The extracellular sulfatase SULF2 promotes liver tumorigenesis by stimulating assembly of a promoter-looping GLI1-STAT3 transcriptional complex

The expression of the extracellular sulfatase SULF2 has been associated with increased hepatocellular carcinoma (HCC) growth and poor patient survival. However, the molecular mechanisms underlying SULF2-associated tumor growth remain unclear. To address this gap, here we developed a transgenic mouse overexpressing Sulf2 in hepatocytes under the control of the transthyretin promoter. In this model, Sulf2 overexpression potentiated diethylnitrosamine-induced HCC. Further analysis indicated that the transcription factor GLI family zinc finger 1 (GLI1) mediates Sulf2 expression during HCC development. A cross of the Sulf2-overexpressing with Gli1-knockout mice revealed that Gli1 inactivation impairs SULF2-induced HCC. Transcriptomic analysis revealed that Sulf2 overexpression is associated with signal transducer and activator of transcription 3 (STAT3)-specific gene signatures. Interestingly, the Gli1 knockout abrogated SULF2-mediated induction of several STAT3 target genes, including suppressor of cytokine signaling 2/3 (Socs2/3); Pim-1 proto-oncogene, Ser/Thr kinase (Pim1); and Fms-related tyrosine kinase 4 (Flt4). Human orthologs were similarly regulated by SULF2, dependent on intact GLI1 and STAT3 functions in HCC cells. SULF2 overexpression promoted a GLI1-STAT3 interaction and increased GLI1 and STAT3 enrichment at the promoters of their target genes. Interestingly, the SULF2 overexpression resulted in GLI1 enrichment at select STAT3 consensus sites, and vice versa. siRNA-mediated STAT3 or GLI1 knockdown reduced promoter binding of GLI1 and STAT3, respectively. Finally, chromatin-capture PCR confirmed long-range co-regulation of SOCS2 and FLT3 through changes in promoter conformation. These findings define a mechanism whereby SULF2 drives HCC by stimulating formation of a GLI1-STAT3 transcriptional complex.

The expression of the extracellular sulfatase SULF2 has been associated with increased hepatocellular carcinoma (HCC) growth and poor patient survival. However, the molecular mechanisms underlying SULF2-associated tumor growth remain unclear. To address this gap, here we developed a transgenic mouse overexpressing Sulf2 in hepatocytes under the control of the transthyretin promoter. In this model, Sulf2 overexpression potentiated diethylnitrosamine-induced HCC. Further analysis indicated that the transcription factor GLI family zinc finger 1 (GLI1) mediates Sulf2 expression during HCC development. A cross of the Sulf2-overexpressing with Gli1knockout mice revealed that Gli1 inactivation impairs SULF2induced HCC. Transcriptomic analysis revealed that Sulf2 overexpression is associated with signal transducer and activator of transcription 3 (STAT3)-specific gene signatures. Interestingly, the Gli1 knockout abrogated SULF2-mediated induction of several STAT3 target genes, including suppressor of cytokine signaling 2/3 (Socs2/3); Pim-1 proto-oncogene, Ser/Thr kinase (Pim1); and Fms-related tyrosine kinase 4 (Flt4). Human orthologs were similarly regulated by SULF2, dependent on intact GLI1 and STAT3 functions in HCC cells. SULF2 overexpression promoted a GLI1-STAT3 interaction and increased GLI1 and STAT3 enrichment at the promoters of their target genes. Interestingly, the SULF2 overexpression resulted in GLI1 enrichment at select STAT3 consensus sites, and vice versa. siRNA-mediated STAT3 or GLI1 knockdown reduced promoter binding of GLI1 and STAT3, respectively. Finally, chromatincapture PCR confirmed long-range co-regulation of SOCS2 and FLT3 through changes in promoter conformation. These findings define a mechanism whereby SULF2 drives HCC by stimulating formation of a GLI1-STAT3 transcriptional complex.
Sulfatase 2 (SULF2) 3 has been identified as an important prognostic marker and regulator of HCC tumorigenesis. Elevated SULF2 expression has been identified in ϳ60% of human HCC. Increased expression of SULF2 in HCC is associated with increased tumor growth, migration, microvascular density, hepatoblast phenotype, and poorer patient survival, with higher rates of tumor recurrence after surgical resection (1). Furthermore, a Sulf2 knockout (KO) mouse was found to have diminished diethylnitrosamine (DEN)-induced liver tumorigenesis (2). SULF2 is an extracellular enzyme with 6-O-desulfatase activity acting on heparan sulfate proteoglycans. Heparan sulfate proteoglycans function as storage sites, harboring various cytokines and growth factors that are released via SULF2 enzymatic activity (3). Given the ability of SULF2 to regulate release of many signaling factors, it is likely that SULF2-mediated tumorigenesis is facilitated through a number of signaling pathways, including ligands of Hedgehog, WNT, and TGF␤ pathways (4). Interestingly, these cascades all have transcription factor GLI1 as a common downstream effector (4 -6). Furthermore, SULF2-mediated WNT signaling, critical for liver regeneration, is GLI1-dependent (6).
We therefore hypothesized that the SULF2-Gli1 signaling axis is a key regulator of HCC tumorigenesis in vivo. Here, we demonstrate that GLI1 is critical for SULF2-potentiated HCC development. Not only is GLI1 important, but it promotes tumor growth via heterodimerization with STAT3 to cooperatively regulate transcription of JAK/STAT-signaling target genes. These findings define a novel mechanism controlling HCC pathobiology and identified this new GLI1-STAT3 complex as a central mediator of this phenomenon.

