Concurrent Disruption of the Ras/MAPK and NF-κB Pathways Induces Circadian Deregulation and Hepatocarcinogenesis

Hepatocellular carcinoma (HCC) is a rising cause of cancer mortality worldwide (1). Large-scale genomic and molecular analyses of HCC specimens suggest the involvement of a heterogenous collection of tumor-initiating oncogenic pathways (2). Consistent with this heterogeneity, no effective targeted therapies have been established; the multikinase inhibitors that block Raf and VEGFR signaling offer only limited therapeutic benefit to HCC patients (3, 4), underscoring the complexity of liver tumorigenesis (5).

The Ras protooncogenes constitute the most frequently mutated gene family in human cancer (6–8). RASopathies represent a group of developmental disorders driven by aberrant Ras/MAPK signaling (8) that are frequently associated with increased risk of cancer. Shp2 is a cytoplasmic tyrosine phosphatase that promotes signaling from receptor tyrosine kinases to the Ras/Erk pathway (9). Dominant active mutations in PTPN11/Shp2 augment Erk signaling in patients with leukemias and solid tumors (9, 10). Concurrent inhibition of Shp2 and other components in the Ras/Erk pathway was shown to effectively suppress tumor cell proliferation (11, 12). The NF-κB family consists of five DNA-binding proteins in homo- or heterodimers (13). The inactive dimers are retained in the cytoplasm by binding to the IκB inhibitor proteins. Ikkβ is a critical component of the IκB kinase (IKK) complex, which mediates phosphorylation and degradation of IκB, leading to nuclear translocation and activation of NF-κB. Overactivation of NF-κB leads to pleiotropic oncogenic processes (13), and disrupting the pathway has been shown to suppress tumor progression in a mouse HCC model (14). Indeed, NF-κB is constitutively active in many types of cancers (13).

However, ablating either Shp2 or Ikkβ in hepatocytes ironically aggravates HCC induced by a chemical carcinogen (15, 16). One possible explanation for the paradoxical antitumoral effect of the oncoproteins is forced activation of compensatory pathway(s) triggered by disruption of one critical oncogenic route (5). Elucidating the compensatory mechanisms will hopefully guide design of more efficacious HCC therapy by simultaneously blocking parallel pathways. Toward this goal, we generated a double-knockout (DKO) mouse line with both Shp2 and Ikkβ deleted in hepatocytes. Surprisingly, concurrent disruption of the Ras/MAPK and NF-κB pathways induced more severe liver tumorigenesis. The DKO mice displayed multiple liver injuries, chronic inflammation, severe cholestasis, changed metabolic pathways, deregulated cell-cycle gene expression, and circadian clock dysregulation. These mutually enhancing disorders and hepatic damages drove spontaneous HCC development in the liver. Thus, this study revealed a previously unrecognized tumor-inhibiting role of baseline Ras/MAPK and NF-κB signaling, associated with hepatic protection and physiologic homeostasis. These two pathways are actually dispensable for hepatocyte proliferation and even malignant transformation in the liver.

C57BL/6 mice were maintained in 12-hour light/dark cycles with standard chow and water ad libitum. Mice with hepatocyte-specific Shp2 deletion (Alb-Cre:Shp2^f/f:SKO) were previously generated by G.S.F's group. Ikkβ^f/f mice were generated by M.K.'s group at UCSD. These lines were crossed to generate double-knockout mice (DKO, Alb-Cre:Ikkβ^f/f:Shp2^f/f). Mice lacking the Alb-Cre transgene were used as wild-type controls (WT). All mouse husbandry and procedures were performed in compliance with protocols approved by the UCSD Institutional Animal Care and Use Committee (protocol S09108).

Genotyping was performed by PCR analysis of genomic DNA extracted from tail snips using primers listed in Supplementary File 1.

Mouse livers and other tissue samples were collected during the same four-hour period each day [Zeitgeiber Time (ZT) 7–11, with vivarium lights-on time designated ZT 0]. For tumor initiation, mice were intraperitoneally injected with 25 mg/kg diethylnitrosamine (DEN; N0258, Sigma-Aldrich) at postnatal day 15.

Liver samples harvested from mice were fixed in Z-Fix (175, Anatech) and paraffin-embedded. Hematoxylin and eosin (H&E) staining was performed by the UCSD Histology Core. Hepatocyte proliferation was assessed by costaining for Ki67 (eBioscience, 47569880; RRID:AB_2688065) and HNF4α (Santa Cruz Biotechnology, 8987; RRID:AB_2116913). Detection of macrophages was performed by staining paraffin sections for F4/80 (eBioscience, 14-4801-81; RRID:AB_467557). Picro-Sirius Red staining for detection of fibrosis was performed according to kit instructions (American MasterTech, STPSR).

