Reduced USP33 expression in gastric cancer decreases inhibitory effects of Slit2‐Robo1 signalling on cell migration and EMT

Abstract Objectives Gastric cancer (GC) is one of the most common cancers in the world, causing a large number of deaths every year. The Slit‐Robo signalling pathway, initially discovered for its critical role in neuronal guidance, has recently been shown to modulate tumour invasion and metastasis in several human cancers. However, the role of Slit‐Robo signalling and the molecular mechanisms underlying its role in the pathogenesis of gastric cancer remains to be elucidated. Materials and methods Slit2, Robo1 and USP33 expressions were analysed in datasets obtained from the Oncomine database and measured in human gastric cancer specimens. The function of Slit2‐Robo1‐USP33 signalling on gastric cancer cells migration and epithelial‐mesenchymal transition (EMT) was studied both in vitro and in vivo. The mechanism of the interaction between Robo1 and USP33 was explored by co‐IP and ubiquitination protein analysis. Results The mRNA and protein levels of Slit2 and Robo1 are lower in GC tissues relative to those in adjacent healthy tissues. Importantly, Slit2 inhibits GC cell migration and suppresses EMT process in a Robo‐dependent manner. The inhibitory function of Slit2‐Robo1 is mediated by ubiquitin‐specific protease 33 (USP33) via deubiquitinating and stabilizing Robo1. USP33 expression is decreased in GC tissues, and reduced USP33 level is correlated with poor patient survival. Conclusions Our study reveals the inhibitory function of Slit‐Robo signalling in GC and uncovers a role of USP33 in suppressing cancer cell migration and EMT by enhancing Slit2‐Robo1 signalling. USP33 represents a feasible choice as a prognostic biomarker for GC.


| INTRODUC TI ON
Gastric cancer (GC) is the third leading cause of cancer-related death and responsible for approximately 723 000 deaths worldwide every year. 1 Nearly half of the cases occur in Eastern Asia and are mostly diagnosed at the advanced stage. 2 As a consequence, the 5-year survival rate for advanced GC patients remains at only 5%-20%. 3 Hence, it is critical to explore the molecular mechanisms of GC development for finding new treatment strategy of GC.
Slit glycoproteins (Slit1-3), originally discovered as neuronal guidance cues, are secreted by midline glia 4 that exert their function by binding to single-pass transmembrane proteins Roundabout family (Robo1-4). 5-7 The Slit-Robo signalling pathway plays important roles not only in neuronal guidance but also during cell migration of a wide range of cell types. [6][7][8][9][10] Recent studies indicate that the inactivation of this pathway is associated with the progression of several cancer types, [11][12][13] including pancreatic cancer, 14 breast cancer, 15 as well as lung tumours. 16 However, the precise function of the Slit-Robo pathway in the development of GC remains ill-defined. A number of studies supported the notion that Slit-Robo signalling plays an important role in anti-tumour processes. 17,18 In contrast, two other reports suggested that Robo1 might promote tumorigenesis. 19,20 Ubiquitin-specific protease 33 (USP33), a member of ubiquitinspecific protease family, was initially identified as a substrate molecule which binds to VHL E3 ligase. 21 Previous studies showed that USP33 is a Robo1-interacting protein that is involved in Slit signalling in midline axons crossing. 22 Furthermore, USP33 is required for Slit-Robo signalling in inhibiting breast cancer cell migration. 15 Together, these studies demonstrate that USP33 plays an important role in the Slit-Robo pathway.
Recently, a study based on data from one patient cohort reported that USP33 expression was found to be reduced in GC and that reduced USP33 expression was associated with poor prognosis. 23 However, the precise molecular mechanisms of how USP33 exerts the anti-tumour function in GC remain to be elucidated.
Here, we set out to investigate the role of Slit-Robo signalling and the precise molecular mechanisms of how USP33 affects the Slit-Robo signalling in GC. and incubated in a humidified chamber at 37°C under 5% CO 2 .

| Wound-healing assay
Cell migration was examined in a modified wound-healing assay.
HEK293 cells which generate the full-length Slit2 protein tagged with 6xMyc tag were cultured in DMEM with 5% FBS. The medium from HEK293 cells was used as a mock control. 3 × 10 5 cells were grown in 6-well plates until approximately 90% confluent. Then we used sterile 200 μL pipet tips to make the scratch at the centre of the plates. The cells were washed with PBS and then incubated in medium with or without Slit2. After a period of time, images were taken under a microscope and the distance between both sides was measured.

