Identification of DUSP4/6 overexpression as a potential rheostat to NRAS-induced hepatocarcinogenesis

The original pT/Caggs-V12Nras [26] had been cloned into the pT3 backbone to generate pT3-CAGGS-NRasV12 [27]. pT3-CAGGS-NRasV12 was applied with pPGK-SB13 for HTVI as described previously [27]. The pT3-CAGGS-NRasV12 vector was furthermore used for the transfection of HCC cell lines. pT3-EF1α constituted the empty vector (EV) control. The control group's data have been published previously [19]. Purification of the plasmids was performed with the Endotoxin-free Maxi Prep Kit (Sigma-Aldrich, St.Louis, MO).

A Rassf1a KO founder breeding pair was kindly provided by Dr. Louise van der Weyden (Wellcome Trust Sanger Institute, Research Support Facility, Hinxton, Cambridge, CB10 1SA, UK). Note that the genetic background of Rassf1a WT and KO mice was C57BL/6 J x 129 Sv. As previously described [19], the control group of Rassf1a WT mice was generated by crossing Rassf1a KO mice with C57BL/6 J mice purchased from Charles River Laboratories (Sandhofer Weg, 97,633 Sulzfeld, Germany). Breeding conditions and PCR verification were identical to our report of Rassf1a KO with PIK3CA injections [19].

We followed exactly the previously described protocol [19]. Mice with an age of 6–8 weeks were subjected to HTVI. The experimental groups comprised: untreated, 1 × phosphate-buffered saline (PBS), empty vector (EV), and pT/Caggs-V12Nras co-injected with pPGK-SB13 in a ratio of 25 to 1 in a total volume of 2 ml 0.9% sodium chloride. The total volume was injected into the lateral tail vein within 5 to 7 s, resulting in temporal right-sided heart failure and increased hydrodynamic pressure in the hepatic veins. The distribution of mice per group is summarized in Supplementary Table 1.

The quality of the injection was defined by the following criteria, which led to the exclusion of the respective mice from further analysis: less than 2 ml of total injection and subsequent unusual behaviour lacking the expected reduction in the movement for at least 60 min. Termination criteria were palpable masses (equivalent to a tumour diameter of about 4 cm), respiratory distress, and lethargy. Animals were euthanized by cervical dislocation, subjected to a standardized autopsy protocol, and livers were photo-documented. Animal breeding and animal experiments were in accordance with protocols by the Mecklenburg-Western Pomeranian federal institution “Landesamt für Landwirtschaft, Lebensmittelsicherheit und Fischerei (LALLF) Mecklenburg-Vorpommern” (protocol number/Aktenzeichen: 7221.3–1.1–052/12).

The human HCC cell lines PLC/PRF/5, HLE, HLF, Snu182, Snu387, Snu449, Hep3b, SK-HEP1, MHCC97-L, HuH6, as well as the human hepatoblastoma cell line Hep-G2 and the human cholangiocarcinoma cell lines KKU-M055, KKU-100, KKU213, HuCC-A1, CC-LP-1, OZ, RBE, SG231, TKKK, and YSCCC were cultured in 5% CO₂ at 37 °C in a humidified incubator. Cell lines were purchased from ATCC (Manassas, Virginia, USA). Cells were grown in Dulbecco’s modified Eagle medium (Gibco, Grand Island, NY) or RPMI 1640 Medium (Gibco) supplemented with 5% fetal bovine serum (Gibco), 100 mg/mL streptomycin, and 100 U/mL penicillin, 1% 1 M HEPES Buffer (Gibco), 1% 100 mM sodium pyruvate (Gibco) and 1% GlutaMAX™ (Gibco).

The AKT/NRAS cell line was derived from a murine HCC that developed after HTVI of myristoylated AKT^flox/flox and NRAS^G12V [27]. Transfection of Cre recombinase resulted in the AKT Cre/NRAS cell line, which was driven solely by the NRAS^G12V oncogene, while AKT was depleted. These cells were maintained in Dulbecco’s modified Eagle medium (Gibco, Grand Island, NY).

