Oncogenic function of the homeobox A13-long noncoding RNA HOTTIP-insulin growth factor-binding protein 3 axis in human gastric cancer

To study the mechanisms of gastric tumorigenesis, we have established CSN cell line from human normal gastric mucosa, and CS12, a tumorigenic and invasive gastric cancer cell line from CSN passages. Many stem cell markers were expressed in both CSN and CS12 cells, but LGR5 and NANOG were expressed only in CS12 cells. Increased expression of homeobox A13 (HoxA13) and its downstream cascades was significant for the tumorigenic activity of CS12 cells, and was associated with recruitment of E2F-1 to HoxA13 promoter accompanied with increased trimethylation of histone H3 lysine 4 (H3K4me3) at the hypomethylated E2F motifs. Knockdown of HoxA13 caused the downregulation of long non-coding RNA HOTTIP and insulin growth factor-binding protein 3 (IGFBP-3) genes, indicating that both were targets of HoxA13. Concurrent regulation of HoxA13-HOTTIP was mediated by the mixed lineage leukemia-WD repeat domain 5 complex, which caused the trimethylation of H3K4 and then stimulated cell proliferation. HoxA13 transactivated the IGFBP-3 promoter through the HOX-binding site. Activation of IGFBP-3 stimulated the oncogenic potential and invasion activity. Increased expression of HoxA13 (63.2%) and IGFBP-3 (28.6%) was detected in human gastric cancer tissues and was found in the gastric cancer data of The Cancer Genome Atlas. Taken together, the HoxA13–HOTTIP–IGFBP-3 cascade is critical for the carcinogenic characteristics of CS12 cells.


