Gas1-Mediated Suppression of Hepatoblastoma Tumorigenesis

Abstract

Hepatoblastoma (HB), the most common pediatric liver cancer, is associated with dysregulated Wnt/β-catenin, Hippo, and/or nuclear factor erythroid 2 ligand 2/nuclear respiratory factor 2 (NFE2L2/NRF2) pathways. In mice, pairwise combinations of oncogenically active forms of the terminal transcription factors of these pathways, namely, β-catenin (B), Yes-associated protein (YAP; Y), and Nrf2 (N), generate HBs, with the triple combination (B + Y + N) being particularly potent. Each tumor group alters the expression of thousands of B-, Y-, and N-driven unique and common target genes. The identification of those most involved in transformation might reveal mechanisms and opportunities for therapy. Herein, transcription profiling of >60 murine HBs revealed a common set of 22 "BYN" genes similarly deregulated in all cases. Most were associated with multiple cancer hallmarks, and their expression often correlated with survival in HBs, hepatocellular carcinomas, and other cancers. Among the most down-regulated of these genes was Gas1, which encodes a glycosylphosphatidylinositol-linked outer membrane protein. The restoration of Gas1 expression impaired B + Y + N–driven HB tumor growth in vivo and in HB-derived immortalized BY and BYN cell lines in vitro in a manner that requires membrane anchoring of the protein via its glycosylphosphatidylinositol moiety, implicating Gas1 as a proximal mediator of HB pathogenesis and validating the BYN gene set as deserving of additional scrutiny in future studies.

Graphical abstract

Hepatoblastoma (HB), the most common pediatric liver cancer, is usually diagnosed in children before 4 years of age.¹ Believed to originate in developmentally restricted fetal hepatoblasts, HBs are currently classified into several histologically distinct subtypes despite being the least molecularly complex of all cancers.1, 2, 3, 4 The most common molecular changes involve the Wnt/β-catenin, nuclear factor erythroid 2 ligand 2/nuclear respiratory factor 2 (NFE2L2/NRF2), and Hippo signaling pathways.⁵ More specifically, they involve missense or in-frame deletion mutations of the CTNNB1 gene, which encodes β-catenin (B), amplification of or missense mutations in the NFE2L2/NRF2 (N) gene, and aberrant nuclear accumulation of the Hippo terminal effector Yes-associated protein (YAP; Y).5, 6, 7, 8 This relative molecular simplicity has been reconciled with histologic and biological diversity of HB by demonstrating that the in vivo expression in mice of different pairwise combinations of B, Y, and N mutants generates tumors with distinct growth rates, histologies, and transcription profiles that recapitulate some of the human subtypes.5, 6, 7^,9 Moreover, the triple combination of B, Y, and N drives a particularly aggressive form of HB closely resembling the so-called crowded fetal histologic subtype, which is accentuated by innumerable fluid-filled cysts—otherwise a rare feature of human HB.^6,10 Further diversity can be generated by different patient-derived missense or in-frame deletion B mutant proteins.^7,9 Distinct and diverse experimental HB characteristics and behaviors can thus be recurrently generated in mice by overexpressing different B, Y, and N oncoprotein combinations and mutants from this otherwise simple oncogenic menu.⁵

The B, Y, and N proteins are all transcription factors (TFs) with hundreds to thousands of unique and shared direct target genes.^6,7,11, 12, 13 The gene expression profiles of HBs can differ widely and in ways that reflect different combinations of the TFs and B mutation identities.5, 6, 7 This raises the question of which downstream genes are directly responsible for transformation. Among the most highly expressed genes in HBs is MYC, a downstream target of B, Y, and N that encodes another oncogenic TF.^14,15 The participation of this TF thus indicates the existence of a transcription cascade that amplifies and extends the activities of upstream factors and confers additional gene expression complexity to different HB subtypes. Unlike B, Y, and N, however, Myc is not mandatory for transformation as HBs can still be efficiently generated from Myc^–/– murine hepatocytes, although tumor growth is slowed.¹⁶ Further augmenting this transcription cascade is the Mlx network, which closely cooperates with the structurally and functionally related Myc network.16, 17, 18 The hepatocyte-specific loss of the ChREBP/Mlxipl gene, which encodes a Myc-like member of this network, and is also a direct target of B, Y, and N, additionally impairs HB growth without affecting initiation as does the inactivation of Mlx, which encodes the obligate heterodimerization partner of carbohydrate-responsive element-binding protein (Chrebp), which is functionally analogous to the dimerization partner of Myc, Max.¹⁷ The ensuing transcription heterogeneity among these various molecular classes of HBs further complicates the identification of the most relevant underlying oncogenic initiators.

The availability of the aforementioned murine models of HB has previously allowed for comparisons of their transcription profiles. Doing so identified 22 common genes that are invariably up- or down-regulated among all murine HBs.^5,6 Eighteen of these so-called BYN genes are associated with a mean of 4.3 cancer hallmarks versus 6.4 for well-known oncoproteins and tumor suppressors (TSs) such as Myc, Kirsten rat sarcoma viral oncogene homolog (Kras), transformation-related protein (Trp)-53, and retinoblastoma-associated protein (Rb).¹⁹ Furthermore, the expression patterns of 10 BYN genes could be used to distinguish between human HBs with favorable and unfavorable outcomes complementing other studies.^6,11,13 Finally, 19 BYN genes were dysregulated in murine hepatocellular carcinomas (HCCs) driven by a human MYC transgene, and 17 were correlated with survival in over a dozen human cancer types.⁶ Many BYN genes encode enzymes and membrane-associated or extracellular proteins, suggesting that they might be less challenging therapeutic targets than any upstream TF.²⁰ Among the more prominent down-regulated BYN genes was the gene encoding growth arrest–specific protein 1 (Gas1), whose transcriptional suppression correlated inversely with the growth rates and aggressiveness of different HB molecular subtypes.⁶

