Analysis of Liver Tumor-Prone Mouse Models of the Hippo Kinase Scaffold Proteins RASSF1A and SAV1

Liver cancer is one of the leading causes of cancer-related death worldwide. The Ras association domain family 1 (RASSF1) gene is located at chromosome 3p21.3 in a region frequently deleted in human solid tumors. The RASSF1 locus encodes two major transcripts referred to as RASSF1A and RASSF1C, which code for two proteins differing in their N-termini (1). The RASSF1A transcript and protein are frequently absent from human tumors due to methylation of the RASSF1A isoform–specific promoter (1, 2). Methylation of the RASSF1A promoter often occurs at an early stage of tumor progression (3). Mice with Rassf1a gene deletion are prone to carcinogen-induced and spontaneous tumorigenesis (4–6).

Although RASSF1A encodes a Ras association domain, this protein is effectively recruited to activated Ras only through heterodimerization with its closest homolog NORE1A/RASSF5 (7). RASSF1A is a potent microtubule binding and stabilizing protein (8, 9). The noncatalytic protein RASSF1A is found in a complex with the kinases MST1 and MST2, the mammalian orthologs of the tumor-suppressive Drosophila Hippo kinase and is associated with several other proteins of the Hippo pathway (10, 11). RASSF1A inhibits the dephosphorylation of MST1/MST2 by phosphatases thus promoting an activated state of MST1/MST2 (12). The core components of the Hippo pathway are highly conserved across species. In mammals, the kinases MST1/MST2 can form a complex with another adaptor protein, SAV1, the mammalian counterpart of the Drosophila protein Salvador (13). MST1/MST2 kinases activate two downstream kinases, LATS1 and LATS2 (14). The best-studied effector of the LATS kinases is YAP, an oncogenic transcriptional coactivator (15). Phosphorylated YAP is retained in the cytoplasm, which prevents it from activating transcription (16). In this model, the Hippo kinases are negative regulators of YAP. Hippo signaling is an important pathway for inhibition of cancer development in the mouse liver. Liver-specific deletion of Mst1 and Mst2 combined, or deletion of Sav1 alone leads to liver enlargement and effective induction of hepatocellular carcinoma or mixed liver tumors in mice (hepatocellular carcinoma/cholangiocarcinoma; refs. 17–20).

However, recent genome sequencing of human hepatocellular carcinomas has failed to uncover mutations in components of the Hippo pathway (e.g., MST1/2, LATS1/2, SAV1; refs. 21–23). Although amplification and/or overexpression of YAP as well as diminished YAP phosphorylation has been seen in a fraction of human liver cancers (19, 24, 25), the overall impact of this pathway on human cancer is still unclear.

Our current report addresses the role of Rassf1a deletion on liver physiology and malignant transformation in the mouse. For comparison, we analyzed mice with single homozygous deletion of Rassf1a or heterozygous deletion of Sav1, and also created a Sav1^+/−Rassf1a^−/− double knockout mouse model to study the function of these genes in liver tumorigenesis.

Rassf1a^−/− mice were derived previously (4). We created Sav1-mutant mice by using standard gene targeting methods to delete exons 1 and 2 in 129S1 mouse embryonic stem cells. Homozygous Sav1 knockouts died in utero as reported previously (26). The Sav1 knockout colony was maintained in the heterozygous state. The Rassf1a- and Sav1-mutant mice were crossed to Cre-Deleter/129 mice (27) to remove the neomycin selection cassette. Typical genotyping results are shown in Supplementary Fig. S1. The Sav1 and Rassf1a-mutant mice were maintained on the 129S1 background. Four mouse cohorts, Rassf1a^+/+ Sav1^+/+ (wild-type), Rassf1a^−/−Sav1^+/+, Rassf1a^+/+Sav1^+/−, and Rassf1a^−/−Sav1^+/−, were established and monitored for spontaneous tumor formation, morbidity, or death. Mice were euthanized when visibly moribund or whenever a growth/tumor became manifest, and a complete necroscopy was performed. Normal appearing liver or tumor tissues were fixed in 10% neutral-buffered formalin, and the remaining parts of liver tumor tissue and adjacent normal tissue were instantly frozen in liquid nitrogen and then stored at −80°C. The formalin-fixed tissues were embedded in paraffin, and then stained with hematoxylin and eosin (H&E).

