B4GALNT1 induces angiogenesis, anchorage independence growth and motility, and promotes tumorigenesis in melanoma by induction of ganglioside GM2/GD2

β-1,4-N-Acetyl-Galactosaminyltransferase 1 (B4GALNT1) encodes the key enzyme B4GALNT1 to generate gangliosides GM2/GD2. GM2/GD2 gangliosides are surface glycolipids mainly found on brain neurons as well as peripheral nerves and skin melanocytes and are reported to exacerbate the malignant potential of melanomas. In order to elucidate the mechanism, we performed functional analyses of B4GALNT1-overexpressing cells. We analyzed ganglioside pattern on four melanoma and two neuroblastoma cell lines by high performance liquid chromatography (HPLC). We overexpressed B4GALNT1 in GM2/GD2-negative human melanoma cell line (SH4) and confirmed production of GM2/GD2 by HPLC. They showed higher anchorage independence growth (AIG) in colony formation assay, and exhibited augmented motility. In vitro, cell proliferation was not affected by GM2/GD2 expression. In vivo, GM2/GD2-positive SH4 clones showed significantly higher tumorigenesis in NOD/Scid/IL2Rγ-null mice, and immunostaining of mouse CD31 revealed that GM2/GD2 induced remarkable angiogenesis. No differences were seen in melanoma stem cell and Epithelial-Mesenchymal Transition markers between GM2/GD2-positive and -negative SH4 cells. We therefore concluded that B4GALNT1, and consequently GM2/GD2, enhanced tumorigenesis via induction of angiogenesis, AIG, and cell motility. RNA-Seq suggested periostin as a potential key factor for angiogenesis and AIG. These findings may lead to development of novel therapy for refractory melanoma.

Malignant melanoma is the most common and lethal skin cancer 1,2 . It is a cancer with one of the biggest rise in incidence 3,4 , and the overall 5-year survival rate is less than 10% for patients with stage IV disease 5, 6 . There have been major advances in the treatment of advanced melanoma including Ipilimumab, an antibody to cytotoxic T-lymphocyte-associated-antigen-4 (CTLA-4), and BRAF inhibitor [7][8][9] . However, the anti-CTLA-4 antibody shows benefit in less than 50% of patients 10 . While BRAF inhibitors increased survival compared to other chemotherapies, its indication is limited to about half of patients with BRAF V600 mutations, and almost all patients develop resistance to these inhibitors 11 . While the combination of Nivolumab (monoclonal antibody against programmed death 1, PD-1) and Ipilumumab has demonstrated an impressive 2-year overall survival rate of 63.8% in stage III-IV patients 12 , further improvement of therapy is still needed for the treatment of advanced melanoma patients.
Because gangliosides including GM2/GD2 require stepwise synthesis reactions (Fig. 1A), a model for induced expression of GM2/GD2 on cell surface via overexpression of B4GALNT1 needs the following conditions; 1) both GM3 and GD3 are positive, and 2) both GM2 and GD2 are negative. To evaluate these conditions accurately in the six cell lines, HPLC-based high-specificity analysis of gangliosides was performed (Fig. 1C). Being that SH4 melanoma cell line showed high expression of both GD3 and GM3 (black arrows) and no expression of GD2 and GM2 (white arrows), SH4 fulfilled the aforementioned conditions and was used in the following study. Other results of neuroblastoma cells were shown in Fig. S1. Induction of morphological change, anchorage independence growth, and cell motility. The SH4 clones overexpressing GM2/GD2, #4 and #5, exhibited a distinct morphological appearance compared to SH4 Wild type (WT) or the mock transduced cells. The cells were round and formed aggregation. More than half of them were detached from the bottom of flask, but still capable of survival and proliferation after detachment ( Fig. 2A). No significant difference was seen between the proliferation of GM2/GD2-positive SH4 clones and control (Fig. 2B). A soft agar colony formation assay demonstrated that GM2/GD2-positive SH4 clones formed larger and greater number of colonies than GM2/GD2-negative cells (#4; 86.6 ± 13.9, #5; 82.5 ± 6.5, Mock; 32.7 ± 6.6, #4 vs Mock; p < 0.0001, #5 vs Mock; p < 0.0001, Fig. 3A). There was no significant difference between the two GM2/GD2-positive SH4 clones (#4 vs #5; P = 0.15). In addition, faster wound closure was observed in the GM2/ GD2-positive SH4 clones than the control cells (#4: 49
No evident difference in major cancer stem cell markers. To evaluate the possibility that B4GALNT1 overexpression might enhance tumor incidence via induction of stemness, several melanoma stem cell markers were analyzed in GM2/GD2-positive and -negative cells by flow cytometry. Previous reports indicated that    www.nature.com/scientificreports www.nature.com/scientificreports/ CD133 29-31 , CD166 29 , CD271 32 , Nestin 30 ,and ABCB5 33 are potential melanoma stem cell markers. Aldehyde dehydrogenase (ALDH) activity has also been reported as a potential marker of melanoma stem cell 34 . However, B4GALNT1 overexpression did not induce any meaningful change in any of these markers ( Fig. S3A-F), there was no significant difference in CD133, CD166, and ABCB5 between GM2/GD2-positive and -negative cells.
