Pharmacophore-guided discovery of CDC25 inhibitors causing cell cycle arrest and tumor regression

CDC25 phosphatases play a key role in cell cycle transitions and are important targets for cancer therapy. Here, we set out to discover novel CDC25 inhibitors. Using a combination of computational methods, we defined a minimal common pharmacophore in established CDC25 inhibitors and performed virtual screening of a proprietary library. Based on the availability of crystal structures for CDC25A and CDC25B, we implemented a molecular docking strategy and carried out hit expansion/optimization. Enzymatic assays revealed that naphthoquinone scaffolds were the most promising CDC25 inhibitors among selected hits. At the molecular level, the compounds acted through a mixed-type mechanism of inhibition of phosphatase activity, involving reversible oxidation of cysteine residues. In 2D cell cultures, the compounds caused arrest of the cell cycle at the G1/S or at the G2/M transition. Mitotic markers analysis and time-lapse microscopy confirmed that CDK1 activity was impaired and that mitotic arrest was followed by death. Finally, the compounds induced differentiation, accompanied by decreased stemness properties, in intestinal crypt stem cell-derived Apc/K-Ras-mutant mouse organoids, and led to tumor regression and reduction of metastatic potential in zebrafish embryo xenografts used as in vivo model.

to keep up with expectations, either due to rapid metabolism in tumor-bearing SCID mice 20 or for not completing clinical trials 21 , and have thus not attained approval. In light of the recent discovery that CDC25 is the therapeutic target of choice in triple-negative breast cancers, namely those that are negative for estrogen-, progesterone-and HER2-receptor expression and that are unresponsive to standard therapy 22 , we set out to develop novel CDC25 inhibitors. To this end, we conducted a pharmacophore-guided drug discovery program that led to the identification of scaffolds of the naphthoquinone group displaying inhibition of CDC25 in enzymatic assays. In cultured cells, the most potent compounds induced inhibition of CDK1 activity and function, with block of mitotic transition followed by cell death. In mouse Apc/K-Ras mutant duodenal organoids, low doses of CDC25 inhibitors caused arrest of proliferation and expression of differentiation markers, whereas high doses induced cell death. In zebrafish embryos, used as in vivo xenograft model, the CDC25 inhibitors led to tumor regression and reduction of metastases.

Results
Pharmacophore-guided library screening and hit selection. To the end of retrieving novel CDC25 inhibitors from an in silico virtual library that was built from a proprietary database of synthetic molecules, we implemented a number of computational strategies (Fig. S1A), according to established protocols 23 . First, CDC25 inhibitors belonging to three classes -natural products, quinones and electrophiles 24 -were subjected to a linear fragmentation process 25 implemented in MOE Suite 26 , in which input structures were split into small pieces by removing the least "scaffold-like" extremity until indivisible essential fragments were obtained. Next, the molecular entities returned by this process, ordered by increasing size, were used to build a series of pharmacophore models (Fig. S1B). The latter were optimized until the achievement of a final model, representative of the chemical features of scaffolds obtained from the fragmentation process. Finally, this model was used to examine a proprietary library through a pharmacophore-guided virtual screening process (MOE Suite). Compounds obtained from the first round of hit selection and belonging to different molecular families were tested at fixed concentration on recombinant CDC25A (Table S1). Reference compound in all tests was the established CDC25 inhibitor NSC-663284, a para-quinonoid derivative of vitamin K 27 . Naphthoquinones UPD-140 (Fig. 1A, 2-(2′,4′-dihydroxyphenyl)-8-hydroxy-1,4-naphthoquinone) and UPD-176 (Fig. 1B, 5-hydroxy-2-(2,4-dihydroxyphenyl) naphthalene-1,4-dione) appeared to be the most effective inhibitors of CDC25 phosphatase activity. Based on the structure of UPD-140 and UPD-176, and exploiting the crystal structure of CDC25B 28 , along with available homology models for CDC25A and CDC25C, we performed hit expansion/optimization through a molecular docking strategy (Fig. 1C). Identification of pockets and surface sites through the localization of regions of tight atomic packing suggested two close cavities suitable to accommodate the compounds. Both cavities are highly conserved in the three enzymes and one is superimposable with the phosphatase catalytic site (Table 1). Starting point for prioritization of scaffolds was the presence of a quinone moiety, which appeared to be a necessary condition for optimal anchoring of compounds in CDC25 catalytic pocket. Reassessment of the library, based on the structure of UPD-140 and UPD-176, followed by in vitro enzymatic assays revealed eight additional compounds as effective inhibitors of CDC25 phosphatase activity, all of them being 1,4-naphthoquinones with hydroxyl groups either in position 5 or 8 (Table S2).  29 . A similar pattern of inhibition was observed on CDC25B and CDC25C, though with slightly higher IC 50 values ( Fig. 2A). Kinetic analysis conducted with UPD-795 revealed a mixed mechanism of enzyme inhibition, as indicated upon data analysis with the mixed-model inhibition equation 30 (Table S3 and Fig. S2B), confirming the predictions of docking studies. Considering that quinones are oxidizing agents, known for their ability to generate reactive oxygen species (ROS) in biological systems, and that the Cys residue in CDC25 active site is highly susceptible to oxidation 31 , we decided to assess a possible role for redox reactions in the inhibition of CDC25 by the most potent naphthoquinones described here. To this end, we treated CDC25A with an excess of the reducing agent DTT before addition of the compounds at a concentration proximal to their IC 50 . The data indicated that the presence of DTT in the reaction prevented inhibition of CDC25 (Fig. 2B).