For DUSP4 silencing, we seeded cells at a density of 4 × 10⁵ cells in 2 ml of medium per well in 6 well plates. Cells were transfected either with DUSP4 and NEG siPOOL (Catalogue numbers: Dusp4-m-002 and N000-c1-059; siTOOLs Biotech GmbH, Planegg, Germany) using Lipofectamine® RNAiMAX (Thermo Fisher Inc). Lipofectamine and siRNA were diluted and combined in OptiMEM® Reduced Serum Medium (Thermo Fisher Inc.). After an incubation period of 48 h, the transfection was repeated. Subsequently, the cells were harvested after an additional 24 h of incubation with cell scrapers. Cells were pelleted by centrifugation (300 g, 5 min). These pellets were used for further downstream analyses.

For transfections, cells were inspected for viability and confluency of about 70–80% and seeded at a density of 4 × 10⁵ cells in 2 mL of medium per well in a 6-well plate. Cells were transfected with pT3-CAGGS-NRasV12 or empty pT3 vector the following day using Lipofectamine® 2000. Lipofectamine and vectors were diluted in OptiMEM® Reduced Serum Medium and combined. Medium in the wells was discarded, and cells were washed with 1xPBS before the addition of the transfection medium. Cells were harvested and pelleted by centrifugation after 48 h incubation using cell scrapers.

For MEK-inhibition experiments, trametinib (Selleck Chemicals GmbH, Berlin, Germany) and mirdametinib (Selleck Chemicals GmbH) were applied. Cells were seeded in 6 well plates at a density of 3–5 × 10⁵ cells in 2 mL medium per well. The control group was treated with DMSO (concentration matching the highest inhibitor concentration as all inhibitors were dissolved in DMSO). After incubation for 24 h at the indicated concentrations, cells were harvested, and cell pellets were obtained.

TPA (12-O-Tetradecanoylphorbol-13-acetate; P1585, Sigma-Aldrich), a phorbol ester, which induces pERK1/2 via the protein kinase C (PKC) [28], was administered to the medium in concentrations from 0.5 – 2 µM for (30 and) 60 min. The dimethyl sulfoxide (DMSO) control was matched to the highest concentration of TPA applied. The displayed data is representative of at least 2 technical replicates in a minimum of 2 biological repeats for all performed experiments. The data of these replicates and repeats are shown in Figure S7 with the designation of the corresponding figures in the main manuscript.

In addition, cell viability assays using the xCELLigence® real-time cell analysis dual plate (RTCA DP) device (OLS OMNI Life Science GmbH & Co KG; Bremen, Germany) were carried out. For impedance-based real-time cell index measurement, cells were grown on E-Plate 16 PET (Agilent Technologies, Inc.). The interval for measurement sweeps was set to 15 min. ~ 5000 cells suspended in a total volume of 150 µl of growth medium were seeded in each well. After 24 h siRNA against DUSP4, 6, the combination of both or SCR was added equivalent to the protocol described above. Afterward, we acquired the measurement for up to 72 h. Raw data were analyzed with the RTCA software (OLS OMNI Life Science GmbH & Co KG). The data were normalized to the timepoint of inhibitor addition.

Total RNA was extracted from 4 samples from each group: empty vector (EV), normal-appearing tissue in NRAS injected mice (NT), and the matched tumorous tissue (T). The concentration was measured on a DeNovix® DS-11 FX + und NanoDrop® ND-1000 (DeNovix Inc., Wilmington, USA), and RIN^e values were obtained on a 200 TapeStation (Agilent, Santa Clara, USA). The 12 mouse RNA samples that were used for further analyses had RIN^e values between 7.1 and 9.9 and were adjusted to a concentration of 100 ng/5 µl. Gene expression analysis was conducted using the nCounter® Mouse PanCancer Pathways Panel, which comprises 770 genes from 13 cancer-associated canonical pathways (NanoString, Seattle, USA).

Total RNA from DUSP4 silencing experiments in AKT/NRAS and AKT Cre/NRAS was analyzed in 3 biological replicates each (SCR vs. DUSP4 siRNA) with the nCounter® Mouse PanCancer Pathways Panel. Following the manufacturer’s instructions, 100 ng of total RNA was hybridized overnight with the reporter and capture code set at 65 °C. Excess probes were washed off using a two-step magnetic bead-based purification on the nCounter Prep station (NanoString Technologies). Elution from the beads, immobilization on a cartridge, and subsequent alignment followed. Data were collected using the nCounter Analyzer (NanoString Technologies) at 555 fields of view (FOV) through epifluorescence microscopy and CCD image acquisition.

