An AAGAB-to-CCDC32 handover mechanism controls the assembly of the AP2 adaptor complex

Significance

Vesicular transport relies on membrane trafficking complexes to capture cargo and to drive vesicle budding and fusion. It has remained largely unexplored how these trafficking complexes are assembled. Here, we unveil the assembly pathway of the AP2 adaptor, a heterotetrameric trafficking complex that regulates clathrin-mediated endocytosis. We demonstrate that AP2 assembly is controlled by a handover mechanism, switching from AAGAB-based initiation complexes to CCDC32-based template complexes. We suggest that a similar mechanism may also govern the assembly of other membrane trafficking complexes exhibiting the same configuration as AP2. Ultimately, these findings could pave the way for developing therapeutics to treat diseases caused by defects in the assembly of membrane trafficking complexes.

Abstract

Vesicular transport relies on multimeric trafficking complexes to capture cargo and drive vesicle budding and fusion. Faithful assembly of the trafficking complexes is essential to their functions but remains largely unexplored. Assembly of AP2 adaptor, a heterotetrameric protein complex regulating clathrin-mediated endocytosis, is assisted by the chaperone AAGAB. Here, we found that AAGAB initiates AP2 assembly by stabilizing its α and σ2 subunits, but the AAGAB:α:σ2 complex cannot recruit additional AP2 subunits. We identified CCDC32 as another chaperone regulating AP2 assembly. CCDC32 recognizes the AAGAB:α:σ2 complex, and its binding leads to the formation of an α:σ2:CCDC32 ternary complex. The α:σ2:CCDC32 complex serves as a template that sequentially recruits the µ2 and β2 subunits of AP2 to complete AP2 assembly, accompanied by CCDC32 release. The AP2-regulating function of CCDC32 is disrupted by a disease-causing mutation. These findings demonstrate that AP2 is assembled by a handover mechanism switching from AAGAB-based initiation complexes to CCDC32-based template complexes. A similar mechanism may govern the assembly of other trafficking complexes exhibiting the same configuration as AP2.

In eukaryotic cells, about one-third of proteins are localized to the endomembrane system, which comprises the endoplasmic reticulum (ER), the Golgi apparatus, endosomes, lysosomes, and other membrane-bound organelles (1, 2). Proteins are transported between these organelles by vesicles that rely on multimeric trafficking complexes such as adaptor protein complexes to capture cargo and facilitate vesicle budding and fusion (3). While the biological functions of trafficking complexes are well established, how they are assembled remains poorly understood.

Protein complex assembly is a highly challenging process in the crowded cellular environment (4, 5). Folding chaperones such as Hsp70s and chaperonins facilitate the folding of individual subunits, but the subunits attain their native three-dimensional conformations only in the context of the fully assembled complex (6, 7). Unpaired subunits and assembly intermediates expose assembly interfaces, which are prone to aggregation, misassembly, and degradation (5, 8). These challenges are exacerbated by macromolecular crowding, which reduces diffusion rates of protein subunits, impedes cognate interactions, and promotes misassembly (9).

Recent studies of adaptor protein complexes suggest that faithful assembly of a trafficking complex in the cell depends on dedicated assembly chaperones (10–12). In a genome-scale CRISPR screen, alpha and gamma adaptin binding protein (AAGAB, also known as p34) was identified as a chaperone essential to the assembly of AP2, a heterotetrameric adaptor protein complex regulating clathrin-mediated endocytosis (CME) (10, 13). AP2 is composed of two large subunits (α and β2), one medium subunit (µ2), and one small subunit (σ2) (14–17). AP2 assembles in the cytosol before being targeted to the plasma membrane to capture cargoes and recruit clathrin to promote vesicle budding (18–22). AAGAB assists AP2 assembly by binding to the α subunit (α-adaptin) to form an AAGAB:α binary complex, which subsequently recruits σ2 to form an AAGAB:α:σ2 ternary complex (10–12). In these binary and ternary complexes, AAGAB stabilizes the α and σ2 subunits. Besides AP2, AAGAB also assists the assembly of AP1, an adaptor protein complex involved in clathrin-mediated vesicle budding at the trans-Golgi network (TGN) and the endosome (12), and AP4, an adaptor protein complex regulating clathrin-independent transport at the TGN (23, 24). Without the assistance of AAGAB, AP1, AP2, and AP4 fail to assemble, leading to their degradation and abrogation of membrane trafficking processes dependent on these adaptor protein complexes (10–12, 23). Heterozygous AAGAB mutations cause punctate palmoplantar keratoderma type 1 (PPKP1), a skin disorder, and are linked to increased incidents of cancer (25–27).

In this work, we found that while AAGAB initiates AP2 assembly by binding and stabilizing α and σ2 subunits, the AAGAB:α:σ2 complex is unable to recruit additional AP2 subunits. We identified coiled-coil domain containing 32 (CCDC32, also known as C15ORF57), an unannotated protein, as another chaperone required for AP2 assembly. CCDC32 mutations cause developmental disorders including the cardiofacioneurodevelopmental syndrome (CFNDS) (28–30), but its function remained unclear. We found that CCDC32 interacts with the AAGAB:α:σ2 complex and its binding leads to the formation of an α:σ2:CCDC32 ternary complex. The α:σ2:CCDC32 complex serves as a template that sequentially recruits µ2 and β2 subunits to complete AP2 assembly, concomitant with CCDC32 release. The function of CCDC32 in AP2 assembly is compromised by a CFNDS-causing mutation. These findings demonstrate that AP2 assembly is controlled by a handover mechanism switching from AAGAB-based initiation complexes to CCDC32-based template complexes.

Results

Identification of CCDC32 as an Essential Regulator of CME.

