The Roles of GSK-3β in Regulation of Retinoid Signaling and Sorafenib Treatment Response in Hepatocellular Carcinoma

Rationale: Glycogen synthase kinase-3β (GSK-3β) plays key roles in metabolism and many cellular processes. It was recently demonstrated that overexpression of GSK-3β can confer tumor growth. However, the expression and function of GSK-3β in hepatocellular carcinoma (HCC) remain largely unexplored. This study is aimed at investigating the role and therapeutic target value of GSK-3β in HCC. Methods: We firstly clarified the expression of GSK-3β in human HCC samples. Given that deviated retinoid signalling is critical for HCC development, we studied whether GSK-3β could be involved in the regulation. Since sorafenib is currently used to treat HCC, the involvement of GSK-3β in sorafenib treatment response was determined. Co-immunoprecipitation, GST pull down, in vitro kinase assay, luciferase reporter and chromatin immunoprecipitation were used to explore the molecular mechanism. The biological readouts were examined with MTT, flow cytometry and animal experiments. Results: We demonstrated that GSK-3β is highly expressed in HCC and associated with shorter overall survival (OS). Overexpression of GSK-3β confers HCC cell colony formation and xenograft tumor growth. Tumor-associated GSK-3β is correlated with reduced expression of retinoic acid receptor-β (RARβ), which is caused by GSK-3β-mediated phosphorylation and heterodimerization abrogation of retinoid X receptor (RXRα) with RARα on RARβ promoter. Overexpression of functional GSK-3β impairs retinoid response and represses sorafenib anti-HCC effect. Inactivation of GSK-3β by tideglusib can potentiate 9-cis-RA enhancement of sorafenib sensitivity (tumor inhibition from 48.3% to 93.4%). Efficient induction of RARβ by tideglusib/9-cis-RA is required for enhanced therapeutic outcome of sorafenib, which effect is greatly inhibited by knocking down RARβ. Conclusions: Our findings demonstrate that GSK-3β is a disruptor of retinoid signalling and a new resistant factor of sorafenib in HCC. Targeting GSK-3β may be a promising strategy for HCC treatment in clinic.

Retinoids are very important for hepatic homeostasis, which effects are mediated by retinoic acid receptors (RARα, RARβ and RARγ) and retinoid X receptors (RXRα, RXRβ and RXRγ) [13]. Deregulated metabolism of retinoids and altered expression of their receptors are implicated in HCC development and progression [14,15]. GSK-3β can inhibit RARα-dependent differentiation of myeloid leukemia [16,17]. Paradoxically, GSK-3β protects RXRα from calpain-mediated truncation in certain solid tumors [18]. Implication of GSK-3β in retinoid signaling and HCC development need further explore. We characterized here RXRα as a direct substrate for GSK-3β. GSK-3β phosphorylates RXRα and impairs its activation of RARβ promoter. Clinically, GSK-3β is overexpressed and associated with RARβ reduction in a majority of HCC. RARβ mediates retinoid action but is frequently silenced during carcinogenesis [14]. Thus, GSK-3β may confer HCC through interfering RARβ-mediated retinoid signalling. This prompted us to further determine whether targeting GSK-3β/RARβ could be of therapeutic significance in HCC.
Sorafenib, a multi-kinase inhibitor, is currently used to treat HCC [19,20]. However, its therapeutic resistance remains a significant problem in clinic [21,22]. Interestingly, sorafenib can stimulate GSK-3β activity in vitro and in vivo. We demonstrated that GSK-3β regulation of RARβ is involved in sorafenib resistance in HCC.

HCC samples
HCC samples (tumors and para-tumor tissues) were collected from The 174 th Hospital affiliated to Xiamen University. The tumors were histologically diagnosed as described [23,24]. All the use of human samples and study protocols were approved by the Hospital Ethics Committee. All patients signed an informed consent form in prior to sample collection. The clinical data were provided in Table S1.

