Identification of SOX4 target genes using phylogenetic footprinting-based prediction from expression microarrays suggests that overexpression of SOX4 potentiates metastasis in hepatocellular carcinoma

Abstract

A comprehensive microarray analysis of hepatocellular carcinoma (HCC) revealed distinct synexpression patterns during intrahepatic metastasis. Recent evidence has demonstrated that synexpression group member genes are likely to be regulated by master control gene(s). Here we investigate the functions and gene regulation of the transcription factor SOX4 in intrahepatic metastatic HCC. SOX4 is important in tumor metastasis as RNAi knockdown reduces tumor cell migration, invasion, in vivo tumorigenesis and metastasis. A multifaceted approach integrating gene profiling, binding site computation and empirical verification by chromatin immunoprecipitation and gene ablation refined the consensus SOX4 binding motif and identified 32 binding loci in 31 genes with high confidence. RNAi knockdown of two SOX4 target genes, neuropilin 1 and semaphorin 3C, drastically reduced cell migration activity in HCC cell lines suggesting that SOX4 exerts some of its action via regulation of these two downstream targets. The discovery of 31 previously unidentified targets expands our knowledge of how SOX4 modulates HCC progression and implies a range of novel SOX4 functions. This integrated approach sets a paradigm whereby a subset of member genes from a synexpression group can be regulated by one master control gene and this is exemplified by SOX4 and advanced HCC.

Introduction

Hepatocellular carcinoma (HCC) is one of the most common malignant neoplasms in humans and is prevalent among Asian populations. Molecular analyses have shown that the HCC pathogenesis is a multifactorial and multistep process reflecting alterations derived from epigenetic instability, chromosomal instability and deregulated transcription (Chen et al., 2002; El-Serag and Rudolph, 2007). HCC death has been ranked the third most common cancer death worldwide as a result of poor prognosis due to the disease’s heterogeneous nature and the high rate of intrahepatic metastasis that occurs in the dense vasculature of the liver (Tang et al., 2004; Kim et al., 2005). Recently, critical molecular signature genes were shown to be involved in organ-specific metastasis (Kang et al., 2003; Minn et al., 2005), supporting the notion that tumor cell–host interaction is the primary key to tumor metastasis as was originally postulated in the seed-and-soil hypothesis (Paget, 1889). Intrahepatic metastasis is unique in that the liver microenvironment is the milieu for HCC cells both before intravasation and after extravasation during the metastasis cascade. Thus, liver-specific signature genes for local invasion are to be expected. Initial efforts to search for genes that functionally contribute to intraheptic metastasis have identified a few candidates (Ye et al., 2003) but a comprehensive portrayal of intraheptic metastasis progression is still lacking.

We compared the expression profiles of six HCCs undergoing intraheptic metastasis and four primary HCCs. Among the differentially expressed genes identified, we further characterized a transcription factor SOX4, a member of a highly conserved transcription factor SOX (Sry-box) family known to have a characteristic DNA-binding HMG domain (Bowles et al., 2000). SOX4 has been shown to be a transcriptional activator involved in the development of the cardiac outflow tract, of pro-B cell expansion (Schilham et al., 1996) and of the central nervous system (Cheung et al., 2000). SOX4 is absent in adult normal livers (Hunt and Clarke, 1999), hence it is not involved in normal liver functions. Recently, overexpression of SOX4 has been found to be associated with several human cancer types (Aaboe et al., 2006; Liu et al., 2006; Pramoonjago et al., 2006). Although SOX4 was the first ‘classical’ transcription factor identified with separable DNA-binding and transactivation domains among the SOX family members, information on its target genes is scanty. Only two SOX4-binding motifs, 5′-AACAAAG-3′ and 5′-AACAATA-3′, have been empirically verified in the promoters of p56^lck (McCracken et al., 1997) and human CD2 (Wotton et al., 1995), respectively. SOX proteins depend on requisite partners for target specificity and combinatorial control with partner factors is the main theme for gene regulation by these proteins (Kamachi et al., 2000; Wilson and Koopman, 2002). Whether induced expression of SOX4 leads to a different and pathological transcription program in HCC compared with the normal hepatocytes from which they originate is not known. In this study, we determined a functional role for SOX4 in liver tumor progression. Furthermore, we investigated transcription regulation by SOX4 using phylogenetic footprinting to obtain target prediction. Finally, we carried out target verification by real-time quantitative PCR, chromatin immunoprecipitation assay and promoter reporter activity assay.

