The Adaptor Protein APPL2 Inhibits Insulin-Stimulated Glucose Uptake by Interacting With TBC1D1 in Skeletal Muscle

Insulin maintains glucose homeostasis by facilitating the uptake of postprandial glucose into adipose tissues and skeletal muscle, the latter of which accounts for ∼75% of total glucose disposal (1). The binding of insulin to its receptors elicits tyrosine phosphorylation of insulin receptor substrates, which in turn activates phosphatidylinositol 3-kinase (PI3K), leading to the membrane recruitment and activation of Akt. Activated Akt subsequently induces the translocation of GLUT4 from intracellular vesicular compartments to the plasma membrane for glucose uptake. In type 2 diabetic patients, the ability of insulin to stimulate glucose uptake is significantly impaired, owing in part to the defective insulin-dependent recruitment of GLUT4 to the plasma membrane (2,3). In rodents, genetic ablation of GLUT4 in skeletal muscle causes insulin resistance and diabetes (4), whereas overexpression of GLUT4 alleviates hyperglycemia and insulin resistance in db/db diabetic mice (5). GLUT4 translocation is tightly controlled by insulin signaling cascades and a series of small GTPase proteins (6). Studies have demonstrated that the Rab-GTPase–activating proteins (GAPs) Tre-2/Bub2/Cdc16 domain family, member 1 (TBC1D1) and member 4 (TBC1D4 [also known as AS160]) may integrate insulin signaling and Rab-GTPase activity, thereby regulating GLUT4 trafficking and glucose uptake (6). However, the detailed intracellular signaling events that confer insulin-elicited glucose uptake are still not fully characterized.

Mounting evidence suggests that the adaptor protein APPL1, which comprises an NH₂-terminal Bin/amphiphysin/Rvs (BAR) domain, a central pleckstrin homology (PH) domain, and a COOH-terminal phosphotyrosine-binding (PTB) domain, is an insulin-sensitizing molecule in multiple insulin-responsive tissues (7). Genetic disruption of APPL1 causes insulin resistance and defective glucose-stimulated insulin secretion, leading to glucose intolerance in mice (8). In contrast, transgenic overexpression of APPL1 prevents obesity-induced deleterious effects on glucose homeostasis and endothelial and cardiac functions (8–10). Hepatic overexpression of APPL1 improves hyperglycemia, glucose intolerance, and insulin sensitivity in db/db diabetic mice, whereas hepatic silencing of APPL1 causes insulin resistance and hyperglycemia in lean mice (11). In pancreatic β-cells, APPL1 promotes glucose-stimulated insulin secretion by upregulating the expression of soluble N-ethylmaleimide-sensitive factor attachment protein receptor proteins in an Akt-dependent manner (8). At the molecular level, the interaction between APPL1 and Akt prevents Akt from binding to the pseudokinase tribble 3 (TRB3), thereby promoting Akt to the plasma membrane for further activation in hepatocytes, endothelial cells, and pancreatic β-cells (8,9,11,12). Furthermore, APPL1 potentiates the insulin-sensitizing effects of adiponectin on promotion of glucose uptake by direct interaction with the two adiponectin receptors (13).

APPL2 is a close homolog of APPL1 abundantly expressed in skeletal muscle. These two adaptor proteins share 52% identity and 72% similarity in amino acid sequence and the same domain organization (7). APPL1 and APPL2 appear to play a similar role in mediating growth factor–induced cell proliferation in fibroblasts and apoptosis in zebrafish (14,15). On the other hand, an in vitro study demonstrated that these two proteins possess distinct or even opposite functions in regulating glucose and lipid metabolism (16). Structural analysis revealed that APPL2 incorporates two homodimers, whereas APPL1 incorporates only one homodimer in the asymmetric unit (17–19). Although APPL1 binds to the adiponectin receptors and increases adiponectin-induced glucose uptake and fatty acid oxidation, APPL2 inhibits adiponectin actions in muscle cells (13,16).