Cell culture and transfection
HepG2 (HB-8065) and HEK293T (CRL-11268) were purchased from ATCC, while SMMC-7721, Bel-7402, and QGY-7703 from Institute of Biochemistry and Cell Biology (SIBS, CAS). All cell lines were obtained between 2008 and 2013 and authenticated by the vendors. The newly received cells were expanded and aliquots of less than 10 passages were stored in liquid nitrogen. All cell lines were kept at low passage, returning to original frozen stocks every 6 months. During the course of this study, cells were thawed and passaged within 2 months in each experiment. QGY-7703 was cultured in RPMI-1640 medium, while other cell lines were grown in Dulbecco's Modified Eagle's Medium. The cultured cells were supplemented with 10% fetal bovine serum. Sub-confluent cells with exponential growth were used throughout the experiments. Transfections were carried out by using Lipofectamine 2000 according to manufacturer's instructions.

Generation of stable lines
GSK-3β stable lines were generated with retroviral vectors. Briefly, HEK293T cells were transfected with PCDH-puro-GSK-3β together with envelope plasmid VsVg (addgene, #8454) and packaging plasmid psPAX2 (addgene, #12260). Retroviral supernatant was harvested at 48 h after initial plasmid transfection and then infected various HCC cell lines. Stable cell pools were selected with 1 μg/ml puromycin (Amresco). The expression efficiency was determined by Western blotting and RT-PCR.

MTT assay and flow cytometry
Cell viability was performed with MTT method as described [15]. For apoptotic analysis, control and treated cells were harvested and washed with precooled PBS twice. The cells were then stained with Annexin V-FITC and propidium iodide (PI) at room temperature for 15 min in the dark. Apoptotic cells were quantitated with flow cytometer analysis (Thermo, Attune NxT).

Dual-luciferase reporter assays
Cells were co-transfected with pGL6-βRARE firefly luciferase reporter constructs, renilla luciferase expression vector (renilla), and ΗΑ-RARα/myc-RXRα in the presence or absence of GSK-3β. The cells were treated with 1 µM 9-cis-RA combined with or without 5 µM tideglusib for 20 h. Cell lysates were measured for luciferase activities. The fluorescence intensity was detected in Multiskan Spectrum (PerkinElmer, USA). The renilla luciferase activity was used to normalize for transfection efficiency.

Co-immunoprecipitation
HCC cells and tumor tissues were lysed and sonicated in 500 µL lysis buffer containing 150 mM NaCl, 100 mM NaF, 50 mM Tris-HCl (pH 7.6), 0.5% NP-40 and 1 mM PMSF. The lysates were incubated with antibodies against endogenous or tags of ectopic proteins and purified with protein A/G beads. For detection of GSK-3β-associated RXRα phosphorylation, HCC tumor lysates were subjected to two rounds of immunoprecipitation (IP). The first round of IP (IP1) was performed with anti-RXRα (D20), which product was then subjected to secondary IP (IP2) with anti-GSK-3β antibody. The lysates or IP2 samples were separated by 10% SDS-PAGE and blotted with anti-p-S/T antibody.

In vitro kinase assays
GFP-RXRα was expressed in and purified from HepG2 cells with immunoprecipitation (IP) using anti-GFP antibody. The cell lysates and IP products were incubated with bacterially purified His-GSK-3β protein in a kinase reaction buffer (pH 7.5, 20 mM Tris-HCl, 10 mM MgCl 2 and 100 mM ATP) at 37°C for 45 min. The reactions were stopped by boiling the samples in loading buffer for 10 min and then separated with 10% SDS-PAGE. GSK-3β-induced RXRα phosphorylation was detected by anti-phospho-ser/thr (p-S/T) antibody.

Animal experiments
Male BALB/c nude mice were injected with HepG2/3β cells (2×10 6 cells) subcutaneously in the posterior flanks and treated with 10 mg/kg sorafenib, 2 mg/kg 9-cis-RA, and 5 mg/kg tideglusib every other day at Day 3 of post-implantation. After three weeks of treatment, the mice were sacrificed. The tumors and various organ tissues were collected for further analysis. Tumor volume was measured twice weekly with a caliper. Tumor samples were immunoblotted with antibodies against RARβ, GSK-3β, p-GSK-3β and GAPDH. Paraffin sections were immunostained using antibodies against Ki-67 and cleaved caspase 3 with DAB Detection Kit (Polymer) (MXB biotechnologies, Fuzhou, China). The expression of p-GSK-3β and RARβ was determined with fluorescent immunostaining. The sections were co-stained with DAPI and detected by Laser Scanning Confocal Microscope (Zeiss). The study protocols were approved by the Institutional Animal Care and Use Committee of University of Xiamen University.

