Identifying and targeting determinants of melanoma cellular invasion

Epithelial-to-mesenchymal transition is a critical process that increases the malignant potential of melanoma by facilitating invasion and dissemination of tumor cells. This study identified genes involved in the regulation of cellular invasion and evaluated whether they can be targeted to inhibit melanoma invasion. We identified Peroxidasin (PXDN), Netrin 4 (NTN4) and GLIS Family Zinc Finger 3 (GLIS3) genes consistently elevated in invasive mesenchymal-like melanoma cells. These genes and proteins were highly expressed in metastatic melanoma tumors, and gene silencing led to reduced melanoma invasion in vitro. Furthermore, migration of PXDN, NTN4 or GLIS3 siRNA transfected melanoma cells was inhibited following transplantation into the embryonic chicken neural tube compared to control siRNA transfected melanoma cells. Our study suggests that PXDN, NTN4 and GLIS3 play a functional role in promoting melanoma cellular invasion, and therapeutic approaches directed toward inhibiting the action of these proteins may reduce the incidence or progression of metastasis in melanoma patients.


IntroductIon
The vast majority of deaths from melanoma follow from the acquisition of invasive behaviour and the establishment of metastases [1,2]. While tumor invasion is a critical event in melanoma metastasis, the molecular determinants remain largely unknown. A broader understanding of the mechanism of invasion will generate novel treatment strategies. Cancer cells usually acquire invasive ability through a phenotypic switch called epithelial-to-mesenchymal transition (EMT) [3]. As a consequence of EMT, stationary epithelial cells acquire a mesenchymal phenotype and the ability to migrate and invade during developmental morphogenesis and tumor metastasis [4,5]. Although melanoma is not an epithelial cancer, recent evidence suggests that melanoma cells exhibit many hallmarks of EMT and conform to the EMT dichotomy of epithelial-and mesenchymallike phenotypes [6][7][8][9]. These two distinct melanoma cell phenotypes differ in their in vitro invasive potential and phenotypic switching between these states has been proposed as a mechanism for tumor progression [10].
The EMT program in tumor metastasis includes many morphological and molecular features found in embryonic cells undergoing EMT and invasion [11,12]. Notably, melanoma cells often show high levels of many molecules reminiscent of the EMT programs undertaken by their embryonic precursor; the neural crest cells (NCCs),

Research Paper
a collection of multipotent and migratory cells [13][14][15]. Studies have also revealed that melanoma cells can revert to a neural crest-like state during metastasis [16,17]. Consequently, the embryonic chicken transplantation model has emerged as a powerful in vivo system for assessing the invasive behaviour and plasticity of melanoma cells [8,[14][15][16][18][19][20][21]. It involves injecting melanoma cells into a microenvironment that is populated with neural crest cells that undergo an EMT to exit from the neural tube and undergo extensive migration, eventually populating a diversity of areas in the embryo [22,23]. Melanoma cells transplanted into this model respond to cues within the host embryonic microenvironment, do not form tumors, and subsequently mimic many aspects of neural crest cell motility [17,19]. The embryonic chicken transplantation model has consequently been utilized to investigate the role of candidate genes in motility and pathfinding in vivo by perturbing gene expression with morpholino or siRNA [8,14,15,21].
We propose that genes over-expressed in mesenchymallike melanoma cell lines that exhibit an invasive phenotype are valid targets for blocking invasion in vitro and in vivo. The aim of our study was to identify candidate genes modulating melanoma cellular invasion. We also investigated whether the candidate genes can be targeted in order to impede melanoma invasion. We have previously utilized a microarray screen to identify genetic transcripts that were over-expressed in mesenchymal-like as compared to epithelial-like melanoma cells [8]. In this study we investigated three genes consistently upregulated in mesenchymal-like melanoma cells including Peroxidasin (PXDN), Netrin 4 (NTN4) and GLIS Family Zinc Finger 3 (GLIS3). PXDN, the human homologue of the Drosophila gene Peroxidasin is a cell surface peroxidase associated with the extracellular matrix. NTN4 belongs to a family of secreted extracellular proteins. GLIS3 is a member of the GLI-similar zinc finger protein family and encodes a nuclear protein with five C2H2-type zinc finger domains. These candidate gene expressions were validated in clinical melanoma samples. We applied small interfering RNA (siRNA) approach to examine the silencing effect of candidate genes on melanoma cellular invasion in vitro. Subsequently, we utilized the chicken embryo model to evaluate the efficacy of candidate gene silencing on melanoma motility in vivo. To our knowledge, an in-depth study of the role of PXDN, NTN4 and GLIS3 in melanoma invasion has not been performed previously and these genes/proteins may be potential drug targets to block melanoma invasion.

