Structural basis for LMBD1-dependent trafficking and cobalamin export of ABCD4 Authors: Qiwei Liu, Xingfan Li, Yingjie Wu, Kai Zheng, Mi Zhou, Tao Long Published: 2026 ABSTRACT AbstractCorrect trafficking of lysosomal transporters is essential for intracellular homeostasis. While most lysosomal membrane proteins are directed to the lysosome via sorting motifs, the cobalamin exporter ABCD4 is distinct, instead relying on LMBD1 as a dedicated chaperone for its trafficking. Dysfunction of either protein causes inherited cobalamin metabolism disorders. Despite its physiological significance, the molecular mechanism underlying this chaperone-dependent trafficking remains unclear. Here, we report the cryo-EM structures of ABCD4 complex with LMBD1 in the lumen-open, substrate-bound and cytosol-open states. LMBD1 contains nine transmembrane-helices (TMs) and a cytosolic domain, both of which engage ABCD4. Cell imaging shows that disruption of these interactions impairs the trafficking of ABCD4 to lysosomes. Structural and biochemical analyses provide insights into cobalamin recognition and reveal conformational states associated with the proposed cobalamin transport cycle. These findings provide molecular insights into cobalamin metabolism and illustrate a chaperone-assisted mechanism that supports proper trafficking of a lysosomal transporter.Subject terms: Biochemistry, Cryoelectron microscopy, Lysosomes Introduction Lysosomes are acidic organelles that degrade biomacromolecules1. Coordinated action of lysosomal membrane proteins and hydrolases maintains macromolecule degradation, nutrient transport, and pH homeostasis. Lysosome biogenesis requires continuous transport of newly synthesized proteins through the ER–Golgi–endosome pathway, guided by sorting signals and recognition factors. Soluble lysosomal proteins are typically directed via the mannose-6-phosphate receptor–mediated pathway, which utilizes specific transport adapters for targeting cellular compartments2. Most lysosomal membrane proteins are guided by tyrosine- or di-leucine–based sorting motifs within their cytosolic domains, recognized by adapter protein complexes (AP1–AP4) that recruit clathrin to form coated transport vesicles3. In contrast to these conventional mechanisms, the cobalamin (vitamin B12) transporter ABCD4 lacks any known sorting motifs and depends on a membrane protein named LMBD1 for its lysosomal localization, highlighting a distinct chaperone-mediated trafficking pathway. Cobalamin is a cofactor for 5-methyltetrahydrofolate-homocysteine methyltransferase (MTR) and mitochondrial methyl-malonyl-CoA mutase (MMUT), which serve key wide-ranging metabolic functions in mammalian cells4. Since humans can only obtain cobalamin from their diet, more than 20 proteins are encoded in an intricate pathway that enables the efficient delivery of cobalamin from its absorption, to the various tissues and cellular compartments5. Somatic complementation analysis has defined eleven known defects in the intracellular processing of cobalamin, which are classified into the following groups: cblA–cblG, cblJ, cblK, cblX, and mut6–8. Among these classes, cblF and cblJ, which are caused by mutations in LMBRD1 and ABCD4, respectively, share a similar deficiency phenotype that is the result of a loss of cobalamin transport out of the lysosome7,9. ABCD4 is an ATP-binding cassette (ABC) transporter that belongs to the D subfamily of ABC half transporters. It mainly functions as a homodimer, with each monomer containing one transmembrane domain (TMD) and one nucleotide-binding domain (NBD)10. To date, four ABCD transporters (ABCD1-4) have been identified in humans11–14. ABCD1-3 are located on the peroxisomal membrane and function in the transport of various acyl-CoAs from the cytosol into peroxisomes for further β-oxidation14–19. Specifically, ABCD1 and ABCD2 are involved in the transport of CoA esters of saturated, monounsaturated and polyunsaturated fatty acids20–22, while ABCD3 transports dicarboxylic acid, branched chain acyl-CoA, and the bile acid intermediates: di- and tri-hydroxycholestanoyl (DHCA and THCA)-CoA23. On the other hand, ABCD4 is a lysosomal protein that plays a key role in the transport of cobalamin from the lysosomal lumen to the cytosol, and dysfunction of this protein results in the accumulation of cobalamin in the lysosomal lumen7. LMBD1, encoded by LMBRD1, is an integral multi-pass