SUMOylation of RNF146 results in Axin degradation and activation of Wnt/β-catenin signaling to promote the progression of hepatocellular carcinoma

Aberrant SUMOylation contributes to the progression of hepatocellular carcinoma (HCC), yet the molecular mechanisms have not been well elucidated. RING-type E3 ubiquitin ligase RNF146 is a key regulator of the Wnt/β-catenin signaling pathway, which is frequently hyperactivated in HCC. Here, it is identified that RNF146 can be modified by SUMO3. By mutating all lysines in RNF146, we found that K19, K61, K174 and K175 are the major sites for SUMOylation. UBC9/PIAS3/MMS21 and SENP1/2/6 mediated the conjugation and deconjugation of SUMO3, respectively. Furthermore, SUMOylation of RNF146 promoted its nuclear localization, while deSUMOylation induced its cytoplasmic localization. Importantly, SUMOylation promotes the association of RNF146 with Axin to accelerate the ubiquitination and degradation of Axin. Intriguingly, only UBC9/PIAS3 and SENP1 can act at K19/K175 in RNF146 and affect its role in regulating the stability of Axin. In addition, inhibiting RNF146 SUMOylation suppressed the progression of HCC both in vitro and in vivo. And, patients with higher expression of RNF146 and UBC9 have the worst prognosis. Taken together, we conclude that RNF146 SUMOylation at K19/K175 promotes its association with Axin and accelerates Axin degradation, thereby enhancing β-catenin signaling and contributing to cancer progression. Our findings reveal that RNF146 SUMOylation is a potential therapeutic target in HCC.


INTRODUCTION
The Wnt/β-catenin signaling cascade is a highly evolutionarily conserved pathway and plays fundamental roles in embryonic development, differentiation and cellular homeostasis [1]. Dysregulation of the Wnt signaling pathway has been frequently observed and implicated in many cancers especially in hepatocellular carcinoma (HCC) [2][3][4]. Notably, aberrant Wnt/β-catenin pathway occurs in almost all stages of carcinogenesis in a cancer type-specific manner, ranging from tumor initiation and progression to metastasis [5]. The Wnt signaling pathway is tightly and dynamically modulated via a variety of key mediators, such as β-catenin, glycogen synthase kinase-3β (GSK-3β), casein kinase 1 (CK1) and Axin [6]. Specifically, β-catenin plays a central role in the activation of Wnt signaling and its functions are precisely modulated at distinct levels through different mechanisms, including transcription control and post-translational modification. For instance, our previous study has revealed that Vps4A directly mediates the plasma membrane localization of β-catenin to inhibit the transcription activity of β-catenin in HCC [7]. Wang et al revealed that the E3 ligase β-TrCP2 can directly mediate the neddylation of β-catenin and promote its subsequent degradation in a proteasome-dependent manner [8].
It has been well documented that Axin, acting as a scaffolding protein interacts directly with APC, CK1α and GSK-3 to assemble the destruction complex responsible for degradation of the substrate β-catenin [5].Importantly, the cytosolic amount of Axin directly determines the activity of the canonical Wnt pathway, and the abundance and activity of Axin are tightly and precisely modulated by several mechanisms, including ubiquitination, methylation and poly(ADP-ribosyl)ation [9][10][11]. For instance, the HECT-type E3 ubiquitin ligase Smurf1 promotes K29-linked polyubiquitination of Axin and disrupts the interaction of Axin with LRP5/6, leading to attenuation of the Wnt/β-catenin signaling [12]. Tankyrase-mediated PARylation of Axin also contributes to the regulation of Axin protein stability via recruitment of the PARylation-dependent E3 ligase RNF146, subsequently resulting in stabilization of β-catenin and activation of Wnt/β-catenin signaling [13][14][15]. In addition, deubiquitination processes mediated by deubiquitinases (DUBs) have also been well established in the regulation of Axin protein stability [16]. Cong et al. showed that the DUB USP7 can directly bind to Axin and reduce its ubiquitination level and increase its stability, thereby regulating cell differentiation by reducing Wnt/β-catenin signaling [17]. Therefore, the precise modulation of Axin turnover is crucial for controlling the activation of the Wnt pathway, and aberrant decrease in the Axin abundance leads to over-activation of the Wnt/β-catenin signaling pathway and tumorigenesis.
SUMOylation is a highly dynamic and reversible posttranslational modification characterized by covalent conjugation of small ubiquitin-like modifier (SUMO) moieties to substrates at specific lysine residues, and it participates in a variety of cellular processes, including DNA repair and replication, cell cycle transition, cell metabolism and anti-tumor immune responses [18][19][20]. Accumulating evidence has indicated that the SUMOylation pathway is constitutively upregulated and highly activated in multiple cancer types, such as HCC, pancreatic ductal adenocarcinoma and lymphomas [21]. Dysregulation of SUMOylation signaling accelerates tumorigenesis and tumor progression by regulating cell cycle progression-related, angiogenic, and metabolic pathways as well as immune tolerance [22][23][24][25][26]. For instance, we have revealed that targeting SUMOylation could enhance the sensitivity of cancer cells to chemotherapy [18]. Therefore, it is reasonable that targeting the SUMOylation pathway may provide a promising therapeutic strategy for cancer, and several SUMOylation inhibitors have been developed and approved for use in clinical trials. Notably, SUMOylation is also implicated in the modulation of Wnt/β-catenin signaling. For instance, SUMOylation of TBL1-TBLR1 blocked it from interacting with the nuclear receptor corepressor (NCoR) complex and increased the binding affinity of the TBL1-TBLR1-β-catenin complex for the promoter of Wnt downstream genes, thereby leading to activation of the Wnt signaling pathway [27]. In addition, the deSUMOylase SUMOspecific protease SENP7S can recognize both SUMOylated β-catenin and SUMOylated Axin, and can maintain the interaction of these complexes, thereby promoting ubiquitylation-dependent degradation of β-catenin and inhibiting the activation of the Wnt/ β-catenin pathway [28].