results transplantation of melanoma cells into the chicken embryo results in the induction of a motile phenotype
We have previously reported the classification of metastatic human melanoma cell lines into epithelial-and mesenchymal-like based on gene expression profiling and functional assays [8]. To compare the motile behaviour of these human metastatic melanoma cell lines, we utilized the in vitro transwell invasion assay and the in vivo embryonic chicken transplantation model. We chose to evaluate ten different melanoma cell lines that were derived from resected melanoma metastases from different locations, as depicted in Table 1. We first evaluated the invasive capabilites of these cell lines using an in vitro transwell invasion assay with reconstituted Matrigel in Boyden chamber inserts. Mesenchymal-like melanoma cell lines LM-MEL-38, -44, -46, -53, and -77 were highly invasive in vitro. In contrast to these cell lines, the epithelial-like melanoma cell lines LM-MEL-28, -34, -42, -62 and -71 lacked invasive ability in vitro (data not shown). Invasive abilities of some of these cell lines have been previously reported [8,21].
To evaluate the behaviours of these cell lines in the embryonic chicken transplantation model, we fluorescently-labelled melanoma cells and cultured them as hanging drops overnight to allow the aggregation of cells. Interestingly, mesenchymal-like melanoma cell lines formed dense clusters that were tightly packed together whereas epithelial-like cell lines formed clusters that were loosely packed. These cells were transplanted into the chick neural tube at the trunk level of the developing embryo. The migratory ability of melanoma cells was assessed 2 days post-injection in wholemount embryos and cross-sections ( Figure 1). Within the embryonic microenvironment, both invasive mesenchymal-like and non-invasive epithelial-like melanoma cells exit from the site of injection in the neural tube, acquire motile abilities and migrate into surrounding tissues ( Figure  1A, 1B). We counted the number of cells that migrated away from the neural tube from both epithelial-like and mesenchymal-like cell lines, and found no difference between these two types of cell lines ( Figure 1C). Neural crest cells migrate from the trunk neural tube and follow two main pathways, one ventrally, beside the neural tube, and the other under the ectoderm [20,24]. Representative cross-section of embryos injected with a mesenchymallike and an epithelial-like melanoma cell line show no differences in the routes followed by the melanoma cells within the embryos, and the routes are largely the same as those followed by neural crest cells; cells were observed beside the neural tube and underneath the ectoderm ( Figure 1D-1F). The timing and the trajectories of melanoma cells do not show apparent differences between epithelial-like and mesenchymal-like melanoma cells. This behaviour is independent of the mutation status of the melanoma cell lines, since the same observation was made using cell lines that were wild-type and cell lines carrying the V600E BRAF mutation (Table 1). Furthermore, no differences in invasive capacities was noted in melanoma cells derived from different metastases, namely, brain, spleen, lymph nodes, rectum, joint, cardiac and skin metastases (Table 1). Taken together, these observations www.impactjournals.com/oncotarget show that melanoma cells can be rewired to acquire a motile phenotype within the chicken embryonic neural microenvironment regardless of their in vitro or in vivo characteristics.

