A novel KLF6-Rho GTPase axis regulates hepatocellular carcinoma cell migration and dissemination

Abstract

The presence of invasion into the extra-hepatic portion of the portal vein or the development of distant metastases renders hepatocellular carcinoma (HCC) patients ineligible for the only potential curative options for this malignancy - tumor resection or organ transplantation. Gene expression profiling of murine HCC cell lines identified KLF6 as a potential regulator of HCC cell migration. KLF6 knockdown increases cell migration, consistent with the correlation between decreased KLF6 mRNA levels and the presence of vascular invasion in human HCC. Concordantly, single-copy deletion of Klf6 in a HCC mouse model results in increased tumor formation, increased metastasis to the lungs, and decreased survival, indicating that KLF6 suppresses both HCC development and metastasis. By combining gene expression profiling and chromatin immunoprecipitation coupled to deep sequencing, we identified novel transcriptional targets of KLF6 in HCC cells including VAV3, a known activator of the RAC1 small GTPase. Indeed, RAC1 activity is increased in KLF6 knockdown cells in a VAV3-dependent manner, and knockdown of either RAC1 or VAV3 impairs HCC cell migration. Together, our data demonstrate a novel function for KLF6 in constraining HCC dissemination through the regulation of a VAV3-RAC1 signaling axis.

Introduction

Hepatocellular carcinoma is a common malignancy worldwide that affects over 800,000 people per year, and is the 2^nd-leading cause of cancer-related death¹. Survival for patients with HCC is generally poor, with a 5-year survival rate of 16.1% in the United States. Importantly, patient prognosis is inversely correlated with tumor stage at the time of diagnosis, where the survival rate increases to 29% for patients with localized disease and eligibility for resection². However, treatment options are limited for other patients, resulting in a 5-year survival rate of 10% and 3% for patients with regional spread or distant metastasis, respectively². Given the prevalence of invasive and metastatic disease ³, and the absence of effective treatment strategies for these patients, studies defining the mechanisms underlying HCC metastasis may provide opportunities to improve the clinical outcome of a significant fraction of HCC patients.

Among the factors previously associated with HCC progression is inactivation of TP53 ^{4, 5}. Previous work from our laboratory demonstrated that while Trp53 gene deletion did not impact tumor development, yet promoted tumor progression and metastasis in a HCC mouse model, consistent with a role in HCC progression ⁶. Other studies from our laboratory demonstrated a role for insulin-like growth factor signaling in HCC cell migration and invasion ⁷. Recent expression profiling and genome sequencing approaches have identified expression changes associated with HCC development and progression^8-13. While these studies identified several factors of potential prognostic and therapeutic significance, functional validation, particularly in vivo, remains lacking for many of these identified genes. Consequently, the mechanisms controlling HCC metastasis remain unclear. Therefore, we sought to identify factors that control the dissemination of HCC cells.

Here, we demonstrate that reduced expression of the zinc finger transcription factor KLF6 is associated with increased HCC cell migration and with the presence of vascular invasion in HCC patients. In agreement, liver-specific deletion of Klf6 promotes HCC dissemination to the lungs in mice. Moreover, shRNA-mediated knockdown of KLF6 in HCC cells results in an increased activity of the RAC1 small GTPase and enhances migration in a manner dependent on its activity. Combined gene expression profiling and chromatin immunoprecipitation experiments identified VAV3, a known activator of RAC1 function, as a novel KLF6 target gene that mediates its impact on HCC cell migration. Together, these findings identify a novel function of KLF6 in regulating Rho GTPase activity, and for the first time connect KLF6 and HCC dissemination.

Results

Identification of factors associated with HCC cell migration

BL185 is a murine HCC cell line, derived from a non-metastatic p53 null tumor, with an intrinsically low level of migration¹⁴. Isolated BL185 cells that migrated through the membranes of either a migration or invasion transwell insert were selected and expanded, generating subpopulations termed BL185-M1 and BL185-I1. These subpopulations display a higher absorbance by MTS assay over time, indicative of an increased proliferation rate (Supplemental Figure 1A). Additionally, the M1 and I1 subpopulations have increased soft agar colony formation relative to the parental cell line (Figure 1A). M1 and I1 also show a ten-fold higher rate of migration than the BL185 parent cell line (Figure 1B). Since migration assays serve as a surrogate for the initial steps of metastasis, these cell lines may serve as useful models for understanding HCC dissemination in vivo.