There was a small decrease in the expression of CD271 and Nestin in the #5 GM2/GD2-positive cell line, but this alone is unlikely to be the cause of enhanced tumor growth. That indicated that B4GALNT1 was unlikely to affect the stemness in the SH4 melanoma cell line.

Promotion of angiogenesis in vivo.
We hypothesized that B4GALNT1 may enhance tumor induction and growth by increasing tumor vascularization. Tumors of similar size from each group were stained by hematoxylin and eosin (H&E) and immunofluorescence for murine-CD31. In H&E staining, the tumors induced by GM2/GD2-positive or -negative SH4 cells did not differ in any characteristics examined: cell shape, number of giant cells, nuclear-to-cytoplasm volume ratio reversal, and hyperchromatism (Fig. 5A). The major blood vessel in the GM2/GD2-positive tumors were more prominent than the one supplying the Mock tumor. Although some tumors showed evidence of local invasion, none of the tumor-bearing mice developed metastasis.
Immunofluorescence staining in tumors derived from GM2/GD2-positive clones by anti-mouse CD31 Ab exhibited many well-structured vessels, while GM2/GD2-negative Mock tumors had much less (Fig. 5B,C). Immunofluorescence performed using selective anti-human-CD31 antibody (non-reactive to mouse CD31) as a negative control did not show any staining of blood vessels (data not shown).

RNA-Seq revealed potential key molecule downstream of B4GALNT1.
To identify the difference of the transcriptional profile between GM2/GD2-positive and -negative cells, we compared them by RNA-Seq and analyzed using edgeR. There were a total of 26,484 genes detected in the individual libraries (Table S1), and 472 genes showed over +/− two-fold change between the two groups (Table S2). The heat map was shown in Fig. S4A. Among the 472 genes, 117 genes were up-regulated and 351 genes were down-regulated by B4GALNT1 overexpression. There was no significant difference between the two Mock clones (Fig. S4B), as well as between two GM2/GD2-positive samples (#4 and #5 clones; Fig. S4C). Genes exhibiting the Top 10 largest fold changes are listed in Table 2. The top up-regulated gene, PTPRD, is a member of the protein tyrosine phosphatase (PTP) family, and is known to be a signaling molecule that regulates a variety of cellular processes including cell growth, differentiation as a tumor suppressor 35 , which was often down-regulated in a variety of tumors. The second highest was B4GALNT1, suggesting that our overexpression of the gene in SH4 had succeeded. The third highest, POSTN (periostin), functions as a ligand for alpha-V/beta-3 and alpha-V/beta-5 integrins to support adhesion and migration [36][37][38] . In addition, it is known to increase angiogenesis. Furthermore, CPVL (Carboxypeptidase Vitellogenic Like), the top down-regulated gene was related to maturation of monocytes into macrophages 39 . Carboxypeptidases are a large class of proteases that act to cleave a single amino acid from the carboxy termini www.nature.com/scientificreports www.nature.com/scientificreports/ of proteins or peptides. The exact function of this protein, however, has not been determined. The second and third downregulated genes were melanoma-associated antigen 12 (MAGEA12 40 ) and chondrosarcoma associated Gene 1 (CSAG1 41 ) which are oncogenes, supposed to be overexpressed in tumors, but in this case they were down-regulated upon B4GALNT1 overexpression.
To further elucidate the role of the 472 genes, we categorized them by function related to cancer malignancy using Ingenuity Pathway Analysis (IPA). This included several genes that functioned as melanoma incidence, proliferation, mobility and colony formation. Some genes were up-or down-regulated as the past reports and others were not (Table 3; "*" means that the fold changes went the opposite direction compared with previous findings). Some of them changed in the opposite direction compared to past reports in melanoma carcinogenesis. These genes may not function as downstream of B4GALNT1, and may have shown a change opposite of other literature due to negative feedback. Pathway analysis was attempted, but no known pathway was found that could fully explain our expression profile suggesting that the revelation of the pathway downstream of glycolipids is not complete.

