If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Division of Hematology/Oncology, UPMC Children's Hospital of Pittsburgh, Pittsburgh, Pennsylvania 15224Tsinghua University School of Medicine, Beijing 100084, China
Division of Hematology/Oncology, UPMC Children's Hospital of Pittsburgh, Pittsburgh, Pennsylvania 15224University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania 15224
Division of Hematology/Oncology, UPMC Children's Hospital of Pittsburgh, Pittsburgh, Pennsylvania 15224University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania 15224
University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania 15224Department of Pathology, UPMC Children's Hospital of Pittsburgh, Pittsburgh, Pennsylvania 15224Pittsburgh Liver Research Center, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania 15213
To whom correspondence should be addressed:Division of Hematology/Oncology, Children's Hospital of Pittsburgh of UPMC, Rangos Research Center, Rm. 5124, 4401 Penn Ave., Pittsburgh, PA 15224. Tel: 412-692-6795
Division of Hematology/Oncology, UPMC Children's Hospital of Pittsburgh, Pittsburgh, Pennsylvania 15224University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania 15224Pittsburgh Liver Research Center, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania 15213Department of Microbiology and Molecular Genetics, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania 15213Hillman Cancer Center, University of Pittsburgh, Pittsburgh, Pennsylvania 15232
7 Please note that the JBC is not responsible for the long-term archiving and maintenance of this site or any other third party hosted site. 1 These authors contributed equally to this work. 2 Supported by a joint visiting medical student research program between UPMC and Tsinghua University School of Medicine, Beijing, China. 3 Present address: Dept. of Internal Medicine, Lankenau Medical Center, Wynnewood, PA 19096. 4 Supported by the Dean's Summer Research Program (DSRP) of the University of Pittsburgh School of Medicine. 6 The abbreviations used are:
HBhepatoblastomaFAPfamilial adenomatous polyposisCRCcolo-rectal cancerHCChepatocellular carcinomaYAPyes-associated proteinSBSleeping BeautyHDTVIhydrodynamic tail-vein injectionCFcrowded fetalPDCpyruvate dehydrogenase complexOxphosoxidative phosphorylationFAOfatty acid oxidationmtDNAmitochondrial DNAETCelectron transport chaineGFPenhanced GFPTCAtricarboxylic acidIPAIngenuity Pathway AnalysisntnucleotidesOCRoxygen consumption rateGAPDHglyceraldehyde-3-phosphate dehydrogenase.
Hepatoblastoma (HB) is the most common pediatric liver cancer. Although long-term survival of HB is generally favorable, it depends on clinical stage, tumor histology, and a variety of biochemical and molecular features. HB appears almost exclusively before the age of 3 years, is represented by seven histological subtypes, and is usually associated with highly heterogeneous somatic mutations in the catenin β1 (CTNNB1) gene, which encodes β-catenin, a Wnt ligand–responsive transcriptional co-factor. Numerous recurring β-catenin mutations, not previously documented in HB, have also been identified in various other pediatric and adult cancer types. Little is known about the underlying factors that determine the above HB features and behaviors or whether non-HB–associated β-catenin mutations are tumorigenic when expressed in hepatocytes. Here, we investigated the oncogenic properties of 14 different HB– and non-HB–associated β-catenin mutants encoded by Sleeping Beauty vectors following their delivery into the mouse liver by hydrodynamic tail-vein injection. We show that all β-catenin mutations, as well as WT β-catenin, are tumorigenic when co-expressed with a mutant form of yes-associated protein (YAP). However, tumor growth rates, histologies, nuclear-to-cytoplasmic partitioning, and metabolic and transcriptional landscapes were strongly influenced by the identities of the β-catenin mutations. These findings provide a context for understanding at the molecular level the notable biological diversity of HB.
2017 PRETEXT: radiologic staging system for primary hepatic malignancies of childhood revised for the Paediatric Hepatic International Tumour Trial (PHITT).
). Other features associated with inferior long-term survival include older age at the time of diagnosis, prematurity, low levels of circulating α-fetoprotein, and high levels of and/or more intense nuclear localization of β-catenin in the tumor (
). Predisposing genetic disorders such as Beckwith–Wiedeman syndrome and familial adenomatous polyposis (FAP) also significantly increase the relative risk of HB development (
). Nonetheless, >80% of HBs contain somatically acquired CTNNB1 gene mutations, and at least one of these has demonstrable in vivo tumorigenicity in mice (
). CTNNB1 encodes β-catenin, a transcriptional co-activator that mediates signaling via the Wnt pathway. In quiescent cells, β-catenin resides in the cytoplasm in a transcriptionally inert form as a short-lived member of the APC complex wherein it is sequentially phosphorylated on several neighboring N-terminal serine and threonine residues and thus marked for rapid polyubiquitinylation and proteasome-mediated degradation (
). Extracellular Wnt ligand signaling recruits the APC complex to the inner surface of the plasma membrane, causing the release of unphosphorylated β-catenin that cannot be recognized by the degradation complex. The now stabilized β-catenin translocates to the nucleus, where it associates with Tcf/Lef transcription factor family members, engages in transcriptional cross-talk with the Hippo tumor suppressor pathway, and drives tumorigenesis by deregulating a large array of targets, including such well-known oncogenes as CCND1 and MYC (
). Mutant forms of β-catenin, often involving one or more of its phosphorylation sites, prevent its association with the APC complex, its ubiquitination, and its proteasomal degradation. Instead, the now stabilized mutant β-catenin is constitutively redirected to the nucleus in a Wnt-independent manner. A small subset of HBs with WT β-catenin harbor acquired mutations in other APC complex components, including Axin1, Axin2, and APC itself, thus leading to the same failure of β-catenin phosphorylation, ubiquitinylation, and rapid proteasomal turnover (
). Indeed, it is the germ-line mutational inactivation of APC in FAP, leading to the development of polyposis and colo-rectal cancer (CRC) later in life, that is responsible for the greatly increased risk of HB tumorigenesis during early childhood in these same individuals (
). Whereas the majority of sporadic CRCs possess acquired APC mutations, those that do not often harbor β-catenin mutations identical to or distinct from those described in HBs (
). These findings underscore the intimate molecular relationship between these seemingly disparate pediatric and adult cancers.