RING-type E3 ubiquitin ligase RNF146 is a PARylationdependent E3 ubiquitin ligase and involved in a variety of cellular processes and signal transduction pathways, including TNKS1/2mediated activation of the Wnt/β-catenin and Hippo-YAP pathways, as well as TNF-induced activation of cell death pathways [10,14,29]. Structurally, the E3 ligase RNF146 is composed of two well-characterized domains, the RING domain responsible for transferring ubiquitin moieties to substrates and the WWE domain mediating the recognition of PARylated substrates by RNF146 [14]. Hence, RNF146 plays critical roles in PARP1/2 and TNKS1/2 mediated cellular processes. For instance, RNF146 can recognize PTEN modified by TNKS1/2-mediated PARylation and promote its ubiquitination and degradation in a proteasome-dependent manner, thereby resulting in activation of the PI3K-AKT pathway and cell proliferation [30]. Our previous study also revealed that RNF146 can recognize and interact with BRD7 modified by PARP1induced PARylation and target this protein for degradation through the ubiquitin-proteasome pathway, leading to activation of AKT phosphorylation and resistance of cancer cells to chemotherapy [31]. Moreover, aberrant expression of RNF146 is frequently observed in multiple cancer types, including colorectal cancer and lung cancer [32,33]. For example, the protein level of RNF146 in non-small cell lung cancer (NSCLC) tissues is positively correlated with nuclear expression of β-catenin, and overexpression of RNF146 promotes NSCLC cell proliferation and invasiveness through the classical Wnt/β-catenin pathway, thereby predicting poor prognosis in NSCLC patients [34]. However, the biological function and clinical significance of RNF146 in HCC remain unknown.
In the current study, we found that RNF146, a key E3 ubiquitin ligase modifying Axin, can be SUMOylated by SUMO3, thereby reducing the stability of Axin and activating β-catenin signaling to promote the progression of HCC. Specifically, PIAS3 and SENP1 mediate the SUMOylation and deSUMOylation, respectively, of RNF146 at lysine 19 and lysine 175. Moreover, SUMOylation promotes the association of RNF146 and Axin, leading to increased ubiquitination and degradation of Axin. Our findings emphasized that targeting RNF146 SUMOylation might be a promising therapeutic strategy for HCC.

RESULTS
RNF146 is SUMOylated by SUMO3 at K19, K61 and K174/175 To determine whether RNF146 is subjected to covalent SUMO modification, we transiently transfected SUMO1, SUMO2, or SUMO3 into HeLa cells. As shown in Fig. 1A, RNF146 was modified strongly by SUMO3 but only moderately by SUMO1/2 (Fig. 1A, B and Supplementary Fig. S1A). We also determined that RNF146 could be modified by endogenous SUMO3 (Supplementary Fig.  S1B). It has been reported that UBC9, the sole SUMO-conjugating enzyme for SUMOylation, directly recognizes the SUMO consensus motif (SCM) and selects substrate lysines for modification [35,36]. Thus, we tested whether the SUMOylation of RNF146 can be modulated by UBC9. As expected, depletion of endogenous UBC9 profoundly decreased SUMO3 conjugation to RNF146 (Fig. 1C). Moreover, inhibiting endogenous UBC9 using the inhibitor 2-D08 significantly reduced the SUMOylation level of RNF146 (Fig. 1D). Collectively, these data indicated that RNF146 was efficiently modified by SUMO3 in vivo.
Furthermore, we identified the SUMOylation site(s) in RNF146. There are thirteen lysine (K) residues, K19, K52, K61, K68, K84, K94, K130, K132, K133, K164, K166, K174 and K175, in the RNF146 protein. To identify all potential lysine site(s) at which RNF146 is SUMOylated, a series of plasmids expressing RNF146 with mutation of these lysine (K) residues to arginine (R) mutations were constructed and cotransfected with HA-SUMO3 into HeLa cells. As shown in Fig. 1E, the SUMOylation levels of the K19R, K61R and K174/175R mutants were significantly decreased compared with those of wild-type RNF146. These results strongly supported the idea that the conjugation of SUMO3 to RNF146 occurs mainly at K19, K61 and K174/175. PIAS3 and MMS21 are the dominant SUMO E3 ligases for RNF146 Next, we sought to identify the SUMO E3 ligase responsible for the SUMOylation of RNF146. Previous studies have revealed that protein inhibitor of activated STAT (PIAS) family members, including PIAS1, PIAS2α, PIAS3, and PIAS4, as well as the methylmethanesulfonate-sensitivity protein MMS21/Nse2, are the major "writers" mediating substrate SUMOylation [37,38]. Therefore, constructs expressing PIAS family members and MMS21 were cotransfected individually with the RNF146 plasmid into HeLa cells and co-IP assay was performed. As shown in Fig. 2A, B, PIAS3 and MMS21 but not PIAS1, PIAS2α and PIAS4 specifically associated with RNF146, leading to upregulation of RNF146 SUMOylation. Furthermore, the association between RNF146 with MMS21 or PIAS3 were clearly detected at the endogenous levels (Fig. 2C). In addition, immunofluorescence staining combined with a proximity labeling approach was employed to detect the colocalization of RNF146 with PIAS3 or MMS21 in SK-hep1 and 293T cells [39]. Both PIAS3 and MMS21 colocalized with RNF146. Importantly, PIAS3 and MMS21 promoted the nuclear localization of RNF146 while reducing its cytoplasmic distribution ( Fig. 2D and Supplementary Fig. S2A). Consistent with this finding, depletion of either endogenous PIAS3 or MMS21 greatly inhibited RNF146 SUMOylation (Fig. 2E, F). Double depletion of endogenous PIAS3 and MMS21 dramatically reduced RNF146 SUMOylation (Fig. 2G).