Statistical analysis
Data were represented as mean ± standard deviation (SD) or median ± SEM. The statistical significances of differences were determined using an analysis of variance or Student t test. A P value of <0.05 was considered as significant. All data were acquired in at least three independent experiments.

GSK-3β is overexpressed and associated with RARβ reduction in HCC
To clarify the role of GSK-3β in HCC, we firstly collected HCC samples (n=18) to examine GSK-3β expression. Our results showed that GSK-3β was upregulated in 66.7% of tumors (≥1.5-fold increase) compared to adjacent liver tissues (Fig. 1A, 1B and  Table S1). This was consistent with other reports [25,26]. In most tumors, increased GSK-3β expression remained significant active (with low Ser 9 phosphorylation level) (Fig. 1A). We noted that high GSK-3β was not correlated to downregulation of total and nuclear β-catenin ( Fig. S1A and B), suggesting that the tumor-suppressing effect of GSK-3β via Wnt/β-catenin was lost in HCC. To study whether GSK-3β could confer HCC growth, we overexpressed or knocked down GSK-3β with various HCC cell lines. We showed that overexpression of GSK-3β could strongly promote colony-forming capability of HCC cells, while siRNA-mediated downregulation of GSK-3β resulted in reduced colony formation (Fig.  S1C). The role of GSK-3β in HCC was further strengthened in two HepG2/siβ clones (Fig. S1D). Importantly, GSK-3β-mediated tumor growth and proliferation was confirmed in in vivo experiment (Fig. S1E-G).
Interestingly, tumor-associated GSK-3β was inversely correlated with RARβ expression ( Fig. 1A and C). To further evaluate the possible clinical relevance of GSK-3β/RARβ, we analyzed GEPIA (Gene Expression Profiling Interactive Analysis) database (http://gepia.cancer-pku.cn/). Agreement with our results, GSK-3β was higher and RARβ lower in HCC than adjacent liver tissue (Fig. 1D), which phenomenon was also observed in many other malignant tumors ( Supplementary Fig. S2). Kalpen-Meier survival plot showed that the patients with high GSK-3β had a shorter overall survival (OS) than those with low GSK-3β (Fig. 1E). Together, our results suggest that GSK-3β play a role in RARβ regulation and HCC development.
We thus proceeded to map GSK-3β-mediated phosphorylation site on RXRα. Deletion mutation analysis showed that GSK-3β could phosphorylate RXRα/ΔN20, ΔN40 and ΔN60, which effect was impaired in ΔN80 and lost in ΔN100, indicating that the putative phosphorylation site is located between 60~100 aa (Fig. 4A). To identify the phosphorylation site, we introduced Ala point mutation into these putative sites. Our results showed that RXRα phosphorylation by GSK-3β was kept at S49A and S66A, but abolished in S78A and S78A-containing mutations ( Fig. 4B and C), thus identifying that Ser 78 is the site for phosphorylation by GSK-3β. Since GSK-3β recognizes sequence motif in the context of S/T-X-X-X-S/T, Thr 82 was expected as a priming phosphorylation site. We thus also introduced Asp mutation into Ser 78 (S78D) and Thr 82 (T82D) to mimic their phosphorylation. As a result, GSK-3β-mediated RXRα phosphorylation was abolished in S78A and T82A, but retained in S78D and T82D (Fig. 4D). The upper-shifted band seen in S78D was due to its acidic carboxylic group-contained aspartic acid, which was unrelated to the activity of GSK-3β [28]. Thus, Thr 82 phosphorylation primes RXRα for subsequent phosphorylation of Ser 78 by GSK-3β. This phosphorylation event was confirmed with anti-p-S/T antibody ( Fig. 4E and data not shown).
RXRα could also form dimer with itself. We thus also study the effect of GSK-3β on 9-cis-RA-induced TREpal luciferase reporter activity, which expression was driven by RXRα:RXRα. We showed that overexpression of GSK-3β could strongly inhibit the formation of RXRα homodimer, which could be rescued when GSK-β was inactivated by tideglusib ( Fig. S3F and S3G). Thus, our results demonstrated that GSK-3β could impair the dimeric capacity and transcriptional activity of RXRα.