Unlike most oncogenes, whose mutational repertoires tend to be quite restricted, β-catenin mutations are highly heterogeneous, comprising both missense substitutions and variably sized in-frame deletions (
Please note that the JBC is not responsible for the long-term archiving and maintenance of this site or any other third party hosted site.
Most occur within the ∼20-amino acid region encompassing the above-mentioned phosphorylation sites. However, a number of less frequent ones reside elsewhere, particularly within the so-called armadillo repeat region (residues ∼141–664), which engages with important cooperating proteins, such as E-cadherin, Tcf4/Lef, and APC (
). These observations suggest that β-catenin's oncogenic activation may have different mutant-specific consequences and be associated with distinct tumor features, behaviors, and outcomes. Recurrent β-catenin mutations are also found in other cancers, including but not restricted to hepatocellular carcinoma (HCC), medulloblastoma, and ovarian and prostate cancer, with many of the mutations not having been previously described in HBs (
Clinically relevant subsets identified by gene expression patterns support a revised ontogenic model of Wilms tumor: a Children's Oncology Group Study.
). Most HBs, HCCs, cholangiocarcinomas, and CRCs show high-level nuclear localization of yes-associated protein (YAP), a terminal effector of the Hippo pathway (
) but showed that upstream compromise of the pathway may nonetheless occur via p19ARF silencing.
The foregoing observations raise several important questions, including the extent to which different β-catenin mutations dictate tumor growth rates and histology. Do they also differentially affect tumor metabolism, which is dramatically altered from that of the normal liver (
)? More directly, how do these mutations affect β-catenin expression, its nuclear:cytoplasmic partitioning, and tumor-specific gene expression profiles (
)? Finally, can non-HB–associated β-catenin mutants generate liver tumors, and, if so, do they differ in any discernible ways from those caused by HB-associated mutants? Answers to these questions could also explain the degree to which the mutant identities influence tumor stage, chemotherapeutic response, the likelihood of recurrence, and the prospects for long-term survival and cure.
To address these issues, we relied here on the above-described mouse model of HB in which Sleeping Beauty (SB) vectors encoding mutant forms of β-catenin and YAPS127A are stably expressed in murine livers following their delivery by hydrodynamic tail-vein injection (HDTVI) (
). We compared tumors generated by each of 14 patient-derived β-catenin variants and WT β-catenin protein. The selected mutants were a combination of those associated with HBs only, those associated with tumors other than HBs, and those shared by HBs and other tumors. We found tumor growth rates, histologic subtypes, β-catenin levels, and nuclear:cytoplasmic distribution, along with metabolic and transcriptional profiles to be profoundly influenced by the identity of the β-catenin mutant. Those previously identified only in non-HB tumors as well as WT β-catenin were also shown to be oncogenic, although the latter was associated with low-level expression, slower growth rate, and a marked HCC-like histology. Our studies identify the extent to which numerous tumor properties are influenced by the nature of the driver β-catenin mutant and provide a rationale for considering β-catenin mutational status in the initial clinical evaluation.
Results
β-Catenin mutants determine HB growth rates, survival, and histology
The previously described Δ(90) β-catenin mutant, encoding an N-terminal, in-frame 90-amino acid deletion was used as the standard against which all other β-catenin mutants were compared (
) (Fig. 1A and Table S1). All mutants were patient-derived and recurrent and were selected based on their location within the β-catenin molecule, whether they were missense or in-frame deletions, and whether they had been previously described in HBs only, non-HB tumors only, or both HBs and non-HB tumors. Δ(90) was Myc epitope–tagged, but all others were C-terminally His6-tagged to allow for comparisons of their expression levels. Purified plasmid DNAs, along with another SB vector encoding YAPS127A and a vector encoding SB translocase, were introduced via HDTVI into 10–15 FVB mice (
). The vast majority of animals developed tumors, and Kaplan–Meier survival curves identified three distinct groups (Fig. 1B and Fig. S1A). Δ(90)-associated tumors, along with those generated by five additional mutants, comprised “Group 1” tumors with a combined median survival time of 87.4 ± 6.9 days. Group 2 tumors, generated by six other mutants, showed combined median survival of 124.1 ± 6.1 days, and Group 3 tumors, generated by the R582W and Δ(36–53) mutants and WT generated tumors with combined median survivals of 210.2 ± 17.7 days. Notably, S33A tumors were associated with significantly shorter median survival than S33Y tumors, and S45A tumors were associated with shorter median survival than S45F and S45P tumors (Fig. 1B and Fig. S1B). The tumorigenicity of WT β-catenin was unsurprising and consistent with its being the primary oncogenic driver in FAP, in most sporadic CRCs, and in HBs harboring APC, Axin1, or Axin2 mutations (
). These initial results indicated that tumor growth rates and survival were dictated by the identities of the β-catenin mutants, that mutants not previously associated with HBs could nonetheless promote liver tumorigenesis, and that, in some cases, alternate missense mutations of the same amino acid could initiate tumors with quite different properties.
Figure 1.Properties of tumors induced by β-catenin mutants.A, relevant landmarks in the β-catenin protein. The majority of the mutations described in the current report reside within the phosphorylation domain (approximately residues 30–50) (Table S1). The mutations N335I, N387K, and R582W are within the armadillo-repeat region that interacts with known partner proteins such as Tcf4, E-cadherin, and APC (
). B, Kaplan–Meier survival curves of mice with HBs driven by the indicated β-catenin mutants. Three tumor groups were identified based on differences in median survival (Fig. S1A). C, immunoblots of total β-catenin levels in whole-tumor lysates. See Fig. S2 for additional sets of tumors. Note that, because Δ(90) is Myc epitope–tagged rather than His6-tagged, it is not detected here. D, total transcript levels for human and murine β-catenin. Bars, mean levels from five individual livers or tumors ± S.E. (error bars). E, p values of each pair-wise comparison from the graph depicted in D. F, relative copy number of SB-β-catenin vectors stably integrated into tumor DNAs from each of the indicated tumor groups. Copy numbers were determined using a TaqMan-based assay and were normalized to the nuclear gene encoding apolipoprotein B. G, p values for all pair-wise comparisons shown in F. H, nuclear:cytoplasmic distribution of β-catenin mutants and WT β-catenin in tumors. Four tumors (T1–T4) were evaluated for each mutant. Except for Δ(90), which was detected with an anti-Myc antibody, all β-catenins were detected with an anti-His6 tag antibody. C, cytoplasmic fraction; N, nuclear fraction. I, quantification of the proportion of β-catenin from F that is nuclear. 4–6 tumors from each group were used to obtain the results. J, p values for all pair-wise comparisons shown in I. K, WT β-catenin localizes mostly to the cytoplasm in rapidly growing Δ(90) tumors. Mice were co-injected with equivalent amounts of SB vectors encoding Myc epitope–tagged Δ(90) β-catenin and His6 epitope–tagged WT β-catenin. The resulting tumors, all of which appeared rapidly (Group 1) (B), were fractionated as described in H.
