c-Jun and Fra-2 pair up to Myc-anistically drive HCC

ABSTRACT

Hepatocellular carcinoma (HCC), a leading cause of cancer-related death with limited therapies, is a complex disease developing in a background of Hepatitis Virus infection or systemic conditions, such as the metabolic syndrome. Investigating HCC pathogenesis in model organisms is therefore crucial for developing novel diagnostic and therapeutic tools. Genetically engineered mouse models (GEMMs) have been instrumental in recapitulating the local and systemic features of HCC. Early studies using GEMMs and patient material implicated members of the dimeric Activator Protein-1 (AP-1) transcription factor family, such as c-Jun and c-Fos, in HCC formation. In a recent report, we described how switchable, hepatocyte-restricted expression of a single-chain c-Jun~Fra-2 protein, functionally mimicking the c-Jun/Fra-2 AP-1 dimer, results in spontaneous and largely reversible liver tumors in GEMMs. Dysregulated cell cycle, inflammation, and dyslipidemia are observed at early stages and tumors display molecular HCC signatures. We demonstrate that increased c-Myc expression is an essential molecular determinant of tumor formation that can be therapeutically targeted using the BET inhibitor JQ1. Here, we discuss these findings with additional results illustrating how AP-1 GEMMs can foster preclinical research on liver diseases with novel perspectives offered by the constantly increasing wealth of HCC-related datasets.

Introduction

Primary liver cancer is a leading cause of cancer-related death worldwide with increasing incidence and limited therapeutic options, particularly for advanced disease [1,2]. Hepatocellular carcinoma (HCC), accounting for a large fraction of liver cancers, often arises in the context of chronic liver diseases, like hepatitis and/or metabolic (dysfunction)-associated fatty liver diseases (MAFLD). The increased relevance of non-viral risk factors for HCC, such as excess body weight, diabetes, and alcohol consumption, is a major concern, aggravated by the lack of biomarkers for early diagnosis and to help stratify patients for targeted therapies [1,3].

Despite numerous studies, therapeutically relevant mechanistic knowledge for this heterogeneous and highly complex disease remains limited. To enable functional evaluation of candidate pathogenic pathways as well as preclinical assessment of therapeutic strategies, a wide range of murine preclinical models have been developed that mimic the major characteristics of human HCC [4–6]. Although no single model can fully replicate HCC heterogeneity, these experimental paradigms can collectively offer clinically relevant insights into HCC mechanisms that can be further validated using human HCC samples and datasets.

Activator protein 1 (AP-1) is a dimeric transcription factor implicated in a plethora of cellular programs. AP-1 comprises a collection of DNA-binding homo- and heterodimers formed by members of the JUN (c-Jun, JunB, and JunD) and FOS (c-Fos, FosB, Fra-1, and Fra-2) families. AP-1 dimers bind to redundant, but also specific sequence elements in the promoters and enhancers of target genes [7]. The composition of AP-1 dimers is determined by protein levels of individual components, often modulated by upstream signaling events. Various AP-1 dimers coexist, and dimer-specific variations in target DNA sequence affinity and/or co-activator/repressor recruitment dictate the genes that are positively or negatively regulated by AP-1 in a given cell type and biological context [8,9].

Understanding AP-1 function in the liver has been the focus of attention for many years. c-Jun, c-Fos, and more recently Fra-1 and Fra-2, have been found increased in human HCC [10–12], and genome-wide expression analysis revealed that AP-1 is central in the oncogenic signaling network of poor prognosis HCCs [13]. In genetically engineered mouse models (GEMMs), hepatocyte-specific gene inactivation experiments demonstrated that individual Jun and Fos genes are largely dispensable for adult liver homeostasis while having important and specific roles in the response of the liver to various stress conditions [14–23]. Concerning liver cancer, mice lacking hepatic c-Jun are resistant to a wide range of experimental HCC paradigms, with alterations in major molecular players involved in cell proliferation, survival, and DNA damage repair, such as p53, Sirt6/Survivin, p21, c-Fos, and c-Myc that are sometimes paradigm-specific [24–28]. On the other hand, hepatocyte-specific inactivation of c-Fos leads to a significant decrease in tumor burden in the diethylnitrosamine (DEN)-induced liver cancer model [29], whereas we have recently shown that Fra-2 inactivation has no overt effect in this setting [24]. Extending these loss-of-function analyses to additional HCC paradigms and/or to more Jun and Fos genes is desirable even if these experiments have largely established that AP-1 dimers, particularly those formed by c-Jun, are essential for hepatocarcinogenesis.