To search for potential chaperones acting in concert with AAGAB to assist adaptor protein assembly, we determined the interactome of AAGAB in HEK 293T cells using coimmunoprecipitation (co-IP) and mass spectrometry (Fig. 1A). As expected, AP1 and AP2 subunits were among the top-ranking AAGAB-interacting proteins (Fig. 1B and SI Appendix, Table S1). Other proteins in the interactome, however, have not been functionally linked to AAGAB. A novel assembly chaperone of adaptor protein complexes is expected to meet three criteria: 1) lacking a transmembrane domain because adaptor protein complexes assemble in the cytosol; 2) ubiquitously expressed according to tissue distribution databases (31); and 3) having no annotated functions in other biological processes. A candidate protein in the AAGAB interactome meeting all the criteria is CCDC32 (Fig. 1B), an evolutionarily conserved soluble protein of ~20 kDa (SI Appendix, Fig. S1). CCDC32 exhibits no sequence similarity with other proteins and possesses no discernable protein domains. The function and mechanism of CCDC32 were unclear.

Fig. 1.

Identification of CCDC32 as an AAGAB-interacting protein required for CME. (A) Proteomic analysis to identify AAGAB-interacting proteins. (B) A volcano plot showing the interactome of AAGAB. Each dot represents an individual protein. Selected proteins are marked by arrows. Full datasets are shown in SI Appendix, Table S1. (C) Representative immunoblots showing the interaction of 3xFLAG-tagged AAGAB with HA-tagged CCDC32. The 3xFLAG-AAGAB protein was transiently expressed in HeLa cells with an empty vector (control) or a plasmid encoding HA-tagged CCDC32. AAGAB was immunoprecipitated from cell lysates using anti-FLAG antibodies, and proteins in the immunoprecipitates were detected using immunoblotting. M.W., molecular weight. All full immunoblots of this work are included in SI Appendix. (D) Normalized surface levels of TfR in WT and mutant HeLa cells measured by flow cytometry. In all figures, data normalization was performed by setting the mean value of WT data points as 100 or 1 and all data points including WT ones were normalized to that mean value. Data are presented as mean ± SD of three biological replicates. ***P < 0.001 (calculated using Student’s t test). (E) Representative confocal images showing the endocytosis of ATTO565-conjugated transferrin (Tf-ATTO565) into WT and CCDC32 KO HeLa cells. The plasma membrane was stained using CF405M-conjugated Concanavalin A. (Scale bars, 10 µm.) (F) Quantification of Tf-ATTO565 fluorescence in WT and CCDC32 KO HeLa cells using flow cytometry. Mean fluorescence of mutant cells was normalized to that of WT cells. Data are presented as mean ± SD of three biological replicates. ***P <0.001 (calculated using Student’s t test). (G) Representative confocal images showing the localization of GLUT4 reporters (ALFA-GLUT4-mGreenLantern) in unpermeabilized adipocytes. mGreenLantern fluorescence reflects total reporters, whereas surface reporters were labeled using mScarlet-anti-ALFA nanobodies. Nuclei were stained using Hoechst 33342. (Scale bars, 10 µm.) (H) Normalized surface levels of GLUT4 reporters measured using flow cytometry. mScarlet fluorescence (surface staining) was normalized to mGreenLantern fluorescence (total). Data of mutant adipocytes were normalized to those of WT cells. Data are presented as mean ± SD of three biological replicates. ***P < 0.001 (calculated using Student’s t test).

The interaction of CCDC32 with AAGAB in the cell was confirmed using co-IP and immunoblotting (Fig. 1C). Next, we deleted the CCDC32 gene in HeLa cells (SI Appendix, Fig. S2) and used flow cytometry to quantify surface levels of transferrin receptor (TfR), a classic CME cargo (32). We observed that surface TfR levels were markedly elevated in CCDC32 KO HeLa cells (Fig. 1D), suggesting a defect in TfR endocytosis. Next, we directly measured the endocytosis of transferrin (Tf), which enters the cell through binding to TfR (32). Tf was labeled using the cleavable fluorescent dye ATTO565-3-[(2-aminoethyl)dithio]propionic acid N-hydroxysuccinimide ester (ATTO565-AEDP-NHS-ester) (33–35). After conjugation to Tf, the ATTO565 dye moiety was linked to a cargo protein through a disulfide bond. Surface-exposed ATTO565 dyes were removed by breaking the disulfide bond using sodium 2-mercaptoethane sulfonate (MESNa), a membrane-impermeable reducing agent, whereas dyes present on internalized Tf were resistant to MESNa cleavage. Using this endocytosis assay, we observed that Tf endocytosis was strongly reduced in CCDC32 KO HeLa cells (Fig. 1 E and F), in agreement with elevated surface levels of TfR in the mutant cells (Fig. 1D).

The role of CCDC32 in Tf uptake is consistent with a previous observation (36), but it was unclear whether CCDC32 selectively regulates TfR/Tf endocytosis or acts as a general regulator of CME. To address this question, we measured surface levels of the glucose transporter GLUT4, a multitransmembrane protein internalized through CME (10). We observed that surface levels of GLUT4 were markedly increased in Ccdc32 KO mouse adipocytes (Fig. 1 G and H), reminiscent of the observations in AP2-deficient cells (10). Thus, the function of CCDC32 is not restricted to a specific cargo or cell type, indicating that CCDC32 is a general regulator of CME.

CCDC32 Regulates AP2 Assembly.

Next, we sought to determine the molecular mechanism by which CCDC32 regulates CME. Since CCDC32 was isolated as an AAGAB-interacting protein, we examined whether CCDC32 is also involved in AP2 assembly. While adaptor protein complexes can assemble spontaneously when expressed at high levels in heterologous expression systems (19, 37), their assembly in eukaryotic cells is strictly dependent on dedicated chaperones (10–12, 23). When an adaptor protein complex fails to assemble due to lack of a chaperone, all subunits are degraded (10, 12). Indeed, expression of AP2 subunits was diminished in CCDC32 KO HeLa cells (Fig. 2 A and B). Confocal imaging revealed abundant AP2 puncta on the surface of wild-type (WT) HeLa cells, but the AP2 puncta were absent in CCDC32 KO cells (Fig. 2 C and D and SI Appendix, Fig. S3 A–D). AP2 expression was fully restored when a CCDC32 rescue gene was expressed (Fig. 2 C and D and SI Appendix, Fig. S3 A–F). We also deleted the CCDC32/Ccdc32 gene in human induced pluripotent stem cells (iPSCs) and mouse preadipocytes, a type of fibroblast. Similar to the HeLa cell data, AP2 subunits were depleted in the mutant iPSCs and preadipocytes (SI Appendix, Fig. S3G). Thus, the requirement of CCDC32 for AP2 adaptor formation is not restricted to a specific cell type. These data suggest that like AAGAB, CCDC32 is essential to the assembly of the AP2 adaptor complex.