We next examined β-catenin protein levels in representative tumors from each group. As seen in Fig. 1C and Fig. S2A, considerable variation was observed, with two of the slowest growing groups, namely Δ(36–53) and WT (Group 3, Fig. S1A), expressing the lowest levels. On the other hand, S33A and K335I tumors also expressed low β-catenin levels but were among the fastest growing (Group 1, Fig. 1B and Fig. S1A). Finally, the slow-growing R582W tumors expressed relatively high levels of mutant protein (Fig. 1C). Thus, tumor growth rates and survival did not necessarily correlate with β-catenin protein levels.
Finally, to assess the degree to which each FLAG-tagged mutant β-catenin was expressed relative to endogenous murine β-catenin and to ascertain the level of expression of Myc-tagged Δ(90) (
), we performed immunoblotting on total tumor or normal liver lysates using an anti-β-catenin antibody. These results showed that all mutant β-catenin proteins were expressed at levels that were about equivalent to or modestly higher than that of the endogenous murine protein (Fig. S2B). Δ(90), which was the only mutant that could be clearly distinguished from its endogenous counterpart as a result of its large deletion, was the most highly expressed.
RNA-Seq data from a subset of nine tumor groups plus livers (see below) were next used to determine whether protein and transcript levels correlated. We quantified both human and endogenous murine β-catenin transcripts from a series of five tumors per cohort (Fig. 1, D and E). Whereas endogenous murine β-catenin transcripts showed little variation among the different groups, human transcripts showed a considerable amount but not always in ways that correlated with protein levels. For example, the biggest transcript disparities were seen in S33A and WT tumors, which expressed comparably low protein levels. Conversely, K335I and Δ(20–32) tumors showed significant differences in human β-catenin protein levels but not in transcript levels. Variations in β-catenin protein thus appear to be due to differential expression and/or stability of both transcripts and protein. The fact that every mutant transcript was expressed at levels significantly lower than those of WT transcripts suggests that many if not all mutant transcripts are unstable. Many patient-derived mutations also appear to exert opposite effects on transcript and protein levels.
To determine whether any of the above transcript and protein expression variability could be attributed to SB-β-catenin vector copy number differences, we quantified these in representative HBs from each tumor group using a TaqMan-based assay. On average, intragroup and intergroup variation seldom exceeded 2–3-fold (Fig. 1, F and G). Exceptions to this were seen with S33A, Δ(36–53) and Δ(90) tumors, which on average contained comparable and significantly fewer copy numbers than other tumor groups, and WT tumors, which contained ∼50% more. However, these variations bore little relationship to RNA or protein levels. For example, S33A and WT tumors showed the greatest differences in integrated vector copy number (∼6-fold) and 8–10-fold differences in β-catenin transcript levels. However, they expressed comparably low levels of β-catenin protein and grew at vastly different rates (Group 1 and 2, respectively). Thus, whereas some differences in integrated plasmid copy number were observed within and among each of the tumor groups, most of the variability in expression appeared to be due to differences in transcript and/or protein levels.
Subcellular fractionation studies showed the nuclear proportion of β-catenin mutants to range from a high of 95% in Δ(45–58) tumors to a low of 39% in S37A tumors (Fig. 1, H–J). The nuclear compartmentalization of WT β-catenin was also high (82%) despite its overall low-level expression (Fig. 1C and Fig. S2). This was surprising, given the expectation that WT β-catenin would be largely, if not exclusively, confined to the cytoplasm by virtue of its efficient association with the APC complex and proteasome. It suggested that tumorigenesis in response to WT β-catenin might require the selective outgrowth of a subpopulation of transfected hepatocytes with the highest nuclear β-catenin content to compensate for its low overall expression. This was consistent with the observation that, after appearing following their initial lag phase, these tumors grow at rates comparable with those of Group 1 tumors (not shown). To test this directly, we co-injected several mice with SB vectors encoding Myc-tagged Δ(90) and His6-tagged WT. As expected, these tumors developed rapidly, indicating that they were primarily Δ(90)-driven. Moreover, the majority of WT β-catenin was now observed to be cytoplasmic (Fig. 1K). This supported the idea that, in the absence of long-term selective pressure, the majority of WT β-catenin remains confined to the cytoplasm.
Among the different tumor groups, the crowded fetal (CF) pattern was the most commonly observed histopathologic subtype (
), although numerous exceptions were noted (Fig. 2 and Table 1). These included a combination of CF, macrotrabecular, and blastemal patterns in S33A tumors and a mix of CF and pleomorphic fetal patterns in Δ(45–58) tumors. Not all β-catenin mutants, however, were associated with HB-like histology; for example, WT tumors primarily demonstrated a classic HCC-like morphology interspersed with small islands of CF HB-like cells. R582W tumors were even more noteworthy for their HCC-like appearance as they contained few discernible HB elements. Several other mutants also generated tumors with significant HCC-like appearance (Table 1). Thus, like tumor growth rates, histology also appeared to be driven by the identities of the β-catenin mutants.
Figure 2.Histology of β-catenin mutant tumors. At least three tumors induced by each β-catenin mutant were examined under blinded conditions. Typical hematoxylin and eosin–stained sections are depicted here. Bars, 50 μm. The most prominent HB histologic subtypes observed in each tumor cohort are summarized in Table 1.