Although the insulin-sensitizing effects of APPL1 are well characterized, little is known about the physiological role of APPL2 in insulin signaling and glucose metabolism. In this study, we provide both in vitro and in vivo evidence that APPL2 is a negative regulator of insulin-stimulated glucose transport in skeletal muscle. Furthermore, we identify TBC1D1 as an interacting partner and a downstream effector that mediates the suppressive effect of APPL2 on insulin-elicited plasma membrane translocation of GLUT4.

Rabbit monoclonal antibodies against total Akt, GAPDH, β-actin and insulin receptor-β (IRβ), rabbit polyclonal antibody against TBC1D1 (#5929), and mouse monoclonal antibodies against phosphotyrosine and GLUT4 were from Cell Signaling Technology. Rabbit anti-HA polyclonal and mouse monoclonal antibodies against FLAG and c-Myc were from Sigma, and a rabbit polyclonal antibody against TBC1D1 (ab56191) was from Abcam. Rabbit polyclonal antibodies against APPL1, APPL2, phospho-TBC1D1 (serine [Ser] 235 and threonine [Thr] 596), phospho-Akt (Ser-473), and recombinant proteins of full-length and truncated APPL1 and APPL2 were obtained from Antibody and Immunoassay Services, The University of Hong Kong (HKU). Human recombinant insulin was from Novo Nordisk. The PI3K inhibitor LY294002 and cytochalasin B were from Sigma, and 2-deoxy-[³H]-glucose and [¹⁴C]-mannitol were from PerkinElmer.

To generate APPL2 transgenic (APPL2 Tg) mice, human APPL2 cDNA was cloned into a transgenic vector under the control of cytomegalovirus immediate early β-actin promoter (9). The transgenic mice were generated as described previously (9) and were screened by PCR analysis with genotyping primers as listed in Supplementary Table 1.

APPL2 knockout (KO) mice were generated by Shanghai Nanfang Center for Model Organisms. The targeting construct containing loxP sites flanking exon 5 of the APPL2 gene and the FRT-flanked selection cassette upstream of the loxP sites was electroporated into embryonic stem cells, followed by selection of positive embryonic stem clones, microinjection, and chimera identification as described previously (9). To generate muscle-specific APPL2 KO mice and their wild-type (WT) littermates, APPL2^floxed mice were crossed with transgenic mice expressing Cre recombinase under the control of muscle creatine kinase promoter (The Jackson Laboratory), and their genotypes were identified by PCR analysis using the primers listed in Supplementary Table 1.

Both APPL2 Tg and APPL2 KO mice were backcrossed onto a C57BL/6 genetic background for at least seven generations and housed in a room with temperature (23 ± 1°C) and light (12-h light-dark cycle) control. Four-week-old male APPL2 Tg mice, muscle-specific APPL2 KO, and their WT littermates were fed with a standard chow (Purina Mills) comprising 20% kcal from protein, 10% kcal from fat, and 70% kcal from combined simple carbohydrates. Glucose tolerance test (GTT) and insulin tolerance test (ITT) were performed in 16-h– and 6-h–fasted animals as previously described (8). All animal experimental protocols were approved by the Animal Ethics Committee of HKU.

The sequences of RNA interference (RNAi) duplex oligos against APPL1, APPL2, and scrambled control (Invitrogen) are listed in Supplementary Table 1. These oligos were transfected into C2C12 or L6 cells by electroporation according to the manufacturer’s protocol (Bio-Rad).

The adenoviruses encoding APPL1 and luciferase were generated in our previous study (11). To construct adenoviral vectors for overexpression of APPL2 or TBC1D1, cDNA encoding human APPL2 or human TBC1D1 were cloned into pAdeasy-1 adenoviral backbone vector (Stratagene) as described previously (11). PCR-based site-directed mutagenesis was performed to introduce S235A, S237A, and T596D mutations in human TBC1D1 by using the mutagenic primers as previously described (12). C2C12 and L6 myotubes were infected with various adenoviruses at a multiplicity of infection of 50.

