Insulin resistance is a critical factor in the development of metabolic disorders, including type 2 diabetes (T2DM). However, its molecular mechanisms remain incompletely understood. In this study, we found that F-box only protein 2 (FBXO2), a substrate recognition component of the Skp1-Cul1-F-box protein (SCF) E3 ubiquitin ligase complex, was upregulated in livers of obese mice. Furthermore, using a protein purification approach combined with high-performance liquid chromatography/tandem mass spectrometry, we carried out a system-wide screening of FBXO2 substrates, in which the insulin receptor (IR) was identified as a substrate for FBXO2. SCFFBXO2 acts as an E3 ligase targeting the IR for ubiquitin-dependent degradation to regulate insulin signaling integrity. As a result, adenovirus-mediated overexpression of FBXO2 in healthy mice led to hyperglycemia, glucose intolerance, and insulin resistance, whereas ablation of FBXO2 alleviated diabetic phenotypes in obese mice. Therefore, our results identify SCFFBXO2 as an E3 ligase for the IR in the liver, which might provide a novel therapeutic target for treating T2DM and related metabolic disorders.
Type 2 diabetes (T2DM), characterized by high blood glucose concentrations, has become a pandemic problem worldwide. Hyperglycemia is usually caused by an insulin secretion deficiency and/or reduced insulin sensitivity. In peripheral tissues, including liver, skeletal muscle, and adipose tissue, insulin binds to its receptor (IR), which then phosphorylates and recruits IR substrates (IRSs) to further activate downstream signaling pathways (1). In the liver, the major node of insulin signaling is activation of phosphoinositide-3-kinase/AKT, which in turn inhibits the expression of phosphoenolpyruvate carboxykinase (PEPCK) and glucose-6-phosphatase (G6Pase), two key gluconeogenic enzymes (2). As a result, hepatic insulin resistance is characterized by excessive hepatic glucose production, contributing to fasting hyperglycemia in T2DM (3). Therefore, identification of novel molecules involved in regulating the hepatic insulin signaling pathway will advance our understanding of the pathogenesis that leads to T2DM.
Polyubiquitination is the formation of an ubiquitin chain on a single lysine residue on the substrate protein, leading to protein degradation (4). It is carried out by a three-step cascade of ubiquitin transfer reactions—activation, conjugation, and ligation—performed by ubiquitin-activating enzymes (E1s), ubiquitin-conjugating enzymes (E2s), and ubiquitin ligases (E3s), respectively (5). The largest subfamily of E3s in mammals is the Skp1-Cul1-F-box protein ubiquitin ligases, which consist of Skp1, Cul1, Rbx1, and one of the F-box proteins (FBPs) (6). Recent studies have shown that FBPs play a crucial role in many biological events, such as inflammation, cell cycle progression, and tumorigenesis, through ubiquitin-mediated degradation of cellular regulatory proteins (7,8). In addition, their dysregulation has been implicated in several pathologies (6–8), suggesting that insights into Skp1-Cul1-F-box protein ubiquitin ligase–mediated biology may provide potential strategies to treat human diseases. Until now, however, whether FBPs play a role in metabolic diseases, especially insulin resistance and T2DM, remains poorly understood.
Research Design and Methods
Male C57BL/6 and db/db mice aged 8–10 weeks were purchased from the Shanghai Laboratory Animal Company and Nanjing Biomedical Research Institute of Nanjing University, respectively. JNK1 knockout mice were obtained from The Jackson Laboratory and backcrossed to a C57BL/6 background for six generations. All mice were housed at 21± 1°C with humidity of 55% ± 10% and a 12-h light/12-h dark cycle. Mice with high-fat diet (HFD)–induced obesity were maintained with free access to high-fat chow (D12492; Research Diets, Inc) containing 60% kcal from fat, 20% kcal from carbohydrate, and 20% kcal from protein. For the depletion of Kupffer cells, C57BL/6 mice were fed an HFD for 12 weeks and then injected with gadolinium chloride (GdCl3; 10 mg/kg, twice each week) or sodium chloride (NaCl) by tail vein for another 2 weeks. All study protocols comply with guidelines and institutional policies prepared by the Animal Care Committee of Shanghai Jiao Tong University School of Medicine.