Figure 1.

(A) Soft agar colony formation by the parent BL185 cell line and the BL185-I1 and M1 subpopulations. Data are from a representative experiment performed in triplicate. (B) Transwell migration activity of the parent BL185 cell line and isolated subpopulations. Data are from a representative experiment performed in triplicate. (C) Validation of selected candidate genes identified by gene expression microarray by qRT-PCR. Fold changes values for the BL185-I1 and M1 subpopulations were calculated relative to the parent BL185 cell line. * p<0.05, ** p<0.01, *** p<0.001 by 2-sided student t-test.

We next identified gene expression changes in common between BL185-M1 and -I1, as compared to the parent BL185 line, using Affymetrix MG 430.2 microarrays. 313 genes were identified that displayed a greater than two-fold change in expression and had an adjusted p-value <0.01. Gene Ontology (GO) term analysis included categories related to cell cycle, angiogenesis, and migration, consistent with the phenotypes described above (Supplemental Table 1). These categories contain several genes that have well-characterized roles in tumor progression and metastasis, such as DLC1 ^{15, 16} and Hif1α ¹⁷. We validated the differential expression of a subset the genes included within these GO terms by qRT-PCR (Figure 1C). Of note, direct evaluation of several factors associated with epithelial-to-mesenchymal transition (EMT) showed no significant changes (Supplemental Figure 1B)¹⁸.

KLF6 is decreased in human HCC and invasive HCC cells

One of the differentially expressed genes was Klf6. KLF6 is decreased at the mRNA level in human HCC^19-21 and has tumor suppressor activity in this tumor type²¹. KLF6 mRNA levels are also decreased in several other human cancers, including those of the prostate, lung, and breast^22-24. In prostate and breast cancers, increased expression of a dominant negative splice variant, KLF6-SV1, is associated with decreased survival and increased rates of metastasis^{25, 26}. Finally, reported KLF6 target genes, including CDH1 (encoding E-Cadherin) and MMP9, are associated with metastasis in vivo^{27, 28}. Thus, KLF6 was an attractive candidate for further analysis.

Quantitative RT-PCR confirmed a decrease in KLF6 mRNA levels in the M1 and I1 subpopulations relative to the BL185 parental cell line (Figure 1C). In agreement with KLF6 mRNA levels, all species of KLF6 protein are decreased in the M1 and I1 subpopulations (Figure 2A). Similarly, full-length KLF6 is decreased in human HCC cell lines when compared to an immortalized hepatocyte cell line, Thle2 (Figure 2B). KLF6 protein levels are also reduced in 3 out of 4 resected human HCC tumor samples compared to their matched normal liver samples (Supplemental Figure 2A). Consistent with previously published findings²¹, data from Oncomine demonstrate that KLF6 is reduced at the message level in HCC samples as compared to normal liver (Supplemental Figure 2B). Moreover, an expression profiling dataset in Oncomine revealed that HCC specimens with vascular invasion have lower KLF6 mRNA than those without vascular invasion (Figure 2C)²⁹. These data are consistent with a reported connection between expression of the dominant-negative KLF6-SV1 splice variant and vascular invasion ³⁰. Thus, reduced KLF6 levels correlate with enhanced HCC dissemination.

Figure 2.

(A) Immunoblot of KLF6 in BL185 cells and the BL185-I1 and M1 subpopulations. (B) Immunoblot of KLF6 in human HCC cell lines. The immortalized hepatocyte cell line Thle-2 is used as a control. (C) KLF6 mRNA levels in human HCC specimens with or without documented vascular invasion. Data are from the Wurmbach Liver dataset in Oncomine²⁹. (D) Immunoblot detection of KLF6 knockdown in BL185 cells infected with lentiviruses encoding KLF6-targeting shRNAs. (E) Soft agar colony formation by BL185 cells following KLF6 knockdown. (F) Transwell migration activity of BL185 cells following KLF6 knockdown. (G) Immunoblot confirming ectopic expression of KLF6-SV1. Cells infected with the pBabe-puro vector are used as a control. β-actin serves as a loading control. (H) Transwell migration activity of BL185 cells with ectopic KLF6-SV1 expression. Migration activity for BL185 cells infected with pBabe-puro is set to 100%. Data are from a representative experiment performed in triplicate. Error bars represent standard deviation. * p<0.05, ** p<0.01 by 2-sided student t-test.