RARβ and p21, two direct target genes of RXRα [29,30], could be induced by 9-cis-RA. Such induction was inhibited when overexpression of GSK-3β ( Fig. 5F and Fig. S4A). GSK-3β-mediated silence of RARβ and p21 could be relieved when GSK-3β was inactivated by tideglusib or LiCl (Fig. 5G and Fig. S4C). In contrast, 9-cis-RA and LiCl alone only played minor role in modulating the expression of p27, Cyclin D1 and Cyclin B1, all of which do not contain RXRα binding sites on their promoters (Fig. S4C). Interestingly, combined treatment of 9-cis-RA and LiCl could synergistically induce expression of these genes (Fig. S4C).
Biologically, sorafenib resistance was observed in various GSK-3β stable cell lines compared to their vector-transfected counterparts ( Fig. 6F and Fig. S6A). Sorafenib response was reestablished in HepG2/3β when GSK-3β activity was inhibited by tideglusib. The enhancement of sorafenib response regarding its anti-proliferation activity (Fig. 6F) and anti-colony formation (Fig. S6B) were achieved by co-treatment of 9-cis-RA and tideglusib, which combination could effectively induce RARβ expression and PARP cleavage (Fig. 6G). When RARβ was silenced by specific siRNA, the dose-dependent effect of sorafenib on inducing PARP cleavage was inhibited even in the presence of 9-cis-RA/tideglusib (Fig. 6G). We then used flow cytometry to evaluate the synergy of 9-cis-RA/tideglusib on enhancing the apoptotic effect of sorafenib. Sorafenib could alone induce 22.1% apoptotic cell death, which effect was promoted to 50.8% by combining with 9-cis-RA/tideglusib. The enhancement of 9-cis-RA/tideglusib on sorafenib apoptotic response was impaired when silencing RARβ (Fig. 6H, 6I and Fig. S6C). Thus, our results demonstrated that GSK-3β-mediated RARβ inhibition was responsible for sorafenib resistance in HCC cells. HepG2/3β stable cells were transfected with pGL6-βRARE, Renilla, HA-RARα and myc-RXRα. After 24 h transfection, the cells were pretreated with vehicle or with different concentrations of tideglusib (2 µM, 5 µM) for 1 h followed by 1µM 9-cis-RA for 20 h. Luciferase activities were similarly detected. ** p<0.01 (vs respective control); ## p<0.01 (GSK-3β vs mock transfection). (C) HepG2 cells were transfected with vector or Flag-GSK-3β in combination with myc-RXRα and HA-RARα for 36 h. The cells were treated with or without 1 µM 9-cis-RA for 6 h. The lysates were immunoprecipitated with anti-myc tag and blotted with anti-HA and anti-myc antibodies. (D) HepG2/3β cells were transfected with siGSK-3β or scramble (500 pmol in 10 cm dish) for 48 h. The cells were then treated with or without 1 µM 9-cis-RA for 6 h. Co-IP was performed with anti-RARα and blotted with anti-RARα and anti-RXRα antibodies. (E) HepG2/3β cells were pretreated with 5 µM tideglusib for 1 h and then treated with vehicle or 1 µM 9-cis-RA for 6 h. The cell lysates were immunoprecipitated with anti-RARα and blotted with anti-RARα and anti-RXRα (D20) antibodies. For (C), (D), and (E), the inputs were detected with 5% of whole cell lysates. (F)(G) RARβ mRNA expression. HepG2 cells were transfected with vector or Flag-GSK-3β for 24 h (F). HepG2/3β cells were pretreated with 5 µM tideglusib for 1 h (G). Both HepG2 and HepG2/3β cells were treated with 1 µM 9-cis-RA or vehicle for 24 h. RARβ and GAPDH transcripts were detected with RT-PCR. (H) CHIP assays. HepG2 cells were transiently transfected with Flag-GSK-3β, while HepG2/3β stable cells were pretreated with 5 µM tideglusib for 1 h. Both cell lines were treated with vehicle or 1 µM 9-cis-RA for 20 h. The Chromatin DNA was purified and immunoprecipitated with anti-RXRα (D20) antibody or nonspecific IgG. The IPs were subjected to RT-PCR analysis by using specific RARβ promoter primers as indicated in Materials and Methods. (I) Different GSK-3β stable cell lines were pretreated with 5 µM tideglusib and then treated with vehicle or with increasing concentrations of 9-cis-RA for 20 h. The lysates were blotted with anti-RARβ, anti-Flag, anti-Ser 9 GSK-3β and anti-GAPDH antibodies. (J) HepG2 and HepG2/3β cells were pretreated with 5 µM tideglusib and then exposed to vehicle or increasing concentrations of 9-cis-RA for 48 h. HepG2/3β cells were also transfected with RARβ siRNA or scramble siRNA and then subjected to similar treatments. The cell proliferation was detected with MTT. ** p<0.01 (vs respective control).