Hepatocytes were isolated by the two-step perfusion method. Briefly, mice were humanely euthanized and livers were immediately perfused with perfusion buffer (HBSS without calcium and magnesium, HEPES, EGTA), then with digestion buffer (HBSS with calcium and magnesium, HEPES, collagenase H; 11074059001, Sigma). Digested livers were suspended in cold phosphate-buffered saline (PBS), then centrifuged three times at 50 × g for three minutes to obtain a pure hepatocyte fraction. For RNA-seq analysis, isolated hepatocytes were immediately flash-frozen. For growth factor stimulation experiments and in vitro circadian experiments, freshly isolated hepatocytes were cultured as follows.

Immediately after isolation, hepatocytes were plated on collagen-coated (C8919, Sigma) plates in Williams' Media E (W1878, Sigma) supplemented with 10% FBS and penicillin/streptomycin (15140, Gibco). For signaling experiments, hepatocytes were serum-starved and then stimulated with mouse recombinant IL6 (BioLegend, 575702), TNFα (BioLegend, 575202), or mouse recombinant HGF (Sigma, SRP3300) before harvesting in RIPA. For circadian time-course experiments, primary hepatocytes were allowed to adhere and recover overnight in complete media. In order to synchronize the cultures, they were serum-starved for 12 hours and then serum shocked with 50% FBS for two hours. Synchronized hepatocytes were harvested in TRIzol reagent at four-hour intervals over a course of 24 hours. Time points were designated by the ZT relative to the start of serum shock (ZT0).

Spleens and liver-draining lymph nodes were collected from humanely euthanized mice between 8 and 12 weeks of age and homogenized through a 70-μm cell strainer and then further suspended using a 27-gauge needle. Liver nonparenchymal cells were isolated by two-step perfusion as described in “primary hepatocyte isolation and culture.” Hepatocytes were removed from the cell suspension by sequential centrifugation at 50 × g; the supernatants containing NPCs were centrifuged at 300 × g for five minutes to pellet the cells. Red blood cells were removed from NPCs, splenocytes, and lymph node suspensions by incubating in ACK lysis buffer for five minutes at room temperature. Cleared cells were pelleted by centrifugation at 500 × g for five minutes and then washed twice in PBS.

Surface marker staining was performed as follows: First, viability staining was performed for 20 minutes on ice in PBS with Tonbo Ghost Dye UV450, with all samples protected from light through this and subsequent steps. Conjugated antibodies were added and cells were incubated 15 to 20 minutes on ice. See Supplementary File S1 for antibody panels used. Cells were washed three times in staining buffer (PBS with 5% FBS and 0.1% sodium azide).

Intracellular staining was performed using the Foxp3 Fix/Perm solution (BioLegend) according to the manufacturer's instructions.

Lysates from whole liver or isolated hepatocytes were prepared in RIPA (10 mmol/L Tris-HCl, 1 mmol/L EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% SDS, 140 mmol/L NaCl) with protease inhibitors and phosphatase inhibitors. SDS-PAGE and immunoblotting were performed according to standard procedures using the antibodies listed in Supplementary File S1.

Serum was collected from two-month-old male mice sacrificed between ZT 7 and 11 and immediately flash-frozen. Serum AST and ALT were measured using commercially available kits (Sekisui).

Serum bile acids were measured using a commercially available kit (Abcam, ab239702) according to instructions. Liver total bile acids were extracted from 50 mg liver and homogenized in 1 mL of 70% ethanol at room temperature. The suspension was incubated for three hours at 50°C and then clarified by centrifugation at 10,000 rpm for 10 minutes at room temperature. The supernatant was used for analysis of bile acid levels.

Whole-liver samples or freshly isolated hepatocytes were collected during the same time period each day (ZT 7–11) and immediately frozen. RNA was extracted from samples of whole liver using the RNeasy Microarray Tissue Mini Kit (73304, Qiagen) or from isolated hepatocytes using RNeasy Plus Micro Kit (74034, Qiagen). Libraries were prepared using Illumina TruSeq v2 kit according to the manufacturer's instructions. Sequencing was performed at UCSD Genomics Core on the HiSeq4000 platform. Raw data quality control was performed using fastqc. Sequenced reads were mapped to the mm9 reference genome using STAR, and gene differential expression analysis was performed using Cuffdiff with fragments per kilobase million (FPKM) values. To compare different expression levels of genes between KO and WT, we used the average FPKM in each group (n = 3). Significantly differentially expressed genes were determined with |log2FC| > 0.5 and q-value < 0.05. Changes unique to DKO were found by overlapping the significantly differentially expressed genes in DKO versus WT, DKO versus IKO, and DKO versus SKO comparisons; for a gene to be considered differentially expressed in DKO, it must have |log2FC| > 0.5 and q-value < 0.05 compared with each other genotype.

