Suppression of tumor metastasis by a RECK-activating small molecule

RECK encodes a membrane-anchored protease-regulator which is often downregulated in a wide variety of cancers, and reduced RECK expression often correlates with poorer prognoses. In mouse models, forced expression of RECK in tumor xenografts results in suppression of tumor angiogenesis, invasion, and metastasis. RECK mutations, however, are rare in cancer genomes, suggesting that agents that re-activate dormant RECK may be of clinical value. We found a potent RECK-inducer, DSK638, that inhibits spontaneous lung metastasis in our mouse xenograft model. Induction of RECK expression involves SP1 sites in its promoter and may be mediated by KLF2. DSK638 also upregulates MXI1, an endogenous MYC-antagonist, and inhibition of metastasis by DSK638 is dependent on both RECK and MXI1. This study demonstrates the utility of our approach (using a simple reporter assay followed by multiple phenotypic assays) and DSK638 itself (as a reference compound) in finding potential metastasis-suppressing drugs.

MS275 failed to suppress metastasis of RM72 ( Fig. 2h-j) even at a dosage where tumor volume and body weight were slightly reduced (Fig. 2k,l), and QSS stimulated metastasis (Fig. 2m-o) at a dosage where tumor volume and body weight were unaffected (1 mg/kg; Fig. 2p,q). Thus, the ability to induce RECK-expression and flat reversion in vitro are not sufficient for suppression of metastasis in vivo.

Differential effects of HDAC inhibitors on RM72 cells in suspension.
In suspension culture 27 , HT1080 cells form aggregates (Fig. 4a, panel 1) while the metastatic RM72 cell line produces two populations of cells: a single cell (SC) population and a population of aggregated cells (AG) (Fig. 4a, panel 2). When separated (Fig. 4a, panels 3 and 4), SC and AG populations proliferate more slowly than the mixed population of parental cells (Fig. 4b), suggesting metabolic cooperation between the two populations. Interestingly, AG populations have a lower metastatic potential than the parental RM72 cells or the SC population ( Fig. 4c; Supplementary  Fig. S3), suggesting that the SC population is the metastatic component of RM72.
When RM72 cells were treated with DSK638 in suspension culture, the size of the aggregates increased, the number of single cells decreased (Fig. 4d, panels 2 vs. 1), and some cells in the periphery of the aggregates died (Fig. 4d, panel 6). When treated with MS275, almost all RM72 cells were single (Fig. 4d, panel 3) and alive (Fig. 4d, panel 7) which was confirmed by flow cytometry (Supplementary Fig. S4). In contrast, numerous dead single cells were found after QSS-treatment (Fig. 4d, panel 8).
When RECK was overexpressed in separated populations, the AG population formed larger aggregates and the SC population formed small aggregates (Fig. 4e, panels 5 and 10). DSK638 induced similar changes in separated populations (Fig. 4e, panels 2 and 7). In contrast, MS275 caused aggregates to disassociate (Fig. 4e, panel 3). QSS had little effect on the AG population, but caused some aggregation of SC cells, although, QSS was less active than DSK638 in converting the SC population to an AG population (Fig. 4e, panels 4, 9 vs. 2, 7).
We also found that in suspension culture, RECK protein was robustly induced by DSK638 but not by MS275 or QSS (Fig. 4f). Such anchorage-independent action of DSK638 may also contribute to its unique bioactivity. Hence, RM72 in suspension culture responded differently to these three HDAC inhibitors.