sirnA-mediated targeting of transcription factor Snail inhibited melanoma invasion
We propose that targeted silencing of genes associated with the invasive melanoma phenotype can be a strategy for blocking melanoma metastasis. As a proof of principle, we first inhibited the expression of the zinc finger transcription factors Snail (SNAI1) and Slug (SNAI2). These have been established as crucial regulators of EMT during embryonic development, organ fibrosis and cancer progression, as they are potent repressors of E-cadherin expression and enhancers of migration and invasion [5,25]. In melanoma cells, both these transcription factors have been implicated in inducing an EMT-like process that would be expected to enhance invasion, tumor progression and metastasis in vivo [26][27][28][29][30].
RNA interference in vitro was used to evaluate the role of Snail and Slug in melanoma invasion. siRNA treatment silenced Slug and Snail expression in two invasive mesenchymal-like cell lines, LM-MEL-44 and -53 ( Figure 2A-2B). We tested the impact of siRNA treatment on invasion using an in vitro transwell Matrigel assay.

Identification of candidate genes for targeting invasion in melanoma cells
We identified potential candidate genes whose function are related to melanoma cell invasion from a previously generated whole genome microarray expression profiling study that compared gene expression of epithelial-and mesenchymal-like melanoma cell lines [8]. Among the 221 differentially expressed genes in mesenchymal-like melanoma cells, 65 genes were selected based on their putative role in regulation of cell plasticity and invasion in development or disease, as determined by literature review. These EMT and invasion related genes included extracellular matrix (ECM) components, membrane bound receptors, cytosolic proteins and transcription factors. To validate the expression of the selected candidate genes in melanoma, we compared their expression by qRT-PCR across a subset of nine mesenchymaland nine epithelial-like melanoma cell lines (data not shown).   In the present study, we selected genes which were highly expressed in at least 6 of 9 mesenchymal-like melanoma cell lines and showed no or little expression in epithelial-like melanoma cell lines. Numerous candidate genes such as ECOP1 and PMEPA1 were eliminated as they were also expressed abundantly in some epitheliallike cell lines (Supplementary Figure S1). Three candidate genes emerged that exhibited varying degrees of higher expression in mesenchymal-like melanoma cell lines. These genes are Peroxidasin (PXDN), Netrin 4 (NTN4) and GLIS Family Zinc Finger 3 (GLIS3). The expression of PXDN, NTN4 and GLIS3 are all associated with motility, however these genes have not been previously described as modulators of EMT and invasion in melanoma and have not been evaluated for their potential for contributing to invasion [31][32][33].

sirnA-mediated PXDN, NTN4 and GLIS3 knockdown impaired invasion of melanoma cells in vitro
High expression of PXDN was found in most mesenchymal-like melanoma cell lines compared with epithelial-like melanoma cell lines ( Figure 3A). We observed diminished PXDN mRNA 72 hours after transfection of PXDN specific siRNA but not with a control siRNA in LM-MEL-12, -33, -38 -45, -46, and -77 ( Figure 3B). Immunofluorescent staining showed reduction in PXDN expression 72 hours after transfection of PXDN specific siRNA but not with a control siRNA (Supplementary Figure S2A). We next evaluated effects of PXDN knockdown on cellular invasion. Invasion is the early crucial step of the metastatic cascade and model systems have been developed to recapitulate melanoma cellular invasion [34,35]. Transwell invasion assays using the reconstituted Matrigel in Boyden chamber inserts have been utilized to study melanoma cell invasion in vitro [8,9,27]. Transfection with a control siRNA did not alter the invasive ability of mesenchymal-like melanoma cell lines, while PXDN silencing in these cell lines resulted in a significant reduction of invasion ( Figure 3C-3D).
NTN4 expression was significantly higher in most mesenchymal-like melanoma cells compared with epithelial-like melanoma cells ( Figure 4A). Two different siRNAs were used in order to control nonspecific effects and tested in six different melanoma cell lines. NTN4 was efficiently inhibited at 72 hours after siRNA transfection in all cell lines tested, namely LM-MEL-12, -38, -44, -46, -53, and -77 ( Figure 4B). Reduction in NTN4 expression was also determined by immunofluorescent staining (Supplementary Figure S2B). Silencing NTN4 reduced the invasive ability of mesenchymal-like melanoma cells compared with control siRNA treated cells in vitro ( Figure 4C-4D).
GLIS3 was also found to be upregulated in the majority of mesenchymal-like melanoma cells compared with epithelial-like melanoma cells that expressed very little or no GLIS3 ( Figure 5A). We next decreased GLIS3 expression using two different siRNAs. GLIS3 expression levels in six mesenchymal-like melanoma cells LM-MEL-12, -38, -46, -45, -59 and -77 was effectively downregulated ( Figure 5B). We further validated reduction of GLIS3 protein expression by immunofluorescent staining (Supplementary Figure S2C). We then analyzed the invasive behaviour of melanoma cells after suppression of GLIS3. We determined that the reduction of GLIS3 expression suppressed melanoma cell invasion in vitro ( Figure 5C-5D). All of the examined genes may therefore play an important role in promoting or establishing melanoma cell invasion.