Results

Identification of genes associated with intrahepatic metastatic HCCs by transcription profiling

To explore the genes contributing to intrahepatic metastasis, we conducted a global transcriptome analysis using Affymatrix chips (HG_U133 Plus 2.0) with RNA samples from four primary HCC (T1), six intrahepatic metastatic HCC (T3) and two normal liver samples (HNL). We focused on a set of 600 genes that were highly expressed in T3 tissues (Supplementary Figure 1 and Table 1). This gene group was marked as T3>T1. Fourteen genes with sixfold higher expression in the T3 tissues were identified and these were enriched with adhesion molecules (NRCAM, CD24, CTHRC1 and ROBO1) and transcription factors (FOXQ1, CDCA7 L, MYBL2 and SOX4); thus, these genes are candidate genes for direct involvement in tumor metastasis (Figures 1a and b). Overexpression of SOX4 has been shown to promote apoptosis in bladder carcinoma (Aaboe et al., 2006); in contrast, RNAi knockdown of SOX4 induces apoptosis in adenoid cystic carcinoma and prostate cancer (Liu et al., 2006; Pramoonjago et al., 2006). This suggests that the SOX4 molecular activity during caracinogenesis is influenced by additional factors that are enriched in a tissue-specific manner. The requirement for SOX4 in a wide range of developmental processes (van De et al., 1993; Ya et al., 1998; Cheung et al., 2000) further argues for a synergy between SOX4 and tissue-selective partner proteins. Among the identified highly upregulated genes in intrahepatic liver tumors, SOX4 was an attractive candidate for further investigation.

Figure 1

Microarray analysis identified SOX4 as upregulated in intrahepatic metastatic liver tumors. (a) Candidate genes associated with intrahepatic metastases. Fourteen candidate genes with sixfold higher expression in T3 tissues were identified in the microarray analysis and their expression levels were verified by reverse transcription (RT)–PCR (b). (c) Further confirmation of increased SOX4 mRNA expression in an additional 61 pairs of hepatocellular carcinoma (HCC) samples (T1, n=15; T2, n=22; and T3, n=24). Results were normalized against the expression level of GAPDH mRNA in each sample. The box plot shows the data distribution across the group classification and presents the 75th and 25th percentile (upper and lower quartile) with the median value in between. N: an average of expression level of all 61 normal adjacent tissues (adjacent-average). There is a statistically significant difference between the level of SOX4 mRNA when T3 tumors are compared to their adjacent tissue and T1 tumors.

To substantiate the significance of SOX4 overexpression, further validation using a larger set of HCC samples was performed. The expression of SOX4 in T3 tumors (n=24) was significantly higher than adjacent normal tissues (n=61, adjacent-average) and T1 tumors (n=15). No statistically significant difference was found between T3 tumors and T2 tumors (n=22) suggesting that elevated expression of SOX4 correlates with tumor progression (Figure 1c).

Suppression of SOX4 expression reduces migration/invasion of HCC cells in vitro

A high level of SOX mRNA was detected in the fetal liver and in five HCC cell lines, with the exception of HepG2 (Figure 2a). This pattern of expression is reminiscent of known oncofetal antigens detected in HCC (Grizzi et al., 2007). RNA interference with VSV-pseudolentiviral shRNA was used to reduce SOX4 expression in Mahlavu, SK-Hep1 and HuH7, which are cell lines known to manifest invasion activity (Shouval et al., 1988; Ye et al., 2003). shSOX4 reduced SOX4 expression at both mRNA (Figure 2d, insets) and protein level (Figure 2b); this happened in both the parental and HA-SOX4-transfected Mahlavu cells. To overcome the poor quality of commercial anti-SOX4 antibody available, which is able to detect endogenous SOX4 protein (← in Figure 2b), the western analysis used HA-SOX4-transfected Mahlavu cells and anti-HA antibody to detect HA-SOX4 fusion protein (◂ in Figure 2b). The SOX4 protein has a molecular weight of 47 kDa, but migrated at about 70 kDa when overexpressed in the transfected cells probably due to the effect of the highly acidic serine-rich domain on protein mobility (Hur et al., 2004). Downregulation of SOX4 did not significantly affect cell growth (Figure 2c) but rather induced distinct morphological changes in the HCC cells (Figure 2d). Five to seven days after shSOX4 lentiviral infection, HuH7 cells exhibited a less-transformed pavement-like cell arrangement (Lee et al., 2003) whereas giant fused cells were noticed in the SK-Hep1 and Mahlavu cultures. Upon close examination with phallodine staining, F-actin positive membrane protrusions were significantly reduced in addition to a loss of F-actin organization in the shSOX4-treated cells (Figure 2e). The level of vimentin, a marker for the epithelial-mesenchymal transition (EMT), was drastically reduced in shSOX4-treated Mahlavu cells (Figure 3a). This suggested that elevated expression of SOX4 impacts on the EMT of HCC cells. Therefore, the effect of shSOX4 on in vitro migration/invasion was examined and a 30–65% reduction in both migration and invasion activity was found in SOX4 knockdown cells (Figure 3b). Mahlavu showed the greatest response to the SOX4 RNA interference.