Frozen tissue was ground in liquid nitrogen by using a pestle, and the total proteins were extracted with extraction buffer [50 mmol/L HEPES, pH 7.4, 137 mmol/L NaCl, 10% glycerol, 0.1% Triton-X 100, and 1× complete protease inhibitor cocktail (Roche)]. Alternatively, soluble cytosolic fractions were prepared following a published procedure (28). Protein samples were resolved on 12% SDS–polyacrylamide gels and transferred onto nitrocellulose membranes (Bio-Rad). Nitrocellulose membranes were blocked with 5% nonfat milk (Bio-Rad) in PBST (PBS with 0.1% Tween 20) and sequentially incubated with primary antibodies and HRP-conjugated secondary antibodies (Jackson ImmunoResearch Laboratories) in 5% nonfat milk in PBST. After incubation, blots were washed three times with PBST. We visualized the immunoreactive bands with Amersham ECL Prime Western Blotting Detection Reagent following the manufacturer's instructions. Immunoprecipitation was performed as described previously (12).

Human liver cancer specimens and adjacent normal liver tissue were obtained from the City of Hope frozen tumor bank on an institutionally approved protocol. Tumor-free liver DNA from accident victims was obtained from Biochain. Genomic DNA was isolated from frozen tissue using the DNeasy Blood & Tissue Kit (Qiagen). One microgram of genomic DNA was treated with sodium bisulfite and the samples were subjected to “combined bisulfite restriction analysis” (COBRA; ref. 29) using primers as described previously (30). PCR products were digested with BstUI or TaqI restriction enzymes (New England Biolabs). These enzymes only cut initially methylated DNA after bisulfite conversion and PCR. The digested PCR products were separated by electrophoresis on 2% agarose gels.

Antibodies against phospho-IκBα (Ser32; catalog no. 2859), MST1 (#3682), YAP (#4912), phospho-YAP (Ser127; #4911), TBK1 (#3013), NF-κB1 (#13586), and phospho-p65 (Ser536; #3033) were purchased from Cell Signaling Technology. Antibodies against YAP1 (sc-15407), TAZ (sc-48805), SAV1 (sc-135394), and NF-κB/p65 (sc-109x) were purchased from Santa Cruz Biotechnology. Lamin B1 (ab16048), TUBB2A (#ab170931), and IκBα (ab32518) antibodies were obtained from Abcam. Anti-β-actin (A5441) and anti-FLAG (F7425) antibodies were from Sigma. Anti-MST2 (#1943-1) was from Epitomics, and anti-GAPDH (627408) was from Genetex. Anti-RASSF1A antibodies (M3-04 and M5-06) were raised against the N-terminal exon of RASSF1A (10).

For total RNA isolation, we followed the instruction of the Ambion Purelink RNA mini kit. cDNA was synthesized from RNA template by using SuperScript III reverse transcriptase (Invitrogen). Real-time PCR amplifications were performed in 96-well optical reaction plates with SYBR Green by using Taq DNA polymerase (Qiagen). We used Affymetrix Gene 1.0-ST microarrays for gene expression analysis of mouse liver RNAs. Microarray data were analyzed using the Bioconductor package “ArrayTools.” Differentially expressed genes were selected by using a cutoff of P ⇐ 0.005 and fold change ≥ 1.5. Heatmaps were created by using Cluster v3.0 and visualized by Java Treeview. Raw data was deposited into the GEO database (accession number GSE73911).