Metastatic melanomas express large amounts of PXdn, ntn4 and GlIs3
Given that PXDN, NTN4 and GLIS3 can regulate invasion of melanoma cells in vitro, we evaluated the in vivo relevance of our data in melanoma tumors. We examined the expression of these genes in a panel of 46 metastatic melanoma patient tumours and detected expression of all three genes in the majority of the samples tested (Supplementary Figure S3A-S3C). We have assessed correlations between the three genes in patient samples and found significant but weak correlation between GLIS3 and PXDN expression, and between GLIS3 and NTN4. No correlation in expression was noted between NTN4 and PXDN (Supplementary Figure S4A-S4C). We then analysed the protein expression patterns of PXDN, NTN4 and GLIS3 in melanoma tumors by immunohistochemical staining of tissue microarrays (TMA) comprising tumors from patients with stage III and IV metastatic melanoma. PXDN expression was detected in 81% (50/62) of metastatic melanoma patient tumors ( Figure 6A-6B). The subcellular location was identified as predominantly membranous. NTN4 expression was detected as both cytoplasmic and membranous in 85% (70/85) of metastatic melanoma tumors ( Figure 6C-6D). GLIS3 was detected as cytoplasmic and nuclear in 76% (38/50) of metastatic melanoma tumors ( Figure 6E-6F). Placenta was used as positive control tissue for PXDN, testis was used as positive control for NTN4 and prostate tissue was used as a positive control for GLIS3 staining (Supplementary Figure S5).

expression of PXdn, ntn4 and GlIs3 is associated with markers of eMt
Cadherin switching is a hallmark of EMT, leading to the down-regulation and replacement of the cell surface adhesion molecule E-cadherin by N-cadherin, enabling motility [3,8]. Since PXDN, NTN4 and GLIS3 are involved in motility and associated with mesenchymal gene signature in melanoma cell lines, we investigated the www.impactjournals.com/oncotarget association of these candidate proteins with markers of EMT in clinical samples.
A cutaneous melanoma dataset consisting of 421 melanoma patients available from The Cancer Genome Atlas (TCGA) (http://www.cbioportal.org) was analysed for association of PXDN, NTN4 and GLIS3 with EMT markers. Large-scale proteomics data from the reverse phase protein array (RPPA) platform are available in the portal for TCGA melanoma studies. For each available protein, the Portal performs a two-sided, two-sample Student's t test to identify differences in protein abundance between tumor samples that have at least one event (alteration) in one of the query genes, and those that do not [36,37]. Analysis revealed that a subset of patients with altered (upregulated) PXDN showed significantly low E-cadherin (p-value: 8.91e-6) and high N-cadherin (p-value: .01) protein expression compared with patients with unaltered PXDN expression (Supplementary Figure  S6). Patients with altered (upregulated) NTN4 showed no difference in E-cadherin (p-value: .5) and N-cadherin (p-value: .1) protein expression (Supplementary Figure S7). In patients with altered (upregulated) GLIS3 significantly low E-cadherin (p-value: 2.53e-6) and high N-cadherin (p-value: .0046) protein expression was detected (Supplementary Figure S8). The significant correlation between PXDN and GLIS3 expression with EMT markers that was observed in our melanoma cell line studies is therefore also confirmed in a large clinical tumor dataset. In TCGA melanoma dataset, mutual exclusivity data from melanoma patients revealed that both PXDN and GLIS3 are co-expressed (Odds ratio = 2.23, p = .010, Fisher's exact test).  Figure 4 and Figure 5 and analysed using ANOVA, with post-hoc Tukey test to identify treatments significantly different from control (**p < .005, ***p < .0005). www.impactjournals.com/oncotarget