Enzymatic characterization and profiling of CDC25 inhibitors.
Furthermore, since the quinone scaffolds so far described in the literature are in part reversible and in part irreversible CDC25 inhibitors, we decided to assess the mode of action of our most potent naphthoquinone inhibitors. To this end, we pre-incubated CDC25A with excess compound (5 μM, corresponding to ~3x IC 50 ) and subsequently assayed for remaining phosphatase activity upon 10-fold dilution of the pre-incubation mix (hence at 0.5 μM compound). The data showed that the enzyme activity remaining upon dilution of the pre-incubation mix was comparable to the activity detected upon direct treatment of the phosphatase with 0.5 μM compounds, indicating a reversible mode of action (Fig. 2C). This was not the case for NSC-663284, which is an established irreversible inhibitor of CDC25 (Fig. 2C).
To assess specificity, we profiled two of the most potent CDC25 inhibitors, UPD-795 and UPD-140, against a panel of protein phosphatases. The data revealed that, when tested at concentrations close to the IC 50 for CDC25A, both compounds also inhibited human PP5 activity to about 50% (Table S4), whereas none of the other phosphatases examined was affected.
To evaluate in a cellular context the oxidation-based mechanism of inhibition, we treated cells with the aminothiol N-Acetylcysteine, an antioxidant, prior to administration of the compounds. At compound doses exceeding 3x the IC 50 (5 μM), cell viability was rescued to ~50% of controls in the presence of the reducing agent (Fig. 3B), hence confirming results obtained in biochemical assays. To assess effects of the CDC25 inhibitors on cell cycle progression we administered compounds to HeLa cells and examined DNA content by flow cytometric analysis. In asynchronous HeLa cells, treatment with UPD-596, UPD-738, UPD-786, UPD-793 or UPD-795 caused an increase of the G1 population, indicating that G1/S was likely the point of action for these compounds. On the other hand, UPD-176, UPD-787 and UPD-790 were most effective in causing accumulation of cells at G2/M (Fig. 3C). Similar results were obtained in U2OS cells (Fig. S4). Since HeLa cells can be conveniently synchronized at specific points in the cell cycle (Fig. S5A), they were treated with UPD-176 or UPD-795 at the time of release from a double-thymidine block (early S-phase). Under these conditions, progression through S-phase was effectively held in check as compared to controls (Fig. S5B). To precisely assess effects of the compounds on the G1/S transition, HeLa cells were synchronized at G2/M, released to allow completion of mitosis and re-entry into the cell cycle, and treated in mid-G1 with UPD-176, UPD-596, UPD-738 or UPD-793. Analysis of cell cycle progression at 12 h post-release, a time when control cells entered S-phase, showed that UPD-176, UPD-738 and UPD-793 effectively blocked cells in G1, whereas UPD-596 appeared to be ineffective in this respect (Fig. 3D). To assess the effect of compounds on the G2/M transition, double-thymidine synchronized HeLa cells were released and treated with UPD-176, UPD-787, UPD-790 or UPD-140 5 h upon release, namely in late S-phase (Fig. S5A). Flow cytometric analysis of cell cycle progression at 12 h post-release, when control cells largely moved to the next G1 phase, showed that all compounds caused accretion of the G2/M peak, with UPD-790 apparently being the most effective of all (Fig. 3E). To exclude the possibility that the observed cell cycle arrest would be secondary to genotoxic effects, we examined the biomarker γ-H2AX at the onset of mitosis in cells treated with the compounds. The data showed that, in comparison to a potent genotoxic agent, no damage to DNA occurred upon treatment with the naphthoquinone compounds ( Fig. S6).
