Identification of Chemical Inhibitors of β-Catenin-Driven Liver Tumorigenesis in Zebrafish

Hepatocellular carcinoma (HCC) is one of the most lethal human cancers. The search for targeted treatments has been hampered by the lack of relevant animal models for the genetically diverse subsets of HCC, including the 20-40% of HCCs that are defined by activating mutations in the gene encoding β-catenin. To address this chemotherapeutic challenge, we created and characterized transgenic zebrafish expressing hepatocyte-specific activated β-catenin. By 2 months post fertilization (mpf), 33% of transgenic zebrafish developed HCC in their livers, and 78% and 80% of transgenic zebrafish showed HCC at 6 and 12 mpf, respectively. As expected for a malignant process, transgenic zebrafish showed significantly decreased mean adult survival compared to non-transgenic control siblings. Using this novel transgenic model, we screened for druggable pathways that mediate β-catenin-induced liver growth and identified two c-Jun N-terminal kinase (JNK) inhibitors and two antidepressants (one tricyclic antidepressant, amitriptyline, and one selective serotonin reuptake inhibitor) that suppressed this phenotype. We further found that activated β-catenin was associated with JNK pathway hyperactivation in zebrafish and in human HCC. In zebrafish larvae, JNK inhibition decreased liver size specifically in the presence of activated β-catenin. The β-catenin-specific growth-inhibitory effect of targeting JNK was conserved in human liver cancer cells. Our other class of hits, antidepressants, has been used in patient treatment for decades, raising the exciting possibility that these drugs could potentially be repurposed for cancer treatment. In support of this proposal, we found that amitriptyline decreased tumor burden in a mouse HCC model. Our studies implicate JNK inhibitors and antidepressants as potential therapeutics for β-catenin-induced liver tumors.


Introduction
Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related death worldwide [1], in large part because of the paucity of effective systemic therapies. To date only one drug, the multikinase inhibitor sorafenib, has been shown to improve survival in patients with advanced HCC [2]. Genome-wide analyses of HCC have enabled the classification of these tumors into subsets and raise the possibility that the distinct signaling pathway alterations in given subgroups might be associated with different responsiveness to targeted molecular therapies [3]. A major subset of HCC is defined by activating mutations in the CTNNB1 gene encoding β-catenin [4,5], an integral component of the canonical Wnt signaling pathway [6]. Studies of human liver tumors suggest that β-catenin mutation may be an early or initiating event in some HCC [4,7,8].
Vertebrate models that recapitulate the specific signaling pathway perturbations that occur in HCC, such as the Wnt/β-catenin signaling pathway, could help further define mechanisms by which these pathways promote HCC. Such models could be highly useful in the identification and characterization of targeted systemic therapies unique to each tumor type. Loss-offunction mutations in APC, which result in increased β-catenin activity, increase hepatic neoplasia in both mice [9] and zebrafish [10]. Interestingly, although expression of activated β-catenin can induce some types of mouse tumors, including intestinal and prostatic neoplasms [6], and mutant β-catenin enhances murine liver tumorigenesis in response to carcinogen treatment [11] and H-Ras co-expression [12], activated β-catenin is not sufficient to induce HCC in mice [6,12,13]. Thus, the use of mouse models to study events in liver tumor formation that may be uniquely and specifically related to β-catenin mutation is limited. The zebrafish represents an excellent model system for studies of human cancer, given its powerful genetics and amenability to chemical screens [14] and the histologic and genetic similarities between zebrafish and human tumors [15]. At early developmental stages, zebrafish larvae are translucent, and internal organs such as the liver and heart can be readily observed in live animals using a dissecting microscope.
To investigate mechanisms by which activated β-catenin signaling promotes liver tumor formation and to identify potential therapeutics for these cancers, we generated transgenic zebrafish expressing hepatocyte-specific activated β-catenin (Tg(fabp10a:pt-β-cat) zebrafish). As adults, these animals show increased liver size, decreased survival, and histologic abnormalities similar to human HCC. The presence of a larval phenotype enabled us to use these zebrafish in American Cancer Society Postdoctoral Fellowship. JS is supported by the Agricultural Research Foundation of Oregon and the John Fryer Salmon Disease Laboratory of the Department of Microbiology at Oregon State University. AG is supported by a V-Foundation Scholar Award and NIH NCI R01-CA170447. This project was supported in part by the UCSF Liver Center, including a Pilot/ Feasibility Award (NIH P30DK026743) to Chunyue Yin and a Tools and Technology Grant to DYRS, and grants from the NIH (DK60322) and the Packard Foundation to DYRS. Flow cytometry was performed at the UCSF Parnassus Flow Cytometry Core, funded in part by Diabetes Research Center (DRC) grant, NIH P30 DK063720. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. a whole-organism chemical screen, in which we tested 960 drugs and identified eight compounds, including two c-Jun N-terminal kinase (JNK) inhibitors and two antidepressants (a tricyclic antidepressant (TCA) and a selective serotonin reuptake inhibitor (SSRI)), that suppressed β-catenin-induced liver enlargement. The β-catenin-specific growth-inhibitory effect of targeting JNK was conserved in human liver cancer cells. Furthermore, activated β-catenin was associated with JNK pathway hyperactivation in zebrafish and in human HCC. Focusing additional chemical screening on serotonergic ligands, we identified two other antidepressants (a TCA and a tetracyclic antidepressant) that suppressed β-catenin-induced liver enlargement. All of our antidepressant hits are FDA-approved, raising the exciting possibility that these drugs or related compounds could potentially be repurposed for cancer treatment. Supporting the hypothesis that these medications might have anti-tumor effects in mammals, we found that the TCA amitriptyline decreased tumor burden in a mouse HCC model. Together, our results show that Tg(fabp10a:pt-β-cat) zebrafish represent a useful tool for studies of β-catenindriven liver tumorigenesis, enabling insights into mechanism and potential therapeutics for liver cancers with β-catenin mutations, including β-catenin-activated human HCC.

