A Novel Zebrafish Model of Metastasis Identifies the HSD11β1 Inhibitor Adrenosterone as a Suppressor of Epithelial–Mesenchymal Transition and Metastatic Dissemination

This article is featured in Highlights of This Issue, p. 341

Overt metastases, the end result of malignant alteration of cancer cells, are responsible for approximately 90% of cancer-associated mortality. Metastasis consists of multiple processes: invasion, intravasation, survival in the circulatory system, extravasation, colonization, and tumor formation in secondary organs with angiogenesis (1, 2). Metastatic dissemination of cancer cells was traditionally viewed as a late stage event of cancer progression (3). However, this paradigm has been challenged by recent studies, in which mammary epithelial cells disseminated systemically from early neoplastic lesions of Her2 and PyMT transgenic mice and from ductal carcinoma in situ in patients with breast cancer (4, 5). Molecular mechanisms that promote dissemination involve the breakdown of local basement membrane, loss of cell polarity, loss of cell–cell adhesions, and induction of epithelial-to-mesenchymal transition (EMT; ref. 6).

EMT plays a central role in early embryonic morphogenesis, its program enables various types of epithelial cells to convert into mesenchymal cells through a downregulation of epithelial markers such as E-cadherin, and an upregulation of mesenchymal markers such as vimentin. Experimental studies demonstrated that EMT also contributes to metastatic progression by conferring invasiveness, motility, and an increased resistance to chemotherapy and apoptosis on cancer cells. EMT is initiated and orchestrated by multiple signaling pathways and transcriptional factor networks: TGFβ, BMP, and Wnt-mediated signaling; and Snail, Slug, and Twist-mediated transcriptional networks (7–9).

Twist, a basic helix-loop-helix transcription factor, plays a critical role in inducing the EMT program. It was first identified in Drosophila melanogaster as an organizer of EMT during fly gastrulation and a regulator of mesoderm differentiation (10). However, in the last decade, experimental studies have demonstrated an additional role of Twist—ectopic expression of Twist confers metastatic properties on cancer cells through induction of EMT. Moreover, clinical studies have revealed elevated expression of Twist is associated with poor survival rates in patients with cancer (8, 11, 12).

Cancer research using zebrafish as a model has attracted attention because this model offers many unique advantages that are not readily provided by other animal models (13). Moreover, the zebrafish has become a popular platform for drug screening to discover anticancer drugs (14); however, a zebrafish model that develops spontaneous metastases has not previously been reported.

Here we report a tamoxifen-controllable Twist1a-ER^T2 transgenic zebrafish as a new animal model for metastasis research, and show that this model serves as a novel platform for discovery of antimetastasis drugs targeting metastatic dissemination of cancer cells. Through in vivo drug screening using this model, we found that adrenosterone has suppressor effects on cell dissemination in both this model and human cancer cells. Finally, we demonstrated that pharmacologic and genetic inhibition of hydroxysteroid (11-beta) dehydrogenase 1 (HSD11β1), a primary target of adrenosterone, suppressed metastatic dissemination of highly metastatic human cancer cells in a zebrafish xenotransplantation model.

Transgenic zebrafish lines Tg(fabp10a:mCherry-T2A-Twist1a-ER^T2) and Tg(fabp10a:mCherry-T2A-ER^T2) were generated through Ac/Ds transposon system (15). Five to 10 pg of either fabp10a:mCherry-T2A-Twist1a-ER^T2 or fabp10a:mCherry-T2A-ER^T2 plasmid was coinjected into wild-type zebrafish embryos with 25–50 pg of in vitro synthesized AC transposase mRNA at the 1–2 cell stage. Tg(fabp10a:TA; TRE:xmrk; krt4:GFP) transgenic zebrafish line known as xmrk was described previously (16). Tg(kdrl:eGFP) zebrafish was provided by Dr. Stainier (Max Planck Institute for Heart and Lung Research). The study protocol was approved by the Institutional Animal Care and Use Committee of the National University of Singapore (protocol number: 096/12).

DNA fragments coding for mCherry-T2A, ER^T2, and zebrafish Twist1a were amplified by PCR with primers containing restriction enzyme recognition sequences. mCherry-T2A–coding fragment was amplified from hsp70l:mCherry-T2A-CreER^T2 plasmid (17). ER^T2 coding DNA fragment was amplified from pCAG-Cre-ER^T2 plasmid (18). The fragments of zebrafish Twist1a was amplified from cDNA that was prepared from wild-type zebrafish embryos at sphere stage. The amplified fragments were cloned into the pDS (19).