Switchable monomers and forced dimers to evaluate AP-1 function

We also set out to investigate whether increased expression of specific AP-1 forming proteins, and, perhaps more importantly, specific AP-1 dimers would potentiate hepatocarcinogenesis. We generated a collection of tetracycline-inducible AP-1 alleles, all following the same embryonic stem cells-based, site-specific recombinase-mediated strategy [30]. These alleles allow switchable and tissue-specific expression of AP-1-forming monomers as well as single-chain AP-1 dimers, where one Jun and one Fos protein are forced to dimerize by virtue of a flexible polypeptide tether [31].

Using these alleles, hepatic expression of a single Fra-1 or Fra-2 monomer prevented and even reverted HFD-induced hepatosteatosis in adult mice by suppressing PPARγ, whereas single inactivation of either of the two genes had no effect [15,32]. In contrast, c-Fos activated hepatic Pparg gene transcription while increasing oxysterols and DNA damage by suppressing LXRα [15,29]. On the other hand, single-chain c-Jun~Fra-2 dimers inhibited, whereas c-Jun~c-Fos, JunB~c-Fos, and JunD~c-Fos dimers activated PPARγ expression and signaling, when expressed in the liver of adult AP-1^hep mice [15,32]. Thus, we could document for the first time in vivo how specific AP-1 dimers exert redundant or antagonistic effects on the expression of a single gene and on lipid metabolism in the liver. These switchable AP-1 alleles have also proven useful to study AP-1 function beyond the liver, in skin epidermis and bone-forming osteoblasts [33–35].

Liver tumors exclusively in c-Jun~Fra-2^hep mice

Surprisingly, considering the similar lipid handling phenotype of these mice, liver tumors were only observed in c-Jun~Fra-2^hep mice kept for several months on normal chow, whereas mice expressing Fra-1/2 monomers or Jun~Fra-1 dimers remained tumor-free [24]. Serum Alpha-fetoprotein (AFP) and Protein-Induced-by-Vitamin-K-Absence-or-antagonist-II (PIVKA), two biomarkers associated with HCC, were specifically and consistently elevated in c-Jun~Fra-2^hep mice, although only late (PIVKA) or to a variable extent (AFP). Liver histology was also heterogeneous with hyperplastic nodules, adenoma, and HCC observed in liver sections and documented by immunohistochemistry for HCC markers. Serum transaminases were increased, consistent with an early onset liver dysfunction and damage. Aberrant cell cycle and replicative stress, senescence, and endoplasmic reticulum (ER) stress were also evident in the livers of c-Jun~Fra-2^hep mice, already at early time points. Genome-wide transcription profiling by bulk RNAseq and subsequent computational analyses using popular analytical tools such as Gene set enrichment analysis (GSEA), WEB-based GEne SeT AnaLysis Toolkit (WebGestalt), Metascape, and CIBERSORTx [36–40] revealed perturbed liver zonation, dyslipidemia, low-grade fibrosis, and mild inflammation with increased myeloid cell abundance in c-Jun~Fra-2^hep livers. Importantly, while some inter-tumoral heterogeneity was noted in the bulk transcriptional profile of the four Jun~Fra-2^hep tumors that were included in the sequencing, these harbored liver cancer gene signatures, particularly those of human HCC with poor outcome and/or with upregulated Myc signaling as well as liver tumors arising in mice expressing a Myc transgene [24].