Results
Zebrafish expressing activated β-catenin develop HCC As a first step to investigate mechanisms by which activated β-catenin promotes liver tumorigenesis, we generated zebrafish that express a mutated form of Xenopus laevis ctnnb1/β-catenin (pt-β-cat) [16] under the control of hepatocyte-specific regulatory sequences from the zebrafish gene fabp10a [17](S1A Fig). Xenopus β-catenin is highly similar to human and zebrafish β-catenin. Its in vivo activity has been extensively studied [16], and its transcript can be distinguished from endogenous zebrafish transcripts. pt-β-cat contains four point mutations affecting putative phosphorylation sites (S33A, S37A, T41A, and S45A); these mutations activate β-catenin by inhibiting its phosphorylation and subsequent degradation, and the affected codons are among those most commonly mutated in human HCC [18]. We engineered the transgene to also encode a fluorescent protein downstream of lens-specific regulatory sequences (cryaa: Venus), such that transgenic zebrafish can be identified by 3 days of age (days post fertilization, dpf) based on eye color [19,20]. We generated three independent lines of Tg(fabp10a:pt-β-cat, cryaa:Venus) zebrafish (hereafter used interchangeably and named Tg(fabp10a:pt-β-cat)), which showed similar phenotypes, supporting the hypothesis that the resulting changes are due to expression of pt-β-cat and not to disruption of another gene during transgenesis. Transgenic zebrafish exhibited Xenopus ctnnb1 mRNA expression in their livers (S1B Fig), while Xenopus ctnnb1 was undetectable in non-transgenic zebrafish (S1B Fig). Zebrafish ctnnb1 mRNA levels were not significantly different in transgenic versus non-transgenic livers (S1C Fig), suggesting that endogenous ctnnb1 transcription was not suppressed by the presence of the transgene. Further, Tg(fabp10a:pt-β-cat) zebrafish exhibit nuclear β-catenin localization and Wnt reporter activation [21](S1 Table) in their hepatocytes by 5-6 dpf (S1D-S1G Fig), indicating hepatocyte-specific expression of activated β-catenin.
By 4 to 5 months of age, Tg(fabp10a:pt-β-cat) zebrafish showed dramatically enlarged livers with increased vasculature and/or discoloration involving 1 or more lobes (Fig 1A and 1D). Microscopically, these transgenic zebrafish livers showed patchy involvement by neoplastic change, with areas of architectural disruption (Fig 1B, 1E, and S2A) and atypical cytological features including nuclear enlargement and nuclear contour irregularities (Fig 1C, 1F, and S2B- S2D Fig). Importantly, similar architectural and cytological abnormalities have been used to diagnose HCC in other transgenic zebrafish models [22] and represent defining features of well-to moderately-differentiated human HCC (S2 Table). For example, zebrafish livers with activated β-catenin often showed abnormal circular gland-like arrangements of hepatocytes ( Fig 1G) resembling the pseudoglands seen in many human HCCs. Like human liver tumors, zebrafish samples occasionally showed evidence of intracellular lipid accumulation, with fat vacuoles displacing hepatocyte nuclei to the edge of the cell (Fig 1H). Overall, the microscopic appearance of Tg(fabp10a:pt-β-cat) zebrafish livers was diverse, mimicking the variety of architectural and cytological features that is seen in human HCC. Thus, our data indicate that Tg (fabp10a:pt-β-cat) zebrafish recapitulate morphologic features of human HCC. To determine the incidence of these gross and microscopic alterations, we examined Tg (fabp10a:pt-β-cat) zebrafish and non-transgenic control siblings at 1 to 6 month intervals. We found that liver size relative to total body mass was significantly increased in Tg(fabp10a:pt-βcat) zebrafish compared to controls at multiple time points (Fig 1I). Tg(fabp10a:pt-β-cat) zebrafish had lower body mass at most time points (S3 Fig and S3 Table), although this difference was only statistically significant at 4 months post fertilization (p<0.01, 2-way ANOVA). Livers were analyzed for HCC by a board-certified pathologist with subspecialty fellowship training in liver pathology (K.J.E.) using histologic criteria based on architectural and cytological abnormalities (S2 Table) as described previously [22]. By 2 months post fertilization (mpf), 33% of Tg (fabp10a:pt-β-cat) zebrafish developed HCC in their livers, and this percentage increased to 78% and 80% at 6 and 12 mpf, respectively ( Fig 1J). Overall, while 25 out of 49 (51%) Tg (fabp10a:pt-β-cat) livers showed HCC, only 1 out of 37 (2.7%) control livers showed such features (Fig 1J; p<0.001, Fisher's exact test). Furthermore, the median adult survival of Tg (fabp10a:pt-β-cat) zebrafish was decreased to 200 days compared to greater than 393 days in non-transgenic control siblings (Fig 1K; p<0.001, logrank test). Together, these results show that activated β-catenin is sufficient to promote HCC in zebrafish.
We next sought to more deeply investigate possible similarities in gene expression between Tg(fabp10a:pt-β-cat) zebrafish livers and human liver cancer. We measured gene expression in Tg(fabp10a:pt-β-cat) zebrafish livers with HCC and non-transgenic control sibling livers by RNA-seq and performed combined clustering analysis of these zebrafish samples with 268 human HCC samples and 243 adjacent non-tumor human liver tissue (GSE25097) [30][31][32]; we based our clustering on 283 differentially-expressed genes in human HCC versus human nontumor liver (absolute log fold-change (logFC) greater than or equal to 2 and false-discovery rate (FDR) less than 0.05) (S1 Dataset). We found that 5 out of 7 (71%) Tg(fabp10a:pt-β-cat) zebrafish liver samples clustered with human HCC, and 4 out of 5 (80%) non-transgenic zebrafish liver samples clustered with non-tumor human liver samples (Fig 2A). The gene expression similarities between Tg(fabp10a:pt-β-cat) zebrafish liver tumors and human HCC are corroborated by a visualization of the first two principal components ( Fig 2B). Next, we identified 485 genes that were significantly dys-regulated in the same direction in both human HCC (GSE25097) and Tg(fabp10a:pt-β-cat) zebrafish livers, when compared to the corresponding non-tumor/non-transgenic controls, at an FDR less than 0.05 (Fig 2C, S1 Dataset). Pathway enrichment analyses of these genes highlighted several pathways that are relevant to cancer, including DNA replication, DNA repair, cell proliferation, telomeres, and cell motility and adhesion (S4 Table, S1 Dataset). Together, these results suggest that there are striking transcriptional similarities between Tg(fabp10a:pt-β-cat) zebrafish liver tumors and human HCC, and that these similarities may be important for liver tumorigenesis in both species.

Chemical screen for suppressors of β-catenin-induced liver enlargement
Zebrafish larvae are small, translucent, and readily absorb compounds from their culture water, facilitating their use in chemical screens [14]. To investigate the utility of Tg(fabp10a:ptβ-cat) larvae as a tool for chemical screening, we examined liver development in these animals. Already at 5 to 6 dpf, Tg(fabp10a:pt-β-cat) livers from all three transgenic lines were 40-70% larger than the livers of control siblings (Fig 3A and 3B). Given the well-established role of activated β-catenin in driving liver growth and proliferation [1,6], we hypothesized that the increase in liver size seen in zebrafish expressing activated β-catenin was due to increased cell proliferation. To test this hypothesis, we performed 5-ethynyl-2´-deoxyuridine (EdU) labeling and cell size measurements. We found that Tg(fabp10a:pt-β-cat) livers showed 130% more EdU labeling than control sibling livers, while hepatocyte size was not significantly altered ( Fig  3C-3E).
To identify druggable pathways that mediate β-catenin-mediated liver enlargement, we next established a protocol for small-molecule chemical screening using Tg(fabp10a:pt-β-cat) larval zebrafish (S6 Fig). For this chemical screen, we also employed Tg(fabp10a:EGFP) zebrafish [17], which express liver-specific green fluorescent protein by 3 dpf, thus facilitating rapid visual assessment of liver size in live animals. Three 3-dpf zebrafish were cultured in each well with drug or vehicle control for 3 days. At 6 dpf we scored liver size using a fluorescence dissecting microscope on a scale from 0 to 4. Concurrently, animals were examined for evidence of drug toxicity including pericardial edema or death. Drugs that substantially decreased average liver size of Tg(fabp10a:pt-β-cat); Tg(fabp10a:EGFP) zebrafish in 2 out of 2 wells (both drug-containing wells) without signs of toxicity were considered potential hit compounds. We then performed a secondary validation screen of these potential hit compounds, using similar parameters with more animals.
As many anti-cancer agents are protein kinase inhibitors, we next applied our screening platform to a commercially available kinase inhibitor library. We tested 160 compounds and identified five drugs that decreased liver size in Tg(fabp10a:pt-β-cat) zebrafish without causing obvious toxicity (S5 Table). We selected SP600125 and EMD 420123 for additional follow-up studies because they were the only 2 of the 5 hits among the kinase inhibitors with a shared mechanism of action. SP600125 is a well-established JNK inhibitor [33]. EMD 420123 also inhibits JNK, albeit with a substantially higher IC50 in vitro [33]. At doses greater than or equal to 10 μM, both drugs caused edema and/or death. All larvae (34/34 and 27/27) treated with 10 μM or 50 μM EMD 420123 died. Five out of 31 and 7/31 larvae treated with 10 μM SP600125 died or showed edema, respectively; 2/21 and 17/21 larvae treated with 50 μM  Zebrafish Chemical Screen and Liver Tumorigenesis SP600125 died or showed edema; and 2/10 and 8/10 larvae treated with 100 μM SP600125 died or showed edema. At lower doses, both SP600125 and EMD 420123 caused a dose-dependent decrease in liver size in Tg(fabp10a:pt-β-cat) larvae (Fig 4A-4C). Although EMD 420123 also decreased liver size in non-transgenic animals, the magnitude of this effect was less substantial (20-40% decrease at 2-5 μM) than the effect on zebrafish expressing activated β-catenin (30-60% decrease at 2-5 μM). Furthermore, at doses up to 5 μM, SP600125 did not affect liver size in non-transgenic animals, suggesting that the decrease in liver size we observed with this compound was related to specific effects in the context of activated β-catenin and not due to nonspecific toxicity.
Activated β-catenin is associated with JNK activity in larval and adult zebrafish Our result that two JNK inhibitors decreased liver size in zebrafish expressing activated β-catenin raised the possibility that activated β-catenin might cooperate with the JNK pathway to increase liver size. Evidence from several different tumor models indicates that JUN is an important phosphorylation target of JNK during carcinogenesis [34]. Thus, we performed whole-mount immunofluorescence staining of zebrafish larvae using phospho-c-Jun-specific antibodies. Tg(fabp10a:pt-β-cat) zebrafish showed a higher index of nuclear phospho-c-Jun staining than controls, supporting the hypothesis that activated β-catenin is associated with JNK pathway activation in the zebrafish larval liver (Fig 5A and 5B). Furthermore, both JNK inhibitors significantly decreased phospho-c-Jun staining in transgenic larvae, consistent with the hypothesis that these compounds decrease liver size in Tg(fabp10a:pt-β-cat) zebrafish larvae at least in part by inhibiting the JNK pathway (Fig 5A and 5B). JNK inhibitor treatment did not significantly affect Wnt reporter activity or Tg(fabp10a:EGFP) expression in Tg(fabp10a:pt-βcat) zebrafish larvae (S7 Fig), arguing that these drugs act downstream of or in parallel to βcatenin/TCF-driven transcription rather than by decreasing fabp10a promoter activity, altering transgene levels, and/or directly inhibiting β-catenin.
JUN autoregulates its own transcription via enhancer elements in its promoter [34,35] and forms a complex with β-catenin and TCF4 to activate JUN transcription in a JNK-dependent manner during mouse intestinal tumor formation [36]. Given our finding that activated β-catenin was associated with JNK pathway activation during zebrafish liver development, we hypothesized that a similar cooperation between activated β-catenin and the JNK pathway might occur during zebrafish liver tumor formation. To assess JNK pathway activity in adult zebrafish, we performed immunohistochemical staining using phospho-c-Jun-specific antibodies. We found that 6 out of 8 (75%) Tg(fabp10a:pt-β-cat) livers with HCC contained foci with moderate or strong nuclear phospho-c-Jun staining, while 0 out of 7 control livers without activated β-catenin showed such staining (Fig 5C; p<0.01, Fisher's exact test). Furthermore, we found that Tg(fabp10a:pt-β-cat) livers had more jun expression than control livers ( Fig 5D). Together, these results support the hypothesis that activated β-catenin is associated with JNK pathway activation in adult zebrafish livers as well.