DNA methylation data packages (Version 13) for the The Cancer Genome Atlas (TCGA) Liver Hepatocellular Carcinoma (LIHC) project were downloaded from the TCGA data portal on Dec 22, 2015. The R package Homo.sapiens was used to obtain the genomic locations of each gene, and GenomicRanges was used to get DNA methylation probes interrogating ±1,500 bp of the genomic range of the target gene. For each heatmap, the probes (rows) are arranged by their genomic location, and samples are clustered with hierarchical clustering to reveal patterns. Normalized isoform-level RNAseq data were downloaded from Firehose (http://gdac.broadinstitute.org). We used uc003dae for RASSF1A and uc003dab for RASSF1C expression levels, and for other genes only expressed isoforms were used. Corresponding Log2-transformed gene expression levels are plotted on top of each sample with a relative scale.

Previous reports suggested that both RASSF1A and SAV1 form a complex through their C-terminal SARAH domains with the Hippo kinases MST1 and MST2 (10, 11). To study the contribution of these proteins to tumorigenesis, we generated mouse strains with germ line (whole body) deletion of Rassf1a (Rassf1a isoform-specific knockout by deletion of exon 1) and/or Sav1. Rassf1a homozygous gene knockout mice are viable and fertile (4, 5), but, for the Sav1 gene knockout, only heterozygous mice could be obtained. This is due to embryonic lethality of complete Sav1 deletion in mice as also reported previously (26). Rassf1a^−/−Sav1^+/− mice were viable and fertile. Mouse colonies of four different genotypes, Rassf1a^+/+Sav1^+/+, Rassf1a^−/−Sav1^+/+, Rassf1a^+/+Sav1^+/−, and Rassf1a^−/−Sav1^+/−, were established on the 129S1 genetic background and were observed for signs of tumor formation over extended periods of time. Overall survival was significantly different between wild-type and Rassf1a^−/−Sav1^+/− mice (P = 0.0063; log-rank test) and between Rassf1a^−/−Sav1^+/+ and Rassf1a^−/−Sav1^+/− mice (P < 0.001) but not between wild-type and Rassf1a^−/−Sav1^+/+ (P = 0.1016), wild-type and Rassf1a^+/+Sav1^+/− (P = 0.1144), or between Rassf1a^+/+Sav1^+/− and Rassf1a^−/−Sav1^+/− (P = 0.287; Fig. 1). Curiously, some Rassf1a^−/−Sav1^+/+ mice survived to very old age.

Liver tumors developed in 19 of 45 mice (42%) of the Rassf1a^−/−Sav1^+/+ genotype (Table 1). These varied from small, solitary nodules to large, multinodular masses that occupied entire lobes of the liver. The tumors were designated as hepatocellular adenomas, hepatocellular carcinoma, mixed hepatocellular adenomas (mHA), or mixed hepatocellular carcinoma (mHCC; Table 1). Microscopic features of hepatocellular adenomas and hepatocellular carcinoma were typical for these types of tumors (e.g., Supplementary Fig. S2) and diagnoses were based upon standard histopathologic criteria (31). Tumors designated as mixed hepatocellular adenomas or mHCC were morphologically distinct and did not clearly fit into any recognized classification of hepatocellular tumors. They are likely similar to tumors described in WW45^+/− (same as Sav1^+/−) mice and designated as “mixed-type” liver tumors thought to arise from hepatic progenitor (oval) cells (17). The mHA and mHCC had a component with obvious hepatocellular morphology and a significant second component of cells with small, oval to occasionally round or elongated nuclei and indistinct cytoplasm. These “small cells” formed solid sheets or irregular bands that penetrated into and sometimes surrounded nodules of neoplastic hepatocytes. The nuclei of the cells were occasionally arranged in rows or formed small, duct-like structures (Fig. 2A–D). Because this second cellular component was uniformly small, highly homogeneous, and lacked mitotic figures, the designation of mHA or mHCC was based solely upon the morphology of the hepatocellular component of the tumor. In comparison with the Rassf1a^−/−Sav1^+/+ mice, only 6 of 30 (20%) wild-type mice had hepatocellular tumors (including one microscopic tumor) and these were all hepatocellular adenomas (Table 1), indicating increased tumor susceptibility and tumor progression in Rassf1a-deleted mice (P = 0.046; χ² test). For the Sav1-targeted mice, 47 of 49 (96%) Rassf1a^+/+Sav1^+/− mice were found with liver tumors at ages greater than 1 year. Similarly, 35 of 39 Rassf1a^−/−Sav1^+/− mice (90%) developed liver tumors, suggesting that loss of one copy of Sav1 is a much more potent driver of liver tumor formation than loss of both copies of Rassf1a, making it likely that loss of Sav1 is the dominant event in driving tumorigenesis in the double knockout mice (Table 1). Most of the livers from Sav1^+/− mice showed signs of oval cell hyperplasia (Fig. 2E and Supplementary Fig. S3). This change was only occasionally observed in Rassf1a^−/−Sav1^+/+ mice and was not seen in wild-type mice. Oval cells are considered bipotential progenitors capable of differentiating into hepatocytes or cholangiocytes. However, distinct features of CC, for example, clear gland formation, mucin production, and/or desmoplastic or fibrotic reaction, were not obvious in mice with a Sav1^+/− or Rassf1a^−/− genotype although most of the livers from Sav1^+/− mice showed signs of oval cell expansion or bile duct hyperplasia (Fig. 2E and F).