Figure 2

Suppression of SOX4 expression by RNA interference resulted in morphological changes and actin organization of hepatocellular carcinoma (HCC) cells. Suppression of SOX4 in different HCC cell lines was achieved with shSOX4 lentiviral infection. A control shLuc lentivirus was also used. (a) Expression level of SOX4 in fetal liver, adult liver and six HCC cell lines detected by reverse transcription (RT)–PCR. (b) The effect of SOX4 knockdown was verified in the parental (Mahlavu) and in HA-SOX4-transfected Mahlavu (Mahlavu/HA-SOX4) cells by western blotting using anti-HA and anti-SOX4 antibodies. Cytosolic (C) or nuclear (N) proteins (30 μg) were loaded. HA: cells transfected with pcDNA3.1-HA vector. Arrowhead (◂): HA-SOX4 fusion protein; arrow (←): endogenous SOX4 protein. Total actin was used as the internal control. An antibody against actin was used as the control in this study; MAB1501 is a pan-actin antibody that can react with all six isoforms of vertebrate actin. (c) SOX4 knockdown did not affect cell proliferation of Mahlavu cells. Cell growth was measured using the MTT assay. (d) The morphologies of HuH7, Mahlavu and SK-Hep 1 cells after downregulation of SOX4 by shRNA. Knockdown of SOX4 in HCC cells was confirmed by RT–PCR (insets). The presence of altered morphologies of the HCC cells was revealed by phase contrast microscopy. Images were taken 5–7 days after infection. Original magnification: × 200. (e) Loss of actin organization in shSOX4-treated cells. Control shLuc-treated and shSOX4-treated HuH7 and Mahlavu cells were fixed and stained with rhodamine-phalloidine or 46-diamidino-2-phenyl indole (DAPI) to detect F-actin and the nucleus, respectively. Right column: merged image of F-actin and nucleus. Bar, 20 μm for shLuc-treated Mahlavu; 10 μm for other treatments.

Figure 3

Suppression of SOX4 resulted in differential expression of vimentin and reduction in migration/invasion activity of Mahlavu cells. (a) Mahlavu cells express high level of vimentin but E-cadherin was undetectable. Less expression of vimentin was detected after SOX4 knockdown (western blot and immunofluorescence). Antibodies to E-cadherin and vimentin were used. Right column of the lower panel: merged image of vimentin and nucleus. Bar, 20 μm for shLuc-treated cells; 10 μm for shSOX4-treated cells. (b) SOX4 knockdown resulted in a 30–65% reduction in migration (upper panel) and invasion (lower panel) activity in three hepatocellular carcinoma (HCC) cell lines. Cells were infected with either control shLuc or shSOX4 for 16 h, which was followed by 2-day puromycin (2 μg/ml) selection before the cells were plated for the migration or invasion assay. Cell migration and invasiveness were assessed as described in Materials and methods. Thirty-five fields were counted for every filter. Data are an average of triplicates for each condition. ^**P<0.01. White bars, Mahlavu cells; black bars, SK-Hep1 cells; gray bars, HuH7 cells.

Prediction of potential SOX4 target genes using computational approach

A transcription factor exerts its actions via regulation of downstream target genes. To fully understand the roles of SOX4 in HCC progression, a comprehensive knowledge of its targets is needed. Our assumption was that the genes regulated by SOX4 are likely to be present in the SOX4 synexpression group, namely, the T3>T1 gene group. The binding motifs of the evolutionarily conserved transcription factors tend to reside in human-mouse conserved noncoding blocks (CNBs) within the promoter regions of the putative target genes (Hardison et al., 1997; Wasserman et al., 2000). To unravel the SOX4 targets, we developed a bioinformatics screening strategy based on phylogenetic footprinting to identify SOX4-binding motifs in the T3>T1 gene group (Fickett and Wasserman, 2000). A degenerate consensus motif, WWCAAWG (A/T A/T CAA A/T G; Wilson and Koopman, 2002), which accommodates the known SOX4 binding motifs, AACAA A/T G was used. A group of 41 genes containing at least one binding motif in their CNBs was retrieved from the T3>T1 gene group (Supplementary Table 2a). In CNBs, 47 binding motifs were identified (Table 1). The motifs demonstrated a broad distribution of spatial locations. Sixteen motifs were located >1 kb upstream from the transcription start site (TSS), six sites were in intron 1 and three sites were in the 5′-UTR regions. Chromatin immunoprecipitation (ChIP) assays were performed to discover if these motifs are functional binding sites for SOX4.