Gas1 was originally identified as being induced in NIH3T3 murine fibroblasts in response to serum starvation, and the only such protein whose enforced re-expression inhibited serum-stimulated G₀→G₁ progression.21, 22, 23, 24, 25, 26 Mature murine Gas1 is an approximately 37-kDa glycosylphosphatidylinositol (GPI)-linked outer membrane protein that exerts pleiotropic effects on survival, apoptosis, and mitogenic signaling.^21,26, 27, 28 It is synthesized as an approximately 40-kDa precursor that, like other GPI-linked proteins, contains an N-terminal approximately 38–amino acid signal peptide that directs Gas1 to the endoplasmic reticulum, where, after signal peptide removal, an additional approximately 28–amino acid C-terminal segment is cleaved and replaced by a serine-linked GPI moiety.^26,29 On further processing by the Golgi apparatus, the mature Gas1 protein, now anchored by this GPI tag, is displayed on the outer plasma membrane. However, whether this mature post-translationally processed and modified protein represents the only form (or the only location) that mediates cell-cycle arrest and other biological functions has been questioned by studies showing that neither C-terminal modification nor GPI addition is necessary for these activities in certain cell types.^26,30,31 In some cases, the suppression of normal and neoplastic cell growth by Gas1 is at least partially dependent on TP53 but does not require the transcriptional activation function of the latter. Rather, it requires a proline-rich domain in TP53 located immediately downstream of the N-terminal transactivation domain.^21,32

This study examined the putative role of Gas1 as a HB TS in the context of its membership in the BYN gene collection. The regulation of Gas1 is found to be both complex and dependent on the coordinate activities of the aforementioned TFs that initiate HB tumorigenesis. This study also shows that GAS1 transcript and protein down-regulation is a common feature of many human and murine HBs and HCCs. Finally, it is demonstrated that the enforced overexpression of Gas1 impairs murine HB tumorigenesis in vivo and HB cell line proliferation in vitro. A more detailed assessment of the latter cell lines shows the growth-inhibitory effects of Gas1 to be dependent on its proper GPI-dependent presentation on the outer surface of the plasma membrane. Collectively, the findings support the conclusion that enforcing the expression of a prominent TS-like BYN member can directly alter HB pathogenesis by inhibiting tumor cell proliferation. The findings are also in keeping with the idea that BYN gene members, either individually or collectively, can contribute to HB pathogenesis, are regulated by each of the major TFs that drive and/or modify the disease, and represent the most proximal factors that contribute to transformation.

Materials and Methods

Animal Care and Husbandry

FVB/N mice aged 4 to 6 weeks were purchased from The Jackson Laboratory (Bar Harbor, ME) and were used for all in vivo tumor studies. Mice were maintained in micro-isolator cages with ad libitum access to standard mouse chow and water. All care and procedures were reviewed and approved by the University of Pittsburgh's Department of Laboratory and Animal Resources (Pittsburgh, PA) and the Institutional Animal Care and Use Committee.

Bacterial Plasmids

Sleeping Beauty (SB) vectors encoding the Δ90 patient-derived in-frame B deletion mutation (B), Y^S127A (Y), N^L30P (N), and a non-SB vector encoding SB transposase (pCMV-transposase) have been previously described.^7,17,18 The same SB vector encoding a C-terminal, V5 epitope–tagged, full-length murine Gas1 protein (exogenous Gas1) was codon optimized, synthesized by GenScript Biotech (Piscataway, NJ), and used as a template for re-cloning of the cDNA into another SB vector, pSBbi-RP (Addgene, Inc., Watertown, MA). The latter was utilized for the expression of the original full-length (343 amino acid), unprocessed murine Gas1 protein (uniprot.org ID Q01721), which included the N-terminal endoplasmic reticulum targeting sequence, the C-terminal moiety that must be cleaved prior to GPI addition, and the C-terminal V5 epitope tag. The absence of V5 epitope-tagged Gas1 protein generated by this vector was used to confirm the efficiency of post-translational cleavage of the C-terminus in the fully processed protein. Finally, a second pSBbi-RP vector encoding a truncated Gas1 protein, lacking the C-terminal 29 amino acids (and the V5 tag), and predicted to be secreted due to its inability to be anchored to the plasma membrane via its GPI modification was used to determine whether soluble Gas1 inhibited HB cell line growth in vitro. cDNAs encoding both of these proteins were amplified from the original SB vector using the common forward primer 5′-TCAAGCCTCAGACAGTGGTTC-3′. The sequence of the reverse PCR primer used to amplify the full-length cDNA was 5′-TAGAAGGCACAGTCGAGG-3′, and the sequence used to amplify the 29 codon–truncated cDNA was 5′-ACTGTCTAGATTAGCCGCTAGAGGAGCTCCGT-3′. All oligonucleotides were synthesized by IDT, Inc. (Coralville, IA). After digestion with NcoI and XbaI, cDNAs were cloned directionally into the pSBbi-RP vector that contained these same sites.