Conditional biallelic deletion of the Sav1 gene in the mouse liver induces liver enlargement and liver tumor formation (17, 18). Although in our work close to 100% of Rassf1a^+/+Sav1^+/− mice or Rassf1a^−/−Sav1^+/− mice developed liver tumors after 12 months of age, we did not find any liver enlargement before tumors developed (Supplementary Fig. S4). This finding suggests that one copy of Sav1 is sufficient to prevent liver enlargement but is insufficient to prevent liver tumor formation. Complete loss of Rassf1a does not lead to liver enlargement either (Supplementary Fig. S4).

As SAV1 and RASSF1A are components and potential regulators of the Hippo pathway, we analyzed levels of MST1 and MST2, YAP protein levels and phosphorylation on serine 127 of YAP in normal-appearing livers of mice of the different genotypes at the age of 6 months (Fig. 3A). If the Hippo pathway is dysfunctional in gene-targeted mice, YAP protein should be less phosphorylated and total levels of YAP should increase owing to diminished degradation of YAP. First, we found that protein levels of both MST1 and MST2 were substantially reduced in the livers of Rassf1a^+/+Sav1⁺⁻ mice. In Rassf1a^−/−Sav1^+/+ mice, MST1 was decreased in only one of three mice and MST2 levels were similar to those in wild-type mice. In the double knockout mice, levels of MST1 were strongly decreased, but MST2 appeared at wild-type levels, suggesting complex regulation of MST levels upon loss of their scaffold proteins (Fig. 3A).

However, levels of YAP and phosphorylated YAP, their ratio, or levels of the YAP homolog TAZ did not differ noticeably in livers of all the genotypes (Fig. 3A). This suggested that although levels of MST1/2 were affected by deletion of SAV1, the standard readouts of the Hippo pathway, YAP and p-YAP, appeared largely unaltered in nontumorous liver. Comparing liver tumors developing in Sav1^+/− mice with adjacent normal liver tissue, we found much reduced levels of SAV1 protein in the majority of the tumors (Fig. 3B). Increased transcript level of the YAP target gene Ctgf is a marker of a dysfunctional Hippo pathway (32). We determined the gene expression levels of Ctgf in liver tumors developing in Rassf1a^+/+Sav1^+/− mice. Thirteen of 14 liver tumor tissues expressed increased levels of Ctgf compared with the adjacent normal tissues (Fig. 3C). The high level of Ctgf expression in two tumors that retained SAV1 protein may be due to Hippo/YAP-independent mechanisms. We did not retain sufficient numbers of frozen tumors from Rassf1a^−/−Sav1^+/+ mice to compare Ctgf levels. The data suggests that the remaining Sav1 copy or perhaps other components of the pathway in the livers of Sav1^+/− mice may eventually become disabled in founder cells giving rise to a dysfunctional Hippo pathway and producing liver tumors. However, the overall protein expression levels of YAP and phosphorylated YAP are not noticeably altered in bulk liver tissue of Sav1 heterozygous or Rassf1a homozygous mutant mice (Fig. 3A).