Total RNA was extracted from various adult organs in Twist1a-ER^T2 transgenic zebrafish and also from the livers of wild-type and ER^T2 transgenic zebrafish. cDNA was synthesized from the total RNA as described previously (20). qPCR using SYBR-Green Master PCR Mix (Applied Biosystems) were conducted in triplicates. All quantitation was normalized to an endogenous control gapdh. Primer sequences are presented in Supplementary Table S1.

Immunofluorescence microscopy assay was performed as described previously (21). Goat anti-mouse and goat anti-rabbit IgG antibodies conjugated to Alexa Fluor 488 (Life Technologies) and diluted at 1:100 were used. Nuclei were visualized by the addition of 2 μg/mL of 4′, 6-diamidino-2-phenylindole and photographed at 100 × magnification by a fluorescence microscope (Zeiss).

Living fry zebrafish were anesthetized using 0.02% phenoxyethanol and were then embedded with 30% methylcellulose in a lateral view orientation. Serial sections were captured in 8 μm Z-step intervals by a Leica TCS SP5X confocal microscope system (Leica). Z-stack images were processed by using image analysis software, Imaris (Bitplane).

FDA, European Medicines Agency, and other agencies approved chemical libraries were purchased from Prestwick Chemical. Ki16425 and Y27632 were purchased from Cayman Chemical and R&D Systems, respectively. Doxycycline, 4-hydroxytamoxifen (4-OHT), adrenosterone, and rabeprazole were purchased from Sigma-Aldrich. Olmesartan was purchased from Cayman Chemical.

Larvae of Twist1a-ER^T2/xmrk double transgenic zebrafish at 8 days postfertilization (dpf) were treated with 30 μg/mL of doxycycline in E3 medium for 3 days to induce xmrk expression and then aliquoted approximately 20 larvae into each well of a 6-well plate with 8 mL of E3 medium containing doxycycline. Drugs from the Prestwick Chemical Library were added to each well of the plates at a final concentration of 5 μmol/L. Twelve hours after drug addition, 4-OHT was added to each well of the plates at a final concentration of 0.1 μmol/L to induce Twist1a-ER^T2 biological activity. Five days after drug addition, the larvae were investigated under a fluorescence microscope and the pattern of cell dissemination was quantified. The number of the larvae showing each dissemination pattern of mCherry-labeled cells from the liver were counted.

Western blotting was performed as described previously (20). Anti-E-cadherin and anti-β-actin antibodies were purchased from BD Biosciences and Sigma, respectively. Anti-GAPDH, anti-Snail, anti-Slug, and anti-Histon H3 antibodies were purchased from Cell Signaling Technology. Anti-HSD11β1, anti-H⁺/K⁺-ATPase beta, anti-Vimentin, and Twist1 antibodies were purchased from Abcam. Anti-PCNA, anti-α-tubulin, anti-Occludin, anti-Claudin6, anti-KRT14, anti-KRT19, anti-MMP1, anti-MMP2, and anti-S100A4 antibodies were purchased from Santa Cruz Biotechnology.

MCF7, MDA-MB-231, MDA-MB-435, MIA-PaCa2, PC3, and SW620 cells were acquired from ATCC. HCCLM3 cells were provided by Dr. Tang, Zhong Shan Hospital of Fudan University (Shanghai, China; ref. 22). All culture methods followed the manufacturer's instruction. Cell viability assay was performed as described previously (23).

Boyden chamber assay was performed as described previously (23). Either 3 × 10⁵ MDA-MB-231, 5 × 10⁵ HCCLM3, or 1 × 10⁶ MDA-MB-435 cells were applied to each top well.

The short hairpin RNA (shRNA)-expressing lentivirus vectors were constructed using pLVX-shRNA1 vector (Clontech). HSD11β1-shRNA_#1–targeting sequence is GCTCCAAGGAAAGAAAGTGAT; HSD11β1-shRNA_#2–targeting sequence is CGAGCTATAATATGGACAGAT. ATP4b-shRNA_#1–targeting sequence is TAACCTTAAGGCCGGATGTTT; ATP4b-shRNA_#2–targeting sequence TGAAACACGGCTTACACTAAT. LacZ-shRNA–targeting sequence is CTACACAAATCAGCGATT.