Discussion
In this study, we analyzed a variety of changes in the SH4 melanoma cell line upon overexpression of GM2/GD2 by transfection of B4GALNT1 gene. One of remarkable findings in vitro is that GM2/GD2-positive SH4 cells showed significant difference of AIG compared to Mock (Fig. 3A), interestingly, Mahata et al. also revealed that GM2/GD2 is associated with AIG by knocking out GM2/GD2 synthase 42 . AIG is often reported as a critical factor for tumorigenesis or exacerbation of malignancy [42][43][44] . Although we initially expected contribution of Epithelial-Mesenchymal Transition (EMT), neither CDH1 (E-cadhelin) nor VIM (vimentin), the major EMT markers, showed change at the mRNA level (data not shown). On the other hand, our RNA-Seq result showed that the expression of POSTN in B4GALNT1-overexpressing cells increased by almost 300 times than Mock cells. Periostin is generally known as a cancer suppresser 45 and it also help to migration in neuronal cell development 46 . While periostin is involved in numerous biological processes, it sometimes contributes to tumorigenesis by promoting cancer cell survival, invasion, and metastasis actively [36][37][38] . It is also known that high expression of periostin protein and/or mRNA is detected in variety of solid tumors 38,47 . Kudo et al. showed that periostin overexpression promoted invasion in head and neck squamous cell carcinoma cells 48 and to explore the genes that are coordinately expressed with periostin, they performed microarray analysis. Among the genes changed in their study, SULF1 was upregulated clearly in our result as well (9.30-fold; Table S2). On top of that, Kotbuki et al. directly revealed that periostin increased cell proliferation and invasion in melanoma cell in vitro and in vivo using overexpression system 49 ,and Fukuda et al. showed that periostin was a key factor in promoting melanoma cell metastasis using shRNA 50 . We therefore speculated that the findings support our conclusion that periostin and its downstream gene overexpression promoted migration induced by GM2/GD2. Furthermore, Bao et al. demonstrated that periostin activated the downstream Akt/PKB pathway via αvβ3 integrin, by in which they observed phosphorylation of Akt1/PKBα on Ser473 to promote cellular survival in colon cancer 51 . Their phosphorylation level, not the total amount, would therefore be contributing to the downstream effect of GM2/GD2. This may explain why our RNA-Seq result did not show a remarkable fold change in Akt/PKB pathway (Table S1).

Symbol
Gene name  www.nature.com/scientificreports www.nature.com/scientificreports/ In our observation, B4GALNT1 overexpression did not affect cell proliferation in vitro, while multiple genes related to tumor cell proliferation-promoting, such as POSTN, IL13RA2, NRP1 were up-regulated, and EMILIN2, which is a proliferation-suppressing gene, was down-regulated notably in mRNA level (Fig. 2B, Table 3). The relationship between ganglioside and cell proliferation is still controversial; some research suggest promotion of proliferation by gangliosides 52-54 , while others show inhibition 55 . These discrepancies may be explained by the difference in the ratio of GD2 + 3 vs GM2 + 3, as well as GD2 vs GD3, which may contribute to differences in cell proliferation. The report by Shibuya et al. which induced GM2/GD2 like ours showed enhanced cell migration 55 , similar to what we observed, supporting that B4GALNT1 and consequently GM2/GD2 intensified migration of SH4 cell (Fig. 3B). Periostin is known to promote motility of several different kinds of cells [56][57][58] . While our IPA analysis did not include the category of melanoma cell migration, RNA-Seq indicated that SDC2 and VCAN might have led to the motility (Table 3).