Molecular mechanisms of metastatic conversion and metastasis suppression. To gain insights
into the molecular mechanisms of the phenotypic changes described above, we analyzed the transcriptomes of the following cell sets and their respective controls: (i) four cell types without treatment: HT1080 (control), RM72, AG, SC; (ii) RM72 cells treated with three drugs (DSK638, MS275, QSS); (iii) RM72 cells infected with RECK-adenovirus; and (iv) RECK-depleted RKD72 cells treated with DSK638. Gene Set Enrichment Analysis (GSEA) 28 yielded three remarkable findings (summarized in Fig. 4g). First, the gene set NEGATIVE_REGU-LATION_OF_METABOLIC_PROCESS was enriched in the transcripts more abundant in the SC population than in the AG population (Supplementary Tables S4, S5); the leading-edge subset of this comparison contains three ID-family genes (ID1, ID2, ID3) encoding transcription factors associated with suppression of cell  Fig. 4c). ID1, ID2, and ID3 were upregulated in RECKdepleted cells (Fig. 4i), suggesting that RECK suppresses these genes. Second, the gene set EXTRACELLULAR_STRU CTU RE_ORGANIZATION_AND_BIOGENESIS was enriched in the transcripts increased by DSK638 (Supplementary Tables S6, S7); the leading-edge subset of this comparison contains eight PCDHB-family genes encoding neural cell-adhesion molecules [30][31][32] . In RM72 cells, PCDHB11 is strongly upregulated by DSK638 ( Supplementary Fig. S5a). The levels of PCDHB11 in the RM72, SC, and AG populations show an inverse correlation with their metastatic potentials: lower in RM72 and SC, and higher in AG ( Third, the gene set MITOTIC_CELL_CYCLE was enriched in transcripts decreased by DSK638 (Supplementary Tables S8, S9) and in transcripts that were less abundant in RM72 than in HT1080 (Supplementary Tables S10, S11); the leading-edge subsets of these two comparisons largely overlapped, suggesting that DSK638 inhibits the mitotic cell cycle, a function already attenuated in RM72 ( Supplementary Fig. S6). Of note, both leading-edge subsets contain E2F1 (a proliferation-associated transcription factor activated by MYC 33 ) and a number of E2F1-targets. We therefore focused on E2F-family and MYC-family genes ( Supplementary Fig. S7) and found the following: (i) E2F7, encoding a "repressor E2F" 34 , is expressed in HT1080 but undetectable in RM72 ( (iv) MXI1, a putative tumor suppressor encoding a MYC-antagonist 35,36 , was strongly upregulated by DSK638 (Fig. 4k, bars 8 vs. 7). Upregulation of MXI1 by MS275 or QSS was less than by DSK638 (Fig. 4k, bars 9, 10 vs. 8). DSK638 upregulates MXI1 in RECK-depleted RKD72 cells ( Supplementary Fig. S7c, bars 1-4), and RECKoverexpression fails to upregulate MXI1 ( Supplementary Fig. S7c, bars 6 vs. 5), indicating RECK-independence of MXI1-upregulation by DSK638. (v) DSK638 upregulates MXI1 and downregulates E2F1 at the protein level (Fig. 4l). These findings support the idea that DSK638 upregulates MXI1 which antagonizes MYC, leading to the downregulation of E2F1, E2F2, and their targets, resulting in suppression of mitotic cell cycle progression.
Taken together, our data indicate that DSK638 activates at least three molecular systems: RECK (a regulator of ID expression and extracellular proteases), PCDHBs (mediators of cell-cell adhesion), and MXI1 (an inhibitor of mitotic cell cycle progression) (Fig. 4m). Consistent with this model, DSK638 failed to suppress the metastasis of RM72 cells with reduced RECK expression (RKD72 in Fig. 2e, 2f, and 2g) or reduced MXI1 expression (compare Fig. 4n with Fig. 2d).

Discussion
Compelling evidence implicates RECK in metastasis suppression 6,[12][13][14]37,38 . Our screen for chemicals capable of inducing RECK expression and flat reversion in tumor cells led to the identification of a metastasis suppressing drug, DSK638. Intra-peritoneal injection of DSK638 suppresses lung-metastasis of a fibrosarcoma cell line, RM72, subcutaneously inoculated into nude mice. Our findings also shed light on mechanisms by which these sarcoma cells acquired metastatic potential and how DSK638 activates RECK expression and suppresses tumor metastasis.