Zebrafish embryos that were derived from Tg (kgrl:eGFP) transgenic zebrafish line were maintained in E3 medium containing 200 μmol/L 1-phenyl-2-thiourea. Approximately 100–400 of red fluorescence protein (RFP)-labeled HCCLM3 or MDA-MB-231 cells were injected into the duct of Cuvier of the zebrafish at 2 dpf.

Data were analyzed by Student t test and P < 0.05 was considered significant.

Twist1a was selected as the metastasis-inducible transgene in the zebrafish model based on the following two criteria: (i) loss of function of Twist1a interfered with metastatic potential of highly metastatic cancer cells in in vivo experiments using mouse metastasis models, and (ii) in clinical studies, elevated expression of Twist1a was observed in metastasized cancer cells in patients suffering from cancer (11, 12). Biological activity of Twist1a is controlled through a tamoxifen-inducible system by fusion of the gene with the estrogen receptor (ER^T2), and the transgene is expressed as a single open reading frame coding for mCherry and Twist1a-ER^T2 separated by a viral T2A peptide sequence under the control of a liver-specific promoter (fabp10a; Fig. 1A).

We microinjected wild-type zebrafish embryos at the single-cell stage of development with a plasmid of either fabp10a:mCherry-T2A-Twist1a-ER^T2 or fabp10a:mCherry-T2A-ER^T2 as a control, generating F₀ founder fish mosaic for expression of these transgenes. Germline transmission with Mendelian ratios for single insertion was confirmed in the F₁ generation. Tg(fabp10a:mCherry-T2A-Twist1a-ER^T2) and Tg(fabp10a:mCherry-T2A-ER^T2) zebrafish are shorted as Twist1a-ER^T2 and ER^T2, respectively, in the following description. To demonstrate liver-specific expression of the transgene, we isolated RNA from the livers and other organs from adult Twist1a-ER^T2 and ER^T2 fish at the F₂ generation and confirmed liver-specific expression of the Twist1a-ER^T2 transgene (Fig. 1B; Supplementary Fig. S1A). We then examined whether Twist1a-ER^T2 activation through tamoxifen (4-OHT) treatment could induce EMT in the liver of the fish through Western blot analysis. It is well known that loss of E-cadherin expression is the most predominant hallmark of EMT, and that Twist causes transcriptional repression of E-cadherin through downregulating the promoter activity of E-cadherin (7, 12). Both 0.1 and 0.5 μmol/L 4-OHT treatments for 48 hours decreased E-cadherin expression in the liver of the fish. This decreased expression was not observed in the liver of ER^T2 fish (Fig. 1C). Moreover, IHC staining revealed that Twist1a-ER^T2 activation disrupted cell–cell adhesion of hepatic cells with a decrease of E-cadherin and an increase of vimentin in the presence of oncogene-mediated signals that were induced by crossing Twist1a-ER^T2 fish with xmrk transgenic zebrafish. In contrast, a morphologic change was not observed in the liver of ER^T2/xmrk double transgenic fish (Fig. 1D; Supplementary Fig. S1B).

These results indicated that Twist1a-ER^T2 activation through 4-OHT treatment for 48 hours enabled the conversion of epithelial cells into mesenchymal cells in the liver.

We examined whether Twist1a-ER^T2 activation could induce dissemination of hepatic cells. Twist1a-ER^T2 fish were treated with 0.1μmol/L 4-OHT from 8 dpf for more than 3 weeks, but none of them (n ≥ 50) showed dissemination from the liver compartment (data not shown). Akin to the fish, it has been reported that in Twist-inducible mice, long-term induction of Twist alone did not give rise to any skin abnormalities, and that only primary tumor cells (papillomas) that were developed through 12-O-tetra decanoylphorbol-13-acetate treatment migrated out of the skin and disseminated throughout the body (24). Therefore, we crossed the fish with xmrk transgenic zebrafish, which develop hepatocellular carcinoma through expression of a constitutively active form of an EGFR homologue in a doxycycline-dependent manner (16). Biological activities of the proteins coded by two transgenes were sequentially induced in an independent manner: first xmrk transcription was initiated through doxycycline treatment from 8 dpf, and then Twist1a-ER^T2 was activated through 4-OHT treatment from 11dpf in presence of doxycycline (Fig. 2A). As shown in Fig. 1D, loss of cell–cell adhesion of hepatic cells with decreased E-cadherin was observed in the liver of Twist1a-ER^T2/xmrk fish 48 hours after 4-OHT treatment; conversely, the abnormalities were not observed in the liver of ER^T2/xmrk fish. Fluorescence and confocal microscopy analysis revealed that Twist1a-ER^T2/xmrk fish showed dissemination of mCherry-positive cells from the liver at day 5 from 4-OHT treatment (Fig. 2B). The dissemination patterns were generally divided into three categories: (i) local dissemination, in which disseminated mCherry-positive cells exist in close proximity to the liver; (ii) abdominal dissemination, in which the cells spread throughout the abdomen; and (iii) distant dissemination, in which the cells were observed over a broad region from the trunk to the tail (Fig. 2C).