Another finding was that GM2/GD2 strongly induced angiogenesis. The effect of B4GALNT1 for tumor incidence in vivo was assessed by injecting GM2/GD2-positive SH4 cells into NSG mice. As shown in Fig. 4A www.nature.com/scientificreports www.nature.com/scientificreports/ tumors injected GM2/GD2-positive cells showed a higher tumor establishment rate. This result corresponds with the fact that B4GALNT1 is a clinical marker for advanced melanoma 59 . Liu Y et al. revealed that gangliosides accelerate tumor angiogenesis in murine cells and demonstrated that GM2/GD2-negative cells formed much smaller tumors, using GM3 synthase and GM2 synthase double knockout low ganglioside tumor model 60 . We assessed murine-CD31 expression in the tumors derived from GM2/GD2-positive cell by immunofluorescence staining and observed that B4GALNT1-overexpressing clones induced many CD31 positive endothelial cells and well-developed vessels. In addition, the surface of the GM2/GD2-positive tumors were better vascularized than that of Mock by observation of the recovered tumor with eyes. Tumor progression requires endothelial cells to be activated for the formation of a vascular system. Lang Z et al. found that the enrichment of human umbilical vein endothelial cell (HUVEC) membranes with ganglioside results in amplified VEGF-induced signaling that is important for angiogenesis, and concluded that ganglioside enhances VEGF-induced endothelial cell proliferation 61 . Liu Y et al. reported that reduction of gangliosides depleted vascularization, while addition of wild type gangliosides restored angiogenesis of ganglioside-poor tumor 60 . To clarify how gangliosides induce blood vessels, we assessed the relations of GM2/GD2 and VEGF. Some other reports indicate that ganglioside enhances VEGF and induces endothelial cell proliferation 61,62 . In addition, Liu Y et al. also reported that periostin induces angiogenesis via Erk/VEGF pathway 63 . However, in our result of RNA-Seq and real-time RT-PCR, the expression of VEGF did not show significant correlation with GM2/GD2 level (Table S1, Fig. S5). There is a possibility that interaction between periostin and integrins directly promoted angiogenesis 64 or GM2/GD2 lowered the threshold for cytokine stimulation 60,65 .
While GM2 and GD2 were reported to be increased greatly in cancer stem cells in breast cancer 66,67 , our data ruled out the possibility that GM2/GD2 enhanced tumor incidence via induction of cell stemness. We assessed some melanoma stem cell markers, such as CD133, CD166, CD271, Nestin, ABCB5, and ALDH activity by flow cytometry, and there was no evidence indicative of GM2/GD2 involvement in stemness (Fig. S3A-F).
In summary, our findings demonstrated that in the SH4 melanoma cell line, overexpression of B4GALNT1 as well as its main products GM2/GD2 promotes AIG and cell migration in vitro and enhances tumor incidence by inducing angiogenesis in vivo. To our best knowledge, this is the first time that RNA-Seq was performed to elucidate the influence of B4GALNT1. This result indicates that GM2/GD2 or B4GALNT1 is upstream regulator of periostin, and it might cause some change of characters related to tumorigenesis mentioned above in melanoma cell line. In this study, we have not only shown how GM2/GD2 exacerbates tumors' malignant characters by using B4GALNT1 artificial expression system, but also reconfirmed RNA-Seq is useful tool to find novel potential target in cancer.

Materials and Methods
All experiments were performed in accordance with relevant guidelines and regulations.
Cell lines. Human melanoma cell lines (A375, RPMI-7951, SH4 and WM115) were purchased from American Type Culture Collection (Manassas, USA). Human neuroblastoma cell lines (IMR32 and RTBM1) were provided by Dr. Hajime Hosoi (Kyoto Prefectural University of Medicine, Japan). The melanoma cell lines were maintained in Dulbecco's modified Eagle's high-glucose medium (DMEM, Corning, USA) and neuroblastoma cell lines were maintained in Eagle's minimal essential medium (EMEM, Corning) supplemented with 10% FBS, 100 U/ml penicillin and 10 mg/ml streptomycin (Corning). HUVEC was maintained in Endothelial Cell Growth Media (Sigma-Aldrich, USA). All cells were cultured at 37 °C in a humidified atmosphere of 5% CO 2 .

Construction of a cDNA expression vector, gene transfection and selection. Human B4GALNT1
cDNA was cloned from IMR32 with the primers listed in Table S3. The fragment was first inserted into Topo vector using Zero Blunt TOPO PCR Cloning Kit (Invitrogen, USA). After confirmation of sequence, the cDNA cut out by BamHI and NotI was inserted into the cDNA3.1(+) expression vector (Invitrogen). SH4 cells were plated in a 60-mm plastic plate (Corning) and then transfected with the plasmids by using Superfect (Qiagen, Germany). Stable transfectants were isolated in the presence of 600 μg/ml G418 (Roche, Germany). www.nature.com/scientificreports www.nature.com/scientificreports/ line (1 × 10 6 cells) and digested with recombinant endoglycoceramidase II from Rhodococcus sp. (Takara Bio, Japan). The released oligosaccharides were labeled with 2-aminopyridine and separated using a HPLC system equipped with a fluorescence detector. Normal-phase HPLC was performed on a TSK gel Amide-80 column (Tosoh, Japan). The molecular size of each PA-oligosaccharide is given in glucose units (Gu) based on the elution times of PA-isomaltooligosaccharides. Reversed-phase HPLC was performed on a TSK gel ODS-80Ts column (Tosoh). The retention time of each PA-oligosaccharide is given in glucose units based on the elution times of PA-isomaltooligosaccharides. Thus, a given compound on these two columns provides a unique set of Gu (amide) and Gu (ODS) values, which correspond to coordinates of the 2-D map. PA-oligosaccharides were analyzed using LC⁄ ESI MS⁄ MS. Standard PA-oligosaccharides, PA-GM1 and PA-GD1a, were purchased from Takara Bio and PA-LST-a and PA-SPG were obtained from our previous study 71 .
