A TNFR2–hnRNPK Axis Promotes Primary Liver Cancer Development via Activation of YAP Signaling in Hepatic Progenitor Cells

Primary liver cancer (PLC) is the fifth most prevalent cancer and the second most lethal tumor type in the world (1, 2). Hepatic progenitor cells (HPC) possess self-renewal capability and differentiation potential (3). It has been well documented that malignant transformation of HPCs can drive liver tumorigenesis, and that HPCs represent a source of liver cancer (4–6). Specifically, HPCs are scarcely detectable under physiologic conditions, but have been found to abnormally proliferate under pathologic conditions such as inflammation and tissue damage (4). For example, chronic inflammation has been demonstrated as a causal factor of PLC (7–9). Long-term and chronic inflammation can lead to abnormal expansion and differentiation of HPCs, which will further undergo malignant transformation to become cancer stem cells (10). Yet, the molecular mechanism that links chronic inflammation to the expansion of HPCs remains elusive.

The Hippo–YAP signaling pathway is known to regulate stem cell homeostasis, tissue regeneration, and tumorigenesis (11). Yes-associated protein (YAP) is a major downstream transcriptional coactivator of the Hippo pathway (11, 12). Phosphorylation of YAP by the tumor suppressor kinase cascade MST1/2-LATS1/2 results in its cytoplasmic retention and subsequent proteolytic degradation (13, 14). Once dephosphorylated, YAP is able to enter the nucleus, where it binds with TEAD family of transcriptional factors to regulate the expression of many target genes (15), usually leading to increased cell proliferation and decreased apoptosis (16). Hyperactivation of YAP has been frequently observed in a variety of tumors including PLC (12, 17). Moreover, genetic activation of YAP in mice causes liver overgrowth and expansion of HPCs, which eventually leads to liver cancer (17, 18). However, whether and how Hippo–YAP signaling (particularly in HPCs) responding to inflammatory stimuli to affect liver tumorigenesis remains poorly understood.

The TNFα is a proinflammatory cytokine important for initiation and development of PLC (19). It is known that TNFα induces either cell proliferation or programmed cell death via engagement with one of its two receptors: TNF receptor I (TNFR1) and TNF receptor II (TNFR2; ref. 20). Notably, a number of studies have evidenced differential roles of the two receptors during liver tumorigenesis (21–25). For instance, we found that TNFα promotes hepatocellular carcinogenesis via activation of HPCs, and that TNFR1 and TNFR2 exert opposite effects in the regulation of PLC (6, 25). Recently, TNFα biology has been further extended to crosstalk with the Hippo–YAP signaling in breast cancer (26), which is intriguing for us to hypothesize that TNFα as an inflammatory signal may trigger, via one of the two receptors, the activation of YAP in HPCs during PLC development.

Here, we demonstrate that TNFR2, but not TNFR1, in HPCs promotes primary liver tumorigenesis via malignant transformation of HPCs. TNFR2, but not TNFR1, is essential for TNFα-induced activation of YAP in HPCs. Mechanistically, TNFα–TNFR2 signaling promoted YAP activation through heterogeneous nuclear ribonuclear protein K (hnRNPK), which directly binds and cooperates with YAP on target gene loci genome-wide. Moreover, TNFR2, hnRNPK, and YAP were all found to be upregulated and positively correlated with one another in PLC. Hyperactivation of the TNFR2–hnRNPK–YAP signaling axis is strongly associated with poor prognosis of PLC including patient overall survival. These observations were further confirmed by single-cell RNA sequencing (scRNA-seq) of epithelial cells from patients with PLC. Thus, our study demonstrated an essential role for TNFR2 in HPCs-mediated PLC development, and further revealed that TNFR2, but not TNFR1, is required for TNFα-induced YAP activation, highlighting hnRNPK as a direct molecular link between TNFα–TNFR2 and YAP signaling.

Additional or detailed methods are described in the Supplementary Materials and Methods.

Male SD rats (10–12 weeks, 220–250g) were obtained from Shanghai Experimental Center, Chinese Science Academy (Shanghai, China). Tnfrsf1a^−/−and Tnfrsf1b^−/−rats were all established by Nanjing Xunqi Biotechnology Co. Ltd, which were used by CRISPR/Cas9 knockout technology. The rats were maintained at an animal facility under pathogen-free conditions. All animals' experiments were according to the animal protocols approved by the Shanghai Eastern Hepatobiliary Surgery Hospital Animal Care Committee. To induce PLC, 100 p.p.m. (95 μg/mL) DEN was added to the drinking water of rats for 16 weeks. Liver tumors were measured with electronic calipers and counted (for tumors with diameters ≥1 mm). Liver sections were preserved in 10% neutral-buffered formalin for histopathologic analysis and IHC assay, and the blood was collected and serum was isolated for biochemical analysis.

To isolate and purify rat HPCs, we removed fetal liver tissue from wild type, Tnfrsf1a^−/− and Tnfrsf1b^−/− rats at day 13.5 of gestation. Then fetal liver tissue was dissociated to obtain hepatic cell suspension after enzymatic digestion for 60 minutes at 37°C. Ep-CAM⁺ cells were isolated from hepatic cell suspension with immunomagnetic beads and seeded into ultra-low attachment dish in rat embryonic fibroblasts condition medium. Hepatic spheroids were formed and collected at day 6 and plated on type I collagen–coated dishes in rat embryonic fibroblasts condition medium. After 2 weeks, hepatic spheroids were treated with trypsin and collagenase and then replanted on type I collagen–coated dishes for another 2 weeks. Rat embryonic fibroblasts condition medium was changed every two days. After that, HPCs were harvested from collagen-coated dishes and cultured to maintain the stemness of cells on normal dishes. We regarded DMEM/F12 medium with 10% FBS, 1% penicillin/streptomycin as standard medium to conduct the further experiments.