Two pDG458 Crispr/spCas9 vectors (pDG1+2 and pDG3+4)³³ were used to express four gRNAs against exon 2 of the murine Cdkn2a gene.^34,35 The top strand sequences were oligo 1: 5′-CGGTGCAGATTCGAACTGCGAGG-3′; oligo 2: 5′-GTCGTGCACCGGGCGGGAGAAGG-3′; oligo 3: 5′-CTTGGGCCAAGTCGAGCGGCAGG-3′; and oligo 4: TGCGATATTTGCGTTCCGCTGGG-3′. Another pDG458 vector was generated that encoded two gRNAs (pDG5+6) against exon 1α and was used to specifically inactivate p16^INK4A.³⁴ Top strand sequences for this vector were oligo 5: 5′-AGGGCCGTGTGCATGACGTGCGG-3′; and oligo 6; 5′-CTCCTTGCCTACCTGAATCGGGG-3′. Each double-stranded oligonucleotide was inserted into its respective site in pDG458 Crispr/spCas9 using a previously described single-step digestion-ligation procedure (https://media.addgene.org/data/plasmids/100/100900/100900-attachment_Yl0i43bWJig3.pdf, last accessed December 13, 2024). After confirmation of all sequences by di-deoxy DNA sequencing, plasmids were purified using Plasmid Plus Midi columns (Qiagen, Inc., Germantown, MD), re-dissolved in sterile Tris-EDTA buffer, and stored at –20°C.

Generation of HBs

HBs were induced in mice aged 6 to 8 weeks, as previously described, using SB vectors encoding the patient-derived in-frame B^Δ90 deletion mutation, the patient-derived N^L30P missense mutation, and the Y^S127A missense mutation (hereafter, B, N, and Y, respectively), all three of which constitutively localize to the nucleus as a result of their respective mutations.^7,17,18 In brief, purified plasmids were delivered to the liver by hydrodynamic tail vein injection in 2 mL of phosphate-buffered saline over 5 to 10 seconds. Each inoculum included 10 μg each of the indicated SB vectors along with 2 μg of the pCMV transposase–encoding vector. Tumors with Cdkn2a locus mutations were generated in the same manner except that 2 μg of each pDG458 vector was included in each inoculum.

Murine HB Cell Lines

BY and BYN murine HB cell lines were generated from individual primary B + Y and B + Y + N tumors in which the Cdkn2a locus was disrupted by targeting exon 2 using the pDG458(1+2) and pDG458(3+4) gRNAs or exon 1α using pDG458(5+6) Crispr/Cas9 vectors described in Bacterial Plasmids. These plasmids were delivered to mice along with the respective B, Y, and/or N SB vectors at the time of tumor generation.³⁴ On reaching maximal size, tumors were excised and ≈2 g was finely minced, washed several times in phosphate-buffered saline, and digested for 30 minutes at 37°C in 0.1% trypsin (Sigma-Aldrich, Inc., St Louis, MO). After further disruption by vigorous pipetting and vortexing, the digested samples were added to tissue culture plates containing standard Dulbecco's modified minimal essential medium–10% fetal bovine serum. Over the next 1 to 2 weeks, macroscopic remnants of the original tumor tissue were discarded. After several additional weeks, both microscopically and macroscopically visible colonies could be identified against the background of nonproliferating cells. These colonies were pooled, expanded, and then frozen in liquid nitrogen within three to five additional passages.³⁴ Cell lines were transfected with 2 μg of the relevant pSBbi-RP–Gas1 expression vectors and 0.2 μg the pCMV-transposase vector using Lipofectamine 3000 according to the manufacturer's instructions (Thermo Fisher Scientific, Inc., Waltham, MA). Control cell lines were established from cells transfected with the empty pSBbi-RP vector alone. All transfected cells were selected in puromycin (2 μg/mL) and expressed dTomato that was also encoded by the pSBbi-RP vector. In other experiments, transiently transfected cells were separated by fluorescence-activated cell sorting based on the intensity of dTomato expression. All cell sorting utilized a FACSAria II instrument (Becton, Dickinson and Company, Franklin Lakes, NJ). Cell counting was performed on the Incucyte live-cell imaging and analysis system (Sartorius Essen, Inc., Ann Arbor, MI) as previously described.³⁴

Immunoblot Analysis

Tissue fragments and cell pellets were lysed in ice-cold 1X SDS-PAGE buffer in the presence of protease and phosphatase inhibitor cocktails.^7,16,17 After standard denaturing SDS-PAGE, separated proteins were blotted to PVDF membranes (Millipore, Inc., Burlington, MA) using a semi-dry apparatus (Fisher Semi-Dry Blotting Apparatus; Thermo Fisher Scientific). Antibodies used for immunoblot analysis and immunofluorescence studies included those for Gas1 (1:1000; catalog number AF2644-SP; R&D Systems, Inc., Minneapolis, MN), V5 epitope tag (1:5000; catalog number R960-25; Thermo Fisher Scientific), GAPDH (1:5000; catalog number G8795; Sigma-Aldrich), Akt (1:500; catalog number sc-5298; Santa Cruz Biotechnology, Dallas, TX), pSer⁴⁷³ Akt [1:500; catalog number 9271; Cell Signaling Technologies (CST), Inc., Danvers, MA], actin (1:5000; catalog number 3700; CST), horseradish peroxidase–conjugated goat anti-mouse IgG (1:5000; catalog number 91196; CST) and horseradish peroxidase–conjugated goat anti-rabbit IgG (1:5000; catalog number 7074; CST). Blots were developed using a chemiluminescence kit (catalog number 34096; Thermo Fisher Scientific) and imaged on the ProteinSimple FluoroChem M system (Bio-Techne, Inc., Minneapolis, MN).