JNK activity in human HCC
Our chemical screen results and follow-up experiments indicate that activated β-catenin is associated with JNK pathway activation in zebrafish larvae and adult livers. We next sought to explore the relevance of this association to human liver tumorigenesis. Diffuse strong glutamine synthetase staining [37] and higher GLUL expression [5] are associated with β-catenin mutations, and these criteria can be used to identify β-catenin-activated tumors [38]. Therefore, we classified β-catenin activation status in human tumors based on glutamine synthetase  staining pattern or GLUL expression levels. We found that 6 out of 15 (40%) of HCCs with activated β-catenin showed at least focal phospho-c-Jun staining ( Fig 5C). We observed phospho-c-Jun staining less often in HCC without β-catenin activation (2 out of 11 tumors; 18%), although this difference was not statistically significant. Next, we investigated JNK activity in a separate, larger set of HCC specimens. Examining expression profiles of 96 HCC samples [39], we determined that tumors in the highest tertile of GLUL expression had significantly higher JUN levels than tumors in the lowest tertile ( Fig 5E). These data suggest that JNK pathway activation is associated with activated β-catenin in human HCC.
Our finding in zebrafish larvae that JNK inhibition decreased liver size and increased survival specifically in the context of activated β-catenin raises the possibility that β-catenin activation might confer sensitivity to JNK inhibition in other systems, including human liver cancer. To test this hypothesis, we used a panel of human liver tumor cell lines with and without activating mutations in β-catenin [40,41], treating cells with the newer generation JNK inhibitor CC401 that is currently in clinical trials. The concentration of CC401 that inhibited cell growth by 50% (GI50) was significantly lower in cells with activating mutations in β-catenin than in cell lines without β-catenin mutations, indicating an increased sensitivity to this drug ( Fig 5F) (p < 0.05, unpaired t test).