PLC and adjacent tumor tissue microarray sections were prepared by Shanghai Outdo Biotech Co. Ltd. (Shanghai, China). This tissue array contains tissues from 77 paired fresh liver carcinoma and adjacent tumor tissue samples. For IHC, TMA sections were incubated with anti-TNFR2 antibody (1:500 dilution), anti-YAP antibody (1:100 dilution), or anti-hnRNPK antibody (1:100 dilution). IHC stains were scored by two independent pathologists who were blinded to the clinical characteristics of the patients. The scoring system was based on the intensity and extent of staining: staining intensity was classified as 0 (negative), 1 (weak), 2 (moderate), or 3 (strong); staining extent was dependent on the percentage of positive cells (examined in 200 cells) and was classified into0 (<5%), 1 (5% to 25%), 2 (26% to 50%), 3 (51% to 75%), or 4 (>75%). According to the staining intensity and staining extent scores, the IHC result was classified as 0 to 1, negative (−); 2 to 4, weakly positive (+); 5–8, moderately positive (++), and 9 to 12, strongly positive (+++).

The cohort in this study contains 77 patients from January 1997 to December 2007. All patients were randomly selected from patients with liver cancer who underwent hepatectomy in the Shanghai Eastern Hepatobiliary Surgery Hospital (Shanghai, China). All samples collected and used in this study were derived from patients after obtaining written informed consent that was approved by the Hospital Research Ethics Committees, and this study was conducted in accordance with the Declaration of Helsinki. None of the patients had preoperative treatment, and recurrence was confirmed by contrast-enhanced imaging studies or cholangiography according to standard guidelines for PLC. Overall survival (OS) was defined as the interval between surgery and death or the last observation taken. The data were censored at the last follow-up period for living patients.

Chromatin immunoprecipitation (ChIP) was performed basically followed the procedures as described previously (27). The remaining supernatants were transferred to new Eppendorf tubes and were incubated with either IgG or YAP antibodies (14074, Cell Signaling Technology) at 4°C overnight. The supernatants were combined as templates for follow-up qPCR analysis.

ChIP-sequencing (ChIP-seq) was performed on the basis of a previous protocol with minor modifications (28). Cells stably expressing Flag-tagged YAP and hnRNPK were subjected the same treatments as described above to get the cell pellets. Sequencing was performed on an Illumina HiSeq 2500 platform.

Cells were transiently transfected with Flag-hnRNPK construct for 48 hours before harvesting. Cells were lysed with denaturing buffer [20 mmol/L Tris·HCl (pH 8.0), 50 mmol/L NaCl, 0.5% Nonidet P-40, 0.5% deoxycholate, 1.0% SDS, and 1 mmol/L EDTA] on ice for 10 minutes, followed by boiling at 95°C for 5 minutes. The cell lysates were cooled on ice for another 5 minutes before incubating with anti-Flag (M2) beads for 4 hours at 4°C. Beads were washed two times with denaturing buffer and then with NETN buffer for another two times to renature the purified proteins. After that, Flag-hnRNPK beads were further incubated with either recombinant MBP-YAP protein (50–504 aa) or MBP for 4 hours before boiling in loading buffer. Samples were separated by SDS-PAGE, and processed for Western blotting using anti-YAP antibody or verified by Coomassie brilliant blue staining.

Sequencing depth of tumor samples was normalized by using CellRanger (version v3.1.0). Highly variable genes were detected according to average expression (between 0.05 and 3) and dispersion (above 0.5) of the genes, followed by data scaling (subtracting the average expression) and centering (divided by SD). Those variable genes were considered accounting for cell-to-cell differences, and were further used for PCA. The first 30 PCs were applied for t-distributed stochastic neighbor embedding (t-SNE) analysis according to the eigenvalues (Supplementary Fig. S6A). We used Seurat package (version 3.0) in R (version 3.5.3) to perform data filtering (both gene and cell), normalization, PCA, and t-SNE. All scRNA-seq data were submitted to the Gene Expression Omnibus (GEO) public database at NCBI (GEO: GSE166635).

We extracted transcriptomic data of cells from the expression profiles of all the single cells. We first selected variable genes across cells, based on criteria of average expression (between 0.05 and 3) and dispersion (above 0.5) of the genes. We annotated the cells based on known cell lineage–specific marker genes as T cells (CD4, CD3E, CD3D, CD3G, CD8A, CD8B), B cells (CD79A, SLAMF7, BLNK, FCRL5), TECs (PECAM1, ENG), CAFs (COL1A2, BGN, ACTA2), TAMs (CD14, CD163, C1QA), as well as HPCs (EPCAM, KRT19, CD24).

Cells defined as endothelia, fibroblast, and macrophage were used as reference to identify somatic copy number variations (CNV) with the R package infercnv (v0.8.2). We scored each cell for the extent of CNV signal, defined as the mean of squares of CNV values across the genome. Putative malignant cells were then defined as those with CNV signal above 0.05 and CNV correlation above 0.5.