Unique and shared features revealed by mRNA profiling

We recently performed RNAseq profiling of 2-months liver samples from Fra1^hep and Fra2^hep mice [41]. Initial unsupervised principal-component analysis (PCA) after inclusion of the 2-months datasets from Jun~Fra-2^hep mice indicated that the three mutant cohorts segregate from all controls and from each other, according to genotype (Figure 1a). Interestingly, Fra2^hep liver samples were closest to controls along the major/first dimension (PC1), whereas Jun~Fra-2^hep liver samples were the most distant (Figure 1a). Consistently, computational analysis of genes changed by at least 2-fold compared to the pool of controls (DEG), revealed that Jun~Fra-2^hep liver samples had the highest number of DEGs followed by Fra1^hep and then Fra2^hep. Interestingly, BioVenn comparison [42] revealed that the vast majority of the Fra1^hep and Fra2^hep DEG were present in the Jun~Fra-2^hep DEG lists, although Fra1^hep and Fra2^hep DEGs displayed a partial overlap (Figure 1b). This suggests that, when ectopically expressed in hepatocytes, Fra1, and Fra2 monomers would form dimers with endogenous, and likely rate-limiting, Jun proteins to regulate both overlapping and distinct gene sets. In line with our previous publications [15,24], Pparg and Fabp1 were among the 32 DEGs downregulated in the three mutant datasets, and preliminary Metascape analysis revealed that the 193 DEGs shared by Fra1^hep and Jun~Fra-2^hep livers were enriched in Gene Ontology terms related to the extracellular matrix, consistent with dysregulated fibrosis-associated genes in Fra1^hep [14] and Jun~Fra-2^hep livers. CIBERSORTx computational deconvolution confirmed that Fra2^hep livers were the most similar to controls with respect to myeloid cell subset distribution, whereas a slight increase in some myeloid subsets was noted in Fra1^hep livers (Figure 1c). Collectively and although it awaits in-depth analysis and experimental evaluation, this initial comparative analysis of hepatic Fra1^hep, Fra2^hep, and Jun~Fra-2^hep transcriptional profiles is in agreement with our previous findings [15,24] and validates the usefulness of unbiased transcriptomic approaches.

Figure 1.

AP-1 (JUN/FOS) complexity in liver diseases.

A. Principal component analysis (PCA) of RNAseq data from Fra-1^hep, Fra-2^hep and c-Jun~Fra-2^hep liver samples after 2 months of transgene expression. PC1 and PC2 account for 41.4% of sample variability. Individual samples from mutants (n = 3 for each genotype) and controls (n = 12) are depicted with squares and circles, respectively. B. Comparison of differentially expressed genes (DEG, Up: left and Down: right) between Fra-1^hep, Fra-2^hep and c-Jun~Fra-2^hep liver RNAseq datasets at 2 months using area-proportional Venn diagrams. DEG with more than 2-fold change and with adjusted p value p < 0.01 were selected. Total DEG (parentheses) and the number of DEG in each subset are indicated. Mutants: n = 3 per genotype, controls: n = 12. C. CIBERSORTx deconvolution of liver myeloid cell subsets in Fra-1^hep, Fra-2^hep and c-Jun~Fra-2^hep liver RNAseq datasets at 2 months (mutants: n = 3 per genotype, controls: n = 12). Bars = means ±SEM. *p < 0.05, ***p < 0.001, two-tailed Student’s t test to controls, total abundance is evaluated in red. D. Working model how Jun/Fos and c-Jun/Fra dimers modulate liver diseases as revealed by AP-1 GEMMs. While c-Jun forms AP-1 dimers with either Fra-1 and Fra-2 to repress Pparγ signaling and suppress fatty liver disease (MAFLD), c-Jun/Fra-2 dimers increase c-Myc transcription to drive liver tumorigenesis and HCC. On the other hand, c-Fos increases Pparγ expression when dimerizing with any Jun protein, thus opposing Fra1/2-containing dimers. c-Fos also promotes hepatocyte transformation by increasing LXR and c-Myc expression. Statins and JQ1 treatment dampen the deleterious effects of ectopic c-Fos or c-Jun~Fra-2 expression, respectively.

