An inflammatory-CCRK circuitry drives mTORC1-dependent metabolic and immunosuppressive reprogramming in obesity-associated hepatocellular carcinoma

Obesity increases the risk of hepatocellular carcinoma (HCC) especially in men, but the molecular mechanism remains obscure. Here, we show that an androgen receptor (AR)-driven oncogene, cell cycle-related kinase (CCRK), collaborates with obesity-induced pro-inflammatory signaling to promote non-alcoholic steatohepatitis (NASH)-related hepatocarcinogenesis. Lentivirus-mediated Ccrk ablation in liver of male mice fed with high-fat high-carbohydrate diet abrogates not only obesity-associated lipid accumulation, glucose intolerance and insulin resistance, but also HCC development. Mechanistically, CCRK fuels a feedforward loop by inducing STAT3-AR promoter co-occupancy and transcriptional up-regulation, which in turn activates mTORC1/4E-BP1/S6K/SREBP1 cascades via GSK3β phosphorylation. Moreover, hepatic CCRK induction in transgenic mice stimulates mTORC1-dependent G−csf expression to enhance polymorphonuclear myeloid-derived suppressor cell recruitment and tumorigenicity. Finally, the STAT3-AR-CCRK-mTORC1 pathway components are concordantly over-expressed in human NASH-associated HCCs. These findings unveil the dual roles of an inflammatory-CCRK circuitry in driving metabolic and immunosuppressive reprogramming through mTORC1 activation, thereby establishing a pro-tumorigenic microenvironment for HCC development.

H epatocellular carcinoma (HCC) is among the most lethal cancers that significantly correlate with obesity [1][2][3] . The pathophysiology begins with obesity-induced hepatosteatosis and non-alcoholic steatohepatitis (NASH), collectively known as non-alcoholic fatty liver disease (NAFLD), which can further develop into cirrhosis and HCC 4 . Notably, HCC is characterized by strong sexual dimorphism in almost all geographic areas where male to female ratios average between 2:1 and 7:1 5, 6 . In a prospective study of >900,000 US adults, men with a body mass index (BMI) of 35 kg/m 2 exhibited a dramatic 4.52fold increase in relative risk of death from liver cancer, while a modest 1.68-fold increase was observed in women 2 . A recent population-based cohort study of 5.24 million adults in United Kingdom confirmed the significant modulation of HCC incidence by gender, in which higher BMI in men but not in women was associated with substantially increased risk of HCC 1 . In addition, another population-based cohort study of 1.2 million Swedish men further showed that a high BMI (≥30 kg/m 2 ) in late adolescence was associated with an increased risk of future severe liver diseases including HCC 3 . These findings consistently underscore the sex disparity in obesity-associated HCC, but the molecular mechanisms underlying HCC development in obese men remain obscure 4,6 .
Using obese mouse models exposed to the hepatic procarcinogen diethylnitrosamine (DEN), Park et al. demonstrated that obesity is a bona fide liver tumor promoter 7 . The obesity-driven HCC development largely depends on a chronic proinflammatory state that results in elevated circulating levels of cytokines, such as tumor necrotic factor-α (TNF-α) and interleukin-6 (IL-6) 7,8 , and the latter of which has recently been shown to correlate with HCC progression in obese people 9 . Chronic IL-6-mediated activation of signal transducer and activator of transcription 3 (STAT3) can cause hepatic insulin resistance critical for the development of glucose intolerance and steatotic HCC 10,11 . Unlike early hepatocarcinogenesis which relies on paracrine nuclear factor kappa B (NF-κB)-regulated IL-6 production by inflammatory cells 12 , HCC progenitor cells in premalignant lesions acquire autocrine IL-6-STAT3 signaling to stimulate cellular proliferation and transformation 13 . Nevertheless, it is unclear how the hepatic IL-6-STAT3 cascade is activated and sustained during malignant transformation.
One of the major IL-6-driven signaling pathways in obesity and HCC development is mechanistic target of rapamycin (mTOR) 7 , which is a key signal transducer in the phosphatidylinositol-4,5-bisphosphate 3-kinase (PI3K)/Protein Kinase B (AKT) pathway. mTOR can assemble with Raptor and Rictor to form two functionally distinct complexes, mTORC1 and mTORC2, respectively. Activation of cap-dependent translation by phosphorylation of 4E-BP1 contributes to mTORC1dependent carcinogenesis 14 . Consistent with the increased de novo lipid synthesis in proliferating cancer cells, mTORC1 has been shown to activate the central lipogenic transcription factor, sterol regulatory element-binding protein 1 (SREBP1), through S6K1 to stimulate lipogenesis and cell proliferation 15 . Animal model and human studies have confirmed the functional significance of mTORC1 activation in NAFLD pathogenesis 7,16 . Stimulation of AKT-mTORC1 signaling, either alone 17 or in combination with β-catenin 18 , induces hepatic lipogenesis and tumorigenesis. Nonetheless, how mTORC1 remains constitutively active in the context of insulin resistance is unresolved 19 . Additionally, mTORC1 was shown to be negatively regulated by glycogen synthase kinase 3β (GSK3β) via phosphorylation of tuberous sclerosis complex 2 (TSC2) 20 , which transmits diverse upstream signals including insulin to mTORC1 21 . Moreover, inactivation of GSK3β was shown to inhibit hepatocellular apoptosis in dietary obesity-promoted HCC 22 . While these findings implicate a causal effect of GSK3β dysregulation in obesity-related hepatocarcinogenesis, the upstream kinase that controls GSK3β/mTORC1 signaling in the obesity-induced inflammatory microenvironment has not been elucidated.