To identify potentially novel RASSF1A-interacting proteins, we performed immunoprecipitation of endogenous RASSF1A from HeLa cells. Coprecipitating proteins were identified by mass spectrometry (Fig. 4A). The most significantly enriched proteins were three kinases, MST1, MST2, and TBK1 (Tank binding kinase 1). Figure 4B shows coimmunoprecipitation of endogenous RASSF1A and TBK1. TBK1 is an oncogenic noncanonical IκB kinase, which phosphorylates IκB to release NF-κB, which can induce the transcription of various proinflammatory and antiapoptotic genes (33, 34). We then analyzed protein expression levels of TBK1 in Rassf1a^−/−Sav1^+/+ mouse liver tissues and found that TBK1 protein expression was consistently and substantially increased in all Rassf1a knockout livers compared with the wild-type livers at the same age (18 months; Fig. 4C). A similar increase of TBK1 levels was consistently found at earlier and later ages of the mice and also was seen in Rassf1a^−/−Sav1^+/− double knockout mice but not in Rassf1a^+/+Sav1^+/− mice (data not shown).

The best-studied functions of TBK1 are through regulating the classical NF-κB pathway via p65 phosphorylation. Therefore, we determined the total protein expression of p65, its phosphorylated form p-p65 and its binding partner, NF-κB1/p50, which complexes with p65 and IκBα in the cytoplasm. p65 or p-p65 did not show a notable increase in Rassf1a^−/−Sav1^+/+ mice. Also, p105 and its truncated form p50 did not consistently show higher protein levels upon Rassf1a deletion (Fig. 4C).

We isolated RNA from livers of the four mouse genotypes at 1 year of age processing four biologic replicates for each genotype. The RNA was analyzed on Affymetrix 1.0-ST microarrays. The four genotypes could be well separated by heatmaps (Fig. 5A). Using a P value cutoff of P < 0.005, we found a relatively small number of upregulated or downregulated genes in Rassf1a^+/+Sav1^+/− mice relative to wild-type mice (Fig. 5B; Supplementary Tables S1 and S2). As expected, levels of Sav1 transcripts were reduced to about 50%. However, at the same significance level, the transcriptomes of Rassf1a^−/−Sav1^+/+ and Rassf1a^−/−Sav1^+/− mice were more substantially altered in comparison with those of Rassf1a^+/+Sav1^+/− mice (Fig. 5B; Supplementary Tables S1 and S2). There was some overlap between the genotypes. Most notably, expression differences in both directions of Rassf1a^−/−Sav1^+/+ and Rassf1a^−/−Sav1^+/− mice relative to wild-type were more similar to each other than to Rassf1a^+/+Sav1^+/− mice (Fig. 5B; Supplementary Table S2).

Gene ontology analysis did not provide any enrichment for specific gene or functional categories in the Rassf1a^+/+Sav1^+/− mice. In contrast, the same analysis for the Rassf1a-deleted genotypes uncovered enrichment of two major functional categories: the circadian clock and microtubule polymerization (Supplementary Fig. S5). This analysis did not reveal a gene category related to the NF-κB pathway or to inflammatory responses in Rassf1a^−/− genotype livers that could perhaps be traced back to increased levels of TBK1. Interestingly, however, we found that several β-tubulin genes, including Tubb1, Tubb2a, Tubb2b, and Tubb5 were among the top 60 upregulated genes in Rassf1a knockout livers (Supplementary Table S1). These β-tubulin genes were significantly upregulated (up to 4-fold) in Rassf1a^−/−Sav1^+/+ and Rassf1a^−/−Sav1^+/− mice but not in Rassf1a^+/+Sav1^+/− livers (Fig. 5C; Supplementary Table S1). In fact, most β-tubulin genes were upregulated except for Tubb3 and Tubb4a, which are β-tubulin isoforms preferentially expressed in the nervous system. Genes coding for α- and γ-tubulin proteins were not significantly different in the knockouts. At the protein level, TUBB2A was increased only slightly in livers of Rassf1a knockout mice (Fig. 5C).