Fig. 2.

Loss of AP2 in CCDC32 KO cells. (A) Representative immunoblots showing the expression of the indicated proteins in WT and mutant HeLa cells. Although expressed in HeLa cells (12), σ2 could not be detected using the antibodies we tested. All full-size blots are included in SI Appendix. (B) Quantification of protein expression in WT and mutant HeLa cells based on immunoblots. Data are presented as mean ± SD of three biological replicates. ***P < 0.001, n.s., P > 0.05 (calculated using One-way ANOVA). (C) Representative confocal images showing AP2 (α staining) puncta on the plasma membrane in WT, CCDC32 KO, and rescue HeLa cells. (Scale bars, 10 µm.) (D) Quantification of AP2 puncta (α staining) on the plasma membrane. Images were captured as in C and analyzed using ImageJ. Each dot represents imaging data of an individual cell focusing on the plasma membrane. Error bars indicate SD. ***P < 0.001, n.s., P > 0.05 (calculated using Student’s t test). (E) Representative structured illumination microscopy (SIM) images showing the colocalization of AAGAB and CCDC32 transiently expressed in HeLa cells. AAGAB was labeled using anti-FLAG antibodies and Alexa Fluor 488–conjugated secondary antibodies. ALFA-tagged CCDC32 was labeled using anti-ALFA nanobodies fused to mScarlet. (Scale bars, 10 µm for main images and 1 μm for enlarged images.) (F) Quantification of AAGAB and CCDC32 colocalization using the Pearson correlation coefficient. Images were captured as in E and analyzed using ImageJ. Each dot represents imaging data of an individual cell focusing on the cytosol. In randomized data, AAGAB images were rotated 90 degrees clockwise, whereas CCDC32 images were not rotated. Error bars indicate SD. ***P < 0.001 (calculated using Student’s t test).

All five adaptor protein complexes (AP1-5) were detected in HeLa cells (Fig. 2 A and B). These complexes exhibit a similar configuration but play distinct roles in vesicle-mediated transport (3, 14, 24). Interestingly, only AP2 was depleted in CCDC32 KO cells, whereas expression of the other adaptor protein complexes remained intact (Fig. 2 A and B). By contrast, expression of AP1, AP2, and AP4 was diminished in AAGAB KO cells (Fig. 2 A and B), in agreement with previous reports (10, 12, 23). Adaptor protein complexes are related to the heterotetrameric F-subcomplex of the COP-I coat (24). However, the COP-I coat was not impacted in CCDC32 or AAGAB KO cells (SI Appendix, Fig. S4). These results suggest that CCDC32 selectively regulates the assembly of AP2 adaptor.

To examine its subcellular localization, CCDC32 was fused to an ALFA epitope tag, and the fusion protein was detected using recombinant ALFA-recognizing nanobodies. CCDC32 was found in both the cytosol and the nucleus (Fig. 2E and SI Appendix, Fig. S3H), suggesting that it does not form large oligomers preventing free diffusion across nuclear pores. CCDC32 partially colocalized with AAGAB in the cytosol (Fig. 2 E and F and SI Appendix, Fig. S3H), consistent with the notion that AP2 assembles in the cytosol. AAGAB expression was not reduced in CCDC32 KO cells (SI Appendix, Fig. S5A), indicating that CCDC32 does not regulate AAGAB expression. Next, AAGAB was overexpressed in CCDC32 KO cells to elevate AAGAB protein levels (SI Appendix, Fig. S5A). AAGAB overexpression, however, did not restore AP2 expression in CCDC32 KO cells (SI Appendix, Fig. S5A). Likewise, overexpression of CCDC32 in AAGAB KO cells did not restore AP2 expression (SI Appendix, Fig. S5B). Thus, AAGAB and CCDC32 play nonredundant roles in regulating AP2 adaptor.

CCDC32 Recognizes the AAGAB:α:σ2 Complex to Form an AAGAB:α:σ2:CCDC32 Quaternary Complex.

To determine whether CCDC32 directly interacts with AAGAB, we coexpressed GST-tagged CCDC32 and His₆-SUMO-tagged AAGAB in Escherichia coli. We first used glutathione beads to pull down GST-CCDC32 (SI Appendix, Fig. S6A). Unexpectedly, GST-CCDC32 did not bind appreciably to His₆-SUMO-tagged AAGAB (SI Appendix, Fig. S6B). Next, we used nickel beads to pull down His₆-SUMO-tagged AAGAB (SI Appendix, Fig. S6C). Consistent with the GST pull-down data, His₆-SUMO-tagged AAGAB did not bind detectable amounts of GST-CCDC32 (SI Appendix, Fig. S6D). Since AAGAB forms homodimers at its resting state through its conserved C-terminal domains (CTD) (11), we examined whether its oligomeric state influences CCDC32 binding. AAGAB dimers were fully dissociated into monomers by a mutation at its dimerization interface, L269R (11). The monomeric AAGAB mutant, however, did not bind noticeably to CCDC32 either, similar to WT AAGAB (SI Appendix, Fig. S6 B and D). Thus, the oligomeric state of AAGAB does not influence CCDC32 binding. These data demonstrate that although found in the AAGAB interactome (Fig. 1 B and C), CCDC32 does not bind to standalone AAGAB, suggesting that they interact through a protein complex involving additional factors.