membrane protein with nine putative TMs. Though LMBD1 mainly localizes to the lysosomal membrane through its AP-2 binding motif (233YERL)9,24,25, it has also been found in the plasma membrane and perinuclear region26. LMBD1 regulates a distinct clathrin-mediated endocytosis of the insulin receptor by interacting with AP-2 on the plasma membrane and mediates microtubule assembly during the process of cell mitosis27,28. Additionally, LMBD1 is essential for the expression of fibroblast growth factor 8, which ensures proper gastrulation and mesoderm formation during mouse embryogenesis28. The functional connection between LMBD1 and ABCD4 was initially established by evidence of their colocalization on the lysosomal membrane7, and their direct interaction was later demonstrated by another study29. These two lysosomal membrane proteins also form a super-complex with the downstream cytoplasmic cobalamin-processing protein MMACHC to mediate the directional transfer of cobalamin6,29,30. ABCD4 can transport cobalamin from the interior to the exterior of lysosomes in a manner that was dependent on its ATPase activity, while LMBD1 showed no cobalamin transport activity and does not affect ABCD4-mediated transport in vitro31. Endogenous ABCD4 is localized to both the lysosomes and the ER. When LMBRD1 is knocked out, the lysosomal localization of ABCD4 is impaired25, suggesting that the translocation of ABCD4 from the ER to lysosomes depends on the lysosomal targeting ability of LMBD1. An earlier cryo-EM structure solved the ATP-bound ABCD4E549Q mutant at 3.6 Å resolution and revealed a lysosome-open conformation, in which the transmembrane domains form a cavity accessible to the lysosomal lumen while the nucleotide-binding domains dimerize with two bound ATP molecules32. This structure provided a structural framework for understanding ABCD4 function, but did not address how ABCD4 interacts with LMBD1 or how cobalamin is exported from the lysosome. Here, we determine the structures of the ABCD4–LMBD1 complex by cryo-EM in three distinct states: lumen-open state at 2.80 Å, substrate-bound state at 2.83 Å and cytosol-open state at 2.98 Å. LMBD1 employs its TMs and cytosolic domain to interact with ABCD4. In contrast to other ABCD family members that coordinate two substrate molecules within a dimer33–38, only a single cobalamin molecule is observed in the substrate-binding cavity of the ABCD4 homodimer in our structure. Our structural and functional analyses establish a structural framework for understanding the working mechanism of the ABCD4–LMBD1 complex, shed light on the molecular basis of disease-associated mutations, and offer atomic-level insights into cobalamin metabolism and chaperone-assisted trafficking of lysosomal transporters. Results The structure of the ABCD4–LMBD1 complex in a lumen-open state To facilitate structural characterization of the ABCD4–LMBD1 complex, we used the ABCD4E549Q mutant, which impairs ATP hydrolysis while preserving ATP binding and stabilizes the transporter in a defined conformational state. We co-expressed Flag-tagged ABCD4E549Q and Strep-tagged LMBD1 in HEK-293F cells, and purified the proteins using Strep-Tactin resin followed by anti-Flag M2 resin. Gel filtration chromatographs revealed that ABCD4E549Q and LMBD1 formed a stable complex that migrated as a single peak (Supplementary Fig. 1), which was suitable for further cryo-EM structural studies. The cryo-EM structure of the ABCD4–LMBD1 complex was determined at 2.80 Å resolution (Fig. 1a–c, Supplementary Figs. 2, 3, and Supplementary Table 1). The cryo-EM structure of the ABCD4–LMBD1 complex in a lumen-open state. a, b Cartoon representation of the overall structure of ABCD4–LMBD1 complex in a lumen-open state from the side of the membrane. The two subunits of ABCD4 and LMBD1 are colored in light pink (protomer A), light blue (protomer B), and pale green, respectively. The six TMs of each subunit are numbered. c Comparison of the ABCD4 structure in the complex (light pink/light blue) and the previously determined structure of ABCD4 alone (gray, PDB: 6JBJ). d Topology of LMBD1 depicting the N- and C-termini, with the TM helices numbered, and the cytosolic helix (CH) labeled. e View of LMBD1 from the side of the membrane (upper) and from the lumen (lower). f Interaction details of the membrane interface between ABCD4 and LMBD1. CLR, cholesterol. g Interaction details of the cytosolic