Genetic and biochemical studies have demonstrated the fundamental roles of androgen receptor (AR) in male predominance of HCC 23 . Using genome-wide location and functional analysis, cell cycle-related kinase (CCRK), the latest cyclin-dependent kinase member (CDK20), was previously underpinned as a direct AR-regulated oncogene in hepatocarcinogenesis through concordant activation of GSK3β/β-catenin and AKT/EZH2 signaling 24,25 . It was further shown that CCRK mediates virus-host signaling to promote hepatitis B virus (HBV)-associated hepatocarcinogenesis 26 , and fosters an immunosuppressive microenvironment for HCC development by induction of myeloid-derived suppressor cells (MDSCs) 27 . Notably, overexpression and hyper-activation of CCRK distinguishes a subset of HCC patients with poor overall and disease-free survival [24][25][26][27] . Based on the biological and clinical significance of CCRK in the male-predominant HBV-associated HCC 28 , here we investigated its potential role in obesity-associated HCC. Our findings demonstrated that knockdown of Ccrk dramatically suppressed hepatic lipid accumulation, inflammation and tumorigenicity in multiple murine NASH and HCC models. Mechanistically, obesity-induced pro-inflammatory STAT3 and AR-induced signaling cooperatively up-regulated CCRK expression, which in turn activated the mTORC1 pathway crucial for lipid/glucose homeostasis, immunosuppression, and tumorigenesis. These findings suggest that CCRK functions as a major signaling hub in obesity-associated hepatocarcinogenesis, providing insights into therapeutic strategies to reduce tumor burden from the worldwide obesity epidemic.
We next investigated whether CCRK promotes NASH development in male mice via administration of lentivirus expressing short-hairpin RNA (shRNA) in the dietary obesity model (Fig. 1a). Lentiviral shRNA-mediated down-regulation of Ccrk (shCcrk) in the livers of the obese mice nearly restored the plasma triglyceride and NEFA concentrations to the basal levels (p < 0.05; Fig. 1c), and reversed the glucose intolerance and insulin resistance when compared to the control mice treated with lentivirus expressing non-specific sequence control (shCtrl; p < 0.01; Fig. 1d, e). Moreover, Ccrk knockdown significantly reduced hepatic lipid accumulation, ballooning degeneration, and spotty necrosis (p < 0.05; Fig. 1g-j), thus supporting a key role of CCRK in promoting NASH.   Fig. 1 Dietary obesity-induced CCRK over-expression promotes lipid accumulation, glucose intolerance, and liver damages in male mice. a Schematic diagram of NASH mouse model with different diets (CD chow diet, HFHC high-fat high-carbohydrate) and lentivirus-mediated Ccrk knockdown (CD+shCtrl, n=8; HFHC+shCtrl, n = 15; HFHC+shCcrk, n = 15). b Body weight, c blood triglyceride and non-esterified fatty acid (NEFA) levels in mice were increased by HFHC at 28 weeks, which could be reduced by Ccrk knockdown. d, e CCRK impaired insulin sensitivity in mice. d Intraperitoneal glucose tolerance test (IPGTT) and e intraperitoneal insulin tolerance test (IPITT) were performed on CD-fed and HFHC-fed mice, blood glucose was measured at indicated time points after glucose or insulin injection (left), and area under the curve (AUC) is shown in a bar chart (right). f CCRK protein expression was induced by HFHC in mouse livers, which could be reduced by shRNA-mediated knockdown. Quantification of CCRK protein levels (relative to β-actin) is shown in a bar chart (bottom). g Representative pictures of Oil Red O, ballooning degeneration, and spotty necrosis of liver tissues in different groups (image magnification = ×200 or ×400, scale bar = 20 μm). h, i Quantifications of Oil Red O, ballooning degeneration and j scoring of spotty necrosis showed increased lipid accumulation, ballooning and spotty necrosis in HFHC-fed mice, which were reduced by Ccrk knockdown. Data are presented as mean ± SD. *p < 0.05; **p < 0.01; and ***p < 0.001 as calculated by unpaired two-tailed Student's t-test (b), one-way ANOVA followed by Bonferroni post-hoc test (c-f, h, j), two-way ANOVA followed by Bonferroni post-hoc tests (d, e), and Chi-square test (i) NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07402-8 ARTICLE CCRK mediates obesity-associated hepatocarcinogenesis. To further investigate the oncogenic function of CCRK, we employed an obesity-associated HCC model 7,30 using HFHC-fed male mice with neonatal DEN treatment (Fig. 2a, b). In this model, hepatic CCRK was markedly up-regulated (p < 0.001; Fig. 2c), which was associated with much higher lipid accumulation and HCC tumorigenicity when compared to the DEN-treated CD-fed mice (>10-fold; p < 0.01; Fig. 2d-g). Strikingly, when compared to shCtrl controls, Ccrk knockdown significantly reduced > 70% obesity-promoted tumor multiplicity and size in HFHC-fed mice (p < 0.01; Fig. 2e, f) but not in CD-fed mice ( Supplementary  Fig. 2a, b). Consistently, the livers of shCcrk-treated mice e Tumor multiplicity and f tumor sizes were higher in the livers of HFHC-fed mice, which could be reduced by Ccrk knockdown. g, h Dietary obesity-induced CCRK increased lipid accumulation and cell proliferation, which could be abrogated by Ccrk knockdown as shown by Oil Red O staining and Ki67 staining in the livers. i, j Blood triglyceride and NEFA levels were elevated by dietary obesity-induced CCRK at 28 weeks, which could be reduced by Ccrk knockdown. k CCRK over-expression impaired insulin sensitivity in mice. IPGTT was performed on CD-fed and HFHC-fed mice, blood glucose was measured at indicated time points after glucose injection (left), and AUC is shown in a bar chart (right). Data are presented as mean ± SD. *p < 0.05; **p < 0.01; and ***p < 0.001 as calculated by unpaired two-tailed Student's t-test (b), one-way ANOVA followed by Bonferroni post-hoc test (c, e-k), and two-way ANOVA followed by Bonferroni post hoc test (k) exhibited significant reduction in lipid accumulation (p < 0.05) and hepatocellular proliferation (p < 0.01; Fig. 2g, h). The reduced tumorigenicity was also associated with significant restoration of normal circulating triglyceride/NEFA levels (p < 0.05; Fig. 2i, j) and glucose tolerance (p < 0.05; Fig. 2k). Taken together, our results suggest that CCRK plays a key role in promoting obesityassociated hepatocarcinogenesis.
IL-6-mediated STAT3 and AR cooperatively induce CCRK. We next investigated the molecular mechanism underlying the upregulation of CCRK in obesity-associated hepatocarcinogenesis. As IL-6 represents a crucial HCC-promoting pro-inflammatory cytokine in obese state 7,9 , we treated both HepG2 and Huh7 HCC cell lines with IL-6, and found increased phosphorylation of STAT3 at Tyr705 (p-STAT3 Tyr705 ) and elevated mRNA and protein expressions of CCRK, which were abrogated by shortinterfering RNA (siRNA)-mediated knockdown of STAT3 ( Fig. 3a, b). Luciferase reporter assay further demonstrated that IL-6 transcriptionally up-regulated CCRK in a STAT3-dependent manner (Fig. 3c). Consistently, we found that the serum IL-6 levels were significantly elevated (p < 0.05) and positively correlated with CCRK protein expression in both dietary obesityinduced NASH and HCC models (p < 0.01; Supplementary  Fig. 2c, d).