Three independent experiments showed that 39.48% ± 18.05% and 45.7% ± 5.60% of Twist1a-ER^T2/xmrk fish showed abdominal and distant dissemination, respectively. In addition, 13.3% ± 5.61% of the fish showed both patterns of dissemination and were redundantly counted in both cases in this analysis. Also, 4.98% ± 1.45% of the fish showed local dissemination and 22.50% ± 4.37% of the fish did not show any cell dissemination. In contrast, only 0.96% ± 0.92% of ER^T2/xmrk fish showed abdominal dissemination and none of the fish showed distant dissemination. In addition, only 6.84% ± 0.44% of the fish showed local dissemination and 92.19% ± 0.86% of the fish did not show any cell dissemination. Neither ER^T2 nor Twist1a-ER^T2 fish showed cell dissemination. Statistical analysis using Student t test revealed that the frequency of Twist1a-ER^T2/xmrk fish showing abdominal and distant disseminations significantly increased over that of ER^T2/xmrk fish (P = 0.02, P < 0.01). In contrast, a significant difference was not observed between the frequencies of Twist1a-ER^T2/xmrk and ER^T2/xmrk fish showing local dissemination (Fig. 2D; Supplementary Table S2).

These results suggested that in this model, Twist1a-ER^T2–driven EMT alone would not be sufficient to induce abdominal and distant cell dissemination and that cooperation of oncogene-driven cellular events would be required for the dissemination. The statistical analysis revealed that these dissemination events were predominantly observed in Twist1a-ER^T2/xmrk fish, but not ER^T2/xmrk fish. This indicates that these dissemination events resulted from Twist1a-ER^T2–driven EMT, but are not affected by xmrk-induced events. Thus, measurement of dissemination to the regions defined in this model might provide a novel way to measure metastatic potential.

We hypothesized that the rapid and high frequency induction of dissemination of mCherry-positive cells from the liver in the Twist1a-ER^T2/xmrk model might provide a novel way to screen chemicals/drugs in vivo for identification of antimetastasis drugs targeting metastatic dissemination of cancer cells. Therefore, we conducted preliminary experiments that investigated whether ki16425 (a LPA1 inhibitor) or Y27632 (an inhibitor of Rho-associated, coiled-coil containing protein kinase), which have been reported to suppress metastasis progression in metastatic mouse models (25, 26), could suppress cell dissemination in the fish model. The basic experimental process followed the experimental design in Fig. 2A except that fish were treated with doxycycline and 4-OHT for 5 days in presence of either vehicle (DMSO), ki16425, or Y27632.

Two independent experiments revealed that the frequency of the fish showing either abdominal or distant cell dissemination in the ki16425-treated group significantly decreased to 20.93% ± 2.93% or 4.71% ± 6.67% when compared with those in the control (vehicle treated) group; 53.10% ± 7.0145% or 45.04% ± 4.39% for abdominal or distant cell dissemination. Similar to the ki16425-treated group, the Y27632-treated group also significantly decreased the frequency of abdominal and distant cell dissemination to 9.15% ± 5.50% and 6.52% ± 9.22%, respectively. In contrast, the frequency of the fish showing local dissemination in ki16425- or Y27632-treated group, slightly increased to 8.99% ± 3.30% or 11.36% ± 3.78%, respectively, when the frequencies were compared with that in vehicle-treated group (3.46% ± 0.34%). Moreover, the frequency of fish that did not show any cell dissemination in the ki16425- or the Y27632-treated group, significantly increased to 68.01% ± 2.81% and 76.21% ± 6.34%, respectively, when the frequencies were compared with that in vehicle-treated group (27.96% ± 7.60%; Fig. 3A and B; Supplementary Table S3).