Table 1 Forty-one genes in the T3>T1 group had SOX4-binding site(s) in the promoter region

Validation of the predicted SOX4 target genes using ChIP and siRNA assays

ChIP is a powerful technique for analysing transcription factor binding sites in living cells. We performed ChIP assays with Mahlavu cells transfected with HA-SOX4 using anti-HA antibody to detect sites binding the HA-SOX4 fusion protein. In parallel, control ChIPs were done using commercially available anti-immunoglobulin G (IgG) antibody. As shown in Figure 4a, SOX4 protein formed complexes with the predicted motifs of 31 genes but failed to bind to the predicted motifs of 10 genes (Table 1, genes underlined). These results indicated a 76% positive prediction rate. The specificity of the ChIP assay was further confirmed using HCC cells transfected with an unrelated HA-FOXQ1 gene construct (data not shown). Occurrences of the motif patterns in ChIP-verified target genes were tallied. Although the sample size is small, nevertheless the frequency of AACAAAG, TTCAAAG and ATCAAAG is slightly higher than all the other patterns combined (Supplementary Table 2b). Based on the percentage of each nucleotide at each position within the core motif (WWCAAWG) and the three flanking nucleotides, the consensus binding motif for SOX4 is A/T T/A CAA A/t G (Table 2a). Motif specificity was validated by comparing the SOX4 motif with the consensus SOX9 motif (Mertin et al., 1999). In contrast to SOX9, no predominant 5′ or 3′ flanking nucleotides were observed for the SOX4 binding motifs in the 31 target genes (Table 2b). This result demonstrates that the binding of the SOX4 protein is specific.

Figure 4

Identification of SOX4 targets in hepatocellular carcinoma (HCC) cells. (a) Chromatin immunoprecipitation (ChIP) assays with anti-HA antibody showed binding of HA-SOX4 to the promoters of 31 genes in Mahlavu cells transfected with HA-SOX4. All of these genes have SOX4 binding motifs in the human-mouse Conserved noncoding blocks (CNBs). LCK is a confirmed SOX4 target gene. Input: input DNA; HA: immunoprecipitation with anti-HA antibody from cells expressing HA-SOX4; IgG: immunoprecipitation with mouse immunoglobulin G (IgG) antibody was used as a negative control. The promoter of CSPG2 and TEAD2 failed to bind to SOX4. Expression of SOX4 target genes was detected using real-time quantitative PCR by the SYBR Green I protocol. All values were normalized against GAPDH mRNA. After shSOX4 knockdown of endogenous SOX4 (b) expression of 20 SOX4 target genes in Mahlavu cells was downregulated, (c) expression of 7 target genes in Mahlavu cells was upregulated and (d) no significant change in expression was found for 4 target genes. C: shLuc was used as the negative control; sh: shSOX4. In all cases, shSOX4 knockdown reduced the level of endogenous SOX4 mRNA by ⩾60%. White bars, shLuc treatment; shaded bars, shSOX4 treatment. The standard deviation is indicated. ^*P<0.05; ^**P<0.01.

Table 2 Analysis of SOX4 binding motifs in confirmed target genes

To further characterize these 31 genes as SOX4 direct targets, we analysed the expression levels of all 31 genes in Mahlavu and HuH7 cells. The expression level of 20 genes was decreased (Figure 4b), whereas that of 7 genes was slightly increased in the SOX4 knockdown Mahlavu cells (Figure 4c). However, the expression levels of four genes, MAP4, NAV3, NPNT and PAM, were not affected (Figure 4d). Unexpectedly, a reduction of SOX4 expression would seem to have a discordant effect on 13 target genes between HuH7 and Mahlavu cells (Table 3).

Table 3 Discordant gene expression after SOX4 RNA knockdown in Mahlavu and HuH7 cells

NRP1 and SEMA3C, two SOX4 target genes, regulate in vitro cell migration

Our results revealed that SOX4 was a very potent transcription factor, which switched on a considerable downstream transcriptional cascade involving multiple cellular pathways (Table 1). A detailed knowledge of the affected genes will require a comprehensive investigation. Genes implicated in mobility and/or metastasis are often associated with axon guidance, cell differentiation, Wnt signaling and microtubule dynamics. In neuronal cells, guidance molecules modulate growth cone motility through cytoskeletal changes (Guan and Rao, 2003). Both semaphorin 3C (SEMA3C) and neuron navigator 3 (NAV3) are involved in axonal guidance (Maes et al., 2002; Gonthier et al., 2007) where neuropilin-1 (NRP1) is a receptor for the semaphorins (He and Tessier-Lavigne, 1997).