Computational Analysis

Consensus TF binding sites within the intron-less Gas1 gene were retrieved from the JASPAR database (https://jaspar.elixir.no) through the University of California–Santa Cruz Genome Browser (https://genome.ucsc.edu, last accessed September 1, 2024). The sites identified included those for β-catenin (TCF4 sites), YAP/TAZ [transcriptional enhancer factor (TEAD) sites], MYC (E boxes), NRF2 [anti-oxidant response element (ARE) sites], SP1, and Myc-interacting zinc finger protein 1 (MIZ1), each with scores ranging from ≈300 to 370. Chromatin immunoprecipitation sequencing (ChIP-seq) data, sourced from ReMap, were obtained through Genome Browser. Representative ChIP-seq peak maps were either downloaded from the Gene Expression Omnibus database (https://www.ncbi.nlm.gov/geo) or subjected to repeat analysis using the provided raw sequencing data. A schematic diagram depicting the binding sites and ChIP-seq peaks was generated using R software version 4.3.1 (https://www.r-project.org), with the ggplot2 package (https://ggplot2.tidyverse.org) employed for visualization.

To quantify Gas1 and other transcript levels in murine HBs induced by various B, Y, and N combinations and in murine HCCs induced by conditional Myc overexpression, previously published RNA-seq results were queried using the Gene Expression Omnibus database.^7,16,17,36 Quantification of Gas1 transcripts in select human cancers from The Cancer Genome Atlas (TCGA) was performed as described previously.⁶ Briefly, RNA-seq expression data [fragments per kilobase of transcript/million mapped reads upper quartile (FPKM-UQ)] and clinical annotation files were downloaded from the TCGA, Genomic Data Commons, and Pan-Cancer Atlas data sets and accessed through Xenabrowser (https://www.cancer.gov/ccg/research/genome-sequencing/tcga, https://portal.gdc.cancer.gov, https://datacatalog.mskcc.org/dataset/10404, and https://xenabrowser.net, respectively; last accessed August 15, 2024). Patients were assigned to groups based on Gas1 expression (high or low) using a cutoff that maximized survival differences. Survival analysis in each group was performed using R version 4.3.1 with the Survival package.

Statistical Analysis

Prism software version 8.0 (GraphPad Software, San Diego, CA) was used for all analyses, and the two-tailed unpaired t-test or the U-test was employed for comparing differences between two groups. For comparisons among three or more groups, one-way analysis of variance with multiple comparisons was used, and adjusted P values are reported. Data are expressed as means ± SEM. Kaplan-Meier survival curves were generated using the log-rank test. At least three biological replicates were performed for each experiment.

Results

Simultaneous Binding of Wild-Type B, Y, N, and Myc to Sites in the Human and Mouse Gas1 Genes Suggests Conserved, Interactive, and Cooperative Transcriptional Regulation

The intron-less human and mouse Gas1 genes each contain multiple and seemingly evolutionarily conserved consensus binding sites for β-catenin (at Tcf4 sites), YAP/TEAD (at TEAD sites), NRF2 (at ARE sites), MYC (at E boxes), and SP1 and MIZ1 (at their eponymous binding sites) (Figure 1, A and B). SP1 and MIZ1 are positively acting TFs directly inhibited by MYC and regulate the majority of negative MYC targets.^16,37,38 ChIP-seq results from ENCODE showed that each of these TFs directly bound both the human and murine GAS1 genes in proximity to their transcriptional start and end sites and, in most cases, also within the genes' coding regions (Figure 1, C–F).³⁹ MYC binding often overlapped SP1 DNA elements while also coinciding with sites of actual SP1 occupancy. This was particularly prominent around the transcription start sites where SP1 binding also overlapped that of other factors. Elsewhere, binding sites aligned imperfectly with their respective DNA binding elements in the gene body. For example, despite the presence of prominent NRF2 footprints, none were associated with ARE binding sites in either the human or mouse genes. This implicated NRF2 binding as being indirect and perhaps occurring in association with other factors such as MYC or at non-consensus DNA elements. The loose coincidence of MYC and NRF2 binding sites supported the former possibility and was in keeping with their previously described interaction.⁴⁰ Collectively, the findings revealed GAS1 gene regulation to be evolutionarily conserved, complex, and responsive to each of the TFs relevant to HB causation, both individually and combinatorically.

Figure 1.

Conserved binding of relevant transcription factors to the murine and human Gas1 genes. A and B: Intron-less murine (A) and human (B) Gas1 genes, with locations of consensus and evolutionarily conserved binding sites for the indicated relevant TFs. C: Representative examples of binding footprints from the experiments shown in D–F for the indicated TFs in relation to the binding sites shown in A and B from four cell types/tissues [murine liver, human A549 lung cancer cells, murine embryonic stem cells (mESCs), and murine medulloblastoma].³⁹D: Summaries of all available ENCODE data on GAS1 gene binding by MYC³⁹ and Yap/Tead from mouse livers, mouse livers overexpressing YAP, and human MYC-driven HCCs (https://www.ncbi.nlm.nih.gov/geo; accession number GSE83863). Hash marks indicate the geometric means of TF-binding distribution across experiments. E: Summaries of ENCODE data on human GAS1 gene binding by NRF2 and SP1 from the A549 lung cancer cell line (https://www.ncbi.nlm.nih.gov/geo; accession numbers GSE141497 and GSE113497; https://www.encodeproject.org; experiment number ENCSR000BPE), and the HepG2 HB cell line (https://www.encodeproject.org; experiment number ENCSR000BJX).³⁹F: Summaries of ENCODE data on murine Gas1 gene binding of Sp1, Tcf4, and Miz1 in MC3T3 cells, mESCs, mesenchymal stem cells (MSCs), murine embryonic fibroblasts (MEFs), hair follicle stem cells (HFSCs), medulloblastoma tumor spheres, and Myc overexpressing 3T9 fibroblasts (https://www.ncbi.nlm.nih.gov/geo; accession numbers GSE76185, GSE126496, GSE137089, GSE48878, GSM1571163, and GSE98419, respectively) (all sites last accessed August 15, 2024).