Since CCRK is a direct transcriptional target of AR 24 , we determined the potential interaction between IL-6/STAT3 and AR signaling in CCRK up-regulation. Knockdown of AR by siRNA abrogated IL-6-induced CCRK transcriptional activity and expression in HepG2 and Huh7 cells (Fig. 3a-c). Notably, deletion of the AR response element (ARE) in the CCRK promoter 24 completely abolished the CCRK transcriptional activation by IL-6 ( Fig. 3c). Moreover, ectopic AR expression in STAT3 knockdown cells restored CCRK expression (Fig. 3d). Conversely, in CCRK-low-expressing LO2 immortalized hepatocytes and SK-Hep1 liver sinusoidal endothelial cells 31 , ectopic expression of constitutive active STAT3 (STAT3C), but not dominant negative STAT3 (DN-STAT3) up-regulated AR and CCRK expression ( Supplementary Fig. 3a), which could be abrogated by knockdown of AR ( Supplementary Fig. 3b). Collectively, these data demonstrate a concerted action by IL-6/ STAT3 and AR signaling to induce CCRK transcription and expression in liver and HCC cells, which was further supported by in vivo data from IL-6 neutralization experiment (Supplementary Fig. 3c, d).
We next investigated whether the auto-regulation occurs at the transcriptional level. Indeed, CCRK induced its own promoter activity, which was abolished by knockdown of either STAT3 or AR in CCRK-expressing LO2 and SK-Hep1 cells ( Fig. 3g and Supplementary Fig. 3f). Deletion of ARE in the CCRK promoter also prevented the transcriptional activation, indicating that the auto-regulation was dependent on direct AR binding ( Fig. 3g and Supplementary Fig. 3f). We next conducted coimmunoprecipitation to determine whether STAT3 physically interacts with AR for transcriptional regulation. WT but not KD CCRK induced a robust physical interaction between p-STAT3 Tyr705 and AR in LO2 and SK-Hep1 cells ( Fig. 3h and Supplementary Fig. 3g). Furthermore, pairwise sequential ChIP assays showed that CCRK-induced co-localization of STAT3 and AR interaction in the ARE of its own promoter in both cell lines ( Fig. 3i and Supplementary Fig. 3h). Taken together, these findings demonstrate that CCRK induces STAT3-AR physical interaction for transcriptional activation via promoter cooccupancy, thus forming an inflammatory-CCRK circuitry (Fig. 3j).
CCRK activates TSC2/mTORC1 signaling through GSK3β. We next elucidated the molecular pathways by which CCRK promotes NASH and HCC. Given the critical role of GSK3β, the direct substrate of CCRK 24 on mTOR signaling in obesityassociated HCC 20,22 , it is conceivable that the CCRK/GSK3β cascade regulates mTORC1 signaling to promote obesityassociated hepatocarcinogenesis. We found that ectopic expression of CCRK in LO2 and CCRK KO Huh7 cells activated mTORC1 signaling as shown by increased levels of phosphorylated mTOR at Ser2448 (p-mTOR Ser2448 ), 4E-BP1 at Thr37/46 (p-4E-BP1 Thr37/46 ), S6K at Thr389 (p-S6K Thr389 ), and the mature form of SREBP1 (mSREBP1), which were impaired by suppression of GSK3β phosphorylation at Ser9 (p-GSK3β Ser9 ) via over-expression of the constitutively-active S9A-GSK3β mutant (Fig. 4a). Moreover, knockdown of CCRK in HepG2 and Huh7 cells reduced mTOR phosphorylation and the downstream signaling cascades, which could be rescued by silencing of TSC2, an GSK3β effector on mTORC1 inhibition 33 (Fig. 4b). To further investigate whether this regulatory pathway is perturbed in vivo, we examined the expression of the signaling molecules in the dietary obesity models. In accordance with CCRK up-regulation, phosphorylation of GSK3β, mTOR, and the downstream signaling molecules were concordantly increased in the liver tissues of both NASH and NASH-HCC models (Fig. 4c, d). Moreover, down-regulation of Ccrk markedly suppressed the GSK3β/mTORC1/SREBP signaling ( Fig. 4c, d) and deregulated multiple lipid metabolic pathways. For instance, Ccrk down-regulation resulted in significant reduction of Srebf1 and its target genes Acc1, Acl, and Fasn, which are major drivers in de novo lipogenesis (Supplementary Fig. 2e). In parallel, Fabp3 was down-regulated suggestive of reduced fatty acid uptake ( Supplementary Fig. 2f), while Abcg1 was up-regulated indicative of enhanced lipid secretion ( Supplementary Fig. 2g). Intriguingly, Cpt2 and Ppara were also down-regulated, denoting suppression of fatty acid beta-oxidation presumably due to decreased amount of lipids ( Supplementary Fig. 2h). Of note, loss of CPT2 was recently shown to promote HCC by protecting tumor cells from lipotoxicity 34 and chemotoxicity 35 . Collectively, these data suggest that CCRK-induced liver steatosis in obesity-associated hepatocarcinogenesis is a net result of increased lipid synthesis and uptake, and decreased lipid secretion.