| Western blot and immunoprecipitation
Total protein lysates were prepared with a protein extraction kit (KGP9100, Key Gene). Proteins were separated on 10% gels by SDS-PAGE and transferred to polyvinylidene difluoride (PVDF) membranes. After blocking in 5% non-fat milk in TBST buffer, the membranes were incubated with specific primary antibodies at 4°C overnight and followed by secondary antibodies. The signals were visualized using the chemiluminescence HRP substrate (WBKL0100; Millipore) and a chemiluminescence detection system.
Cell lysates were used for immunoprecipitation using the Dynabeads Protein G Immunoprecipitation kit (Invitrogen) following the manufacturer's guidelines. Immunoprecipitated proteins were then detected by Western blot.

| Immunohistochemical analysis
All specimens were fixed in 4% formalin and then embedded in paraffin. The 4 μm sections were incubated with primary antibodies at 4°C overnight. After washing with PBS, the sections were incubated with HRP-polymer-conjugated secondary antibody at room temperature for 1 hour. Next, sections were stained with DAB solution for 3 minutes and the nuclei were counterstained with haematoxylin. The results were evaluated by both the intensity of cell staining (graded as 0, no staining; 1, weak; 2, moderate; and 3, strong) and the percentage of positive tumour cells (graded as 0, <5%; 1, 5%-25%; 2, 26%-50%; 3, 51%-75%; and 4, >75%). Intensity score and percentage score were calculated.

| Immunofluorescence microscopy
Cells were cultured on collagen-coated glass coverslips for 24 hour and then rinsed with PBS twice before fixation with 4% formaldehyde for 20 minute at 37°C. Subsequently, cells were rinsed with PBS for three times and permeabilized with 0.2% Triton X-100 for 10 minute. The cells were incubated with PBS containing 1% BSA for 30 minute and then incubated with the primary antibody at 4°C overnight. Afterwards, cells were washed and incubated with fluorophore-conjugated secondary antibodies (Cy3™-Goat Anti-Rabbit IgG or Cy2™-Goat Anti-Mouse IgG Jackson Immunoresearch) for 2 hour and then stained with DAPI for 5 minute. After the final wash, a fluorescent microscope was used (Nikon, Japan) to collect images.

| Statistical analysis
All data were analysed using SPSS 20.0 software (SPSS Inc, Chicago, IL, USA). The results obtained from cell line experiments and animal assays were analysed using Student's t test (for two groups) or ANOVA (for more than two groups). Mann-Whitney U test was used to analyse differences in immunohistochemical (IHC) scores.
Chi-square test was used to analyse association of the expression of Robo1 and USP33 with clinicopathologic features. The Kaplan-Meier method was used the survival analyses. The optimal cut-off F I G U R E 2 Slit2 inhibits migration of GC cells in a Robo-dependent manner and suppresses EMT markers. A, Cell migration was examined in a wound-healing assay using MGC-803 cells in the medium containing the mock control (Ctrl), Slit2 and Slit2 with RoboN. Original magnification, 40×; scale bar: 100 µm. B, The migration of BGC-823 cells tested by wound-healing assays. C, Quantification of the distance of MGC-803 cell migration. D, Quantification of the distance of BGC-823 cell migration. E, Cell migration was examined in MGC-803 and BGC-823 cells transfected with Slit2 plasmid or control in a transwell assay. Original magnification, 100×; scale bar: 200 µm. F, Cell invasion was examined in MGC-803 and BGC-823 cells in the transwell assay. G, Cell migration was quantified. H, Cell invasion was quantified. I, Immunofluorescent microscopy was used to detect expression of E-cadherin (red) and vimentin (green) in MGC-803 and BGC-823 transfected with Slit2 plasmid or control, DAPI (blue) was applied for nuclear staining. Original magnification, 400×; scale bar: 50 µm. J, The expression of epithelial cell marker (E-cadherin), mesenchymal cell markers (N-cadherin, Vimentin) and related transcription factors (Snail, Slug) was analysed by Western blotting. GAPDH was used as an internal control. All data are shown as mean ± SEM and analysed by Student's t test, *P < 0.05, **P < 0.01, ***P < 0.001 values of USP33 expression were generated by X-tile software. Data are presented as the mean ± SD. P < 0.05 was considered significant.

| Expression of Slit2 and Robo1 is downregulated in gastric cancer
To investigate the role of Slit-Robo family in GC, we first measured Slit2, 3 and Robo1, 2, 3 expression in 54 paired cancer tissues and matched adjacent non-cancer tissues from GC patients.