Identification of SOX4 target genes in the category of axon guidance is of great interest. To clarify the SOX4 transcriptional regulatory mechanism, we conducted promoter reporter assays with the NRP1 and SEMA3C promoter regions and also generated a series of mutant constructs with a core binding-site mutation (WWCAAWG to WWTGGWG). In the presence of exogeneous SOX4, the promoter activity of NRP1 was stimulated by 2.5-fold and this was reduced significantly by a mutated SOX4 binding site (Figure 5a). The SEMA3C promoter has three motifs, one (−845 to −839) in the conserved block and two additional motifs (−917 to −911 and −1104 to −1098) in nonconserved regions. Mutations of the core binding motifs resulted in small reductions in luciferase activity but a 50% reduction was detected when all three binding sites were mutated (M3; Figure 5b). This indicates that all three SOX4 binding motifs in the SEMA3C promoter directly influence SEMA3C transcription activation. The basal levels of both NRP1 and SEMA3C promoter activity were partially reduced in SOX4 knockdown cells (Figures 5a and b). Surprisingly, shSOX4 also reduced the luciferase activities of the promoter reporters bearing the mutated binding sites. A definitive cause to this observation is not clear. The cloned promoter fragments of NRP1 and SEMA3C contain overlapping binding sites for SOX4, FOXA1, HNF4 and POU2F1. The CAA to TGG mutation may have created poor accessibility for other transcription factors because a general reduction of reporter activity was detectable under SOX4 knockdown. This effect needs further investigation. Nevertheless, this result indicates that SOX4 can directly induce NRP1 and SEMA3C transcription activation through the SOX4 binding motifs found within their promoters.

Figure 5

SOX4 target genes, NRP1 and SEMA3C regulate cell migration. The promoters of NRP1 and SEMA3C are transactivated by SOX4. HEK293 T cells were transiently co-transfected with HA-SOX4 or pcDNA3.1-HA vector and (a) wild-type pGL3-NRP1 promoter (▪) or pGL3-NRP1 promoter-mutant (), (b) wild-type pGL3-SEMA3C promoter (▪) or pGL3-SEMA3C promoter-mutants. The SEMA3C promoter constructs with the mutated core binding sequences were designated as pGL3-SEMA3C promoter-M1 (), -M2 () or M3 (). Luciferase activity was measured 48 h after transfection. The transfection efficiency was normalized against pRL-TK activity. Normalized luciferase activity from triplicate samples is presented relative to that of cells transfected with the pGL3-basic construct (□). The experiment was repeated twice with same results. ^*P<0.05; ^**P< 0.01. (c) Suppression of either NRP1 or SEMA3C expression effectively reduced cell migration activity in hepatocellular carcinoma (HCC) cell lines. The effect of shRNA was verified by reverse transcription (RT)–PCR. HuH7 cells were not treated with shSEMA3C due to the low expression level of SEMA3C in these cells. (d) Knockdown of NRP1 or SEMA3C resulted in a nearly 50% reduction of migration activity in HCC cell lines. Data are an average of triplicates for each condition. ^**P<0.01. White bars, Mahlavu cells; black bars, SK-Hep1 cells; gray bars, HuH7 cells.

To understand whether NRP1 and SEMA3C expression confers biological activity on SOX4, we examined the effect of silencing NRP1 and SEMA3C on cell migration. Knockdown of NRP1 and SEMA3C (Figure 5c) drastically reduced cell-migration activity in HCC cells (Figure 5d). This result lends support to the idea that SOX4 affects cell migration partly via regulation of its downstream target genes; thus NRP1 and SEMA3C are effectors whereas SOX4 is an upstream instigator in the chain of regulation.

SOX4 knockdown suppresses in situ tumor invasion and metastasis

To determine whether SOX4 knockdown affects metastatic potentials, we initially infected the Mahlavu and HuH7 cells with shLuciferase (shLuc) or shSOX4 lentiviruses. The infected cells were injected into the flanks of nude mice or, alternatively, orthotopically into the livers of nude mice. Knockdown of SOX4 significantly reduced Mahlavu tumor growth (Figure 6a). Invasion into deeper smooth muscles layers (Figure 6b, blue arrowheads) was only found with the shLuc tumors but not with the shSOX4 tumors. Similar findings were also observed in HuH7 cells infected with shSOX4 lentivirus (data not shown). Local invasion with visible multiple tumor foci in the liver parenchyma was observed after intrahepatic injection of shLuc-infected Mahlavu cells (blue arrowheads, Figure 6c, mice 1 and 2; Figure 6d, mouse 2) but not in mice injected with shSOX4-Mahlavu cells (Figure 6c, mice 3 and 4: Figure 6d, mouse 3, Figure 6e, mice 3 and 4). The edges of the Mahlavu tumor foci were found to be actively invading the normal parenchyma and also the major blood vessel cavities of the liver (Figure 6e, mouse 1). Only a remnant in the needle track (Figure 6e, lower-left panel, black arrow) and the residual tumor mass (Figure 6e; lower-middle panel, yellow arrowheads) were observed in mouse 3 and mouse 4, respectively.