CCRK promotes lipid accumulation and insulin resistance via mTORC1. Recent studies underline a strong link between mTORC1 signaling and NAFLD pathogenesis 36 , which is associated with abnormalities in hepatic lipid and glucose metabolism as well as insulin resistance 37 . We thus hypothesized that over-  Vector HepG2 Huh7 HepG2 Huh7 L02

CCRK KO Huh7
Huh7 a b c d e f g h i j Fig. 3 IL-6 activates STAT3 and AR signaling to stimulate CCRK expression. a Western blot and b qRT-PCR analysis of HepG2 and Huh7 cells treated with or without IL-6, combined with siRNA-mediated knockdown of STAT3 or AR. c Luciferase reporter assay was used to measure CCRK promotor activity in HepG2 and Huh7 cells expressing either WT or AR response element (ARE)-deleted mutant CCRK promoter, treated with or without IL-6, combined with siRNA-mediated knockdown of STAT3 or AR. d CCRK expression was blocked by knockdown of STAT3, but rescued by over-expression of AR. e STAT3 phosphorylation suppressed by AR knockdown could be rescued by over-expression of CCRK. f WT but not kinase-defective (KD) CCRK-induced STAT3 phosphorylation and AR expression. g CCRK induces its own promoter activity, which was abolished by deletion of ARE or knockdown of either STAT3 or AR in CCRK-expressing cells. h Co-immunoprecipitation of p-STAT3 Tyr705 and AR in LO2 cells transfected with WT CCRK, but not in those transfected with empty vector or KD CCRK. IgG is a control for non-specific immunoprecipitation. i ChIP-re-ChIP assay demonstrated an increased co-occupancy of STAT3 and AR at the ARE of CCRK promotor in LO2 cells transfected with WT CCRK relative to those transfected with vector only. IgG is a control for nonspecific immunoprecipitation. The antibodies used in the 1st and 2nd ChIP are as indicated. j Schematic representation of the IL-6/STAT3/AR/CCRK circuitry in hepatocarcinogenesis. Data are presented as mean ± SD. *p < 0.05 and **p < 0.01 as calculated by one-way ANOVA followed by Bonferroni post hoc test (i) expression of CCRK contributes to the pathophysiology of NAFLD via mTORC1 signaling. To test this, we first constructed a CCRK-inducible LO2 liver cell model (LO2-CCRK) in which doxycycline treatment readily induced CCRK, leading to increased p-GSK3β Ser9 , p-mTOR Ser2448 , p-4E-BP1 Thr37/46 , p-S6K Thr389 , and mSREBP1 (Fig. 5a). Exposure of doxycyclineinduced LO2-CCRK cells with fatty acid mixture of oleic and palmitic acids (2:1 ratio), which mimics the liver steatosis of NAFLD patients 38 , significantly increased hepatocellular lipid accumulation compared to non-induced cells (p < 0.001; Fig. 5b, c). Notably, inhibition of the mTORC1 signaling by shRNAmediated knockdown of the positive regulatory subunit Raptor (shRaptor) or, rapamycin treatment (Fig. 5a), abolished the lipid over-accumulation in CCRK-expressing liver cells (p < 0.001; Fig. 5b, c). In parallel, doxycycline-induced LO2-CCRK cells also exhibited significantly higher glucose uptake (p < 0.001; Fig. 5d), which was completely abolished by mTORC1 inhibition (p < 0.001; Fig. 5d).
To determine whether CCRK promotes hepatic insulin resistance, we utilized CCRK KO Huh7 cells, which exhibited reduced GSK3β/mTORC1 signaling compared to the WT cells (Fig. 5e). We treated both WT and CCRK KO Huh7 cells with high dose of insulin for 24 h, followed by assessment of insulin sensitivity by phosphorylation of Akt at Ser473 (p-Akt Ser473 ) 7,30,39 . High insulin exposure reduced subsequent insulinstimulated p-Akt Ser473 in Huh7 WT cells (Fig. 5f, lane 4 vs. lane 2), which could be restored in CCRK KO cells (Fig. 5f, lane 6 vs. lane 4). To investigate whether mTORC1 signaling mediates CCRK-promoted insulin resistance, we treated LO2-CCRK cells in hyperinsulinemia condition and found that induction of CCRK further impaired insulin sensitivity (Fig. 5g, lane 6 vs. lane 4). Notably, both genetic and pharmacological inhibition of   tumors (p < 0.01; Fig. 5h, i), whereas inhibition of mTORC1signaling by either Raptor knockdown or Rapamycin treatment abrogated the induced tumorigenicity in CCRK-expressing cells (p < 0.01; Fig. 5h, i). Consistently, tumor nodules from doxycycline-induced LO2-CCRK cells showed significantly more proliferating cells when compared with non-induced, Raptorablated and Rapamycin-treated groups as assessed by Ki-67 staining (p < 0.05; Fig. 5j). Western blot analysis on the xenograft tumor tissues further confirmed that the reduced tumorigenicity was associated with the inactivation of GSK3β/ mTORC1 signaling and its downstream molecules, and the cellular proliferation marker PCNA (Fig. 5k). Taken together, we demonstrate that CCRK promotes hepatic metabolic dysregulation and tumorigenicity in an mTORC1-dependent manner.
We further validated these findings in the dietary obesity-HCC model (Fig. 2a). Consistently, the level of PMN-MDSCs, but not M-MDSCs, was also significantly increased in HFHC-fed mice (p < 0.05) but normalized to basal level by Ccrk knockdown (p < 0.05; Fig. 6g and Supplementary Fig. 5g). Overall, the level of PMN-MDSCs but not M-MDSCs significantly correlated with tumor multiplicity in the dietary obesity-HCC model (p < 0.0001; Fig. 6h and Supplementary Fig. 5g). To examine how hepatic CCRK expression induces PMN-MDSCs, we performed cytokine profiling and found a significant induction of granulocyte-colony stimulating factor (G-csf) 42 in the peri-tumoral liver tissues of CCRK TG mice (p < 0.05; Supplementary Fig. 5h). Consistently, the hepatic G-csf expression was significantly increased in HFHCfed compared to CD-fed mice (p < 0.05), which was abrogated by Ccrk knockdown (p < 0.01; Fig. 6i). Using both ectopic expression and KO cell models, we further demonstrated that CCRK positively regulated G-CSF expression, which could be abolished by Raptor knockdown (Fig. 6j, k). Taken together, these results suggest that hepatic CCRK signaling recruits pro-tumorigenic PMN-MDSCs to liver microenvironment through mTORC1dependent G-CSF expression.