Immuoprecipitation and In-Solution Digestion
The standard immunoprecipitation (IP) purification procedure has been previously described (9). In brief, HEK293T cells stably expressing Flag-tagged wild-type (WT) or mutant (MUT) F-box only protein 2 (FBXO2) were lysed in 5 mL lysis buffer (50 mmol/L Tris-HCl [pH 7.5], 150 mmol/L NaCl, 0.5% Nonidet P40, and 100 mmol/L phenylmethylsulfonyl fluoride) with gentle rocking at 4°C for 20 min. Lysates were cleared and subjected to IP with 50 μL of anti-FLAG M2 beads overnight at 4°C. Beads containing immune complexes were washed with 1 mL ice-cold lysis buffer. Proteins were eluted with 100 μL 3X FLAG peptide (Sigma-Aldrich, St. Louis, MO) in Tris-buffered saline for 30 min and precipitated with cold acetone. The precipitated proteins were digested in solution with trypsin, and the tryptic peptides were centrifuged in a vacuum to dryness for further analysis.
High-Performance Liquid Chromatography/Tandem Mass Spectrometry Analysis
Nanoflow liquid chromatography/tandem mass spectrometry (MS) was performed by coupling an Easy nLC 1000 liquid chromatograph (Thermo Fisher Scientific, Waltham, MA) to an Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific). Tryptic peptides were dissolved in 20 µL of 0.1% formic acid, and 10 µL were injected for each analysis. Peptides were delivered to a trap column (2-cm length with a 100-µm inner diameter, packed with 5 µm C18 resin) at a flow rate of 5 µL/min in 100% buffer A (0.1% formic acid in high-performance liquid chromatography–grade water). After 10 min of loading and washing, the peptides were transferred to an analytical column (17 cm × 79 μm, 3-μm particle size; Dikma Co, Beijing, China) coupled to an Easy nLC 1000 system (Thermo Fisher Scientific). The separated peptides were ionized using a nanospray ionization source, then analyzed in an Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific) with a top speed 3s data-dependent mode. For MS/MS scanning, ions with an intensity above 5,000 and charge states 2–6 in each full MS spectrum were sequentially fragmented by higher collision dissociation, with normalized collision energy of 32%. The dynamic exclusion duration was set at 60 s, and the precursor ions were isolated by quadrupole with a 1-Da isolation window. The fragment ions were analyzed in the ion trap with automatic gain control 7,000 at rapid scan mode. The raw spectra data were processed by Thermo Proteome Discoverer 2.1 and MS/MS spectra data were searched against the Uniprot human database (88,817 sequences) by Mascot (v.2.4; Matrix Science, London, U.K.).
Glucose and Insulin Tolerance Tests
Glucose tolerance tests (GTTs) were performed by intraperitoneal injection of d-glucose (Sigma-Aldrich) at a dose of 2.0 mg/g body weight after a 16-h fast. For insulin tolerance tests (ITTs), mice were injected with regular human insulin (Eli Lilly & Company, Indianapolis, IN) at a dose of 0.75 U/kg body weight after a 6-h fast. Blood glucose was measured using a portable blood glucose meter (LifeScan; Johnson & Johnson, New Brunswick, NJ).
Hepatic tissues or cells were lysed in radioimmunoprecipitation buffer containing 50 mmol/L Tris-HCl, 150 mmol/L NaCl, 5 mmol/L MgCl2, 2 mmol/L EDTA, 1 mmol/L NaF, 1% NP-40, and 0.1% SDS. Western blotting was performed using antibodies against FBXO2 (ab133717; Abcam), IRβ (ab131238; Abcam), AKT (13038, 4821; Cell Signaling Technologies), and GAPDH (5174; Cell Signaling Technologies). Tyrosine phosphorylation of IRS1 was analyzed by IP of IRS1 with anti-IRS1 from total lysate, followed by Western blotting with anti-pTyr antibody (PY100).