INTRODUCTION
Homeobox genes are known as the transcriptional regulators of mammalian embryogenic development and are deregulated in tumorigenesis. There are few known direct targets of Hox proteins, and their mechanism of regulation is incompletely understood. The homeobox A13 (HoxA13) gene is the most posterior of the HOX clusters in 7p15.2. This gene is expressed in the genital tubercle during embryogenesis [1,2] and plays an essential role in skeletogenesis, interdigital programmed cell death, and cell sorting of autopod formation. The loss of HoxA13 function in mice causes missing phalanx elements and affects the carpal and tarsal regions [3]. In humans, mutations in HoxA13 are associated with dominantly inherited hand-foot-genital syndrome (HFGS; OMIM #140000) [4,5] and Guttmacher syndrome (GS; OMIM #176305), which include limb and genitourinary abnormalities [6,7]. Similar malformations have also been observed in the spontaneous mouse mutants, hypodactyly [8], and in engineered HoxA13 -null mouse models [9,10]. HoxA13 is essential for placental vascular patterning and labyrinth endothelial specification through direct regulation of tyrosine kinase with immunoglobulinlike and epidermal growth factor-like domain 1 and forkhead box F1 [11].
The role of HoxA13 in cancer progression has been reported in hepatocarcinogenesis [12], especially in the liver stem-like cell lines [13], and in prostatic neoplasia [14], leukemogenesis [15], and esophageal squamous cell carcinoma [16]. HoxA13 is a prognostic marker of the aggressive phenotype of gastric cancer [17]. However, the mechanism underlying HoxA13-mediated gastric carcinogenesis and progression of gastric cancer is unclear.
Long noncoding RNAs (lncRNAs) that do not encode proteins are defined as transcripts containing > 200 nucleotides. lncRNAs account for more than 90% of the transcriptome and are typically transcribed by RNA polymerase II. They play an essential role in the control of gene expression involved in various physiological processes, including development, differentiation, and metabolism [18]. HOTTIP lncRNA is located at the 5′-end of the HoxA cluster and is associated with the polycomb repressive complex 2 (PRC2) and WD repeat domain 5 (WDR5) [19]. The interaction between HOTTIP and the WDR5-mixed lineage leukemia (MLL) complex increases histone H3 lysine 4 trimethylation and activates the expression of multiple 5′-HoxA genes [19]. Recent reports have shown that HOTTIP is associated with cancer metastasis and is a negative prognostic factor in patients with liver and tongue cancer [20,21]. In addition, HOTTIP expression promotes cancer progression and drug resistance by regulating HoxA13 in pancreatic cancer [22]. Another study shows that HOTTIP increases pancreatic cancer cell proliferation, survival, and migration through HoxA family genes other than HoxA13 [23].
The insulin-like growth factor-binding protein-3 (IGFBP-3) influences several molecular mechanisms or signaling pathways that determine cell death or survival, particularly in the context of cancer. Whereas the biological activity of IGFBP-3 is attributed in part to its ability to bind and neutralize insulin-like growth factors (IGF), thereby inhibiting IGF receptor (IGFR) activation, there is other evidence that IGFBP-3 also has intrinsic IGF-or IGF1Rindependent effects that influence cell fate. IGFBP-3 inhibits cell growth and apoptosis in some circumstances but stimulates cell growth and survival in others [24][25][26]. IGFBP-3 is known to bind nuclear receptors of retinoic acid, vitamin D, peroxisome proliferatoractivated receptor γ, nuclear hormone receptor 77, and epidermal growth factor receptors as well as the protein kinase catalytic subunits of DNA repair enzymes [25]. IGFBP-3 is known as a transcriptional target of the tumor suppressor protein p53, which modulates IGFBP-3 [26,27]. However, the relationship between HoxA13 and IGFBP-3 remains elusive.
The progression of gastric cancer is recognized as a multistep process that involves the activation of oncogenes and inactivation of tumor suppressor genes [28,29]. We have previously established a nonmalignant gastric cell line, CSN, from the stomach mucosa of a patient with mild gastritis, which exhibits features of stem/progenitor cells [30]. After a prolong expansion of CSN cells, a tumorigenic subline CS12 was generated, which exhibited anchorage-independent growth, xenograft tumor formation in nude mice, duplication of the short arm of chromosome 7 (7p15.1-15.3 and 7p22. 1-22.3) on chromosome 12, and increased expression of HoxA cluster genes when compared with the nontumorigenic CSN cells [31]. Thus, the increased expression of HoxA genes may contribute to gastric tumorigenesis. Here, we examined the role of HoxA13 in contributing to the cancerous characteristics of CS12 cells and identified the HoxA13-HOTTIP-IGFBP-3 axis as the underlying mechanism.
site. Activation of IGFBP-3 stimulated the oncogenic potential and invasion activity. Increased expression of HoxA13 (63.2%) and IGFBP-3 (28.6%) was detected in human gastric cancer tissues and was found in the gastric cancer data of The Cancer Genome Atlas. Taken together, the HoxA13-HOTTIP-IGFBP-3 cascade is critical for the carcinogenic characteristics of CS12 cells.

CS12 cells exhibited more aggressive cancerous features than CSN cells
To characterize the cancerous features of CS12 cells in vitro, cell growth, colony formation, cell motility, and chemoresistance between CSN and CS12 cells, were compared. A trypan blue dye exclusion assay showed that CS12 cells proliferated more rapidly than CSN cells ( Figure 1A). A colony-formation assay showed that CS12 (5 × 10 4 cells) generated about 200 colonies, but CSN cells did not produced any colonies ( Figure 1B). Cellcycle analysis showed that 27%-32% of CS12 cells were in S-phase, whereas only 18%-22% of CSN cells were in S-phase ( Figure 1C). All data were consistent with the more proliferative nature of CS12 cells. A Transwell invasion assay demonstrated that both the invasion and the migration efficiencies were 1.4-fold higher in CS12 cells than in CSN cells ( Figure 1D), which showed that CS12 exhibited increased migration and invasion activities. These findings were supported by the elevated expression of matrix metalloproteinase (MMP) and epithelial to mesenchymal transition (EMT) genes, such as Snail and Zeb 1 in CS12, but not the expression of Twist (Supplementary Figure 1A). Their sensitivity to 5-fluorouracil, a common anticancer drug used for treatment of gastric cancer [32], was examined. The results showed that CS12 were more resistant to 5-fluorouracil than CSN cells ( Figure 1E). Their in vivo tumorigenicity was examined using a xenograft transplantation test and only CS12 cells generated tumors in SCID mice ( Figure 1F) consistent with previous findings [30,31]. The tumor showed 10% malignant cells that contained little cytoplasm (Supplementary Figure 1B). Taken together, these data indicate that CS12 cells exhibit more cancerous characteristics than CSN cells.