The enrichment of circadian factors in Rassf1a-deleted livers (Supplementary Fig. S5) was unexpected. One differentially expressed gene was Usp2. The ubiquitin-specific protease USP2 is a key driver of the circadian clock (35). Interestingly, USP2 is also a binding partner and negative regulator of the Rassf1a-interacting kinase, TBK1 (36). Although Tbk1 itself was not transcriptionally misregulated in Rassf1a knockout mice, we observed that at the transcript level, factors regulating TBK1 protein stability, including Usp2 and Dtx4, were dysregulated. Usp2 was the second most significantly altered gene in Rassf1a^−/−Sav1^+/+ livers (downregulated ∼16-fold; Supplementary Table S1). Dtx4, the sixth most significantly altered (upregulated) gene (Supplementary Table S1) in this genotype is an E3 ubiquitin ligase and key suppressor of TBK1 targeting this kinase for degradation (37). The data suggest that a regulatory feedback loop may exist between RASSF1A, TBK1, and its modifiers DTX4 and USP2, which is perturbed upon loss of RASSF1A.

Promoter hypermethylation of tumor suppressor genes is a potential contributor to tumor formation. As mutations in genes of the Hippo pathway are uncommon in human cancers (16), including liver cancer, we analyzed several genes of the pathway for methylation at their promoters in 22 pairs of human normal tissue and tumor tissue of the liver (Supplementary Figs. S6 and S7; Supplementary Table S3). Liver cancer patients often have predisposing conditions including cirrhosis, HBV, or HCV (Supplementary Table S3). We analyzed promoter methylation of the target genes by using COBRA on normal tissues and tumor tissues (Supplementary Figs. S6 and S7). Seventeen of 22 (77%) of the liver cancer patients had higher RASSF1A methylation in the tumor tissues compared with the adjacent normal tissues (Supplementary Fig. S6A). Partial methylation of RASSF1A was seen in some tumor-adjacent livers and in liver DNA from an 83-year-old tumor-free liver. Our RASSF1A promoter methylation data are consistent with other groups' earlier findings showing extensive methylation of RASSF1A in hepatocellular carcinoma (38–40). However, we detected no methylation at the SAV1 promoter (Supplementary Fig. S6B). Furthermore, we found no methylation in human liver tumors at the promoters of the genes MST1 and MST2 (Supplementary Fig. S7).

To support this analysis, we analyzed hepatocellular carcinoma DNA methylation data from TCGA (Fig. 6). This analysis shows methylation of the RASSF1A but not the RASSF1C or SAV1 promoters in hepatocellular carcinoma in the TCGA dataset. The more highly methylated tumors (indicated by a dotted line in Fig. 6A) had the lowest expression of RASSF1A. Although the TCGA sample number for normal liver is still limited, there seems to be a trend for lower expression of SAV1 in some hepatocellular carcinomas. However, about half of the hepatocellular carcinomas still express high levels of SAV1 (Fig. 6B). Analysis of TCGA data for MST1 and MST2 also show a lack of methylation at their promoter-associated CpG islands (Supplementary Fig. S8). MST1 is expressed at rather high levels and is downregulated in very few hepatocellular carcinomas. Unexpectedly, MST2 is upregulated in about half the hepatocellular carcinomas relative to normal liver tissues (Supplementary Fig. S8B). These data suggest that the RASSF1A gene is frequently and extensively methylated and silenced in human liver tumors but that SAV1 or MST1/2 are less commonly affected.