E2F7 is a feedback regulator of E2F1 capable of inducing cell-cycle-arrest under the control of RB and p53 34,39 . The absence of E2F7-expression in RM72 cells is likely to contribute to their malignant properties. If so, Figure 2. Metastasis-suppressing activity of DSK638. (a) Effects of DSK638 on the level of RECK protein in RM72 and RKD72 (RECK-depleted) cells. The cells were exposed to medium containing 10 µM DSK638 for 48 h, and the lysates were subjected to immunoblot assay. Note that DSK638 fails to induce RECK expression in RKD72 cells (lane 4). (b-d) Effects of DSK637 and DSK638 on the growth and lung metastasis of RM72 cells inoculated subcutaneously into nude mice. (b) The IVIS images of two typical cases per group after 2-weeks treatment with the indicated compound at the indicated dose. Upper panel: whole body. Lower panel: resected lung. (c) Relationship between tumor volume and lung metastasis (relative bioluminescence intensity; each dot represents one animal). (d) The ratio between lung metastasis and tumor volume. Bar represents mean ± s.e.m. Note the activity of DSK638 (data in green) to suppress lung metastasis of RM72 cells. (e-g) Effects of DSK637 and DSK638 on the growth and lung metastasis of RECK-depleted RKD72 cells. Experiments similar to those shown in b-d were performed using RKD72 cells. Note that DSK638 (data in green) failed to suppress lung metastasis of RKD72 cells. (h-l) Effects of MS275 on the growth and lung metastasis of RM72 cells inoculated subcutaneously into nude mice. Experiments similar to those shown in b-d were performed using MS275 (2.5 mg/kg). The effects on tumor sizes (k) and body weights (l) are also shown. Note that MS275 (data in orange) failed to suppress lung metastasis of RM72 cells at a dosage where tumor growth and body weight were slightly reduced. (m-q) Effects of QSS on the growth and lung metastasis of RM72 cells inoculated subcutaneously into nude mice. Experiments similar to those shown in h-l were performed using QSS at three doses (0.03, 0.5, and 1 mg/kg). Note that QSS (data in dark blue) stimulated lung metastasis of RM72 cells at a dosage (1 mg/kg) where tumor growth and body weight were unaffected.   S2a) and weakens KLF2 inhibition of RECK promoter activity (Fig. 3h), suggesting that reduction in KLF2-mediated transcriptional repression underlies DSK638-mediated RECK gene activation. This model was supported by the observation that in the cells overexpressing KLF2, the level of RECK protein was lower than the control when incubated in regular medium but higher in medium containing DSK638, making the ratio of induction greater (Fig. 3i). Unexpectedly, however, when KLF2 was depleted, the level of RECK protein was somewhat reduced in the presence of DSK638 (Fig. 3k, lane 4), as if under these conditions, KLF2 is required for RECK expression. A possible model to explain these findings is that DSK638 converts KLF2 from a repressor to an activator essential for DSK638-stimulated RECK-upregulation (Fig. S9a). Given that recruitment of transcriptional regulators by a closely related DNA-binding protein, KLF1, is switched by acetylation 40 and that the relevant acetylation sites (K288 and K302 in KLF1) are conserved in KLF2 (Fig. S2b), it is tempting to speculate that DSK638, probably acting as an HDAC inhibitor (Table S2), might convert KLF2 from a repressor to an activator by inhibiting deacetylation of these sites.   www.nature.com/scientificreports/ KLF2 is frequently mutated in splenic marginal zone lymphoma 39,41 , and downregulated in ovarian 41 and prostate 42 cancers. KLF2 inhibits invasion and/or metastasis of prostate 43 and colon 44 cancer cells. Our findings raise the possibility that at least a part of such anti-oncogenic functions of KLF2 may be mediated by RECK. In support of this premise, we found that RECK had a negative effect on ID-expression: in many cell types, ID proteins block differentiation, maintain stemness, and exerts oncogenic/pro-metastatic effects 29 . KLF2 is also known to play important roles in various physiological processes, including lung development 45 , cardiovascular development 46 , hematopoiesis 47 , immune cell differentiation/functioning 26,[48][49][50] , and atheroprotection 26,51 . Possible roles for RECK as a conditionally regulatable effector of KLF2 in these processes should be interesting subjects for future studies.