Differential expression of stemness genes and pluripotency in CSN and CS12 cells
Acquired chemoresistance and EMT are wellknown as hallmarks of cancer stem cell-like cells [32,33]. Both CS12 and CSN cells express OCT4 stemness genes [30,31]. Here, we further characterized whether CS12 possessed typical stem cell features. Immunofluorescence analysis showed the expression of stemness markers including OCT4, SOX2, KLF4, SSEA-3, SSEA-4, TRA-1-60, and TRA-1-81 in both CSN and CS12 cells, but LGR5 and NANOG were only detected in CS12 cells (Supplementary Figure 2A, 2B and 2C). Semiquantitative reverse transcription-polymerase chain reaction (RT-PCR) analysis showed that the transcript levels of SOX2, NANOG, hTERT, and REX1 were significantly higher in CS12 compared with CSN cells (Supplementary Figure 2D). These data suggest that CS12 cells exhibit more stem-like cell characteristics.
We next examined the pluripotency of CS12 cells by inducing teratoma formation. CS12 differentiated into cells including osteoblasts, muscle cells, and megakaryocytes that were derived from two germ layers (Supplementary Figure 2E). Neither CSN nor CS12 cells stained positively for alkaline phosphatase (data not shown), suggesting that CSN and CS12 cells were not pluripotent stem cells.

Upregulation of HoxA13 was critical for the tumorigenic properties of CS12 cells
We previously reported that CS12 cells exhibited duplicated chromosome 7 short arm where HoxA genes reside [4]. Both the qPCR and western blot data showed that HoxA13 was 4.7-to 15-fold upregulated in CS12 than in CSN cells (Figure 2A and 2B). These results suggested that HoxA13 was upregulated through mechanisms other than simply duplicated gene dosage.
Because of the oncogenic role of HoxA13, we suspect that high expression of HoxA13 may contribute to gastric tumorigenesis of CS12 from CSN cells. To test this hypothesis, HoxA13 expression was knocked down in CS12 cells and then the cell growth, cell mobility, colony formation, and tumor formation in SCID mice were examined. Knockdown by HoxA13 siRNA significantly reduced HoxA13 expression at protein ( Figure 3A) and mRNA ( Figure 3B) levels, but the scrambled siRNA and off-target C-Jun siRNA did not affect the expression of HoxA13. After knockdown of HoxA13 expression in CS12 cells, the cell proliferation activity ( Figure 3C) and colony formation ability ( Figure 3D) were reduced by about 50%. The migration and invasion activities of HoxA13-knockdown CS12 cells were decreased to around 30% of controls ( Figure 3E and 3F). Knockdown of HoxA13 expression induced by shHoxA13 lentivirus also significantly impaired tumor formation ability of CS12 cells in SCID mice ( Figure 3G and 3H). Hematoxylin and eosin staining of tumor sections demonstrated that tumor regions shrank by knockdown of HOXA13 ( Figure 3H). These results indicate that elevated HoxA13 expression is critical for the cancerous features of CS12 cells.