To determine if KLF6 could function as a potential prognostic marker for HCC, we performed immunostaining on a human tissue microarray containing 106 human HCC specimens. We found that HCC specimens displayed a range of nuclear, cytoplasmic, and negative staining, in roughly the same proportions as the tumor-associated normal tissue (where available) (Supplemental Figure 2C, D). Neither the absence of KLF6 staining nor KLF6 localization correlated with HCC grade or overall survival (data not shown), in contrast to previous studies that demonstrated an inverse correlation between KLF6 mRNA levels and tumor stage²¹.

KLF6 is a repressor of transformation-related phenotypes

To test whether reduced KLF6 levels in HCC cells stimulated phenotypes associated with tumor dissemination, KLF6 was stably knocked down using two independent shRNAs in the BL185 cell line (Figure 2D). KLF6 knockdown increased soft agar colony formation and enhanced cell migration (Figure 2E, F). Similarly, KLF6 knockdown increased cell migration in two additional HCC cell lines (Supplemental Figure 3A), confirming that this is not a cell line-specific effect.

The KLF6 SV1 splice variant is postulated to act as a dominant-negative protein and has been previously shown to enhance transformation and metastasis in other tumor types^{25, 26}. Ectopic expression of KLF6-SV1 in BL185 cells increased cell migration by approximately 50% (Figure 2G, H), consistent with our findings following KLF6 knockdown.

Reduced expression of the KLF6 target CDH1 is associated with EMT¹⁸. Immunoblotting demonstrated that KLF6 knockdown cells have reduced E-cadherin levels (Supplemental Figure 3B). However, the levels of other EMT-associated markers are not significantly different between KLF6 knockdown cells and controls (Supplemental Figure 3B), suggesting that a classical EMT is not associated with KLF6-regulated cell migration in HCC cells.

Single-copy loss of Klf6 enhances HCC tumor formation and decreases survival

We next determined if decreased KLF6 levels promote HCC progression and metastasis in vivo using our previously described PyMT-driven RCAS-TVA HCC mouse model^{6, 14}. In this model, hepatocytes and their progenitors are uniquely susceptible to RCAS virus infection^{6, 14}. We previously demonstrated that delivery of RCAS-PyMT to Albumin-tva, Albumin-cre, Trp53^flox/flox compound mice induces the development of metastatic HCC¹⁴. We crossed a conditional Klf6 allele into this model, such that half of the resulting progeny are Klf6 heterozygous in the liver while half are Klf6 wild type (WT) ³¹. (All progeny are also Albumin-tva, Albumin-cre, Trp53^flox/flox.)

Mice were injected with RCAS-PyMT-producing DF1 chicken fibroblasts and a cohort of Klf6^flox/WT and Klf6^WT animals evaluated for tumor-free survival. We observed that Klf6^flox/WT animals had significantly reduced survival relative to their Klf6^WT counterparts (p=0.0398, Figure 3A). Necropsy demonstrated that these animals had HCC at the time of euthanasia, and proportionally more Klf6^flox/WT mice developed HCC compared to Klf6^WT mice (74% versus 38%) (Figure 3B). Analysis of RNA isolated from HCCs arising in either Klf6^flox/WT or Klf6^WT livers demonstrated that KLF6 was typically expressed at 50% lower levels in tumors arising in heterozygous livers, similar to the levels observed in non-tumor liver tissue (Supplemental Figure 4A, B).

Figure 3.

(A) Kaplan-Meier plot comparing tumor-free survival of mice of the indicated genotypes. p=0.0398 for difference between RCAS-PyMT injected Klf6^WT and Klf6 ^Flox/WT mice (calculated by Log-rank test). (B) Tumor incidence in mice of the indicated genotypes. p=0.0258 by Fisher's exact test. (C) H&E stained tissue sections from HCCs arising in Klf6^WT and Klf6 ^Flox/WT mice.