Antidepressants suppress β-catenin-induced liver enlargement
Having established our zebrafish model as a useful tool for identifying drugs that suppress βcatenin-induced liver tumorigenesis, we next expanded our chemical screen to search for additional, novel compounds with anti-HCC effects. We tested 800 compounds from a commercially available library of pharmacologically active compounds, many of which are FDAapproved for treatment of non-cancer illnesses, reasoning that hits from such a screen could more easily be repurposed for HCC treatment. We identified two drugs, amitriptyline (a TCA) and paroxetine (an SSRI) that decreased liver size in Tg(fabp10a:pt-β-cat) zebrafish without causing obvious toxicity (S5 Table). Both antidepressants caused a dose-dependent decrease in liver size in Tg(fabp10a:pt-β-cat) larvae (Fig 6A and 6B). As we found with SP600125, amitriptyline treatment did not significantly affect Wnt reporter activity or Tg(fabp10a:EGFP) expression in Tg(fabp10a:pt-β-cat) zebrafish larvae (S7 Fig), arguing that this drug acts downstream of or in parallel to β-catenin/TCF-driven transcription rather than by decreasing fabp10a promoter activity, altering transgene levels, and/or directly inhibiting β-catenin.
To identify additional related compounds that suppress β-catenin-induced larval liver enlargement, we screened 80 compounds from a serotonergic ligands library. We found two additional drugs, mirtazapine (a tetracyclic antidepressant) and clomipramine (a TCA), that decreased liver size in Tg(fabp10a:pt-β-cat) zebrafish without causing obvious toxicity (S5 Table). The identification of four antidepressants, all of which inhibit serotonin reuptake, that suppress β-catenin-mediated liver growth raises the possibility that serotonin signaling or another shared target of these drugs may inhibit β-catenin-induced liver tumorigenesis.
Combining multiple drugs in cancer treatment represents an important tool for achieving maximal therapeutic benefit. To investigate the possibility that JNK inhibitors and serotonergic antidepressants could be used together to decrease β-catenin-mediated liver growth, we examined the effects of SP600125 plus amitriptyline on Tg(fabp10a:pt-β-cat) larval liver size. Treatment with high doses of both drugs (5 μM SP600125 plus 30 μM amitriptyline) resulted in death in 20 out of 20 larvae treated with drugs from 3 dpf to 6 dpf. However, treatment with slightly lower doses of both drugs (2 μM SP600125 plus 20 μM amitriptyline) resulted in a decrease in liver size that was significantly greater than the one we observed with either drug alone (p<0.001 versus 20 μM amitriptyline and p<0.01 versus 2 μM SP600125, 2-way ANOVA) (Fig 6C). This drug combination did not significantly alter liver size in non-transgenic animals (2-way ANOVA) (Fig 6C). Livers of Tg(fabp10a:pt-β-cat) zebrafish treated with the two-drug combination were similar in size to livers of non-transgenic siblings ( Fig 6C). These data suggest that amitriptyline and SP600125 decrease liver size in Tg(fabp10a:pt-β-cat) zebrafish via mechanisms that are at least partially distinct.
We next sought to test the hypothesis that amitriptyline, which caused the most robust decrease in liver size in Tg(fabp10a:pt-β-cat) zebrafish among the antidepressants we tested, had an anti-tumor effect in human liver cancer cells. To test this hypothesis, we treated a panel of human liver tumor cell lines with and without activating mutations in β-catenin [40,41], as well as immortalized non-transformed human hepatocytes, with amitriptyline. Interestingly, every human liver cancer cell line had a GI50 that was at least slightly lower than the nontumor HepTert cell line, indicating an increased sensitivity to this drug (S8 Fig). While on average cell lines with activating mutations in β-catenin had a lower GI50 (17.7 μM) than cell lines without β-catenin mutations (21.7 μM), this difference was not statistically significant, raising the possibility that other genetic or epigenetic differences besides β-catenin mutation might contribute to amitriptyline sensitivity in liver cancer cells.
In contrast to our other class of hits, the JNK inhibitors, antidepressants are FDA approved and are widely used for treatment of a variety of common, non-neoplastic conditions including major depressive disorder, obsessive-compulsive disorder, and anxiety disorders. In addition, these compounds are highly soluble in water and easy to administer. We reasoned that together these attributes could facilitate translation of our amitriptyline studies to other vertebrates, including mammals. Although β-catenin alone is not sufficient to induce HCC in mice, hydrodynamic transfection of MET and activated β-catenin together (HDT-Met/β-Cat) induces multi-focal HCC in mice [42], resulting in innumerable (greater than 100) coalescing tumors that replace the majority of the hepatic parenchyma ( Fig 6D, top panels, and S9A- S9B Fig top  panels). Therefore, we next sought to test the hypothesis that amitriptyline could decrease tumorigenesis in HDT-Met/β-Cat mice. We treated these mice continuously with 200 mg/L amitriptyline in their drinking water, to achieve plasma levels of drug that improve behavior in mouse models of depression [43,44] and are similar to the therapeutic levels in humans treated for depression [43].
Strikingly, we found that amitriptyline treatment caused a significant decrease in the liver tumor burden of HDT-Met/β-Cat mice, decreasing mean liver-to-body mass ratio from 0.29 to 0.24 (p = 0.0288) (Fig 6D and 6E). Amitriptyline treatment decreased mean liver mass from 6.8 g to 4.9 g (p = 0.0012) and decreased mean body mass from 23.0 g to 20.1 g (p = 0.0023) in HDT-Met/β-Cat mice (S10 Fig). In mice that had not been hydrodynamically transfected, amitriptyline treatment did not significantly change liver-to-body mass ratio (mean = 0.061 in vehicle-and amitriptyline-treated mice; p = 0.8819) or liver mass (mean = 1.7 g in vehicletreated mice; mean = 1.6 g in amitriptyline-treated mice; p = 0.6724), although in these mice amitriptyline caused a slight decrease in body mass (mean = 27.3 g in vehicle-treated mice; mean = 26.0 g in amitriptyline-treated mice) that was not statistically significant (p = 0.0749) (Figs 6D, 6E and S10). Upon histologic examination of HDT-Met/β-cat livers, we noted that amitriptyline treatment did not substantially alter tumor morphology (S9A- S9B Fig bottom  panels). However, amitriptyline treatment significantly decreased tumor cell proliferation (Figs 6F, 6G, and S11), reducing the amount of Ki-67 labeling from 30 to 16 Ki-67-positive cells per high-power field (HPF) (p = 0.0201). At the same time, amitriptyline-treated tumors demonstrated slightly increased cell death, with 2.9 TUNEL-positive cells per HPF compared to 1.8 TUNEL-positive cells per HPF in control tumors, although this difference was not statistically significant (p = 0.0907) (Figs 6F, 6H, and S11). Together, these findings indicate that amitriptyline, in addition to decreasing β-catenin-induced liver enlargement in zebrafish, inhibits liver tumorigenesis in a mammalian model.

Discussion
In this study, we report a new transgenic zebrafish model with hepatocyte-specific activation of β-catenin. We found that activation of β-catenin promotes liver overgrowth in zebrafish, consistent with mouse models where its activation leads to increased hepatocyte proliferation and liver growth [13]. While β-catenin activation alone does not give rise to HCC in mouse models [6,13], our study reveals that it is sufficient to promote HCC in adult zebrafish. Considering that β-catenin activation may be an early or initiating event in some human HCC [4,7,8], our new zebrafish transgenic line represents a unique and useful tool to study mechanisms of βcatenin-mediated liver tumorigenesis and identify potential therapeutics for these tumors.
Using these transgenic zebrafish, we performed a small-molecule screen and identified several kinase inhibitors that decreased liver size in larvae expressing hepatocyte-specific activated β-catenin. We focused our follow-up studies on two JNK inhibitors, and found that expression of activated β-catenin was associated with JNK pathway hyperactivation in larval and adult livers. Prompted by our observations in zebrafish, we examined human tumors for evidence of JNK pathway activity and noted increased JUN expression predominantly in HCCs with activated β-catenin, demonstrating the link between the two in human disease. Although previous studies had reported a role for JNK in liver tumorigenesis [45], a link between JNK activation and a specific HCC subtype had not been previously established. Our studies raise the possibility that JNK activity may be of particular importance in HCC with activated β-catenin.
Phospho-c-Jun staining in HCC is associated with poor response to sorafenib [46], the only drug currently approved for treatment of advanced HCC [1], underscoring the importance of identifying more effective therapies for HCCs with JNK activation. Our results connecting activated β-catenin and the JNK pathway raise the possibility that sorafenib resistance may be linked to activating mutations in β-catenin in human HCC. We propose that inhibiting the JNK pathway may represent an especially useful therapeutic strategy for this sorafenib-resistant HCC subgroup. SP600125 suppresses liver tumor formation in N-diethylnitrosamine (DEN) treated rats [47], indicating its anti-tumor potency in mammals. However, as the tumors in that study were induced by a chemical carcinogen and not by a specific genetic manipulation, the relevance of these findings to any of the genetically distinct human HCC subtypes was not clear. Now, our findings provide the first evidence that SP600125 or related compounds might be effective in treating human HCC with β-catenin mutations.
Repurposing existing FDA-approved drugs for cancer treatment has several advantages, including decreased drug development time. The need for new systemic therapies is particularly urgent for HCC, given its increasing incidence and its high mortality rate when localized treatment fails. In the second round of our chemical screen, we tested compounds with known mechanisms of action, including many FDA-approved drugs, and identified two antidepressants, a TCA and an SSRI, that suppressed β-catenin-induced liver growth in our zebrafish model. We also found that one of these drugs, amitriptyline, decreased tumor burden in a mouse HCC model. Our finding that these antidepressants decreased liver growth in these tumor models was somewhat surprising, because both TCAs and SSRIs are thought to act at least in part by increasing serotonin levels in the brain via inhibition of the serotonin reuptake transporter (SERT), and serotonin promotes liver growth during regeneration [48]. However, TCAs and SSRIs have other biologically important roles besides SERT inhibition, including direct and/or indirect effects on serotonin receptors (5-HT2A and 5-HT2B) [49,50]. In addition to blocking both serotonin and norepinephrine reuptake, TCAs bind to multiple G-protein-coupled cell surface receptors, mediating induction of apoptosis in lung cancer cells [51]. Furthermore, it is possible that in the models we used, TCAs and SSRIs are exerting at least part of their anti-tumor effects via non-tumor cell types, such as hepatic stellate cells, which store, release, and respond to serotonin [52]. Here we show that amitriptyline inhibited β-catenininduced liver growth and/or tumorigenesis using two distinct, in vivo, vertebrate systems. Further experiments are required to determine which cell types and molecular targets mediate the anti-tumor effects of amitriptyline in these models. Although the myriad targets of antidepressants complicate mechanistic investigations, the pleiotropic effects of these drugs may render them particularly suitable for repositioning as cancer therapeutics. Our findings raise the exciting possibility that amitriptyline or related antidepressants could be effective in treating or preventing a major subset of human HCC.
Although we found that liver cancer cell lines with β-catenin mutations had a slightly lower GI50 in response to amitriptyline than cell lines without β-catenin mutations, this difference was not statistically significant, and we noted that 5 out of 6 liver cancer cell lines had a significantly lower GI50 in response to amitriptyline than immortalized hepatocytes. These results raise the possibility that amitriptyline might be generally toxic to cancer cells, as opposed to having an effect that is specific to the presence of activated β-catenin. In keeping with this possibility, previous reports have shown that amitriptyline inhibits multiple cancer types in vitro, including liver cancer [53], multiple myeloma [54], melanoma [55], small cell lung cancer [51], glioma [56], and neuroblastoma [56]. Amitriptyline has direct actions on tumor cells, such as inducing mitochondrial damage, increasing reactive oxygen species, and inhibiting the antioxidant machinery of human cervical, lung, and liver cancer cells [57]. We discovered that amitriptyline caused a significant decrease in liver tumor burden in mice hydrodynamically transfected with Met and β-catenin, but we did not test the effects of amitriptyline on mouse liver tumors lacking activated β-catenin. Future studies examining the effects of amitriptyline in other mouse models could be helpful to determine the specificity or generalizability of this drug's anti-tumor effects.
Previous chemical screens for inhibitors of Wnt/β-catenin signaling have primarily employed cell-based assays [58,59]. These approaches are relatively fast, enabling testing of very large numbers of compounds. While we were able to test only a modest number of compounds in our chemical screen, the whole organism approach facilitates examination of organ-specific phenotypes and enables the identification of compounds that may more indirectly affect Wnt/ β-catenin signaling, for example via downstream targets, parallel cooperating pathways, or effects on non-hepatocyte cell types. Furthermore, our approach enabled us to concomitantly assess toxicity in live vertebrate animals.
During our study, we focused on mechanistic characterization of two compounds identified in our chemical screen because these drugs shared a single well-described mechanism of action, JNK inhibition. Our subsequent discovery that JNK activity is relevant to both the larval and adult phenotypes of our model as well as to human HCC validates this chemical screen as a productive approach to identify and explore additional mechanisms translatable to human βcatenin-mediated liver tumorigenesis.
We amplified pt-β-catenin by polymerase chain reaction (PCR) from the pCS2-XE49 plasmid [16], a kind gift from Ben Cheyette, adding NotI and ClaI restriction sites at the 5' and 3' ends, respectively. We placed the pt-β-catenin insert between the~2.8 kB fabp10a upstream region [17] and polyadenylation sequences into a vector containing, in reverse orientation, the crystallin alpha A eye specific promoter (cryaa) followed by coding sequence for Venus fluorescent protein [19,20]. The resulting plasmid also contained I-SceI restriction enzyme sites, facilitating generation of transgenic zebrafish as previously described [62]. One-cell stage embryos were microinjected with plasmid and I-SceI enzyme (New England Biolabs, Ipswich, MA) using a Leica dissecting microscope (Allendale, NJ). Injected embryos with yellow-green eyes at 2-3 dpf were raised to adulthood. By crossing these injected animals to wild-type zebrafish, we identified three founders representing independent transgenic lines 3-3 (s985), 4-1 (s986), and 4-5 (s987). Transgenic descendants of these founders were distinguished from control siblings based on cryaa:Venus expression at 3 dpf or later.