CF, crowded fetal; MT, macrotrabecular; BL, blastemal; PF, pleomorphic fetal; HCC, hepatocellular carcinoma. In cases where multiple histologic subtypes were observed, the most prominent one is listed first.
a CF, crowded fetal; MT, macrotrabecular; BL, blastemal; PF, pleomorphic fetal; HCC, hepatocellular carcinoma. In cases where multiple histologic subtypes were observed, the most prominent one is listed first.
β-Catenin mutants generate tumors with distinct metabolic features
Δ(90) HBs as well as HCCs generated by c-Myc overexpression undergo profound metabolic reprogramming involving higher rates of glycolysis and Warburg-type respiration in association with increased pyruvate dehydrogenase complex (PDC) activity (
). Concurrently, oxidative phosphorylation (Oxphos), particularly that utilizing fatty acids as an energy source, is significantly reduced. This fatty acid oxidation (FAO)-to-glycolysis switch occurs in the presence of and correlates with a pronounced decrease in mitochondrial DNA (mtDNA) content that presumably reflects the reduction in mitochondrial mass seen in numerous other cancers (
In control livers, mitochondrial Complex I was responsible for ∼30–40% of electron transport chain (ETC) activity, whereas Complex II was responsible for the remaining ∼60–70% (Fig. 3, A–D) (
). Nearly all tumor groups had signature patterns of Complex I and Complex II activities, and in some cases, these appeared to be independent of one another. For example, the Complex I activities of S45A, S45F, and S45P tumors were higher than those of the Δ(20–32) tumors, which in turn were similar to livers (Fig. 3, A and B). In contrast, Complex II activities of Δ(20–32) tumors were lower than those of S45A, S45F, and S45P tumors (Fig. 3, C and D). Importantly, the Complex II activities for S33A and S45F tumors were also similar to those of livers. These results indicate not only that β-catenin mutants differentially affect Complex II activity but that its down-regulation is context-dependent and does not invariably accompany transformation.
Figure 3.Metabolic behavior of HBs driven by different β-catenin mutants.A, Complex I activity. OCRs were determined on tissue homogenates following the addition of pyruvate, malate, glutamate, and ADP (
). B, significant pair-wise comparisons between individual tumors groups based on the results shown in A. C, Complex II activity. Following the addition of succinate to achieve the maximal combined activities of Complexes I + II, rotenone was added to inhibit Complex I. The residual activity (i.e. that of Complex II) is shown here. D, significant pair-wise comparisons between individual tumor groups based on the results shown in C. E, FAO activity in the indicated tissues. F, significant pair-wise comparisons between individual tumors groups based on the results shown in E. G, PDH activity in the indicated tissues. H, significant pair-wise comparisons between individual tumors groups based on the results shown in G. I, mtDNA quantification. A TaqMan assay was used to amplify a region of the mitochondrial D-loop region as described previously (
). Results were normalized to an amplicon from the nuclear gene encoding apolipoprotein B. J, significant pair-wise comparisons between individual tumors groups based on the results shown in I. K, transcript levels of mitochondrial ribosomal protein subunits. The indicated liver and tumor tissues (five per group) were subjected to RNA-Seq analysis. Levels of 78 transcripts encoding each of the mitochondrial ribosomal protein subunits are expressed as mean values and depicted as Z-scores. Error bars, S.E.
The degree to which tumor groups down-regulated FAO and up-regulated PDC activity was also highly variable. For example, S33Y and S37A tumors down-regulated FAO to a lesser degree relative to Δ(90) tumors, whereas T41A, S45A, and WT tumors did so to a greater degree (Fig. 3, E and F). PDC activity was similarly highly mutant-dependent, as evidenced by both S45A and R582W tumors having activities similar to those of livers, whereas Δ(36–53) and S33A tumors had activities 2.5–2.8 times higher, respectively (Fig. 3, G and H). These results indicate that different metabolic activities are controlled to different degrees, and in some cases independently, by the underlying β-catenin mutant. They also demonstrate that, as was true for FAO and Complex II activity, the up-regulation of PDC activity is not a prerequisite for transformation.
mtDNA DNA content was also evaluated using a TaqMan-based approach with PCR primers that amplified the D-loop region of the mitochondrial genome (
). Although all other tumor cohorts were associated with a reduced mtDNA content relative to livers, the magnitude of this loss was significantly less pronounced in S45F tumors.
Finally, we transcriptionally profiled mitochondrial ribosomal protein subunit transcripts. In keeping with the overall loss of mtDNA content in all tumors, these were also reduced, although the patterns of these changes were again highly β-catenin mutant–specific (Fig. 3K).
β-Catenin mutants differ in their transactivation potential
β-Catenin mutants might vary in their transactivation potential for various nonmutually exclusive reasons, including differences in their expression levels, the extent of their nuclear localization, and their ability to engage Tcf family members and other components of the transcriptional machinery. To examine this, we employed a vector (7TGC) that expresses enhanced GFP (eGFP) driven by a Tcf4/β-catenin–responsive promoter (
). We co-transfected this with each SB-β-catenin vector into HEK293 cells and quantified eGFP expression 48 h later. Significant differences were observed in the degree to which each β-catenin mutant increased eGFP expression (Fig. 4, A and B). For example, Δ(36–53) was the most potent activator of eGFP, whereas R582W and WT were the weakest. In parallel studies, we repeated these same transfections but included the SB-YAPS127A vector. YAPS127A exerted little effect on the reporter except in the case of the S45A mutation, where it modestly but significantly suppressed eGFP expression (Fig. 4, A and C).
Figure 4.Transactivation potential of different β-catenin mutants.A, eGFP/mCherry ratios in response to the co-expression of each of the indicated β-catenin mutants and in the absence (black circles) or presence (red circles) of co-expressed YAPS127A. Each point represents the results of a single transfection and the evaluation of at least 10,000 mCherry-positive cells. At least three transfections for each mutant were performed on different days. B, significant pair-wise comparisons between transfections without YAPS127A shown in A. C, significant pair-wise comparisons between transfections performed in the presence of co-expressed YAPS127A shown in A. Error bars, S.E.