To visualize the execution of mitosis, we administered compounds to HeLa cells 5 h upon release from a double-thymidine block, and monitored cells by time-lapse microscopy for the subsequent 14 h. We observed that whereas control cells timely progressed through mitosis and moved to the next cycle, cells treated with UPD-787 could not complete mitotic transition and died before reaching G1 (Fig. 4A). A similar pattern was obtained with compounds UPD-790 and UPD-176 ( Fig. S7A and data not shown). To better appreciate the response to inhibitors at the onset of mitosis, we administered low doses (5 μM) of UPD-787 or UPD-790 to synchronized HeLa Kyoto cells, which carry mCherry-H2B and GFP-α-tubulin. We observed a failed attempt to round up and execute mitosis, which was followed by membrane blebbing and death ( Fig. S7A and Movies M1-M3). Western blot analysis revealed caspase-mediated cleavage of PARP-1, a signature of apoptosis, in cells treated with the compounds (Fig. S7B). Biochemical analysis of mitotic markers at 10 h from the double-thymidine block-release showed decreased MPM2 epitopes in treated cells, a read out for CDK1 activity, and reduced Histone H3 phosphorylation at Ser 10 , a read-out for chromosome condensation (Fig. 4B). To directly assess cellular CDK1 activity under these conditions we used antibodies recognizing phosphorylated Thr 14 or Tyr 15 , two sites in the P-loop of CDK1 catalytic domain, the phosphorylation of which hampers kinase activity and that are selectively dephosphorylated by CDC25 32,33 . As expected, at 12 h from the double-thymidine block-release point, control cells progressed to the G1 phase of the next cycle displaying low pThr 14 15 -CDK1 in cells treated with CDC25 inhibitors appeared to be intermediate between that of cells treated with Ro-3306, a specific inhibitor of CDK1, and of cells treated with Nocodazole, an agent that interferes with tubulin polymerization causing arrest at pro-metaphase with active CDK1 (Fig. 4E).

CDC25A overexpressing cell lines are sensitive to CDC25 inhibitors.
To determine whether cancer cells overexpressing CDC25 are sensitive to treatment with the compounds identified in this study, we initially selected cell lines that carry activating bi-allelic mutation of K-Ras (G12S) and are reported to express high level of CDC25B (http://www.proteinatlas.org/). However, Western blot analysis of cytoplasmic and nuclear extracts of such cell lines (A549 and Colo741) compared to HeLa, revealed lack of correspondence between the mRNA levels reported in databases and actual protein expression (Fig. S8). Similar results were obtained upon analysis of CDC25A and CDC25C protein expression (Fig. S8). Given the lack of an appropriate cell line model, we resorted to a system where CDC25A is expressed in a tetracycline-dependent manner and has been previously described 34 . Viability assays showed that cells grown in the absence of tetracycline, a condition inducing ectopic expression of HA-CDC25A to high extent, remained sensitive to treatment with UPD-176 or UPD-787 (Fig. 5A,B). P-values, determined by non-linear regression analysis of the data and curve fitting comparison, resulted to be < 0.0001 for both UPD-176 and UPD-787 (Fig. S9), indicating that the difference between tet-induced and non-induced cells was significant.

CDC25 inhibitors arrest growth of Apc/K-Ras intestinal organoids.
Having established the effect of CDC25 inhibitors on the growth of cells in monolayer cultures, we addressed the ability of selected compounds to affect the growth of organoid cultures. Since CDC25A cooperates with, and is rate-limiting in tumorigenesis induced by Ras 10 , we generated and cultured 3D-organoids from stem cells of the intestinal crypts of Apc/K-Ras mice 35 . The latter carry the loss-of-function mutation Apc1638N combined with the villinCre-driven gain-of-function K-RasG12V mutation 36 . Upon seeding Apc/K-Ras organoids along with UPD-176 (5 μM), we observed that growing spheroids partially invaginated (Fig. 6A), indicating reduction of stem cell renewal accompanied by increased differentiation 37 , whereas higher doses of the compound caused massive death (data not shown). Consistent with the pattern of invagination at low compound treatment, immunofluorescence confocal imaging revealed augmented expression of lysozyme, a marker for cells exiting the proliferative compartment of the crypt and acquiring a differentiated state (Fig. 6B). Quantitative RT-PCR performed on the stemness marker Lgr5 and on the differentiation markers Lysozyme and Cryptdin confirmed the confocal microscopy data (Fig. 6C).