These results indicated that the reported antimetastasis drugs (ki16425 and Y27632) could suppress abdominal and distant dissemination, and that the cells that failed to disseminate to these regions were stuck either in a region close to the liver or within the liver compartment by the effect of ki16425 or Y27632. These results suggested that indeed, this model provided a valid and novel approach to screen chemicals/drugs in vivo for identification of antimetastatic potential.

Next, we subjected 67 FDA-approved drugs to this screening. A protocol for the screening followed the above experiment except approximately 20 Twist1a-ER^T2/xmrk double transgenic zebrafish were placed in each well of a 6-well plate and were treated with vehicle or each of the drugs for 5 days in presence of doxycycline and 4-OHT. On the basis of the three categories that are indicated in Fig. 2C, the fish in vehicle (DMSO) or drug-treated groups were subdivided, and then the effects of the drugs were evaluated through comparing the frequencies of the fish showing the abdominal and distant dissemination patterns with those in the vehicle-treated group.

The screening assay showed that 63 drugs did not affect the frequency of the fish showing cell dissemination, and one drug had a lethal effect on the fish. Only three drugs; adrenosterone, rabeprazole, and olmesartan affected the frequency of the fish showing cell dissemination, and in each of these drugs the treated group was lower than that in the vehicle-treated group.

To obtain more precise data, we conducted large scale experiments with each of these three drugs to determine whether they could suppress cell dissemination in this model.

Two independent experiments revealed the frequencies of the fish showing abdominal and distant dissemination in the adrenosterone-treated group significantly decreased to 2.94% ± 4.16% and 0% when compared with those in vehicle-treated group; 53.33% ± 2.11% and 46.41% ± 2.78% for abdominal and distant dissemination, respectively. Similar effects were observed in rabeprazole- and olmesartan-treated groups: dissemination in the rabeprazole-treated group significantly decreased to 4.74% ± 2.42% or 0%; in the olmesartan-treated group the frequencies decreased to 27.79% ± 12.55% or 6.35% ± 5.16%. Conversely, the frequency of the fish that did not show any cell dissemination in adrenosterone-, rabeprazole-, or olmesartan-treated group, significantly increased to 94.92% ± 1.15%, 85.97% ± 1.59%, or 64.10% ± 1.08% when the frequencies were compared with that in the vehicle-treated group (27.77% ± 2.70%). The frequency of the fish showing local dissemination in rabeprazole-treated group significantly increased to 9.38% ± 0.41% but that in olmesartan-treated groups only slightly increased to 6.79% ± 9.55% when the frequencies were compared with that in the vehicle-treated group. In the adrenosterone-treated group the frequency was 2.13% ± 3.01% (Fig. 4A and B; Supplementary Table S4).

To eliminate the possibility that the suppressor effects of these drugs might result from inhibition of xmrk-driven primary tumor growth of the liver in the fish, we investigated whether each of these drugs might affect growth. Each of the drug-treated fish showed enlarged livers that were similar to those of vehicle-treated fish. Imaging analyses revealed that the size of the liver and the frequencies of proliferating cell nuclear antigen and cleaved caspase 3–positive cells in the livers of these fish were same. Moreover, survival analysis showed each of the drug-treated fish survived as long as vehicle-treated fish (Supplementary Fig. S2).

These results demonstrate that this model offers a new high-throughput platform for identifying antimetastasis drugs, and suggests that these drugs might have a potential to suppress metastatic dissemination of cancer cells without affecting primary tumor growth.

Zebrafish have orthologues to 86% of 1,318 human drug targets as reported previously (27). However, it is unclear whether the identified drugs through this zebrafish-based screening would show the same suppressor effects on metastatic dissemination of human cancer cells. We, therefore, examined whether the identified drugs could suppress metastatic dissemination and invasion of highly metastatic human cell lines. Adrenosterone, rabeprazole, and olmesartan are reported to block HSD11β1, hydrogen/potassium-transporting ATPase (H⁺/K⁺-ATPase) and angiotensin 2 type 1 receptor (AT1), respectively (28–30). AT1-mediated signaling has been demonstrated to promote metastasis progression through in vivo and vitro studies (31). However, the involvement of HSD11β1 and H⁺/K⁺-ATPase in metastasis progression is not known. We investigated HSD11β1 and H⁺/K⁺-ATPase beta expression in human cell lines possessing different metastatic properties. Western blot analysis revealed that highly metastatic cell lines (MDA-MB-231, MDA-MB-435, MIA-PaCa2, PC3, SW620, and HCCLM3) expressed a higher amount of HSD11β1 when compared with cell lines with low metastatic potential (MCF7). H⁺/K⁺-ATPase beta expression was observed in all of the investigated cell lines except PC3, but the expression level did not correlate with metastatic properties (Fig. 5A). Although these drugs did not affect cell viability of HCCLM3, MDA-MB-231, and MDA-MB-435 cells, these drugs inhibited cell motility and invasion of these cells in dose-dependent manners (Fig. 5B and C; Supplementary Fig. S3). Olmesartan also inhibited cell motility and invasion of these cells without affecting their cell viability (data not shown).