The nSolver path view module was employed to display the undirected global significance score of the included Kyoto Encyclopaedia of Genes and Genomes (KEGG) gene as a heatmap [29]

Please refer to the supplementary methods section for an overview of these techniques.

Upon HTVI of NRAS^G12V, macroscopic hepatic tumours were detected as early as 3 months after the injection in the Rassf1a wildtype (WT) group, while the first grossly visible tumours were detected at the 6 months time point in Rassf1a knockout (KO) mice. Overall, 22 out of 60 (36.7%) mice developed grossly discernible tumours in the WT group, similar to 23 out of 62 (37.1%) mice in the KO group (Fig. 1b). Histologically, preneoplastic lesions (see definition below) were detectable in 7 WT mice (11.7%) and 5 KO mice (8.1%). Evaluation of the incidence per time point also yielded similar results between WT and KO, as inferred from Fig. 1c. Several mice required euthanasia before the predetermined time point and are represented in the intervals 3–6, 6–9, and 9–12 months. The liver weight/ body weight ratio was used as a surrogate parameter for tumour burden; again, no apparent differences between KO and WT mice were observed (Fig. 1d). Overall, tumour latency was 3–6 months, after which most mice displayed macroscopic tumours. As reported previously, the control group, which comprised untreated, 1xPBS, and EV injected mice did not develop tumours in the WT group, while 1 tumour each was detectable at 9 and 18 months in the Rassf1a KO cohort [19]. Histologically, the NRAS-induced tumours mostly had a well-circumscribed margin with a solid (partly also pseudocystic) growth pattern. They comprised cells reminiscent of hepatocytes with limited to moderate cytological atypia. Nuclei were hyperchromatic, and the ratio of the nuclear-cytoplasmic ratio was increased. Occasionally, vacuolated tumour areas could be observed, which is indicative of lipid accumulation inside the cells, while being much less pronounced when compared to our phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) injection model [19]. Often tumour cells had amorphous eosinophilic inclusion bodies. Some tumours also showed intravascular growth, although we did not detect metastases outside the liver by macroscopic and microscopic evaluation during the autopsy. A subset of tumours possessed a central fibrotic core, populated by a distinct tumour cell population growing in strands or a ductular to cribriform pattern. These tumour cells were weakly positive for cytokeratin 7, thus imparting a biliary phenotype, while showing reduced expression of CPS1, a hepatocyte-specific marker (Figure S1). Apart from tumours, we defined the category of preneoplastic lesions. This distinction was based on a morphological comparison: tumours displayed the occasional presence of necrosis, expansive growth with evident compression or more diffuse infiltration of surrounding tissue, and macroscopic correlation as well as more overt cytologic signs of malignancy compared with preneoplastic lesions. The size of preneoplastic lesions was defined as more than 0.2 and less than 0.5 cm. Examples are shown in Fig. 2a. Rassf1a WT and KO tumours were indistinguishable at the histological level. As discussed above, the observation that NRAS^G12V injections alone led to tumorigenesis was deemed highly interesting, since the previous assumption had been that it would elicit senescence and consecutive clearance of early preneoplastic foci by the immune system [12]. The resulting tumours offered the possibility to study exclusively RAS-dependent hepatic tumorigenesis, so we continued with an in-depth analysis of NRAS^G12V−induced tumours in Rassf1a WT mice.

On immunohistochemical (Fig. 2a) and Western blot evaluation (Fig. 2b), both tumours and preneoplastic lesions shared a common profile. An overexpression of pan-RAS and NRAS could be confirmed, which indicates the incorporation of the NRAS^G12V transgene. The immediate effector p-RAF1 was consequently increased at the protein level. In addition, augmented p-MEK positivity was hinted in the nuclei, while being more evident in the Western blot. Moreover, enhanced p-ERK cytoplasmic and nuclear positivity was evident compared to the surrounding tissue. Ki67 positive nuclei appeared to be more frequent in tumours when compared to preneoplastic lesions (Fig. 2a).