Discussion
The rising prevalence of metabolic risk factors, especially obesity and diabetes, among HCC patients underscores the urgent need for novel therapeutic strategies 3,44,45 . Strikingly, the risk of obesity-associated HCC is even higher in males 1-3 . Here we show that hepatic CCRK cooperatively induced by the proinflammatory IL-6/STAT3 and AR signaling promotes HCC development by reprogramming lipid metabolism and immune microenvironment. Mechanistically, STAT3-AR co-binding stimulated by CCRK transcriptionally activates its own promoter, which in turn triggers the mTORC1/4E-BP1/S6K/SREBP1 cascades via GSK3β phosphorylation to augment hepatic lipid accumulation, glucose intolerance, insulin resistance, and tumorigenicity. Moreover, over-expression of CCRK reshapes the liver microenvironment by up-regulating mTORC1-dependent G-CSF expression to recruit pro-tumorigenic PMN-MDSCs, resulting in immune evasion. As we also showed concordant STAT3-AR-CCRK-mTORC1 signaling activation in patients with NASH-associated HCC, our findings unveil a central role of CCRK as a molecular hub of multiple metabolic and immunosuppressive cascades, and offer a novel therapeutic target for HCC intervention in the era of obesity epidemic. Mounting evidence suggests that obesity is associated with chronic low-grade systemic inflammation leading to metabolic syndrome and insulin resistance 46 . The pro-inflammatory IL-6 levels are higher in mouse and human NAFLDs 10 , and a positive correlation between hepatic IL-6 expression and disease severity was evident in patients with NASH 47 . Naugler et al. have previously shown that estrogen effectively suppressed IL-6 expression in female mice via MYD88/NF-κB signaling, and thereby reduced liver damages and HCC multiplicity 48 . In this murine HCC model, the IL-6 response was transient that occurred only during the first 48 h after DEN stimulation, whereas hepatocarcinogenesis was a long process that took almost 8 months 48 . These results implied that a sustainable oncogenic signal is required but not clearly defined. In this study, we addressed this enigma by showing in the male NAFLD context (Figs. 1 and 2) that IL-6 can induce a self-reinforcing STAT3/AR/CCRK oncogenic circuitry, in which CCRK facilitates the physical interaction of phosphorylated STAT3 and AR and their co-occupancy at CCRK promoter for sustained transcriptional activation (Fig. 3). As we also observed CCRK up-regulation in male HFHC-fed mice (Supplementary Fig. 1a) and NAFLD patients (Supplementary Fig. 6c), the IL-6/STAT3 and androgen/AR crosstalk in CCRK activation may explain why obese men are at greater risk of developing HCC.
The diverse functional effects of mTOR pathways in liver metabolism and hepatocarcinogenesis are well-established [15][16][17][18][19] , however, the upstream drivers have not been fully defined. We found that in both NASH and HCC murine models, mTOR signaling was consistently activated and driven by CCRK (Fig. 4). Mechanistically, CCRK phospho-inactivates GSK3β, thereby subverts the mTOR inhibitor TSC2 to up-regulate p-4E-BP1, p-S6K, and mSREBP1 levels. As a negative regulator of a key rate-limiting initiation factor eIF4E, 4E-BP1 phosphorylation by mTORC1 leads to the dissociation of 4E-BP1 from eIF4E, allowing the formation of cap-dependent translation initiation complex at 5′ end of mRNAs 14 . Once phosphorylated and activated by mTORC1, p-S6K1/2 is another major effector that regulates RPS6, as well as other regulators of translation initiation 15 . Notably, both of these critical pathways for protein synthesis have been shown to play indispensable roles for mTORC1-dependent hepatocarcinogenesis 49 . Additionally, increased de novo lipogenesis via the mTORC1/S6K1/SREBP1 axis has pathogenic and prognostic significance in HCC 15,17,50 . A recent study has further shown that either genetic or pharmacological inhibition of the SREBP pathway dramatically reduced HCC progression 51 . Importantly, we demonstrated that in vivo ablation of Ccrk simultaneously circumvents these metabolic and lipogenic mTORC1-dependent cascades to diminish hepatic lipid accumulation, glucose intolerance, insulin resistance, and HCC tumorigenicity (Fig. 5). Notably, CCRK also controlled the βcatenin-AR regulatory loop (Supplementary Fig. 7) as previously shown in HBV-associated hepatocarcinogenesis 26 , which may contribute to the lipogenic tumor phenotype 18 in the HFHC diet model. Overall, these mechanistic and functional data highlight CCRK as a crucial mTORC1 regulator in the development of NASH and NASH-associated HCC.
Both cellular and non-cellular components of the tumor microenvironment determine tumor development and progression, as well as anti-tumor immunity and response to cancer therapy 40 . Interestingly, in addition to the tumor-intrinsic roles, CCRK also exerts tumor-extrinsic functions in the liver microenvironment 52 . Using a CCRK TG model, we showed that CCRK effectively induced mTORC1-dependent G-csf expression to recruit PMN-MDSCs to the liver, leading to enhanced tumorigenicity potentially via immune escape (Fig. 6). In the dietary obesity HCC model, we further found that the PMN-MDSC level (Fig. 6g) and hepatic p-p65 ser536 expression ( Supplementary  Fig. 7) were positively regulated by CCRK. As PMN-MDSCs were also increased in male CCRK TG mice under HFHC diet (Supplementary Fig. 4c), our data suggest that the obesity-triggered AR/CCRK signaling may establish an immunosuppressive barrier through induction of MDSC expansion and recruitment via NF-κB/IL-6 27 and mTORC1/G-CSF 42 pathways, respectively. Ma et al. have recently depicted the role of lipid dysregulation on adaptive immune responses, especially the selective loss of intrahepatic CD4 + T lymphocytes in NAFLD-promoted HCC 53 . Thus, whether and how CCRK-driven signaling network modulates and impairs T cell-mediated tumor immune surveillance warrant further investigation.
Our findings of both intrinsic effects on hepatocytes and hepatocyte-extrinsic functions in the liver microenvironment highlight the therapeutic potential of targeting CCRK in HCC. Pharmacological inhibition of the aberrantly activated mTOR in HCC has been proposed, but the in vivo effects have not been clearly elucidated 54 . Rapamycin was even shown to produce unwanted adverse effects in HCC pre-clinical models 55 . As an upstream driver of mTOR pathway, Ccrk inhibition is sufficient to reverse the metabolic and oncogenic phenotypes in multiple NASH and HCC models. Moreover, a recent study showed that co-blockade of CCRK improved immune-checkpoint therapy for large hepatoma eradication 27,56 . There are currently a number of pan-CDK inhibitors that display binding to CCRK, but CCRKspecific inhibitor has yet been developed 56 . Given the functional and clinical significance of CCRK in both viral 26 and NASHrelated HCCs (Fig. 7 and Supplementary Fig. 6d), the development of CCRK-targeted agents and characterizations of their effects in combination immunotherapy will ultimately lead to more effective treatment for patients with HCC.