Glucose uptake assays were performed in C2C12 and L6 myotubes using 2-deoxy-[³H]-glucose as tracer as described in our previous study (20). For ex vivo glucose uptake assay, mice fasted for 4 h were anesthetized. Isolated extensor digitorum longus (EDL) or soleus muscles were stimulated without or with insulin 60 μU/mL followed by measurement of 2-deoxy-[³H]-glucose uptake (21). For determination of GLUT4 translocation, an antibody-coupled densitometric assay was used to measure the content of surface Myc-GLUT4 in L6 myotubes stably expressing Myc-tagged GLUT4 as described in previous studies (20,22).

C2C12 myotubes or human embryonic kidney (HEK) 293 cells were subjected to immunoprecipitation as described in our previous study (12). The immunocomplexes were eluted and subjected to immunoblotting analysis with various antibodies as specified in each figure legend or by mass spectrometry analysis for identification of interacting partners of APPL2 as previously described (23).

All experiments were performed routinely, with four to six repeats in each group. Data are presented as mean ± SE. Statistical significance was determined by Student t test or two-way ANOVA (for the experiments that involved two factors) followed by Bonferroni post hoc tests. In all statistical comparisons, P < 0.05 indicated a significant difference.

To compare the effects of APPL1 and APPL2 on glucose uptake and metabolism, we used the adenoviral gene delivery system for overexpression of these two adaptor proteins in L6 myotubes. The APPL1 and APPL2 protein levels in cells with ectopic expression of both APPL1 and APPL2 were increased by approximately threefold relative to their endogenous levels in L6 myotubes (Fig. 1A). Consistent with a previous study (24), ectopic overexpression of APPL1 enhanced insulin-stimulated glucose uptake and GLUT4 translocation to the plasma membrane compared with cells with ectopic expression of luciferase controls. On the contrary, overexpression of APPL2 inhibited such insulin actions (Fig. 1B and C). A similar result was also found in C2C12 myotubes (data not shown).

To verify these in vitro findings in cultured cells, we evaluated the impact of APPL1 and APPL2 overexpression on insulin-stimulated glucose uptake in skeletal muscle in mice. Transgenic mice with overexpression of FLAG-tagged human APPL1 driven by cytomegalovirus-β-actin promoter were generated in our previous study (9,10). We used a similar strategy to generate transgenic mice with overexpression of human APPL2 (Supplementary Fig. 1A), which was confirmed by PCR (Supplementary Fig. 1B) and immunoblotting (Supplementary Fig. 1C). The expression levels of APPL2 in EDL and soleus muscles of the transgenic mice were elevated approximately four- to sixfold compared with WT controls (Supplementary Fig. 1C). Consistently, ex vivo studies in isolated muscles showed that transgenic expression of APPL2 suppressed, whereas APPL1 enhanced, insulin-stimulated glucose uptake in EDL muscle compared with WT controls (Fig. 1D). A similar result was also observed in soleus muscle (data not shown).

APPL2 Tg mice exhibited a trend of increased fasting glucose and insulin levels (Supplementary Fig. 1D and E) and displayed a modest, but significant impairment in both glucose tolerance and insulin sensitivity compared with WT littermates (Supplementary Fig. 1F and G). Of note, overexpression of APPL2 had no effect on protein abundance of GLUT4 in EDL muscles (Supplementary Fig. 2).

Transfection of L6 myotubes with the duplex RNAi against rat APPL1 and APPL2 led to a reduction of APPL1 and APPL2 expression by 76% and 81%, respectively, compared with scrambled controls (Fig. 1E). Of note, knockdown of APPL2 expression potentiated, whereas suppression of APPL1 expression inhibited, insulin-stimulated glucose uptake and plasma membrane recruitment of GLUT4 (Fig. 1F and G). Likewise, the potentiating effects of APPL1 and the inhibitory effects of APPL2 on insulin-evoked glucose uptake were observed in C2C12 myotubes (data not shown).