Genetically engineered mouse model with transthyretin promoter (TTR)-driven Sulf2 overexpression potentiates liver tumorigenesis
To determine the pathophysiologic role of Sulf2 in HCC, we developed a transgenic mouse expressing murine Sulf2 under control of the hepatocyte-specific TTR (Fig. 1A). Nontransgenic controls are referred to as Sulf2-WT and have physiologic expression of Sulf2. Histopathologic assessment of liver tissue from 8-month-old Sulf2-WT and Sulf2-TTR mice did not reveal any major phenotypic differences. Analysis revealed normal liver architecture without evidence of cellular atypia or inflammation (Fig. 1B). Overexpression of Sulf2 was confirmed by quantitative PCR (Fig. 1C). A cohort of 16 mice was observed long-term, consisting of six Sulf2-TTR and 10 Sulf2-WT mice. One of the six Sulf2-TTR mice developed spontaneous liver tumorigenesis at 19 months, demonstrating typical hepatocellular carcinoma with enlarged round hyperchromatic nuclei, high nuclear/cytoplasmic ratio, and moderate microvesicular fat globules in the cytoplasm. Tumor cells were arranged in thick trabecular, solid, and pseudoglandular growth patterns (Fig. 1D).
Given the low rate of spontaneous liver tumorigenesis, DEN was used to induce HCC. Briefly, at 14 days of age, male mice were treated with intraperitoneal DEN and sacrificed at 8 months ( Fig. 2A). Gross examination of murine livers demonstrated substantial tumor burden in Sulf2-TTR mice relative to Sulf2-WT mice (Fig. 2C). Histopathologic analysis revealed no obvious morphologic differences in tumors between mouse genotypes. Pathology consistently demonstrated well-differentiated HCC (Fig. 2D). Interestingly, there was no significant difference in overall tumor number between the two mouse genotypes. However, in line with our gross observations, Sulf2-TTR mice had significantly larger tumor volume, weight, and tumor/liver weight ratio relative to Sulf2-WT (Fig. 2E). Finally, the number of mice with metastatic disease in the lung was not clearly different between the two mice ( Fig. 2F). Taken together, hepatocyte-specific Sulf2 overexpression potentiates DEN-induced liver tumorigenesis.

Gli1 loss impairs Sulf2-potentiated liver tumorigenesis
Consistent with prior studies between SULF2 and GLI1 in vitro, overexpression of Sulf2 correlated with increased Gli1 expression in Sulf2-TTR mice relative to Sulf2-WT mice (6, 7) (Fig. S1A). Furthermore, as Sulf2-WT and Sulf2-TTR mice age, relative expression levels of Sulf2 and Gli1 remain stable without statistical differences. With exposure to DEN, there appears to be a trend for increased expression of both Sulf2 and Gli1 with increased variability of expression relative to baseline. However, these findings were also not significantly different (Fig. S1B).
To further delineate the interplay between Sulf2 and Gli1 in liver tumorigenesis in vivo, Sulf2-TTR mice were crossed with Gli1 knockout mice (8) (Fig. S1C). Resultant genotypes of interest included Gli1 WT, heterozygous (HET), and homozygous KO mice, all with TTR-driven Sulf2 overexpression. Gli1 knockout did not significantly alter Sulf2 expression levels (Fig.  2B). As described above, male mice were treated with intraperitoneal DEN to induce liver tumorigenesis. Gross analysis revealed marked differences between genotypes. As noted above, Sulf2-TTR, Gli1 WT mice demonstrated significant liver tumor burden compared with Sulf2-WT, GLI1 WT controls; however, knockout of Gli1 impaired DEN-induced tumorigenesis in a Gli1 allele dose-dependent manner, with homozygous knockouts having significantly lower gross tumor burden (Fig. 2C). Although significant differences in tumor burden were observed, once again histopathologic analysis of tumors across genotypes did not differ significantly, demonstrating well-differentiated HCC (Fig. 2D).
Further quantitative analyses reinforced gross observations of tumor burden. With knockout of Gli1, there was an allele dose-dependent decrease in tumor number, tumor volume, tumor weight, and tumor/liver weight ratio (Fig. 2E). Further, several Gli1 WT mice developed metastatic disease, Gli1 KO SULF2 promotes liver cancer through GLI1-STAT3 complex SULF2 promotes liver cancer through GLI1-STAT3 complex mice lacked visible metastases, and Gli1 HET mice demonstrated an intermediate phenotype (Fig. 2F). Thus, intact Gli1 is critical for Sulf2-potentiated liver tumorigenesis.