Figure 6

Effects of SOX4 knockdown in subcutaneous tumors and orthotopic liver tumors. (a) In vivo subcutaneous tumor growth curves of shLuc and shSOX4 infected Mahlavu cells (n=7). ^* P<0.05; ^** P<0.01. (b) H&E stained images of representative subcutaneous tumors that were extracted at 6 weeks after cell injection. Note shLuc control infected Mahlavu cells showed muscular infiltration. It should also be noted that tumor cells after shSOX4 knockdown formed multiple layers of cells with a flat morphology (yellow arrowheads) but did not invade into the smooth muscle. (c) Gross morphology of livers and abdominal cavity organs from representative mice after intrahepatic injection of shLuc (mice 1 and 2) or shSOX4 (mice 3 and 4) infected Mahlavu cells. (d) Gross morphology of representative livers from mice 2 and 3 in (c) after formaldehyde fixing for one day. (e) H&E stained images of representative livers that were extracted at 7 weeks after cell injection. Metastatic tumor foci (upper-left and middle-left panels) and invasive edges (upper-right and middle-middle panels, blue arrowheads) in mice 1 and 2 are identified. In the upper-middle panel, it is important to note the presence of tumor cells invading the hepatic vessels and forming small metastatic foci. Middle-right panel shows the morphology of the disseminating tumor in the abdominal cavity. Lower-right panel shows no invasion of tumor cells into liver parenchyma. Bar represents 100 μm. White arrows represent the injection sites of tumor cells. Black arrow shows the needle track. Blue arrowheads mark the tumor foci and invasive edges of tumor cells in the shLuc group. Yellow arrowheads represent the smooth and flat edges of the tumor cells in shSOX4 knockdown group.

Our findings suggest that SOX4 plays an important role in regulating progression of liver tumors. The potential of SOX4 as a novel pathological staging marker and a therapeutic target for liver cancer merits further investigation.

Discussion

In an effort to identify genes differentially expressed in intrahepatic metastasis, we found that SOX4 is overexpressed in the liver tumors with local invasion. The pattern of expression was validated in a larger set of HCC samples and the clinical relevance of SOX4 expression explored using mechanistic studies. Empirical evidence suggested that SOX4 plays an important function in liver tumor metastasis as RNAi knockdown reduced HCC cell migration, invasion and intrahepatic metastasis in an orthotopic liver cancer model. This SOX4 function also seems to operate during breast cancer metastasis (Tavazoie et al., 2008) but not in other human cancer types (Aaboe et al., 2006; Liu et al., 2006; Pramoonjago et al., 2006). This suggests that SOX4 has diverse activities across various human cancers and that these are likely to be cell context-dependent, involving differential target activation. SOX4 knockdown neither induced apoptosis (data not shown) nor decreased cell growth in Mahlavu cells (Figure 2d). It is tempting to speculate that shSOX4 reduced in vivo tumor growth and metastasis by a failure of angiogenesis as reported for siNRP1 (Hong et al., 2007).

To delineate the impact of SOX4 overexpression in the transcription regulation of HCC, a target gene search platform incorporating doctrines frequently implemented in eukaryotic transcription regulation was developed. These were (1) co-expressed genes are likely to be coregulated (Niehrs and Pollet, 1999; Karaulanov et al., 2004) and (2) noncoding regulatory sequences tend to be evolutionarily conserved (Hardison et al., 1997; Wasserman et al., 2000). The assumption that overexpression of SOX4 modulates the transcriptional activity of member genes within its synexpression group was approached using both a bioinformatics and a biochemical strategy. Experimental evidence from gene profiling, binding-site computation prediction, ChIP verification and gene ablation led to the recognition of previously unidentified SOX4 target genes. Stringent adherence to cross-species conserved sequence paradigm improved substantially the motif prediction. In the T3>T1 group, SOX4 binding motifs were recognized in 343 genes, of which 41 genes had motifs in the human–mouse CNBs. ChIP verification further confirmed 31 genes as true targets. Although the importance of several SOX family proteins is well characterized in vertebrate developmental processes (Bowles et al., 2000), fully defined SOX target genes are few. There are four known SOX4 targets, p56^lck, CD2, TLE1 and PUMA but the SOX4 binding motifs for only p56^lck and CD2 are known (Wotton et al., 1995; McCracken et al., 1997; Liu et al., 2006). This study is the first to report 31 experimentally confirmed direct target genes for SOX4 in one publication. SOX4 is a pleiotropic transcription factor regulating a wide variety of biological processes. Two SOX4 target genes, SEMA3C and NRP1, are associated with tumorigenesis or tumor progression (Bielenberg et al., 2006; Herman and Meadows, 2007) but only their role in HCC progression is explored in this study. High levels of SEMA3C and NRP1 correlated with HCC cell migration, supporting a favorable role in promoting intrahepatic HCC metastasis; the implications of these results needs to be further tested.