Gas1 Transcripts and Protein Are Down-Regulated in Liver Cancer

Gas1 transcripts were markedly reduced in murine HBs that were previously generated with each of the various pair-wise or triple combination of oncogenic B, Y, and N mutant proteins (specifically, the patient-derived in-frame Δ90 B deletion mutation, the patient-derived L30P missense N mutation, and the S127A missense YAP mutation, hereafter referred to unless otherwise stated as B, Y, and N). This was particularly prominent in B + Y + N tumors, the most aggressive of the 4 subtypes (Figure 2A).⁶ HBs generated by a combination of Y and other in-frame deletion and missense mutations of B that previously have been characterized showed a similar reduction of Gas1 transcripts as did slowly growing B + Y tumors generated in Myc^−/−, Chrebp^−/−, and Myc^−/− × Chrebp^−/− double-knockout hepatocyte backgrounds (Figure 2B).^6,16,17 Consistent with these findings, analysis of two previously reported RNA-seq studies of 59 primary human HBs and matched livers showed Gas1 transcript levels to be significantly lower in tumors with higher molecular risk scores (Figure 2C).^12,13

Figure 2.

Gas1 transcript and protein levels decline in liver cancers. A: Expression of Gas1 transcripts in murine livers and HBs generated by the indicated combinations of B, Y, and N and obtained from RNA-seq profiling.⁶ B:Gas1 transcript levels in previously reported murine HBs generated by Y and the indicated patient-derived β-catenin mutants.⁷ Also shown are Gas1 transcript levels in B + Y HBs generated in mice with hepatocyte-specific knockout (KO) of Myc, Chrebp, and Myc × Chrebp (DKO).^16,17C:Gas1 transcript levels in 59 human HBs previously stratified by risk group based on prior molecular profiling.^12,13D:Gas1 transcripts in control livers (L) and HCCs generated by doxycycline-regulated overexpression of a human MYC transgene at days 3 (D3) and 7 (D7) after MYC induction but prior to tumor development. R3 and R7 refer to regressing tumor, 3 days and 7 days after silencing Myc. RT refers to a recurrent tumor induced once again by Myc induction ∼3 months after the original tumor had completely regressed.³⁶E:Gas1 transcripts in human HCCs and adjacent matched livers from the TCGA database. F: Gas1 protein levels in two sets of normal livers and HBs driven by the indicated oncoprotein combinations. the bottom lane of each set shows the expression of Gas1 in B + Y + N tumors in mice that were also injected via hydrodynamic tail vein injection with an SB vector encoding V5 epitope-tagged Gas1 (exogenous Gas1). G: Gas1 protein expression in the liver (L) and HCC tissues from D. Data are expressed as means ± SEM. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, and ∗∗∗∗P < 0.0001. CPM, counts per million; ns, nonsignificant; T, large tumors obtained at ≈30 days after MYC induction; TPM, transcripts per million; WT, wild type.

The induction of undifferentiated HCCs, with some HB-like characteristics, in response to doxycycline-regulated overexpression of a human MYC transgene in mice that were previously reported was also associated with dramatic declines in Gas1 transcripts (Figure 2D).³⁶ This occurred in tumors but not in livers as late as 7 days after MYC induction. Regressing tumors after MYC silencing showed partial normalization of Gas1 expression by day 7. Gas1 transcript suppression was also seen in recurrent tumors induced 3 to 4 months after the original ones had completely regressed. The transcripts were also down-regulated in a large subset of human HCCs from TCGA (Figure 2E) but did not correlate with survival (data not shown). The greater variability of Gas1 expression among these tumors compared to MYC-induced murine HCCs or HBs may reflect their greater molecular complexity and heterogeneity, more variable etiologies and accompanying pathologies.^2,3,5,9,41

Endogenous Gas1 protein expression in murine HBs tended to mirror that of Gas1 transcripts (Figure 2F). B + Y HBs stably generated with a SB vector encoding V5-epitope–tagged Gas1 protein demonstrated 5- to 10-fold higher levels of this exogenous Gas1 than control livers. In both cases, the sizes of the protein were identical, indicating that the V5 epitope tag of the exogenous protein had been efficiently removed during post-translational processing.^26,42

Endogenous Gas1 protein expression in MYC-induced murine HCCs also reflected that of transcripts at each time point (Figure 2G). Thus, declines in Gas1 protein were not appreciated until tumors had developed at around day 30. Small increases in Gas1 protein expression during the early induction phase (D3 to D7) suggested that this might represent an abortive attempt by incipient tumor to restrain uncontrolled growth. As observed with Gas1 transcripts in regressing tumors, protein levels were partially normalized by day 7 and then again became undetectable in recurrent tumors.

Gas1 Inhibits Murine HB Growth

To determine whether Gas1 functions as a TS in vivo as hypothesized, the survival of mice bearing B + Y + N + Gas1 HBs was compared to that in those bearing control B + Y + N tumors, where B, Y, and N represent previously described B^Δ90, Y^S127A, and N^L30P proteins.^6,7 The former group showed ≈50% longer survival and their tumors expressed high levels of the exogenously expressed Gas1 protein (Figure 3, A and B).

Figure 3.