Migratory behaviour of melanoma cells in the chicken transplantation model is reduced following knockdown of PXdn, ntn4 or GlIs3
The preceding experiments indicated that PXDN, NTN4 and GLIS3 are highly expressed in metastatic melanoma and play a role in promoting melanoma cell invasion in vitro. To further assess this at a functional level we used the chicken transplantation model to evaluate the roles of these candidate genes. Melanoma cells were transfected with siRNA targeting PXDN, NTN4 or GLIS3, cultured as a hanging drop for 24 hours and introduced into the trunk neural tube of a developing chick embryo.  (b) Melanoma cell lines were transfected with either 10 nM control siRNA or NTN4 specific siRNA and NTN4 qRT-PCR was performed after 72 h (t-test, *p < .05, ***p < .0005, n.s = not significant). (c-d) The in vitro invasive ability of these cells lines was tested using a Matrigel assay. (C) Representative images of invasive cells were taken (scale bar = 100 µm) and (D) average intensities of invasive cells calculated in K counts mm 2 using Odyssey Software. Bars indicate mean +/-SEM of three independent experiments in triplicate. Data was combined with data from same cell lines in Figure 3 and Figure 5 and analysed using ANOVA, with post-hoc Tukey test to identify treatments significantly different from control (*p < .05, **p < .005, ***p < .0005). towards fewer migratory GLIS3 siRNA treated cells in LM-MEL-77 compared to control siRNA treated cells. These observations strongly indicated that PXDN, NTN4 and GLIS3 expression is associated with invasive ability in melanoma cells.