To eliminate the possibility that the metastasis-suppressing effects of the identified drugs might result from off-target effects of the drugs, we conducted validation experiments to determine whether knockdown of the gene encoding the protein that the drugs target would show the same effects. HCCLM3 cells expressing shRNA for either HSD11β1 or H⁺/K⁺-ATPase beta showed decreased cell motility and invasion without affecting cell viability (Fig. 5D–F; Supplementary Fig. S3).

These results demonstrated that the drugs that were identified through the zebrafish-based screening assay were able to show the same suppressor effects on cell motility and invasion of human cancer cells.

Although the identified drugs inhibited cell motility and invasion of highly metastatic cell lines in vitro, it was unclear whether the drugs could suppress metastatic dissemination of human cancer cells in vivo. Therefore, we examined whether the identified drugs could suppress metastatic dissemination of these cells in a zebrafish xenotransplantation model. Adrenosterone was selected as a test drug in the examination because HSD11β1 was overexpressed only in highly metastatic cell lines, suggesting it could be a novel target for blocking metastatic dissemination of cancer cells (Fig. 5A). RFP-labeled HCCLM3 (HCCLM3R) cells were injected into the duct of Cuvier of Tg(kdrl:eGFP) zebrafish at 2 dpf and then maintained in the presence of either vehicle or adrenosterone. Twenty-four hours after injection, the frequencies of the fish showing metastatic dissemination of the inoculated cells were measured under fluorescence microscopy (Fig. 6A). In this model, the dissemination patterns were generally divided into three categories: (i) head dissemination, in which disseminated HCCLM3R cells exist in the vessel of the head part; (ii) trunk dissemination, in which the cells were observed in the vessel radiating from the trunk to the tail; and (iii) end-tail dissemination, in which the cells were observed in the vessel of the end-tail part (Fig. 6B).

Two independent experiments revealed that the frequencies of the fish in adrenosterone-treated group showing head, trunk, or end-tail dissemination significantly decreased to 55.3% ± 7.5%, 28.5% ± 5.0%, or 43.5% ± 19.1%, respectively, when compared with those in vehicle-treated group; 95.8% ± 5.8%, 47.1% ± 7.7%, 82.6% ± 12.7%. Conversely, the frequency of the fish in adrenosterone-treated group not showing any dissemination significantly increased to 45.4% ± 0.5% when compared with those in the vehicle-treated group, 2.0% ± 2.9% (Fig. 6C; Supplementary Table S5).

Similar effects were observed in another xenograft experiment using RFP-labeled MDA-MB-231 (231R) cells. This experiment followed the experimental design in Fig. 5A. In adrenosterone-treated group, the frequencies of the fish showing head, trunk, or end-tail dissemination, significantly decreased to 27.9% ± 25.3%, 33.3% ± 11.7%, or 51.2% ± 15.9%; conversely, the frequency of the fish not showing any dissemination, significantly increased to 39.1% ± 8.2% when compared with those in the vehicle-treated group; 75.2% ± 23.2%, 57.7% ± 6.7%, 85.7% ± 3.4%, or 2.0% ± 2.9%, respectively (Fig. 6D; Supplementary Table S5).