Methods
Patient specimens. Paired tumor/non-tumor tissues from 23 patients with NASHassociated HCCs who underwent hepatectomy (Supplementary Table 1), and liver biopsy tissues from 23 NAFLD patients (Supplementary Table 2), were collected at the Prince of Wales Hospital (Hong Kong) for analyses. All HCC patients reported a history of metabolic syndrome involving diabetes, hypertension, dyslipidemia, and hepatic steatosis. HCC patients with chronic hepatitis B and C, or record of excessive alcohol intake were excluded from this cohort. Informed consent was obtained from all human subjects, and the study was approved by the Joint CUHK-NTEC Clinical Research Ethics Committee. Liver tissues were graded by the NASH Clinical Research Network scoring system 57 .
Sense and antisense strands of siRNA oligonucleotides were synthesized and purified by Techdragon Company. The lyophilized powder of siRNAs was dissolved in DNase-/RNase-free water and aliquoted for storage at −80°C. siRNAs specifically targeting human CCRK, AR, STAT3, and TSC2 were designed. A scramble siRNA (siCtrl) was synthesized as a negative control, which does not target any human gene mRNAs. Cells were transfected with siRNAs using Lipofectamine 2000 reagent (Invitrogen) according to the manufacturer's protocol. In brief, at 24 h before transfection, cells were seeded (in antibiotic-free media) at 70-90% confluency. siRNAs and Lipofectamine 2000 were individually incubated with serum-free MEM medium (Gibco) for 5 min at room temperature, then they were mixed together and incubated at room temperature for 20 min. The mixtures were then pipetted onto the cells evenly, and the cells were cultured in the incubator until indicated time point.
Construction of inducible and stable cell lines. Generation of the doxycyclineinducible stable cell line was performed as described 61 . In brief, CCRK cDNA was amplified by PCR and used to replace the PAF gene of the vector PB-T-PAF to form the new construct PB-T-CCRK. LO2 cells were seeded and grown in T75 flasks overnight to reach 60-70% confluency. On the following day, cells were cotransfected with pCyL43, PB-RN together with or without PB-T-CCRK using Lipofectamine 2000 following the manufacturer's instructions. Two days after transfection, cells were subjected to dual drug selection, i.e. 1 μg/ml puromycin (Gibco) and 800 μg/ml G418 (Roche), until drug-resistant colonies formed and stabilized.
For constructing shRaptor and shControl (shCtrl) stable cell lines, doxycyclineinducible CCRK-expressing LO2 cells (LO2-CCR) were seeded and grown in sixwell plates overnight to reach 50-70% cell confluency. Cells were transfected with shRaptor or shCtrl plasmid using FuGENE ® HD Transfection Reagent and cultured for 48 h. After that, transfected cells were detached from the plates with 0.25% trypsin-EDTA, and transferred to a 100-mm tissue culture plate and maintained in 400 μg/ml hygromycin (Life Technologies Corporation) for 4 weeks until resistant colonies formed. Then, the stable cells (shRaptor or shCtrl-LO2) were isolated and transferred into new dishes and cultured in antibioticscontaining medium for further experimental use.
CCRK KO Huh7 cell line was constructed as previously described 62 . Briefly, human CCRK-targeted sgRNA was designed online (http://crispr.cos.uniheidelberg.de) and cloned into the vector lentiCRISPR v2 (Addgene Plasmid #52961). This construct was confirmed by sequencing. Next, Huh7 cells were transfected with this construct or control vector. Two days after transfection, puromycin (1 μg/ml) was used to select transfected colonies. Approximately 2 weeks later, individual cells were plated in 96-well plates and cultured for 4 weeks. CCRK KO Huh7 cells were confirmed by sequencing and Western blot.
Quantitative reverse transcription PCR (qRT-PCR). Total RNAs were extracted using TRIZOL Reagent (Invitrogen). In each reaction 500 ng of total RNA was used to generate cDNA. RNA samples were mixed with DNase (Invitrogen, CA, USA) to get rid of unwanted genomic DNA and incubated with PrimeScript RT Master Mix (Takara). The mixture was incubated at 37°C, 15 min; 85°C, 5 s. The cDNA was diluted in 150 µl DNase/RNase-free water and stored at −20°C until use. Aliquots (3 µl each) of cDNA were amplified using Power SYBR Green PCR Master Mix (Takara) and ViiTM7 Real-Time PCR System (Applied Biosystems). Each reaction was performed in triplicate and GAPDH was used as an internal control. The primers are listed in Supplementary Table 3. . Protein lysates from cell lines and tissues were prepared using lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% NP-40, 0.5% Na-deoxycholate) and T-PER Tissue Protein Extraction Reagent (Thermo Scientific) supplemented with protease inhibitor cocktail, respectively. Protein concentration was determined by the Bradford method (Bio-Rad Laboratories). 10-100 μg of protein was separated by 6-12% sodium dodecyl sulfate polyacrylamide gel electrophoresis and electroblotted onto equilibrated nitrocellulose membrane (Bio-Rad Laboratories). The indicated primary antibodies were incubated at 4°C overnight followed by secondary antibodies for 2 h at room temperature. Then, the antibody-antigen complexes were detected with enhanced chemiluminescence (ECL, GE Healthcare Life Sciences) and exposed to X-ray film (Fuji). β-actin was used as loading control. Signals were quantified by Image J software and defined as the ratio of target protein to β-actin. Uncropped Western blots are shown in Supplementary Fig. 8.
Co-immunoprecipitation. Cells were collected and lysed in 1 ml co-IP lysis buffer (20 mM Tris, pH 7.5, 15 0 mM NaCl, 1.0% Triton X-100, 1 mM EDTA, and protease inhibitor cocktail) on ice. The lysates were centrifuged at 12,000×g, 15 min, 4°C and immunoprecipitated with IgG, anti-p-STAT3 Tyr705 or anti-AR antibodies at 4°C overnight. On the following day, 20 μl of protein A/G beads (Santa Cruz) were added into the protein-antibody complex and incubated for 6 h at 4°C. The complexes were then washed with Co-IP lysis buffer four times for 10 min each at 4°C and collected by centrifuging at 2500 rpm, 5 min, 4°C. Proteins were detached from beads by denaturing with 5 × protein loading buffer at 100°C for 10 min. Protein interaction was then detected by western blot using indicated antibodies.