Gli1 abrogation down-regulates Stat3 target genes
To elucidate transcriptional target genes mediating HCC tumorigenesis via the Sulf2-Gli1 signaling axis, we determined the differential gene expression between tumors derived from Sulf2-TTR, Gli1 WT, and Sulf2-TTR, Gli1 KO mice. RNA-Seq was performed on these tumors in triplicate. Unsupervised clustering of transcriptomic profiles resulted in appropriate segregation between Gli1 WT and Gli1 KO tumors (Fig. S2A). Relative to Gli1 WT tumors, in Gli1 KO tumors, 11 genes were significantly up-regulated, whereas 34 genes were significantly down-regulated (Fig. 3A). Ingenuity pathway analysis revealed the canonical Stat3 pathway as one of the most differentially regulated. Similarly, KEGG pathway analysis revealing significant down-regulation of Jak-Stat signaling (Fig. S2B), including known canonical targets of Stat3 such as Socs2, Socs3, Pim1, and Flt4 (Fig. 3A).
Validation was then carried out in homogenized tumor tissue derived from three mice of each Gli1 genotype. Quantitative PCR confirmed an allele dose-dependent reduction in expression of the canonical Jak/Stat signaling components Socs2, Socs3, Pim1, and Flt4 (Fig. 3B). Additional differentially expressed genes with evidence of Stat3 regulation in the literature (listed above), were also validated. This revealed several candidates that have reduced expression in Gli1 KO relative to Gli1 WT tumors (Fig. S2D). Thus, these findings suggest that increased Stat3-regulated gene expression underlies Sulf2-Gli1 axis mediated liver tumorigenesis.

SULF2-mediated up-regulation of STAT3 target genes depends on both GLI1 and STAT3
The significance of the observed changes for human HCC was first investigated using the Cancer Genome Atlas. Assessing liver cancer Cancer Genome Atlas transcriptomic data (n ϭ 438 cases) revealed positive correlations between SULF2 expression and not only GLI1, but also JAK/STAT3 signaling components, including SOCS2, SOCS3, FLT3, FLT4, and STAT3. The only exception was PIM1, which did not seem to be related to SULF2 expression (Fig. S3).
Thus, the role of SULF2, GLI1, and STAT3 in regulation of these human orthologues was investigated in the HCC cell lines Huh-7 and Hep3B. HCC cells were transfected with either nontargeting siRNA (siNT) or siRNA against GLI1 (siGLI1). Knockdown efficiency was evaluated by Western blotting (Fig.  S4A). 24 h after transfection, cells were transduced with either control adenovirus or viral particles expressing SULF2. Expression of SOCS2, SOCS3, PIM1, and FLT3 was assessed by qPCR 72 h after transfection. The expression patterns of all four genes were similar. As seen in murine tissue, SULF2 transduction increased expression of the above STAT3 targets, and GLI1 knockdown abrogated this induction (Fig. 3C).
Using both the in silico tool INSECT 2.0 (28) and manual analysis of the 3000 base pairs upstream of each gene's transcriptional start site (TSS), putative GLI1 consensus sites were identified in promoters of SOCS2 (1 site; Ϫ2737), SOCS3 (4 sites; Ϫ2871, Ϫ1820, Ϫ1314, Ϫ399), PIM1 (1 site; Ϫ669), and FLT3 (3 sites; Ϫ2173, Ϫ1589, Ϫ374). These sites are graphically depicted in Fig. 3D. Binding was confirmed by ChIP studies. Relative to cells transduced with control adenoviral particles, SULF2 transduction resulted in GLI1 enrichment within at least one of the putative binding sites identified (Fig. 3D). The exception was PIM1, where GLI1 was not significantly enriched at its only putative binding site as identified by our methodology (Fig. 3D). This finding suggests either a more indirect regulatory mechanism or a more distant GLI1-binding site. Thus, transcriptomic analysis revealed a STAT3 signature among a cohort of genes positively regulated by SULF2 and dependent on intact GLI1. SULF2 expression results in GLI1 enrichment at the promoters of these genes.
Given the numerous STAT3-regulated genes that appear to be important in GLI1-dependent liver tumorigenesis, the requirement for STAT3 in gene expression was also interrogated. SULF2-mediated increased expression of SOCS2, SOCS3, PIM1, and FLT3 was abrogated by siRNA-mediated STAT3 knockdown (Fig. 4A). Knockdown efficiency was evaluated by Western blotting (Fig. S4A). Expression patterns of additional human orthologues of genes of interest discovered during RNA-Seq of mouse tissues ( Fig. 3 and Fig. S2) were also assessed under these experimental conditions. A similar pattern was observed to the canonical JAK/STAT signaling genes for most of these. Exceptions included DERL3 and SERINC2. DERL3 expression was unaffected by GLI1 or STAT3 knockdown or SULF2 overexpression, whereas SERINC2 expression was decreased by not only GLI1 and STAT3 knockdown but also SULF2 overexpression (Fig. S4B).
Furthermore, putative STAT3 consensus binding sites were identified. Several sites were found in SOCS2 (1 site; Ϫ1043),  . Interestingly, much as with GLI1, STAT3 occupancy was enriched at all gene promoters tested with SULF2 transduction relative to controls (Fig. 4B).
In addition to these cis regulatory elements within promoter regions, we also used in silico analyses to identify enhancer elements for the canonical JAK/STAT signaling genes. The genomic coordinates for enhancer and double-elite regions annotated to regulate the JAK/STAT genes were identified