Taken together, these results suggest that PIAS3 and MMS21 are the primary SUMO E3 ligases for RNF146. SENP1/2/6 are the dominant deSUMOylase responsible for removing SUMOylation from RNF146 It has been documented that SUMOylation is a highly dynamic and reversible process and that deSUMOylation is accomplished by SENP family members, namely, SENP1, 2, 3, 5, 6 and 7, in human cells [40,41]. To determine which SENP catalyzes the deSUMOylation of RNF146, we cotransfected SENP constructs individually with the RNF146 plasmid into HeLa cells and performed co-IP assays. As shown in Fig. 3A, SENP1/2/6 but not SENP3/5/7 specifically interacted with RNF146 and greatly reduced the SUMOylation levels of RNF146. Moreover, these interactions were further confirmed by a reciprocal co-IP assay using anti-MYC beads (Fig. 3B). Furthermore, the endogenous complex containing RNF146 and SENP1, SENP2 or SENP6 were also detected by a co-IP using an anti-RNF146 antibody (Fig. 3C). Moreover, TurboID-based proximity labeling revealed that SENP1 promoted the cytoplasmic translocation of RNF146 and was primarily colocalized with RNF146 in the cytoplasm (Fig. 3D). However, both SENP2 and SENP6 enhanced the nuclear accumulation of RNF146 and formed discrete nuclear puncta exhibiting obvious colocalization with RNF146 ( Supplementary Fig. S3A). Consistent with this observation, depletion of endogenous SENP1, SENP2 and SENP6 markedly increased the SUMOylation level of RNF146 ( Fig. 3E and Supplementary Fig. S3B, C). Triple depletion of endogenous SENP1, SENP2, and SENP6 further enhanced the level of RNF146 SUMOylation compared with that in cells with depletion of SENP1, SENP2 or SENP6 alone (Fig. 3F). Taken together, these results indicate that SENP1/2/6 are the bona fide deSUMOylases of RNF146.   (SFP-RNF146) or the SFB-tagged vector were transfected with HA-SUMO1, HA-SUMO2 or HA-SUMO3 for 24 h. The harvested cells were lysed with NETN buffer and subjected to immunoprecipitation (IP) and Western blot with indicated antibodies. B HeLa cells stably expressing SFB-RNF146 or SFB-tagged vector were transfected with HA-SUMO3 for 24 h, and the cell lysates were then subjected to IP using anti-HA beads and detected with the indicated antibodies. The red asterisk indicates the light chain of IgG. C HeLa cells stably overexpressing SFB-RNF146 were transfected with either the scrambled or UBC9 siRNA for 48 h and were then transfected HA-SUMO3 for 24 h. The cells were harvested and subjected to IP using anti-S beads prior to Western blot analysis. D HeLa cells stably expressing SFB-RNF146 or the SFB-tagged vector were transfected with HA-SUMO3 for 24 h and were then treated with 100 μM 2-D08 for 24 h. Cell lysates were then analyzed as indicated. E HeLa cells were cotransfected with the indicated SFB-RNF146 plasmids and HA-SUMO3 for 24 h. Then, the cell lysates were subjected to IP using anti-S beads and western blot with the indicated antibodies.
RNF146 SUMOylation at lysine 19 and lysine 175 promotes the degradation of Axin and upregulation of β-catenin It has been reported that Axin is a scaffolding protein in the β-catenin destruction complex and that its stability is regulated by RNF146 [10,42]. However, the molecular mechanism by which RNF146 regulates Axin degradation remains unknown. Considering the pivotal role of aberrant Wnt/β-catenin signaling in HCC progression, we sought to determine whether SUMOylation modulates the role of RNF146 in regulating Axin degradation and promotes HCC progression. As shown in Fig   The harvested cells were lysed with NETN buffer, and subjected to IP using anti-S beads and Western blot. B HeLa cells stably expressing SFB-RNF146 were transiently cotransfected with HA-SUMO3 and either 3MYC-tagged PIAS3 or 3MYC-tagged MMS21 for 24 h. Then, whole-cell lysates were subjected to IP using anti-MYC beads and detection with the indicated antibodies. The red asterisk indicates the heavy chain of IgG. C HeLa cells were lysed with RIPA buffer, and whole-cell lysates were subjected to co-IP using IgG or an anti-RNF146 antibody and analyzed by Western blot. The red asterisk indicates the light chain of IgG. D SK-hep1 cells were cotransfected with mCherry-RNF146 and V5-turbo-PIAS3/MMS21 for 36 h. After fixation, immunofluorescence staining was performed using the indicated antibody and DAPI; the scale bar indicates 10 μm. HeLa cells stably overexpressing SFB-RNF146 were transfected with HA-SUMO3 for 24 h prior to treatment with scrambled siRNA and either the PIAS3 (E) or MMS21 (F) siRNAs for another 48 h. The cells were harvested and subjected to IP using the indicated antibodies. G HeLa cells stably overexpressing SFB-RNF146 were transfected with scrambled, PIAS3, MMS21 or PIAS3/MMS21 siRNAs for 48 h and lysed with NETN buffer, prior to IP and western blot analysis with the indicated antibodies.