Hypomethylation of the HoxA13 promoter at the p53/E2F-binding site in CS 12 cells
To investigate the mechanism underlying HoxA13 overexpression in CS12 cells, we first examined the DNA methylation of HoxA13 promoter by sodium bisulfite conversion followed by pyrosequencing analysis. The results showed a decreased DNA methylation (ratio; 0.1 vs 0.32) at the CpG position 191 of HoxA13 promoter in CS12 when compared with CSN cells ( Figure 4A). Notably, the CpG position 191 is a composite p53/E2Fbinding site (27,200,830; designated as the E1 site) ( Figure 4B). To examine the differential binding of p53 and E2F1 at this position in CS12 and CSN cells, chromatin immunoprecipitation (ChIP) assays were www.impactjournals.com/oncotarget The drug resistance capacity of CS12 cells was measured as the survival rate of cells when exposed to drug such as 5FU. (F) Tumor sizes with time in SCID mice subcutaneously injected with CS12 cells and CSN normal cells. CS12 cells, not CSN cells, form tumors. Data in A-E were derived from five independent experiments and are presented as mean ± SEM (two-tailed Student t test; * p < 0.05; ** p < 0.01).
Oncotarget 36053 www.impactjournals.com/oncotarget conducted. The results showed that recruitment of p53 to the E1 site of HoxA13 promoter was decreased in CS12 cells, whereas the interaction between E2F1 and E1 site was increased ( Figure 4C). The recruitment of RB1 was similar in CSN and CS12 cells and the nonspecific (NS) site in the HOXA13 promoter did not recruit p53 or E2F1. Thus, the recruitment of E2F-1 to the E1 site is critical for the activation of HoxA13 promoter in CS12 cells. In attempt to confirm this observation further, we performed the forced expression of E2F-1 to observe the enhanced expression of HoxA13 promoter. We generated the HoxA13 promoter-luciferase constructs of wild type-(WT-) and its E1 mutant (mE1)-promoter luciferase and examined the effects of E2F-1. The expression of WT-HoxA13 promoter was more greatly increased by overexpression of E2F-1 than that of mE1 mutant promoter in CS12 cells ( Figure 4D) and CSN cells (Supplementary Figure 3A). Thus, a decrease of p53 binding at the E1 site by ChIP assay was consistent with the reduction of p53 and p21 expression in CS12 cells (Supplementary Figure 4A and 4B). By contrast, further addition of p53 significantly repressed the expression of WT-HoxA13 promoter in CSN cells (Supplementary Figure 3B) and the transactivation activity of p53 and the p53-regulated p21 Cip1 promoter activity were also decreased in CS12 when compared with those in CSN cells (Supplementary Figure 4C and 4D). These results suggest that expression of E2F-1 was increased and expression of p53 was decreased in CS12 cells as compared with CSN cells. These results were consistent with the increase of S phase in CS12 compared with CSN cells ( Figure 1C).

Differential recruitment of DNA/histone methyltransferases and altered histone modification at the E1 site of HoxA13 promoter
Because the hypomethylation at the E1 site of HoxA13 promoter was found in CS12 cells, we next conducted ChIP assays to compare the recruitment of DNA methyltransferases to the E1 site in CS12 and CSN cells. The results showed that the recruitment of DNMT1 and DNMT3b to the E1 site was one-half to one-third lower in CS12 than in CSN cells ( Figure 4E), supporting the observed hypomethylation of HoxA13 promoter even though the expression of DNMT1 and DNMT3b was 1.8-fold higher in CS12 cells than in CSN cells (Supplementary Figure 5).   Figure 1B. The effects of siRNA-HoxA13 on migration (E) and invasion (F) activities were assessed as described in Figure 1D.  Oncotarget 36056 www.impactjournals.com/oncotarget Next, we examined the recruitment of MLL1 and WDR5 and the methylation of histone H3 at the E1 site in CS12 and CSN cells because the previous study showed that MLL1/WDR5 complex mediates trimethylation of histone H3 lysine 4 (H3K4me3) at the 5′-HoxA cluster and activates HOXA gene expression. The results showed that MLL1 and WDR5 densely occupied at the E1 site but not at the NS site, which was coincident with increased H3K4me3 and HoxA13 expression in CS12 but not in CSN cells (Figures 2 and 4E).