We constructed single-cell trajectory of each tumor by using reversed graph embedding method implemented in R Monocle package (version 2.6.3; ref. 29). We created a Cell Data Set object for single-cells of each tumor with the parameter expression Family as negbinomial. To detect genes that could provide important information in shaping the trajectory, we conducted PCA and t-SNE (the first 10 PCs were used) based on the genes expressed in at least 10% of all the cells of each tumor, and further applied density peak clustering method to identify clusters in the t-SNE space. The derived top 1,000 genes were considered crucial for defining the progress of cells. The second step was dimensionality reduction and trajectory construction with the obtained genes. Reversed graph embedding technique was applied in this process, by projecting cells to a low dimensional space while simultaneously learning smooth tree-like manifold as well as assigning cells onto the manifold.

Statistical analysis was performed using SPSS software version 20.0 (SPSS) and Python 3.6. Data are presented as mean ± SEM. Differences were analyzed by the Student t test and one-way ANOVA. Tumor incidence (%) was analyzed by Fisher exact test. Kaplan–Meier method was used to calculate the survival rate and log-rank test for the different significance. Correlation between the expression of YAP signature and hnRNPK was calculated using Pearson correlation, and a linear model was built to fit the data and test significance and was plotted as a trendline with the 95% confidence intervals (95% CI). A P < 0.05 was considered statistically significant.

Data are available in a public, open access repository. The accession number for the raw data reported in this article have been deposited in the GEO under accession number GSE166635 (scRNA-seq data) and GSE166627 (RNA-seq data).

Our observations demonstrating the role of TNFR during DEN-induced PLC (25) led us to further explore the impact of TNFR on liver tumorigenesis. Consistent with our previous reports, depletion of TNFR2 had a dramatic inhibitory effect toward DEN-induced PLC in rats, whereas, in contrast, TNFR1 ablation obviously promoted PLC formation (Supplementary Fig. S1A and S1B). Moreover, the numbers of both hepatic progenitor cells (HPCs, OV6⁺) and tumor-associated fibroblasts (TAFs, α-SMA⁺) were markedly reduced in PLC tissues from Tnfrsf1b^−/− rats; yet no significant change was observed for these cells in tumors from Tnfrsf1a^−/− rats (Supplementary Fig. S1C and S1D). The markedly opposite roles of TNFR1 and TNFR2 in PLC development clearly suggest distinct mechanisms through which these receptors regulate PLC.

Given the well-established pathogenic role of HPCs in PLC (4–6, 21–25), and the above observations of TNFR2 deficiency-induced loss of HPCs during liver tumorigenesis, we next focused on how TNFR2 was linked to PLC development. To test the hypothesis that lacking of TNFR2 directly resulted in disturbed growth and differentiation of HPCs, we attempted to extract HPCs from fetal livers (13.5 days) of wild-type (WT), Tnfrsf1a^−/− or Tnfrsf1b^−/− rats (Fig. 1A and B). After purification of HPCs in vitro, around 1 × 10⁸ cells were then injected into DEN-treated male rats by tail vein, of which, 0.9% NaCl was used as mock control (Fig. 1C, red arrow). Of note, to dynamically evaluate the effects of exogenous HPCs on PLC development, we collected liver tissues at 4, 8, 12, and 16 weeks after DEN treatment to examine the presence of parental HPCs and corresponding liver tumor formation incidence (Fig. 1C, black arrow). Immunofluorescence analysis revealed that GFP-labeled all types of HPC were detected in liver tissues around one week after injection (Fig. 1D and E). Interestingly, presence of HPCs derived from WT and Tnfrsf1a^−/− rats were stably located in PLC tissues and Tnfrsf1a^−/− HPCs group gradually increased to 60% of PLC tissues at final point of analysis (Fig. 1D and E). However, HPCs derived from Tnfrsf1b^−/− rats were unstable and barely detectable at 12 weeks after injection (Fig. 1D and E), suggesting that TNFR2 deficiency reduced self-renewal ability and malignant transformation of HPCs.

In regard to tumor formation incidence, we found rats receiving WT HPCs developed severe PLC symptoms than those from mock group, which only displayed several small, white, cyst-like lesions, at 12 weeks after DEN exposure, indicating that the injected parental HPCs accelerate the PLC development (Fig. 1F and G). Moreover, rats receiving Tnfrsf1a^−/− HPCs further increased tumor formation incidence and tumor size than those receiving WT HPCs (Fig. 1F–H). In sharp contrast, rats injected with Tnfrsf1b^−/− HPCs did not develop PLC as severe as those observed in rats receiving WT or Tnfrsf1a^−/− HPCs (Fig. 1F–H). Quantification of GFP⁺ cells within all observed tumors from indicated groups revealed that more than 60% of tumors in WT or Tnfrsf1a^−/− HPCs receipt rats were originated from exogenously injected HPCs, whereas only less than 20% were GFP positive in rats receiving Tnfrsf1b^−/− HPCs (Fig. 1I). Meanwhile, we extracted WT and Tnfrsf1b^−/− HPCs from female rats pretreated for 8-week DEN (Supplementary Fig. S1E), and found that self-renewal ability of Tnfrsf1b^−/− HPCs were much lower than WT HPCs via sphere formation analysis (Supplementary Fig. S1F and S1G), further excluding the possibility that gender difference may affect the tumor-suppressive effect of TNFR2-deficiency in malignant evolution of HPCs. Taken together, these results indicate that TNFR2 is required for malignant transformation of HPCs during PLC development, and that HPCs lacking TNFR2 were disabled to mediate PLC.