For gene-expression heatmaps, we first compared gene expression among the four groups (WT, SKO, IKO, and DKO) with function scale () in RStudio, where we scaled the genes in range (−1 to 1), as indicated in each color key. Each column in the heatmap represents average expression level of the three biological replicates within each genotype. Heatmaps were made with R package heatmap.2. For pathway analysis, we first separated up- and downregulated genes, then analyzed each list using the gene set enrichment analysis (GSEA) software's preranked tool. The top dysregulated gene sets for each genotype comparison, or for the DKO-unique dysregulated genes, were ranked in bar plots in RStudio, with package ggplot. The abbreviated pathway names are listed in Supplementary File S1 with respective original names. Enrichment plots for immune- and metabolism-related pathways were selected from the top 20 positively and negatively enriched pathways in the DKO versus WT comparison, respectively. See Supplementary File S2 for the full pathway enrichment results for all whole-liver and hepatocyte comparisons between genotypes.

The raw counts of TCGA liver hepatocellular carcinoma (LIHC) patient transcriptomic data set were downloaded from the Genomic Data Commons Data Portal (https://portal.gdc.cancer.gov/). This data set includes 371 HCC patient samples and 50 adjacent normal liver tissues. The R package DESeq2 was used to normalize raw counts across all samples. Next, we divided all HCC samples into three groups: (i) circadian clock-normal, (ii) circadian clock-mid, (iii) circadian clock-dysregulated, based on whether circadian clock genes were dysregulated or not in these patient samples compared with nontumor tissues. In detail, we checked expression of seven core clock genes, including CLOCK, ARNTL, PER1, PER2, PER3, CRY1, and CRY2. For each gene, the mean and standard deviation (SD) of normalized values in nontumor tissues were calculated. We set (mean − SD) as lower bound and (mean + SD) as upper bound, and then we defined an HCC patient with this circadian clock gene considered dysregulated if its normalized value was higher than the upper bound or lower than the lower bound. We repeated this process for all seven circadian clock genes. Next, we assigned patients with at most two dysregulated circadian clock genes into “normal” group, patients with at least five dysregulated circadian clock genes into “dysregulated” group, and others into “mid” group. With subclustering of all HCC patients, we first performed differential expression analysis between “normal” and “dysregulated” groups using DESeq2. The genes with adjusted P < 0.05 were defined as differentially expressed. We specifically checked components from the Ras/MAPK signaling pathway and the NF-κB pathway and then made volcano plots with differentially expressed genes in pathways above. Next, in order to further verify whether activities of these two pathways were inhibited in the dysregulated group, we collected downstream targets of the Ras/MAPK pathway as well as the NF-κB pathway from the Msigdb gene sets HALLMARK_TNFA_SIGNALING_VIA_NFKB and BILD_HRAS_ONCOGENIC_SIGNATURE, respectively. The pathway activity was defined as the sum of rank-transformed normalized values of all genes in the corresponding gene set. The boxplot was made using the Wilcoxon rank sum test.

Survival analysis was performed using OncoLnc (http://www.oncolnc.org/), with “high” and “low” expression defined as the upper and lower quartiles of expression for each gene. For expression levels of individual clock genes linked to altered survival, the raw counts in tumor and nontumor liver samples from the TCGA cohort were normalized using DESeq2, as described above. To assess the association of individual clock gene disruption with broader circadian deregulation within tumors, patients with upper- or lower-quartile expression of selected individual clock genes were divided into the “normal,” “mid,” and “dysregulated” groups, as previously defined.

RNA from whole liver or from freshly isolated hepatocytes was isolated using a standard phenol-chloroform extraction with TRIzol reagent according to the manufacturer's protocol. cDNA was prepared using the High-Capacity cDNA Reverse Transcription Kit (Life Technologies, 4368813). Quantitative reverse transcription was performed using PowerUP SYBR Green PCR Master Mix (Invitrogen, A25743) and the primers listed below on the Stratagene MX3005P. Ct values for each transcript were normalized to β-actin, and fold change was determined by the ΔΔCt method compared with control. Primer sequences are listed in Supplementary File S1.

Multiple comparisons were performed using one-way ANOVA with Tukey post-hoc analysis in GraphPad Prism. P < 0.05 was considered statistically significant (*, P < 0.05; **, P < 0.01; ***, P < 0.001). Data are presented as boxplots with center line representing mean and whiskers indicating range.

