of commercially available compounds was downloaded from SPECS website (www.specs.net) in a sdf format and prepared using the Maestro LigPrep tool by energy minimizing the structures (OPLS_2005 force field), generating possible ionization states at pH 7 ± 2, generating tautomers and low-energy ring conformers. After removing the Ndc80-SPC25 chimera protein, a 12Å docking grid (inner-box 10Å and outer-box 22Å) was prepared using as centroid the area defined by Glu411 and Trp407 (α-tubulin chain) and Asp199, Ile165, Asp163 and Arg253 (β-tubulin chain) in the isolated tubulin dimer structure. Molecular docking of the prepared SPECS library was performed using Glide HTVS. The best 5000 ranked compounds obtained were docked using the more accurate Glide SP precision setting 1 as output pose per input structure. The output poses were then finally docked with Glide XP scoring function and the output database saved as mol2 file. The best 500 compounds according to the Glide XP results were visually inspected for their ability to bind the active site and 15 derivatives were selected and purchased.
MD simulations were performed on Supermicro Intel Xeon CPU ES-46200 @ 2.20 GHz × 12 running Ubuntu 14.04 using the Desmond package for MD simulation: OPLS-AA force field in explicit solvent, employing the TIP3 water model was used. The initial coordinates for the MD simulation were taken from the best docking result for compound SM15 at the tubulin dimer interface. A cubic water box was used for the solvation of the system, ensuring a buffer distance of~10 Å between each box side and the complex atoms. The system was neutralized adding 27 sodium counter ions. The system was minimized and pre-equilibrated using the default relaxation routine implemented in Desmond. A 100 ns MD simulation was performed, during which the equations of motion were integrated using a 2 fs time step in the NPT ensemble, with temperature (300 K) and pressure (1atm) constant. All other parameters were set using the Desmond default values. Data were collected every 40ps (energy) and every 160 ps (trajectory). Visualization of protein-ligand complex and MD trajectory analysis were carried out using Maestro. The MD simulation was performed in presence of Hec1 protein isolated from the Ndc80-SPC25 chimera protein (124 amino acids from Met79 to Ser202). L-glutamine and antibiotics. MCF10A cells were cultured in DMEM/F12 medium supplemented with 5% horse serum, 10 μg/ml insulin, 0.5 μg/ml hydrocortisone and 20 ng/ml epidermal growth factor. Cells were grown at 37°C in a humidified atmosphere containing 5% CO 2 and routinely tested for mycoplasm contamination. All SMs were stored as 10 mM stock solutions in DMSO at − 20°C and added after serial dilutions in order not to exceed 1% DMSO in medium. A total of 6 × 10 5 HeLa cells were seeded in 25 cm 2 flasks and exposed 24 h later to different concentrations of SMs. Cells were harvested 24 h later to obtain IC 50 values or after 24, 48, 72 h from SM addition in time course experiments. Cell counting was performed by a Z1 Coulter Counter (Beckman Coulter, Brea, CA, USA) and cell viability was assessed by trypan blue exclusion. Cell cycle progression and apoptosis were evaluated on the same samples by flow cytometry on an Epics XL apparatus (Beckman Coulter) using propidium iodide staining. Ten thousand events were collected from each sample.

SPR experiments
SPR experiments were carried out using a SensiQ Pioneer system. The sensor chip (COOH5) was activated chemically by a 35 μl injection of a 1:1 mixture of N-ethyl-N'-3-(diethylaminopropyl)carbodiimide (200 mM) and Nhydroxysuccinimide (50 mM) at a flow rate of 5 μl/min. Ligands, that is, tubulin (3900 RU) and assembled MTs (6050 RU), were immobilized on activated sensor chips via amine coupling as previously reported: 32 immobilizations were carried out in 10 mM sodium acetate at pH 4.0; flow cell Fc2 was left empty and used as a reference surface. The remaining ester groups were blocked by injecting 1 M ethanolamine hydrochloride (35 μl). SM15 was dissolved in DMSO to a final concentration of 10 mM, and then diluted 1:50 in buffer A (10 mM Hepes pH 7.4, 150 mM NaCl+0.05% surfactant P20+2 mM MgCl 2 ). Further dilutions of SM15 were carried out in running buffer (buffer A+2% DMSO); each sample was injected in duplicate on the sensor chip for 240 s (contact time) followed by a dissociation of 1000 s, at a flow rate of 30 μl/min. Regeneration procedures included two long (1000 s and 300 s) injections of buffer, separated by a 30 s injection of 10 mM glycine pH 1.5. The sensorgrams were analyzed using the SensiQ Qdat program.