To explore the molecular mechanisms through which TNFR2 mediates the malignant transformation of HPCs, we performed RNA-seq analysis to assess the global impacts of TNFR2 on HPCs' response to TNFα. Upon TNFα treatment, 519 genes were significantly upregulated in wild-type HPCs (Fig. 2A; Supplementary Fig. S2A). Surprisingly, 39% of these genes (203 of 519) were further upregulated in Tnfrsf1a^−/− HPCs treated with TNFα (Fig. 2A and C). However, these 203 genes were dramatically downregulated in TNFα-treated Tnfrsf1b^−/− HPCs relative to their expression in either TNFα-treated Tnfrsf1a^−/− HPCs or WT HPCs (Fig. 2A and C), further indicating an inhibitory role for deletion of TNFR2 in the regulation of these genes. Given the known regulatory role of the Hippo signaling pathway in liver tumorigenesis (13), we speculated that TNFR2 may control HPCs through Hippo signaling pathway. Supporting this notion, gene-set enrichment analysis (GSEA) revealed that knock out of TNFR2 in TNFα-treated HPCs significantly reduced the expression of a group of YAP target genes (Fig. 2B and C), whereas TNFα stimulation in either WT or deletion of TNFR1 HPCs efficiently increased the expression of these genes (Fig. 2C; Supplementary Fig. S2B and S2C).

To validate the above observations, we evaluated the effects of TNFR2 on YAP activation at both cellular and animal levels. First, IHC analysis showed that nuclear accumulation of YAP in HPCs (OV6⁺) was significantly decreased in tissues from DEN-treated Tnfrsf1b^−/− rats compared with those from DEN-treated WT and Tnfrsf1b^−/− rats (Fig. 2D). Consistently, in cell-based assays, TNFα stimulation dramatically reduced YAP phosphorylation (Supplementary Fig. S2D and S2E), and promoted YAP nuclear retention by either cell fractionation assay or indirect immunofluorescence assay (Fig. 2E and F). Nevertheless, activation of YAP was completely abolished in Tnfrsf1b^−/− HPCs (Fig. 2E and F), validating the essential role of TNFR2 in activating YAP signaling in response to TNFα. Importantly, reconstitution of TNFR2 expression in Tnfrsf1b^−/− HPCs efficiently restored TNFα-induced YAP activation (Fig. 2G), increased its nuclear retention (Fig. 2H), as well as elevated CTGF and CYR61 mRNA transcription (Fig. 2I). Collectively, these results clearly indicate the essential role of TNFR2 in TNFα-induced YAP activation in HPCs.

Hippo signaling limits YAP activation via a cascade of phosphorylation events in response to stimuli. As TNFR2 is required for TNFα-induced YAP activation, we then examined the involvement of Hippo signaling in this process via Western blotting analysis of key phosphorylation events (Supplementary Fig. S2D and S2E). YAP activation, exhibited by a dramatical drop of the ratio of phosphorylated YAP to YAP, in WT and Tnfrsf1a^−/− HPCs, but not Tnfrsf1b^−/− HPCs, was reproducibly detected after TNFα treatment (Supplementary Fig. S2D and S2E). However, we failed to noticeably observe increased phosphorylation of MST1 and MOB1 of Hippo pathway (Supplementary Fig. S2D and S2E), suggesting that TNFα–TNFR2 regulation of YAP via a Hippo-independent manner.

To dissect the molecular mechanism through which TNFR2 mediates YAP activation in HPCs, we carried out immunoprecipitation-mass spectrometry (IP-MS) by using YAP-specific antibody, and detected 156 potential YAP-interacting proteins in HPCs (Fig. 3A). Subsequent integrative analysis by combining MS data, RNA-seq data (Fig. 2A), and publicly available liver cancer database from The Cancer Genome Atlas (TCGA; ref. 30) identified 6 high-confidence candidates (BSG, hnRNPH, hnRNPK, PNP, SORD, SMPD3; Fig. 3A). Among them, hnRNP proteins have been closely associated with TNFα signaling (31, 32). Therefore, we reasoned that TNFR2 mediates YAP signaling possibly through hnRNP family members. To further identify the specific hnRNP family member involved in TNFR2-mediating signaling, we then examined their expression in response to TNFα, and found mRNA levels of hnRNPA and hnRNPK were significantly increased (Supplementary Fig. S3A). Moreover, hnRNPK protein level was also robustly increased upon TNFα treatment (Supplementary Fig. S3B), strongly hinting its involvement in TNFα–TNFR2–YAP signaling axis.

To verify the interplay of hnRNPK–YAP in participating TNFα–TNFR2 signaling, we performed coimmunoprecipitation experiments and confirmed interaction of YAP with hnRNPK (Fig. 3B), and further revealed that TNFα stimulation enhanced such interaction (Fig. 3B and C). Furthermore, MBP-pull-down assay using purified proteins indicated that YAP directly interacts with hnRNPK (Fig. 3D). Subsequent domain-mapping analysis revealed that the N-terminal DNA-binding domain (amino acids 1–239) of hnRNPK interacts specifically with the C-terminal region of YAP (amino acids 291–504; Fig. 3E; Supplementary Fig. S3C). Further IHC analysis revealed strong colocalization between YAP and hnRNPK at different time points in DEN-treated rats, confirming interactions between these two proteins in vivo (Supplementary Fig. S3D). Together, these findings revealed a novel YAP-hnRNPK interaction in response to TNFα.