Two important findings emerged from these comparative analyses. The first one is that tumors develop in c-Jun~Fra-2^hep livers despite the dyslipidemic, low hepatic fat context, whereas an overall similar context does not lead to tumorigenesis in Fra1^hep and Fra2^hep livers. The second is the pathogenic link between c-Jun/Fra-2 and c-Myc, an important and crucial tumor-initiating event, that is observed exclusively in c-Jun~Fra-2^hep livers and likely contributes to early pre-neoplastic events. These include hepatocyte proliferation, replicative and ER stress, inflammation, and the expression of senescence and dedifferentiation markers. Whether sustained expression of Jun~Fra-2 is needed for the progression of these pre-neoplastic cells to fully transformed hepatocytes is certainly worth assessing. Tet-switchable alleles allow such experiments where c-Jun~Fra-2 expression is restricted to a short-time frame and the outcome evaluated several months later. Using Fos^hep mice, we have shown that early and time-restricted c-Fos expression is not sufficient to drive hepatocarcinogenesis up to 6 months later, while it strongly potentiates DEN-tumorigenesis [29]. Combining short c-Jun~Fra-2 expression with DEN or with treatments modulating inflammation or fibrosis would therefore also be insightful.

Switching tumors ON/OFF to investigate oncogene addiction

In our published study, we exploited the “tetracycline switch” to address another question: is c-Jun~Fra-2 expression necessary to maintain a tumor once fully established? In other words, would switching off the initiating oncogenic driver induce rapid and sustained tumor regression, a phenomenon known as oncogene addiction? The rationale was two-fold: first, while c-Jun/AP-1 is a bona fide oncogene in HCC, inducible loss-of-function GEMMs revealed that c-Jun was only required in the early stages of DEN-tumorigenesis but not in established tumors [25,27]. On the other hand, as oncogene-addiction is largely demonstrated in Myc-driven HCCs [43], the involvement of Myc in c-Jun~Fra-2-driven tumorigenesis might render these tumors sensitive to the suppression of the initial oncogenic hit, even if partially, and this could have important implications for HCC therapy.

Longitudinal blood and ultra-sound monitoring of c-Jun~Fra-2^hep mice together with end-point analyses revealed that most tumors appeared initially addicted to c-Jun~Fra-2, with a sharp and rapid drop in serum AFP and tumor volume. However, when hepatic c-Jun~Fra-2 expression was switched off the fate of individual tumors was heterogeneous and independent of their initial size. Some tumors regressed, but others resumed growing after a variable time, along with the appearance of neo-tumors and a “rebound” in serum AFP. Strikingly, reversion-escapers analyzed at the end-point had no detectable c-Jun~Fra-2, but maintained c-Myc expression in the tumoral area, while it was decreased to control levels in adjacent non-tumoral areas [24]. These data indicate that tumors initiated by the c-Jun/Fra-2 AP-1 dimer are more addicted to increased c-Myc than to the initial oncogenic driver. Consistently, interfering pharmacologically with c-Myc expression and activity by treating tumor-bearing c-Jun~Fra-2^hep mice with the BET bromodomain inhibitor JQ1 resulted in a tumor-static effect without notable impact on c-Jun~Fra-2 expression. JQ1 also slowed c-Jun~Fra-2^hep liver tumors when co-administered with Sorafenib, a widely used HCC drug. Furthermore, JQ1 had a beneficial effect on pre-neoplastic events occurring in c-Jun~Fra-2^hep mice, when administered in a preventive setting, reducing hepatocyte proliferation and DNA-damage indexes [24]. This is in line with the idea that the early increase in c-Myc expression observed in c-Jun~Fra-2^hep livers but not in c-Jun~Fra-1^hep, Fra-1^hep or Fra-2^hep livers, is functionally relevant for the tumor phenotype observed at later stages.