To eliminate the possibility that the metastasis-suppressing effects of adrenosterone might result from off-target effects of it, we conducted validation experiments to determine whether knockdown of HSD11β1 would show the same effects. The basic experimental process followed the experimental design in Fig. 5A except that subclones of HCCLM3R cells that express shRNA targeting either LacZ or HSD11β1 were injected into the fish at 2 dpf and the fish were maintained without drug. In the fish that were inoculated with shHSD11β1 HCCLM3R cells, the frequencies of the fish showing head, trunk, and end-tail dissemination, significantly decreased to 27.3% ± 4.9%, 14.8% ± 0.7%, or 49.8% ± 16.5%; conversely, the frequency of the fish not showing any dissemination, significantly increased to 49.2% ± 4.4% compared with those fish that were inoculated with shLacZ HCCLM3R cells; 80.7% ± 27.1%, 49.3% ± 4.5%, 79.1% ± 7.7%, or 0% (Fig. 6E; Supplementary Table S5).

These results indicated that adrenosterone could suppress metastatic dissemination of human cancer cells in vivo.

It was unclear how adrenosterone suppressed metastatic dissemination of cancer cells. Therefore, we elucidated the mechanism of action of it. The screen reported here relies upon cell dissemination that is induced by Twist1a-driven EMT. Previously, HCCLM3, MDA-MB-231, and MDA-MB-435 cell lines have been reported to express Twist and possess mesenchymal properties (12, 32). Thus, adrenosterone was expected to suppress cell motility and invasion of these cells that had already transitioned to mesenchymal-like traits via EMT, not inhibit EMT itself. HSD11β1 is NADPH-dependent enzyme that catalyzes the interconversion of the steroid pair of the inactive metabolite cortisone and the stress hormone cortisol. It is reported that a transcriptional activation of the glucocorticoid receptor, which is a receptor for cortisol, upregulates EMT-inducing transcriptional factors such as snail and slug (33). Therefore, we speculate that adrenosterone might interfere with the acquired mesenchymal traits of these cells through downregulation of these genes. Western blot analysis revealed Snail and Slug expression in adrenosterone-treated HCCLM3 and MDA-MB-231 cells was markedly decreased compared with vehicle-treated cells (Fig. 7A). Subsequently, E-cadherin expressions in adrenosterone-treated these cells were restored at both the mRNA and the protein levels compared with that in vehicle-treated these cells (Fig. 7B and C). Also, knockdown of HSD11β1 showed the same effects (Fig. 7D). However, adrenosterone-treated cells did not show decreased expression of mesenchymal makers such as N-cadherin and vimentin when compared with vehicle-treated cells (Supplementary Fig. S4A).

A recent study demonstrated that cancer cells can stably exist in a hybrid epithelial–mesenchymal state where they coexpress epithelial and mesenchymal markers (34). Therefore, we further investigated whether adrenosterone-treated cells might exit in such a hybrid state. FACS analysis demonstrated that the percentage of E-cadherin–positive cells (E-cad⁺ cells) in adrenosterone-treated HCCLM3 (∼30%; Fig. 7E), and immunofluorescence analysis revealed the E-cad⁺ cells showed cell–cell adhesion; conversely, E-cadherin–negative cells maintained a loss of cell–cell contact (Supplementary Fig. S4B). mRNA expression of other EMT-related epithelial markers: CDH1, OCLN, CLDN3, CLDN6, CLDN7, KRT14, and KRT19 in the E-cad⁺ cells was more than 2.5-fold increase when compared with vehicle-treated cells. Conversely, expression of mesenchymal markers, MMP1, MMP2, and S100A4, in the E-cad⁺ cells was decreased to less than half of that in vehicle-treated cells (Fig. 7F), while expression of a few other mesenchymal markers (CDH2, VIM, FN1, SMA, MMP7, MMP9, and MMP14) was similar to that in vehicle-treated cells (Supplementary Fig. S4A). In addition, consistent changes at protein levels of these genes were also observed (Fig. 7G).

These results indicated a fraction of adrenosterone-treated HCCLM3 cells existed in a hybrid epithelial–mesenchymal state through recovery of epithelial markers but not lose of mesenchymal markers with the exception of MMP1, MMP2, and S100A4. Accumulating evidence demonstrates that reexpression of either E-cadherin or claudin7 is enough to interfere with metastatic progression of cancer cells (35, 36). Therefore, we conclude that adrenosterone could suppress metastatic dissemination of HCCLM3 and MDA-MB 231 cells through recovery of epithelial markers.