Analysis of Tg(fabp10a:pt-β-cat) adult liver morphology
For analysis of gross and microscopic changes in adult zebrafish over time, we raised and maintained 5 clutches of Tg(fabp10a:pt-β-cat) s987 zebrafish and parallel tanks of non-transgenic control siblings. At each time point, zebrafish were randomly selected from the Tg(fabp10a:pt-βcat) tank and the control tank containing zebrafish from the same clutch or clutches. Zebrafish were euthanized at the following time points by rapid chilling and weighed: 2 months (6 control zebrafish, 6 transgenic zebrafish); 3 months (8 control zebrafish, 8 transgenic zebrafish); 4 months (10 control zebrafish, 15 transgenic zebrafish); 5 months (8 control zebrafish, 8 transgenic zebrafish); 6 months (9 control zebrafish, 9 transgenic zebrafish); 12 months (10 control zebrafish, 10 transgenic zebrafish). Livers were dissected using a Leica dissecting microscope, rinsed in phosphate buffered saline (PBS), and weighed, and the entire liver or representative portions were fixed in 10% formalin for at least 24 hours. Remaining liver portions were used for RNA preparation and/or cryosections as described below. Paraffin embedding, sectioning, and hematoxylin and eosin (H&E) staining were performed at the Gladstone Histology and Light Microscopy Core Facility. Slides were blinded and histologic changes were scored by a pathologist (K.J.E.). A second pathologist (J.S.) reviewed a subset of the slides to confirm the diagnoses. Each specimen was assigned to one of the following categories: (1) Hepatocellular carcinoma (HCC), defined as the presence of architectural abnormalities, such as pseudoglands, sheets of cells, and decreased or absent bile ducts, involving greater than 5% of the tissue, plus cytological abnormalities, including nuclear enlargement, nuclear contour irregularities, and mitotic figures; (2) Minimal/mild changes, defined as the presence of cytological abnormalities in the absence of substantial architectural disruption or architectural abnormalities without substantial cytological atypia; or (3) No changes, defined as no substantial cytological or architectural abnormalities. Lesions assigned to the first category (HCC) fulfilled diagnostic criteria established by a board-certified pathologist with subspecialty fellowship training in hepatopathology (K.J.E.) (S2 Table) as well as by other pathologists and veterinary pathologists [22]. Quantification of liver-to-body mass ratios and microscopic changes over time were performed in the same experiment. In some cases, only a subset of zebrafish were examined microscopically, so N values are lower for microscopic analysis of the following groups: transgenic zebrafish at 4 months (N = 8); control zebrafish at 3, 4, and 5 months (N = 4, each time point).
Representative photographs of livers situated within the body cavities of 4-month-old Tg (fabp10a:pt-β-cat) s986 zebrafish and control non-transgenic siblings were taken with a Zeiss Stemi SV 11 microscope and Zeiss MRc5 camera (Carl Zeiss, Oberkochen, Germany) at lowest magnification. Representative photographs of adult zebrafish and human liver sections were taken with an Olympus BX51 microscope using the 40X objective and an Olympus DP72 camera (Olympus, Tokyo, Japan).

Flow cytometric analysis
Two separate independent flow cytometry experiments were performed, which showed similar results. For experiment #1, we harvested cells from 4-month-old Tg(fabp10a:pt-β-cat) s987 zebrafish and control siblings by dissecting out livers, dissociating cells in 0.25% trypsin, and filtering cells through a 40 µm filter. The concentration of live cells in each sample was determined using a Countess Automated Cell Counter (Life Technologies, Carlsbad, CA, USA), in order to prepare 500 µL cell suspensions with equal numbers of live cells. Cell were fixed overnight in 95% ethanol at -20ºC and then stained with propidium iodide in the presence of RNase A for 1 hour at room temperature. Flow cytometry was performed using a BD LSRII with FACSDiva Software (BD, Franklin Lakes, NJ); 5,500 to 30,000 events were collected for each sample. For experiment #2, we harvested, fixed, and stained cells from 4-month-old Tg(fabp10a:pt-β-cat) s985 zebrafish, 18-month-old Tg(fabp10a:pt-β-cat) s987 zebrafish, and 18-month-old control siblings using the same protocol as for experiment #1. Flow cytometry was performed using a BD Fortessa with FACSDiva Software (BD); 10,000 events were collected for each sample. We then spiked each liver sample with a few drops of fixed, propidium-iodide-stained sperm obtained from non-transgenic adult zebrafish in order to provide a 1N reference; 10,000 additional events were collected for each sample. We visually confirmed that plots for samples with and without sperm appeared similar except for the presence of a 1N peak.
Raw data collected from both experiments were gated sequentially as follows: first, dead and dying cells were excluded by plotting side-scatter area (SSC-A)(y-axis) versus forward-scatter area (FSC-A)(x-axis) and gating out cells on the far left; second, doublets were excluded by plotting forward-scatter width (FSC-W) versus FSC-A and gating out cells with high FSC-W; third, remaining doublets were excluded by plotting side-scatter width (SSC-W) versus SSC-A and gating out cells with high SSC-W. Histograms showing live, doublet-free cells were generated using FACSDiva Software.