MXI1 is a known MYC-antagonist 33,36 , and MXI1-upregulation is likely a critical event in metastasis-suppression by DSK638: Mxi1-deficient mice are hyperplastic in multiple tissues 33 , suggesting a role for MXI1 in tumor suppression. Among the three HDAC inhibitors examined, the strengths of MXI1 upregulation and E2F1/E2F2 downregulation are correlated (Fig. 4k, bars 8-10 vs. Figure 4j, bars 3-5, 8-10), supporting our model that MXI1 is involved in the E2F1-downregulation induced by DSK638 (Fig. 4m). Although the relevance of MXI1 to cancer metastasis remains unexplored, our data, together with the previous observations that MYC and E2F1 promote epithelial-mesenchymal transition and metastasis 52,53 , implicate MXI1 in metastasis-suppression.
DSK638 upregulates multiple PCDHB-family genes. Protocadherins (PCDHs) play important roles in specific cell-cell adhesions during neural development [30][31][32] . Anti-oncogenic functions of some PCDHs have also been reported, although, their mechanisms of action remain obscure 54 . Our data indicate that PCDHB11, a member strongly upregulated by DSK638 in RM72 cells, may contribute to metastasis-suppression by enhancing their cell-cell adhesion. Thus, DSK638 promotes two pathways that affect extracellular interactions: PCDHB11, as described here, and RECK, a membrane-anchored protein that regulates peri-cellular proteolysis 6-10, 55 .
We found that the single cell population of RM72 cells had a higher metastatic potential than the aggregate population, although cell clustering has previously been implicated in breast cancer metastasis 56 . The apparent disparity may reflect the differences in experimental systems employed and/or the step(s) of metastasis being focused on.
Recent studies have yielded several important advances in understanding the role of RECK in angiogenesis. Characterization of tissue-selective knockout mice revealed that RECK expression in both endothelial cells and mural cells is required for embryonic vascular development and survival; in mural cells RECK is essential as early as the mid-gestation period and in later embryonic stages RECK is required in endothelial cells for proper brain angiogenesis 57 . Aortic ring assays revealed that Reck-deficiency resulted in the formation of excessive vascular sprouts that are prone to fuse with each other and are poorly associated with mural cells and the ECM 57 , implicating RECK in blood vessel maturation. Endothelial RECK was found to serve as a ligand-selecting component of the WNT7 receptor required for brain angiogenesis and blood-brain-barrier formation [58][59][60][61] . RECK in neural precursor cells, a WNT7-producer, was also found to be essential for forebrain angiogenesis, probably by facilitating the delivery of active ligand to the receptor expressed on the surface of endothelial cells 62 . Hence, RECK protein is expressed in multiple cell types where it probably has different functions but acts in concert to help coordinate proper vascular development.
On the other hand, forced RECK expression in cancer cells was found to result in suppression of tumor angiogenesis and metastasis in a mouse xenograft model 7    www.nature.com/scientificreports/ vessels well-covered by basement membrane 7 , suggesting that RECK expression in the surrounding tissues (i.e., tumor cells) suppresses the branching of tumor vessels derived from the host animal. Thus, RECK promotes normal angiogenesis but suppresses tumor angiogenesis. Although the exact mechanism(s) by which RECK suppresses tumor metastasis remains unclear, downregulation of ID (this study) and normalization of tumor vessels (see above) may contribute to this suppression. Other RECK-mediated processes relevant to tumor metastasis include the regulation of ECM-degradation 65 , Notch-signaling 9 , STAT3-signaling 66 , cell migration 67 , mesenchymal phenotype 68 , fibrillin fiber formation 55 , and cellular senescence 69 . In addition to these RECK-mediated events, DSK638 may also affect other processes such as regulation of cell cycle progression (via MXI1) and cell-cell adhesion (via PCDHBs). Consequently, RECK may be considered a useful marker as well as an effector for drug screening (Fig. S9b). This study has also demonstrated the practical value of our approach using a molecular marker (e.g., RECK) for a primary screen and multiple phenotypic assays for secondary screening.