Next, we examined whether CCDC32 interacts with AP2-associated AAGAB. AAGAB binds to the α and σ2 subunits of AP2 to form an AAGAB:α:σ2 ternary complex (10–12). We expressed and purified recombinant CCDC32 from E. coli and added it to the AAGAB:α:σ2 complex (Fig. 3A). When incubated at 4 °C, CCDC32 bound to the AAGAB:α:σ2 complex to form an AAGAB:α:σ2:CCDC32 quaternary complex (Fig. 3 B–E). Using co-IP of epitope-tagged proteins, we observed that CCDC32 interacted with α and σ2 subunits in HeLa cells (Fig. 3 F and G), in addition to its binding to AAGAB (Fig. 1C). The interaction of CCDC32 with α and σ2 was also observed in co-IP using endogenous proteins (SI Appendix, Fig. S2B), confirming that CCDC32 is found in the same protein complexes as AP2 α and σ2 in the cell. By contrast, CCDC32 did not interact with AP1 or AP3 subunits (SI Appendix, Fig. S7), in agreement with the observation that CCDC32 is dispensable for the formation of AP1 and AP3 adaptors (Fig. 2 A and B). These data demonstrate that although not binding to standalone AAGAB, CCDC32 interacts with the ternary AAGAB:α:σ2 complex to form an AAGAB:α:σ2:CCDC32 quaternary complex.

Fig. 3.

CCDC32 recognizes the AAGAB:α:σ2 ternary complex. (A) Diagram of a GST pull-down assay detecting CCDC32 binding to the AAGAB:α:σ2 ternary complex. His₆-SUMO-tagged AAGAB was coexpressed with the GST-tagged α trunk domain and untagged σ2 in E. coli. Proteins were isolated from E. coli lysates using glutathione beads before purified recombinant CCDC32 was added and incubated at 4 °C for 1 h. (B) Representative Coomassie blue–stained gels showing the binding of CCDC32 to the AAGAB:α:σ2 ternary complex as depicted in A. (C) Representative Coomassie blue–stained gels showing that CCDC32 does not bind to GST. GST proteins were isolated from E. coli lysates using glutathione beads before purified recombinant CCDC32 was added and incubated as in A and B. (D) Recombinant CCDC32 was added to the AAGAB:α:σ2 ternary complex as in A and B at the indicated molecular ratio and the interaction was measured using immunoblotting. (E) Quantification of CCDC32 binding to the AAGAB:α:σ2 ternary complex based on immunoblots from three independent experiments. Levels of proteins were normalized to those of AP2 α. Error bars indicate SD. (F) Diagrams of 3xFLAG-tagged CCDC32 and HA-tagged FL α and σ2 used in co-IP. (G) Representative immunoblots showing the interaction of 3xFLAG-tagged CCDC32 with HA-tagged α and σ2. The 3xFLAG-tagged CCDC32 protein was transiently expressed in CCDC32 KO HeLa cells with an empty vector (control) or plasmids encoding HA-tagged AP2 subunits. CCDC32 was immunoprecipitated using anti-FLAG antibodies, and proteins in the immunoprecipitates were detected using immunoblotting.

CCDC32 Directly Binds to the α and σ2 Subunits.

Next, we examined whether CCDC32 directly binds to the α and σ2 subunits of AP2 using GST pull-down assays (Fig. 4A). When coexpressed in E. coli, CCDC32 directly interacted with α and σ2 but only moderately stabilized them, in contrast to the strong stabilizing effects of AAGAB (Fig. 4B). Similar results were obtained when nickel bead pull-down was used to detect the interaction of AP2 subunits with CCDC32 and AAGAB (Fig. 4 C and D). These data are consistent with the observation that CCDC32 cannot replace AAGAB to stabilize these AP2 subunits (SI Appendix, Fig. S5B). In AAGAB KO HeLa cells, CCDC32 still interacted with transiently expressed α and σ2 (Fig. 4E), confirming that CCDC32-AP2 binding occurs in the cell even in the absence of AAGAB. Thus, while less efficient than AAGAB in stabilizing α and σ2, CCDC32 can directly interact with these AP2 subunits. Next, we tested a CCDC32 mutant recapitulating a mutation causing the human disease CFNDS (28). We observed that the CCDC32 mutant was defective in binding AP2 α and σ2 when expressed at similar levels as WT CCDC32 (SI Appendix, Fig. S8), indicating that the disease-mimicking mutant is defective in regulating AP2 assembly.

Fig. 4.

CCDC32 binds and stabilizes AP2 subunits. (A) Diagram of the GST pull-down assay measuring the interaction of AAGAB or CCDC32 with α and σ2. His₆-SUMO-tagged AAGAB or CCDC32 was coexpressed with GST or GST-tagged α trunk domain and untagged σ2 in E. coli. Proteins were isolated from E. coli lysates using glutathione beads. (B) Representative Coomassie blue–stained gels and immunoblots showing the binding of GST-tagged α and untagged σ2 to His₆-SUMO-tagged AAGAB or CCDC32 as depicted in A. (C) Diagram of the nickel bead pull-down assay measuring the interaction of GST-tagged α and untagged σ2 with His₆-SUMO-tagged AAGAB or CCDC32. Proteins were coexpressed in E. coli as in A and B and isolated using nickel beads recognizing a His₆ tag on AAGAB or CCDC32. (D) Representative Coomassie blue–stained gels and immunoblots showing the binding of His₆-SUMO-tagged AAGAB or CCDC32 to GST-tagged α and untagged σ2 as depicted in C. (E) Representative immunoblots showing the interaction of 3xFLAG-tagged CCDC32 with HA-tagged AP2 subunits in AAGAB deficient cells. The 3xFLAG-tagged CCDC32 protein was transiently expressed in AAGAB KO HeLa cells with an empty vector (control) or plasmids encoding HA-tagged AP2 subunits. CCDC32 was immunoprecipitated using anti-FLAG antibodies, and proteins in the immunoprecipitates were detected using immunoblotting.

Handover of the α and σ2 Subunits From AAGAB to CCDC32.