Similarly, we next examined the regulatory effect of TNFR2 on hnRNPK–YAP interaction after TNFα stimulation. Consistent with the positive regulation of TNFR2 on YAP activation (Fig. 2), we also found that deficiency of TNFR2 robustly diminished TNFα-induced upregulation of hnRNPK (Supplementary Fig. S3B), its binding ability to YAP (Supplementary Fig. S3E), as well as the nuclear localization of both hnRNPK and YAP in HPCs (Fig. 3F). Besides, conspicuous nuclear co-localizations of YAP and hnRNPK were observed in DEN-induced tumor samples only from WT rats but not from Tnfrsf1b^−/− rats (Supplementary Fig. S3F). More importantly, though TNFR2-depleted HPCs displayed reduced self-renewal ability upon TNFα treatment, reintroduction of either hnPNPK or YAP efficiently rescued the sphere formation efficiencies in Tnfrsf1b^−/− HPCs (Fig. 3G and H; Supplementary Fig. S3G). Collectively, these findings clearly indicate the hnRNPK–YAP interaction functions downstream of TNFα–TNFR2 signaling.

Given that p38 and ERK signaling have been associated with hnRNPK regulation (33, 34) and the Hippo-independent YAP regulation (Supplementary Fig. S2D and S2E), we thus hypothesized that p38 and ERK may act upstream of hnRNPK/YAP interplay after TNFα stimulation. To corroborate the exact roles of p38 and ERK under TNFα-induced hnRNPK/YAP activation condition, we systemically measured the temporal changes of p38 and ERK activation (increased phosphorylation), YAP activation (decreased phosphorylation), and hnRNPK upregulation after TNFα treatment. Specifically, we collected samples at different time points upon TNFα stimulation (0, 5, 30, 60, and 180 minutes) and examined these events via Western blotting analysis (Supplementary Fig. S4). We first examined the dynamic changes of ERK phosphorylation but only observed a very minor increase of ERK phosphorylation (∼1.2-fold at 30 minutes; Supplementary Fig. S4A and S4B). Nevertheless, TNFα treatment could still robustly induce hnRNPK accumulation and YAP activation when coincubated with ERK1/2 inhibitor SCH772984 (Supplementary Fig. S4A and S4B). On the contrary, we found TNFα markedly activated p38 signaling (∼3 fold) as early as 30 minutes after treatment, whereas YAP activation was observed at a relative later stage (∼60 minutes), indicating that p38 activation, as well as hnRNPK upregulation, appeared earlier than YAP activation (Supplementary Fig. S4C and S4D). Importantly, treatment with p38 inhibitor (BMS-582949) not only abrogated TNFα-mediated p38 activation, but significantly suppressed hnRNPK accumulation and YAP activation (Supplementary Fig. S4C and S4D). Further immunofluorescence assay also demonstrated that treatment with p38 inhibitor, but not ERK1/2 inhibitor, strongly reversed the TNFα-stimulated YAP and hnRNPK nuclear localization (Supplementary Fig. S4E and S4F). Taken together, these results firmly suggest that p38 acts downstream of TNFα–TNFR2 to activate YAP signaling.

Next, we assessed the regulatory effect(s) of hnRNPK on TNFα-induced YAP activation. To this end, we first established the hnRNPK knockdown HPCs (Supplementary Fig. S4G). Then, we performed RNA-seq analysis and identified 1,853 upregulated and 1,743 downregulated genes, respectively, in hnRNPK-depleted TNFα-treated HPCs. GSEA revealed a significant negative enrichment of YAP target genes upon hnRNPK depletion (Fig. 4A). Further qPCR analysis confirmed that knockdown of hnRNPK significantly inhibited TNFα-induced transcription of YAP target genes CTGF and CYR61 (Fig. 4B). Consistently, TNFα-induced YAP nuclear localization was also dramatically suppressed in hnRNPK-depleted cells as detected by either indirect immunofluorescence assay (Fig. 4C) or nuclear/cytosol fractionation assay (Fig. 4D). Furthermore, hnRNPK knockdown also impaired the endogenous YAP–TEAD association (Supplementary Fig. S5A). Together, these results indicate that hnRNPK is required for TNFα-induced nuclear translocation and activation of YAP.

In addition to the cell-based in vitro assays, we further applied the tail vein injection liver tumor model as described previously to validate oncogenic roles of hnRNPK during liver tumorigenesis in vivo. To this end, around 1 × 10⁸ WT and hnRNPK-depleted HPCs were first injected into DEN-treated male rats (Fig. 4E, red arrow). The HPCs receiving rats were sacrificed at 12 and 16 weeks treated with DEN to collect samples for analysis of liver tumor incidence and presence of GFP⁺ HPCs (Fig. 4E, black arrow). In line with our notion that hnRNPK mediates the TNFα-stimulated liver tumorigenesis, we found rats receiving WT HPCs developed severe liver tumors as early as 12-week after injection, whereas rats injected hnRNPK-depleted HPCs barely formed tumors (Fig. 4F–H). Moreover, at 16-week after DEN induction, significantly more and larger tumors were quantified in WT receipt groups than hnRNPK-depleted or mock groups (Fig. 4F–H). Quantification of GFP⁺ cell population in tumors from indicated treatment also revealed a markedly higher percentage of parental HPCs in WT than in the hnRNPK-depleted group (Fig. 4I), suggesting hnRNPK is essential for malignant transformation of HPCs in vivo. Meanwhile, we also performed qPCR analysis to examine the mRNA expression of YAP target genes including CTGF and CYR61 in these tumors, and found they were robustly stimulated in WT but not hnRNPK-depleted group (Fig. 4J). Taken together, these results clearly demonstrate an oncogenic role for hnRNPK in TNFα/YAP-mediated PLC development.