SULF2 promotes liver cancer through GLI1-STAT3 complex
using the GeneHancer data (version 4.4) from GeneCards or UCSC Genome Browser. Using the Homer tool findMotifs-Genome, these genomic coordinates were searched for transcription factor enrichment. Ultimately, this exploratory search of a small gene set (n ϭ 6) did not reveal an enrichment of GLI1 or STAT3 motifs associated with enhancer regulation of SOCS2, SOCS3, PIM1, ITGB3, PTGES2, or FLT3. Taken together, these findings support a promoter-based co-regula-tion by GLI1 and STAT3. This is further supported by a nonsignificant enrichment of GLI motifs in the promoters (Ϫ2000 to ϩ100 bp around the TSS) of these six genes using the Homer tool findMotifs.
Given expression of SOCS2, SOCS3, PIM1, and FLT3 are dependent on both GLI1 and STAT3; co-regulation was hypothesized to be mediated by GLI1 and STAT3 heterodimerization. Therefore, HCC cell lines were co-transfected

SULF2 promotes liver cancer through GLI1-STAT3 complex
with FLAG-tagged STAT3 and His-tagged GLI1. Immunoprecipitation of either FLAG or His did not result in detectable levels of GLI1 or STAT3, respectively (Fig. 4C). However, transduction with adenovirus expressing SULF2 after co-transfection of transcription factors resulted in increased co-immunoprecipitation (Fig. 4C). Together, these findings confirmed interaction between STAT3 and GLI1 and define these factors as common effectors of SULF2.

SULF2 expression induces GLI1 and STAT3 enrichment at reciprocal consensus binding sites
The relative location of GLI1 and STAT3 consensus sites, where enrichment was confirmed with SULF2 expression, was mapped on promoters of SOCS2 (Fig. 5A), SOCS3 (Fig. 5B), PIM1 (Fig. 5C), and FLT3 (Fig. 5D). As depicted, the relative distribution of the confirmed binding sites is significantly dispersed. This led to the hypothesis that either GLI1 or STAT3 could bind to its respective consensus site with the other transcription factor in complex with it to promote transcription. Alternatively, GLI1 and STAT3, bound at disparate sites, could physically augment the promoter to come into close proximity and regulate gene expression.
To address the former hypothesis, ChIP-PCR was carried out to determine whether enrichment of GLI1 or STAT3 was observed at reciprocal transcription factor-binding sites. Enrichment of GLI1 at STAT3-binding sites and STAT3 at  GLI1 sites was assessed both with transduction of control adenovirus and a vector expressing SULF2. Interestingly, depicted in the top row of bar graphs for each panel, GLI1 was enriched at all STAT3 consensus sites with SULF2 expression relative to controls. STAT3 occupancy was also enriched at all GLI1-binding sites (Fig. 5, A-D). Further, the same patterns were seen in a majority of other putative STAT3 target genes (Fig. S5).