Luciferase Reporter and Chromatin IP Assays
All the transient transfections were conducted using Lipofectamine 2000 (Invitrogen, Shanghai, China). The FBXO2 promoter was amplified from the mouse genomic DNA templates and inserted into pGL4.15 empty vector (Promega). Luciferase activity was measured using the Dual-Luciferase Reporter Assay System (Promega). For chromatin IP (ChIP) assays, a commercial kit was used (Upstate, Billerica, MA). In short, mouse primary hepatocytes (MPHs) were fixed with formaldehyde, and chromatin was incubated and precipitated with antibodies against p65 (ab16502; Abcam) or control IgG (ab172730; Abcam). DNA fragments were subjected to real-time PCR using primers flanking the nuclear factor (NF)-κB binding site in the FBXO2 promoter. The primer sequences were 5′-ACCAGCGCGACGCGG TATGGGA-3′ (forward) and 5′-TGGGGCAGCCGGACTAAAAGCT-3′ (reverse).
Values are shown as mean ± SEM. Statistical differences were determined using the Student t test. Statistical significance is considered at P < 0.05, P < 0.01, or P < 0.001.
Upregulation of FBXO2 in Livers of Obese Mice
To identify genes that are differentially expressed in obesity, we previously performed a clustering analysis of Affymetrix arrays, which showed that a large number of mRNAs were markedly dysregulated in the liver of mice fed an HFD compared with mice fed a normal chow diet (12,13). Here we describe work on the FBPs. More than 70 FBPs are present in mammals (6). Our data showed that 11 FBPs were significantly changed (P < 0.05), of which 8 were increased and 3 were decreased (Supplementary Table 1). Here, FBXO2 was chosen for further experiments because its expression was enriched in the liver and hepatocytes (Supplementary Fig. 1A and B). By contrast, its expression in other tissues, including skeletal muscle, white adipose tissue, heart, and kidney, was relatively low (Supplementary Fig. 1A). Increased mRNA and protein expression of FBXO2 in HFD-fed mice was further confirmed by quantitative real-time PCR and Western blotting, respectively (Fig. 1A and B). Upregulation of FBXO2 was also detected in the livers of db/db mice (Fig. 1C and D), a well-established genetic model of T2DM, suggesting that abnormal expression of FBXO2 represents a typical feature of insulin resistance in obese animals.
Identification of the IR as a Novel Substrate for FBXO2
FBXO2 was shown to preferentially target N-linked high-mannose oligosaccharides in glycoproteins for ubiquitination and degradation (14). The F-box–associated domain of FBXO2 is essential for its activity of recognizing glycoprotein, which is completely abolished by mutations of two residues (15,16). To systematically identify the FBXO2-interacting proteins, HEK293T cells were transfected with retroviruses expressing Flag-tagged WT FBXO2 or an F-box–associated domain mutant (MUT), which could not recognize glycoprotein, as previously described (15,16). IP against Flag was subsequently performed with the lysates of cells carrying WT or MUT FBXO2 proteins, respectively. As depicted in Supplementary Fig. 2, all purification procedures were monitored by Coomassie Brilliant Blue staining as well as Western blotting with anti-Flag antibody, showing that both WT and MUT FBXO2 proteins were highly enriched in the final elution fraction. Consistent with previous results (16), concanavalin positivity signals accumulated dramatically in the WT, but not MUT, final elution fraction. The final immunoprecipitates from WT and MUT cells were further subjected to MS analysis. Proteins were identified using Mascot software, and identified proteins filtered with an overall false discovery rate <0.01% were considered as potential interacting candidates. Using these criteria, we finally identified 2,643 proteins from WT samples and 1,138 proteins from MUT samples (Supplementary Table 2). To exclude the unspecific binding, we then focused on the proteins that were exclusively identified in WT cells, resulting in 1,569 potential substrates. Importantly, through comparison with the Uniprot database, we found that more than one-third of these proteins (528, or 33.7%) were glycoproteins. By contrast, only 82 proteins (7.6%) from MUT elutes were classified as glycoproteins in the Uniprot database (Fig. 2A). Together, our data indicated that glycoproteins were significantly enriched among the proteins interacted with WT but not MUT FBXO2. Interestingly, the Kyoto Encyclopedia of Genes and Genomes pathway showed that a portion of these glycoproteins was involved in N-glycan biosynthesis and oxidative phosphorylation, suggesting a potential role for FBXO2 in energy metabolism (Fig. 2B and Supplementary Table 3). Bioinformatics analysis further showed that these glycoproteins were highly enriched in membrane, endoplasmic reticulum (ER), and lysosome (Fig. 2C and Supplementary Table 3). Given the relevance of FBXO2 in obese animals, we questioned whether any molecules involved in the insulin signaling pathway are potential substrates of FBXO2. Intriguingly, we found that IR, a large transmembrane glycoprotein containing multiple N-linked glycosylation sites (17,18), was coeluted with only WT FBXO2, and not MUT FBXO2, in two replicates (Fig. 2D).