Since RM72 may represent only a part of the metastatic properties of malignant cancers 4 , testing DSK638 in other metastasis assay systems would be an important next step in developing metastasis-suppressing drugs of clinical value 5 . Such drugs, once proven to have reasonably low side effects, may be useful for preventing or retarding recurrence in cancer patients. Drugs that prevent metastatic colonization may be useful for adjuvant chemotherapy. We expect that drugs that prevent steps after metastatic colonization should have wide therapeutic applications. Continued efforts to develop bioassay systems that represent defined step(s) of metastasis will be of great benefit in achieving this goal.

Methods
Chemicals. The DSK compounds were synthesized by SUMIKA TECHNOSERVICE (Osaka): the DSK Project was a joint research project between Kyoto University and Dainippon Sumitomo Pharma, Co., Ltd. Sources and catalogue numbers of the other chemicals are listed in Supplementary Table 1. For in vitro use, these chemicals were dissolved in DMSO (Sigma Aldrich) to a concentration of 10 mM, stored at -20˚C, and further diluted to an appropriate final concentration in appropriate medium upon use. For in vivo studies, chemicals were dissolved in olive oil and administrated daily by intraperitoneal (ip) injection for 14 days.
Drug screening. Details of high throughput screening will be described elsewhere (manuscript in preparation). Briefly, a 1-kb RECK promoter fragment was inserted into the multicloning sites of the pL4.10 vector (Promega) to generate pL4.10-phRECK. HT1080 cells plated on the previous day were transfected with a mixture of two plasmids, pL4.10-phRECK and pTK-RLuc (Renilla luciferase reporter vector). After 6 h, the cells were exposed to the test chemical (3 µM) for 24 h and then subjected to a dual luciferase assay to identify compounds that activate the RECK promoter. The 43 compounds that gave rise to more than twofold increase in luciferase activity by pL4.10-phRECK were examined further using immunoblot assays.
Immunoblot assay. HT1080 cells plated on the previous day were treated with the test chemical or the vehicle (DMSO) in growth media for 48 h. The cells were lysed as described previously 69 . The protein extracts were separated by electrophoresis on a 10% SDS polyacrylamide gel. Protein detection was performed using the antibodies described in Supplementary Table S1. For visualization, the Enhanced Chemiluminescence kit (Millipore) was used. Images were recorded and analyzed using LAS-4000 and the MultiGauge software (Millipore) according to the manufacturer's instructions.
Quantitative flat reversion (qRev) assay. DT cells were stably transfected with pmCherry (Clontech 632,522), and a clone (named DSK4b) uniformly emitting bright red fluorescence was isolated. DSK4b cells (2,000 cell/well) were seeded onto 96-well tissue culture plates, and test chemicals (in serial dilutions) were added at 8 h. After incubation for 72 or 96 h, the nuclei were stained with Hoechst33342 (0.5 µg/ml) just before image analysis. The number of nuclei per unit area (blue fluorescence; n) and the total area of red fluorescence per unit area (A) were recorded using a cell image analyzer (Cytell, GE or ArrayScan VTI, Thermo Fischer). The area per cell (A/n) represents the activity of a drug to induce flat reversion, and the number of nuclei (n) is related to the cytotoxicity of the test chemical.
Matrigel invasion assay. HT1080 cells plated on the previous day were treated with the test chemical (10 µM) or the vehicle (DMSO) in growth media for 48 h. FluoroBlok Transwell Inserts (8 -mm pore size; Corning) were pre-coated by adding 100 µl diluted Matrigel (BD Biosciences; 25 µg/100 µl) onto the membranes and air drying overnight. The coated inserts were placed into 24-well plates containing growth media as a chemoattractant. The cells treated with the test chemical were pre-labeled for 30 min with CellTracker Green CMFDA (Life Technologies) and suspended in DMEM containing 0.1% FBS and the test chemical and plated onto the insert. After 24-h incubation, the cells that had invaded to the lower side of the membrane were photographed (4 fields/2 inserts/sample) using an inverted fluorescent microscope (KEYENCE BZ-9000). The area occupied by the fluorescent, invading cells was quantified using Image J.