The transcriptomic data generated in this publication have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE158381 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE158381).

An original mouse line (Alb-Cre;Ikkβ^f/f;Shp2^f/f) was generated in this study; please contact the corresponding author for resource sharing.

To interrogate pathway cross-talk, we crossed Shp2^f/f:Albumin-Cre mice (17) with Ikkβ^f/f mice (15), to generate a DKO mouse line lacking both Shp2 and Ikkβ in hepatocytes (Supplementary Fig. S1A and S1B). The DKO mice were born at the expected Mendelian frequency with no obvious health problems at young ages, similar to single Shp2 (SKO) or Ikkβ (IKO) hepatocyte-specific knockouts. We first examined the susceptibility of all three mutant mouse lines to DEN-induced carcinogenesis, which entails multiple mutations frequently detected in human HCCs (18). Because both the Ras/MAPK and NF-κB pathways constitute major signaling nodes controlling cellular growth and survival, we speculated that dual inhibition of these pathways might counteract the compensatory proliferation and/or malignant transformation previously reported in either single knockout, as is the case for combined deletion of Shp2 and Stat3 (16). We compared liver tumor loads in WT, SKO, IKO, and DKO mice at five and eight months after DEN injection (Fig. 1A and B; Supplementary Fig. S1C and S1D). The DKO mice unexpectedly exhibited much higher tumor burdens than IKO and SKO and showed tumors at the liver surface as early as five months after DEN injection, when only a few small macroscopic nodules and microscopic lesions were detected in some IKO and SKO mice (Fig. 1A–C; Supplementary Fig. S1D). Enhanced HCC development in DKO mice became even more striking at eight months after DEN (Fig. 1A and B; Supplementary Fig. S1C). Histopathologic examination of liver sections identified the tumors as HCC (Fig. 1C; Supplementary Fig. S1E). These results indicate that, contrary to our predictions, simultaneous disruption of Ras/MAPK and NF-κB signaling further aggravated chemical carcinogenesis.

Even more surprising was the detection of spontaneous HCC development in DKO mice, without any exogenous tumor-promoting treatment, which was not observed in either IKO or SKO livers (Fig. 1D and E; Supplementary Fig. S1F and S1G). When examining the mouse colonies without DEN injection, we observed spontaneous liver tumors in DKO mice at nine months of age, a time at which no tumor nodules were detected in WT, IKO, or SKO mice (Fig. 1D and E; Supplementary Fig. S1G). At 12 months, 75% of DKO mice showed large liver tumors, whereas only 29% of SKO and no IKO mice showed small benign hepatocellular adenomas (HCA; Fig. 1D and E; Supplementary Fig. S1H). The DKO livers exhibited substantial cytological atypia, and the spontaneously developed tumors in DKO liver exhibited typical HCC morphology (Fig. 1F; Supplementary Fig. S1G). These results indicate, ironically, that concurrent inactivation of two prooncogenic pathways, Ras/MAPK and NF-κB, drives carcinogenesis in the liver.

To dissect the factors underlying spontaneous liver tumorigenesis, we examined the liver phenotypes of young, non–DEN-treated mice of the four genotypes prior to tumor emergence. At two months of age, the DKO livers already displayed irregular lobe edges and visible crevices (Fig. 2A). The IKO livers had no gross structural or developmental defects and the SKO livers exhibited mild fibrosis around the portal triads at the early stage; these fibrotic and periportal changes were dramatically enhanced in DKO mice (Fig. 2B and C). Consistent with the anatomic lesions, qRT-PCR detected progressively increased levels of fibrosis-related Col1α1 and Tgfβ mRNAs in IKO, SKO, and DKO livers at two months (Fig. 2D); however, additional molecular and pathologic analyses are required to characterize the fibrosis phenotype in DKO livers. Serum AST and ALT levels were slightly elevated in single KO and significantly higher in double KO mice (Fig. 2E). SKO mice displayed modestly increased serum and liver bile acids (BA; Fig. 2F and G), reinforcing a role for Shp2 in control of BA biosynthesis (19). Although deleting Ikkβ alone did not alter BA levels, both the intrahepatic and serum BA amounts were markedly elevated in DKO mice (Fig. 2F and G), indicative of severe cholestasis induced by synergistic effect of dual gene deletion.