CDC25 inhibitors cause tumor regression and block the formation of metastasis in an in vivo
model. Next, we evaluated the effects of UPD-176, UPD-140 and UPD-738 on growth and metastatic potential of HCT116 tumors in vivo using a zebrafish xenograft model 38 . The compounds were well tolerated by the developing zebrafish embryos at the effective dose (10 µM) and did not cause signs of overt toxicity. Treatment of tumor-bearing fish larvae with UPD-140 for three days (from day 2 to day 5) caused a dramatic regression of the tumors, which were reduced to approximately half the size of the initial volume (p < 0.00784, Fig. 7A,B). The concurrent reduction in metastasis achieved borderline significance (p < 0.0604, Fig. 7A,C). Treatment with UPD-176 or UPD-738 displayed a tendency towards tumor regression, although not significant in comparison to vehicle (DMSO)-treated embryos (p < 0.1830 and p < 0.0609, respectively). For these compounds, also the variation of metastatic burden upon treatment with either compound did not appear to be significant ( Fig. 7A-C). Taken together, these data demonstrate that inhibiting CDC25 may be an effective strategy to induce tumor regression and inhibit metastasis, in vivo.

Discussion
Precision oncology is centered on the principle of continuous molecular interrogation of tumors and on the use of dedicated pharmacological tools to maximize success in therapy. Constant monitoring is essentially intended to assess resistance 39 and reveal pathways to which tumors may become addicted during treatment 40 . CDC25 phosphatases are overexpressed in a variety of human cancers 9,41 , are rate-limiting in tumorigenesis induced by Ras 10 , and were recently proposed as target of choice in unresponsive, triple-negative breast cancer 22 . In this study, we set out to identify novel CDC25 inhibitors. Taking advantage of knowledge gained in previous drug discovery programs 13,42 , we initially defined a pharmacophore that is common to compounds belonging to three distinct classes of established CDC25 inhibitors (Fig. S1). Performing a ligand-based virtual screen of a proprietary library, we identified naphthoquinone inhibitors of CDC25 phosphatases and conducted molecular docking studies on the crystal structure of CDC25B (Fig. 1, Tables S1 and S2). The phenyl moiety present in our scaffold represents a novel feature with respect to the chemical structure of previously described naphthoquinone inhibitors of the CDC25 family 14 . The interactions established by the R groups of the phenyl ring with residues located in the two cavities of CDC25 (Table 1) likely allow efficient anchoring of the naphthoquinone scaffold ( Fig. 1), contributing to ameliorate potency in enzymatic and cellular assays. Biochemical studies on the identified compounds revealed a mixed mechanism of inhibition for all CDC25 members (Fig. S2), possibly indicating binding of the compounds to catalytic (Fig. 1C) and allosteric sites of the phosphatase, as also reported for other CDC25 inhibitors 18,19,27,43,44 . Mechanistically, we observed that compounds acted reversibly on CDC25 (Fig. 2C), likely through oxidation of the catalytic cysteine (Fig. 2B), a mode of action that we confirmed in cells (Fig. 3B). Such mechanism is compatible with a model proposed for CDC25 inhibitors 31 , according to which the thiolate group of the active-site cysteine undergoes very rapid conversion to sulfenic acid and it is protected from further conversion into irreversibly inactivated sulfinic acid by a back-door cysteine located in close proximity to the catalytic residue. Cellular experiments showed that the most potent compounds arrested cells at the G1/S or the G2/M transition (Fig. 3D,E). Considering that these compounds displayed similar kinetics of inhibition on the three purified CDC25 isoforms ( Fig. 2A), the reason for such preferential point of action in the cell cycle remains to be investigated. Interestingly, administration of compounds at the time of release from double-thymidine treatment showed minimal progression to S-phase at 6 hours of release (Fig. S5B). This may indicate that either turnover of the compound occurred during this time or, more likely, that the amount of CyclinE/CDK2 complex built up while cells accumulated at the point of forced arrest (i.e., early S-phase) was sufficient to support cell cycle progression despite inhibition of CDC25. The latter hypothesis is corroborated by the observation that when compounds were added in mid-G1, transition to S-phase was effectively blocked (Fig. 3D). Since quinones undergo reduction to form semiquinone radicals or hydroquinones that, in turn, can react with oxygen to form superoxide radicals and/or hydrogen peroxide 45 , hence affecting a number of metabolic processes in the cell, we examined DNA damage induction as a possible side-effect relevant to the overall cellular response observed. Specifically, we quantified the biomarker γ-H2AX in synchronized cells that were treated with compounds. The data excluded the possibility that the cell cycle arrest observed in response to compound treatment could be secondary to genotoxic effects (Fig. S6).