dIscussIon
Despite extensive studies on melanoma, the molecular mechanisms of melanoma tumor invasion are still not fully understood. Through a knockdown approach and in vivo motility model, we demonstrate 3 novel molecular players PXDN, NTN4 and GLIS3 as critical factors of melanoma invasion. Our results show that PXDN, NTN4 and GLIS3 are overexpressed in metastatic melanoma clinical specimens. These genes promote motility, a crucial cellular feature associated with metastatic disease, thus implicating these gene products as potential drug targets to block melanoma cell motility.
Metastasis is a complex multi-step event and invasion of a tumor through basement membrane into surrounding tissue is the first crucial step of the metastasis process [4,38]. We and others have previously shown that in melanoma, invasion and proliferation are uncoupled, such that highly proliferative melanoma cells are less likely to be invasive, and vice versa [8,10,39], but the properties are not mutually exclusive, and invading cells undergo proliferation [40]. The transition between each state is referred to as phenotype switching, which is similar to the EMT process. Many studies have shown that melanoma cells with mesenchymal traits such as a heightened ability to invade (as detected by in vitro Matrigel coated Boyden based invasion chamber assays) possessed increased tumor metastasis in vivo in mouse  Figure 3 and Figure 4 and analysed using ANOVA, with post-hoc Tukey test to identify treatments significantly different from control (**p < .005, ***p < .0005). www.impactjournals.com/oncotarget models of melanoma [27,30,34,41]. Furthermore, microarray profiling of the melanoma tumor derived from patients reveals characteristic invasive and mesenchymal expression signatures associated with metastatic outcome [42]. Such observations are consistent with a crucial role for invasion in melanoma metastasis. In this study we have identified 3 candidate genes that functionally correlated with heightened invasiveness in mesenchymallike melanoma cells. Knocking down expression of each of these genes reduced cell motility in vitro and in vivo; therefore further investigation of the role of these genes in cell invasion and other aggressive melanoma traits such as drug resistance, adhesion and proliferation is warranted.
Studies using gene expression arrays have revealed coordinated upregulation of numerous genes involved in cell motility in mesenchymal-like melanoma cell lines. This coordinated upregulation of genes probably reflects the activation of EMT transcriptional programs that promote cell motility by either environmental cues such as matrix components present in mesenchymal-like melanoma cells or by changes in some master regulatory genes [8,42]. The invasive behaviour of melanoma cells has been mainly attributed to their origin from the embryonic neural crest cells [17]. Hence, the embryonic microenvironment populated with neural crest cells is considered to be an ideal and clinically relevant micro-compartment for in vivo studies of melanoma cells [16,[43][44][45].
The chicken transplantation model has generated new ways to understand the basic biology of melanoma motility and to test strategies that aim to block the spread of melanoma [8,14,15,20,21]. Tumor cells are either intrinsically able to migrate (mesenchymal cells) or acquire this capability during tumor progression as part of promigratory programs such as EMT, often in response to microenvironmental cues [46]. When transplanted into the chicken embryonic neural crest microenvironment, melanoma cells respond to cues from surrounding host tissues and follow migratory routes travelled by NCCs [16]. We demonstrate that 10 different melanoma cell lines regardless of their site of metastasis, BRAF mutation status or invasive behaviour in vitro, gain motility when transplanted into the trunk chicken microenvironment, reflecting the underlying plasticity of these cell types.
Interestingly, we observed that mesenchymal-like melanoma cell lines formed smoother and more tightly packed clusters when cultured as a hanging drop compared to epithelial-like cell lines. We have previously shown that culturing both epithelial-and mesenchymal-like melanoma cell lines as hanging drops does not promote an EMT [20]. However, this difference in appearance between the cell lines in hanging drops was not associated with differences in migration, as melanoma cells from both epithelial and mesenchymal-like lines migrated following transplantation into chick trunk neural tube. It is likely that the melanoma cells exhibit motility in vivo by undergoing an EMT process in response to the chick microenvironmental cues. This is in accordance with a recent study demonstrating that 13 known neural crest genes associated with EMT and cell migration were induced in melanoma cells following exposure to the neural crest microenvironment [15,47]. Assessing changes in EMT traits after harvesting melanoma cells exposed to chick microenvironment warrants in-depth investigation.