Western blotting
Western blot experiments were performed as previously described. 33 In brief, cells were lysed in RIPA buffer and 40 μg of total proteins were resolved in 4-12% gradient gels by SDS-PAGE. Proteins were transferred onto nitrocellulose membranes and incubated with rabbit anti-poly ADPribose polymerase (11835238001, Roche Diagnostics GmbH, Mannheim, DE, USA), active caspase 3 (559565, BD Biosciences, St Jose, CA, USA) and goat anti-actin (Sc1616, Santa Cruz Biotechnology, Santa Cruz, CA, USA) antibodies. Protein lysates from tumors were probed with poly ADP-ribose polymerase cleavage antibody (AB3565, Millipore, Billerica, MA, USA) and HSP72/73 antibody (HSP01, Millipore). The membranes were subsequently probed with the appropriate horseradish peroxidase-conjugated secondary antibodies (Santa Cruz Biotechnology) and antigens on the membrane were revealed by enhanced chemiluminescence (RPN2209, GE Healthcare, Little Chalfont, UK).
Live cell microscopy A total of 3 × 10 4 HeLa or U2OS cells/well were seeded in four-well microslides (80426, Ibidi, Martinsried, DE) and time-lapse observation started 24 h later, when 5 μM SM15 was added to the medium. Cells were recorded under an Eclipse Ti inverted microscope (Nikon, Tokyo, Japan), using a Plan Fluor 40 × /0.6 NA or 60 × /0.7 NA objective (Nikon) for DIC and fluorescence, respectively; during the whole observation cells were kept in a microscope stage incubator (Basic WJ, Okolab, Naples, Italy) at 37°C and 5% CO 2 . Images were acquired over 24 h at 3 min for phase contrast or DIC and 30 min for fluorescence. Videos and still images were processed using NIS-Elements AR 3.2.

Tumor xenografts and TUNEL in nude mice
HeLa cells in exponential growth phase were harvested from the culture, washed and resuspended in PBS and 4 × 10 6 cells were subcutaneously injected into 6-8-week-old male CD-1 nude (nu/nu) mice. After the appearance of palpable tumors, animals were allocated into treatment groups (n = 8/group) on the basis of their tumor volume to obtain two similar groups, that subsequently received either no treatment or SM15. Two different experiments were performed. Animals were intraperitoneally treated with SM15 (10 mg/kg) sonicated and dissolved in 1% carboxymethyl cellulose for five consecutive days for three weeks starting from tumor palpability. Diet consumption, body-weight loss, and postural and behavioral changes were monitored daily. Animals were sacrificed 25 days after cell injection, tumors were removed and proteins extracted. We applied the Shapiro-Wilk test to evaluate the normality of the weight distribution of the treated mice and controls. Both distributions, at the two time points, were not normally distributed. Thus, in order to explore potential differences between the two treatment groups we used the nonparametric Mann-Whitney test. Results were considered to be statistically significant if P o0.05. Mice were purchased from Charles River Laboratories (Calco, Italy). All procedures involving animals and their care were authorized and certified by D.lgs 26/2014 (816/2015-PR del 11/08/2015) of the Italian Health Ministry. The immunohistochemical detection of apoptosis was performed by TUNEL assay using a commercial kit (In Situ Cell Death Detection Kit, POD, Roche). The assay was performed according to the manufacturer's instructions. For each tumor, three different 5 μm frozen sections were analyzed and examined by light microscopy. Sections were scanned at × 200 magnification. Apoptosis of tumor cells was counted in four high-power fields (×400 magnification) per section.

Statistical analysis
All data are presented as mean ± s.e.m. Significance of differences between experimental variables was determined by unpaired two-tailed Student's ttest unless otherwise stated. A P-value of 40.05 was considered significant. Statistical analyses were performed using the GraphPad Prism program. The sample sizes were determined by power analysis, based on variation shown in our previous experiments. 12 No data were excluded from any analysis. Evaluation of immunofluorescence and TUNEL samples was performed independently and in blinded manner by two investigators.