Microarray analysis
We isolated RNA from 4-month-old Tg(fabp10a:pt-β-cat) s986 zebrafish livers with HCC and control sibling livers. Preparation of cDNA and subsequent microarray hybridizations using Agilent zebrafish (V3) microarrays were performed at MoGene (St. Louis, MO). Two biological replicates were performed, and the fold-change for each probe was determined by dividing the processed Cy3 (transgenic) signal by the processed Cy5 (control) signal. The average foldchanges of probes with significant signals above background were inputted into Ingenuity Pathways Analysis (IPA) (Ingenuity Systems, www.ingenuity.com, release date 2012-11-01).
Zebrafish RNA-seq analysis, human microarray analysis, and crossspecies comparison, hierarchical clustering, and pathway enrichment analyses We isolated RNA from 4-to 6-month-old Tg(fabp10a:pt-β-cat) s986 and Tg(fabp10a:pt-β-cat) s987 zebrafish livers with HCC and control sibling livers. Bioanalyzer trace and RNA-seq library preparation were performed at the Gladstone Genomics Core Laboratory. The UCSF Genomics Core Facility performed single-read, 50-base-pair sequencing using Illumina HiSeq technology; 8 pooled barcoded libraries were run in one lane of the flow cell. Raw FASTQ files, corresponding to each sample, were aligned to Release 79 of the zebrafish genome made available by the Ensembl project, using STAR v2.4 [63]. Overlaps of the resulting alignments with protein-coding genes were then quantified using the GenomicAlignments R package [64]. Differential gene expression analysis between Tg(fabp10a:pt-β-cat) animals and their wild-type siblings was performed using the DESeq2 R package [65]. Regularized log-transformed values of the raw counts were used for all subsequent cross-species analyses.
Gene expression profiles for 268 human HCC tumors and 243 adjacent non-tumor liver tissue, measured using a Rosetta/Merch RSTA Affymetrix microarray, were sourced from the Gene Expression Omnibus (GEO-Accession: GSE25097) [30][31][32]. Probes corresponding to multiple Entrez IDs were collapsed using the 'MaxMean' method in the WGCNA R package [66][67][68]. Differential gene expression analysis between HCC samples and the adjacent nontumor liver tissue was then performed on log2-transformed values using the limma R package [69].
To enable cross-species comparison, zebrafish genes were matched to their human orthologs based on annotations available from the Ensembl database, using the biomaRt R package [70,71]. One-to-many relationships were resolved by choosing the orthologous pairs with the largest confidence of orthology and amino acid sequence identity with respect to the human protein, as indicated in the Ensembl annotations. A total of 13710 orthologous pairs with oneto-one relationships between zebrafish and human genes were obtained as such.
For cross-species clustering analysis, human and zebrafish gene expression values were first individually median-centered and then scaled by dividing each value by the median absolute deviation for the corresponding gene. The transformed datasets were then merged and subjected to principal component analysis (PCA) and hierarchical clustering using a Pearson correlation distance metric with average linkage, based on a subset of 283 orthologous gene pairs that are significantly dys-regulated in human HCC samples when compared to adjacent nontumor liver samples (absolute log fold-change ! 2, false discovery rate (FDR) < 0.05). PCA and hierarchical clustering were performed using the FactoMineR [72] and cluster [73] R packages respectively.
To probe for conserved genetic pathways, we focused on genes that were significantly dysregulated in the same direction in both human HCC samples and Tg(fabp10a:pt-β-cat) animals when compared to their non-tumor counterparts (FDR < 0.05). Expression profiles of these genes are illustrated in heatmaps for human and zebrafish samples respectively, using the gplots [74] and cluster [73] R packages. We then performed pathway enrichment analysis on this gene list. Overrepresented Gene Ontology (GO) terms [75] and pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG) [76] and Reactome [77] databases were identified using a p-value cutoff of 0.05 as determined by the Fisher's Exact Test. . In the initial screen, for each compound we tested at least 2 wells containing Tg(fabp10a:pt-β-cat); Tg (fabp10a:EGFP) larvae and at least one well containing Tg(fabp10a:EGFP) control siblings. Three larvae were placed into each well. Larvae were sorted under a Leica MZ16F fluorescence dissecting microscope based on cryaa:Venus and fabp10a:EGFP expression at 3 dpf, pipetted into wells, cultured at 28°C, and examined 3 days later at 6 dpf using the same microscope. We scored liver size qualitatively on a scale from 0 to 4, with 1 indicating typical liver size in 6 dpf Tg(fabp10a:EGFP) larvae without activated β-catenin in the absence of drug, 0.5 indicating smaller-than-typical liver size, and 2-4 indicating increased liver size. For each plate, we included at least 4 control wells containing Tg(fabp10a:pt-β-cat); Tg(fabp10a:EGFP) larvae in DMSO, and experimental compounds were only considered potential hits if the average liver size score in these control wells was greater than or equal to 2. Drugs that decreased average liver size of Tg(fabp10a:pt-β-cat); Tg(fabp10a:EGFP) zebrafish in both wells (average liver size score less than 2) without causing toxicity/death in any wells were considered potential hit compounds.

Chemical screen and follow-up
The secondary validation screen was performed using a similar protocol, except for every drug we tested at least three wells, each containing three larvae. Drugs that resulted in an average liver size score of less than or equal to 2 in Tg(fabp10a:pt-β-cat); Tg(fabp10a:EGFP) zebrafish without causing toxicity/death in control Tg(fabp10a:EGFP) siblings were considered confirmed hits. Again, compounds were only considered hits if the average liver size score in the corresponding Tg(fabp10a:pt-β-cat); Tg(fabp10a:EGFP) DMSO control well(s) was greater than or equal to 2.
Additional follow-up experiments were performed with SP600125 (Sigma-Aldrich), EMD 420123 (EMD Millipore/Calbiochem), amitriptyline (Sigma-Aldrich) and paroxetine (Sigma-Aldrich). We dissolved these compounds in DMSO and stored them as concentrated solutions at -20°C or -80°C. Larvae were cultured in egg water (40 g/L Instant Ocean) with drugs at a final concentration of 0.5-100 μM from 3 dpf until euthanasia and fixation; controls were cultured with the same DMSO concentration (0.2-1%).

Analysis of Tg(fabp10a:pt-β-cat) survival
To generate adult survival curves, we raised and maintained 4 clutches of Tg(fabp10a:pt-βcat) s986 zebrafish and parallel tanks of non-transgenic control siblings. Beginning at adulthood (2-4 months post fertilization), zebrafish were counted and/or checked 2-3 times per week and the number of dead and missing zebrafish was recorded. Zebrafish with evidence of distress including altered swimming or feeding behavior or severe abdominal distention were euthanized and counted as dead. The experiment was terminated when all zebrafish were greater than 11 months old. Survival curves were generated and analyzed using MedCalc statistical software.

Larval liver size measurements, immunofluorescence, and EdU labeling
In general, experiments involving zebrafish larvae were planned to result in a final sample size of approximately 7-10 zebrafish. With respect to randomization, animals were sorted at 3 dpf (before increased liver size of transgenic animals is obvious) and assigned to experimental groups immediately after sorting. Animals were excluded from analysis if they died during drug treatment (10 μM and 50 μM EMD 420123) or if livers became detached or poorly oriented during fixing, staining, or mounting. Exclusion of samples was established before quantification of results.
Larvae were euthanized with tricaine methanesulfonate (0.03%), fixed in 4% paraformaldehyde (PFA) in PBS for at least 12 hours at 4°C and rinsed in PBS. Skin and any residual yolk surrounding the liver were removed using forceps under a Leica dissecting microscope. Brightfield photographs of larvae embedded in 3% methylcellulose were taken using a Zeiss Stemi SV 11 microscope and Zeiss MRc5 camera at highest magnification. We quantified larval liver size in transgenic zebrafish (Tg1 = s985; Tg2 = s986; Tg3 = s987) and control siblings by outlining livers on these photographs and measuring the area using FIJI/ImageJ. For dose-response experiments, photographs were blinded before measuring liver size. Dose response experiments for each drug and drug combination were done at least twice on separate clutches of zebrafish, and showed similar trends; results shown are pooled.
For studies of hepatocyte proliferation, larvae were incubated in 10 μM 5-ethynyl-2'-deoxyuridine (EdU) dissolved in embryo medium for 3 hours immediately before euthanasia. EdU was visualized using the Click-iT EdU Imaging Kit (Life Technologies), with the Click-iT reaction performed before incubation with primary antibodies.
Samples were imaged on a Zeiss LSM5 Pascal confocal microscope (Carl Zeiss) or a Nikon Eclipse 90i confocal microscope (Nikon Inc., Melville, NY).