Altogether, through this comprehensive characterization, we were able to underline the strong upregulation of major known canonical effectors of RAS. Nonetheless, signalling via the PI3K-mTOR and Hippo pathways was also observed (detailed in the figure legend), suggesting a considerable level of crosstalk [32, 33]. Thus, despite being exclusively Ras-triggered, our model outlines the complexity of effects on oncogenic signalling via various effectors.

It should be noted, that the distinction between preneoplastic lesions and tumours was perfomed exclusively on the above-mentioned morphological criteria. At the molecular level, oncogenic pathways were gradually activated. Thus, in this model, we could not demonstrate a stepwise process of carcinogenesis. On these grounds, using the terms "early and late tumours" would also be justifiable as a replacement for preneoplasia and tumour.

Evaluation of the pathway scores based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) gene sets showed a progressive increase in the activity of several major cancer-related pathways when comparing preneoplastic tissue and tumour tissue against normal liver tissue. Transcriptional Misregulation and Apoptosis were the most prominently altered pathways (Fig. 3d).

The differentially expressed genes analysis of preneoplastic lesions versus normal tissue only yielded 3 genes with an adjusted p-value less than or equal to 0.05, including Prom1, Cdkn2b, and Ngf. In contrast, the comparison of EV-injected samples with tumours detected 59 significantly differentially expressed genes with an adjusted p-value less than or equal to 0.05. Of note, among the most highly overexpressed genes, we again detected Prom1, Cdkn2b, and Ngf (Fig. 3a and 3b, Supplementary Data 1), which implies an early and sustained contribution throughout the process of carcinogenesis. The marked upregulation of Cdkn2b, while Cdkn1a and Cdkn1c were found elevated (Supplementary Data 1), strongly suggested an active oncogene senescence pathway, which was potentially counter-regulated by enhanced protein levels of CyclinD1 evident by immunohistochemistry (Figure S2). Presumably, the net effect of the combined changes was an evasion of the senescence mechanism with the development of preneoplastic lesions and tumours. The upregulation of Ngf argues for the presence of a positive feedback loop, since Ngf can cause sustained ERK1/2 activation and nuclear translocation [37]. Finally, the strong and early upregulation of Prom1 (encoding CD133), which was also confirmed by immunohistochemistry (specific membranous expression), RT-qPCR, and Western blot (Fig. 4), was of particular interest. While CD133 is one of the most established markers for detecting and isolating cancer stem cells, including liver cancer, its function remains enigmatic [38]. For instance, it has been connected to oncogenic β-catenin and JAK/STAT pathways [39, 40]. Moreover, it has been purported that CD133 activity is enhanced by NFκB, which is also upregulated in our model [41]. Nonetheless, the exact interrelation with the RAS/MAPK pathway has not been reported, so we analysed this interaction further in the subsequent in vitro experiments.

Besides these genes, we also observed an upregulation of DUSP4 and 6 in comparing tumour samples and normal tissue (Fig. 3c). This upregulation is a compelling finding, since a digenic dependence on DUSP4 and DUSP6 has recently been reported, representing a potential therapeutic target [20]. While both DUSP4 and DUSP6 represent negative feedback regulators of ERK [42], they possess different substrate specificities. DUSP6 is a cytoplasmic ERK-specific phosphatase, and DUSP4 can additionally dephosphorylate JNK/p38 kinases [43]. Indeed, by immunohistochemistry, we could confirm an enhanced cytoplasmic positivity for DUSP6 in NRAS-induced tumours (Fig. 4a). At the same time, DUSP4 reacted predominantly in the nuclei (Fig. 4a lower panel), but in part also aberrantly in the cytoplasm of tumours (Fig. 4a upper panel) as it has been reported previously [44]. Upregulation in tumours was confirmed by RT-qPCR (Fig. 4b). Western blots displayed stronger bands in 2 out of 4 tumour samples, which were also evident in preneoplastic lesions in the case of DUSP6 (Fig. 4c). Finally, we confirmed the upregulation of Nupr1 in tumours by RT-qPCR, a stress-induced protein involved in sorafenib resistance and NFκB and ERK pathway crosstalk [45]. DUSP4 and DUSP6, as well as CD133, were subsequently investigated more in-depth.