A dichotomy of SOX4 transcriptional activity, whereby it acts as both an activator and a suppressor, was observed in two HCC cell lines with dissimilar differentiation statuses (Table 3). Several questions remained to be addressed. These include how SOX4 modulates the transcriptional activity of genes with divergent biological activities and how is the mode of SOX4-mediated transcription activity, either activation or suppression, determined. Combinatorial control integrating the concerted actions of multiple proteins to achieve transcription repression or activation is the key to the higher eukaryote complex regulatory networks. The molecular roles in the combinatorial assembly of transcription regulators are exemplified by the POU and SOX protein partnership and others (Kamachi et al., 2000; Wilson and Koopman, 2002; Remenyi et al., 2004). The mechanistic basis of SOX functional specificity mainly rests on context-specific binding partner proteins (Wilson and Koopman, 2002). Syntenin has been shown to directly associate with Sox4 and this association is necessary for IL-5-mediated activation of Sox4 transcriptional activity in B cells (Geijsen et al., 2001). Whether this particular cooperation occurs with other SOX4-mediated gene regulation events has not been tested. We explored the binding motifs for potential partner proteins that bind in the vicinity of the SOX4 binding site (<50 bp). Several POU2F1 binding motifs emerge as common cis elements near the SOX4 motif in the majority of SOX4 target genes identified in this study (data not shown). Whether SOX4 and POU2F1 synergistically regulate SOX4 targets is under investigation. A comprehensive analysis of the regulatory modules surrounding SOX4 binding sites ought to shed further light on the molecular basis of SOX4 regulation.

Materials and methods

Cell lines and human liver tissues

The human HCC cell lines, Mahlavu, HuH7, SK-Hep1 and human HEK293T cells were cultured as described previously (Chau et al., 2007). Paired samples of tumor/nontumorous liver tissues (71) were obtained from patients who had undergone primary HCC curative hepatic resection at Taipei Veterans General Hospital, Taiwan. The study was approved by the Committee for the Conduct of Human Research and patient informed consent was obtained. On analysis, 19, 22 and 30 of the samples were classified at stages 1 (T1), 2 (T2) and 3 (T3) HCC, respectively (AJCC, 6th edn, TNM classification). Immediately after surgical resection all tissues were snap frozen in liquid nitrogen and stored at –80 °C until use.

Analysis of gene expression

Gene expression of SOX4 in cell lines and after shSOX4 knockdown was examined using RT–PCR and standard gel electrophoresis. Expression of SOX4 target genes was detected with real-time quantitative PCR (RT-qPCR) by the SYBR Green I protocol (Bio-Rad, Hercules, CA, USA). All values were normalized against GAPDH mRNA. The primer sequences are listed in Supplementary Table 3.

Plasmid constructs

The full-length human SOX4 (NM_003107) gene was subcloned into pcDNA3.1(B) HA vector (Invitrogen, Carlsbad, CA, USA) and designated HA-SOX4. Luciferase reporter constructs containing the promoter regions of NRP1 (nucleotides −1 to −1216 with the SOX4 binding site AACAATG at position −1091 to −1097 bp) and SEMA3C (nucleotides −1 to −1173) were subcloned in pGL3-basic vector (Promega, Madison, WI, USA) and designated wild-type pGL3-NRP1 promoter and pGL3-SEMA3C promoter, respectively. SOX4 core binding site mutations (CAA) within the promoter regions of NRP1 and SMEA3C were generated using a QuickChange Site-Directed Mutagenesis Kit (Stratagene, La Jolla, CA, USA). The SEMA3C promoter has three SOX4 binding sites at −845 to −839, −917 to −911 and −1104 to −1098. The CAA of the core binding site was mutated at the −845 site (M1), at both the −845 and −917 sites (M2) and at all three sites (M3). The RT–PCR primers used in mutagenesis are listed in Supplementary Table 4.

RNA interference with shRNA

HCC cells were plated and infected with lentiviruses expressing shSOX4, shNRP1, shSEMA3C or shLuc in the presence of 8 μg/ml protamine sulfate for 24 h, which was followed by puromycin (2 μg/ml; 48 h) selection. RT–PCR and/or western blotting were performed to validate the knockdown efficiency. The shRNA constructs are described in the Supplementary Materials and methods.