Despite increased tumor incidence and reduced survival in the Klf6^flox/WT animals, tumor burden per mouse was not significantly different than that observed in Klf6^WT mice (Supplemental Figure 4C). Moreover, Klf6 heterozygous HCCs do not display different histologies than Klf6^WT tumors (Figure 3C). HCCs arising in Klf6 heterozygous and Klf6^WT livers display similar frequencies of tumors with vacuolated cytoplasm as previously observed in the RCAS-TVA model (Figure 3C, top panels)^{6, 14}. Tumors arising in both genotypes also commonly displayed more cellular, dense, and fibrotic areas with similar frequencies (Figure 3C, lower panels). Thus, reduction of Klf6 gene dosage does not impact HCC histology.

To determine if Klf6 deletion accelerated HCC development, we assessed an independent cohort of animals euthanized at 4.5 months of age, a time before most animals succumb to HCC. We observed that Klf6^flox/WT mice had larger tumors than Klf6^WT mice at this time point (Supplemental Figure 4D). Ki67 staining demonstrated no difference in cell proliferation between tumors arising in the two genotypes (Supplemental Figure 4E). Cleaved caspase 3 staining also revealed no differences (data not shown). Therefore, we infer that the difference in tumor size at 4.5 months reflects earlier initiation of tumor development in Klf6^flox/WT livers, consistent with previous findings²¹.

Single-copy loss of Klf6 facilitates the growth of lung metastases

Despite no differences in tumor burden, proportionally more mice from the Klf6^flox/WT survival cohort developed lung metastases than from the Klf6^WT group (p=0.0453, Chi-Square test) (Figure 4A). A single metastatic focus was identified in the lungs of a Klf6^WT mouse with HCC upon serial sectioning of the tissue and H&E staining (Figure 4C). In contrast, many of the tumor-bearing Klf6^flox/WT mice had grossly visible lung metastases (Figure 4C).

Figure 4.

(A) Incidence of lung metastasis in HCC-bearing mice of the indicated genotypes. (B) Fold change in mRNA levels of genes associated with poor HCC prognosis in Klf6 ^Flox/WT HCCs as compared to Klf6^WT HCCs. p<0.0001 by F-test. (C) H&E images (25X magnification) of lung metastases observed in Klf6^WT and Klf6 ^Flox/WT mice.

HCCs isolated from the survival cohort of mice were assayed for expression of α-fetoprotein (AFP) and insulin-like growth factor 2 (IGF2), as these markers are associated with poor prognosis in HCC patients ^{32, 33}. AFP and IGF2 mRNA levels are increased in several of the HCCs from the Klf6^fl/wt group (Figure 4B) when compared to the average levels of the Klf6^WT tumors. The variance in expression in tumors from the Klf6^flox/WT group is significantly different from those in the wild-type group for both poor-prognosis genes (p<0.0001 by F-test).

Taken together with our in vitro data demonstrating that KLF6 knockdown increases HCC cell migration, our in vivo findings support a novel role for KLF6 in suppressing HCC dissemination in vivo.

Knockdown of KLF6 impacts activation of RAC1 and CDC42

We next sought to characterize the mechanism of KLF6-mediated suppression of migration and dissemination in HCC. Rho family GTPases regulate cancer cell migration and metastasis, and have been implicated in HCC cell migration and metastasis^34-36. We therefore assessed whether Rho GTPase activity is altered in HCC cells following KLF6 knockdown. We found that the levels of CDC42 and RAC1 are unchanged and the levels of RHOA are modestly decreased in KLF6 knockdown cells (Figure 5A, left). However, the activity of both CDC42 and RAC1 is increased in KLF6 knockdown cells, while RHOA activity is decreased relative to controls (Figure 5A, right). To determine the relative contributions of CDC42, RAC1, and RHOA to cell migration after KLF6 inhibition, BL185 cells with KLF6 knockdown were additionally infected with lentiviruses encoding shRNAs targeting these genes (Figure 5B). We observed that knockdown of RAC1, but not CDC42 or RHOA, reduced cell migration to baseline levels (Figure 5C). Moreover, treatment of KLF6 knockdown cells with a RAC1 inhibitor similarly reduced cell migration (Figure 5D). Combined knockdown of CDC42 and chemical inhibition of RAC1 further suppressed HCC cell migration, suggesting a context-dependent role for CDC42 in HCC cell migration (Figure 5D).

Figure 5.