FBXO2 Negatively Regulates the Stability of the IR
Next, we confirmed the specific interaction between FBXO2 and the IR in transiently transfected HEK293T cells using coimmunoprecipitation (Fig. 3A). The endogenous interaction of these two proteins was also detected in MPHs (Fig. 3B). Because FBXO2 could interact with the IR, we tested whether FBXO2 could regulate IR stability or accelerate its protein degradation. Indeed, endogenous IR protein contents were dramatically decreased in MPHs transfected with adenovirus expressing FBXO2 (Fig. 3C), whereas its mRNA levels remained unchanged (Fig. 3D). In addition, abundance of IRS1, IRS2, Glut1, and Glut4 proteins were not affected by FBXO2 overexpression (Fig. 3C). The IGF-I receptor, which is closely related to the IR and has overlapping functions, was slightly reduced, suggesting the specificity of FBXO2-induced IR degradation (Fig. 3C). The ubiquitination of IR was also increased by ectopic expression of FBXO2 in MPHs treated with MG132, a proteasome inhibitor (Fig. 3E). Furthermore, overexpression of FBXO2 reduced the half-life of IR to less than 2 h (Fig. 3F), supporting the notion that FBXO2 could regulate IR stability and promote its degradation. In agreement, posttranscriptional downregulation of hepatic IR was also observed in obese mice (Supplementary Fig. 3A–D).
Moreover, insulin inhibited dexamethasone/foskolin-induced glucose production, which was largely attenuated by the overexpression of FBXO2 (Fig. 4A). In agreement with this, FBXO2 expression also blocked the suppressive effects of insulin on dexamethasone/foskolin-induced expression of gluconeogenic enzymes (PEPCK and G6Pase) (Fig. 4B). In addition, FBXO2-induced downregulation of IR protein was attenuated by MG132 (a proteasome inhibitor), but not leupeptin (an inhibitor of lysosomal protease) (Fig. 4C). MG132 treatment also restored insulin-suppressed glucose production and gluconeogenic gene expression (Fig. 4D and E), indicating the involvement of the proteasome system in FBXO2-mediated inhibition of insulin signaling.
Liver-Specific Overexpression of FBXO2 Promotes Hyperglycemia and Insulin Resistance
To investigate the role of FBXO2 in regulating insulin signaling in vivo, FBXO2 or green fluorescent protein adenovirus was administered to C57BL/6 mice via a tail vein injection. As shown in Fig. 5A, the level of FBXO2 protein was dramatically increased while IR was decreased in the liver, but not in other tissues, including white adipose tissues and skeletal muscles (data not shown). Overexpression of hepatic FBXO2 did not affect body weight or food intake (Supplementary Fig. 4A and B), but significantly increased circulating concentrations of glucose and insulin, indicating insulin resistance (Fig. 5B and C). A dramatic reduction in insulin sensitivity was also revealed by GTTs and ITTs (Fig. 5D). These changes were accompanied at a molecular level by phosphorylation of IRS1 and AKT, two crucial molecules in the insulin signaling pathway, in response to acute intraperitoneal insulin injection (Fig. 5E). Moreover, the mRNA expression of PEPCK and G6Pase was upregulated by FBXO2 overexpression (Fig. 5F).
Ablation of FBXO2 Enhances Insulin Sensitivity in db/db Mice
To further confirm the effects of FBXO2 in an independent setting, we disrupted its expression in the liver of db/db mice by delivering adenovirus-expressing FBXO2-specific short hairpin RNA (shRNA) or a nonspecific control shRNA. FBXO2 shRNA treatment significantly reduced hepatic FBXO2 protein levels and increased IR protein expression compared with negative control shRNA-injected littermates (Fig. 6A). As a result, loss of FBXO2 dramatically improved hyperglycemia, hyperinsulinemia, glucose tolerance, and insulin resistance (Fig. 6B–D). Well-improved insulin signaling and downregulation of gluconeogenic enzymes were also observed in db/db mice with FBXO2 deficiency (Fig. 6E and F). Similar effects on glucose homeostasis were observed in mice with HFD-induced obesity that were transduced with FBXO2 shRNA (Supplementary Fig. 5A–D), suggesting that knockdown of FBXO2 in the liver could alleviate the diabetic phenotype in obese mice.