After establishing the positive regulation of hnRNPK on YAP nuclear translocation and activation, we further explored the biological functions of nuclear hnRNPK–YAP interaction on target gene expression. To this end, we first examined whether hnRNPK still promotes YAP (5SA)-driven CTGF and CYR61 transcription as this mutant disrupted the Hippo-mediated inhibitory effects and dominantly localized in nucleus (Supplementary Fig. S5B). Surprisingly, depletion of hnRNPK dramatically diminished both endogenous and YAP (5SA)-driven CTGF and CYR61 transcription (Supplementary Fig. S5C), indicating that the hnRNPK cooperates with YAP in nucleus to regulate target gene expression. Next, we performed ChIP-seq assay to profile the chromatin binding signatures of YAP and hnRNPK in HPCs. Analysis of the distribution of YAP- or hnRNPK-binding sites relative to genes annotated in the human genome revealed that most peaks were located closely (<1 kb) to transcription start sites (Fig. 5A). Notably, among the 4,343 hnRNPK sites, 2,373 (47.5%) were also occupied by YAP, implying that hnRNPK may promote gene expression together with YAP (Fig. 5B). More strikingly, both YAP and hnRNPK were found to be enriched at the promoter regions of several key factors involved in liver tumorigenesis, such as ELK1 (35), HNF4a (36), STAT3 (23), and Smad2 (37), further supporting that YAP acts in complex with hnRNPK to coregulate PLC development (Fig. 5C). We also performed a de novo motif analysis to further reveal the coregulation of gene expression by YAP and hnRNPK. Interestingly, among the top five enriched motifs, four of them including ELK1, ELK4, HNF4a and STAT3 were shared by both proteins, whereas ELK1, ELK4, and HNF4a ranked as the 1#, 2#, and 3# enriched motifs, respectively (Fig. 5D). Taken together, these data revealed a genome-wide association between YAP and hnRNPK in HPCs.

We next performed ChIP-qPCR assays to validate the ChIP-seq profiles. First, we found both YAP and hnRNPK efficiently bind to promoters of YAP target gene CTGF and CYR61 as well as those identified by sequencing including ELK1, ELK4, and STAT3 (Fig. 5E and F; Supplementary Fig. S5D and S5E). Strikingly, overexpression of hnRNPK further robustly enhanced YAP binding at these gene promoters (Fig. 5E; Supplementary Fig. S5D), and vice versa (Fig. 5F; Supplementary Fig. S5E). Importantly, TNFα treatment significantly promoted transcription of ELK1, ELK4, and STAT3 in WT but not Tnfrsf1b^−/− HPCs (Fig. 5G), further validating the coregulation of hnRNPK-YAP interplay in driving target gene expression. To better corroborate the effects of the TNFR2–hnRNPK axis on YAP-associated chromatin binding, we then performed ChIP-qPCR assays in Tnfrsf1b- and hnRNPK-depleted HPCs. Consistently, we found TNFα-induced YAP binding at gene promoters were greatly impaired in both deficient cells (Fig. 5H and I; Supplementary Fig. S5F and S5G). Furthermore, the occupancies of YAP on gene promoters can be rescued upon reintroduction of hnRNPK in Tnfrsf1b^−/− HPCs (Fig. 5J; Supplementary Fig. S5H). Meanwhile, mRNA expressions of YAP target genes including ELK1, ELK4, and STAT3 were also dramatically reduced in tumors derived from rats receiving hnRNPK-depleted HPCs compared with those from WT HPC-induced tumors (Supplementary Fig. S5I). Collectively, these results indicate that hnRNPK functionally acts downstream of TNFα–TNFR2 signaling to interact with YAP, forming a complex on the promoter regions of YAP target genes in HPCs.

To further corroborate our conclusion that HPCs serve as a cell origin of PLC and that TNFR2-induced YAP signaling is required for HPC-mediated PLC, we performed scRNA-seq analysis using clinical samples of patients with PLC. We first generated scRNA-seq profiles of freshly isolated PLC from patients with 10x Genomics. A total of 25,189 single-cell transcriptomes from tumors of two patients were obtained after initial quality controls. Then, we performed t-SNE analysis of all cells, which were grouped mainly according to their cell types as annotated on the basis of known cell lineage–specific marker genes unique to CD4⁺/CD8⁺ T cells, regulatory T (Treg) cells, B cells, tumor-associated fibroblasts (TAF), tumor-associated macrophages (TAM), liver sinusoidal endothelial cells (LSEC; Fig. 6A). In particular, we noticed a population of cells with an unknown entity but express hepatic progenitor cells (HPC-like) markers (Fig. 6A). These results are consistent with previous findings in other tumor types (29). We further confidently discriminated 1,206 malignant cells by inferring large-scale chromosomal CNVs based on transcriptomes (Fig. 6A and B). The inferred CNV profiles are consistent with the CNV profiles in HCC from previously published studies (38). We next determined the trajectory of HPC-like and malignant cells based on the reversed graph embedding method. Interestingly, we found HPC-like branches tend to be aggregated at the trunk of the trajectory tree, while malignant cells were accumulated at the branches of the tree (Fig. 6C), suggesting that the malignant cells may derive from HPC-like population.