The above data suggest two distinct CCDC32-AP2 binding modes. In the first binding mode, CCDC32 exists in a quaternary complex containing AAGAB, α, and σ2. In the second binding mode, CCDC32 associates with α and σ2 in a ternary complex free of AAGAB. To determine the functions of these binding modes, we further characterized the AAGAB:α:σ2:CCDC32 quaternary complex using purified recombinant proteins. We observed that at 37 °C CCDC32 rapidly associated with the AAGAB:α:σ2 complex to form the AAGAB:α:σ2:CCDC32 quaternary complex (Fig. 5A). Interestingly, during incubation at 37 °C, AAGAB gradually dissociated from the AAGAB:α:σ2:CCDC32 quaternary complex (Fig. 5A). In the absence of CCDC32, by contrast, the AAGAB:α:σ2 complex remained stable at 37 °C (Fig. 5A and SI Appendix, Fig. S9). Thus, the AAGAB:α:σ2:CCDC32 quaternary complex is metastable at 37 °C and AAGAB ultimately dissociates from AP2 subunits. However, there was a significant delay (over 20 min) between the binding of CCDC32 and the dissociation of the majority of AAGAB proteins (Fig. 5A). These data suggest that the AAGAB:α:σ2:CCDC32 quaternary complex is a key intermediate in the AP2 assembly pathway and are consistent with the interactions and colocalization of AAGAB and CCDC32 observed in the cell (Figs. 1 B and C and 2 E and F). Together, these findings demonstrate that AAGAB initiates AP2 assembly by stabilizing α and σ2 but is subsequently displaced by CCDC32, forming a new ternary complex of α:σ2:CCDC32 (Fig. 5B).

Fig. 5.

Handover of the α:σ2 hemicomplex from AAGAB to CCDC32. (A) His₆-SUMO-tagged AAGAB was coexpressed with GST-tagged α (trunk domain) and untagged σ2 in E. coli. The AAGAB:α:σ2 ternary complex was isolated from E. coli using glutathione beads and incubated with recombinant CCDC32 at the indicated molar ratios at 37 °C. After incubation for the indicated periods, the glutathione beads were washed, and proteins bound to the beads were detected using immunoblotting. Left: Representative immunoblots. Right: Quantification of proteins based on immunoblots from three independent experiments. Levels of proteins were normalized to those of AP2 α. Error bars indicate SD. (B) Diagrams depicting two AP2-binding modes of CCDC32.

The α:σ2:CCDC32 Complex Sequentially Recruits the μ2 and β2 Subunits.

Next, we examined whether CCDC32 interacts with the μ2 and β2 subunits of AP2 (Fig. 6A). Using co-IP, we detected CCDC32 binding to μ2 but not to β2 in HEK 293T cells (Fig. 6 B and C and SI Appendix, Fig. S10A). The binding of CCDC32 to μ2 was confirmed using GST pull-down of endogenous μ2 proteins (SI Appendix, Fig. S10 B and C). By contrast, AAGAB interacted with neither μ2 nor β2 in HEK 293T cells (Fig. 6 B and C and SI Appendix, Fig. S10A), consistent with the results of our proteomic analysis and previous studies (Fig. 1B and SI Appendix, Table S1) (10). These data further suggest that AAGAB initiates AP2 assembly by stabilizing α and σ2 but the AAGAB:α:σ2 complex is unable to recruit μ2 or β2. When α and σ2 subunits are transferred from AAGAB to CCDC32, the newly formed α:σ2:CCDC32 complex resumes AP2 assembly by recruiting μ2.

Fig. 6.

The α:σ2:CCDC32 complex sequentially recruits µ2 and β2 to complete AP2 assembly. (A) Diagrams of HA-tagged FL AP2 µ2 and β2 subunits and 3xFLAG-tagged AAGAB and CCDC32 used in co-IP experiments. (B and C) Representative immunoblots showing the interaction of 3xFLAG-tagged AAGAB and CCDC32 with HA-tagged µ2 (B) or β2 (C). The 3xFLAG-tagged AAGAB or CCDC32 protein was transiently expressed in HEK 293T cells with an empty vector (control) or plasmids encoding the indicated HA-tagged AP2 subunits. AAGAB and CCDC32 were immunoprecipitated using anti-FLAG antibodies, and proteins in the immunoprecipitates were detected using immunoblotting. (D) Representative immunoblots showing the interactions of µ2 and β2 subunits with the α:σ2:CCDC32 complex. His₆-SUMO-tagged CCDC32 was coexpressed with GST-tagged α (trunk domain) and untagged σ2 in E. coli. The α:σ2:CCDC32 ternary complex was isolated from E. coli using glutathione beads and incubated with E. coli lysates containing His₆-SUMO-tagged µ2 and β2 (trunk domain, a.a. 1–591) at 4 °C for the indicated periods. The glutathione beads were washed, and proteins bound to the beads were detected using immunoblotting. (E) Quantification of proteins based on immunoblots from three independent experiments. Levels of proteins were normalized to those of AP2 α. Error bars indicate SD. (F) Representative SIM images showing the subcellular localization of clathrin heavy chain (CHC), AP2 α, and CCDC32, focusing on the plasma membrane. Endogenous CHC was stained using anti-CHC and Alexa Fluor 647–conjugated secondary antibodies, whereas endogenous AP2 α was stained using anti-α antibodies and Alexa Fluor 568–conjugated secondary antibodies. FLAG-tagged CCDC32 was labeled using Alexa Fluor 488–conjugated anti-FLAG antibodies. Colocalization of CHC and AP2 α, but not with CCDC32, was indicated by arrows. (Scale bars, 10 μm for main images and 1 μm for enlarged images.) (G) Profile analysis plot comparing the distributions of CHC, α and CCDC32 within the rectangular areas of F. (H) Quantification of the colocalization of CCDC32 with AP2 adaptor (α staining) and CHC using the Pearson correlation coefficient. Images were captured as in F and analyzed using ImageJ. Each dot represents imaging data of an individual cell focusing on the plasma membrane. In randomized data, CHC and AP2 α images were rotated 90 degrees clockwise, whereas CCDC32 images were not rotated. Error bars indicate SD. n.s., P > 0.05 (calculated using Student’s t test).