The recruitment of lncRNA HOTTIP was involved in the upregulation of HoxA13 in CS12 cells
The recruitment of the WDR5-MLL complex to the 5′-end HOXA cluster is mediated by the lncRNA HOTTIP [19,20]. The q-PCR results showed that the expression level of HOTTIP was 17.8-25.1-fold higher in CS12 cells than in CSN cells ( Figure 5A). However, the expression level of HOTARMI, the lncRNA resides on the 3′-end HoxA cluster, was similar between both cells. Interestingly, another lncRNA H19 was also highly expressed in CS12 cells.
To confirm the involvement of HOTTIP in the upregulation of HoxA13 in CS12 cells, siRNA against HOTTIP was introduced to reduce the HOTTIP, but not the H19 RNA levels ( Figure 5B). Upon HOTTIP knockdown, the expression of 5′-end HOXA genes including HoxA13 were reduced in CS12 cells when compared with the effect of control siRNA ( Figure 5C). ChIP analyses showed that both the recruitment of WDR5 and MLL1 and the level of H3K4me3 at the E1 site, but not at the NS site, were decreased in CS12 cells after HOTTI knockdown ( Figure  5D). In addition, the recruitment of DNMT3b, but not DNMT1, was restored by HOTTIP knockdown at the E1 site. Thus, HOTTIP might affect the recruitment of DNA methyltransferase DNMT3b but not DNMT1. These results demonstrated that HOTTIP was involved in the upregulation of HoxA13 in CS12 cells.

IGFBP-3 was a HoxA13 downstream target and was important for the cancerous features of CS12 cells
To investigate the mechanism underlying HoxA13 downstream genes including those for Annexin A2 (ANXA2) [34,35] and IGFBP-3. siRNA against HoxA13 decreased the expression of HOTTIP, H19, IGFBP-3, and ANXA2 in CS12 cells ( Figure 6A). The endogenous expression of IGFBP-3 was about two-to threefold higher in CS12 than in CSN cells ( Figure 6B). To verify the effect of HoxA13 on IGFBP-3 expression, IGFBP-3 promoter was cloned to a luciferase reporter, and then cotransfected with a HoxA13 expressing construct or vector control into CSN cells. The results showed that ectopic HoxA13 expression significantly activated IGFBP-3 promoter activity ( Figure 6C). However, this transactivation was impaired by mutation of two putative HOX-binding sites on the IGFBP-3 promoter. These results confirmed that IGFBP-3 was a HOXA13 target gene. We also examined the effect of HoxA13 on the expression of IGFBP-3 in HoxA13 siRNA transfected CS12 cells and found that HoxA13 siRNA reduced the expression of IGFBP-3 by 80 to 85%, but scramble and off-target siRNA did not show the significant reduction ( Figure 6D). To search for IGFBP-3 mediated cancer related activity, we constructed IGFBP-3 siRNA. The expression of IGFBP-3 in IGFBP-3 siRNA-transfected CS12 cells was significantly reduced, but the scrambled and off-target siRNA did not change the expression of IGFBP-3 ( Figure 6E). Importantly, siRNA against IGFBP-3 reduced cell growth of CS12 cells significantly, but scrambled siRNA did not reduce cell growth ( Figure 6F). Knockdown of IGFBP-3 also reduced migration ( Figure 6G) and invasion ( Figure 6H) activities of CS12 cells. These results suggest that IGFBP-3 is critical for the cancerous features of CS12 cells.

Increased expression of HoxA13 and IGFBP-3 in human gastric cancer
The expression of HoxA13 was examined by immunohistochemistry (IHC) in gastric cancer obtained from 57 patients. HoxA13 was detected in 73.7% (42/57) of these samples ( Figure 7A and 7B). The expression of IGFBP-3 was stained for 28 specimens and the positive rate was 50.0% (14/28). By scoring expression levels (0 to 7) by positivity and intensity of IHC staining, HoxA13 and IGFBP-3 were highly expressed (score ≥ 4) in 63.2% (36/57) and 28.6% (8/28), respectively, of these gastric cancer specimens (Supplementary Table 1). We also analyzed HoxA13 and IGFBP-3 expression using the gastric cancer data of The Cancer Genome Atlas and found that both genes were overexpressed (Supplementary Figure 6).