EPCAM⁺ cells have been regarded as HPC-like as they can form dense round colonies when cultured and are bipotent progenitors of hepatoblasts, which eventually differentiate into cholangiocytes or hepatocytes both in vitro and in vivo (39, 40). We identified biliary and potential liver progenitor cell surface marker genes that correlated with EPCAM expression (Fig. 6D). We then examined the expression levels of TNFRSF1B (Fig. 6E), TNFRSF1A (Fig. 6F), HNRNPK (Fig. 6G), and YAP1 (Fig. 6H) in these cells and found that the expression of TNFRSF1B was much higher than TNFRSF1A (Fig. 6E and F). Moreover, we found a cooccurrence of TNFR2, hnRNPK, and YAP in patient-derived HPC-like population (Fig. 6E, G, and H). Consistent with the positive regulation of TNFR2 on YAP activity, GSEA revealed that the expression of a group of YAP target genes was reduced in TNFRSF1B⁻ HPC-like (Fig. 6I). Then, we continued to demonstrate the expression patterns of CTGF (Fig. 6J), CYR61 (Supplementary Fig. S6B), ELK1 (Supplementary Fig. S6C), ELK4 (Fig. 6K), and STAT3 (Fig. 6L) at single-cell level of liver cancer tissue from patients. The situation of ELK4 (Fig. 6K) and STAT3 (Fig. 6L) are extremely similar with that of YAP1. Collectively, these results indicated cooccurrence of TNFR2, hnRNPK, YAP, as well as those oncogenic target genes, in HPC-like populations, further supporting the notion that the TNFR2–hnRNPK–YAP signaling is essential for HPC-mediated PLC development.

To further verify the correlation between hnRNPK and YAP in PLC, we analyzed the expression profiles of hnRNPK and YAP in PLC datasets available from the TCGA (30). Indeed, the mRNA levels of hnRNPK and YAP were both significantly elevated in PLC patient specimens (n = 367) compared with adjacent tumor tissues (n = 50, Supplementary Fig. S7A), and there was a positive correlation between hnRNPK and YAP expression (R² = 0.3957, P < 0.0001; Supplementary Fig. S7B).

To determine the potential association of TNFα-induced TNFR2-hnRNPK-YAP signaling with clinical outcomes, we performed IHC staining of TNFR2, hnRNPK, and YAP on tissue microarrays containing 77 PLC specimens that have long-term clinical follow-up records. Consistent with the well-established oncogenic role of YAP in liver tumorigenesis, we observed dramatically increased YAP expression in patients with PLC (Fig. 7A and B). Notably, elevated expression of hnRNPK and its colocalization with YAP in the nucleus were observed in PLC tissues (Fig. 7A and B). Meanwhile, highly similar staining patterns were observed for TNFR2, strongly indicating its coaccumulation with hnRNPK and YAP in patients with PLC (Fig. 7A and B). Furthermore, we collected and preprocessed patients' data by extracting six available clinical factors in two categories, that is, the clinical background (age, gender), and cancer's stage information (tumor size, lymph node metastasis, distant metastasis, tumor stage). Our results showed that the staining of TNFR2 was significantly associated with tumor size, lymph node metastasis, tumor metastasis, and TNM stage (all P < 0.05, Supplementary Table S1), but not significantly correlated to gender and lymph node metastasis. Staining of hnRNPK was associated with tumor size, tumor metastasis, and TNM stage (P < 0.05), but not significantly correlated to age, gender, lymph node metastasis (Supplementary Table S2). YAP staining was only associated with tumor size and TNM stage (P < 0.05), but not significantly correlated to age, gender, lymph node metastasis, and tumor metastasis (Supplementary Table S3). Moreover, the protein level of either TNFR2 or hnRNPK was positively correlated with that of YAP (P < 0.05; Supplementary Tables S4 and S5). A similar positive association was also revealed between TNFR2 and hnRNPK (P < 0.05; Supplementary Table S6).

Next, we performed Kaplan–Meier analysis to assess the association of the TNFR2–hnRNPK–YAP signaling axis with overall survival of patients with PLC. Our results showed a clear association of high expression level hnRNPK with poor prognosis of patients with PLC (all P < 0.05; Fig. 7C; Supplementary Fig. S7C). In addition, patients with PLC with high levels of both TNFR2 and hnRNPK/YAP had shorter survival, whereas those with low levels of TNFR2 and hnRNPK/YAP had longer survival. Collectively, these results highlight the hyperactivation of TNFR2–hnRNPK–YAP signaling axis as a PLC prognostic marker (relative risk, 0.3496; 95% CI, ∼0.1687 to ∼0.7246; P = 0.0013), and demonstrate a close pathologic association of this axis with PLC pathogenesis.