SFB-RNF146
HA-SUMO3 -  endogenous RNF146 significantly increased the protein level of Axin, accompanied by downregulation of β-catenin, and these effects were reversed by reintroducing SFB-RNF146-WT. Interestingly, reintroduction of RNF146-K19R or RNF146-K175R failed to reverse the change caused by knocking out endogenous RNF146, but reintroduction of RNF146-K61R or RNF146-K174R did reverse these changes ( Fig. 4A and Supplementary Fig. S4A). Thus, SUMOylation at K19/K175 but not SUMOylation at K61/K174 contributes to the function of RNF146 in the degradation of Axin.
As β-catenin signaling is negatively regulated by Axin and the nuclear localization of β-catenin determines its activity, we further investigated the effect of RNF146 SUMOylation on β-catenin signaling. We found that overexpressing RNF146-WT but not RNF146-K19R or RNF146-K175R in SK-hep1 cells significantly increased the protein level of β-catenin in the nucleus but only slightly increased the protein level of β-catenin in the cytoplasm (Fig. 4B, C). Then, the enzymes mediating the SUMOylation and deSUMOylation of RNF146 were knocked out. Strikingly, silencing PIAS3 but not MMS21 severely blocked RNF146-induced β-catenin accumulation while upregulating the expression of Axin (Fig. 4D). Additionally, silencing SENP1 but not SENP2 or SENP6 dramatically decreased Axin expression and increased β-catenin expression (Fig. 4E). GSK-3β phosphorylates Ser33, Ser37 and Thr41 in β-catenin to promote its degradation and inhibit Wnt signaling [43]. On the other hand, phosphorylation of β-catenin at Ser675 promotes its interaction with various transcription factors such as TCF4 and TBP [44]. To investigate whether SUMOylation of RNF146 is involved in regulating Axin stability and Wnt signaling in HCC cells, we depleted endogenous UBC9 to impair RNF146 SUMOylation. Consistent with previous observation, depletion of UBC9 decreased the total β-catenin level and blocked β-catenin signaling as shown by the increased phosphorylation of β-catenin at S33/S37/T41 and reduced phosphorylation at S675 (Supplementary Fig. S4B). Importantly, the abolition of RNF146 SUMOylation stabilized Axin by decreasing its ubiquitination ( Supplementary Fig. S4B). Previous studies have shown that SUMOylation alters the stability or protein-protein interactions of the target proteins [24]. We found that mutation of RNF146 had little effect on the total protein level and protein half-life compared with those of RNF146-WT ( Supplementary Fig. S4C)  The expression of the indicated proteins was determined by western blot. B Cytoplasmic and nuclear distribution of β-catenin and SFB-RNF146 in SK-hep1 cells stably expressing SFB-vector or SFB-RNF146. C Endogenous RNF146 was depleted in SK-hep1 cells. The cytoplasmic and nuclear distribution of β-catenin was analyzed after transfection with SFB-vector or the indicated SFB-RNF146 plasmids. D SK-hep1 cells stably expressing SFB-RNF146 were treated with scrambled siRNA or siRNAs for PIAS3 and MMS21 for 72 h. Then, the cells were lysed with RIPA buffer, and the expression of Axin and β-catenin was analyzed by Western blot. E SK-hep1 cells stably expressing SFB-RNF146 were treated with scrambled siRNA or siRNAs for SENP1, SENP2 or SENP6 for 72 h. The expression of Axin and β-catenin were analyzed by Western blot. F SK-hep1 cells with depletion of endogenous RNF146 and stable expression of SFB-vector, wild-type SFB-RNF146 or mutant SFB-RNF146 plasmids were lysed with NETN buffer. Then, whole-cell lysates were subjected to IP using anti-S beads and analyzed by Western blot to evaluate the interaction between RNF146 and Axin. G SK-hep1 cells with depletion of endogenous RNF146 and stable expression of SFB-vector, wild-type SFB-RNF146 or mutant SFB-RNF146 plasmids were transfected with HA-Ub for 24 h. Cell lysates were subjected to co-IP using an anti-Axin antibody, and Western blot was performed using the indicated antibodies. The heavy and light chains of IgG were denoted with red asterisk.
the association of RNF146 with its substrate Axin. Indeed, mutation of either K19 or K175 severely interfered with the RNF146-Axin interaction (Fig. 4F). Moreover, the K19 and K175 mutations of RNF146 failed to mediate the ubiquitination of Axin (Fig. 4G). Additionally, through sequence alignment, we identified a SUMO-interacting motif (SIM) in Axin, which was conserved among multiple species (Supplementary Fig. S4D). Collectively, these results indicate that PAIS3 and SENP1 mediate RNF146 SUMOylation and deSUMOylation at K19 and K175. And, RNF146 SUMOylation promotes its interaction with and degradation of Axin, leading to upregulated β-catenin.