Outlook: what can AP-1^hep mice teach us

The mechanistic basis for sustained c-Myc expression in escaping tumors is likely complex with tumor cell intrinsic and extrinsic mechanisms. Increased Fos expression was observed in some c-Jun~Fra-2^hep escapers, whereas pre-neoplastic c-Fos^hep livers had elevated c-myc mRNA. Whether c-Fos-containing dimers promote hepatic c-Myc transcription directly through the c-Jun/Fra-2-binding site in the c-myc enhancer and/or the putative AP-1 site in the c-myc promoter that is not bound by c-Jun~Fra-2 [24] awaits experimental evaluation. We previously reported that c-Jun~c-Fos, JunB~c-Fos, and JunD~c-Fos dimers increased Ppary2 expression in the liver, counteracting Fra-1- and Fra-2-containing dimers on the same AP-1 binding site [15]. Ongoing work using mice with hepatocyte-specific, switchable Jun/Fos forced dimer alleles will certainly shed light on the role of individual Jun/Fos dimers in modulating c-Myc expression and their effects on liver homeostasis. We hypothesize that increased c-Myc expression in c-Fos^hep livers potentiates the deleterious effects of LXR-mediated dysregulated cholesterol metabolism that occurs in these livers. Since Statins only partially reversed the pre-neoplastic phenotype in c-Fos^hep mice [29], JQ1 co-treatment could lead to synergistic benefits in these mice (Figure 1d).

Many questions remain unanswered: The cellular and molecular events occurring immediately after switching off c-Jun~Fra-2 expression in tumors, the involvement of Fos-containing dimers or other molecules in the maintenance of Myc expression, the contribution of tumor-extrinsic events, such as inflammation, or cell types, like macrophages, and the connection of all these events to the various pro-tumorigenic functions of c-Myc are only a few examples. Not least is tumor heterogeneity that contributes significantly to late diagnosis, drug resistance, and treatment failure in HCC [44]. We plan to tackle these important questions by unbiased DNA, RNA, and proteome profiling of a large number of tumors and non-affected liver samples collected from AP-1^hep GEMMs in different ON and OFF settings. Using more advanced technologies, such as single-cell and/or in situ OMICs, coupled with computer-assisted multiplatform analyses will certainly help dissecting tumor sub-types, immune cell infiltrates and cell–cell interactions within and around the tumors. We previously documented how c-Fos/AP-1 expression in premalignant keratinocytes promotes inflammation-mediated pre-neoplastic lesions by recruiting T cells [34]. Whether a similar bidirectional cross-talk between premalignant hepatocytes and infiltrating immune cells would also operate in the liver is certainly worth investigating.

The datasets generated using AP-1^hep mice will also be compared with the wealth of OMIC data generated using c-Myc-switchable liver mice [45,46], HCC GEMMs and patient material. Including tumor samples from mice treated for a short period with JQ1 or compounds directly targeting Myc, such as the recently described in vivo delivery-compatible Myc antisense oligonucleotides [47], might also provide additional insights into the Myc-dependent gene sets involved in reversion escape mechanisms. Given the involvement of Myc signaling in the dysregulation of tumor cell microenvironment and in immune evasion [43,45], a particular focus will be set on the interactions between pre-neoplastic or fully transformed hepatocytes and their non-parenchymal environment. Importantly, c-Myc and AP-1 are among the transcription factors that can modulate the expression of immune checkpoint molecules in some tumor types, thus facilitating tumor evasion [48]. Evaluating the expression and transcriptional regulation of immune checkpoint molecules in the above described experimental settings will also be informative.

An interesting clue into the tumorigenesis pathway via which c-Jun/Fra upregulates c-Myc to elicit HCC involves the RNA-binding protein FMRP, which has been recently reported to orchestrate an immunosuppressive TME [49], in addition to stimulating invasion and metastasis in other tumor types [50,51]. FMRP is broadly upregulated in human HCC [49], and c-Myc has been implicated in its upregulation in human cancer cells, via siRNA K/D and inhibition with JQ1 [49], and in a GEMM of PDAC with a switchable c-MYC oncogene [49,52]. While preliminary analysis revealed an increased FMRP HCC signature score [49] in c-Jun~Fra-2^hep liver tumors, future studies are warranted to illuminate FMRP’s roles in HCC phenotypes, such as immune evasion and invasiveness, and its regulation by the Jun/Fra-Myc axis described herein.