(A) Quantification of relative RHO family protein levels in BL185 cells with KLF6 knockdown and non-silencing control (left panel). Quantification of the relative amounts of RHO family proteins in the active conformation (right panel). (B) Immunoblot of RAC1, CDC42, and RHOA in KLF6 knockdown cells with specific shRNAs targeting individual RHO family members. (C) Transwell migration activity of KLF6 depleted BL185 cells following subsequent knockdown of RAC1, CDC42, or RHOA. In columns 2 and 3, the indicated shRNAs are in the pGIPz vector. In all remaining lanes the indicated shRNAs are in the pLKO.1 vector. (D) Transwell migration activity of KLF6 depleted cells infected with a non-silencing control (shGFP) or shRNA targeting CDC42 and treated with the RAC1 inhibitor NSC2376 (gray bars), or water as a vehicle control (black bars). * p<0.05, ** p<0.01, *** p<0.001 by 2-sided student t-test.

The above findings suggest that KLF6 controls the expression of regulators of RAC1 and CDC42 activity. We therefore sought to identify KLF6 transcriptional targets that impact RHO activity, cell migration, and HCC metastasis.

KLF6 represses the expression of VAV3 and CDC42EP3

To determine transcriptional targets of KLF6 in HCC, we utilized two genome-wide approaches: gene expression profiling and ChIP-Seq. We performed gene expression profiling to identify genes differentially expressed upon KLF6 knockdown, utilizing three independently generated populations of KLF6 knockdown cells. Comparison with their corresponding non-silencing controls identified over 600 genes with a > 1.5-fold change in mRNA levels and an adjusted p-value < 0.01. To perform ChIP-Seq, we expressed a V5 epitope-tagged version of full-length KLF6 in BL185 cells and performed ChIP using an anti-V5 antibody. Libraries from two independent IP experiments were prepared for deep sequencing and compared to input chromatin as a control. ChIP peaks that were enriched in the KLF6 IP libraries were called and annotated to the nearest transcriptional start site (TSS). KLF6 binding was enriched around the TSS but was also found to occur at distant sites both upstream and downstream of TSSs (Supplemental Figure 5A). The location of KLF6 binding did not correlate with whether the target gene had increased or decreased expression in response to KLF6 knockdown (data not shown).

We then overlapped the gene lists from both experimental approaches to identify a collection of putative KLF6 transcriptional targets (Supplemental Table 2). Using ChIP coupled to PCR, we validated KLF6 binding to several of the putative target genes (Supplemental Figure 5B and data not shown). We also verified, using qRT-PCR, differential expression between cells with KLF6 knockdown and controls for a subset of the putative target genes (data not shown). Importantly, among the target genes identified was Cdh1, a known KLF6 target, providing validation of the robustness of the dataset ²⁷.

Among the novel KLF6 target genes were genes encoding two modulators of RHO family GTPase function, VAV3 and CDC42EP3. KLF6 binds directly downstream of Cdc42ep3, and within the Vav3 gene, indicating that KLF6 is likely directly regulating their expression. VAV3 is a guanine exchange factor for several small GTPases, including RAC1 (and potentially CDC42), and is known to drive migration in several cancer cell lines^37-39. CDC42EP3 is known to bind to active small GTPases and impact migration in fibroblasts^{40, 41}.

In BL185 cells, VAV3 mRNA and protein levels are increased following KLF6 knockdown (Figure 6A, B), however the expression of the related Vav1 and Vav2 genes is not impacted (Supplemental Figure 6A). Like VAV3, CDC42EP3 levels are also increased following KLF6 knockdown (Supplemental Figure 6B). In addition, VAV3 levels are elevated in the highly migratory BL185-M1 and I1 cell lines, whereas CDC42EP3 levels were only modestly increased (Supplemental Figure 6C). Moreover, VAV3 mRNA levels are increased in tumors generated from Klf6^flox/WT mice as compared to Klf6^WT mice, although CDC42EP3 levels did not increase in this setting (Figure 6C, Supplemental Figure 6D). In human HCC cell lines, VAV3 and CDC42EP3 levels are increased as compared to an immortalized hepatocyte line Thle2, correlating with the reduced KLF6 levels in these HCC cell lines (Figure 6D, Supplemental Figure 6E). Together, these findings suggest that VAV3, and to a lesser extent CDC42EP3, may mediate, at least in part, the enhanced HCC cell migration induced following KLF6 inhibition.

Figure 6.