SULF2 promotes liver cancer through GLI1-STAT3 complex
Knockdown experiments were then performed with transfection of siNT, siGLI1, or siSTAT3 prior to SULF2 transduction. Results are shown in the bottom row of each panel of Fig. 6. A common pattern across promoters was observed. If studying a GLI1 consensus site, STAT3 knockdown did not result in significantly reduced GLI1 enrichment. However, GLI1 knockdown resulted in depletion of STAT3 at the binding site. The opposite was true at STAT3-binding sites. GLI1 knockdown had no significant effect on STAT3 binding, whereas STAT3 knockdown reduced GLI1 enrichment (Fig.  5, A-D). Notable exceptions were the GLI1 and second (from the TSS) STAT3-binding sites within the SOCS3 promoter. At these sites, GLI1 or STAT3 knockdown had no significant effect on STAT3 or GLI1 occupancy, respectively (Fig. 5B). This can be explained by the close proximity of the Because the ligation product of interest is rare, data are presented as -fold change in amplification normalized to the undigested control. Huh-7 cells were transfected with siNT (n ϭ 6), siGLI1 (n ϭ 6), or siSTAT3 (n ϭ 6). Afterward, they were transduced with either Adv-Null (n ϭ 6) or Adv-SULF2 (n ϭ 18). DNA was either undigested as a negative control or digested with the restriction enzyme depicted. Results are expressed as means Ϯ S.E. (error bars). C, diagram depicting a working model of SULF2-potentiated liver tumorigenesis. SULF2 overexpression results in increased GLI1 expression and enrichment at target genes. Co-regulation of genes occurs with STAT3 through heterodimerization and either co-enrichment or promoter DNA looping to mediate gene expression.

SULF2 promotes liver cancer through GLI1-STAT3 complex
consensus sites, prohibiting the design of primer sets to interrogate them individually.
These results support the hypothesis that enrichment of GLI1 at a STAT3 consensus site depends on intact STAT3 and vice versa. However, this does not rule out the possibility of distant, transcription factor-bound sites being structurally brought into close proximity to regulate gene expression.

GLI1 and STAT3 coordinately control promoter DNA looping in HCC cells
The relative positions of confirmed GLI1 and STAT3 consensus binding sites, as depicted in Fig. 5, prohibit direct interaction of DNA-bound transcription factors in a number of cases. This suggested two possibilities. First, one of the two transcription factors may simply recruit the other, as supported by enrichment of one transcription factor at reciprocal consensus sites. Alternatively, GLI1 and STAT3 may augment promoter structure to bring the factors into close proximity, forming heterodimers. The latter hypothesis was tested with chromatin conformation capture (3C) qPCR. Among the four genes of interest, we chose to investigate SOCS2 and FLT3 loci, given the more feasible transcription factor-binding orientation. Specifically, each gene promoter contains a single GLI1binding site and STAT3-binding site separated by Ͼ1000 nucleotides (Fig. 5, A and D). The schematics for SOCS2 and FLT3 are depicted in Fig. 6A. In the SOCS2 promoter, the GLI1binding site is located 1740 bp upstream of the STAT3-binding site. Primers designed to anneal between these sites were located 1097 nucleotides apart. Two BsrI digestion sites were identified between this primer set. Successful promoter digestion and ligation between these sites would result in reduction of a 1222-bp amplicon to 145 bp (Fig. 6A, left). Similarly, the GLI1 site in the FLT3 promoter is located 1305 bp upstream of the STAT3-binding site. Primers designed to amplify the region between the sites would result in a 1097-bp amplicon. Using BsrGI digestion and subsequent ligation, this would reduce the amplicon to 500 bp (Fig. 6A, right).
To determine whether GLI1 and STAT3 were interacting at these distances in each promoter, 3C-qPCR was performed on genomic DNA isolated from Huh-7 cells under various experimental conditions. First, cells were transfected with siNT, siGLI1, or siSTAT3. 24 h after transfection, cells were transduced with either control adenoviral particles or a vector expressing SULF2. PCR was performed 48 h after transduction. Isolated genomic DNA was either left undigested or incubated with BsrI (for SOCS2 studies) or BsrGI (for FLT3 studies) with subsequent ligation and qPCR. Under conditions of transduction with control adenovirus, effectively no amplification was detected. However, transduction with SULF2 resulted in a significant increase in amplification under conditions of BsrI or BsrGI digestion with minimal detection without digestion. Furthermore, siRNA-mediated knockdown of both GLI1 and STAT3 abrogated this signal, supporting their importance in co-regulation (Fig. 6B). Confirmation of these results was conducted by gel electrophoresis of PCR products (data not shown). These findings indicated likely physical interaction between GLI1-and STAT3-binding sites of the SOCS2 and FLT3 promoters in the context of SULF2 expression.