We hypothesized that the above-reported differences in tumor phenotypes could be explained by variations in their global transcript expression profiles, as already seen for the response of 7TGC (Fig. 4). We further surmised that the overall transcriptional profiles of these tumors would resemble one another more closely despite their variable histologic features (Fig. 2) (
). We chose tumors generated by nine of the β-catenin mutants and subjected five of each along with five normal control livers to RNA-Seq. The former included the Group 1 tumors S33A, S45A, K335I, N387K, Δ(20–32), and Δ(90) and the Group 2 and 3 tumors S33Y, Δ(36–53), and WT. K335I and N387K were chosen because their mutations resided within the armadillo repeat region, which is far removed from the main phosphorylation domain located between residues 33 and 45 and whose coordinated multisite phosphorylation is critical for cytoplasmic retention by the APC complex and its subsequent degradation by the proteasome (
) (Fig. 1A). WT tumors were also selected because of their pronounced HCC-like histology (Fig. 2 and Table 1). This allowed a comparison of their transcriptomes with those from tumors generated by other β-catenin mutants and HCCs induced by ectopic Myc overexpression (
Relative to livers, tumor cohorts showed a mean of 5646 significant gene expression differences (range = 4222–6820, q < 0.05) (Fig. 5A and Table S2). As hypothesized, the differences were much less among individual tumor cohorts, ranging from as many as 1515 in the case of S33A versus S33Y tumors to as few as 1 in the cases of S33A versus S45A, S33A versus Δ(20–32), S45A versus Δ(20–32), and S45A versus Δ(90) tumors and 0 in the case of K335I versus N387K (Fig. 5B) and Table S2). Interestingly, S33Y showed the second fewest differences relative to normal liver (4766) but the largest number of differences relative to the eight other tumor groups (mean = 1071) (Table S2). When S33Y was excluded from the comparison, the tumor group showing the smallest number of differences relative to others was Δ(90) (average = 48, p < 10−9).
Figure 5.Transcriptomic differences among select HBs.A, gene expression differences among select β-catenin–induced tumors and normal control livers. Each column represents the mean transcript expression level of five tissue samples from that cohort. Gene expression differences were identified from RNA-Seq data using CLC Genomic Workbench (Qiagen), DeSeq2, and EdgeR, and only those transcripts identified by all three methods as being differentially expressed are included. The number above each tumor cohort indicates the differentially expressed transcript differences between that group and normal liver (also see Table S3). The mean number of differences among all groups was 5646. B, gene expression differences among each of the individual cohorts. Numbers above each tumor group indicate the number of gene expression differences compared with the Δ(90) cohort. See Table S2 for a comparison of gene expression differences among all cohort comparisons. C, gene expression differences between each cohort of β-catenin–induced tumors and HCCs. The latter were induced by induction of a doxycycline-regulated human Myc transgene as described previously (
). Numbers below each cohort indicate the number of gene expression differences relative to the HCC group. This along with principle component analysis (PCA; supporting Fig. S3) indicated that, despite their histologic differences, all β-catenin–induced tumors were much more HB-like than HCC-like at the molecular level and more closely resembled liver than they did HCC (Fig. S3). D, HBs deregulate a majority of genes identified as direct and expressed targets of β-catenin/Tcf4. Transcripts in the “liver” cohort include the entire 613 repertoire of expressed genes from normal liver whose proximal promoters were bound by β-catenin/Tcf4 (
). Numbers below the tumor cohorts indicate the number of transcripts that were deregulated relative to those in the liver. See Table S3 for a pairwise comparison of Myc target gene differences among all cohorts.
Despite the pronounced HCC-like histologic appearance of WT tumors and the variable but significant HCC-like content of several other tumor groups (Fig. 2 and Table 1), all tumors induced as a result of WT or mutant β-catenin overexpression resembled one another at the whole-transcriptome level much more closely than they resembled HCCs induced by Myc overexpression (
Ingenuity pathway analysis (Qiagen) was used to identify the top pathways whose component transcripts were most deregulated in at least one tumor cohort relative to that of liver. In keeping with the biochemical and metabolic abnormalities characterized above, seven of the top 15 pathways identified by this approach involved mitochondria-associated processes, including Oxphos, the TCA cycle, branched chain amino acid catabolism, and β-FAO, and all were predicted to be down-regulated in tumors (Fig. S4). The degree to which each component pathway's transcripts were deregulated (i.e. the Z-score) was also found to be highly characteristic of each β-catenin mutant.
) have identified 613 expressed genes whose proximal promoters directly bound β-catenin/Tcf4 complexes in primary murine hepatocytes. The differentially expressed genes in each of our tumor groups overlapped with 44.4–61.0% of these (mean = 324 (53%); Fig. 5D). Thus, on average, the majority of previously identified β-catenin/Tcf4 target genes in liver were dysregulated in HBs by different β-catenin mutations, with each cohort possessing a distinct expression profile.
Among the most highly up-regulated transcripts in Δ(90) tumors is Myc, whose hundreds of target genes include many that encode metabolic enzymes and their regulators (
). We speculated that, by altering pathways involving glycolysis, FAO, Oxphos, and ribosomal biogenesis, Myc balances the growth of tumors with their metabolic needs. Large differences in Myc expression were seen among different tumor groups, although nearly all expressed substantially more Myc than livers (Fig. 6A). All three slowly growing, Group 3 tumors were among those with the lowest Myc levels, and five of the six Group 1 tumors were among the highest Myc expressers. For the remaining tumors, however, this correlation was imperfect. T41A and Δ(20–32) tumors were among the fastest growing despite expressing quite low levels of Myc (Group 1, Fig. S1A). Four of the six Group 2 tumors expressed low levels of Myc, but two (S45P and Δ(45–58)) were among the highest expressers. Myc levels also did not always correlate with the β-catenin mutants' activation of the 7TGC promoter (Fig. 4, A and B). For example, Δ(36–53) generated tumors with low Myc levels despite being the most potent 7TGC activator, whereas the reverse was true for R582W and Δ(90). These studies suggest that β-catenin mutants differentially up-regulate Myc and that some do so independently of their effects on other genes. In keeping with this idea, l-Myc expression was high in control livers, down-regulated in most but not all tumors, and did not necessarily correlate with those of Myc (Fig. 6A).