The zebrafish system has been increasingly recognized as a platform for chemical screening because it provides the advantage of high-throughput screening in an in vivo vertebrate setting with physiologic relevance to humans (14). This study demonstrated that approximately 80% of Twist1a-ER^T2/xmrk double transgenic zebrafish induced dissemination of mCherry-labeled hepatic cells from the liver within 5 days through induction of EMT. This rapid and high frequency induction of cell dissemination enabled us to perform an in vivo chemical screen for identification of antimetastasis drugs targeting metastatic dissemination of cancer cells. Indeed, this screen identified three FDA-approved drugs; adrenosterone, rabeprazole, and olmesartan, that suppressed cell dissemination. The antimetastatic potential of these drugs was confirmed in in vitro experiment using highly metastatic human cell lines. Especially, adrenosterone was able to suppress metastatic dissemination of these cells in a zebrafish xenotransplantation model, demonstrating that the drugs identified through this zebrafish-based screen showed similar suppressor effects on human cancer cells. Thus, this model offers a novel platform for discovery of antimetastasis drugs targeting metastatic dissemination of human cancer cells.

This study revealed a new role for adrenosterone as a potential inhibitor of metastatic dissemination of cancer cells. Adrenosterone was originally isolated from the adrenal cortex of the fish as a steroid hormone with a weak androgenic effect, and currently used as a daily supplement for bodybuilders (28, 37). As a competitive inhibitor of HSD11β1, which catalyzes the interconversion of the steroid pair of the inactive metabolite cortisone and the stress hormone cortisol, adrenosterone decreases the amount of cortisol. HSD11β1 is normally expressed in key metabolic tissues including the liver, adipose tissue, and the central nervous system (28). Here, we provide the first evidence that elevated expression of HSD11β1 was observed only in highly metastatic human cell lines but not in poorly metastatic cell lines. Pharmacologic inhibition of HSD11β1 enabled suppression of both spontaneous cell dissemination from a primary tumor site in Twist1a-ER^T2/xmrk double transgenic zebrafish and metastatic dissemination of human cancer cells in a zebrafish xenotransplantation model, and genetic inhibition of HSD11β1 showed the same effects. Therefore, HSD11β1 could be a novel therapeutic target for inhibiting metastatic dissemination of cancer cells.

We also showed the effects of adrenosterone on HCCLM3 cells were heterogeneous and E-cad⁺ cells in adrenosterone-treated HCCLM3 cells showed elevated expression of several epithelial makers compared with adrenosterone-treated E-cadherin–negative cells. These results indicate effects of adrenosterone on these cells are heterogeneous and some of the cells were able to recover epithelial markers in the presence of adrenosterone. The heterogeneity might result from clonality of HCCLM3 and MDA-MB-231 cell lines, which are composed of multiple clones and each clone shows different gene expressions and different metastatic potencies (38, 39).

This report did not address whether transcriptional activation of glucocorticoid receptor, which is a receptor for cortisol, might induce EMT in epithelial cells and be essential for maintaining mesenchymal-like traits in HCCLM3 and MDA-MB-231 cells. Previous studies demonstrated that transcriptional activation of glucocorticoid receptor upregulates EMT-inducing transcription factors such as Snail and Slug, and cortisol promotes metastatic progression of breast cancer cells (33, 40). This evidence supports adrenosterone as a suppressor of metastatic progression of human cancer cells.

No potential conflicts of interest were disclosed.

Conception and design: J. Nakayama, Z. Gong

Development of methodology: J. Nakayama

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Nakayama, J.-W. Lu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Nakayama, J.-W. Lu, H. Makinoshima

Writing, review, and/or revision of the manuscript: J. Nakayama, J.-W. Lu, H. Makinoshima, Z. Gong

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Nakayama, H. Makinoshima

Study supervision: J. Nakayama, Z. Gong

We sincerely appreciate Drs. Joshua Collins (NIH/NIDCR) for helping in this research; Diane Palmieri (NIH/NCI), Daniel Fitzgerald (Otsuka Maryland Medical Laboratories), Natascia Marino (Indiana University), and Takashi Hoshino (Takeda Pharmaceutical Company, Ltd.) for editing this article; Motomi Osato (National University of Singapore), Michael Brand (Dresden University of Technology, Germany), and Zhao-You Tang (Fudan University) for providing pCAG-Cre-ER^T2, hsp70:mCherry-T2A-CreER^T2 vector, and HCCLM3 cells; and Shu Wang for providing cell culture facilities. This study was funded by National Medical Research Council of Singapore (R-154000547511) and Ministry of Education of Singapore (R-154000A23112) to Z. Gong.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.