Next, we set out to verify the regulation of the proteins mentioned above in human HCC cell lines (Figure S3). Six of 12 cell lines had detectable DUSP4 protein levels with weak expression in 3 cell lines (HUH7, HLE, and SNU-387), moderate expression in one cell line (PLC/PRF/5), and very high basal expression in 2 cell lines (MHCC97-L and SNU-449). Of note, the amount of DUSP4 protein did not correlate with pERK/ERK ratios. DUSP6 expression was detectable in 11 out of 12 HCC cell lines. All cell lines were positive for CD133, while some were more strongly positive than others. For the consecutive NRAS overexpression experiments, we used PLC/PRF/5, HUH-7, and SNU-182 with absent to moderate DUSP4 baseline expression. Transfection of these cell lines with pT3-CAGGS-NRasV12 led to a strong upregulation of NRAS protein expression (Fig. 5a). In response, the two cell lines, HUH-7 and SNU-182, exhibited an increase in the canonical effectors p-MEK1/2 and pERK1/2. Surprisingly, however, PLC/PRF/5 experienced a reduction in both p-MEK1/2 and p-ERK1/2. DUSP4 showed a strongly enhanced signal in all cell lines, and DUSP6 in PLC/PRF/5 and SNU-182. Apparently, additional NRAS^G12V transfection represented such a strong stimulus in PLC/PRF/5 cells that a negative feedback overcompensation occurred, likely mediated by DUSP4 and DUSP6. Interestingly, such overcompensation was not observed in the other two cell lines, suggesting that a threshold of “oncogene overdose” is required to reach the tipping point.

To investigate the specificity of these findings to NRAS activation, we subjected the cell lines to another method of ERK hyperactivation by treatment with the phorbol ester TPA, which induces pERK1/2 via protein kinase C (PKC) [28]. Strong activation of ERK1/2 was detected in all cell lines (Fig. 5b), and protein levels of DUSP4 were strongly upregulated in a dose-dependent manner. In contrast to NRAS-mediated ERK-activation, protein levels of DUSP6 consistently decreased. Compellingly, in this experiment, no overcompensation was evident in PLC/PRF/5 cells, which indicates that this phenomenon could be specific to the RAS/MEK/ERK cascade triggered directly by strong oncogenic signalling through NRAS^G12V. Moreover, this phenomenon could also rely on increased DUSP6 expression, which was contrarily regulated in the TPA treatments.

In addition, we studied the effects of MEK inhibition on the selected target proteins. For this experiment, we employed the AKT/NRAS cell line, derived from a murine HCC, which developed after HTVI of myristoylated AKT^flox/flox and NRAS^G12V [27], and its derivative AKT Cre/NRAS. Here AKT had been removed by the Cre recombinase, leading to a pure NRAS addiction as opposed to the combined AKT and NRAS oncogene dependency in the original cell line. The baseline characteristics of AKT/NRAS and AKT Cre/NRAS regarding DUSP4, DUSP6, and CD133 were very distinct (Fig. 5c). Indeed, AKT Cre/NRAS displayed a much higher activation of ERK1/2 and correspondingly high protein levels of DUSP4 and DUSP6, while CD133 was slightly increased. This effect argues for an exclusive role of DUSP4 and DUSP6 in NRAS dependency, which is attenuated upon additional concurrent activation of AKT.

We treated these two cell lines with the MEK inhibitors mirdametinib and trametinib for 24 h (Fig. 5d). As a result, DUSP4 and DUSP6 protein levels were strongly diminished in both cell lines. Also, pERK1/2 levels were reduced as expected in all treatment groups (although the effect of mirdametinib on ERK phosphorylation was limited in AKT Cre/NRAS). CD133 levels decreased considerably upon MEK inhibition in the purely NRAS-addicted AKT Cre/NRAS cell line. This finding argues for a direct relation between RAS-signalling and CD133 expression levels, confirming the results obtained in the mouse model.

Finally, we performed siRNA-mediated gene silencing of DUSP4 in these two murine cell lines. Moderate silencing efficiency was achieved on the protein level of DUSP4 and pERK1/2 (Fig. 6a). Downregulation of DUSP4 elicited an increase in pERK1/2 levels in both cell lines, especially in the AKT Cre/NRAS cell line. Protein levels of CD133 and DUSP6 were unaltered.