To examine the cellular mechanism of action for the compounds described in this study, we focused on activation of CDK1 and entry into mitosis. Analysis of the extent of Thr 14 /Tyr 15 phosphorylation, as indicator of kinase activation, and phosphorylation of CDK1 substrates, as indicator of kinase activity, revealed that the inhibitors effectively impaired both responses (Fig. 4B-E). Chromosome condensation normally occurring in prophase was also impaired in compound-treated cells, in line with flow cytometry data attesting a G2/M arrest under these conditions (Fig. 4B,F). Visual inspection of cells treated with the compounds showed that they could not execute mitosis, in line with our biochemical evidence, but rather underwent massive death (Fig. 4A). Administration of low concentration of the inhibitors to HeLa cells carrying mCherry-H2B allowed appreciating failed attempts to round up for mitosis, followed by membrane blebbing and death according to an apoptotic pattern (Fig. S7, Movies M2 and M3).
Enzymatic profiling of two of the most potent CDC25 inhibitors on a panel of phosphatases revealed PP5 as the only other target of the compounds (Table S4). PP5 controls a number of cellular processes including proliferation, migration and DNA damage. Interestingly however, PP5 activity is normally off due to folding of an N-terminal inhibitory domain onto the catalytic site, with ligand-mediated release of auto-inhibition occurring in response to cellular cues 46 . Hence, we argue that the specific G1/S and G2/M responses that we describe in this study are genuine effects of the identified compounds on CDC25 phosphatases.
Considering that CDC25A and CDC25B are overexpressed in a variety of human cancers 9 and CDC25A was demonstrated to be a rate-limiting oncogene in transformation by RAS 10 , we examined the response of a cell line overexpressing CDC25A, as paradigm for the demonstration of compounds potency. The data confirmed that the compounds could effectively decrease viability of CDC25 overexpressing cells (Fig. 5).
Finally, we conducted studies in 3D-organoids, a system that reproduces architecture and function of the tissue of origin in reduced scale 47 and in zebrafish as in vivo model of K-Ras-dependent tumors. Organoids obtained from stem cells of the intestinal crypts of Apc/K-Ras mice feature an internal cavity, corresponding to the crypt's lumen, but lack the symmetry characteristic of this organ 48 . Fluorescence microscopy revealed decreased organoids size and acquisition of a differentiated state in response to CDC25 inhibitors, a pattern confirmed by qRT-PCR on selected markers (Fig. 6). These data are reminiscent of observations made on Cdc25B -/-/Cdc25C -/knock-out mice where Cdc25A was conditionally disrupted (Cdc25A fl-) in all tissues of the adult organism 8 . In this triple knock-out background, the authors reported large loss of the small intestine and crypts atrophy due to arrest at G1 and G2 phases of the cell cycle, which was paralleled by an increase of epithelial cell differentiation. As a whole, the triple knock-out studies and our data support the concept that blocking cell cycle progression through inhibition of CDC25 activity is beneficial to target tumor growth driven by mutant Ras and dependent on CDC25.
In zebrafish transparent embryos implanted with fluorescently labeled HCT116 cells, tumor growth/regression and metastasis can be accurately followed in vivo, over time, at single-cell resolution 38 . In this system, the efficacy of potential drug candidates can be accurately determined 49,50 . We obtained evidence that compounds bearing a pharmacophore common to NSC-663284, which was shown to have minimal antitumor activity in mice due to rapid metabolism 20 , are well tolerated and show clear anti-tumor efficacy in zebrafish xenograft models. Interestingly, compounds displaying the most potent effect in biochemical assays, showed the highest efficacy in the zebrafish model (Fig. 7), suggesting that these, and UPD-140 in particular, may be promising candidates for further investigation.
As a whole, the data reported in this study reveal the potential of novel naphthoquinone scaffolds acting as CDC25 inhibitors to target Ras-dependent tumors.