Chromatin immunoprecipitation (ChIP)-re-ChIP assay. ChIP-re-ChIP assay was performed as previously described 30 . At 48 h post-transfection of CCRK and control plasmids, cells grown at the confluency of 80% in 150-mm dishes were cross-linked with 1% formaldehyde for 10 min at room temperature on the shaker, followed by adding 1 M glycine to a final concentration of 125 mM for 2 min to quench formaldehyde. Next, cells were washed twice using PBS at 4°C and harvested by scraping. A total of 1 × 10 8 cells were collected by centrifugation at 700×g, 5 min, 4°C and lysed with 10 ml cold lysis buffer I (50 mM HEPES-KOH, pH 8.0, 1 mM EDTA, 140 mM NaCl, 10% glycerol, 0.5% NP-40, 0.25% Triton X-100, supplemented with protease inhibitors), and rotated at 4°C for 10 min. Then, the cell debris was harvested through centrifugation at 1350×g, 5 min, 4°C, and then resuspended in 10 ml cold lysis buffer II (10 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, supplemented with protease inhibitors) and rotated for 10 min at 4°C. The nuclear pellets were collected by centrifugation at 1350×g, 5 min, 4°C and resuspended with 3.6 ml lysis buffer III (10 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% Na-Deoxycholate, 0.5% N-Lauroylsarcosine, supplemented with protease inhibitors). DNA fragmentation was achieved using a Bioruptor ultrasonicator (Diagenode). 50 μl of cell lysate was saved as a reference sample. 5 μg of STAT3 antibody (Cell signaling) or AR antibody (Millipore) attached to 100 μl Dynabeads Protein G (Invitrogen) were used for immunoprecipitation of protein-DNA complexes overnight. For Re-ChIP assay, the beads were eluted in 10 mM DTT and mixed with Re-ChIP dilution buffer (50 mM Tris/HCl at pH 7.4, 150 mM NaCl, 1% Triton, 2 mM EDTA, and protease inhibitors) and subjected to a second ChIP. 5 μg of AR antibody NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07402-8 ARTICLE NATURE COMMUNICATIONS | (2018) 9:5214 | DOI: 10.1038/s41467-018-07402-8 | www.nature.com/naturecommunications (Millipore), STAT3 antibody (Cell signaling), or IgG antibodies (Santa Cruz) attached to 100 μl Dynabeads Protein G (Invitrogen), which was subjected to re-ChIP assay to bind with protein-DNA complexes for immunoprecipitation overnight. The IP DNA or input DNA was subjected to elution, reverse crosslink and purification. PCR primers targeting a region within 150 bp of the putativebinding site were designed to test the IP and input DNA. Equal amounts of IP and diluted input DNA were used as templates for quantitative PCR by Power SYBR Green-based detection (Applied Biosystems).
Luciferase reporter assay. The WT and androgen-responsive element (ARE)deleted CCRK promoter luciferase reporters were constructed as previously described 24 . For IL-6-induced CCRK transcription, cells were transiently transfected with WT or ARE-deleted CCRK promoter constructs and Renilla luciferase reporters. After 24 h, the cells were further treated with human recombinant IL-6 for 3 h. For CCRK self-reinforced transcription, cells were co-transfected with CCRK construct, WT or ARE-deleted CCRK promoter constructs and Renilla luciferase reporters for 48 h. The treated cells were harvested using passive lysis buffer for 15 min and assayed by the Dual Luciferase Reporter Assay System (Promega) using GloMax microplate luminometer (Promega). The experiments were replicated three times in two independent experiments.
Lipid accumulation assay. A free fatty acid (FFA) mixture was prepared at 2:1 ratio of oleic acid (OA) to palmitic acid (PA) that mimics benign chronic steatosis with low toxicity described previously 38 . Briefly, 100 mM PA (Sigma) and 100 mM OA (Sigma) stocks were prepared in absolute ethanol at 70°C and filter-sterilized. Five percent (w/v) FFA-free BSA solution was prepared in double-distilled water and filter-sterilized. A 5 mM stock solution for each fatty acid was prepared in 5% BSA solution in distilled water at 37°C then the mixture was stored at 4°C for further experiments. Nile Red (Sigma) is a dark purplish-red powder, the stock solution was prepared in DMSO at 1 mg/ml and then diluted 1:500-1:5000 in PBS for immunofluorescence staining of lipids. Cells were plated on coverslips in sixwell plates and grown overnight. On the following day, cells were treated by conditional medium (DMEM supplemented with 10% FBS containing doxycycline or control solution with or without 10 nM rapamycin) for 36 h and starved for 12 h with FBS-free conditional medium. The starved cells were co-cultured with 200 µM FFA for 24 h to induce steatosis, and the BSA solution was used as control. Cells grown on coverslips were washed with PBS twice and fixed with 1 ml (for each well) of 3.7% formaldehyde/PBS for 15 min at room temperature. After this, cells were washed with PBS twice and stained with Nile-red solution for 30 min (samples were covered with foil to avoid bleaching by room light). Nuclei were counterstained by Hoechst (H33342, Calbiochem). The positive cells were calculated at random at ×400 under the fluorescence microscope and more than 1000 cells were assessed. The ratio of positive cells in every groups was used for statistical analysis. The experiments were replicated three times in three independent experiments.
Insulin resistance assay. Cells were plated in six-well plates and grown overnight. On the following day, cells were treated by conditional medium (DMEM supplemented with 10% FBS-containing doxycycline or control solution with or without 10 M rapamycin) for 24 h and starved for 24 h with FBS-free conditional medium. Then, the starved cells were co-cultured with or without 100 nM insulin (Sigma) for 24 h to induce insulin resistance. In the last 1 h, the cells were incubated with fresh DMEM and treated briefly for 10 min with or without 10 nM insulin. The cells were then harvested, and the lysates were prepared as described in western blot assay (10 µl lysates blotted with indicated antibodies).