RNF146 SUMOylation activates β-catenin signaling and promotes the proliferation of HCC cells in vitro
As both β-catenin signaling and SUMOylation play an important role in the progression of cancers especially in HCC, we sought to determine whether RNF146 SUMOylation can modulate β-catenin signaling in HCC and contribute to HCC progression. First, RNF146 was overexpressed in SK-hep1 and HCC-LM3 HCC cells. Indeed, RNF146 downregulated Axin and decreased the S33/S37/T41 phosphorylation but increased the S675 phosphorylation of β-catenin, indicating the activation of Wnt signaling (Fig. 5A). Conversely, knocking down or depleting RNF146 with siRNAs or sgRNAs promoted β-catenin phosphorylation at S33/S37/T41 while reducing its phosphorylation at Ser675 (Fig. 5B and Supplementary Fig. S5A). Next, we validated the role of RNF146 in the progression of HCC. The colony formation assay showed that overexpressing or depleting RNF146 significantly promoted or inhibited HCC cell proliferation, respectively (Supplementary Fig. S5B, C). It has been reported that β-catenin activation contributes to the G1-S transition and counteracts the increased level of replication stress in cancer cells [45,46]. Thus, the DNA fiber assay was performed and the results showed that overexpression of RNF146 markedly increased but depletion of RNF146 significantly decreased the replication fork speed in HCC cells (Fig. 5C, D and Supplementary Fig. S5D, E). Taken together, these results suggest that RNF146 promotes the activation of β-catenin signaling and the subsequent progression of HCC. Next, we sought to validate whether the tumor-promoting function of RNF146 depends on its SUMOylation at K19 and K175. To directly determine the role of RNF146 SUMOylation in the progression of HCC, a series of cell lines with stable depletion of endogenous RNF146 and reintroduction of control vector, the wild-type RNF146 plasmid or RNF146 mutant plasmids were generated ( Supplementary Fig. S5F). To directly determine the transcriptional activity of β-catenin, TOP/FOP assay was performed. We found that overexpressing RNF146-WT could significantly enhance the activity of β-catenin. Consistently, knocking out endogenous RNF146 reduced the activity of β-catenin. And, these effects could be completely reversed by reintroducing RNF146-WT but not RNF146-K19R or RNF146-K175R ( Fig. 5E and Supplementary Fig. S5G). In addition, several common target genes of β-catenin including Axin2 and CCND1 were examined by performing RT-qPCR in both HCC-LM3 and SK-hep1 cell lines. As expected, overexpression of RNF146 promoted the expression of AXIN2 and CCND1, while knocking out RNF146 inhibited their transcription. Also, RNF146-WT but not RNF146-K19R or RNF146-K175R could completely rescue these effects ( Supplementary  Fig. S5H, I).
Furthermore, the BrdU incorporation and colony formation assays showed that RNF146 depletion significantly suppressed the proliferation of HCC cells, which was completely rescued by reintroduction of RNF146-WT but not RNF146-K19R or RNF146-K175R (Fig. 5F, G and Supplementary Fig. S5J, K). Moreover, the DNA fiber assay showed that RNF146-K19R and RNF146-K175R failed to rescue the RNF146 depletion-induced replication fork stress in HCC cells (Fig. 5H, I). We found that reducing RNF146 SUMOylation by knocking down PIAS3 significantly inhibited the proliferation of HCC cells. In addition, combined depletion of PIAS3 and RNF146 further augmented the cell growth inhibition caused by knocking down RNF146 alone (Supplementary Fig. S5L). Collectively, these results indicate that SUMOylation of RNF146 at K19 and K175 is essential for its role in the progression of HCC cells.

Abolishing RNF146 SUMOylation inhibits HCC tumorigenesis
To further confirm the function of RNF146 SUMOylation in the progression of HCC in vivo, SK-hep1 cell lines with stable depletion of endogenous RNF146 and reintroduction of wild-type RNF146, RNF146-K19R or RNF146-K175R were used. The results showed that RNF146-WT but not RNF146-K19R or RNF146-K175R significantly promoted xenograft tumor growth ( Fig. 6A-C). In addition, we measured the expression of Axin and β-catenin, and found that RNF146-WT but not RNF146-K19R or RNF146-K175R downregulated the expression of Axin and increased the expression of β-catenin (Fig. 6D). Consistent with these findings, RNF146-WT, but not RNF146-K19R or RNF146-K175R promoted the nuclear localization of β-catenin and cell proliferation as indicated by Ki67 (Fig. 6E). Thus, SUMOylation of RNF146 is vital for its role in HCC progression in vivo. As shown earlier, the SUMOylation inhibitor 2-D08 significantly blocked RNF146 SUMOylation (Fig. 1D), and we then determined whether targeting SUMOylation can inhibit the progression of HCC. The colony formation and EdU imaging assays showed that the proliferative capacity of SK-hep1 and HCC-LM3 cells was decreased by 2-D08 treatment in vitro ( Supplementary Fig. S6A, B). The TOP/FOP assay showed that 2-D08 could significantly reduce the luciferase activity in both HCC-LM3 and SK-hep1 cell lines (Supplementary Fig S6C). In addition, 2-D08 could reduce xenograft tumor growth with no obvious toxicity (Fig. 6F, G and Supplementary Fig. S6D, E). Moreover, the protein level of Axin was increased, while the protein level of β-catenin was decreased in tumor xenografts from mice treated with 2-D08 ( Supplementary Fig. S6F). Also, 2-D08 treatment inhibited nuclear localization of β-catenin and reduced Ki67 positive cells (Supplementary Fig. S6G). Collectively, these results confirm that RNF146 SUMOylation plays a pivotal role in the progression of HCC and could serve as a therapeutic target in HCC.