Slit1 was excluded for its limited expression in nervous tissues, and
Robo4 was excluded for lacking of Slit binding site. It was found that Slit2 and Robo1 showed the most significant decrease in GC ( Figure 1A,B and Figure S1A). Meanwhile, Slit2 and Robo1 were also decreased in patients of stage III than patients of stage I and II ( Figure  Both mRNA and protein levels of Robo1 in GC cell lines were found to be lower than those determined for GES-1.

| Slit2 inhibits GC cell migration in a Robodependent manner and suppresses EMT
To investigate the role of Slit2-Robo1 signalling in GC progression, we used two independent GC cell lines, MGC-803 and BGC-823 expressing Robo1 receptor for the following studies ( Figure 1H). We  Figure 2J).
Together, these results clearly demonstrated that Slit2 inhibits GC cell migration and invasion in a Robo-dependent manner and suggest that Slit2 signalling suppresses EMT in GC.

| USP33 expression is down-regulated in GC and correlates with Robo1 expression
Our previous studies suggested that USP33 regulates the expression of Robo1 and is therefore essential for the activation of the Slit-Robo pathway. 15,27,28 The observation that Robo1 expression is reduced in GC samples prompted us to examine whether USP33 affects the development of GC.
Analysis of USP33 expression in the public GC datasets showed that USP33 is commonly down-regulated in GC samples  Figure 3A,B). We further examined the USP33 mRNA expression in our own cohort of paired GC samples. Unsurprisingly, USP33 mRNA expression was also found to be lower in GC tissues ( Figure 3C).
Moreover, linear regression analysis revealed that the relative expression levels of USP33 correlated well with Robo1 ( Figure 3D).
USP33 protein levels were also reduced in GC tissues by both immunohistochemistry and Western blot analysis ( Figure 3E-G).
Consistently, both mRNA and protein levels of USP33 were lower To assess the overall frequency of genetic alterations of USP33 in GC patients, we analysed large datasets from cBioPortal for Cancer Genomics (http://cbioportal.org). 29 As shown in Figure 3J, mutations of the USP33 gene in GC patients were detected in five independent cohorts, ranging from 1% to 4.55%, while copy number alteration (CNA) was observed in three cohorts. Interestingly, ten USP33 mutations were identified in GC patient samples, with five mutations inside the catalytic domain of USP33 and two additional frameshift (fs) mutations upstream of the catalytic domain ( Figure 3K).

| USP33 interacts with Robo1 in GC cells
To examine the relationship between USP33 and Robo1, we transfected MGC-803 and BGC-823 with two different siRNAs against USP33 (siUSP33 #1 and #2). Quantitative RT-PCR analysis showed that the USP33 siRNAs reduced mRNA expression of USP33 ( Figure 4A). Robo1 mRNA levels, however, were not affected by USP33 siRNAs ( Figure 4B). In comparison, Western blot showed that USP33 knock-down caused a decrease in both USP33 and Robo1 protein levels ( Figure 4C), suggesting that USP33 regulates Robo1 protein expression through post-translational modification.
We next examined whether USP33 interacts with Robo1 in GC cells by co-immunoprecipitation assay. While the control antibody showed no precipitation, the anti-Robo1 antibody specifically coimmunoprecipitated USP33, indicating that USP33 interacts with Robo1 in MGC-803 cells ( Figure 4D). To test whether Slit2 affects Robo1-USP33 interaction, co-immunoprecipitation experiments were carried out using untreated MGC-803 cells, cells treated with Slit2 containing media or cells transfected with Slit2 plasmid.
Robo1 and USP33 were determined in immunoprecipitated proteins by Western blot. In comparison, neither Slit2 treatment nor expression failed to affect interaction between Robo1 and USP33 (see Figure S2A).