Discussion
Taken together, in HCC, we have previously described the role of SULF2 in releasing growth factors such as Hedgehog (SHH) ligands, TGF␤, and Wnt3A, which can increase expression of GLI1 (4, 6, 7). Here, using a novel Sulf2 transgenic mouse model, we found that Sulf2 potentiates HCC tumorigenesis. Furthermore, this process depends on intact Gli1. Interestingly, Gli1 promotes a gene expression profile including many wellcharacterized Stat3 target genes both in the mouse model and in human orthologues. Regulatory interaction between GLI1 and STAT3 is not without precedence (33)(34)(35), but out study demonstrates that SULF2 expression results in increased heterodimerization with GLI1. Furthermore, GLI1 and STAT3 are both found to be enriched not only at their respective consensus promoter-binding sites, but also at a number of reciprocal binding sites for many target genes. Thus, co-regulation of this gene cohort is likely mediated by multiple mechanisms. GLI1 and STAT3 may directly interact at a single consensus binding site. Alternatively, we show that spatially distant STAT3 and GLI1 sites can be structurally brought into close proximity to promote gene expression (Fig. 6C).
This work interweaves a number of critical signaling pathways in HCC pathogenesis. GLI1 is a common downstream effector of both Hedgehog and WNT signaling, whereas STAT3 is a key downstream mediator of inflammatory signaling via the IL-6/JAK/STAT axis. In hepatocellular adenomas, benign lesions with potential for malignant transformation, multiple molecular subtypes have been defined. Interestingly, mutations within the aforementioned pathways define some of these subtypes: activating mutations in ␤-catenin, activating mutations within JAK2/STAT3 signaling, and a more recently described subtype characterized by increased GLI1 expression as a consequence of INHBE and GLI1 fusion (29). Not only do aberrations in these pathways play a role in defining adenoma subtypes, but they are also found in a large proportion of HCC.
In fact, adenomas harboring mutations in ␤-catenin carry a higher risk of malignant transformation (29). It follows that WNT/␤-catenin signaling activation, most commonly caused by mutations in CTNNB1 (␤-catenin), has been observed in approximately half of HCC cases (30 -32). Inactivating mutations of TP53 are also abundant in HCC (30,31). Recently, multi-omics characterization of HCC by the Cancer Genome Atlas Research Group identified an association between TP53 mutation status and SHH signaling activity. In the context of inactive TP53, the known target gene and negative regulator of SHH, PTCHD4, had markedly reduced expression. Furthermore, gene set enrichment analysis revealed up-regulated SHH expression (32). Finally, although not as common as in hepatocellular adenomas, activating mutations in IL-6 signal transducer (IL6ST), which lead to increased JAK/STAT signaling, are among the top 15 recurrently mutated genes in HCC (32). Consequently, recurrent mutations in HCC resulting in aberrantly activated WNT/␤-catenin, SHH, and JAK/STAT signaling may converge on transcriptional effectors GLI1 and STAT3, which co-regulate genes.
Unfortunately, current therapeutic modalities for HCC have advanced only incrementally over the last decade (36 -40).

SULF2 promotes liver cancer through GLI1-STAT3 complex
However, as molecular pathogenesis of HCC is better understood, novel treatment approaches have been developed. Given the relevance of SULF2 in pathogenesis and prognosis of HCC (1-4, 42, 43), targeting SULF2 has been investigated and seems promising in preclinical studies (7). In addition, small-molecule inhibitors for WNT signaling are also being investigated (44). However, whether WNT, JAK/STAT, or SHH signaling is clinically targeted in HCC, the current study suggests the possibility of cross-talk limiting the efficacy of such approaches. Instead, given that these frequently aberrantly activated pathways appear to have the common transcriptional complex of GLI1 and STAT3 as a downstream mediator of pathogenesis, novel approaches disrupting these effectors may be less susceptible to treatment resistance.

Experimental procedures
Animals FVB strains overexpressing Sulf2 under control of the hepatocyte-specific transthyretin promoter (Sulf2-TTR mice) were developed by the Mayo Clinic transgenic and knockout core facility. Mice were created with the Ϫ3 kb TTR promoter region driving the expression of Sulf2 cDNA, which was cloned into the TTR second exon. Cloning was achieved by amplifying the coding sequence of murine Sulf2 with restriction sites for BamHI on the 5Ј end and XbaI on the 3Ј end using mouse kidney cDNA. The PCR product was digested with BamHI and XbaI and ligated into vector pcDNA3.1. Sulf2 was then cut out of the pcDNA3.1 vector using BamHI and XbaI, and 5Ј overhangs were filled using T4 DNA polymerase and dNTPs and subsequently ligated into the pTTR exV3 vector at the StuI blunt cut site. Insertion and orientation were confirmed by sequencing (45).
FVB founders were back-crossed into the C57/BL6 background to the N6 generation. This was done to ensure uniform susceptibility to carcinogenesis, as C57/BL6 mice are known to have a higher propensity for developing tumors with DEN treatment when compared with FVB mice. The care and use of the animals for these studies were reviewed and approved by the Mayo Clinic Institutional Animal Care and Use Committee.
The Sulf2 transgenic mice were not found to have any major phenotypic changes when compared with WT. Mice did not show any signs of liver injury or inflammation. WT littermates were used as controls. Mice were maintained in a temperature-controlled (22°C), pathogen-free environment and fed a standard rodent chow diet and water ad libitum. The care and use of the animals for these studies were reviewed and approved by the Mayo Clinic Institutional Animal Care and Use Committee.
Gli1 knockout mice have a C57/BL6 background (8) and were raised and maintained in the animal facilities of Mayo Clinic. Gli1 heterozygous null and Sulf2-TTR mice were crossed, ultimately producing three groups of mice: Sulf2-TTR, Gli1 WT; Sulf2-TTR, Gli1 HET; Sulf2-TTR, Gli1 KO. Primers used for genotyping of Sulf2 and Gli1 status are listed in Table 1.