Figure 6.Regulation of metabolic pathways is β-catenin mutant–specific.A, c-Myc and l-Myc protein levels in individual tumors. B, expression of 613 direct Myc target genes previously identified in hepatocytes (
). Transcripts are arranged from highest to lowest expression based on levels in liver. Numbers at the bottom indicate the number of significant expression differences versus liver. Table S3 shows the actual number of differentially regulated transcripts among all pairwise comparisons. C, transcripts encoding TCA cycle enzymes. Table S4 shows the number of similarly regulated transcripts between any two tissues relative to those in liver. D, transcripts encoding ETC and Complex V subunits. Table S5 shows the number of transcript differences among all pairwise comparisons. E, differential regulation of transcripts encoding mitochondrial FAO-related transcripts. Table S6 shows the number of transcript differences among all pairwise comparisons. F, differential regulation of (endosomal) FAO (β-oxidation)-related transcripts. Table S7 shows the number of transcript differences among all pairwise comparisons. G, expression of p19ARF in liver and tumor samples. H, murine HBs most closely resemble the C1 molecular subtype of human HBs. The CLC Genomics Workbench hierarchical clustering program was used to compare and organize the murine orthologs of the 16-transcript set of Cairo et al. (
). Human tumor results are from individual tumors, whereas murine results represent the mean expression value of five tumors for each group as done for other panels in the figure. Human C1 tumors are indicated by the green bars at the bottom of the heat map, human C2A and B tumors are indicated by the yellow bars, and murine tumors are indicated by magenta bars. The scale bar here indicates the -fold change rather than the Z-score.
We next assessed the expression of 634 direct Myc target genes in the liver using data from the Ingenuity Pathway Analysis (IPA, Qiagen, Inc.) Knowledge Base (
). Between 204 and 277 of these (mean = 243 (38.3%)) were deregulated in each of the tumor cohorts (Fig. 6B and Table S3). In keeping with the fact that liver expresses very low levels of Myc (Fig. 6A), most of these tumor transcripts were expressed in the opposite direction. Interestingly, each HB group expressed its own signature profile of these transcripts that may be in part reflect competition between Myc and l-Myc (Table S3). Each tumor group's expression of these individual target genes as a group was unique and in keeping with our results regarding Myc protein expression and 7TGC responses discussed above. This was again consonant with the idea that each β-catenin mutation affected each target gene's response in a specific and largely individualized manner that accounts for this distinct transcript patterning.
IPA was again used to identify families of functionally related transcripts that differed among the nine tumor groups. Over half of the top pathways so identified were involved in aspects of metabolism involving activities such as Oxphos, lipid and xenobiotic metabolism, and FAO. Consistent with the overall reduction of Complex II activity and mtDNA content (Fig. 3), transcripts encoding most TCA cycle enzymes, ETC subunits, and FAO-related enzymes were down-regulated, although the patterns of these changes were again highly mutant-specific (Fig. 6, C–E). The vast majority of transcripts encoding enzymes involved in ω-FAO, a nonenergy producing and largely cytoplasmic process (
), we failed to identify de novo mutations in any transcripts expressed by the above 45 tumors (not shown). However, we have previously demonstrated that expression of the p19ARF tumor suppressor is markedly attenuated in Δ(90) tumors without being mutated and that tumorigenesis was strongly abrogated by SB-mediated ectopic p19ARF expression (
). To determine whether p19ARF loss was a general finding, we evaluated its levels in representative samples from each tumor cohort and verified its down-regulation in all Δ(90) tumors relative to that of liver (Fig. 6G). Similar findings were made in most other tumor groups but not all. For example, S45P, Δ(20–32) and Δ(48–58) tumors (all Group 1 or Group 2, Fig. 1B) showed, at best, only modest suppression of p19ARF. Thus, whereas p19ARF down-regulation occurs frequently and reproducibly in certain HB cohorts, it is not a universal finding. Together with the results presented in Fig. 6A, this suggests that the pathogenesis of HBs by different β-catenin mutants does not invariably involve the inactivation of the same tumor suppressor pathways or the activation of the same oncogene pathways.
) previously described a 16-transcript signature that allowed classification of tumors into two groups, C1 and C2, with favorable and unfavorable outcomes, respectively. The latter group is associated with adverse clinical features, such as higher stage at the time of diagnosis and evidence for vascular invasion. A subsequent study of 24 additional HBs by Hooks et al. (
) subdivided the C2 group into two subgroups based on the expression of four transcripts. C1 HBs displayed a predominantly fetal histology, whereas C2 tumors possessed more immature embryonal features, more mitoses, and more intense nuclear β-catenin staining. Murine embryonal liver stem cells demonstrated a C1-like profile prior to their ectopic overexpression of β-catenin, Myc, or N-Myc and a C2-like profile afterward. C2 patterns were seen in the human HB cell lines Huh6 and HepG2, which reverted to C1 when either Myc or β-catenin was silenced (
) and assessed each of our tumor groups for the expression of the murine orthologs of the signature 16 transcript collection identified by Cairo et al. (
). We used the CLC Genomics Workbench 12.0 program to cluster these in an unbiased manner into groups with related expression patterns. As seen in Fig. 6H, this approach indicated that murine HBs appeared most closely related to the C1 human molecular subtype, irrespective of their underlying β-catenin mutation.
Discussion
Survival in HB is associated with certain clinical and laboratory features including age, stage, histologic subtype, and serum α-fetoprotein levels (
Frequent deletions and mutations of the β-catenin gene are associated with overexpression of cyclin D1 and fibronectin and poorly differentiated histology in childhood hepatoblastoma.
). Newer transcriptome-based classifications have identified the C1/C2 groups and the C1/C2A/C2B groups as having distinct long-term survival differences based on 16-gene or four-gene signatures, respectively (
Frequent deletions and mutations of the β-catenin gene are associated with overexpression of cyclin D1 and fibronectin and poorly differentiated histology in childhood hepatoblastoma.