Another notable phenotype in the DKO mice was enlargement of spleen and liver-draining lymph nodes as early as two months of age (Supplementary Fig. S2A–S2C). We performed flow-cytometric analysis of the immune cell subpopulations in the liver, spleen, and lymph nodes (Supplementary Fig. S2D–S2F). We detected lowered CD4/CD8 cell ratios and NKT percentages in SKO and DKO livers, relative to WT and IKO livers. Tregs were elevated in SKO but not in DKO liver (Supplementary Fig. S2F). However, we did not observe significant or biased changes in the spleen of DKO mice, with only lower dendritic cells (DC) in IKO and monocytes in SKO mice (Supplementary Fig. S2D). Similar trends were detected in the lymph nodes, with slightly higher ratios of B and NKT cells in DKO mice (Supplementary Fig. S2E). Thus, the enlarged spleen was likely related to portal hypertension due to liver injuries rather than a specific immune response.

To decipher the molecular mechanisms underlying the histopathologic changes, we performed RNA-seq analysis of the whole liver from two-month-old mice without DEN treatment. We compared the WT liver transcriptome with those of IKO, SKO, or DKO livers, and performed GSEA for differentially expressed genes (Fig. 3A and B; Supplementary Fig. S3A–S3C). With the complete results presented in Supplementary File S2, the analysis of canonical pathways (CP) showed relatively fewer significant transcriptomic changes in the IKO/WT than the SKO/WT comparison (Supplementary Fig. S3B), consistent with the severity of respective liver damages and tumor loads. Shared changes in both IKO and SKO livers included cytokine and interferon signaling and innate immune systems, indicating hepatic injuries and inflammatory responses (Supplementary Fig. S3B and S3C). The DKO liver had most of the changes detected in IKO and SKO; however, the DKO/WT comparison also revealed more prominent changes in cell-cycle and metabolic pathways (Fig. 3B; Supplementary Fig. S3C). Indeed, the DKO liver was particularly distinguished by upregulation in cell cycle, cell division, and chromosome segregation pathways. Homeostatic processes and cytokine responses were altered in DKO compared with WT, consistent with the trends seen in the single knockouts (Fig. 3B; Supplementary Fig. S3C).

We then focused on changes unique to the DKO liver transcriptome only, by identifying genes that were differentially expressed in the DKO livers compared with all other genotypes, and then by determining to which functional gene sets the 258 DKO-unique upregulated genes belonged (Fig. 3C). These upregulated genes were involved in extracellular matrix deposition and various immune-related pathways such as myeloid and leukocyte activation (Fig. 3C and D). In particular, mRNAs for the macrophage-recruiting cytokine Ccl2 and its cognate receptor Ccr2 were significantly upregulated in the DKO liver tissue only, as detected by qRT-PCR (Fig. 3E). Additionally, expression of the monocyte chemoattractant Ccl8 showed a trend toward upregulation in DKO (Fig. 3E). Immunostaining for the macrophage marker F4/80 also demonstrated increased macrophage presence in DKO livers (Supplementary Fig. S4A).

Also noted in the DKO liver were altered metabolic pathways (Fig. 3B and F). By probing the gene set segregation of DKO-specific downregulated transcripts, we found that perturbed metabolic gene sets included metabolism of lipids and amino acids, transport of metabolic components, and oxidative processes (Fig. 3F and G; Supplementary Fig. S4B). Heatmaps of selected metabolic pathways revealed modestly altered patterns of gene expression in each single knockout, with the DKO most profoundly perturbed compared with WT (Supplementary Fig. S4C). qRT-PCR validated expression changes in selected metabolic genes identified by the RNA-seq data, especially the downregulated expression of the fatty acid metabolic genes Aacs and Fabp in DKO liver (Fig. 3H). Together, these results suggest that removal of both Ikkβ and Shp2 caused more severe hepatic damages, metabolic changes, chronic inflammation, and immune suppression, showing additive or synergistic effects of dual gene deletion. These changes collectively contribute to an apparently tumor-promoting niche in DKO livers.

The above-described hepatic disorders vividly illustrate the drastically augmented susceptibility to chemical carcinogenesis in the DKO liver. However, these pathogenic alterations are apparently not sufficient to drive spontaneous HCC development. To identify hepatocyte-intrinsic oncogenic signals, we isolated fresh hepatocytes from untreated two-month-old mice of each genotype and performed RNA-seq analysis. This approach allowed us to parse out hepatocyte-autonomous changes from microenvironmental factors; the results of these comparative pathway analyses are shown in Supplementary File S2. Comparison of the different transcriptomes (Supplementary Fig. S5A and S5B) revealed higher ROS levels, disordered metabolic pathways, and reduced cell proliferation in IKO hepatocytes, whereas SKO hepatocytes displayed changes in innate immunity and interferon response, as well as epigenetic regulators. Notably, comparing DKO hepatocytes with WT controls revealed markedly increased expression of genes involved in cell mitosis, chromosome segregation, and antigen processing (Fig. 4A).