Since β2 does not associate with CCDC32 in the cell, we considered the possibility that β2 recruitment is accompanied by CCDC32 dissociation from AP2 subunits. To test this possibility, we prepared recombinant α:σ2:CCDC32 complexes with a GST tag on the α subunit and examined whether the ternary complex binds to recombinant μ2 and β2 expressed in E. coli (SI Appendix, Fig. S10D). Consistent with the co-IP results, μ2 interacted with the α:σ2:CCDC32 complex (SI Appendix, Fig. S10E). Interestingly, when both μ2 and β2 were added, AP2-bound CCDC32 was diminished (Fig. 6 D and E and SI Appendix, Fig. S10 D–F), supporting the notion that CCDC32 is released when β2 joins the other three AP2 subunits to complete AP2 assembly. Next, we examined whether CCDC32 associates with mature AP2 adaptors on the plasma membrane. In HeLa cells, SIM imaging showed that CCDC32 did not colocalize with endogenous mature AP2 adaptors on the plasma membrane (Fig. 6 F–H), confirming that CCDC32 does not continue associating with AP2 when the latter is fully assembled.

Altogether, these data demonstrate that the CCDC32:α:σ2 complex serves as a template that sequentially recruits μ2 and β2 subunits. β2 binding completes AP2 assembly and triggers CCDC32 release. Thus, like AAGAB, CCDC32 meets the criteria of an assembly chaperone: interacting with assembly intermediates and promoting complex assembly without being part of the final protein complex.

Discussion

In the crowded cytosol of eukaryotic cells, AP2 subunits cannot spontaneously assemble into the heterotetrameric AP2 complex through random collision (10–12). Instead, AP2 assembly is strictly dependent on the assistance of the assembly chaperones AAGAB and CCDC32 (Fig. 7). Prior to AP2 binding, AAGAB forms as a homodimer through its CTD, burying hydrophobic α-binding residues in the dimeric interface (11). AAGAB initiates AP2 assembly by using its CTD to bind the α subunit, forming an AAGAB:α binary complex. Next, the AAGAB:α binary complex recruits the σ2 subunit, forming an AAGAB:α:σ2 ternary complex. In these binary and ternary complexes, AAGAB stabilizes the α and σ2 subunits. Subsequently, CCDC32 recognizes the AAGAB:α:σ2 ternary complex and its binding leads to AAGAB release and formation of a new ternary complex of α:σ2:CCDC32. The α:σ2:CCDC32 complex serves as a template to recruit the μ2 subunit. The resulting α:σ2:CCDC32:μ2 quaternary complex in turn serves as a template to recruit the β2 subunit. β2 binding completes the formation of full AP2 complex and is accompanied by CCDC32 release (Fig. 7). AP2 assembly occurs in the cytosol and only fully assembled AP2 complexes are recruited to the plasma membrane (19), which may prevent aberrant binding of assembly intermediates to CME cargo proteins (10).

Fig. 7.

Model of chaperone-assisted AP2 adaptor assembly.

Chaperone-assisted AP2 assembly is an unexpectedly complex process with at least five distinct chaperone–AP2 intermediates en route to full AP2 formation (Fig. 7). Despite the complexity, the entire AP2 assembly pathway is driven by free energy changes induced by protein–protein interactions (SI Appendix, Fig. S11). As assembly chaperones, AAGAB and CCDC32 possess distinct properties and play complementary roles in AP2 assembly. AAGAB controls the first half of the AP2 assembly pathway by binding and stabilizing the α and σ2 subunits in the AAGAB:α and AAGAB:α:σ2 complexes (Fig. 7). These initiation complexes are essential intermediates in the AP2 assembly pathway, but they are unable to recruit μ2 and β2 subunits (Fig. 6). CCDC32 assists the second half of AP2 assembly by forming template complexes that consecutively recruit μ2 and β2 subunits to complete AP2 assembly. Unlike AAGAB, CCDC32 cannot effectively stabilize individual α and σ2 subunits and thus must rely on AAGAB to supply these subunits (Fig. 4). CCDC32, however, stabilizes the α and σ2 subunits once the latter are transferred from AAGAB, implying that it engages an α:σ2 hemicomplex rather than individual α and σ2 subunits. With their distinct AP2-binding properties, AAGAB and CCDC32 direct AP2 assembly to follow a precise order of subunit incorporation, which was likely evolutionarily selected to minimize misassembly.

Structural and biochemical studies are needed to establish the molecular details of chaperone–AP2 interactions and to determine whether alternative assembly routes exist. Nevertheless, it is conceivable that AAGAB and CCDC32 stabilize AP2 subunits and assembly intermediates by recognizing assembly interfaces normally hidden in mature AP2 complexes, analogous to how the chaperone AHSP engages unpaired α-globin (38). By recognizing assembly interfaces, AAGAB and CCDC32 fulfill two fundamental roles of assembly chaperones: 1) maintaining individual subunits and assembly intermediates at soluble and assembly-competent states to avoid misassembly and aggregation and 2) protecting individual subunits and assembly intermediates from degradation by the cellular quality control system, which likely also recognizes exposed hydrophobic interfaces (8, 39). In mammalian cells deficient in AAGAB or CCDC32, all AP2 subunits are degraded (Fig. 2 A and B) (10, 12). Notably, while stable in E. coli and in vitro, AAGAB–AP2 complexes do not accumulate in CCDC32 KO mammalian cells (Fig. 2 A and B), suggesting the existence of quality-control mechanisms detecting stalled AAGAB–AP2 intermediates and targeting them for degradation.

Unlike folding chaperones, an assembly chaperone selectively regulates one or a limited number of protein complexes (Fig. 2 A and B and SI Appendix, Figs. S4 and S7), reflecting its binding to target-specific sequences rather than generic hydrophobic stretches. AAGAB assists the assembly of AP1, AP2, and AP4, whereas CCDC32 selectively regulates AP2. Since adaptor protein complexes adopt a similar configuration (3, 20, 24, 37), their assembly is expected to face the same challenges in the crowded cytosol. Thus, we anticipate that AP1 and AP4 also rely on CCDC32-like chaperones to form template complexes recruiting their respective μ and β subunits. No protein exhibiting sequence homology with CCDC32 is found in the human proteome. However, such a chaperone could resemble CCDC32 in function rather than in protein sequence. While AP3 and AP5 do not require AAGAB or CCDC32 to assemble, their assembly is expected to rely on dedicated chaperones as well. Although adaptor protein complexes may rely on distinct assembly chaperones, the mechanism uncovered in this work is likely conserved in their assembly pathways, which are collectively referred to as chaperone-assisted adaptor protein assembly (CAPA). The genetic and proteomic tools that uncovered AAGAB and CCDC32 in AP2 assembly will also be instrumental in identifying assembly chaperones in other CAPA pathways.