Enforced Gas1 expression extends the survival of HB-bearing mice and inhibits tumor cell growth in vitro. A: Kaplan-Meier survival curves. HBs were generated by hydrodynamic tail vein injection–mediated hepatic delivery of SB vectors encoding B + Y + N or B + Y + N + V5 epitope-tagged exogenous Gas1.^6,7B: Expression of Gas1 protein in four sets of tumors from A. See Figure 2F for additional examples. C: H&E staining of representative tumors from (A). Note fluid-filled cysts with adjacent areas of necrosis (arrow) in B + Y + N HBs as previously described⁷ and their absence in B + Y + N + Gas1 HBs. Higher-power magnification showed similar tumor cell morphologies resembling the crowded fetal variant of HB.¹ See Supplemental Figure S1 for additional examples. D: Expression of Gas1 in the indicated HB cell lines. Each of the indicated cell lines was stably transfected with an empty pSBbi-RP vector or the pSBbi-RP–Gas1-V5 expression vector and selected in puromycin. The panel above the immunoblots shows the structure of the vectors. E: Gas1 overexpression minimally inhibits BY and BYN cell line proliferation. Stably transfected cells were selected in puromycin for 1 week and seeded at equal number for growth curve. F: HB cell growth is inhibited by Gas1 in a dose-dependent manner. Two days after stable transfection with empty pSBbi-RP or pSBbi-RP–Gas1-V5 vectors, cells from the indicated lines were stratified by dTomato expression (low or high). Cell growth was then monitored as in E during puromycin selection. See Supplemental Figure S2 for images of the surfaces of representative wells at the termination of the study. Data are expressed as means ± SEM (some error bars were too small to be displayed). ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, and ∗∗∗∗P < 0.0001. Scale bars: 400 μm (low magnification); 50 μm (high magnification). EF1a, elongation factor 1 alpha promoter; ns, nonsignificant; OE, overexpression; PuroR, puromycin resistance cassette.

B + Y and B + Y + N murine HBs resemble the crowded fetal subtype of human HB.^1,6,7,43 The latter, however, also contain innumerable fluid-filled cysts, which are only rarely observed in human HBs.^6,10 These were again seen in the control B + Y + N tumors generated for the current study (Figure 3C and Supplemental Figure S1). Also as previously reported, the cysts were often adjacent to well-demarcated regions of necrosis.⁶ B + Y + N + Gas1 tumors contained neither of these features, suggesting that they were either properties of the high growth rates of B + Y + N HBs or that they were suppressed by Gas1.

To evaluate the consequences Gas1 restoration in greater detail, several immortalized cell lines were generated from primary B + Y and B + Y + N HBs.³⁴ This was enabled by the in vivo Crispr-mediated targeting of the Cdkn2a gene, which inactivated its two encoded TS proteins (p16^INK4A and p19^ARF).^35,44 Like the primary tumors from which they originated, these cell lines expressed little to no endogenous Gas1 but did express high levels of the exogenous protein (Figure 3D). Yet, in no case did this inhibit growth by more than ≈15% (Figure 3E). It was noted, however, that the intensity of dTomato expression declined significantly in Gas1-transfected cells, suggesting that Gas1 was in fact exerting more growth suppression than initially supposed and that cells with the highest levels of expression were being selected against. The above growth curves were therefore repeated after first sorting high- and low-intensity dTomato⁺ populations into separate wells shortly after their transfection. Doing so showed that low-dTomato cells grew at about half the rate as did control cells, whereas high-dTomato cells grew at only about one-eighth the rate as in BY1, with similar levels of growth suppression being observed in BY and BYN cell lines (Figure 3F and Supplemental Figure S2). It is thus concluded that high-level Gas1 expression is selected against but that maintaining it efficiently suppresses tumor cell proliferation in a manner that is both p16^INK4a- and p19^ARF-independent.

HB Growth Suppression Is Dependent on GPI-Dependent Attachment of Gas1 to the Outer Plasma Membrane

Growth inhibition by Gas1 is pleiotropic by virtue of its cell surface interactions with membrane receptors for the glial cell line–derived neurotrophic factor (GDNF) and/or hedgehog signaling pathways.^{26,28,42,45,46} In the former case, it blocks GDNF-mediated signaling by the co-receptors glial cell line-derived neurotrophic factor family receptor (GFR)-α₁ and rearranged during transfection (Ret) and suppresses the downstream growth-promoting activation of the phosphatidylinositol 3 kinase (PI3K)-Akt axis in a manner that requires GPI-mediated anchoring of Gas1 to the outer membrane.^28,47,48 However, soluble Gas1, lacking the GPI anchor, can also act in a paracrine and/or autocrine manner in association with GDNF and RET.^23,30,49,50 To examine this in HB cells, the expression, subcellular localization, and growth-inhibitory properties of full-length Gas1 were compared to those of an attachment-defective and secretable form lacking the GPI modification (Figure 4A). Both proteins were expressed at similar levels within cells with only the truncated version of Gas1 being detected extracellularly (Figure 4B). However, its apparent molecular weight now corresponded to that of the mature GPI-linked wild-type protein seen in whole-cell lysates. This finding suggests that the absence of a GPI tag led to aberrant, non–GPI-related post-translational modification(s) prior to secretion. Further examination of stably transfected cells by immunofluorescence staining showed the mature, GPI-linked protein to be both cytoplasmic- and membrane-localized, whereas the GPI attachment–defective form was less abundant, more diffuse, and absent from the membrane (Figure 4C). Unlike the growth inhibition imposed by constitutively expressed mature Gas1 (Figure 3F), GPI attachment–defective Gas1 inhibited growth minimally, if at all (Figure 4D). Tissue culture fluid from these latter cells, even when concentrated >10-fold, was unable to inhibit control BY1 cell growth (Figure 4E). Moreover, no differences in either total or phosphorylated Akt were seen between control cell and those overexpressing either form of Gas1 (Figure 4F). Finally, based on previously reported RNA-seq data,^6,16,17 virtually no RET or GDNF transcripts were detected in livers or in BY and BYN HBs (range, 0 to 0.02 reads per kilobase million [RPKM]) (data not shown).

Figure 4.