In this study, we created three mouse models in which scaffold protein binding partners of the mammalian Hippo kinases MST1 and MST2 were deleted. Liver-specific deletion of Mst1 and Mst2 together (18–20), as well as liver-specific homozygous deletion of Sav1 (17, 18), produced hepatomegaly and effectively induced hepatocellular carcinomas in mice. These deletions increase the pool of hepatic progenitor cells (oval cells), a phenomenon we also observed in our Sav1^+/− mice and to a lesser extent in Rassf1a^−/− mice. Sav1^+/− mice also developed liver cancer in a previous study (17). It remains unknown why constitutive deletion of Sav1 in the mouse has such a prominent effect on liver tumor formation and rarely affects other tissues even though YAP, its best-studied downstream effector, plays a general role in many cell types.

In our comparative mouse studies, we determined that homozygous deletion of Rassf1a causes liver tumor susceptibility. In previous studies, liver tumors were not observed as a common type of tumor in Rassf1a^−/− mice (4, 5). These mice were on a different genetic background (129SvJ × C57BL/6). The heterozygous deletion of Sav1 was much more potent in promoting liver cancer (Table 1) even though both RASSF1A and SAV1 are binding partners and activators of the MST kinases (10, 14, 41). Although we are unable to offer a definitive explanation for these differences in tumor susceptibility between the Sav1 and Rassf1a deleted genotypes at this time, one likely possibility is the redundancy of the RASSF protein family. While there is only one SAV1-like protein in mammalian genomes, there are 10 RASSF proteins defined by the presence of a RAS association domain (42). Six of these proteins (RASSF1-6) have a C-terminal RAS association domain. The proteins most closely related to RASSF1 are RASSF3 and RASSF5 (also known as NORE1). We found that the Rassf5 gene was upregulated 1.24-fold in the liver of Rassf1a knockout mice (Supplementary Table S1) but other Rassf genes were not changed. Deletion of multiple members of the C-terminal Rassf family may be necessary to elicit a more dramatic phenotype in animal models.

Mice with heterozygous deletion of Sav1 or homozygous deletion of Rassf1a did not show any signs of liver enlargement prior to development of tumors in contrast to mice with conditional ablation of Sav1 or Mst1 and Mst2 in the liver (17–20) or liver-specific overexpression of Yap (43). These data suggest that the core Hippo pathway restraining tissue overgrowth by inhibiting YAP activity in the liver is functional in the absence of Rassf1a or in presence of only one copy of Sav1. However, these deletions led to the formation of liver tumors. It is likely that the remaining functional copy of Sav1 is lost or disabled in tumor-initiating cells of Sav1^+/− livers. The downstream consequences are activation of YAP-dependent expression of progrowth genes as exemplified by increased expression of Ctgf (Fig. 3C). The phenotype of Rassf1a homozygous deletion is milder than that of Sav1 single copy deletion but is still pronounced. It can be hypothesized that loss of Rassf1a cooperates with other tumor-driving events in liver cancer formation as previously demonstrated for hematologic malignancies (6).

Besides its involvement in the Hippo pathway, RASSF1A has additional functions that when impaired may contribute to malignant transformation. As an established microtubule binding and stabilizing protein (8), loss of RASSF1A may interfere with events such as cell migration (44) and mitosis (8, 10, 45, 46). Defects in cell migration or mitosis are likely contributors to oncogenesis. Interestingly, we observed that loss of Rassf1a leads to increased expression of a number of β-tubulin isoform-encoding genes including Tubb1, Tubb2a, Tubb2b, and Tubb5. This finding suggests that loss of microtubule stabilization by RASSF1A may invoke compensatory events by increasing transcription of β-tubulin genes. However, this compensation may be insufficient contributing to the tumor-promoting phenotype of Rassf1a loss.