Real time RT-PCR. Total RNA was extracted from a tumor specimen with RNeasy mini kit (Qiagen) and complementary DNA (cDNA) was synthesized by the use of the SuperScript VILO cDNA Synthesis Kit (Thermo Fisher Scientific) according to the manufacturer's instructions, respectively. The primers used in this experiment are listed in Table S3. Real-time RT-PCR was carried out using LightCycler 480 System (Roche) with SYBR Green (Applied Biosystems, CA) as previously described 72 . Thermal cycling conditions were: initial denaturation for 10 min at 95 °C, and 40 cycles of 15 sec at 95 °C, and 1 min at 60 °C. Data were analyzed with the Light Cycler software.
Anchorage-independent soft agar colony formation assay. Cells were cultured in a two-layer soft agar system 73 . It consisted of a 1% agarose (RPI, USA) underlayer and a 0.7% agarose overlayer containing 1 × 10 4 cells in 60-mm dishes (Corning). Colonies were allowed to form for 2 weeks with fresh media added every 3 days. Plates were stained with crystal violet and colonies more than 0.1 mm in diameter were counted.
Wound-healing assay. Wound-healing assays were carried out as described previously 74,75 . Immediately after scratching (0 h), the plates were photographed and the distance between the edges of the wound area was measured. At 6 h and 24 h after scratching, the plates were photographed and the distance between the edges of the wound region was again measured.
In vivo tumorigenesis. Tumors were induced in 5-6 week old female and male NSG mice (Jackson Laboratory, USA). Each mouse was injected subcutaneously with SH4 cell lines transformed with pcDNA3.1(+) empty vector or the one expressing B4GALNT1 suspended in 0.1 ml of PBS at a single site (2 × 10 6 cell; left, 2 × 10 5 cell; right) to the lower flank. Tumor diameter was monitored every 2-3 days on onset of tumor formation. Mice were sacrificed when the largest tumor size reached 16 mm in diameter along with IACUC approved protocol. At the end point of the experiments, tumors were extracted. At least 4 mice were used in each group (Mock, #4 and #5). The animal experiments were performed in accordance with the institutionally approved animal experimental protocol.
Histopathology and immunohistochemistry. Histological specimens were fixed in 10% formalin and routinely processed for embedding in paraffin. The sections were stained with H&E. Some portions of tumors were embedded in OCT and frozen at −20 °C for immunofluorescence analyses. The tumors were sectioned 30-μm-thick by CM 1800 Cryostat (Leica Biosystems, Germany). The sections were fixed by 50 μl of ice-cold acetone for 5 min. After washing the slides in 1 x PBS, they were incubated in blocking buffer (2% bovine serum albumin (BSA) serum in PBS) for 1 h. The sections were incubated with PE-conjugated anti-mouse CD31 antibody (#102507, BioLegend, 1:500 at 4 °C overnight). After washing the slides in 1 x PBS three times, the slides were incubated by 20 μl VECTASHIELD Antifade Mounting Medium with DAPI (H-1200, Vector Laboratories, USA) and coverslipped. The tissues were observed with automated upright microscope System DM5500 B (Leica Biosystems).
RNA sequencing (whole transcriptome shotgun sequencing, WTSS) and analysis. RNAs derived from SH4 with (#4 and #5) or without (Mock; two samples) B4GALNT1 over-expression were analyzed by RNA sequencing as described in ref. [76][77][78] . Quality tested with Bioanalyzer 2100 (Agilent Technologies, USA). Sequencing was accomplished on the MiSeq 500 (Illumina, USA). 50 bp FastQ paired-end reads (n = 23.6 Million per sample) were trimmed using Trimmomatic (v 0.33). Quality control checks on raw sequence data were performed with FastQC. Read mapping was performed via Hisat2 (2.1.0) using the Human UCSC genome (hg38) as reference. Differentially expressed genes were identified using the edgeR (Bioconductor, www.bioconductor.org) feature in CLCGWB (Qiagen) using raw read counts. The generated list was filtered based on a minimum 2 x absolute fold change and false discovery rate (FDR) corrected p < 0.05. Pathway analysis was performed in IPA (Qiagen) using fold change and FDR corrected values. Statistical analysis. Statistical analysis was performed using the unpaired Student's t-test. A P-value of less than 0.05 was considered statistically significant.