Gas1-mediated growth suppression requires its GPI-mediated tethering to the outer plasma membrane. A: Structure of murine Gas1 proteins. Top: Unmodified, full-length Gas1 precursor protein encoded by the pSBbi-RP vector. In the endoplasmic reticulum (ER) and Golgi, it is converted to the processed/mature, GPI-linked form after removal of its N-terminal residues (≈38; red), its C-terminal residues (≈28; green), and the 14-residue V5 epitope tag (blue). It is then further modified by the addition of a GPI moiety to a newly exposed C-terminal Ser₃₁₅. Bottom: Mature version of the GPI attachment–defective, secreted form of Gas1, which terminates at Gly₃₁₄, and cannot be modified by GPI addition. B: Expression of the processed forms of Gas1 shown in A in transfected BY1 cells. Given that the extracellular compartment into which soluble Gas1 was secreted contained no β-actin, Ponceau S staining of the membrane was performed instead to confirm equal amounts of serum proteins on the membrane. C: Confocal immunofluorescence images of Gas1 localization (green) in BY1 cells expressing the Gas1 proteins shown in A. Arrows in the left panel indicate regions of plasma membrane localization. dTomato expression (red) is superimposed, as is nuclear staining with Hoechst 33342 (blue). In the right panel, cells expressing the GPI attachment–defective, secreted form of Gas1 show much less intense Gas1 staining, which is only cytoplasmic. D: Growth curves of control BY1 cells and those expressing the GPI attachment–defective, secreted form of Gas1. the brightest populations of dTomato⁺ cells were selected and plated as described in Figure 3F, and cell proliferation was quantified over a 4-day period. E: Lack of BY1 cell inhibition by soluble Gas1. Control cells or those expressing the indicated forms of Gas1 (A–D) were grown to 100% confluence. After reduction of the serum concentration to 0.1%, the supernatants were collected daily over 3 days, combined, and concentrated 10-fold. Nontransfected naïve BY1 cells were then exposed to the collected media, and cell counts were performed as in D. F: Akt phosphorylation is not impacted by the overexpression of either mature GPI-linked or GPI attachment–defective Gas1. Control BY1 cells or those overexpressing the indicated Gas1 proteins were subjected to immunoblotting for total and pSer⁴⁷³ Akt. G: Up-regulation of hedgehog pathway direct-target genes in human HBs. Heatmaps of RNA-seq data from two studies totaling 59 human HBs^12,13 comparing the expression of 224 known Hedgehog pathway direct-target genes from the Molecular Signatures Database (https://www.gsea-msigdb.org/gsea/msigdb, last accessed August 1, 2024). All values were expressed relative to matched liver samples uninvolved with tumor. Data are expressed as means ± SEM. ∗∗P < 0.01. Scale bars = 50 μm (C). ns, nonsignificant; WT, wild type.

The growth-suppressive effects of Gas1 via the hedgehog signaling pathway is GPI-dependent.^46,51, 52, 53 This was investigated in the previously mentioned RNA-seq data sets from 59 human HBs,^12,13 with evidence for variable degrees of up-regulation of 224 direct target genes of the pathway (Figure 4G), indicating that it was active in many if not all of these tumors and in keeping with the down-regulation of Gas1 transcripts that were originally documented (Figure 2C). Consistent with this finding, transcripts encoding zinc finger proteins Gli1, Gli2, and Gli3, the three TFs involved in activating these hedgehog target genes were also up-regulated (≈2.5- and ≈3.1-fold, respectively; q < 10⁻³).^51,54 The two studies differed somewhat with regard to whether the up-regulation of the target genes was more pronounced in tumors considered to be at high risk versus those with more favorable outcomes. Differences in the ways that patients were classified, the methods of analyses, the quality of the mRNAs used for analyses, or underlying molecular heterogeneity may have accounted for this finding. Nonetheless, the studies did establish up-regulation of hedgehog target gene expression to be a common feature of human HBs as previously reported,^54,55 and that it generally correlates with the concerted down-regulation of Gas1 expression.

In contrast to the results discussed in the previous paragraph, significant up-regulation of hedgehog signal-related transcripts was not demonstrated in any of the murine HBs, including those from the most aggressive BYN molecular subtype, and Gli1-3 transcripts were undetectable (<0.01 RPKM in each case) (data not shown). These results led to the conclusion that HB growth suppression by Gas1 requires anchoring to the membrane via its GPI moiety and does not mediate its effects via suppression of the GDNF-GFRα₁-RET-PI3K-Akt signaling axis. in contrast, it may utilize hedgehog signaling to suppress growth in human tumors.

Discussion

B, Y, and N are TFs with hundreds or thousands of unique and shared positive and negative targets.6, 7, 8^,17,36,41,43 The search for genes that are always regulated in the same manner among a variety of different HB types, that is, BYN genes, thus represented an attempt to distill the immense transcriptomic complexity of these tumors so as to allow the identification and prioritization of the most relevant and proximal drivers of tumorigenesis.⁷ An underlying premise of this approach is that such genes are regulated in the same direction regardless of the TF combinations driving the tumor. This regulation was predicted to be evolutionarily conserved and independent of the growth rate of the tumor, its genetic background, and the direction of expression of other transcripts without direct roles in transformation.^5,6 By comparing the gene expression profiles of multiple individual murine HBs generated by all possible combinations of B, Y, and N, by different B mutants and in Myc−/− and/or Chrebp−/− hepatocyte backgrounds, the 22-member BYN gene set that met these criteria was identified.^6,7,16,17 Most of its members possessed previously described functions that could potentially contribute to one or more cancer hallmarks.^6,19 Many of them were previously identified as influencing normal and/or neoplastic proliferation in addition to being deregulated in HCC and other cancers in ways that often correlate with survival.^5,6,11,13