DEN-induced liver tumors in mice
DEN was used as a carcinogen to induce primary liver tumors. DEN is a genotoxic drug, and its use has been well-established in mouse models for hepatocarcinogenesis (46). At 14 days of age, mice received a single intraperitoneal injection of DEN (15 mg/kg body weight). At 21 days of age, mice were separated by sex and genotyped. Littermates with negative genotypes were used as WT controls. Only male mice were included in the analysis, as female mice have a substantially lower rate of tumor development after DEN treatment. All mice were sacrificed at 8 months, and their liver and lungs were examined for tumors. The liver weight, number of visible tumors, and size of visible tumors were recorded.

Histopathology
Liver tissue from mouse models were removed and fixed in 10% neutral buffered formalin. Tissues were sent to Mayo Clinic Histology Core Laboratory for paraffin embedding and sectioning for hematoxylin and eosin staining. Histopathology was independently reviewed by two pathologists, R. G. and M. V., who were blinded to genotype.

RNA-Seq
S. W. provided histopathologic assessment of murine liver tissue prior to RNA extraction for RNA-Seq. Replicates submitted were considered morphologically similar. cDNA libraries were generated in accord with the Illumina RNA-Seq preparation protocol. The cDNA fragments were amplified by PCR and sequenced at both ends using an Illumina Genome Analyzer IIx. All samples had over 90 million reads with over 75% of the reads mapping to the mouse genome (obtained from the UCSC Genome Browser assembly ID: mm10). Bioinformatics analysis was done using Qiagen Ingenuity Pathway Analysis. Reads were analyzed using MAPRSeq version 2.0.0 against the mouse reference for calculating gene expression counts. Differential expression using edgeR was performed between WT and Gli1 KO mouse liver tumors. Samples were sequenced in triplicate. Gene expression levels were filtered using a false discovery rate of Ͻ0.1 and a -fold change whose magnitude is greater than 1 and log counts per million ϾϪ1. Data have been deposited in the GEO database, GSE139374 ("Investigating the role of GLI1 in SULF2-potentiated liver tumorigenesis") ( Table 2).

Quantitative PCR
Total RNA was extracted from cultured cells or primary murine tissue using TRIzol reagent (Invitrogen) following the manufacturer's protocol. 2 g of total RNA was reverse-transcribed using a high-capacity cDNA synthesis kit (Applied Biosystems). A portion of the total cDNA was amplified by quan-

SULF2 promotes liver cancer through GLI1-STAT3 complex
titative PCR. Samples were prepared with 1ϫ SYBR Green Supermix (Bio-Rad). Primers used for validation of RNA-Seq findings in mouse model-derived liver tumors are listed in Table 3. Primers used to assess levels of human orthologues in HCC cell lines are listed in Table 4. Amplification was performed using the C1000 thermal cycler (Bio-Rad) under the following reaction conditions: 95°C for 3 min followed by 40 cycles of 30 s at 95°C, 30 s at 60°C, and 20 s at 72°C. Each mRNA level was normalized by comparison with glyceraldehyde-3-phosphate dehydrogenase and TATA-binding protein RNA levels in the same sample. The results were calculated following the 2⌬Cp method.

Tissue culture
The Hep3B cell line was obtained from ATCC and cultured in complete minimum essential medium with 10% FBS. The Huh-7 cell line was obtained from the Japan Health Science Research Resources Bank (HSRRB, Osaka, Japan) and grown in Dulbecco's modified Eagle's medium with 10% FBS.