), these and the tumor's rarity and extensive histologic variability have made it impossible to correlate specific mutations with survival or other features. Early surgical and/or chemotherapeutic intervention further preclude any assessment of the natural history of tumors associated with specific β-catenin mutations. Our results in mice indicate that several HB biological, biochemical, metabolic, and transcriptional behaviors are predetermined by the identity of the β-catenin mutation. It remains unclear whether our failure to detect metastases in any cases is the result of the inadvertent selection of mutants that favored the generation of only localized disease or because most endogenously arising murine tumors tend to have low metastatic propensity (
Clinically relevant subsets identified by gene expression patterns support a revised ontogenic model of Wilms tumor: a Children's Oncology Group Study.
). Whereas some of these are shared with HB, others are not, thus raising the questions of whether they are all equally oncogenic in the liver and, if so, whether they differentially influence tumor behaviors. It seemed reasonable to surmise that in HB, where few if any other recurrent driver mutations exist (
), different β-catenin mutants influence not only growth rates but additional phenotypes. Examples of other oncoproteins exhibiting similar behaviors exist, including non-small-cell lung cancer, where survival and response to erlotinib and gefitinib correlate with missense or deletion mutations, respectively, in the epidermal growth factor receptor (
Exon 19 deletion mutations of epidermal growth factor receptor are associated with prolonged survival in non-small cell lung cancer patients treated with gefitinib or erlotinib.
Clinical course of patients with non-small cell lung cancer and epidermal growth factor receptor exon 19 and exon 21 mutations treated with gefitinib or erlotinib.
With regard to β-catenin specifically, previous transient transfection studies demonstrated differences in the abilities of nine β-catenin mutants, including S45F and S45P, to activate a synthetic reporter similar to 7TGC (Fig. 4) (
). These findings were limited, however, in that some of the β-catenin mutants were not patient-derived, nuclear:cytoplasmic distribution of the mutants was not correlated with total expression levels, and other target gene responses were not examined. Consistent with our results, a recent in vivo study showed that Δ(90) and S45Y, in combination with c-Met, induced HCCs with similar growth rates and histologies but distinct gene expression profiles (
). Thus, our finding that each β-catenin mutant was associated with signature transcriptional profiles (Figure 4., Figure 5., Figure 6.) is not without precedent, likely explains the distinct features of each tumor group and provides an overarching mechanistic explanation for the phenotypic differences observed among each tumor group.
Two of the fastest growing tumor cohorts, namely S45A and N387K (Group 1, Fig. S1A), were driven by mutants not previously associated with HBs. By comparison, the common HB-associated mutants S33Y and T41A drove more slowly growing Group 2 tumors (Fig. 1B and Fig. 1) (
). The implication of these findings is that mutations previously identified in non-HB tumors may be as tumorigenic as “classical” HB-associated mutants. The recurrent nature of certain mutants therefore implies neither that they are the more efficient oncogenes nor that they drive particularly aggressive neoplasms. Rather, their frequency may reflect greater context-dependent mutagenic susceptibility of certain codons. The identities of these need not be fixed and might well vary as a function of age and/or context. Alternatively, some mutations detected thus far only in adult tumors may yet be identified in HBs as testing for these becomes more commonplace.
Like growth, HB histology was also β-catenin mutant–dependent. For example, both R582W and WT HBs were almost exclusively HCC-like, as were to a lesser extent T41A and K335I tumors (Table 1 and Fig. 2). Yet no other factors appeared to be predictive of these histologies. The lack of greater histopathologic diversity among our tumor groups more akin to that of the seven HB subtypes (
) suggests either that different β-catenin mutants and/or other genes are required for such differences to be fully manifested. These could include TP53, TERT, and NFE2L2, which are recurrently mutated in a small minority of HBs (
). Given the inability to detect such cooperating mutations in the tumors studied here using the CLC Genomics Workbench Basic Variant Detection algorithm, it is tempting to speculate that the histologic variability we did encounter is a result of the inherent transcriptional differences of the β-catenin mutants and perhaps epigenetic differences that account for the altered expression of important cooperating genes, such as Myc, l-Myc, and p19ARF (Fig. 6, A and G). It remains unclear to what extent different tumor histologies are interconvertible and whether they are influenced by factors such as the metabolic state or microenvironment, which can vary over quite small intratumoral distances (Table 1 and Fig. 2) (
A notable feature of WT and R582W tumors was their undeniable retention of HB-like transcriptomes despite histologically resembling HCCs (Table 1, Figure 2., Figure 5. (A and C), and Fig. S3). This contrasts with the well-known observation that FAP-associated liver tumors, which express WT β-catenin, are always HBs (
). The discrepancy may be at least partly explained by the long latency period of WT murine tumors, which arise well into adulthood. HBs are believed to originate from short-lived primary hepatoblasts or fetal-like multipotent progenitors, thus explaining the strict age dependence even in individuals with FAP (
). This population might be depleted and/or altered in mice by the time WT and R582W tumors appear, at which point they more closely resemble HCCs histologically while retaining HB transcriptomic profiles. Early liver progenitors and fetal-like hepatocytes may also degrade WT β-catenin more efficiently within the cytoplasmic APC complex, thus delaying the onset of its stabilization and nuclear translocation (
). This was suggested by our finding that Δ(90) tumors showed more pronounced cytoplasmic retention of co-expressed WT β-catenin than did WT tumors (Fig. 1, compare H with K). Slowly growing R582W tumors appear even more HCC-like than WT tumors (Table 1 and Fig. 2), and four of the five remaining cohorts with less conspicuous HCC-like histology were also categorized as belonging to the slower-growing Group 2 or 3 (Table 1). A nonmutually exclusive explanation is that YAPS127A, while incapable of HB induction alone (
), can nevertheless induce HCC in older mice and that its contribution becomes increasingly important with age, whereas WT β-catenin continues to dominate the transcriptional output (
). There may exist a specific age-dependence for each β-catenin mutant such that the histology of early onset tumors is more HB-like and more β-catenin–dependent, whereas later onset tumor are more HCC-like and YAPS127A-dependent. A similar “window” of susceptibility may also apply to the defects that deregulate other facets of the Hippo pathway (
). That Δ(90) generates HBs or HCCs based on whether it is co-expressed with YAPS127A or c-Met lends further credence to the argument that β-catenin–driven tumorigenesis can be promiscuous and influenced by other co-expressed oncoproteins as well as age- and tissue-specific factors (
) but think it likely that it would also drive HB tumorigenesis in association with YAPS127A, given the robust tumorigenicity of S45A, S45F, and S45P (Fig. 1B).