| USP33 deubiquitinates and stabilizes Robo1
We next examined whether USP33 affects the stability of Robo1. The main proteolytic systems responsible for intracellular protein degradation are the ubiquitin-proteasome system (UPS) and the lysosomal system. 30 To examine the role of lysosomes vs the UPS system in Robo1 degradation, MGC-803 cells were treated with chloroquine (CHQ, a lysosome inhibitor) or MG132 (a proteasome inhibitor). The Robo1 protein level increased after MG132 treatment, whereas chloroquine treatment failed to show visible effects ( Figure 4G). These effects were also confirmed in additional 4 GC cell lines (see Figure S2B), suggesting that Robo1 is degraded mainly via the ubiquitin-proteasome system in GC cells.
Furthermore, the decrease in the Robo1 protein level induced by siUSP33 was blocked by MG132 ( Figure 4H). We then examined the levels of ubiquitinylated Robo1 after co-transfecting Flag-tagged ubiquitin (Flag-Ub) together with either control siRNA (Ctrl) or siUSP33. Downregulation of USP33 increased the level of ubiquitinylated Robo1 in the presence of MG132 ( Figure 4I).
Together, these data support the notion that USP33 stabilizes Robo1, preventing it from ubiquitin-proteasome-mediated degradation.

| USP33 mediates Slit2 signalling in inhibiting GC cell migration and EMT process in vitro
To test the involvement of USP33 in Slit2-Robo1 signalling, we performed a wound-healing assay. USP33 expression was reduced in GC cells by lentiviral vector containing a small hairpin sequence targeting USP33 (LV-shUSP33). The efficiency of transfection was confirmed by qRT-PCR and Western blot (see Figure S2C Together, these results clearly demonstrate that USP33 mediates Slit2 signalling in inhibiting GC cell migration and EMT in cultured cells.

| USP33 mediates the inhibitory function of Slit2 signalling on metastasis in vivo
To investigate the role of Slit2 and USP33 in GC metastasis, we used an in vivo xenograft model. Control MGC-803 or BGC-823 cells, cells stably co-expressing Slit2 with shControl or shUSP33 were injected into the caudal veins of athymic BALB/c nude mice (6 mice per group).
Mice were monitored for 6 weeks using an IVIS Imaging system. Six weeks after tumour cell injection, mice were euthanized with the lung tissues harvested for histological examination. The numbers of lung metastatic foci were quantified ( Figure 6C,D), and representative images are shown in Figure 6A Together, these results strongly suggest that USP33 mediates Slit2 signalling in inhibiting GC metastasis in vivo.

| USP33 is required for Slit2-Robo1 signalling in inhibiting TGF-β pathway
To further elucidate the potential pathway regulated by Slit2-Robo1 signalling, a gene set enrichment analysis was performed using Slit2 expression as a phenotype label in GC cohorts from TCGA database.
Higher Slit2 expression was significantly correlated with negative regulation of TGF-β pathway and pathway that restricted Smad protein phosphorylation (P < 0.05; Figure 6E,F). Transforming growth factor-β (TGF-β) is widely upregulated in several human cancers 31 and could promote invasion and metastasis by inducing EMT in cancer cells, 32 while the phosphorylation of Smad proteins plays a key role in the TGF-β pathway. 33 These data suggested that Slit2 may inhibit migration and EMT via inhibiting TGF-β pathway in GC.
We then measured the key proteins involved in TGF-β pathway.
These results indicated that USP33 is required for Slit2-Robo1 signalling in inhibiting TGF-β pathway.

| USP33 expression is inversely correlated with tumour size, lymph node metastasis and neural invasion in GC, and low USP33 expression predicts poor survival
To explore the clinical significance of USP33, we examined the correlation between the USP33 expression and clinicopathological characteristics in our GC cohort. As shown in Table 1, USP33 expression was inversely correlated with tumour size, lymph node metastasis and neural invasion. From TCGA dataset, higher USP33 expression significantly correlates with longer overall survival ( Figure 6H). Furthermore, KM-plotter analysis of additional GC datasets 34 also shows that high USP33 expression was associated with extended patient survival in two independent datasets, GSE62254 and GSE15459 ( Figure 6I,J). Together, these results suggest that USP33 represents a suitable choice as a prognostic marker for GC.