To test whether deletion of APPL2 in skeletal muscles protects mice from glucose intolerance, we generated muscle-specific APPL2 KO mice by crossing APPL2^floxed mice with transgenic mice expressing Cre recombinase under the control of muscle creatine kinase promoter, which resulted in the disruption of the APPL2 gene at exon 5 (Fig. 2A). Immunoblotting analysis confirmed the dramatic reduction of APPL2 protein in EDL and soleus muscles but not in the brain of APPL2 KO mice (Fig. 2B). The residual expression of APPL2 in EDL and soleus muscles in APPL2 KO mice is perhaps due to its ubiquitous expression in other nonmyocyte cells (16). Genetic ablation of APPL2 in muscles had no obvious effects on food intake, body weight, and fasting glucose and insulin levels (Supplementary Table 2). GTT revealed that APPL2 KO mice exhibited a significant improvement of glucose tolerance in response to glucose challenge compared with WT littermates (Fig. 2C and D). Serum insulin levels during the GTT were similar between the two groups of mice (Fig. 2E). Insulin sensitivity, as determined by ITT, was also enhanced by APPL2 deletion (Fig. 2F). Ex vivo studies demonstrated that insulin-stimulated glucose uptake in EDL muscles of APPL2 KO mice was significantly increased compared with WT littermates (Fig. 2G). A similar result was observed in soleus muscles (data not shown). Insulin-elicited phosphorylation of Akt and IRβ in EDL muscle was comparable between the two groups (Fig. 2H).

To identify the proximal downstream effectors of APPL2, we established a stable cell line expressing FLAG-tagged human APPL2 for affinity pull-down purification of its potential interacting partners in HEK293 cells. Tandem mass spectrometry identified several putative interacting partners of APPL2, including heat shock protein (HSP) 70, HSP90, APPL1, centaurin delta 1, TBC1D1, and son of sevenless homolog 1. Among these APPL2-interacting proteins, TBC1D1, a member of the TBC1 Rab-GTPase family of proteins abundantly expressed in skeletal muscles, is an important regulator of insulin signaling and glucose metabolism (25,26). Of note, our coimmunoprecipitation analysis showed that TBC1D1 binds to APPL2 but not to APPL1 (Fig. 3A). The specificity of immunoprecipitation for TBC1D1 was confirmed by RNAi-mediated knockdown of TBC1D1 expression in C2C12 myotubes, leading to a substantial decrease in immunoprecipated TBC1D1 (Supplementary Fig. 3). On the other hand, APPL2 did not bind to TBC1D4 (Fig. 3B), a paralog of TBC1D1 that is also involved in the regulation of glucose transport in muscle cells and adipocytes (6). The APPL2–TBC1D1 interaction was enhanced by insulin stimulation in EDL muscle of C57 mice (Fig. 3C) and C2C12 myotubes (Fig. 3D), and such an enhancement was largely blocked by the PI3K inhibitor LY294002 (Fig. 3D).

To determine which domain of APPL2 is responsible for TBC1D1 binding, we generated a series of vectors that express various domains of APPL2 (Fig. 4A). Coimmunoprecipitation analysis revealed that TBC1D1 interacted with the BAR domain but not with the PH or PTB domain of APPL2 (Fig. 4B). The pull-down assay further confirmed the direct interaction between TBC1D1 and the BAR domain of APPL2 (Supplementary Fig. 4). To test whether the BAR domain mediates the suppressive effect of APPL2 on insulin-stimulated glucose uptake, we transduced L6 myotubes with adenovirus encoding the BAR domain or the PH-PTB domain or luciferase as control. Similar to full-length APPL2 (Fig. 1B and C), ectopic overexpression of the BAR domain but not the PH-PTB mutant inhibited insulin-stimulated glucose uptake and GLUT4 translocation to the plasma membrane (Fig. 4C and D). Taken together, these findings suggest that the BAR domain of APPL2 exerts an inhibitory effect on insulin-evoked glucose uptake by interacting with TBC1D1.