Regulation of Hepatic FBXO2 in Obesity
The results described above demonstrate that FBXO2 was upregulated in obese livers, and manipulation of FBXO2 could modulate insulin sensitivity. Finally, we sought to determine the signaling pathway that regulates FBXO2 expression. T2DM is tightly associated with high circulating concentrations of glucose, fatty acids, insulin, and proinflammatory cytokines. Therefore we performed a screen to assess whether these cellular factors and hormones could affect FBXO2 expression. As a result, tumor necrosis factor-α (TNF-α) and interleukin (IL)-1β, but not high glucose, fatty acids, insulin, or dexamethasone, induced FBXO2 expression in MPHs (Fig. 7A and Supplementary Fig. 6A–D), suggesting that inflammation might be responsible for the upregulation of FBXO2 in obese mice. To confirm this point, we deleted Kupffer cells in HFD-fed mice by administering GdCl3 (19). Consistent with previous reports that Kupffer cells are the primary source for hepatic inflammation in obesity (19–21), GdCl3 treatment significantly reduced the expression of proinflammatory markers including TNF-α, IL-1β, and F4/80 in liver tissues (Supplementary Fig. 6E). Under this condition, there was a marked decrease in FBXO2 expression in the livers of obese mice compared with controls (Fig. 7B).
Growing evidence has noted the roles of inflammation-mediated c-Jun N-terminal kinase (JNK) 1 and IKKβ/NF-κB signaling pathways on the regulation of liver metabolic homeostasis (20,22–24). Hence, it is interesting to determine whether JNK1 and/or IKKβ/NF-κB activation may underlie the upregulation of FBXO2. As shown in Supplementary Fig. 7A, FBXO2 mRNA expression showed similar changes after TNF-α treatment in JNK1 knockout MPHs compared with JNK1 WT MPHs, suggesting that JNK1 might not be essential for the regulation of FBXO2 expression. In agreement with this, the induction of FBXO2 was largely blocked by BAY 11–7082 (an NF-κB inhibitor), but not SP600125 (a JNK inhibitor) or U0126 (an extracellular signal–regulated kinase inhibitor) (Supplementary Fig. 7B), suggesting that the canonical IKKβ/NF-κB pathway mediates the effects of proinflammatory cytokines to induce FBXO2 expression.
Next, we speculated that FBXO2 is a molecular target of IKKβ/NF-κB. To confirm this, we examined the promoter region of FBXO2 and found that a canonical NF-κB DNA-binding motif (5′-GGGRNNYYCC-3′) exists in the proximal promoter region of the FBXO2 gene (Fig. 7C). We then created luciferase plasmids controlled by the FBXO2 promoter and found that IKKβ increased the transcriptional activities of these promoters when transfected into HEK293T cells (Fig. 7D). On the other hand, mutagenesis of the NF-κB DNA-binding motif abrogated the effect of IKKβ/NF-κB in activating the transcriptional activities of these promoters (Fig. 7D). Similarly, inhibition of NF-κB activation by BAY 11–7082 abolished the TNF-α–induced activity of the FBXO2 promoter (Fig. 7E), further suggesting that hepatic inflammation regulates FBXO2 through NF-κB signaling. The association of p65 with the FBXO2 promoter was also confirmed by ChIP assays (Fig. 7F). Considering these data together, we speculate that chronic hepatic inflammation–mediated IKKβ/NF-κB activation may be an important mechanism leading to upregulation of FBXO2 in obesity.