Subsequently, NanoString® gene expression analysis was conducted to identify differentially expressed genes following siRNA-mediated Dusp4 silencing in the AKT Cre/NRAS cell line (Fig. 6b). Some markers of cell cycle progression, including Ccnd2 and Cdk6, were upregulated. On the other hand, apoptosis-related genes were diminished, including Casp12 and Endog, implying a pro-proliferative and anti-apoptotic effect mediated by Dusp4 gene silencing. Furthermore, several genes related to the MAPK pathway were upregulated, including Map3k5, Map3k, Mapk9, and Prkca (except for downregulated Map2k6), which suggests a release of formerly active negative feedback regulation (Supplementary Data 1). Additionally, we compared the differentially expressed genes between AKT Cre/NRAS vs. AKT/NRAS (Figure S4, Supplementary Data 1). This comparison verified the reduction in Akt1 expression. Moreover, we found a strong upregulation of Cdkn2a in the AKT Cre/NRAS cell line, which could imply an increased tendency towards senescence and cell cycle inhibition in this cell line. Again, the purely NRAS defined oncogene addiction appeared to be more prone to senescence response, which can be overcome by the additional activity of the AKT pathway and was mirrored by much slower growth characteristics of the AKT Cre/NRAS cell line.

Subsequently, we evaluated the effects of DUSP4 and DUSP6 silencing, alone and combined, in the HCC cell line PLC and the murine AKT Cre/NRAS cell line using an impedance-based real-time measurement (xCELLigence®, Fig. 6c). We detected an increased growth rate for DUSP4 and DUSP6 silencing and the combination of both compared to SCR. In a parallel measurement with adjusted parameters, we further validated the effectiveness of the silencing in both instances (Figure S5).

Since in human HCC RAS mutations are rare, we resorted to CCA cell lines to verify the effect of RAS mutations on DUSP4 transcriptional levels. Therefore, we studied 12 CCA cell lines, of which we had previously confirmed the KRAS mutational status [4]. We found a significant increase (p = 0.043, Mann–Whitney test) of DUSP4 protein levels relative to β-actin in the 7 KRAS mutated cholangiocarcinoma cell lines when compared to KRAS wildtype CCA cell lines (Fig. 7).

In addition, we analysed our local CCA cohort, which had been screened before for mutations in KRAS, using a high-resolution melting curve analysis. From this cohort, we compared 35 KRAS WT to 19 KRAS mutated cases by qPCR (Fig. 7c). A statistically significant upregulation of DUSP4 expression levels in the KRAS-mutated tumours was detected (p = 0.048, Mann–Whitney test).

Furthermore, we confirmed the regulation of DUSP4 in response to NRAS^G12V. For this purpose, we selected the CCA cell line KKU-M055, which does not harbour KRAS or NRAS mutations (Figure S6) and has a low DUSP4 baseline expression. Following the transfection of NRAS^G12V, we observed a statistically significant (p = 0.026, Mann- Whitney-Test) upregulation of Dusp4 protein levels, while Dusp6 remained unaltered. Thus, the observation of Dusp4 overexpression under the influence of activated NRAS could be confirmed in a CCA cell, similar to the regulation in HCC cell lines described earlier.

Finally, we investigated further HCC and CCA mouse models to corroborate the dependency of DUSP expression on the injection of NRAS^G12V across entities. Specifically, we analysed PIK3CA^E545K/ MET and PIK3CA^E545K/ NRAS^G12V, which induce HCC [46]. An increase of Dusp4 and Dusp6 transcripts was detected. Similarly, a statistically significant difference for Dusp4 expression was observed in the NICD/NRAS CCA model [47] compared to EV (adj p = 0.023, Kruskal–Wallis test, Dunn’s multiple comparisons). No statistically significant difference was observed in the NICD only and NICD/AKT models [48]. Dusp6 showed a trend of being elevated in the NRAS combined model. Thus, we could clearly define a dependency of Dusp4 upregulation based on the presence of activated NRAS. This observation was specific to NRAS and could not be mirrored by e.g., AKT activation.