We then compared the transcriptome of DKO hepatocytes against the WT, IKO, and SKO groups to identify changes induced specifically by dual Ikkβ and Shp2 removal (Fig. 4B; Supplementary Fig. S5C; Supplementary File S3). This analysis revealed unique upregulation of 148 genes only in DKO hepatocytes, a large proportion of which were integral promitotic genes or part of cell-cycle–related pathways (Fig. 4B, Supplementary Fig. S5C; Supplementary File S3). These data suggest that a distinguishing feature of DKO hepatocytes is, surprisingly, enhanced cell-cycle progression and cell division, despite concurrent disruption of two proliferative pathways. Consistent with the RNA-seq data, immunostaining of liver sections from two-month-old mice detected significantly increased hepatocyte proliferation in DKO livers using Ki67 and HNF4α antibodies (Fig. 4C). qRT-PCR analysis showed marked upregulation of transcripts for multiple growth factors, including Hgf, IL6, TNFα, PDGF-B, and PDGF-D, which are known to spur compensatory hepatocyte proliferation in damaged livers (Fig. 4D). As many of these factors mediate liver regeneration via pathways expected to be impaired in the DKO model, we isolated hepatocytes from WT, IKO, SKO, and DKO livers and examined the activation status of Ras/MAPK and NF-κB pathway components in response to several growth factors. Removal of Shp2 indeed blunted Erk activation by HGF, and loss of Ikkβ suppressed IκBα and p65/RelA phosphorylation following TNF stimulation (Fig. 4E and F). Following IL6 stimulation, p-Erk and p-Akt levels were severely reduced in SKO and DKO hepatocytes, but Stat3 phosphorylation increased in SKO and DKO (Fig. 4G). Therefore, both the Ras/MAPK and NF-κB pathways were indeed disrupted in hepatocytes deficient for Shp2 and Ikkβ. Furthermore, the expression of oncofetal genes Afp and Igf2 was upregulated in DKO livers in young mice at the preneoplastic stage, before detection of macroscopic tumors (Fig. 4H). These results argue that the Ras/MAPK and NF-κB pathways are less critical for driving cell division than previously thought, and are actually dispensable for hepatocyte proliferation and transformation in the liver.

Besides the upregulated transcripts, we also examined the functions of genes downregulated only in DKO hepatocytes. In contrast to the enhanced proliferation of DKO hepatocytes, IKO and SKO hepatocytes shared downregulation of cell division processes (Supplementary Fig. S5B). In comparison with WT controls, DKO hepatocytes demonstrated the metabolic changes already detected in whole liver (Fig. 5A). Intriguingly, despite being freshly isolated during the same time interval as all other genotypes (ZT 7–11), DKO hepatocytes exhibited changes to circadian pathways (Fig. 5A). To confirm that these processes were unique to the DKO hepatocytes, we interrogated the 90 genes that were downregulated only in DKO hepatocytes, as compared with WT, IKO, and SKO (Fig. 5B; Supplementary File S3). Interestingly, the most significant gene set changes were those involved in control of circadian clock (Fig. 5B). Specifically, we observed dramatically reduced expression of 13 circadian genes in DKO hepatocytes, including inhibitory components Bhlhe40 (Dec1) and Bhlhe41 (Dec2); transcription factor Klf10; core clock inhibitors Cry2, Per2, and Per3; and the clock-implicated nuclear receptor Rorc (Fig. 5C). Among them, the majority (10/13) were transcription factors that are negative or positive clock regulators. In agreement with the RNA-seq data, qRT-PCR analysis detected very low levels of mRNAs for Per2, Per3, Cry2, and Dbp in primary hepatocytes isolated from two-month-old DKO mice, as well as altered levels of the clock effectors Bmal1 and Clock (Fig. 5D). Of note, the transcripts encoding negative regulatory genes were expressed at similar or even higher levels in IKO and SKO hepatocytes than the WT (Fig. 5C), suggesting that their drastic inhibition in DKO hepatocytes was unlikely due to a simple additive effect of the two gene deletions. Rather, it appears to be a specific defect in hepatocytes induced only by combined disruption of the two pathways. To further examine the effect of dual pathway blockade on the expression patterns of circadian genes, we monitored clock gene expression over a 24-hour period in serum-shock-synchronized primary hepatocytes isolated from WT and DKO mice. The expression of the negative clock regulatory genes Per2, Cry2, and Dec1 was consistently lower in DKO than in WT hepatocytes (Fig. 5E). Although Bmal expression was only slightly altered in DKO, Clock mRNA levels were markedly higher at most time points in DKO hepatocytes (Fig. 5E). Modest induction of circadian transcripts was observed in DKO hepatocytes in vitro, after serum synchronization, which displayed similar kinetics to WT. However, the amplitude of induction was generally lower in DKO hepatocytes. These results suggest that circadian deregulation is a cell-autonomous effect of Ikkβ and Shp2 deletion. Given that disruptions of circadian rhythm or deletion of clock regulators in murine liver were sufficient to induce HCC (20), clock dysfunction may be a key contributor to spontaneous HCC development in DKO mice (Fig. 5F).