| D ISCUSS I ON
Gastric cancer remains a common cause of tumour-related death and a major health problem in the world, especiallyin Eastern Asia. 1,2 Although great efforts have been made, the mechanisms underlying the tumorigenesis and development of GC remain implicit, which partially accounts for the poor prognosis of GC patients. Epithelial-mesenchymal transition (EMT) has emerged as a critical process of cell invasion and metastasis in most epithelial tumour, including GC, 25,26 and could serve as a potential target for cancer pharmacological intervention. 35 In this study, we first report the involvement of Slit-Robo signalling in the EMT of GC. We also demonstrate the mechanism of how USP33 mediates Slit-Robo signalling in GC. As illustrated in Figure 7, USP33 interacts with Robo1 and stabilizes Robo1, preventing it from ubiquitin-proteasome-mediated degradation.
The first indication that Slit-Robo signalling might play an important role in cancer derived from studies by Sundaresan and colleagues, which identified and cloned the DUTT1 gene (later renamed as ROBO1) and used probes to detect two homozygous deletions at the 3p12 locus in lung and breast carcinomas. 36,37 Subsequent studies have confirmed the involvement of Slit-Robo signalling in several types of cancer. 14,17,[38][39][40] Overwhelming evidence suggests that Slit expression is reduced in different types of cancers. 38,39 However, the role of Slit-Robo signalling in GC remains controversial.
For example, it was reported that POU2F2 promotes GC metastasis through a positive regulation of Robo1, 19 whereas another study showed that down-regulating Slit2 increases growth and motility of GC cells by activating AKT/β-catenin. 17 In our study, we clearly demonstrate the downregulation of Slit2 and Robo1 expression in multiple datasets and our samples at both mRNA and protein levels. Our data indicate that Slit2 inhibits the migration of GC cell in a Robo-dependent manner. This is consistent with our previous studies of lung cancer 16  GAPDH was used as an internal control. All data are shown as mean ± SEM and analysed by Student's t test, *P < 0.05, **P < 0.01, ***P < 0.001 as medulloblastoma 43 and glioma. 44 Moreover, we found that Slit2 inhibits the EMT process, which may support for the clinical application of Slit-Robo signalling.
Several Robo-interacting molecules, such as srGAP, 8 Abl, 9 ERK1/2, 10 USP33 15,22 and Myo9b, 16 have been found to mediate Slit-Robo signalling by different mechanisms. USP33 was initially identified as a substrate molecule which binds to VHL E3 ligase. 21 To date, a considerable number of proteins interacting with USP33 have been identified, including beta-arrestin, 45 hSP56, 46 RALB, 47 ADRB 48 and DIO2. 49 The findings of these studies suggest that USP33 possesses biological functions critical for a wide range of human physiological and pathological processes.
As a Robo1-interacting protein, our previous studies have demonstrated that USP33 regulates the expression of Robo1 15,27,28 and is required for Slit-Robo signalling in modulating axon midline crossing 22 and inhibiting cell migration in breast cancer, 15 colorectal cancer 27 and lung cancer. 28 In this study, the results that Robo1 expression is reduced in GC prompted us to explore whether USP33 affect the Slit-Robo signalling in GC. We also found that knock-down of USP33 reduced the protein level of Robo1, while failed to affect Robo1 mRNA level, suggesting that USP33 regulates Robo1 protein expression through the post-translational modification. Subsequent experiments proved the hypothesis that Robo1 is degraded mainly via the ubiquitin-proteasome system. Transforming growth factor-β (TGF-β) has been proved as a critical factor during malignant progression in many types of cancer; meanwhile, the increased level and tumour-promoting function of TGF-β in gastric cancer have also been reported. 50,51 Furthermore, TGF-β signalling is closely related to EMT and contributes to distant metastatic of tumours 32,53 and Smad protein phosphorylation is a key step during the activation TGF-β signalling. 54,55 In this study, by the gene set enrichment analysis and Western blot of the key proteins, we demonstrated that the inhibitory functions of Slit2-Robo1 on cell migration and EMT are mediated partially by the inactivation of TGF-β signalling and USP33 is required for these effects.
The degradation of many intracellular short-lived proteins relies on the ubiquitin-proteasome system (UPS). 56 The therapy targeting the ubiquitin system has developed into a promising strategy for cancer treatment. 57 Data from our patient samples together with analyses of multiple independent datasets show that higher USP33 expression is significantly associated with longer patient survival, suggesting the potential applications of USP33 for GC therapy and predicting prognosis. Future studies are needed to investigate the potential value of Slit2-Robo1-USP33 in diagnosis and treatment of GC.
In summary, our data reveal the new molecular mechanism of USP33 in GC and Slit2-Robo1-USP33 pathway in suppressing GC cell migration and EMT. In addition, higher USP33 expression is significantly associated with extended patient survival. These results support the suppressive role of USP33 in GC and suggest the potential of USP33 as a prognostic biomarker and therapeutic target for GC.

ACK N OWLED G EM ENTS
This work was partially supported by the National Natural Science RK is supported by National Natural Science Foundation of China (31501133 and 31671452). We would like to thank Torsten Juelich for the professional language editing service.

CO N FLI C T O F I NTE R E S T
The authors have no conflicts of interest to declare.