Previous studies have created mice, via the Cre-loxP system, with tissue-specific disruption of the IR gene. Intriguingly, hyperglycemia and insulin resistance were only exhibited in liver-specific IR knockout mice and not in skeletal muscle– or fat-specific IR knockout mice (25–27), suggesting that hepatic IR has a critical role in regulating glucose homeostasis and insulin sensitivity. Although downstream signaling pathways of insulin are well established, molecular determinants that directly regulate IR expression remain poorly elucidated. In this study we provide in vitro and in vivo evidence showing a critical role of FBXO2 as a posttranscriptional regulator of hepatic insulin signaling. First, a protein purification approach combined with the high-performance liquid chromatography/tandem MS assay was used to identify IR as a novel interacting protein of FBXO2, which was further confirmed by coimmunoprecipitation assays. FBXO2 interacts with the IR to enhance its ubiquitination-mediated protein degradation. Second, the physiological role of FBXO2 is further revealed by both gain-of-function and loss-of-function studies of mice. Overexpression of FBXO2 in the liver led to hyperglycemia, hyperinsulinemia, glucose intolerance, and insulin resistance in healthy mice, whereas selective knockdown of FBXO2 in obese mice improved these symptoms. Third, FBXO2 was upregulated in obese livers, suggesting that inhibiting the expression or activity of FBXO2 might represent a potential therapeutic target for enhancing insulin sensitivity.
Several studies since the 1970s have reported the abnormal number and function of IRs in various tissues of insulin-resistant mice, including liver, adipose tissue, skeletal muscle, leukocytes, and endothelial cells, whereas its mRNA levels were not found to be decreased (28–31). These results suggest that the small number of receptors could be due to posttranscriptional levels. Indeed, it has been shown that IR protein expression could be targeted and inhibited by several microRNAs in adipocytes, heart, and liver (32–34). Song et al. (35) demonstrated that IR is ubiquitinated by Mitsugumin 53 (MG53) in skeletal muscle because IR ubiquitination and insulin-elicited downstream signaling are inversely changed in MG53 transgenic mice and MG53 knockout mice. A recent study identified nuclear ubiquitous casein and cyclin-dependent kinase substrate as regulators of IR expression, thereby regulating energy homeostasis and glucose metabolism (36). Therefore, along with these studies, molecular interventions that selectively increase IR expression might provide an attractive avenue to treat T2DM. Although both we and another group (35) found that proteasome inhibitor administration could efficiently prevent the degradation of IR by different E3 ligases, how IR gets into the proteasome for degradation remains unclear. Moreover, our bioinformatics analyses showed that the glycoproteins that interacted exclusively with WT FBXO2 were highly enriched in the membrane, ER, and lysosome, suggesting other membrane glycoproteins might also be ubiquitinated by FBXO2. Membrane proteins are subject to a complex series of sorting, trafficking, quality control, and quality maintenance systems, which are largely controlled by ubiquitination (37). Retrotranslocation of misfolded membrane proteins from the ER into the cytoplasm and processive cleavage by the 26S proteasome also participate in ubiquitination-mediated degradation (38). Interestingly, it has been reported that FBXO2 ubiquitinates N-glycosylated proteins that are translocated from the ER to the cytosol and functions in an ER-associated degradation pathway (14). Therefore, the degradation of IR might take place in the ER via retrotranslocation, which needs to be determined in future studies.
Our data also indicate that aberrant expression of FBXO2 is attributed, at least in part, to the activation of IKKβ/NF-κB by proinflammatory factors. Numerous studies have demonstrated that low-grade and chronic inflammation plays a positive role in the glucose intolerance and insulin resistance seen in obesity (39). While several potential mechanisms have been proposed (39), our results may provide a novel insight whereby inflammation inhibits the hepatic actions of insulin. In addition, whether FBXO2 expression could be regulated by other factors such as ER stress and autophagy remains to be determined.
To our knowledge, we have for the first time identified FBXO2 as a functional E3 ligase for IR in the liver. Several recent reports showed that FBXO2 plays an important role in the brain by controlling the abundance of the N-methyl-D-aspartate receptor and amyloid precursor protein (40,41). However, its role in other biological events remains largely unexplored. Therefore, future studies directed at understanding its tissue-specific downstream targets are needed.
Acknowledgments. The authors are grateful to Xiaoying Li from Zhongshan Hospital, Fudan University, Shanghai, for helpful discussion of the manuscript.
Funding. This study was supported by grants from the National Natural Science Foundation of China (grant nos. 81402478, 31401185, and 81570769), the Shanghai Rising-Star Program (grant no. 16QA1402900), and the Research Foundation of Hubei Polytechnic University for Talented Scholars (grant no. 9666).
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. B.L., H.L., and L.G. performed animal and cellular experiments and analyzed the data. D.L. and X.X. provided technical advice on the cellular studies. Z.W. and Y.L. conceived the research ideas, supervised the project, and wrote the manuscript. Y.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.