Quantification of hepatocyte proliferation, cell size, Wnt reporter activity, and phospho-c-Jun immunofluorescence staining in zebrafish larvae
To quantify hepatocyte proliferation and cell size, we used Tg(fabp10a:pt-β-cat) s987 ; Tg (fabp10a:rasGFP) larvae and Tg(fabp10a:rasGFP) control siblings. Larvae were fixed at 6 dpf, and proliferating hepatocytes and cell membranes were visualized by EdU labeling and anti-GFP immunofluorescence staining, respectively. Larvae were blinded and imaged using the same pinhole, laser power, slice number (50, beginning at ventral surface of liver), slice interval (1 μM), and scan speed for each larva, but gain was adjusted to obtain the clearest image. Images were blinded again before quantifying EdU positivity, which was determined by counting the number of EdU positive hepatocytes and the total number of hepatocytes in slice 8 using ImageJ. To quantify average cell size, we used the same set of blinded images. For each larva, twenty representative cells in slice 2 were outlined and measured using ImageJ.
The EdU and cell size experiments were done twice, and only representative results (experiment 2) are shown in Fig 2. We observed an increase in EdU labeling in transgenic zebrafish in both experiments, but this difference was not statistically significant in experiment 1. Only 4 zebrafish were examined for each group in experiment 1. Cell size was increased slightly in transgenic zebrafish versus non-transgenic zebrafish in experiment 1 but the difference was not statistically significant (N = 8 (Non-Tg) and N = 7 (Tg)). For cell size, quantification was done slightly differently in experiment 1; we measured 100 cells in slice 1 to calculate the average cell size for each larva.
To quantify Wnt reporter activity, we used Tg(fabp10a:pt-β-cat) s986 ; Tg(7xTCF-Xla.Siam: GFP) zebrafish or Tg(fabp10a:pt-β-cat) s987 ; Tg(7xTCF-Xla.Siam:GFP) zebrafish and Tg (7xTCF-Xla.Siam:GFP) control siblings. Larvae were fixed at 6 dpf, and nuclei were visualized by anti-PHH3 immunofluorescence staining or by incubating larvae for 30 minutes at room temperature in TO-PRO-3 stain (Life Technologies; 1:70 dilution) before rinsing and mounting. Larvae were blinded and imaged using the same parameters. Wnt reporter activity was quantified for each zebrafish by selecting a representative area of the liver using the anti-PHH3/TO-PRO-3/far red channel and quantifying the mean signal of the same area in the GFP/green channel using ImageJ. The Wnt reporter activity experiment was done at least twice for each drug dose with similar results; representative results from one experiment are shown.
To quantify fabp10a promoter activity, we fixed Tg(fabp10a:pt-β-cat) s987 ; Tg(fabp10a: EGFP) zebrafish and Tg(fabp10a:EGFP) control siblings at 6 dpf. Larvae were blinded and imaged using the same parameters. We quantified fabp10a promoter activity for each zebrafish by quantifying the mean signal in a representative area of the liver in the GFP/green channel using ImageJ. The fabp10a promoter activity experiment was done at least four times for each drug dose with similar results; representative results from one experiment are shown.
To quantify phospho-c-Jun staining in zebrafish larvae, we used Tg(fabp10a:pt-β-cat) s987 zebrafish and sibling controls, treated with 5 µM SP600125, 5 μM EMD 420123 or 0.5% DMSO from 3-5 dpf. Larvae were fixed at 5 dpf and stained with phospho-c-Jun-specific antibody. Nuclei were visualized by incubating larvae for 30 minutes at room temperature in TO-PRO-3 stain (Life Technologies; 1:70 dilution) before rinsing and mounting. Larvae were imaged using the same parameters, the image files were blinded, and phospho-c-Jun staining for each larva was quantified by counting the total number of hepatocyte nuclei and the number of phosphoc-Jun-positive hepatocyte nuclei in the first slice of each image using ImageJ/FIJI. Phospho-jun staining of transgenic and non-transgenic larvae was done at least twice with qualitatively similar results, although only quantified in this manner once.

Immunofluorescence and immunohistochemistry of adult liver sections
For immunofluorescence of adult zebrafish livers, representative samples of livers from 6--12-month-old Tg(fabp10a:pt-β-cat) s986 zebrafish and control siblings were fixed in 4% PFA in PBS, incubated in 30% sucrose plus 0.02% sodium azide in PBS for at least 1 hour, and embedded in OCT compound. Cryosections were cut, air-dried, and stained with primary and secondary antibodies as described above, except Triton X-100 was omitted from rinsing and staining solutions and TO-PRO-3 stain was used at 1:500 dilution. Samples were imaged on a Zeiss LSM5 Pascal confocal microscope. Immunofluorescence experiments were done at least twice with qualitatively similar results.
For immunohistochemical analysis, we selected 3-6-month-old Tg(fabp10a:pt-β-cat) adult zebrafish livers that showed HCC on H&E-stained sections. As controls, we selected 3--6-month-old non-transgenic zebrafish that did not show substantial cytological or architectural abnormalities on H&E-stained sections. Immunohistochemistry was performed on formalin-fixed, paraffin-embedded tissue sections using phospho-c-Jun (Ser73) (D47G9) XP rabbit antibody (Cell Signaling Technology; Catalog Number 3270; 1:200 dilution) at the Gladstone Histology and Light Microscopy Core Facility following standard protocols including 0.3% H 2 O 2 blocking, citrate steaming antigen retrieval, and counterstaining with Mayer's hematoxylin. Slides were blinded, and the amount of nuclear staining was scored by a pathologist (K.J.E.) on a scale from 0-4+ based on the most intensely stained area of the slide: 0, no nuclear staining or rare scattered hepatocytes with nuclear staining; 1+, occasional scattered hepatocytes with nuclear staining; 2+, nuclear staining in 10-50% of hepatocytes; 3+, nuclear staining in greater than 50% of hepatocytes. For statistical analysis, 0 and 1+ staining patterns were considered phospho-c-Jun negative, and 2+ (moderate) and 3+ (strong) staining patterns were considered phospho-c-Jun positive.

Immunohistochemistry of human HCC samples
Human HCC specimens were selected from the pathology files of the University of California, San Francisco, and had been previously evaluated for nuclear β-catenin staining and glutamine synthetase expression by immunohistochemistry for clinical or research [81] purposes. Cases were defined as β-catenin active if the tumor showed diffuse strong glutamine synthetase staining (greater than 90% of tumor cells positive) or diffuse intermediate glutamine synthetase staining (50 to 90% of tumor cells positive). Cases were defined as β-catenin inactive if the tumor showed glutamine synthetase staining in less than 50% of tumor cells. Immunohistochemistry for phospho-c-Jun was performed on formalin-fixed, paraffin-embedded tissue sections using phospho-c-Jun (Ser73) (D47G9) XP rabbit antibody (Cell Signaling Technology; Catalog Number 3270; 1:200 dilution) at the Gladstone Histology and Light Microscopy Core Facility as described above. Slides were scored in a blinded fashion by K.J.E. and confirmed by a second pathologist (S.K.); cases were considered phospho-c-Jun positive if definitive nuclear staining was present in any tumor cells.