(A) mRNA levels of VAV3 in BL185 cells following KLF6 knockdown. (B) Immunoblot detection of VAV3 in BL185 cells following KLF6 knockdown. (C) qRT-PCR determination of VAV3 mRNA levels in tumors induced in Klf6^WT and Klf6 ^Flox/WT mice. (D) Immunoblot detection of VAV3 in human HCC cell lines.

VAV3 and CDC42EP3 regulate HCC cell migration

To determine whether VAV3 functions downstream of KLF6 loss to stimulate HCC cell migration, we introduced lentiviral vectors encoding two independent VAV3 targeting shRNAs into BL185 cells with KLF6 knockdown and confirmed VAV3 knockdown by immunoblot (Figure 7A). Cell migration assays demonstrated that knockdown of VAV3 reduced cell migration to baseline levels (Figure 7B). Similarly, in 045-2a HCC cells with KLF6 knockdown, which also show increased VAV3 expression (Supplemental Figure 7A), concomitant VAV3 knockdown reduces cell migration (Supplemental Figure 7B). Likewise, introduction of two independent siRNAs targeting CDC42EP3 into BL185 cells reduced cell migration to baseline levels (Supplemental Figure 7C).

Figure 7.

(A) Immunoblot confirming VAV3 knockdown in KLF6 depleted BL185 cells. (B) Transwell migration activity of KLF6 depleted BL185 cells following subsequent knockdown of VAV3. (C) RAC1 activation, as detected by immunoprecipitation of RAC1-GTP and normalized to total RAC1, in BL185 cells with simultaneous knockdown of KLF6 and VAV3 and ectopic expression of wild type VAV3, VAV3 ISOIII (GEF mutant) or VAV3 W493L (PH mutant). The activity of cells with KLF6 knockdown is set to 100%. Data are the average of 3 independent experiments. (D) Transwell migration activity of the cell lines noted in (C). Migration activity for BL185 cells with KLF6 knockdown is set to 100%. Data are from a representative experiment performed in triplicate. Error bars represent standard deviation. * p<0.05, ** p<0.01, *** p<0.001 by 2-sided student t-test.

To ascertain whether VAV3-mediated activation of RAC1 activity potentially regulates HCC cell migration, we determined the impact of VAV3 expression on RAC1 activity. Concomitant VAV3 knockdown reduced RAC1 activity in BL185 cells with KLF6 knockdown (Figure 7C). Ectopic expression of wild type human VAV3 in these double knockdown cells significantly enhanced RAC1 activity, whereas GEF-inactive and pleckstrin homology domain mutant proteins failed to do so (Figure 7C) ⁴². In agreement, expression of wild type, but not mutant, VAV3 increased cell migration in double knockdown cells (Figure 7D).

Together, these data demonstrate that VAV3 is a critical factor that stimulates RAC1-mediated HCC cell migration downstream of KLF6 inhibition.

Discussion

The presence of invasive and metastatic disease renders HCC patients ineligible for any of the currently available curative options², yet, the mechanisms that control the invasive and metastatic capabilities of HCC cells remain poorly understood.

Previous work demonstrated that heterozygous deletion of the gene encoding the transcription factor KLF6 in vivo increased hepatocyte proliferation and liver mass, and deletion of Klf6 in primary hepatocytes enhanced their proliferation in vivo^{21, 30}. In mice with liver-specific deletion of Klf6, treatment with the carcinogen diethylnitrosamine resulted in increased hepatocarcinogenesis as compared to mice with intact KLF6³⁰. These data demonstrated a role for KLF6 as a tumor suppressor in HCC, yet data connecting KLF6 to HCC dissemination and metastasis were lacking. Here we demonstrate that KLF6 suppresses HCC cell migration, and that deletion of a single Klf6 allele promotes metastasis to the lungs, consistent with our observation that HCCs with vascular invasion have lower KLF6 mRNA levels than HCCs without vascular invasion. Given the frequency of invasive and metastatic disease in HCC patients, our work has important ramifications for understanding the mechanisms underlying HCC progression.

Our study is the first to demonstrate KLF6 regulation of metastasis in autochthonous mouse models, whereas prior studies that showed a role for KLF6 in regulating breast and prostate tumor metastasis utilized cancer cell lines implanted into immunocompromised mice^{25, 26, 43}. Moreover, while these prior studies utilized the manipulation of the KLF6 splice variant KLF6-SV1, our study demonstrates for the first time that reduction in KLF6 levels is sufficient to promote metastasis. Thus, the effects previously observed following ectopic KLF6-SV1 expression likely reflect dominant-negative effects and not gain-of-function properties.