Antibodies, immunoblotting, immunostaining and staining for F-actin

Immunoblotting was performed as described previously (Lee et al., 2003). Nuclear and cytosolic lysates were prepared using a CelLytic Nuclear Extraction System Kit (Sigma). Protein lysate (30 μg) was electrophoresed on 10% SDS polyacrylamide gels and transferred onto polyvinylidene difluoride membranes (Millipore, Bedford, MA, USA). The membranes were incubated with primary antibodies overnight at 4 °C and then with horseradish peroxidase-conjugated secondary antibody (PerkinElmer Life Sciences, Boston, MA, USA). The primary antibody against HA was purchased from Covance (Berkeley, CA, USA), SOX4 from Abnova (Taipei, Taiwan), vimentin from Lab Vision (Fremont, CA, USA), E-cadherin from BD Biosciences (Franklin Lakes, NJ, USA) and actin from Chemicon (Temecula, CA, USA). Signals were detected by an enhanced chemiluminescence kit (PerkinElmer, Waltham, MA, USA). The relative level of protein expression was normalized against actin. Immunostaining was performed as described previously (Chau et al., 2007). Incubation with rhodamine-phalloidin (1 U/ml; Invitrogen) was done at room temperature for 15 min followed by counterstaining with 46-diamidino-2-phenyl indole. The fluorescent images were captured using an Olympus FV1000 confocal microscope ( × 60 oil immersion lens).

Cell migration assay and invasion assay

Cells were tested for migration and invasion abilities in vitro in a Minicell (Millipore, Billerica, MA, USA). The lower side or the upper side of the polycarbonate membranes (containing 8-μm pores) of the Minicell coated with 50 μg/ml of type I collagen or 80 μg per well of Matrigel was used for migration assay or invasion assay, respectively. Cells were added to the upper chamber of a Minicell. After incubation for 16 h at 37 °C, the cells at the lower side were prepared for Gimsa stain. The level of migration or invasion was determined using a microscope at × 200 magnification. All experiments were repeated three times.

Prediction of SOX4 binding sites using cross-species comparison

A bioinformatics method that estimates the binding site conservation tendency by phylogenetic analysis of the human–mouse orthologous promoters to predict the conserved transcription factor binding sites was developed. In total, 19073 orthologous gene pairs were analysed. The promoters of the human and mouse orthologous gene pairs were retrieved from Ensembl genome database V32 (Homo sapiens core 32.35e and Mus musculus core 32.34). The promoter region was defined as a 3-kb upstream sequence from both the TSS and from the translation start site. The upstream sequences of paired genes that shared 80% conservation based on local alignment using BLAST (version 2.2.12) were identified as CNBs. The putative promoter sequences of CNBs were scanned for the SOX4 binding motif, WWCAAWG. The genes scored were classified into three categories, namely ‘Motifs in CNB’, ‘Motifs not in CNB’ and ‘No Motif’. The final category was that where no SOX4-binding motif was found.

Chromatin immunoprecipitation reactions

The ChIP assays were performed as described previously (Nelson et al., 2006). Cells were fixed with 1% formaldehyde, harvested and lysed in SDS buffer containing 50 mM Tris-HCl (pH 8.1), 0.5% SDS, 100 mM NaCl, 5 mM EDTA and protease inhibitors. The pellet was sonicated using 1 s pulses separated by 5 s, for 4 min at output level 6 using a Sonicator3000 (Misonix, Farmingdale, NY, USA). The sheared chromatin was precleared with 30 μl protein A beads (Amersham Biosciences, Piscataway, NJ, USA) followed by incubation with 5 μg anti-HA mAb or mouse IgG (PerkinElmer). Untreated sonicated chromatin was saved as input. Purified DNA was subjected to PCR reactions using the primers described in Supplementary Table 5.

Promoter reporter assay

HEK293 T cells (5 × 10⁴ per well) were seeded in 24-well plate and co-transfected with 0.25 μg of HA-control vector or HA-SOX4 construct, 0.25 μg of pGL3-NRP1 (wild-type or mutant) or pGL3-SEMA3C (wild-type or mutant) and 0.05 μg of pRL-TK (Promega) using jetPEI reagent (Polyplus-Transfection, Illkirch, France). After 48 h, the luciferase activity was measured using the Dual-Luciferase Reporter Assay System Kit (Promega).

In vivo turmorigenesis assay

Mahlavu and HuH7 cells (10⁷) infected with shLuc or shSOX4 lentiviruses were implanted subcutaneously into the flanks of the nude mice. Primary tumor growth rate was analysed by measuring tumor length (L) and width (W), and tumor volume was calculated according to V=0.4 × ab² as previously described (Attia and Weiss, 1966). Orthotopical intrahepatic injection was conducted as previously described (Lu et al., 2007). Using 20 μl of serum-free medium containing 50% Matrigel (BD Biosciences, San Jose, CA, USA), 10⁶ cells were slowly injected into the liver of anesthetized mice with a 27-gauge needle. Mice were killed 6–7 weeks later for histopathological evaluation.

Statistical analysis

All data are expressed as mean±s.d. and compared among groups using the Student's t-test. Categorical variables were compared using the Wilcoxon rank test. A P-value <0.05 was considered statistically significant.