To further delineate how the APPL2–TBC1D1 interaction regulates insulin-stimulated glucose uptake, we generated a series of truncated mutants of TBC1D1 (Fig. 5A) to map the minimal domain mediating its binding to APPL2. Coimmunoprecipitation analysis revealed that all the mutants of TBC1D1 containing the linker region between the two PTB domains (amino acids 165–279) were able to interact with APPL2, whereas those mutants without the linker region lost their APPL2-binding property (Fig. 5B). Furthermore, the linker region alone was sufficient to bind with APPL2 (Fig. 5B).

TBC1D1 is a hyperphosphorylated protein, and insulin induces its phosphorylation through Akt activation (6). Because insulin prompted the APPL2–TBC1D1 interaction in a PI3K-dependent manner, we searched for the phospho-Akt substrate site within the linker region that may be crucial for its association with APPL2. Of note, a previous study showed that Ser-229 of mouse TBC1D1 (equivalent to Ser-235 on human TBC1D1) within this region is potentially phosphorylated by Akt (27). We confirmed that insulin stimulated TBC1D1 phosphorylation at Ser-235 in a time-dependent manner in both EDL muscle of C57 mice and C2C12 myotubes (Fig. 5C and D), and such an insulin action was largely abolished by the PI3K inhibitor LY294002 (Fig. 5D). Substitution of Ser-235 with nonphosphorylatable alanine (S235A) markedly attenuated insulin-stimulated interaction between APPL2 and TBC1D1, whereas mutation of Ser-237 to alanine had no obvious effect (Fig. 5E).

We next compared the effects of TBC1D1 and its mutant S235A on insulin-stimulated glucose uptake in L6 myotubes. Consistent with a previous study (27), we found that adenovirus-mediated expression of TBC1D1 did not affect insulin-evoked Akt phosphorylation (data not shown) but led to a significant inhibition of insulin-stimulated glucose uptake compared with cells expressing luciferase control (Fig. 5F). The inhibitory effects of TBC1D1 on insulin-evoked glucose uptake were abrogated by the S235A but not S237A mutation (Fig. 5F).

Several previous studies have shown that insulin-induced Akt activation leads to direct TBC1D1 phosphorylation on Thr-596 (equivalent to Thr-590 in mouse TBC1D1) (6,26–28), which in turn promotes the transport of GLUT4 to the plasma membrane by activating Rab-GTPases (6,26–28). To investigate the effects of APPL2 on insulin-elicited phosphorylation of TBC1D1, we injected APPL2 Tg and APPL2 KO mice intraperitoneally with insulin. Immunoblotting demonstrated that insulin-stimulated TBC1D1 phosphorylation on Thr-596 in EDL and soleus muscles was enhanced by APPL2 deletion but diminished by APPL2 overexpression (Fig. 6). On the other hand, APPL2 overexpression or deletion had no obvious effect on expression levels of total TBC1D1 or TBC1D1 phosphorylation on Ser-235 (Fig. 6).

To further investigate the interplay between APPL2 and TBC1D1 in regulating insulin-stimulated glucose uptake, we co-overexpressed APPL2 and TBC1D1 or its S235A mutant in L6 myotubes by adenoviral gene transfer system (Fig. 7A). Similar to the findings in APPL2 Tg mice, insulin-stimulated TBC1D1 phosphorylation on Thr-596 was significantly impaired by overexpression of APPL2 compared with cells overexpressing luciferase controls (Fig. 7A and B). However, the suppressive effect of APPL2 overexpression on insulin-stimulated phosphorylation of Thr-596 was abolished in cells expressing the TBC1D1-S235A mutant (Fig. 7A and B). Of note, overexpression of APPL2 caused only a modest suppressive effect on insulin-elicited Akt phosphorylation and had no effect on IRβ phosphorylation in L6 myotubes expressing either TBC1D1 or its S235A mutant (Fig. 7C and D). As expected, the inhibitory effect of TBC1D1 overexpression on insulin-stimulated glucose uptake was further aggravated by overexpression of APPL2 (Fig. 7E). However, this suppressive effect of APPL2 was lost in cells expressing the TBC1D1-S235A mutant (Fig. 7E).