Our study differs from published studies that failed to demonstrate a role for KLF6 in constraining HCC dissemination^{21, 30}. These differing results may reflect the different models employed (GEMM vs. carcinogen induced). However, they may also reflect differences in p53 gene status, as p53 is deleted in our model, but was intact in the previously published studies. Thus, the ability of KLF6 deficiency to promote HCC metastasis may be enhanced by inactivation of p53.

To elucidate the mechanisms underlying KLF6's ability to suppress HCC metastasis, we utilized gene expression profiling and ChIP-Seq to identify KLF6 regulated genes. Among the novel putative KLF6 target genes identified by our studies is VAV3, which has been shown to promote RAC1 and CDC42 activation ³⁹. Indeed, our data demonstrate that RAC1 activity is increased following KLF6 knockdown in a VAV3-dependent manner. Consistent with an important role for VAV3 and RAC1 downstream of KLF6, HCC cell migration induced following KLF6 knockdown is dependent on these factors. Thus, we have identified a novel function for KLF6 in regulating the activity of RHO family GTPases, and link this function to a metastasis-associated phenotype. Confirmation that VAV3 and RAC1 regulate HCC metastasis downstream of KLF6 inhibition will require future in vivo studies.

Our studies described here highlight a novel role for a commonly inactivated tumor suppressor in restraining HCC metastasis. They also identify for the first time a transcriptional program regulated by KLF6 in tumor cells. Further characterization of the identified target genes and their potential roles in HCC development and dissemination will provide important insight into this deadly malignancy.

Materials and Methods

Mouse breeding, tumor analysis

All mouse strains used have been previously described^{6, 31, 44, 45}. Klf6^fl/fl mice were crossed with the RCAS-TVA HCC model⁶ to obtain Albumin-tv-a, p53^fl/fl, Albumin-cre, Klf6^fl/wt mice. These mice were then crossed with Albumin-tv-a, p53^fl/fl, Albumin-cre, Klf6^wt/wt mice to obtain littermates that were either heterozygous or wild type for Klf6 to be used for direct comparison. Mice were injected in the liver at 3 days old with 2×10⁶ DF1 cells producing RCAS-GFP or RCAS-PyMT virus¹⁴ and assigned to either a survival study or designated for analysis at 4.5 months of age following genotyping. Mice were not randomly assigned to the studies. Group sizes were estimated based on investigators’ prior experience. In the survival study, mice were euthanized when illness became apparent. At 9 months of age, all remaining mice were euthanized and assessed for the presence of HCC. Kaplan-Meier statistics were performed using the Log-rank test to establish a p-value. Euthanized mice that did not have liver tumors were censored from the analysis. The pathologist was not blinded to mouse genotypes for tissue analysis.

Animals were housed in specific pathogen free facilities with abundant food and water. All studies were approved by the University of Massachusetts Medical School Institutional Animal Care and Use Committee.

Tumor volume was calculated using the formula for the volume of an ellipsoid, and tumor burden was calculated by summing the tumor volumes per mouse.

To count and measure lung metastases, paraffin-embedded lungs from each mouse with a primary liver tumor were sectioned in 200μm steps. A section from each step was H&E stained and examined for the presence of metastases to lung. Using an eyepiece graticule, lung metastases were measured at their longest and shortest axes. Areas of the metastases were calculated using the formula for area of an ellipse.

Cell culture

All mouse HCC cell lines were cultured in high-glucose DMEM (Life Technologies) with 10% FBS and 1% Pen/Strep. All mammalian cell lines were grown at 37°C. Human HCC cell lines were obtained from ATCC. All other cell lines were from Lewis lab collections.

DF1 chicken fibroblast cells were cultured in growth medium at 39°C in high-glucose DMEM (Life Technologies) with 10% FBS and 1% Pen/Strep.

Stable knockdown cell lines were generated from lentiviral delivery of shRNAs directed to a mouse target mRNA. shRNA IDs are detailed in Supplemental Table 3.

The RAC1 inhibitor NSC23766 (Millipore, 553502) was added to cells as they were plated in a transwell migration chamber to a final concentration of 50μM.