DISCUSSION
We found DNA hypomethylation at the p53-E2F-1 responsive element (E1 site) of the HoxA13 promoter in CS12 cells when compared with CSN cells, together with increased recruitment of E2F, but exclusion of p53 binding at the E1 site, which were consistent with increased HoxA13 expression in CS12 cells. Forced expression of E2F-1 activated, but p53 inhibited HoxA13 promoter in CS12 and CSN cells ( Figure 4D, Supplementary Figure 3). The identification of differential DNA methylation at the E1 site of the HoxA13 promoter in CSN and CS12 cells is a new finding, which provides a mechanism to explain the upregulation of HoxA13 expression during the development of gastric cancer (Figure 2). Moreover, expression of HoxA13 was observed in gastric cancer specimen ( Figure 7A, Supplementary Figure 6, and www.impactjournals.com/oncotarget    Table 1). One previous study reported that increased HoxA13 expression was a poor prognostic factor in gastric cancer [17], this study further demonstrated that HoxA13 enhanced the migration and invasion ability of gastric cancer cells.
The hypomethylation at the E1 site of the HoxA13 promoter in CS12 cells was concurrent with decreased recruitments of the DNA methyltransferases DNMT1 and DNMT3b ( Figure 4E). In addition, increased binding of WDR5 and MLL complex together with elevated levels of H3K4me3 were observed at the E1 site, which were dependent on increased HOTTIP expression in CS12 cells ( Figure 5D). These result showed that HOTTIP activated HoxA13 expression through epigenetic mechanism including DNA methylation and histone modification. Interestingly, knockdown of HoxA13 led to a decrease of HOTTIP expression (Figure 6), demonstrating a positive feedback control of HoxA13 and HOTTIP expression.
More than 230 sequence-specific and specially expressed lncRNAs have been reported to be associated with the HOX gene family [39]. In liver, pancreas, and tongue squamous cell cancers, HOTTIP is positively associated with HoxA13 expression [20,22,23].
Up-regulation of HOTTIP is a negative prognostic factor for hepatocellular carcinoma patients [20]. Overexpression of HOTTIP in human pancreatic cancers increases cell proliferation, invasion, and EMT activity [23]. These studies support the oncogenic role of HOTTIP, which induces expression of another oncogene HoxA13 in gastric cancer.
We identified both IGFBP-3 and HOTTIP are the target genes of HoxA13 in gastric cancer. HoxA13 transactivated IGFBP-3 gene expression via Hox-binding elements in the GFBP-3 promoter ( Figure 6C). Despite extensive investigation showing the involvement of IGFBP-3 in cancers, it is not a currently used cancer biomarker because it is debatable whether IGFBP-3 is upor down-regulated in cancers. In gastric cancer, IGFBP-3 has been reported to be a suppressor of migration, invasion, and the EMT through suppression of invasive factors including MMP14 and urokinase-type plasminogen activator [40]. We here found contradictory results in CS12 cells, where both the expression of IGFBP-3 and cell migration were increased significantly. Besides, knockdown of IGFBP-3 inhibited cell proliferation, colony formation, migration, and invasion ( Figure 6E-6G). IGFBP-3 may potentiate gastric cancer cell division and invasion that contradicts previous findings of its role as a tumor suppressor [26]. There may be multiple factors that can influence IGFBP-3 expression, and its expression may have both positive and negative effects on tumor development as reported previously [24,25,41].
OCT4 was expressed in CSN and CS12 cells, indicating that both cell lines may have the stemness characteristics [36]. However, only CS12 expressed NANOG and LGR5 (Supplementary Figures 1A, 1B, and 2C) and expressed higher levels of SOX2, NANOG, REX1, and hTERT than CSN cells (Supplementary Figure 2D). Only CS12 cells showed differentiation of two germ layers and exhibited strong tumor formation of xenografts in SCID mice (Supplementary Figure 2E). Thus, we conclude that CS12 cells bear the cancer stem cell-like characteristics including characteristic patterns of cell proliferation, cell cycle progression, invasion and migration, and drug resistance. However, CS12 cells are not stem cells because the staining of alkaline phosphatase is negative.
In conclusion, the previously established gastric cancer CS12 cell line showed the characteristics of stemness gene expression but was not fully pluripotent because only two germ layers were differentiated. We also identified the IGFBP-3 as the target of HoxA13 and a positive regulator of gastric cancers. Both HoxA13 and IGFBP-3 were overexpressed significantly in human gastric cancer specimens of Taiwan (Supplementary Table 1) and The Cancer Genome Atlas (Supplementary Figure 6). Thus, the HoxA13-HOTTIP-IGFBP-3 axis might be an oncogenic pathway in the gastric cancer and a potential new oncotarget for gastric cancer therapy.