Unlocking HCC: the power of public data and bioinformatics

Our efforts to delineate the mechanistic basis of hepatocarcinogenesis in c-Jun~Fra-2^hep mice has greatly benefited from publicly available data analysis tools and databases, in particular, those related to human liver cancer cell lines and tumor samples. Web-based or downloadable packages have made data access, analysis, and interpretation accessible to non-computer experts and allowed our bioinformatics collaborators to concentrate on more elaborated tasks. Mining publicly available cancer-related databases and datasets is invaluable to basic research as it enables the rapid identification of patterns and associations that can subsequently be validated experimentally. Two open access and largely publicly funded, ground-breaking initiatives were instrumental for our recent publication: the Encyclopaedia of DNA Elements (ENCODE) project [53] for human MYC gene transcription analyses and the Cancer Genome Atlas (TCGA) [54], to validate the pathogenic interaction between JUN/FRA2 and MYC in HCC samples. Nevertheless, our analysis of HCC datasets has two limitations. The first is that only mRNA could be analyzed for JUN and FRA2. Post-transcriptional events, including phosphorylation-mediated modulation of protein stability, are important regulators of AP-1 [8,9], and it would be ideal to additionally probe the correlations between JUN and FRA2 protein co-expression and MYC mRNA levels. The second limitation is that the actual LI-TCGA cohort does not cover the full spectrum of HCC, or at least not in sufficient numbers to allow informative analyses across HCC subtypes or underlying liver diseases. Given our observations that c-jun inactivation decreases c-Myc expression in the HBV^Tg but not in the DEN models of HCC, assessing the connection between JUN/FRA2 and MYC across HCC etiologies, for example, viral hepatitis, MAFLD, or alcohol consumption, could be of particular interest.

On a global scale, more than two-thirds of liver cancer cases are secondary to chronic Hepatitis B or C Virus infections, and the majority of diagnosed HCC patients are Asians, but non-HBV-related samples and/or Western patients are still over-represented in the current large-scale analyses. Various initiatives, such as the Pan-Cancer Analysis of Whole Genomes (PCAWG), the Indian Cancer Genomics Atlas (ICGA), and GenomeAsia100K [55–57], have been launched to correct the etiological and ethnic bias in previous genomic research and collect OMIC data from patients and healthy individuals across the globe. Notably, initial findings from the deep whole-genome sequencing of 494 hCCs were recently published by the Chinese Liver Cancer Atlas (CLCA) project and included a valuable comparison of potential driver mutations with TCGA and PCAWG HCC cohorts [58]. The integration of large-scale genomic, transcriptomic, and proteomic data from multiple sources will certainly promote a better understanding of HCC and accelerate the translation of basic research into patient-tailored clinical applications, and we are therefore looking forward to the public release of protein and mRNA expression datasets from such large HCC cohorts.

In conclusion, GEMMs have significantly advanced cancer research by allowing scientists to study the genetic and molecular mechanisms underlying cancer in a well-defined, controlled, and living organism. GEMMs have contributed to the identification of key oncogenes and tumor suppressor genes, provided insights into cancer metastasis and resistance, and facilitated the development and preclinical evaluation of targeted therapies. Switchable AP1 loss- and gain-of-function models, such as the c-Jun~Fra-2^hep mouse, have still not been exploited to their full potential. Together with the unlimited prospects of large-scale profiling, GEMMs will certainly yield further useful insights to HCC complexity and accelerate the translation of basic research into patient-tailored clinical applications.

Ethical approval

All animal experiments were performed in accordance with institutional, national, and European guidelines for animals used in biomedical research and approved by the Spanish National Cancer Research Centre (CNIO) – Instituto de Salud Carlos III Ethics Committee for Research and Animal Welfare and the Austrian Federal Ministry of Education, Science and Research (approved project Wagner E. 171/18).

Funding Statement

The Wagner laboratory is supported by the European Research Council [AdG 2016-741888-CSI-Fun], a H2020 – MSCA grant [ITN 2019-859860-CANCERPREV] and the Medical University of Vienna.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

RNAseq datasets used to generate figure panels A-C were generated and analyzed as described in [24] and are deposited in the Gene Expression Omnibus (GEO) archive https://www.ncbi.nlm.nih.gov/geo/ as series: GSE269844 [41], available at the time of publication, and GSE261005 [36].