In vitro migration, soft agar colony formation, and proliferation analysis

Transwell migration assays were performed as previously described ¹⁴. Soft agar colony formation was also performed as previously described ^{14, 46}. Proliferation experiments were conducted via MTS assay in 96-well plates as previously described ¹⁴.

All experiments were performed using at least 3 biological replicates. Data presented are from a representative experiment performed in triplicate. Error bars represent the standard deviation of the dataset. Student's t-tests were used to calculate p-values.

Immunoblotting, Immunohistochemistry, and GTPase Activity Assays

Immunoblotting was performed as previously described ¹⁴. Antibodies for Western blotting were used as described in Supplemental Table 4.

Immunostaining was performed as previously described ⁶. KLF6 primary antibody (sc-7158) was added at a dilution of 1:250 in PBS with goat serum, and incubated overnight at 4°C.

Human HCC tissue microarrays were stained as described above. Samples were scored as either positive or negative for KLF6. Staining in KLF6 positive samples was classified as either primarily cytoplasmic or nuclear.

GTPase activity assays were performed as previously described ³⁶.

KLF6 Expression in human HCC datasets

The Oncomine database was used for KLF6 expression data in Figure 2C and Supplemental Figure 2B. The data in Supplemental Figure 2B were derived from the Roessler Liver-2 dataset, and features 220 normal liver and 225 hepatocellular carcinoma samples ¹⁰. Figure 2C was derived from the Wurmbach Liver dataset and features 17 non-invasive and 18 invasive hepatocellular cancers ²⁹.

Gene Expression Microarrays

Gene Expression profiling was conducted using Mouse Genome 430 2.0 arrays (Affymetrix, 900495). For profiling of the M1 and I1 populations, two independent RNA isolations were used as technical replicates. For expression profiling of KLF6 knockdown cells, three independent sets of control and shRNA infections were performed for analysis.

The RMA method in the Affymetrix package from Bioconductor ⁴⁷ was used in R to summarize the probe level data and normalize the dataset to remove across-array-variation. Log transformed data were used in subsequent analyses. Moderated T statistics in the Limma package from Bioconductor ⁴⁸ was used to determine whether a gene's expression level differs between treatments. Genes with an adjusted p-value <0.01 using the B-H method ⁴⁹ and at least 1.5 fold difference in expression were considered significantly changed.

The raw data files in addition to normalized expression data have been deposited in the NCBI GEO Archive under accession numbers GSE54757 and GSE54762.

Gene Ontology terms were determined through the use of the Gene Ontology Enrichment Analysis Software Toolkit (GOEAST) program ⁵⁰.

Quantitative RT-PCR

qRT-PCR was performed using SYBR Green (VWR, 95072) in an ABI 7300 machine using 50ng cDNA. Primers used for analysis are listed in Supplemental Table 5.

C_t values for each sample were averaged and normalized to β-actin C_t values. The Comparative C_t Method was used to calculate fold change, where fold change = 2^−ΔΔCT.

ChIP-Sequencing

BL185 cells were transfected with pcDNA3.1 constructs containing KLF6-V5 or LacZ-V5. Chromatin was isolated from transfected cells according to the protocol specified in the SimpleChIP Enzymatic Chromatin IP Kit (Cell Signaling). Either 10μL of V5 antibody (Abcam, 9116) or Rabbit IgG was added to pre-cleared chromatin for immunoprecipitation.

Immunoprecipitated DNA was prepared for deep sequencing using Illumina Genomic DNA Adapters and Primers. Libraries were sequenced on the Illumina platform using a single-end, long 76-basepair read. Sequence reads were mapped to the mm9 genome using the Illumina Genome Analyzer Pipeline. Reads that were uniquely mapped with two or fewer mismatches were retained for peak detection. Regions of KLF6 enrichment over input (referred to hereafter as “peaks”), were determined using the Model-based Analysis of ChIP-Seq algorithm (MACS, version 1.3.6 ⁵¹) with the default settings and the following modified parameters: --gsize=2500000000 --tsize=76, --bw=150, --mfold=10. Peaks were assigned to nearest genes using the ChIPpeakAnno package ⁵² from Bioconductor. ChIP-sequencing raw files and annotated peaks are accessible in the GEO Archive under accession GSE54763.

Statistics

All error bars represent standard deviation, and data meets the assumptions of the tests unless otherwise noted. For all in vitro experiments, p-values were calculated using 2-sided student t-test.