To further investigate whether APPL2 suppresses insulin-elicited glucose uptake by modulating TBC1D1 phosphorylation on Thr-596, Thr-596 of TBC1D1 was mutated to aspartic acid (T596D) to mimic its phosphorylation. The level of adenovirus-mediated expression of the TBC1D1-T590D mutant was comparable to WT TBC1D1 in L6 myotubes (Fig. 8A). However, the suppressive effects of APPL2 overexpression on insulin-induced glucose uptake was observed in only L6 myotubes with ectopic expression of WT TBC1D1, not in L6 cells expressing the TBC1D1-T590D mutant (Fig. 8B).

In this study, we provide both in vivo and in vitro evidence showing that APPL2 negatively regulates insulin-stimulated glucose transport by interacting with TBC1D1. In both skeletal muscle and cultured muscle cells, insulin-induced glucose uptake is diminished by overexpression of APPL2 but is enhanced by suppression of APPL2 expression.

Despite of the high similarity in domain organization and amino acid sequences between APPL1 and APPL2, we demonstrate that these two adaptor proteins exert opposite effects on insulin-stimulated glucose uptake in skeletal muscle. APPL1 exerts its insulin-sensitizing effects by competing with TRB3 for binding to Akt, thereby promoting the translocation of Akt to the plasma membrane for further activation (11,16,24). On the other hand, the present study shows that APPL2 suppresses insulin-dependent glucose uptake at a step downstream of Akt. Unlike APPL1, APPL2 does not interact with Akt or the regulatory subunit p85 and catalytic subunit p110 of PI3K (7,29,30). Indeed, overexpression of APPL2 results in only a modest decrease in insulin-elicited Akt phosphorylation in myotubes, whereas targeted deletion of APPL2 in skeletal muscle has no obvious effect on Akt phosphorylation. Such a modest effect of APPL2 on Akt activity in cultured cells is perhaps a result of its heterodimerization with APPL1, which in turn prevents the binding of APPL1 to Akt (16). We demonstrate that APPL2 but not APPL1 interacts with TBC1D1, a downstream substrate of Akt that is critically involved in GLUT4 vesicle trafficking. Therefore, the opposite effects of APPL1 and APPL2 on insulin-stimulated glucose uptake are attributed to the differential binding of these two adaptor proteins to Akt and TBC1D1. In line with these findings, APPL1 and APPL2 have been shown to bind to various types of Rab-GTPases involved in membrane trafficking (17,19). The binding of APPL1 and APPL2 to various sets of intracellular signaling molecules may be due to their differences in oligomerization, surface charges, and/or subcellular localization (17,18,31,32).

TBC1D1 and TBC1D4, both of which are Rab-GAP proteins sharing ∼47% sequence identity and a similar domain organization, are the important regulators of both insulin- and contraction-induced trafficking of GLUT4 vesicles (6,33,34). Upon insulin stimulation, activated Akt induces phosphorylation of both TBC1D1 and TBC1D4 at multiple sites flanking the second PTB domain (26–28,35,36). Substitution of these phosphorylation sites with nonphosphorylatable alanine in TBC1D1 and TBC1D4 abolishes insulin-induced GLUT4 translocation to the plasma membrane in both adipocytes and muscle cells (25–27,36–38). Although the precise mechanisms by which TBC1D1 and TBC1D4 regulate GLUT4 vesicle trafficking remain unclear, it has been proposed that nonphosphorylated TBC1D4 in the basal state binds to GLUT4-containing vesicles to maintain its substrate Rab-GTPases in their inactive guanosine diphosphate–loaded form, thereby trapping GLUT4 inside cells (27,35–38). Insulin-evoked TBC1D4 phosphorylation on Thr649, possibly through interaction with 14-3-3 proteins, inhibits its GAP activity, which in turn allows guanosine triphosphate loading and activation of Rab-GTPases required for docking and fusion of GLUT4-containing vesicles to the plasma membrane (36–38). A previous study demonstrated that mice with TBC1D4-Thr649Ala knockin mutation, in which Thr649 was mutated to nonphosphorylatable alanine, display impaired glucose disposal and insulin sensitivity as a result of reduced GLUT4 trafficking to the cell surface in skeletal muscles (39). Likewise, insulin-elicited phosphorylation of TBC1D1 at Thr-596, a site equivalent to Thr-649 of TBC1D4, is obligatory for GLUT4 trafficking possibly by inactivation of GAP activity (26–28). In the current study, we found that the inhibitory effects of APPL2 overexpression on glucose uptake in muscle cells and skeletal muscle are associated with impaired phosphorylation of TBC1D1 at Thr-596 in response to insulin stimulation. Furthermore, mutation of Thr-596 of TBC1D1 to aspartic acid reverses the inhibitory effect of APPL2 on insulin-stimulated glucose uptake, suggesting that the APPL2–TBC1D1 interaction prevents Akt-mediated phosphorylation of TBC1D1 at Thr-596, thereby impairing insulin-evoked GLUT4 translation to the plasma membrane.