Cell lines, reagents, and animals
Human gastric normal cells CSN and cancer cells CS12 cells were cultured as described elsewhere [30] with a slight modification to include Keratinocyte-SFM (Gibco-Invitrogen Co., Carlsbad, CA, USA). 293T cells were obtained from the RIKEN Cell Bank (Tsukuba, Ibaraki, Japan) and were cultured in Dulbecco's modified Eagle's minimal essential medium (DMEM) (Gibco) supplemented with 10% charcoal-stripped FBS (Gibco) with or without 1% penicillin and streptomycin (Gibco). The animal welfare guidelines for the care and use of laboratory animals were approved by the Animal Care Committee of the RIKEN BioResource Center in Japan, the National Laboratory of Animal Center and the Kaohsiung Medical University in Taiwan.

Patient samples
This study enrolled patients with gastric cancers from the Kaohsiung Medical University Hospital (KMUH) from June 2010 to August 2013. The study of human subjects was approved by the Institutional Review Board of the KMUH (KMUHIRB-960343, Kaohsiung, Taiwan). All patients gave their informed consent, and the ethics and scientific committees of the participating institutions approved the study. Tumor types were determined according to the World Health Organization classification. At the time of surgery, all tissue samples were immediately flash-frozen in liquid nitrogen and stored at -80°C until

Cell proliferation, colony assay and cell cycle analyses
The living cells were counted using the trypan blue dye-exclusion method, and were analyzed by flow cytometry to identify the sub-G population of cells [43]. MTT assay was assayed as followed to the manufacturers' instructions as described elsewhere [44,45]. A colony assay was performed as described elsewhere [43]. Briefly, cells were plated in duplicate at 5 × 10 2 or 5 × 10 3 cells per gelatin-coated dish. Two weeks later, colonies with a diameter > 2 mm were counted after staining with Giemsa staining solution (Wako Chemical Co., Tokyo, Japan). For analysis of the cell cycle [46], serum-starved cells were cultured in DMEM containing 15% FBS and collected at the indicated times. Harvested cells were stained with propidium iodide (PI; 1 µg/mL), and subjected to a fluorescence-activated analysis of DNA content in a flow cytometer (EPICS XL-MCL; Beckman Coulter, Miami, FL, USA).

Migration, invasion, and chmoresistance assays
Cells (1 × 10 4 cells) cultured in DMEM without FBS were seeded in the upper Transwell plate coated with or without matrix gel (Corning, Inc., NY, USA; 1 mg/mL). The lower plate contained DMEM plus 10% FBS. Three days later, the cells on the lower plate of the Transwell were fixed with 4% formaldehyde, stained with 1% crystal violet, and the cells were counted under a microscope. Regarding chemoresistance assay, cells were seeded in 96-well plates and incubated for 24 h, to allow cell attachment. DMEM containing a serial dilution of 5-fuluorouracil (20 μg/mL) was added, and the cells were incubated for an additional 48 h in 5% CO. Cell viability was examined using the MTT assay.

Teratoma formation assay, alkaline phosphatase and immunohistochemistry
Induced pluripotent stem cells (iPS-like cells) (200 cells; one colony/spot) were injected subcutaneously into the dorsal flank of severe combined immunodeficiency (SCID) mice, as described elsewhere [47]. The teratomas that formed after the injection were fixed in 4% paraformaldehyde overnight and embedded in paraffin. Sections were stained with hematoxylin and eosin. Measurement of alkaline phosphatase activity and immunocytochemistry were performed as described elsewhere [47]. The antibodies used in this work were listed in Supplementary Table 2.