The clinical significance of RNF146 SUMOylation/Axin/ β-catenin axis in human HCC tissues To determine the clinical significance of RNF146 SUMOylation/ Axin/β-catenin signaling axis, we analyzed the expression of Axin, RNF146 and enzymes mediating RNF146 SUMOylation in tissues of HCC patients. We found that PIAS3 and UBC9 were significantly upregulated, while SENP1 and Axin were obviously downregulated in 12 HCC tissues, compared with paracancerous tissues (Fig. 7A, B and Supplementary Fig. S7A, B). RNF146 was higher expressed in 5/12 HCC tissues, yet no significantly change was examined, consistent with its expression in public database GEPIA2 (http://gepia2.cancer-pku.cn/) and CPTAC (http:// ualcan.path.uab.edu/index.html) (Fig. 7A, B and Supplementary  Fig. S7C). Nonetheless, by performing immunohistochemistry (IHC) analysis in 99 HCC tissues, it was showed that the expression of RNF146 was negatively correlative with that of Axin. Consistently, both UBC9 and PIAS3 were negatively correlative with Axin Fig. 5 RNF146 SUMOylation activates β-catenin signaling and promotes the proliferation of HCC cells in vitro. A SFB-vector and SFB-RNF146 overexpression plasmids were stably transfected into SK-hep1 and HCC-LM3 cells. Axin and β-catenin signaling was analyzed by Western blot using the indicated antibodies. B RNF146 was knocked out with two sgRNAs in SK-hep1 and HCC-LM3 cells. Cell lysates were analyzed by Western blot using the indicated antibodies. SK-hep1 or HCC-LM3 cells with stable overexpression (C) or depletion (D) of RNF146 were sequentially labeled with IdU and CldU for 20 min. Three independent experiments were performed, and the replication fork speed was calculated and analyzed. E The SFB-vector, SFB-RNF146 wild-type or SFB-RNF146 mutant plasmids were stably expressed in SK-hep1 and HCC-LM3 cells in which endogenous RNF146 was depleted by sgRNA. TOP/FOP assay was performed to directly detect the effect of RNF146 on β-Catenin/TCF activity. F The indicated stable cell lines were used for the BrdU staining assay. Three independent experiments were performed, and the percentage of BrdU-positive cells was quantified and analyzed. G Colony formation assays were performed using the indicated stable cell lines. n = 3 independent experiments. SK-hep1 and HCC-LM3 stable cell lines were used, and the DNA fiber assay was performed. Representative images were shown in (H). The replication fork speed was quantified and analyzed as shown in (I). n = 3 independent experiments. n.s. not significant; *p < 0.05; **p < 0.01; ***p < 0.001. expression, whereas SENP1 was positively correlated with Axin expression (Fig. 7C, D). By using the public database GEPIA2 and CPTAC, we found that the mRNA and protein level of UBC9 are significantly overexpressed in HCC specimens and its high expression is correlated with poor prognosis of HCC patients ( Fig. 7E and Supplementary Fig. S7D, E). Also, the expression of UBC9 is positively correlative with the expression of RNF146 and β-catenin target genes such as MMP7, CD44 and MYC (Supplementary Fig. S7F). More importantly, HCC patients with high expression of both RNF146 and UBC9 have the worst prognosis (Fig. 7F). These results imply that SUMOylation might play an important role in HCC progression and RNF146 might be a key downstream target of SUMOylation. Taken together, these data provide strong evidences that SUMOylation cooperates with RNF146 to downregulate the expression of Axin, promote the activation of β-catenin signaling and contribute to the poor prognosis of HCC patients (Fig. 8).

DISCUSSION
In the current study, we found that RNF146 can be modified by SUMO3 and that PIAS3 and SENP1 mediate the SUMOylation and deSUMOylation of RNF146, respectively. In particular, SUMOylation of RNF146 at K19/K175 controls its interaction with Axin. By accelerating the ubiquitination and degradation of Axin, RNF146 SUMOylation activates Wnt/β-catenin signaling and contributes to the progression of HCC. Our study emphasizes that targeting SUMOylation might be a promising therapeutic strategy for cancer.