Although a previous study has identified Ser-235, a highly conserved amino acid located within the linker region of TBC1D1, as a likely phospho-Akt substrate site responsive to insulin stimulation (27), its physiological relevance has never been explored. In the current study, we further confirm that insulin induces TBC1D1 phosphorylation on Ser-235 in a time-dependent manner by using an antiphospho-Ser-235 antibody, and this phosphorylation is suppressed by pharmacological inhibition of PI3K. Furthermore, we found that mutation of Ser-235 to alanine not only abolishes the APPL2–TBC1D1 interaction but also abrogates the suppressive effects of APPL2 on insulin-induced TBC1D1 phosphorylation on Thr-596 and glucose uptake in muscle cells. On the basis of these findings, Akt-dependent phosphorylation of Ser-235 at N-terminal TBC1D1 likely triggers the interaction between TBC1D1 and APPL2, which in turn prevents phosphorylation of Thr-596 of TBC1D1 by either conformational changes or direct blockage of Akt access to Thr-596 and its surrounding motif. Of note, Ser-235 is absent in TBC1D4, which may partly explain the preferential interaction of APPL2 with TBC1D1 but not with TBC1D4. On the other hand, mutation of Ser-237 to alanine has no effect on the APPL2–TBC1D1 interaction as well as on the suppressive effect of TBC1D1 on insulin-stimulated glucose uptake. Of note, Ser-237 is phosphorylated by AMPK in response to contraction but not by insulin stimulation (40–42), suggesting that APPL2 may only regulate glucose uptake through the TBC1D1 signaling pathway in response to insulin and not contraction in skeletal muscle.

In summary, the present study identifies APPL2 as a negative regulator of insulin-stimulated GLUT4 translocation to the plasma membrane by interacting with and modulating phosphorylation of TBC1D1 (Supplementary Fig. 5). Insulin stimulation induces phosphorylation of both Ser-235 and Thr-596 of TBC1D1. Phosphorylation of TBC1D1 at Ser-235 triggers its interaction with APPL2, which in turn blocks further phosphorylation of Thr-596 required for plasma membrane targeting of GLUT4. Such a feedback regulatory loop may be an important mechanism for insulin to fine tune glucose homeostasis in mammals.

Funding. This work was supported by General Research Fund Grant HKU 783010M, HKU matching funds for State Key Laboratory of Pharmaceutical Biotechnology, the National Science Foundation of China (81270881), and the National Basic Research Program of China (2011CB504004 and 2010CB945500).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. K.K.Y.C. and W.Z. contributed to researching data and writing the manuscript. B.C. and B.W. contributed to researching data. Y.W. contributed to researching data and the discussion. D.W. contributed to researching data and editing the manuscript. G.S. contributed to researching data and reviewing the manuscript. K.S.L.L. contributed to supervising the study and editing the manuscript. A.X. contributed to the design and supervision of the study and writing of the manuscript. K.K.Y.C. 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.