Accumulating evidence has suggested that HPCs represent a major cell origin of PLC (4–6, 23). But most of researches observed the tumorigenic ability of HPCs in vitro or in nude mice models (6, 23), which could not mimic the inflammatory microenvironment during PLC development in vivo. Here, we identified TNFR2 as a specific receptor linking inflammatory stimuli such as TNFα to malignant transformation of HPCs, which drives PLC development in rats. Our study uncovered a TNFR2–hnRNPK–YAP signaling axis essential for TNFα-induced malignant transformation of HPCs and highlighted the pathologic association of this axis with PLC development (Fig. 7D).

PLC frequently arises in the presence of chronic injury and inflammation. Our previous studies reported that TNFα markedly drove liver tumorigenesis (25, 41, 42). Furthermore, we discovered TNFα promoted PLC dependence in a manner dependent on TNFR2 (25). On the basis of these findings, here we show that TNFR2 in HPCs are required for PLC development. Interestingly, our study demonstrated opposite phenotypic effects for TNFR1 and TNFR2 in HPCs during PLC development. Such observations were consistent with the mechanistic observations showing that TNFR1 deficiency promoted, but TNFR2 deficiency abolished, TNFα-induced accumulation and activation of YAP during PLC formation. This phenomenon may relate to the selective orientation of receptors, which need further study.

The transcriptional regulator YAP is pervasively activated in human malignancies, and recent work has indicated that, remarkably, YAP activation is apparently essential for cancer initiation and/or the growth of most solid tumors (16). Inflammation is known to be a powerful factor for inducing YAP activation (17). Recently, the effect of TNFα on Hippo–YAP pathway received increasing attention (26). But the mechanisms underlying both the nuclear translocation and transcriptional activation functions of YAP remain poorly understood. We found that TNFR2 triggers YAP activation via hnRNPK, which binds with YAP to form a complex inducing the transcription of YAP target genes. We further revealed that hnRNPK promotes the nuclear retention of YAP and also coregulates the expression of a large group of Hippo pathway target genes, indicating that hnRNPK is a biomolecular link between TNFR2-depenent TNFα signaling and Hippo signaling.

hnRNPK, a member of Heterogeneous nuclear ribonuclear proteins (hnRNP) family, takes part in both physiologic and pathologic processes such as spermatogenesis, ovary development, erythroid differentiation, and carcinogenesis (43, 44). Current understanding holds that the aberrant cytoplasmic localization of hnRNPK is associated with poorer prognosis for patients with cancer (45). It has also been reported that hnRNPK is involved in the development of hematologic disorders, and may even act as a tumor suppressor (46). Excitingly, our study reveals that hnRNPK is required to promote YAP's transcriptional activity in the nucleus. Moreover, considering our observations that hnRNPK levels are positively correlated with YAP in clinical PLC patient samples and that elevated hnRNPK levels are associated with poor prognosis, the therapeutically attractive speculation that depletion of hnRNPK may inhibit PLC growth awaits confirmatory study.

In summary, we have established TNFR2 as a specific receptor downstream of TNFα in driving HPC-mediated PLC, and demonstrated that TNFR2 regulates YAP's activity via hnRNPK. hnRNPK directly binds to and stabilizes YAP on target gene loci to promote their transcription. The TNFR2–hnRNPK–YAP signaling was hyperactivated in HPC-like cells driving PLC and associated with poorer prognosis.

Y. Meng reports grants from National Natural Science Foundation of China during the conduct of the study. Q. Zhao reports grants from The National Natural Science Foundation of China during the conduct of the study. L. An reports grants from National Natural Science Foundation of China and grants from Shanghai Science and Technology Committee (STCSM) during the conduct of the study. S. Jiao reports grants from National Natural Science Foundation of China during the conduct of the study. R. Li reports grants from The National Natural Science Foundation of China during the conduct of the study. Y. Sang reports grants from National Natural Science Foundation of China during the conduct of the study. J. Liao reports grants from Science and Technology Department of Fujian Province during the conduct of the study. Z. Zhou reports grants from National Natural Science Foundation of China, Ministry of Science and Technology, and grants from Chinese Academy of Sciences during the conduct of the study. L. Wei reports grants from National Natural Science Foundation of China and grants from Ministry of Science and Technology of the People's Republic of China during the conduct of the study. No disclosures were reported by the other authors.

Y. Meng: Conceptualization, data curation, validation, writing–original draft, project administration, writing–review and editing. Q. Zhao: Conceptualization, resources, data curation, validation, methodology, writing–original draft, project administration, writing–review and editing. L. An: Resources, data curation, validation, methodology, writing–original draft. S. Jiao: Resources, validation, methodology, writing–original draft. R. Li: Resources, investigation, writing–original draft. Y. Sang: Resources, data curation, software, investigation. J. Liao: Data curation, software, methodology. P. Nie: Data curation, methodology. F. Wen: Data curation, formal analysis, methodology. J. Ju: Formal analysis. Z. Zhou: Conceptualization, formal analysis, writing–review and editing. L. Wei: Conceptualization, writing–review and editing.

The authors thank staffs of the Core Facility for Cell Biology of Shanghai Institute of Biochemistry and Cell Biology for technical support. This project was supported by National Key Program of China (grant no. 2017YFA0504503, 2017YFA0504504, 2018YFA0107500, 2020YFA0803200), National Natural Science Foundation of China (81630070, 81972599, 81725014, 81822035, 31930026, 81802737, 81822035, 81902806, 32070710). Shanghai Pujiang Program (19PJ1408300), and the “Strategic Priority Research Program” (XDB19020202). Natural Science Foundation of Fujian Province, China (grant no. 2019J01298).

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.