Adenoviral transduction
Adenoviral transduction was performed using null adenovirus (Adv-Null) (1060, Vector Biolabs) or Adv-SULF2 (ADV-224703, Vector Biolabs). Viral particles were applied to culture medium at a multiplicity of infection of 50 based on optimization experiments. Transduction efficiency was confirmed by qPCR and Western blotting.

siRNA knockdown studies
Knockdown experiments were performed using transfection reagent DharmaFECT 3 (T-2003-04, Dharmacon). Briefly, on day 0 of transfection, 1 ϫ 10 6 cells were plated in a 10-cm dish. On day 1 of transfection, two 500-l aliquots of Opti-MEM were prepared in 1.7-ml microcentrifuge tubes. 6.5 l of each siRNA was added to one tube, whereas 10 l of transfection reagent was added to the other. Medium aliquots were mixed and incubated at room temperature for 20 min prior to dropwise application to a single 10-cm dish. Knockdown of GLI1 was achieved using two ON-TARGETplus siRNA constructs (J-003896-05-0002 and J-003896-06-0002, Dharmacon), whereas reagents used against STAT3 included two ON-TAR-GETplus siRNA constructs (J-003544-08-0002 and J-00354-10-0002, Dharmacon). Knockdown efficiency was confirmed by Western blotting and qPCR as described.

ChIP PCR
ChIP was conducted as described previously (5). Primers used for assays are listed in Table 5.

SULF2 promotes liver cancer through GLI1-STAT3 complex Co-immunoprecipitation
Huh-7 and Hep3B cells were transfected with both GLI1-His and STAT3-FLAG on day 1 followed by transduction with Adv-Null or Adv-SULF2 on day 2. Cells were then harvested and lysed in 50 mM Tris, pH 7.5, 300 mM NaCl, 0.5% Nonidet P-40 with Complete inhibitor. Lysates were passed five times through a 27 1 ⁄2-gauge needle and then diluted with 50 mM Tris, pH 7.5, 0.5% Nonidet P-40 to 150 mM NaCl. Samples were centrifuged at 17,000 ϫ g for 10 min. Supernatants were subjected to immunoprecipitation following the Dynabeads Protein G immunoprecipitation kit protocol (Invitrogen). The following antibodies were cross-linked to

SULF2 promotes liver cancer through GLI1-STAT3 complex
Dynabeads Protein G for 1 h at room temperature: anti-GLI1 (as above), anti-STAT3 (as above), and normal goat and rabbit IgGs (Santa Cruz Biotechnology, Inc.). Dynabeads-antibody complexes and lysates were incubated overnight at 4°C with rotation. Proteins were eluted by the addition of SDS sample buffer and incubation at 95°C for 5 min. These elu-ates were subjected to Western blot analysis using primary antibodies as mentioned above.

3C PCR
Experiments were conducted as described previously (41). Briefly, single-cell suspensions of 1 ϫ 10 7 cells under each of the experimental conditions described were prepared. Cross-linking with 2% formaldehyde, 10% fetal bovine serum/PBS was performed for 10 min at room temperature while tumbling. On-ice quenching was achieved by adding 1 M ice-cold glycine. After isolation of the cell pellet from the supernatant, cell lysis was conducted on ice for 10 min. Lysis buffer consisted of 10 mM Tris-HCl, pH 7.5, 10 mM NaCl, 5 mM MgCl 2 , 0.1 mM EGTA, 1ϫ complete protease inhibitor (11836145001, Roche Applied Science). After centrifugation and removal of supernatant, the pellet was resuspended in restriction enzyme buffer using NEBuffer 3.1 (B7203S, NE BioLabs) or Cut Smart Buffer (B7204S, NE BioLabs) and incubated at 37°C with added SDS to a concentration of 0.3% while shaking for 1 h. Then Triton X-100 was added to a concentration of 2%, and the reaction was incubated again for 1 h at 37°C while shaking. Selected restriction enzyme, BsrI (R0527S, NE BioLabs) or BsrGI-HF (R3575S, NE BioLabs), was then added at 400 units/ml and incubated overnight at 37°C while shaking. SDS was added to a concentration of 1.6% and incubated for 25 min at 65°C while shaking. Ligation buffer was then added with Triton X-100 to a concentration of 1%, followed by incubation for 1 h at 37°C shaking. 100 units of ligase was added and incubated for 4 h at 16°C followed by 30 min at room temperature. 300 g of proteinase K was then added, and the mixture was incubated at 65°C overnight to de-cross-link. We then added 300 g of RNase and incubated for 45 min at 37°C. Phenol-chloroform was added and mixed vigorously. The supernatant was saved, and two washes were performed. The first wash was with 2 M sodium acetate, pH 5.6, distilled water, and ethanol followed by incubation at Ϫ80°C for 1 h. The pellet was then washed again with 70% ethanol. The pellet was dried and resuspended in 10 mM Tris, pH 7.5. qPCRs were set up using Prime Time Gene Expression 2ϫ Master Mix (1055772, IDT) and the primers and TaqMan probes listed in Table 6.