The physiological importance of CAPA is highlighted by human diseases caused by AAGAB and CCDC32 mutations. These diseases represent a subtype of coatopathy (3), caused not by mutations in adaptor proteins themselves but by impaired assembly of otherwise normal adaptor protein subunits. In PPKP1, heterozygous AAGAB mutations are expected to subtly impact vesicle-mediated transport such that their effects are restricted to a small group of cargo proteins in selected tissues. Although AAGAB regulates three adaptor protein complexes, its PPKP1 connection is likely through AP1 because partial reduction in AP1 activities is also linked to skin diseases (40–45). Since AP2 is essential to mammalian development (46), biallelic CCDC32 mutations in CFNDS might be partially compensated for by upregulations of other factors involved in CME. A future research direction is to define which cargo proteins are sensitive to the disease-causing mutations and how their imbalances contribute to the pathogenesis of the diseases.

Materials and Methods

Cell Culture.

Mouse preadipocytes, HEK 293T cells, and HeLa cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% Fetal Bovine Essence (FBE, VWR, #10803-034) and penicillin/streptomycin (Thermo Fisher Scientific, #15140122). WTC Gen1c iPSCs were cultured on Matrigel (Corning, #354277) in mTeSR media (STEMCELL Technologies, #100-0274) supplemented with 10 µM Y-27632 dihydrochloride (TOCRIS, #1254). All cell lines were maintained in a humidified 37 °C incubator with 5% CO₂. Adipocytes were differentiated using a previously established protocol (47).

Gene KO Using CRISPR-Cas9.

CRISPR constructs and lentiviral particles were prepared as previously described (10, 47). Oligonucleotide sequences of gRNAs targeting the human CCDC32 gene are 5′-GACTCTACAGCCACAAGATC-3′ and 5′-GAAGTGTATTTAGCATCTCT-3′. Oligonucleotide sequences of gRNAs targeting the mouse Ccdc32 gene are 5′-GACTCTTCGGCTACAAAGTC-3′ and 5′-GTGCTGGGATCGATTCCTCC-3′.

Flow Cytometry.

Cells were washed with KRH buffer (12 mM HEPES, pH 7.0, 121 mM NaCl, 4.9 mM KCl, 1.2 mM MgSO₄, and 0.33 mM CaCl₂) and blocked with KRH buffer containing 5% FBE at 4 °C. To measure surface TfR, unpermeabilized cells were stained using monoclonal anti-TfR antibodies (DSHB, #G1/221/12) and allophycocyanin (APC)-conjugated secondary antibodies (Thermo Fisher Scientific, #17-4015-82). After dissociation from plates using Accutase, APC fluorescence was measured on a CyAn ADP Analyzer (Beckman Coulter).

To generate the ALFA-GLUT4-mGreenLantern reporter, an ALFA tag (SRLEEELRRRLTE) was inserted between a.a. 67 and 68 of human GLUT4, whereas a mGreenLantern fluorescent protein was fused to the C terminus of GLUT4. The ALFA tag was chosen for the reporter because it could be recognized by versatile high-affinity nanobodies (48). A DNA fragment encoding the ALFA-GLUT4-mGreenLantern reporter was subcloned into the BamHI and KpnI sites of a modified version of the pLKO.1-puro vector (Sigma-Aldrich, #SHC001), in which the puromycin resistance gene was deleted. Lentiviral particles were prepared as in CRISPR gene KO and used to infect preadipocytes. Preadipocytes stably expressing the GLUT4 reporter were differentiated into adipocytes and washed three times with the KRH buffer, followed by incubation in the KRH buffer at 37 °C for 2 h. Surface GLUT4 reporters were stained using recombinant mScarlet-tagged anti-ALFA nanobodies. Cells were dislodged using Accutase, and mGreenLantern and mScarlet fluorescence was measured on a CyAn ADP Analyzer. To calculate normalized surface levels of GLUT4 reporters, mean mScarlet fluorescence was divided by mean mGreenLantern fluorescence.

Endocytosis Assays.

Tf-ATTO565 was generated as previously described (35). For endocytosis assays, cells were washed three times with KRH buffer and incubated in KRH buffer for 30 min before being chilled on ice. Next, the cells were incubated with prewarmed 1 µg/mL Tf-ATTO565 at 37 °C for 8 min. The cells were chilled on ice and washed three times with ice-cold 150 mM MESNa (Sigma-Aldrich, #M1511) in PBS and once with ice-cold PBS. Subsequently, the cells were disassociated using Accutase, and ATTO565 fluorescence (excitation/emission: 563 nm/589 nm) was measured using the PE channel on a MACSQuant Analyzer. To measure endocytosis using confocal imaging, cells were grown on coverslips (VWR, # 89015-725) in 24-well plates. Endocytosis assays were carried out as described above. After washing with MESNa three times and once with PBS, cells were fixed with 4% paraformaldehyde (PFA) without permeabilization. After washing three times with PBS, the cells were incubated with 50 µg/mL CF405M-conjugated Concanavalin A (Biotium, Cat# 29074) at RT for 15 min to stain the cell membrane.

Immunostaining and Imaging.