Gelatin zymography. RM72 cells plated on the previous day were treated with the test chemical (10 µM) or the vehicle (DMSO) in growth media for 24 h. The cells were then exposed to serum-free DMEM for 24 h, and the supernatants were subjected to electrophoresis on a 1% SDS-polyacrylamide gel containing 0.1% gelatin under non-reducing conditions. The gel was washed twice in 2.5% Triton X-100, 10 mM Tris-HCl pH 8.0 for 30 min and once in 10 mM Tris-HCl pH 8.0 for 30 min at room temperature. The gel was then incubated in

Spontaneous tumor metastasis model in nude mice.
The experiments using mice were approved by the Animal Research Committee, Kyoto University, and were performed in accordance with MEXT Notice No. 71 and the Act on Welfare and Management of Animals, Japan; the present report was prepared in compliance with the ARRIVE guideline. Spontaneous metastasis assays were performed as previously described 15 . In brief, RM72 cells or RKD72 cells (3 × 10 6 ) suspended in 0.1 ml PBS were injected subcutaneously into the right posterior flank of Balb/c nu/nu mice (6 weeks old, male, Charles River). Small tumors (~ 3 × 3-mm diameter) developed 5 days after injection. The mice were randomly divided into groups (n ≥ 5 animals per group) and treated with vehicle or the test chemical via intra-peritoneal injection. After 14-d treatment, the fates of tumor cells in the mice were assessed by bio-imaging (see next section). The tumor volume (length x width x height) and body weight were measured once a week. The sample size of at least five animals (with visible tumors at day 5 after cell inoculation) per group was chosen to make experiments manageable (especially, the daily treatment with drugs) and to obtain reproducible results based on our experience with this assay. The animals with no visible tumor at day 5 were excluded from the experiments. Data for all individual animals were plotted on a graph (X-axis: tumor volume; Y-axis: photon flux from resected lung). All cages were kept under comparable conditions. The order of treatments and measurements were not randomized, but all data retrospectively confirmed that the observed differences could not be explained by such artefacts. Animal experiments were carried out by Y.Y., K.Y., and M.N.; measurements were not blinded, but all animal experiments were not hypothesis-driven and performed objectively.
Tumor imaging in vivo. Mice were anesthetized and injected intraperitoneally with 75 mg/kg of d-luciferin (Promega) in PBS (-). Bioluminescence images were acquired with the IVIS Imaging System (Xenogen) at 5 min after injection. Photons emitted from living mice or isolated organs were measured with a recording period of 60 s using the Living Image software (Xenogen). Fig. 3a were generated as follows. No. 3 (-177 to -1): pL4.10-phRECK was digested with XhoI and SmaI, and the cohesive end was filled in followed by selfligation (i.e., circularization). No. 2 (-1173 to -48) and No. 5 (-48 to -1): The 1.3-kb KpnI-PvuII fragment from pL4.10-phREC was digested with EaeI to isolate the KpnI-EaeI (1.14 kb) and EaeI-PvuII (0.16 kb) fragments whose cohesive ends were then filled in. The KpnI-EaeI (1.14 kb) fragment was ligated into pL4.10 digested with Kpn I and EcoRV to obtain construct No. 2. The EaeI-PvuII (0.16 kb) fragment was further digested with HindIII and ligated into pL4.10 digested with EcoRV and Hind III to obtain construct No. 5. No. 4 (-177 to -48): Construct No. 2 was digested with Xho I and Sma I, and the cohesive end was filled in followed by self-ligation (i.e., circularization). The Sp1 site mutants shown in Fig. 3b were generated using QuikChange Lighting Site-Directed Mutagenesis Kit (Agilent). For the small segments of RECK promoter tested in Fig. 3c (sequences are shown in Supplementary Fig. S1), annealed synthetic oligonucleotides were cloned into pGL4.10.

Luciferase reporter assay.
In experiments with DSK638 a duel luciferase assay using firefly luciferase as the reported vector and the Renilla luciferase vector (pRL-TK) as the control vector could not be used: DSK638 activated Renilla luciferase activity from pRL-TK to such an extent that it could not serve as an internal control. Therefore, in experiments with DSK638 firefly luciferase activity alone was measured. HT1080 cells (40,000