Both absolute β-catenin level and nuclear localization are of prognostic significance in some HB studies, although the degrees to which these are mutually and mutationally dependent have not been determined (
). Missense and in-del mutations can profoundly affect mRNA stability by altering folding, splicing, recognition by micro-RNAs, and translational efficiency (
). Independently, steady-state protein levels can be affected by mutation-dependent effects on folding, protein-protein associations, and post-translational modifications (
). This latter is particularly germane in the case of β-catenin mutations, which often involve key sites of APC complex–mediated phosphorylation that determine β-catenin's stability and nuclear translocation (
). However, our results suggest that the consequences of such mutations are more complex and nuanced than might have been previously surmised. For example, Δ(20–32) and Δ(45–58) possess comparably sized deletions, with the former affecting none of the known APC phosphorylation sites and the latter affecting only Ser-45. Accordingly, Δ(20–32) would have been expected to be expressed at a lower level than Δ(45–58), to be cytoplasmically restricted and to be less pathogenic. Yet, whereas it was indeed more cytoplasmically confined, its total expression was equivalent to Δ(45–58)'s while generating faster-growing tumors (Fig. 1, A, C, and H–J). Other examples were seen with S33A and S45F. Although each bore only a single phosphorylation site mutation, S33A was expressed at a lower level despite being more nucleus-localized. Yet it also generated faster-growing tumors.
β-Catenin mRNA and protein steady-state levels also did not often correlate. For example, WT transcripts were ∼10-fold more abundant than those of S33A, although tumor protein levels were similar (Fig. 1, C and D). Conversely, S45A and K335I transcript levels were equivalent, although S45A protein was more highly expressed. Low β-catenin protein levels were associated with three of the four slowest growing tumor groups (R582W, Δ(36–53), and WT), although this was not always so, as evidenced by the rapid growth of K335I tumors (Group 1, Fig. 1B and Fig. S1A). All of these proteins, however, efficiently localized to the nucleus (Fig. 1, H–J), suggesting that this could serve as a pro-oncogenic mechanism to compensate for overall low β-catenin expression. This was supported by our findings with WT β-catenin, which was virtually all nuclear in slowly growing tumors but mostly cytoplasmic when co-expressed with Δ(90) in rapidly growing tumors (Fig. 1, H and K). The long latency of WT tumors may allow clones with higher expression of nucleus-localized β-catenin to be selected and become dominant. The nuclear content of β-catenin may thus vary over time. Nonetheless, our findings support the idea that the oncogenic potency of any particular β-catenin mutant is more influenced by the amount that is nucleus-localized than by the total amount that is expressed. The mutant-dependent nature of β-catenin expression could have significant implications for newer therapeutic approaches that abrogate its interaction with the transcriptional machinery (
). It remains to be determined whether these assumptions will hold for non-HB β-catenin-driven tumors. More precise determinations of mutant β-catenin transcript and protein half-lives will require carefully controlled conditions using metabolic inhibitors, such as actinomycin D and cycloheximide, or transient affinity label-tagging methodologies.
In tumor groups such as S33A, S45A, Δ(20–32), Δ(45–58), and Δ(90), the number of stably integrated SB-β-catenin copy numbers was remarkably consistent among individual tumors, and in other cohorts, it seldom differed by more than 2–3-fold (Fig. 1F). Importantly, mice from each cohort were injected at different times and with different DNA preparations during tumor induction. The highly reproducible intragroup β-catenin copy number for the above-mentioned cohorts is testament to the fact that neither of these variables exerted significant influence over the final copy number. Interestingly, those tumor groups with the least copy number variation were all associated with rapidly growing tumors, whereas more slowly growing tumors, such as S33Y, S37A, R582W, and WT, showed greater variation.
Each β-catenin mutant also impacted metabolic pathways in distinct ways (Figure 3., Figure 6. (C–F)). Mechanistically, this likely reflects differences in the degree to which each mutant regulates these pathways' component transcripts (Figs. 3K and 6 (B–F)). Independent of total transcript levels, the expression patterns of small numbers of transcripts in cancer-related pathways have recently been shown to better predict survival across more cancer types (
). Considering only the differences in β-catenin, Myc, and l-Myc protein levels among the different tumor groups (Figs. 1C and 6A and Fig. S2), we can envision a model in which β-catenin mutants differentially interact with Tcf/Lef family members, each with its own inherent DNA binding affinity and target gene subsets (
). Together, our findings suggest that the mechanisms underlying each β-catenin mutant's distinct effects on transformed hepatocyte behavior may involve nonmutually exclusive combinations of total protein levels, subcellular partitioning, interaction with the transcriptional machinery, cross-talk with the Hippo-YAP pathway, the affinity for target genes, and the magnitude and patterns of their response.
In conclusion, we provide the first direct evidence showing that much of the diversity that distinguishes HBs can be attributed to the underlying identity of their underlying β-catenin mutations. Those previously identified only in non-HB tumors are also tumorigenic in the liver, thus suggesting that much of β-catenin's oncogenic capacity is neither mutant- nor tissue-restricted. Additional work will be necessary to determine whether these mutations can be used in meaningful ways to predict tumor behavior, response to and durability of chemotherapy, and long-term survival in patients.
Experimental procedures
Animals and HDTVI
HDTVIs of purified plasmid DNAs were performed as described previously (
). Injections, in a total volume of 2 ml in 0.9% sodium chloride, contained 10 μg each of SB vectors encoding the indicated β-catenin mutant and the previously described YAPS127A</
30744-4&id=gr1.jpg){kind=link}
30744-4&id=gr2.jpg){kind=link}
30744-4&id=gr3.jpg){kind=link}
30744-4&id=gr4.jpg){kind=link}
30744-4&id=gr5.jpg){kind=link}
30744-4&id=gr6.jpg){kind=link}