Cells grown on coverslips were fixed using 4% PFA and permeabilized in PBS containing 5% FBE and 0.2% saponin. Transiently expressed 3xFLAG-tagged AAGAB was labeled using anti-FLAG M2 antibodies (Sigma-Aldrich, #F1804, RRID: AB_262044) and Alexa Fluor 488–conjugated secondary antibodies (Thermo Fisher Scientific, #A11008, RRID: AB_142165). To express ALFA-tagged CCDC32, a DNA fragment encoding human CCDC32 and a C-terminal ALFA tag was subcloned into the NheI and SalI sites of the SHC003BSD-DelGFP vector (Addgene, #133301). Transiently expressed ALFA-tagged CCDC32 was labeled using mScarlet-anti-ALFA nanobodies. Endogenous CHC was labeled using anti-CHC antibodies (Cell Signaling Technology, #2410, RRID: AB_2083156) and Alexa Fluor 647–conjugated secondary antibodies (Thermo Fisher Scientific, #A32733, RRID: AB_2633282). The endogenous AP2 α subunit was labeled using anti-α-adaptin (Santa Cruz Biotechnology, #sc-17771, RRID: AB_2274034) and Alexa Fluor 488–conjugated secondary antibodies. Endogenous β-COP was labeled using mouse monoclonal anti-β-COP antibodies (clone: CM1A10) and Alexa Fluor 568–conjugated secondary antibodies (Thermo Fisher Scientific, #A11004, AB_2534072). To visualize surface GLUT4 reporters, preadipocytes grown on coverslips were differentiated into mature adipocytes, starved in KRH for 2 h, fixed with 4% paraformaldehyde, and blocked with 5% FBE without permeabilization. Surface reporters were stained using mScarlet-anti-ALFA nanobodies, whereas nuclei were stained with Hoechst 33342 (Sigma-Aldrich, #D9642). Confocal images were captured using a 100× oil immersion objective on a Nikon A1 laser scanning confocal microscope. In SIM, cells were grown on coverslips, labeled as in confocal microscopy, and imaged using a 100× oil immersion objective on a Nikon SIM microscope.

Mass Spectrometry.

The 3xFLAG-AAGAB protein was transiently expressed in HEK 293T cells and immunoprecipitated using anti-FLAG M2 antibodies and Protein A/G agarose beads (Santa Cruz Biotechnology, #sc-2003). Proteins in the immunoprecipitates were eluted from resins using an elution buffer (50 mM Tris, pH 8.5, 5% SDS, 10 mM Tris-(2-carboxyethyl) phosphine, 40 mM 2-chloroacetamide, 2 mM biotin, and 1 mM DTT). Samples were prepared for mass spectrometry using the protein aggregation capture (PAC) method (49). Peptides were prefractionated using high-pH fractionation before they were resolved on a Thermo Ultimate 3000 RSLCnano system in a direct injection mode (49). False discovery rates were set to 0.01 for both protein and peptide identifications with a minimum peptide length of four residues and a minimum peptide number of two.

Immunoblotting and Immunoprecipitation.

Immunoblotting was carried out as previously described (10, 47). Primary antibodies used in immunoblotting included polyclonal anti-AAGAB antibodies (Bethyl Laboratories, #A305-593A RRID: AB_2782752), polyclonal anti-CCDC32 antibodies (GenScript, custom antibody generated using a synthetic peptide corresponding to a.a. 167–180 of human CCDC32), polyclonal anti-AP1-γ antibodies (Bethyl Laboratories, #A304-771A), polyclonal anti-AP1-σ antibodies (Bethyl Laboratories, #A305-396A, RRID: AB_2631787), monoclonal anti-AP2-α antibodies (BD Biosciences, #610502, RRID: AB_397868), polyclonal anti-AP2-β antibodies (Bethyl Laboratories, #A304-719A, RRID: AB_2620914), monoclonal anti-AP2-μ antibodies (BD Biosciences, #611350), polyclonal anti-AP2-σ antibodies (Abcam, #ab128950, RRID: AB_11140842), monoclonal anti-AP3-δ antibodies (DSHB, #SA4), monoclonal anti-AP4-ε antibodies (BD Biosciences, #612018), monoclonal anti-AP5-ζ antibodies (Thermo Fisher Scientific, #66533-1-IG), and monoclonal anti-α-tubulin antibodies (DSHB, #12G10, RRID: AB_1210456). Secondary antibodies used in immunoblotting included horseradish peroxidase (HRP)-conjugated anti-rabbit antibodies (Sigma-Aldrich, #A6154, RRID: AB_258284), and HRP-conjugated anti-mouse antibodies (Sigma-Aldrich, #A6782, RRID: AB_258315). FLAG-tagged proteins were directly detected using HRP-conjugated anti-FLAG M2 antibodies (Sigma-Aldrich, #A8592, RRID: AB_439702). HA-tagged proteins were directly detected using HRP-conjugated anti-HA antibodies (Roche, #12013819001, RRID: AB_390917).

Mammalian expression plasmids encoding 3xFLAG-tagged AAGAB and HA-tagged AP1/2 subunits were developed in our previous work (10, 12). To express HA- or 3xFLAG-tagged CCDC32, DNA fragments encoding human CCDC32 with a C-terminal HA or 3xFLAG tag were subcloned into the NheI and ApaI sites of the pcDNA3.1 vector (Thermo Fisher Scientific, #V79020) or the NheI and SalI sites of the SHC003BSD-DelGFP vector. IP experiments were carried out as previously described (10, 47). Endogenous CCDC32 was immunoprecipitated using custom anti-CCDC32 antibodies and protein A/G agarose beads. In GST pull-down of endogenous AP2, GST and GST-CCDC32 bound to glutathione beads were used to pull down AP2 subunits from HeLa cell extracts.

Recombinant Protein Expression and Pull-Down Assays.

Plasmids expressing His₆-SUMO-tagged AAGAB, GST-tagged AP2 α trunk domain (a.a. 1–621), and untagged σ2 subunit were described in our previous work (10, 11). Human CCDC32 was subcloned into the BamHI and XhoI sites of the pGEX-4T-3 vector (GE Healthcare) and the pET-SUMO expression vector (50, 51). AP2M1 and AP2B1, which encode full-length (FL) human µ2 subunit and the trunk domain of human β2 subunit (a.a. 1–591), were subcloned into the BglII and SalI sites of the pET-SUMO vector. To express mScarlet-tagged anti-ALFA nanobodies, cDNAs encoding the mScarlet fluorescent protein and the nanobody were subcloned into the BamHI and XhoI sites of the pET-SUMO vector. The nanobody sequence was based on a previous report (48). Recombinant proteins were expressed and purified using established procedures (10). Expression of μ2 and β2 subunits in E. coli was induced at 16 °C for 16 h using 0.2 mM IPTG.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information.