We identified the kinase TBK1 as a strong RASSF1A-interacting protein (Fig. 4). This interaction was further proven to be relevant in vivo by the clearly enhanced levels of TBK1 protein in every liver sample obtained from mice with Rassf1a deletion. The absence of a protein commonly leads to diminished levels of its binding partner(s) owing to reduced protein stability in absence of the partner. However, this is not the case with TBK1, which showed increased levels upon deletion of Rassf1a. One possible explanation is that RASSF1A promotes degradation of TBK1 upon binding, a scenario, which requires additional investigation. In any event, the data suggest that RASSF1A is a negative regulator of the oncogenic kinase TBK1 because in the absence of RASSF1A, the levels of this kinase are increased. TBK1 is best known as an activator of the NF-κB pathway and has oncogenic properties (33, 34). Deletion of Rassf1a did not lead to a noticeable activation of the pathway and of NF-κB target genes in the liver (Fig. 4; Supplementary Table S1). We also did not observe an increase of inflammatory cells in sections of livers from Rassf1a^−/− mice. For these reasons, NF-κB pathway–independent functions of TBK1 may be important for tumor promotion in the liver and these may be increased upon loss of Rassf1a. Mouse models have shown that TBK1 is critical in protecting the liver from apoptosis (47). TBK1 can directly activate AKT by phosphorylating the activation loop (48). Furthermore, it has been shown that TBK1 is a mitotic kinase that phosphorylates and activates Polo-like kinase 1 (PLK1; ref. 49). Interference with mitotic progression is generally a potential tumor-promoting event. Importantly, a role of RASSF1A and the Hippo pathway in mitotic exit and prevention of polyploidy has been proposed previously (10), suggesting that loss of RASSF1A may have a multipronged effect on correct execution of mitosis, which may explain aspects of the tumor-predisposing phenotype of Rassf1a loss.

Finally, our studies add information to the possible role of the Hippo pathway in human liver cancer. Several mouse models with deletions of core components of the pathway clearly suggest that failure of Hippo signaling promotes liver cancer (50). However, mutations of SAV1, RASSF1A, MST1, MST2, or LATS1, LATS2 are extremely rare in human cancer (16). Although upstream Hippo signaling regulators such as FAT1, FAT4, or NF2 may be mutated in certain types of malignancies, they are not commonly mutated in human liver cancer either. Therefore, one may consider epigenetic inactivation of Hippo pathway tumor suppressor genes as a potential tumor-driving event. However, our own studies (Supplementary Figs. S6 and S7) along with previous data (38–40) and data obtained from TCGA (Fig. 6 and Supplementary Fig. S8) suggest that the only Hippo pathway gene commonly silenced by epigenetic processes in hepatocellular carcinoma is RASSF1A. This is particularly puzzling inasmuch as epigenetic inactivation of SAV1 would be expected to be a much more efficient tumor-promoting event than silencing of RASSF1A, given the data from our mouse models. Although SAV1 seems to be downregulated in a subset of hepatocellular carcinoma, MST2 is overexpressed rather than downregulated in many of the TCGA samples. Furthermore, although Sav1-mutant livers are characterized by an expansion of the oval cell pool, recent data has cast doubt onto whether these progenitor cells can be considered general precursors of hepatocellular carcinoma in various liver cancer models, as opposed to mature hepatocytes, which may undergo dedifferentiation (51). Future studies are needed to resolve the complex roles of the Hippo pathway in human hepatocellular carcinoma.

S.L. Vonderfecht has provided expert testimony for UCLA. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Guo, G.P. Pfeifer

Development of methodology: L. Liu, R. Dammann, S.L. Vonderfecht

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Zhang, C. Guo, X. Wu, L. Liu, W. Tsark, S.L. Vonderfecht

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Zhang, X. Wu, A.X. Li, L. Liu, H. Shen, S.L. Vonderfecht, G.P. Pfeifer

Writing, review, and/or revision of the manuscript: X. Zhang, X. Wu, A.X. Li, L. Liu, R. Dammann, S.L. Vonderfecht, G.P. Pfeifer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.L. Vonderfecht

Study supervision: G.P. Pfeifer

This work was supported by a grant of the University of California Tobacco-Related Disease Research Program (17RT-0116) and by NIH grant CA084469 to G.P. Pfeifer.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.