As one of the most highly down-regulated BYN genes (>22-fold in B + Y + N HBs),⁶ Gas1 exerts growth-suppressive effects in a variety of both transformed and untransformed cell types.^{21,24,26,28,30,31,42,49} The finding that restoring Gas1 expression both in vivo and in vitro, inhibited tumor cell growth (Figure 3, A and F) not only confirmed this role in the context of HB but also validated the argument that BYN genes are relevant for tumor initiation and/or maintenance as originally proposed.^5,6

Gas1 is invariably suppressed in all examined molecular subtypes of HB and in Myc-driven HCCs to a degree that reflects the identities of the underlying oncogenic driver(s) (Figure 2, A–D).⁶ This finding agrees with the observation that multiple, evolutionarily conserved binding motifs for β-catenin/Tcf, YAP/Taz, NRF2, MYC, SP1, and MIZ1 exist within the intron-less GAS1 gene, with many of these sites being occupied by their respective factors in livers, liver cancers, and other normal and neoplastic cell types (Figure 1). This suggests that each factor exerts substantial, independent, and redundant negative transcriptional control over Gas1, but that cooperative and more nuanced suppression can be achieved combinatorically. This was readily seen in HCCs in which Myc overexpression alone was sufficient to achieve a degree of Gas1 suppression rivaling that seen in any of the HB groups (Figure 2, A, B, and D). Notably, most of these factors, as well as the Myc-interacting cofactors Sp1 and Miz1,^37,38 also bound at the 3′-end of the gene in proximity to the transcription end site (Figure 1). It was recently reported that many human and murine genes show Myc and/or Max binding in proximity to these sites and in association with numerous other TFs, cofactors, and histone modifiers; an open chromatin environment; and local nuclease susceptibility.⁵⁶ Functionally, Myc binding around transcription end sites contributes to chromatin looping while regulating total as well as read-through transcription that has been reported to modulate nucleus→cytoplasm transport and translation in response to various stresses, including oncogene overexpression.^56,57

The immortalized HB cell lines used in this work could be generated only in the face of concurrent Cdkn2a locus inactivation.³⁴ Cdkn2a encodes two TSs that are crucial regulators of proliferation and survival, with p16^INK4a promoting replicative and oncogene-induced senescence via the Rb pathway and p19^ARF indirectly stabilizing Tp53 to promote cell-cycle arrest and apoptosis.^44,58 These pathways share considerable crosstalk, with mutations or blocks in each being extremely common in human cancers and immortalized cell lines.^35,59,60 GAS1 relies on TP53 to promote growth suppression in several cell types.^27,32 This may explain some of the growth-suppressive effects of Gas1 on primary HBs, which contain no Cdkn2a locus mutations and express elevated levels of wild-type p16^INK4A and p19^ARF (Figure 3A).³⁴ In contrast, the immortalized HB cell lines derived from these tumors contain heterogeneous Crispr/Cas9-generated Cdkn2a mutations and express no wild-type p16^INK4a or p19^ARF proteins.³⁴ Thus, the growth inhibition that is mediated by mature GPI-linked Gas1 in primary HBs, and that cooperates with Cdkn2a-encoded p16^INK4A and p19^ARF, may rely on pathways that are independent of Cdkn2a, Rb, and Tp53 in HB cell lines. It is thus possible that the lack of any discernible growth inhibition when HB cell lines are cultured in the presence of extracellular GPI detachment–defective Gas1 (Figure 4E) reflects, at least in part, the Cdkn2a-related differences between primary HB tumors and cell lines.

Soluble, secreted Gas1, lacking the GPI anchor, has been reported to inhibit the growth of certain tumor cells, particularly those of neural origin.29, 30, 31 This inhibition occurs via the engagement of Gas1 with the cell-surface protein RET, a GDNF co-receptor, which leads to the suppression of RET-mediated PI3K-Akt signaling.47, 48, 49 The inability to suppress HB cell growth in vitro with high concentrations of GPI attachment–defective secreted Gas1 or to demonstrate any changes in Akt signaling in Gas1-producing cells (Figure 4, E and F) likely reflects the tissue-related differences between these cells and those of neural origin, as well as the fact that HBs express little, if any, RET or GDNF. The fact that mature, GPI-linked Gas1 does markedly inhibit HB cell proliferation both in vivo and in vitro (Figures 3F and 4A, and Supplemental Figure S2) suggests that it may instead function via an alternate pathway involving inhibitory interactions with the hedgehog pathway leading to cell cycle arrest.⁴⁶ However, the fact that the activation of this pathway was demonstrated only in human HBs and not in experimentally induced murine tumors suggests that Gas1-mediated growth suppression in the latter involves novel and perhaps cytoplasmicly localized Gas1 functions, as well as yet-to-be identified membrane-localized and Cdkn2a-independent ones.

Importantly, the ectopic restoration of Gas1 slows, but does not prevent, HB tumorigenesis driven by the triple combination of B + Y + N that is associated with particularly aggressive HBs and robust Gas1 suppression (Figure 3A).⁶ Thus, the current work, while demonstrating directly that Gas1 does function to suppress tumor growth as originally postulated,^5,6 also indicates that other factors, quite likely other members of the BYN gene group, must be concurrently normalized to achieve a deeper and more durable extent of tumor suppression. In the future, it will be important to better identify and characterize the interrelationships among these.

Disclosure Statement

None declared.

Supplemental Data

Supplemental Figure S1.

Additional examples of the differential histologic analyses of the B + Y + N and B + Y + N + Gas1 tumors shown in Figure 3C. in control B + Y + N tumors, note the presence of numerous cysts and commonly associated regions of adjacent necrosis. Scale bar = 900 μm.

Supplemental Figure S2.

A and B: Images of the surfaces of representative wells of the study in Figure 3, E (A) and F (B). Scale bars = 300 μm.