SUMOylation has been identified in thousands of proteins and is involved in a variety of physiological and pathological processes. However, only a few enzymes are known to mediate the reversible conjugation-deconjugation of SUMO moieties to target proteins [41]. Herein, we found that the E3 ubiquitin ligase RNF146 is predominantly modified by SUMO3, whose conjugation is catalyzed by PIAS3 and MMS21 and whose deconjugation is catalyzed by SENP1/2/6. Of note, the Ψ-K-x-E/D (where Ψ represents a hydrophobic residue) motif is proposed to be the most common consensus motif for SUMOylation on target substrates [47,48]. However, only half of all SUMOylation sites precisely follow this pattern, and SUMOylation can occur at other lysine residues [49,50]. Thus, we mutated all lysines in RNF146 and, consistent with these previous findings, K19, K61, K174 and K175 could be modified by SUMO3, and three of these lysines (K19, K61, K174) did not conform to the above consensus motif. Seven days after injection, the mice were intraperitoneally injected with 2-D08 (5 mg/kg) or vehicle (10% DMSO, 40% PEG300, 5% Tween 80, 45% saline) every 2 days for 16 days. Tumor growth (F) and weight (G) were measured as indicated (n = 8). Statistical analyses were performed using GraphPad Prism. Tumor growth was compared using two-way ANOVA. Statistical significance of the tumor weight between two groups was determined by unpaired t-test (two-tailed). n.s. not significant; *p < 0.05; **p < 0.01; ***p < 0.001.
Previous studies have shown that both E3 ligases and SENPs are substrate specific and cannot compensate for each other [18,41]. Thus, we speculate that distinct enzymes might mediate SUMOylation at different lysines in RNF146 and subsequently play different roles. RNF146 binds directly to and ubiquitinates Axin, thereby accelerating its degradation [10]. Interestingly, we found that only PIAS3 and SENP1 can act at K19/K175 in RNF146 and affect the ubiquitination and stability of Axin, while mutation of K61 or K174 had no effect on the association between RNF146 and Axin. Thus, our findings add a new layer of complexity to the SUMOylation machinery. Notably, few post-translational modifications of RNF146 have been reported. We had reported that K94 of hSSB1 could be modulated by both SUMOylation and acetylation. Importantly, SUMOylation and acetylation of K94 have a synergistic effect on hSSB1 stability [18]. Considering the important role of RNF146 in regulating the stability of multiple substrates, whether more post-translational modification exists on RNF146 or even these SUMOylation sites have other modification and what is the relationship between these modifications remain important issues. Fig. 7 The clinical significance of RNF146 SUMOylation/Axin/β-catenin axis in human HCC tissues. The expression of RNF146, UBC9, PIAS3 and SENP1 in HCC tumor and adjacent specimens (n = 12) was analyzed by western blot using indicated antibodies (A), and their expression were quantified and analyzed as shown in (B). C Representative IHC images of RNF146, Axin, PIAS3, UBC9 and SENP1 staining using HCC tissues. Scale bar indicates 100 μm. D The association between Axin expression and UBC9, RNF146, PIAS3 or SENP1 (n = 99). The percentage of positive staining and p value based on Pearson's χ 2 test and Pearson's correlations are shown in the tables. E UBC9 expression in HCC (n = 369) and non-tumor tissues (n = 50) analyzed by the GEPIA2 web tool. F The correlation between the expression pattern of RNF146 and UBC9 with the overall survival of HCC patients was analyzed using data from TCGA. The high and low grouping of RNF146 and UBC9 was based on the median of the gene expression. n.s. not significant; *p < 0.05; **p < 0.01; ***p < 0.001.
SUMOylation plays a pivotal role in orchestrating the activity, stability, localization, protein-protein interactions and, ultimately, the function of target proteins, thereby mediating a variety of cellular processes. We found that SUMOylation of RNF146 promotes its nuclear localization and association with Axin, thereby influencing the stability of Axin (Figs. 2D and 4F). Conversely, deSUMOylation of RNF146 induces its cytoplasmic localization (Fig. 3D). It has been reported that RNF146 interacts with and ubiquitinates Axin with tankyrase-mediated PARsylation and promotes Wnt signaling by contributing to the degradation of Axin [10]. Here, our study showed that SUMOylation of RNF146 increased its association with Axin and promoted the ubiquitination and degradation of Axin, which in turn inhibited the degradation of β-catenin and promoted Wnt signaling. It has been proposed that SUMO is localized predominantly in the nucleus [41]. Moreover, in addition to its role as a cytoplasmic anchor for β-catenin, Axin was also reported to function as a molecular chaperone to promote nuclear export of β-catenin [51]. Thus, RNF146 SUMOylation mediated degradation of Axin might promote the activity of Wnt/β-catenin signaling by enhancing the stability of β-catenin as well as the nuclear localization of β-catenin. Considering endogenous tankyrase is localized mainly in the nucleus [52], it is conceivable that PIAS3-mediated SUMOylated RNF146 might interact with and ubiquitinate PARsylated Axin in the nucleus (Fig. 8). In addition, K61/K174 in RNF146 could be SUMOylated, and MMS21 together with SENP2/6 could modify the SUMOylation and localization of RNF146. Unlike the diffuse distribution of RNF146 in the nucleus induced by PIAS3, MMS21 was colocalized with RNF146 as scattered puncta throughout the nucleus, implying that MMS21 might SUMOylate RNF146 to play different roles. RNF146 is an E3 ubiquitin ligase and has multiple substrates, such as BRD7 and SH3BP5 [31,53]. As our group and other groups have reported, both BRD7 and MMS21 participate in DNA repair; [54,55] thus, the function of MMS21-mediated RNF146 SUMOylation in DNA damage response and whether this SUMOylation occurs at K61/K174 will be elucidated in the future.