Targeting GSK-3β enhances the anticancer effect of sorafenib
The significance of GSK-3β/RARβ involved in regulation of sorafenib treatment response was finally determined in vivo. Targeting GSK-3β could significantly inhibit tumor growth of subcutaneous or orthotopical xenografts (Fig. 7A and Fig. S6D). We found that inactivation of GSK-3β by tideglusib could significantly shrink tumor (22.3% inhibition), inhibit Ki-67 expression (14.2%) and induce caspase 3 activation (9.1%), further supporting that overexpression of GSK-3β might be a tumor promoter in HCC (Fig. 7A and B). Sorafenib treatment resulted in tumor inhibition by 48.3%, which effect could be largely enhanced to 93.4% by combining 9-cis-RA/ tideglusib (Fig. 7A). Consistently, sorafenib-induced Ki-67 inhibition and caspase 3 activation were greatly promoted by 9-cis-RA/tideglusib from 28.4% to 50.2% and from 12.4% to 30.3% respectively (Fig. 7B). 9-cis-RA was alone inefficient to improve the anti-tumor effect of sorafenib when GSK-3β remained active (Fig. 7A, B and D). Combination of 9-cis-RA/tideglusib/sorafenib did not affect the mouse weight and change the normal histological characteristics of various tissues including liver, lung, kidney, heart, and spleen, demonstrating that this strategy has less toxic side effect ( Fig. S7A and B). Mechanistically, tideglusib and 9-cis-RA/tideglusib could efficiently inactivate GSK-3β, but only combination could strongly induce RARβ expression ( Fig. 7C-E). Interestingly, RARβ expression by 9-cis-RA/tideglusib was extensively translocated into cytoplasm (Fig. 7C), implying that the nuclear export of RARβ is responsible for apoptotic induction and tumor growth inhibition. The involvement of GSK-3β/RARβ in sorafenib action was summarized in Fig. 7F.