Materials and Methods
Chemistry. Commercially available chemicals were purchased from Sigma-Aldrich (Milan, Italy). For work-up and chromatographic purification, commercial grade solvents were used. Semi-preparative and preparative purifications of the synthesized compounds were carried out on Isolera One automated flash chromatography system (Biotage, Upsala, Sweden). The analytical profile of the synthesized molecules was in accordance with literature data (see below). 1 H and 13 C[ 1 H] NMR spectra were recorded on a Bruker Avance III 400 MHz and a Bruker AMX 300 MHz spectrometers 51 . All spectra were recorded at room temperature. High-resolution mass spectra were recorded on an ESI-TOF Mariner from PerSeptive Biosystem (Stratford, Texas, USA), using electrospray ionization (ESI). Purity was assayed by HPLC, using a Varian Pro-Star system equipped with a Bio-Rad 1706 UV-VIS detector and an Agilent C-18 column (5 mm, 4.6 mm 150 mm). Water (A) and acetonitrile (B) were used as mobile phases with an overall flow rate of 1 mL/min and the following analytical method: 0 min (90% A-10% B), 15 min (10% A-90% B), 20 min (10% A-90% B), 21 min (90% A-10% B), 255 min (90% A-10% B). Purity was over 97% (HPLC area).

Generation of virtual library and hit selection.
An in silico virtual library was built from a proprietary database of synthetic molecules (2075 compounds) upon 2D to 3D conversion of the molecules chemical structure, optimization of compound conformers and addition of Gasteiger partial charges (MOE Suite). The strategy for hit selection is described in the Results section.
In silico hit expansion/optimization. The structure of the first two hits, UPD-140 and UPD-176, provided a clue for the hit expansion/optimization phase through a Site Finder approach (MOE Suite), followed by a Molecular Docking protocol (Glide, Schrödinger Suite 59 ). Site Finder allows the identification of pockets and surface sites by identifying regions of tight atomic packing using Alpha Shapes, a generalization of convex hulls developed in 60 that has been successfully validated against several unrelated targets 61,62 . The procedure suggested two close cavities that are highly conserved in the three enzymes, one of them corresponding to the phosphatase catalytic site. Based on the Site Finder indications, a molecular docking process was performed using the Glide package XP procedure 63 (Schrödinger Suite) and focusing on compounds featuring a quinone moiety.
Microscopy. Cells were grown in 35 mm CellView ™ cell culture dishes with glass bottom (# 627870, Greiner-BioOne) at a density of 2.5 × 10 5 cells/ml in a humidified cell incubator maintained at 37 °C and 5% CO 2 . Cell were viewed by phase contrast microscopy with a 10x objective using an Olympus IX 81 motorized inverted microscope (Olympus, Hamburg, Germany) equipped with external temperature control chamber and CO 2 bottle to maintain cells at 37 °C with 5% CO 2 . Transition through mitosis was documented by acquisition of four frames per hour over a period of 14 h using a CCD camera (Orca AG, Hamamatsu) and cellR ® software (Olympus). HeLa-Kyoto cells (mCherry-H2B/EGFP-α-tubulin) were visualized by fluorescence microscopy with an Olympus IX 81 microscope using a 10x objective and selecting Ex. = 492 ± 18 nm/Em. = 535 ± 50 nm for the green channel and Ex. = 572 ± 23 nm/Em. = 645 ± 75 nm for the red channel.
In vivo tumor model. Implantation of HCT116 cells into zebrafish embryos were carried out essentially as previously described 38 . Briefly, HCT116 cells were cultured in McCoy's medium supplemented with 10% FCS and penicillin/streptomycin until ~80% confluence, then labeled for 30 min at 37 °C in 6 μg/mL 1,1′-dioctadecyl-3,3, 3′,3′-tetramethylindocarbocyanine perchlorate (DiI, Sigma) in PBS followed by washing 3x in PBS. Labeled cells were implanted in the perivitelline space of 48 hours post-fertilization (hpf) wildtype (AB) zebrafish embryos, kept from the 1-cell stage in 1-phenyl-2-thiourea (PTU)-containing E3-water. Approximately 200-500 cells were implanted in each embryo, which were then transferred to E3 medium containing PTU as well as 10 μM UPD-176, UPD-140 or UPD-738 appropriately diluted from stock solutions made in DMSO. As control, embryos were incubated in PTU-E3 with 0.1% DMSO (vehicle). Tumor-bearing embryos were incubated at 36 °C for three days and the change in tumor volume was evaluated as the size of the tumors at three days post-implantation (3 dpi) relative to the size immediately after implantation (0 dpi). Metastasis was evaluated at 3 dpi as the number of tumor cells present in the caudal hematopoietic plexus, the main metastatic site for tumors implanted in the perivitelline space.