In the present study, we elucidated that SUMOylation of RNF146 plays a crucial role in regulating Wnt/β-catenin signaling and the progression of HCC. Thus, targeting SUMOylation might be a promising strategy for cancer treatment, just as we and others have reported [18,56]. 2-D08 was first reported as an SUMOylation inhibitor which prevents the transfer of SUMO from the UBC9-SUMO thioester to the substrate and has since been widely used in the study of SUMOylation [18,[57][58][59]. In the present study, we found that 2-D08 blocks the SUMOylation of RNF146 and promotes the degradation of β-catenin by stabilizing Axin both in vivo and in vitro (Figs. 1D, 6F, G, and Supplementary Fig. S6). Notably, it was reported that 2-D08 could inhibit the activity of Axl kinase, though the underlying mechanism is elusive [60]. Axl signaling could contribute to stabilizing β-catenin via the Akt/GSK-3β/β-catenin cascade [61]. Therefore, in addition to targeting RNF146 SUMOylation/Axin/β-catenin pathway, it is possible that 2-D08 might exert its role in β-catenin stability by targeting the Axl/Akt/GSK-3β/β-catenin cascade. Additionally, some other kinases could also be inhibited by 2-D08 in biochemical kinase assays in vitro [18]. Thus, more specific inhibitors targeting SUMOylation are needed in cancer treatment.
In summary, our findings reveal that reversible SUMOylation of RNF146 at K19/K175 plays a pivotal role in its localization and functional regulation. Specifically, SUMOylation of RNF146 promotes its association with Axin and subsequently accelerates the degradation of Axin, thereby enhancing β-catenin signaling and contributing to cancer progression (Fig. 8). Therefore, our study uncovers a new layer of regulation upstream of the RNF146-Axinβ-catenin axis and indicates that inhibiting the SUMOylation of RNF146 could serve as a proposing therapeutic strategy for HCC.

MATERIALS AND METHODS Cell culture and transfection
HeLa and HEK293T cells were obtained from ATCC (Manassas, VA, USA). SK-hep1 cells were provided by the Stem Cell Bank of the Chinese Academy of Sciences (Shang Hai, China), and HCC-LM3 cells were a gift from Prof. Peng Li (Sun Yat-Sen University Memorial Hospital, Guangzhou, China). All cells were cultured in DMEM (Thermo Fisher Scientific, USA) containing 10% fetal bovine serum (FBS, LONSERA) in 5% CO 2 at 37°C. The Mycoplasma PCR Detection Kit (Sigma, USA) was routinely employed to exclude mycoplasma contamination. STR profiling was performed for authentication of all cell lines. Plasmid transfection was performed using Viafect (Promega, USA). siRNA transfection was performed using Lipofectamine RNAiMAX (Thermo Fisher Scientific, USA) according to the manufacturer's protocol. The sequences of the indicated siRNAs are shown in Supplementary Table S1.

Antibodies and reagents
The antibodies and reagents used in this paper are listed in Supplementary  Table S2.
amplification with the indicated primers. In addition, cDNAs coding for RNF146, SENP1/2/6, PIAS3 and MMS21 were constructed using Gateway technology as described previously [54] (Invitrogen, USA). The related K-to-R mutants of SFB-RNF146 were cloned into the destination vector using the a TaKaRa MutanBEST Kit (TaKaRa, Japan). To establish the RNF146knockout cell line, RNF146 CRISPR vectors (Santa Cruz Biotechnology, USA) were transiently transfected into SK-hep1 and HCC-LM3 cells, prior to puromycin treatment for 48 h. After selection, the cells were harvested by trypsinization and cultured to obtain clones. Several clones were picked, cultured and validated by Western blot analysis. All constructed plasmids are validated by sequence and listed in Supplementary Table S3.

Tumor xenograft model
This study was approved by the Animal Research Committee of Sun Yat-Sen University Cancer Center and was performed according to the Guide for the Care and Use of Laboratory Animals. Briefly, 8 × 10 7 cells/mL were mixed with equal volume of Matrigel (Corning, USA). The mice were randomly divided into different groups and a total of 2 × 10 6 cells were subcutaneously injected into the right flank of 4-6-week-old male NOD/ SCID mice. Tumor volumes were measured every 3 days and calculated as the formula V = length × width 2 × 0.52. All mice were euthanized and tumors were isolated 21 days after injection. For the 2-D08 treatment assay, 1.5 × 10 6 cells were subcutaneously injected into each mouse. Seven days after injection, 2-D08 (5 mg/kg) or vehicle (10% DMSO, 40% PEG300, 5% Tween 80, 45% saline) was intraperitoneally injected into the mice every other day for 16 days. The researchers conducting the injection were blinded to the grouping and the content of injected material. After being photographed and weighed, the tumor tissues were fixed in 4% paraformaldehyde and analyzed by Immunohistochemical staining or the total proteins of the tumor tissue were extracted and analyzed by immunoblot using indicated antibodies. No animals were excluded from the analyses during the study.

Statistics
All statistical results are shown as the mean ± SDs. Student's t-test or twoway ANOVA or χ 2 tests were performed with GraphPad Prism 8.0 or SPSS 22.0 software. Sample size for each statistical analysis was shown as indicated. Differences with p < 0.05 were considered to be statistically significant (n.s. not significant; *p < 0.05; **p < 0.01; ***p < 0.001). Additional methods are presented in the Supporting Information.