To explore the relevance of the mouse data to human HCC, we analyzed 371 HCC patient samples and 50 adjacent normal liver tissues in the TCGA data set. Based on expression levels of circadian clock genes in tumor relative to nontumor tissues, we divided HCC samples into three groups: (i) circadian clock-normal, (ii) circadian clock-mid, (iii) circadian clock-dysregulated. Specifically, we checked expression of seven core clock genes, including Clock, Arntl, Per1, Per2, Per3, Cry1, and Cry2. Patients with at most two dysregulated circadian genes were assigned into “normal” group, patients with at least five dysregulated clock genes into “dysregulated” group, and the others into “mid” group. With subclustering of all HCC patients, we performed differential expression analysis between “normal” and “dysregulated” groups using DESeq2. The genes with adjusted P < 0.05 were defined as differentially expressed. We specifically checked components in the Ras/MAPK or NF-κB pathways and generated volcano plots with differentially expressed genes in the two pathways (Fig. 6A and B). Multiple constituents from both pathways exhibited altered expression in clock-dysregulated HCC patients (Fig. 6A and B). Next, to verify whether activities of these two pathways were inhibited in the “clock-dysregulated” group, we collected downstream targets of the Ras/MAPK and NF-κB pathways from Msigdb gene sets, respectively. The pathway activity was defined as the sum of rank-transformed normalized values of all genes in a corresponding gene set. These analyses revealed a correlation between dysregulated circadian genes with downregulation of both Ras/MAPK and NF-κB pathways in HCC patients (Fig. 6C and D).

To determine whether this DKO tumor model resembled clock-disrupted human HCC, we further assessed enrichment for gene sets that were previously reported to associate with low expression of a signature of clock genes in TCGA HCC data (Fig. 6E; ref. 21). The majority of these gene sets were also altered in DKO hepatocytes, with lower enrichment seen in IKO or SKO hepatocytes. We then determined whether individual circadian genes downregulated in the DKO mouse model were associated with HCC prognosis. Indeed, lower-quartile expression of Per1, Rorc, and Cry2 was significantly associated with poorer overall survival (Fig. 6F). We also examined the expression levels of these genes in the TCGA cohort and confirmed that expression of these clock genes was significantly lower in tumors than in nontumor liver tissues (Fig. 6G). Finally, we asked whether these individual circadian gene disruptions were associated with broader circadian deregulation in HCC. To address this, we compared the proportion of patients with normal or dysregulated clock gene sets, as defined for Fig. 6C and D, between the upper- and lower-quartile expression levels of individual clock genes. The patients with low expression of these genes in tumors tended to have disruptions to other clock genes, as indicated by a higher proportion of patients falling within the mid- or dysregulated groups (Fig. 6H). Thus, deregulated expression of clock genes was linked to disease progression and prognosis in HCC patients. Together, these results suggest that this novel mechanism of HCC development unveiled in our mouse model is very likely implicated in the pathogenesis of human liver cancer.

A large body of literature has documented critical roles of the Ras/MAPK and NF-κB pathways in driving carcinogenesis in various cell types and organs (6–8, 13). Accordingly, these two pathways are currently hot drug targets pursued for oncological treatment worldwide. However, this study showed that a compound mutant mouse line with both pathways disrupted in hepatocytes was hypersensitive to DEN, which was known to induce oncogenic mutations frequently detected in HCC patients (18). Consistent with a previous prediction (5), the augmented susceptibility to DEN was associated with more severe cholestasis, apparent fibrosis and chronic inflammation in the young DKO liver. Transcriptomic analysis revealed broad hepatic metabolic disorders, including changes in metabolism of fatty acids, amino acids, and BA. Although some changes were shared with single-knockout livers, the DKO mice exhibited the most severe perturbations, due to additive or synergistic effects of dual gene deletion.