Immunoprecipitation and western blotting
Immunoprecipitation and western blotting were performed as described elsewhere [48,49].

Reverse transcriptase PCR (RT-PCR) and quantitative polymerase chain reaction (qPCR)
Total RNA was extracted from cells using the TRIzol reagent (Thermo Fisher scientific). RNA was reverse transcribed to cDNA using a reverse transcription kit (Promega). PCR was performed using the GoTaq ® green www.impactjournals.com/oncotarget master mix (Promega). qPCR was performed using the Quantifast SYBR green PCR kit (Qiagen, Gaithersburg, MD, USA) as described elsewhere [43][44][45][46]. Amplification curves and gene expression were normalized to those of β-actin or GAPDH which was used as an internal control. The primers used for qPCR are listed in the Supplementary Tables 3-6.

Transient transfection and luciferase assay
Transient transfection and luciferase assay were performed as described [43][44][45][46]. Cells were plated into each well of a 12-well plate and cultured for 24 h. The cells were then cotransfected with the indicated amount of constructs carrying the IGFBP3-promoter, HoxA13promoter-luciferase reporters and with or without increasing dose of HoxA13, using Lipofectamine 2000 (Invitrogen). The total amount of transfected DNA was kept constant at 1 μg/well by the addition of pBluescript. After 48 h or the indicated period of incubation, the cells were harvested and the activities of luciferase were measured in an illuminometer (Berthold Technologies GmbH and Co. KG, Bad Wildbad, Germany) using the Dual-Luciferase Reporter Assay System (Promega). Luciferase activity values were normalized to transfection efficiency.

Quantitative DNA methylation analysis by mass ARRAY epityping
High molecular weight DNA was isolated from CSN and CS12 cells using the PureGene kit from Qiagen (Hilden, Germany). Bisulfite conversion was performed by using the EZ DNA Methylation-Gold kit (Zymo Research Co., Irvine, CA, USA, Cat. no. D5005) [49]. The promoter region of the HoxA13 gene was PCR-amplified from bisulfite-treated human genomic DNA using primers that incorporated the T7 promoter sequence. The DNA methylation analysis was performed on Mass ARRAY (Sequenon) through MALDI-TOF mass spectrometer for data acquisition [50]. Comparative sequence analysis was done by using EpiTYPER software for the percentage of specific site methylation [47]. MassAray primers were designated to cover the promoter regions of the indicated genes.

Chromatin immunoprecipitation assay (ChIP)
The ChIP assay was performed as described elsewhere [46]. The immunoprecipitated protein-DNA complexes were washed twice with binding buffer (10 mM HEPES, pH 7.9, 10 mM Tris-HCl, pH 7.9, 12.5% glycerol, 0.25% NP-40, 0.5% Triton X-100, 0.24 M NaCl, 0.75 mM MgCl 2 , 1.1 mM EDTA, and protease inhibitor mixture) and then washed twice with Tris-EDTA buffer (10 mM Tris-HCl, pH 7.9, and 1 mM EDTA). The protein-DNA complexes were disrupted with proteinase K (Sigma-Aldrich) DNA was extracted with phenol and chloroform, precipitated in ethanol, and analyzed by real-time PCR using the Power SYBR Green Master Mix (Invitrogen). The PCR conditions consisted of 1 cycle of 2 min at 50°C and 1 cycle of 10 min at 95°C followed by 40 cycles of 95°C for 15 sec, and 55-60°C for 60 sec. The primers used in these experiments are shown in the Supplementary Table 7.

Statistical analysis
The data are presented as the mean ± SEM from triplicate experiments and additional replicates as indicated. Significance was assessed using two-way ANOVA (P < 0.0001) followed by two-tailed student's t-tests. Survival analysis was performed using the Kaplan-Meier method, and the curves were compared using the log-rank test. A P value < 0.05 was considered statistically significance.