Discussion
Functional GSK-3β is recently demonstrated to confer tumor development and poor prognosis in a wide range of solid tumors [34,35]. It is thus widely attempted to design GSK-3β inhibitor for cancer treatment [36][37][38]. However, the concern is its another important function in suppressing tumor growth [2]. Inhibition of active GSK-3β, whether beneficially or detrimentally, is highly dependent on contextual environment and clinical settings [35,39]. Disclosing the role and mechanism of GSK-3β in tumor will help develop new therapeutic strategy. HCC is the fourth most common tumor worldwide but with very limited treatment options [24,40]. The expression and therapeutic significance of GSK-3β in HCC remain largely unexplored. Although the samples we examined are small, we could consistently demonstrate that GSK-3β is increased in almost every tumor and upregulation of ≥1.5-fold is seen in 66.7% HCC (Fig. 1A, B and E). Increased GSK-3β is closely associated with shorter overall survival (OS) (Fig. 1E). Consistently, it was recently demonstrated that GSK-3β is overexpressed in HCC and targeting GSK-3β can induce degradation of c-FLIPL, a master anti-apoptotic regulator [25]. Our study further showed that overexpression of GSK-3β conferred HCC cell proliferation, colony formation and tumor development, while targeting GSK-3β by tideglusib can significantly induce about 22.3% of growth inhibition in HepG2/3β xenografts (Fig. 7A). Thus, we demonstrated that overexpression of functional GSK-3β supports HCC growth. Since overexpression of GSK-3β renders HCC resistant to certain chemotherapies like retinoid and sorafenib, the therapeutic significance of targeting GSK-3β may lie on its combination with other anticancer drugs.
Hepatocarcinogenesis is closely linked to impaired retinoid metabolism and altered retinoid receptors [15,41,42]. GSK-3β is recently suggested to be a modulator of retinoid signaling as it strongly inhibits RARα-dependent myeloid leukemia differentiation in response to all-trans retinoic acid treatment [16,17]. However, the roles of retinoid receptors in leukemia and solid tumors can be quite different. The therapeutic effects of retinoids are usually less efficacy in solid tumors than leukemia. The mechanism and implication of GSK-3β-mediated impairment of retinoid signaling in solid tumors have not been reported. We demonstrated here that overexpression of GSK-3β can inhibit RARβ expression and impair retinoid signaling in HCC ( Fig.  1 and 5). RARβ expression is required for mediating retinoid action [30], but this protein is frequently down-regulated in HCC with poorly understood mechanism [43]. It was demonstrated that chromatin hypermethylation can impact negatively on RARβ expression [44]. Interestingly, GSK-3β was shown to play a fundamental role in maintaining DNA methylation [45]. There are currently no reports on GSK-3β regulation of RARβ. We demonstrated here that GSK-3β-mediated RARβ inhibition is attributed to its direct inactivation of RXRα (Fig. 5D), suggesting that a functional RXRα is required for RARβ induction. We identified RXRα as a new substrate for GSK-3β. GSK-3β can directly interact with and phosphorylate RXRα at Ser 78 within its N-terminal proline-directed context of S/T-X-X-X-S/T (Fig. 4). Such modification renders RXRα incapable of heterodimerizing with RARα to activate retinoic acid response element on RARβ promoter (Fig. 5). Targeting GSK-3β can recover the function of RXRα (Fig. 5D) and promote retinoid-induced RARβ expression in vitro (Fig. 5G) and in vivo (Fig. 7D). Our results thus disclosed a novel mechanism by which GSK-3β regulates RARβ expression in HCC.
Deregulation of RARβ-mediated retinoid signaling by GSK-3β may at least partially explain why clinical trials of some classical retinoids like β-retinoic acid have no proven benefit in HCC [46]. Interestingly, clinical trial of acyclic retinoid, a synthetic analog of retinoids that target at phosphorylated RXRα, revealed a promising effect in reducing the incidence rate of secondary HCC by about 20% [47,48].
Sorafenib, a multi-kinase targeted anti-cancer drug, is being widely used to treat HCC [19,20] but with significant treatment resistance. We thus asked if GSK-3β-mediated RARβ inhibition could impact on sorafenib treatment response. We found that sorafenib can extensively activate GSK-3β both in vitro (Fig. 6A-D) [32] and in tumor microenvironment (Fig.  7E). Since GSK-3β is highly expressed in HCC, sorafenib treatment will generate abundantly hyperactive GSK-3β. Overexpression of functional GSK-3β strongly inhibits sorafenib action as indicated in various GSK-3β stable liver cell lines vs their vector-transfected counterparts ( Fig. 6F and Fig. S6A). 9-cis-RA cannot alone induce RARβ expression in GSK-3β stable cell lines, in which RARβ is silenced by GSK-3β. Targeting GSK-3β by tideglusib can greatly potentiate 9-cis-RA activation of RARβ-dependent signaling. Importantly, reactivation of RARβdependent signaling that is inhibited by overexpression of GSK-3β returns profoundly unexpected sorafenib treatment outcome (tumor inhibition raised sharply from 48.3% to 93.4%) (Fig.  7A). In this study, we only used low dose of tideglusib in animal experiment by considering that GSK-3β is normally a critical regulator of cell metabolism and homeostasis. In addition, tideglusib has been demonstrated to have fewer side effects under phase II trial in Alzheimer's disease treatment [49,50]. On the other hand, normal tissues are resistant to 9-cis-RA-induced cytoxicity [51]. Combination of tideglusib and 9-cis-RA do not exacerbate deleterious effect of sorafenib in liver and other normal tissues (Fig. S7B).
In summary, our findings suggest that HCC may take advantage of GSK-3β overexpression to support its growth possibly through interfering RARβmediated retinoid signalling. The discovery of GSK-3β/RARβ in sorafenib treatment response may help design improved strategy to overcome the significant treatment resistant problem of sorafenib in clinic.