Analysis of JUN expression in human HCC samples
We used Microsoft Excel to sort genome-wide expression profiles of 96 human HCC samples [39] based on GLUL expression levels. We defined "GLUL high" tumors as the 32 tumors with the highest GLUL expression, and "GLUL low" tumors as the 32 tumors with the lowest GLUL expression. We normalized JUN expression levels for each of tumor sample by subtracting the average JUN expression level of all 96 tumors.
To determine the dose at which CC401 or amitriptyline inhibited growth by 50% (GI50), for each experiment cells were treated with 4-8 serial dilutions of drug with 6 replicates per concentration. CellTiter-Glo (Promega) Luminescent Cell Viability Assay was performed in 96-well plates, using a Safire2 plate reader (TECAN) and Magellan software. GI50 for each experiment was determined using GraphPad Software, and at least three experiments were performed for each drug for each cell line. For Western blot analysis, cultured cells were washed with PBS and collected into radioimmunoprecipitation assay (RIPA) buffer containing protease and phosphatase inhibitors as described [82]. Protein concentrations were determined using DC Protein Assay (Bio-Rad Laboratories, Hercules, CA) with bovine serum albumin standards; equal loading was confirmed by β-actin and Ponceau S staining of each membrane.

Mouse hydrodynamic transfection and drug treatment
Hydrodynamic transfection (HDT) was performed as described previously [42]. Female FVB/N mice, obtained from Taconic Biosciences (Hudson, NY) or Jackson Laboratory (Bar Harbor, ME), were injected at 6 to 8 weeks of age with 100 μL per gram of 0.9% normal saline containing plasmids encoding Sleeping Beauty transpose (800 ng/mL), human MET (10 μg/mL), and ΔN90-CTNNB1 (10 μg/mL) [42]. One month post-HDT, at 10 to 12 weeks of age, the animals were randomized into control or experimental groups and their drinking water was replaced with sterile water containing 200 mg/L amitriptyline plus 0.5-2% saccharine (experimental) or sterile water containing 0.5-2% saccharine alone (control). Mice were sacrificed by regulated administration of carbon dioxide followed by dissection of a major organ (liver) approximately 13 weeks post-HDT, when palpable tumors were present in control animals, except for two control mice and one amitriptyline-treated mouse, which were found dead at 13 weeks post-HDT. Within each experiment, the same concentration of saccharine was used for all mice and mice were sacrificed at the same time point. Four independent experiments were pooled. Three control mice and five amitriptyline-treated mice were excluded from analysis because of very low tumor burden, indicating unsuccessful HDT. One amitriptyline-treated mouse was excluded from analysis because it was found dead.
As non-HDT control animals, 11-week-old female FVM/N mice were obtained from Taconic Biosciences and given sterile water containing 200 mg/L amitriptyline plus 0.5% saccharine or sterile water containing 0.5% saccharine alone. After 12 weeks, mice were sacrificed by regulated administration of carbon dioxide followed by dissection of a major organ (liver).

Immunohistochemistry of murine samples
Mouse livers were fixed in 4% paraformaldehyde in PBS. Paraffin-embedded blocks and hematoxylin-and-eosin-stained slides were prepared at the Gladstone Histology and Light Microscopy Core Facility. All HDT animals included in the liver and body weight analyses were analyzed by immunohistochemistry, except for three animals (two control mice and one amitriptyline-treated mouse), which were found dead at 13 weeks post-HDT. TUNEL assay was performed using the ApopTag Peroxidase In Situ Apoptosis Detection Kit (EMD Millipore/ Calbiochem) following the manufacturer's protocol. Immunohistochemistry for Ki-67 was performed using pre-diluted Ki-67 antibody (Thermo Scientific; Catalog Number RM-9106-R7) following deparaffinization, rehydration, quenching of endogenous peroxidase activity by hydrogen peroxide incubation, and blocking in 1.5% normal goat serum. Goat anti-rabbit IgG-Biotin secondary antibody was used (Santa Cruz; Catalog Number sc-2040). Samples were detected with VECTASTAIN Elite ABC Reagent (Vector Labs) and Vector DAB substrate kit (Vector Labs) and counterstained with hematoxylin. To quantify proliferation and cell death, slides were blinded and a pathologist (K.J.E.) counted the number of Ki-67-or TUNEL-positive cells in 10-30 high-power fields for each mouse. Photographs of mouse liver sections were taken using an Olympus BX53 microscope with UPlanSAPpo objectives, Olympus DP72 camera, and Olympus cellSens Standard software.

Statistical methods
Student's two-tailed t-tests were performed using Microsoft Excel. Fisher's exact tests, Mann-Whitney tests, and 2-way ANOVA were performed using GraphPad Prism 6 Software. Survival curves were generated and analyzed using MedCalc statistical software and GraphPad Software. Comparisons were considered statistically significant if p-values were less than 0.05. Standard error of the mean is reported for each group of data. In general, sample size and variance were similar between all groups being statistically compared.

Ethics statement
Animal studies were carried out following guidelines and standard procedures set forth by the University of California at San Francisco Institutional Animal Care and Use Committee (UCSF IACUC). The UCSF IACUC approved all protocols involving live vertebrate animals (AN106527, AN090555). Zebrafish were euthanized by tricaine or rapid chilling, following UCSF IACUC guidelines. Mice were euthanized by carbon dioxide inhalation/administration followed by cervical dislocation and/or dissection of a major organ (liver), following UCSF IACUC guidelines. Human studies were approved by the institutional review board of UCSF, the Human Research Protection Program Committee on Human Research (IRB #10-04369). The requirement for informed consent was waived because the research involved at most minimal risk to the subjects, the waiver would not adversely affect therights and/or welfare of the subjects, the research could not practically be carried out without the waiver, and the subjects would be provided with additional pertinent information after participation, if appropriate. for compounds that suppress larval liver enlargement caused by activated β-catenin. Transgenic zebrafish expressing activated β-catenin (Tg(fabp10a:pt-βcat, cryaa:Venus)) were crossed to zebrafish expressing liver-specific GFP (Tg(fabp10a:EGFP). At 3 days old, larvae with green livers (fabp10a:EGFP+) were selected for drug treatment. Zebrafish with activated β-catenin, identified by their fluorescent eyes, were treated in parallel to control zebrafish without activated β-catenin. Three zebrafish were placed in each well, and 4 wells were tested for each experimental compound and vehicle (DMSO) control. Liver size was scored qualitatively 3 days post treatment (dpt). Drugs that decreased average liver size of Tg (fabp10a:pt-β-cat); Tg(fabp10a:EGFP) zebrafish in both wells compared to DMSO controls without causing toxicity/death in any wells were considered potential hit compounds. (TIFF) S7 Fig. SP600125 and amitriptyline do not significantly affect fabp10a promoter activity or Wnt reporter activity. (A) Representative photographs of fabp10a:EGFP expression (the same fabp10a promoter element used to drive pt-β-cat expression) in control (top row) and Tg (fabp10a:pt-β-cat) (bottom row) zebrafish livers at 5 dpf. (B) Fluorescence intensity ± standard error of the mean (SEM) was quantified using ImageJ, and samples were compared using 2-way ANOVA; p>0.05 for each group compared to every other group. N values are shown above the x-axis. (C) Representative photographs of 7xTCF-Xla.Siam:GFP (Wnt reporter) expression in control (top row) and Tg(fabp10a:pt-β-cat) (bottom row) zebrafish livers at 6 dpf. (D) Fluorescence intensity ± standard error of the mean (SEM) was quantified using ImageJ, and samples were compared using 2-way ANOVA (p>0.05 for all drug treatments compared to DMSO control of same genotype.) Scale bars, 20 μm. Six zebrafish were analyzed for each group.   Table. Body mass, liver mass, and liver-to-body mass ratios of Tg(fabp10a:pt-β-cat) zebrafish and sibling controls.