Melanomas modify their microenvironment through secretion of specific extracellular matrix proteins and extracellular matrix modifying proteases to promote tumor growth and metastasis [48]. Alterations in extracellular matrix remodelling have been shown to be an important event during EMT [46]. Two of the candidate genes identified in this study are extracellular matrix proteins, namely PXDN and NTN4. PXDN is a unique member of the heme-containing peroxidases enzyme family that was initially identified to play a role in embryonic development in Drosophila melanogaster [49]. PXDN is involved in the formation and stabilization of the ECM in a way that does not seem to be mediated by its peroxidase activity [50]. In addition to melanoma, PXDN was also detected in tumors such as breast cancer, colon cancer, glioblastoma and ovarian cancer [51,52]. A gene-set enrichment analysis (GSEA) of our previous study of mesenchymal melanoma cell lines [8] revealed association of PXDN with breast cancer [53] and colon cancer [54]. PXDN, also known as Melanoma-associated gene-50 (MG50), encodes a new melanoma antigen containing at least six naturally expressed melanoma HLA class 1-restricted epitopes that are recognized by human cytolytic T lymphocytes, further indicating its potential clinical applications [51]. Although PXDN was expressed in melanoma samples its physiological role remains largely unexplored. Silencing PXDN in mesenchymal melanoma cells blocked cellular invasion and this observation is consistent with a previous study showing that loss of PXDN in BeWo choriocarcinoma cells blocked cellular invasion and adhesion [32]. Transforming growth factor -β1 (TGF-β1) treatment induced the production of PXDN by differentiating myofibroblasts undergoing an EMT during tissue fibrosis [50]. TGF-β1 induced EMT in melanoma by modulating ECM molecules and it is likely that PXDN represents an ECM molecule in melanoma mediating an EMT in a TGF-β1-dependent manner [8,21]. Since PXDN is a surface protein accessible to extracellular pharmaceutical compounds, it may prove to be a novel target for abrogating tumor invasion in melanoma.
NTN4 is a member of the netrin family that are secreted chemotropic guidance cues with roles in embryogenesis and tumor development [55,56]. NTN4 is emerging as an important regulator of epithelial and endothelial migration, adhesion, proliferation and apoptosis [57,58]. In cancers, the expression and functional role of NTN4 has been controversial. Recent studies have suggested a role for NTN4 in regulating metastasis, angiogenesis and tumor growth [31,58]. However, other studies examining expression of NTN4 in clinical samples have reported that it is markedly down-regulated in prostate, breast and cervical cancers [59][60][61]. The variable expression patterns of NTN4 in different cancers could depend on many factors including the receptors available on the responsive cells, multiple signaling pathways activated and the concentration of protein [56,58]. NTN4 functions as an ECM molecule, is a component of basement membranes and regulates cell interaction with the ECM [62]. NTN4 has been implicated in activating multiple oncogenic pathways in gastric cancer, namely Jak/Stat, PI3K/Akt and ERK/MAPK [63], and it interacts with integrin β4 to promote glioblastoma cell proliferation via the AKT-mTor pathway [56]. Our previous GSEA of melanoma cell lines revealed that NTN4 was downregulated in osteosarcoma cells following knockdown of HDAC2 [64], suggesting that it is epigenetically regulated in cancer cells. Our data suggest a role for NTN4 in regulating melanoma invasion, and NTN4 being an ECM protein, peptides or antibodies that inhibit or limit its action might be among the strategies to control or limit melanoma metastasis.
Gli-similar (GLIS) 1-3 proteins constitute a subfamily of Krüppel-like zinc-finger transcriptional regulators that play a critical role in embryonic development and in several pathologies including kidney fibrosis [65]. During embryonic development GLIS3 regulates neurulation, is prominently expressed in the dorsal neural tube and is implicated in cellular motility [33]. GLIS3 has been reported to be highly expressed in ependymomas and glioblastomas, suggesting a role for GLIS3 in tumorigenesis [66,67]. GLIS3 interacts with the transcriptional coactivator with PDZ-binding motif (TAZ) which has been implicated in EMT [68,69]. GLIS3 shares a great degree of homology with members of the Gli family and provides a mechanism for crosstalk between Gli and Glis signaling pathways.
Recently, GLI2, a TGF-β target has been implicated in increased melanoma invasion and metastatic capacity [41]. Our previous GSEA of melanoma cell lines revealed association of GLIS3 with breast cancer [53] (the same gene set that showed upregulation of PXDN), supporting its role in multiple tumour types. Elucidation of GLIS3 and its interaction partners in melanoma may provide new opportunities for the development of therapeutic strategies in the treatment of this disease.
Our study suggests that Snail (SNAI1) rather than Slug (SNAI2) plays a role during later stages of melanoma invasion and metastasis. This is consistent with reports demonstrating a role for Slug early during melanoma development, rather than during later stages of metastasis [9,26,30,39]. This is in contrast to other studies implicating Slug as pro-invasive gene responsible for the metastatic behaviour of melanomas [28,29,70]. Increase in the mesenchymal phenotype and metastatic ability of melanomas with Snail overexpression has been demonstrated [71,72]. Both Snail and Slug are crucial for the neural crest development and functional redundancies between the family members have been well documented [22,73,74].
Although this study shows an association between PXDN, NTN4, GLIS3 and SNAI1 and melanoma invasion, their roles in melanoma are most likely multifaceted. We observed some correlation in gene expression in patient samples between GLIS3 and PXDN, and GLIS3 and NTN4. How these proteins are regulated and how they act in metastasis is not yet clear. Thus further understanding of the molecular crosstalks between these proteins and EMT-related signaling pathways may also provide effective approaches to targeting a broader signaling pathway underlying melanoma plasticity. www.impactjournals.com/oncotarget