The most unanticipated result in this study is the spontaneous hepatocarcinogenesis in DKO mice without any exogenous challenge. This finding is surprising because the severe liver disorders that aggravated DEN-induced tumors were likely not sufficient to be carcinogenic drivers. Transcriptomic analysis of isolated hepatocytes identified severe disruption of core circadian control gene expression as the most prominent feature in DKO hepatocytes at two months, prior to tumor appearance but concurrent with accelerated proliferation and expression of HCC marker genes. Of note, previous studies demonstrated that chronic jetlag in mice was sufficient to drive HCC development, and targeted deletion of clock regulators or effectors was also found to drive spontaneous HCC development in mice (20). Therefore, we propose here that clock disruption contributes to the spontaneous HCC development in DKO mice. Consistent with the mouse results, TCGA data analysis revealed a potentially similar mechanism of tumorigenesis involved in human HCC. In human liver cancer, as in our DKO mouse model, circadian dysfunction was coupled to enhanced proliferation and disruption of numerous metabolic processes (21). Epidemiologic links of chronic circadian disruption, such as shift work, to increased rates of human cancer have prompted the IARC to list circadian disruption as a probable carcinogen. Lower expression of several clock genes deregulated in the DKO model was associated with worse prognosis in human HCC. Together, these observations suggest a functional link between the circadian disruption, broad liver pathologies, and spontaneous tumor phenotype in our DKO model.

This study unexpectedly disclosed a physiologic function of the Ras/MAPK and NF-κB pathways working in concert to maintain circadian homeostasis at the ground state. A variety of negative regulatory clock TFs were consistently downregulated, which could contribute to the observed increased expression of Bmal1. As downregulation of multiple clock genes occurred only in the DKO hepatocytes, it is unlikely that the circadian deregulation was secondary to liver damages. Given that it was not observed in either IKO or SKO hepatocytes, the dramatic dysregulation in DKO hepatocytes was evidently not an additive effect of dual gene deletion; rather, it appears to be a systemic disorder that stems from disruption of an intricate signaling network constituted by two classic signaling pathways. This finding is consistent with the highly heterogeneous nature of HCC-associated oncogenic mutations (22), which frequently does not follow a clearly defined mutational signature.

Contrary to the previously defined proproliferative effects of the MAPK and NF-κB pathways, the DKO hepatocytes displayed dramatically increased proliferation and upregulated expression of cell-cycle genes. Tumor nodules were detected in DKO livers in nine months, displaying a phenotype much faster than that produced by deletion of clock genes (20). We propose that this accelerated tumor phenotype was driven by the mutually promoting effects of circadian dysregulation and multiple liver disorders, especially the strikingly elevated BA levels in the DKO liver (Fig. 2G). Circadian dysfunction has been shown to disrupt BA homeostasis (23), and aberrant BA production influenced hepatic expression of clock genes (24). Therefore, the circadian disruption and cholestasis-associated liver damages may constitute a destructive feedback loop spurring excessive compensatory hepatocyte proliferation and eventually driving tumorigenesis (Fig. 5F). The initial trigger leading to this pathogenic downward spiral was ironically the combined suppression of two well-characterized oncogenic pathways. Although circadian deregulation is apparently a contributor to the pathologic phenotype, we cannot exclude contributions from as-yet-unidentified pathways. For example, whereas canonical NF-κB signaling is suppressed in this model, we cannot exclude a possibility that Ikkβ deletion might influence noncanonical NF-κB signaling (25, 26), which remains to be explored.

Although each signaling pathway has received extensive analysis individually, how they intertwine is complex and unpredictable. While deleting both Shp2 and Pten in blood cells showed mutually neutralizing effects in myeloproliferation and leukemogenesis (27), simultaneous removal of Shp2 and Pten in hepatocytes accelerated NASH-HCC development (28). Although loss of either Shp2 or β-catenin promoted DEN-induced HCC (5), deleting both suppressed liver tumorigenesis (Chen and colleagues, unpublished data). This study has unveiled previously unrecognized roles of two central signaling pathways in sustaining circadian rhythm and physiologic homeostasis, providing a fresh view on their baseline gatekeeper function. Due to the limited efficacy of monotherapy that targets a single oncogenic pathway, one prevailing view in cancer therapy is to simultaneously block multiple pathways to optimize the antitumor effect. However, the data presented here raise a cautionary note on the concept of combinatorial targeting.

No disclosures were reported.

K.L. Hanley: Formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. Y. Liang: Data curation, formal analysis, investigation, and methodology. G. Wang: Data curation and formal analysis. X. Lin: Data curation, formal analysis, and visualization. M. Yang: Formal analysis, investigation, and methodology. M. Karin: Resources, writing–review and editing. W. Fu: Resources, methodology, writing–review and editing. G.-S. Feng: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing.

RNA sequencing was performed at the UCSD Genomics Core. Histologic consultation was provided by Dr. N. Varki from the UCSD Mouse Phenotyping Core. Research reported in this publication was supported by NIH award F31CA228420 to K.L. Hanley and NCI grants R01CA239629 and R01CA236074 to G.-S. Feng.

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