Generation of conditional activation of CCRK TG mouse. To overexpress CCRK specifically in adult mouse liver, a liver-specific transferrin promoter (pTf) together with transcription stopping cassette flanked by loxP sites (LSL) were used to control the expression of human CCRK. The 0.6 kb mouse transferrin promoter from C57BL/6 J genome DNA was amplified and subcloned in front of the 1.5 kb human CCRK cDNA in pcDNA3.1. The LSL stopping cassette was amplified from the genomic DNA of Rosa26LacZ mouse, and was inserted between the 0.63 kb transferrin promoter and the human CCRK cDNA. The lineralized DNA carrying pTF-LSL-CCRK was released by PmeI, diluted into 1 ng/µl by TE buffer (10 mM Tris pH 7.5, 0.1 mM EDTA), before being subjected to Pronuclear injection. The expression of CCRK in Tg(pTf-LSL-CCRK) was blocked by the transcription stopping cassette. Removal of the stop cassette by tamoxifen-induced Cre recombinase will turn on the expression of CCRK in a time-specific manner in mouse liver. To achieve this, Rosa26-CreERT2 mice were used to cross with Tg(pTf-LSL-CCRK) to establish the mouse line Rosa26-CreERT2/+; Tg(pTf-LSL-CCRK)/+. Primary hepatocytes were isolated from control and CCRK TG mouse as previous described. Adenovirus coding with Cre was used to infect primary hepatocytes for 4 h and then washout. Cells were collected 48, 96 and 144 h post infection for further analysis.
NASH and NASH-HCC mouse model. The obesity-promoted NASH and HCC models were constructed as described previously 7,30 . For the dietary obesity-NASH model, 6-week-old C57BL/6 male mice were randomly assigned to receive regular CD or high fat high carbohydrate diet (HFHC; Surwit diet) and drinking water enriched with high-fructose corn syrup for 22 weeks. For the dietary obesity-HCC model, C57BL/6 male mice were injected with DEN (25 mg/kg; Sigma-Aldrich) at 14 days of age. After 4 weeks, mice were separated into two dietary groups and fed with CD or HFHC and drinking water enriched with high-fructose corn syrup for 22 weeks. Lentiviruses encoding shRNA against Ccrk (shCcrk) or control sequence (shCtrl) were packaged according to the manufacturer's instructions (Dharmacon) for transduction in the dietary obesity models. At the age of 6 and 18 weeks, 5 × 10 7 transducing units of lentiviruses in 100 ml PBS were administered via tail vein injection as previously described. Intraperitoneal glucose or insulin tolerance test (IPGTT/IPITT) was performed at the age of 26 weeks. All mice were sacrificed at the end of 28 weeks. All animal experimentation ethics approvals had been obtained from the Chinese University of Hong Kong (CUHK) Animal Experimentation Ethics Committee.
Intraperitoneal glucose and insulin tolerance tests. For intraperitoneal glucose tolerance test (IPGTT), 26-week-old mice were transferred to new clean cages and fasted overnight for~16 h. On the following day, 1.5 g glucose per kg body weight was treated intraperitoneally. For intraperitoneal insulin tolerance test (IPITT), 26-week-old mice were transferred to new clean cages and fasted for 6 h. 75 U insulin per kg body weight was treated intraperitoneally. Tail blood was then collected under normal condition at indicated time-points. Glucose levels were measured using one touch blood glucose strips (Johnson and Johnson).
Metabolic profiling. Blood was collected from heart or tail vein, and kept at room temperature for not more than 4 h. The blood was centrifuged at 3000 rpm, 15 min, room temperature, and the serum was transferred to new tubes and stored at −20°C until use. The concentration of serum insulin, triglyceride, and non-essential fatty acid (NEFA) were measured using respective ELISA kits (LabAssay) according to the manufacturer's instructions.
Xenograft mouse models. For the xenograft mouse model, 6-week-old female athymic nude mice were injected with 5 × 10 5 LO2-CCRK-shRaptor and LO2-CCRK-shCtrl cells into the right flanks. 1 week after implantation, the drinking water of doxycycline-treated group mice was replaced by Dox drinking water containing 5% sucrose and 2 mg/ml doxycycline, which was changed every 3 days. The control group mice were treated by sucrose-enriched drinking water (5% sucrose) lacking doxycycline. For the rapamycin treatment, the freshly-made rapamycin (20 mg/ml stock in ethanol) diluted in 0.25% polyethylene glycol plus 0.25% Tween-20 was IP injected three times per week at 6 mg/kg, as previously described. Vehicle in the same volume was injected as control. Tumor length and width were measured using clipper and then tumor volume was calculated using the formula V = (L × W × W)/2. Tumor size was measured every other day using a caliper, and then tumor volume was calculated using the formula V = (L × W × W)/2, with L indicating length and W indicating width. The mice were sacrificed after 1 month for tumor size measurement.
Ki67 immunohistochemistry. Tissue sections (5 mm) from formalin-fixed paraffin-embedded liver tissues were prepared with a microtome (Leica). Then, the liver sections were deparaffinized, rehydrated, and washed in distilled water. Antigen retrieval was done using a pressure cooker with EDTA antigen retrieval buffer (1 mM EDTA, pH 8.0) for 10 min and rinsed by PBS. The endogenous peroxidase activity was then blocked by incubating the slides in 3% hydrogen peroxide in distilled water for 10 min at room temperature and rinsed using PBS. The sections were then blocked using blocking buffer for 30 min at room temperature and stained with polyclonal antibody against Ki-67 (1:100, Labvision) at 4°C for 16 h. The universal HRP Multimer Ultraview Kit on Benchmark XL (Ventana Medical System) was used for chromogen development. For the nuclei staining, the sections were stained with the alum haematoxylin for 2 min, differentiated with 0.3% acid alcohol for several seconds and blue-up with Scott's tap water substitute for 1 min. The Ki67 indexes were assessed by counting the percentages of nucleus-stained (brown) cells/loci per total number of cells. At least 10 fields (magnification ×400) were counted per each liver section and the average indexes were calculated.
Orthotopic HCC mouse model and MDSC expansion analysis. Six-week-old CCRK transgenic mice were injected with tamoxifen (100 mg/kg body weight) daily for 3 consecutive days. At 30-day post tamoxifen injection, 5 × 10 6 Ccrk-depleted Hepa1-6 cells were intrahepatically injected into the liver of male CCRK TG mice. Mice were sacrificed at one month post-tumor injection. Blood, tumor, and matched non-tumor liver tissues were harvested. Single cell was isolated from