MAterIAls And Methods cell culture and melanoma patient samples
Melanoma cell lines were established from resected melanoma metastases by mechanical dissociation of tissue with subsequent overnight digestion in media containing collagenase IV at 37°C. Established cell lines were Mycoplasma-tested using the MycoAlert test (Lonza Rockland, Inc., USA). All tissue donors provided written informed consent for tissue collection and research, which was covered by protocols approved by the Austin Health Human Research Ethics Committee, Melbourne, Australia (approval number H2012/04446). All cell lines were matched with their donors by HLA-typing and STR-profiling. Cells were cultured in RPMI1640 supplemented with 10% fetal calf serum (FCS) as described previously [75].

Invasion assays
Invasion assays were performed in Boyden chamber inserts with Matrigel coating (Becton, Dickinson and Company, USA). Insert membranes were stained with 4ʹ, 6-diamidino-2-phenylindole (DAPI) or a 0.1% crystal violet solution (Sigma, USA). Cells were photographed with a monochromatic Olympus camera. Analysis of intensity of transwell membrane calculated in K counts mm 2 with Odyssey LI-COR Scanner System (LI-COR Biosciences, USA) was utilized as a measure of invasion.

Immunohistochemistry and pathological evaluation
Paraffin embedded tissue slides were deparaffinised and rehydrated, endogenous peroxidise activity was blocked with 3% Hydrogen peroxide, antigen retrieval was performed in 10 mmol/L citrate buffer, and nonspecific binding was blocked with blocking reagent. Anti-PXDN antibody (NBP1-84316, Novus Biologicals, USA), anti-NTN4 antibody (NBP2-13680, Novus Biologicals, USA) and anti-GLIS3 antibody (NBP2-16668, Novus Biologicals, USA) was applied at 5 µg/mL, 1 µg/mL and 8 µg/mL concentrations, respectively and incubated overnight at 4°C, followed by 60 minute incubation with secondary anti-mouse antibody HRP (Dako). The chromogen used was 3-amino-9-ethylcarbazole (AEC). Human placenta was used as the positive control for PXDN and NTN4. Human prostate tissue was used as positive control for GLIS3. Negative control for which the primary antibody was substituted with the same concentration of mouse IgG was also used for all antibodies tested. Slides were scanned using a ScanScope XT (Aperio) and immunohistochemical reactivity was evaluated by two independent investigators.

In vivo chick embryo model
Melanoma cells were treated with PXDN, NTN4 or GLIS3-specific siRNAs or scrambled control siRNA as described above and labelled with CM-DiO or Dil as per manufacturer's instructions (Invitrogen, USA). Cells were grown overnight in a hanging-drop fashion to allow the formation of aggregates. Fertile chicken eggs were incubated at 38°C for 48 hours prior to transplantation. Cell aggregates consisting of similar sized small aggregates from approximately 500-800 cells were harvested and carefully injected with a glass pipette into the trunk neural tube lumen of developing chicken embryos. The eggs were then sealed with adhesive tape and re-incubated for 2 days. After incubation, embryos were removed from the eggs and fixed with 4% paraformaldehyde and whole mounts were analyzed for the localization of melanoma cells using Lumar V12 Zeiss microscope as described previously [20]. Segments of embryos containing the melanoma cells were then washed in PBS, incubated overnight in 30% sucrose at 4°C, frozen and 20 μm transverse sections cut using a cryostat. Sections were rinsed in PBS, mounted in fluorescent mounting media and imaged using a Zeiss Axio M1 microscope and AxioVision software.

statistical analysis
Statistical comparisons of data were performed using Student's t-test or ANOVA in Prism software version 5.00 (GraphPad Software Inc). Comparisons of pooled mesenchymal cell lines versus epithelial cell lines and gene expression differences between control siRNA treated and knockdown treated cells were analysed using Student's two-tailed t-test.
Comparisons between the invasiveness of siRNA treated cells in vitro and in vivo were tested using ANOVAs, with Tukey's post-hoc test. For grouping of data, some ANOVAs were run with data pooled from multiple graphs, this is indicated in the relevant figure legends. The significance threshold was set at 0.05.