interface between ABCD4 and LMBD1. Residues are represented as sticks; a dashed line represents hydrophilic interactions. The conformation of ABCD4 in complex with LMBD1 adopts a lumen-open state, closely resembling the previously reported structure of ABCD432, with a Cα root-mean-square deviation (RMSD) of 0.977 Å (Fig. 1c). ABCD4 forms a homodimer, with each protomer containing six TMs followed by one NBD. Two endogenous ATP molecules and magnesium ions were observed in the NBDs, despite the absence of supplemented ATP and magnesium during sample preparation. (Supplementary Fig. 3d). During data processing, an additional density was observed on each ABCD4 protomer, suggesting that the complex adopts a dimeric arrangement with one LMBD1 protein per ABCD4. Thus, we attempted to introduce C2 symmetry during data processing. However, the cryo-EM densities of LMBD1 were weaker as the local resolution of the whole complex became higher (Supplementary Fig. 2a). Consequently, we refined the data using C1 symmetry and applied a focused mask on the transmembrane regions of one ABCD4 protomer and LMBD1 for 3D classification without image alignment. This approach resolved four similar conformations of LMBD1 with a 2:1-ABCD4:LMBD1 stoichiometry in the cryo-EM maps, only one of which enabled atomic model building for the whole complex (Supplementary Fig. 2a). To further explore the binding stoichiometry between ABCD4 and LMBD1, we performed a chemical cross-linking assay. Consistent with our cryo-EM analysis, the results indicate the coexistence of both 2:1 and 2:2 stoichiometries in the ABCD4–LMBD1 complex, with the 2:1 assembly being the predominant form (Supplementary Fig. 4). In our structure, LMBD1 consists of 9 TMs and a cytosolic α-helix bundle (designated as CH1 and CH2) with the dimensions of ~50 Å × 40 Å × 100 Å (Fig. 1d, e). Due to flexibility and limited local resolution, several loops between the TMs, including the putative glycosylation sites and the AP-2 binding motif, were not visible and therefore were not modeled. A DALI search39 for structural homologs reveals that the overall structure of LMBD1 does not exhibit any significant fold similarity with previously characterized membrane protein structures, implying that LMBD1 presents, to the best of our knowledge, a previously unreported fold. The cryo-EM map of the lumen-open structure complex revealed two interfaces between ABCD4 and LMBD1: (1) TM4 and the β-strands between TM7 and TM8 of LMBD1 interact with the TMD of ABCD4 (Fig. 1f and Supplementary Fig. 5a); (2) the cytosolic bundle of LMBD1 forms a complex with the NBD of ABCD4 (Fig. 1g and Supplementary Fig. 5b). In the cytosolic interface, Pro260/Leu261/Pro262 in the loop between the two α-helix contribute hydrophobic contacts with five hydrophobic residues (Pro453, Leu478, Tyr482, Pro483, and Tyr539) of ABCD4 (Fig. 1g). In addition to these hydrophobic interactions, the side chains of three polar residues (Arg259, Arg264 and Lys270) and the main chain of Pro260 in LMBD1 form hydrogen bonds with four residues of ABCD4: Asp450, His454, Glu480, and Gln541 (Fig. 1g and Supplementary Fig. 5b). Furthermore, residue Lys270 also forms a π-cation interaction with Tyr482 of ABCD4. In the membrane interface, Leu156/Val157/Phe160 in TM4 and Trp391/Trp394 in the β-strand region engage in hydrophobic interactions with Phe84 in TM2 and Leu41/Met42 in TM1 of ABCD4, respectively (Fig. 1f and Supplementary Fig. 5a). Additionally, one cholesterol molecule was observed in the membrane interface, which may also contribute to the hydrophobic interactions between LMBD1 and ABCD4 via Phe393 of LMBD1 and Phe99 of ABCD4 (Fig. 1f and Supplementary Fig. 5a). The substrate-bound state By adding cobalamin, we obtained the structure of the substrate-bound ABCD4E549Q–LMBD1 complex (Fig. 2a, b). We determined the cobalamin-bound structure at a resolution of 2.83 Å without symmetry (Fig. 2a, b and Supplementary Figs. 6, 7, and Supplementary Table 1). Akin to the lumen-open form, only one LMBD1 was clearly resolved in the cryo-EM map. Structural alignment of ABCD4 in this complex with the lumen-open form revealed a RMSD of 0.468 Å across 738 Cα atoms (Fig. 2c). The NBDs exhibit minimal conformational changes relative to the lumen-open structure, likely due to the retention of endogenous ATP molecules and magnesium ions (Supplementary Fig. 7b). Upon substrate binding, TM1 and TM6 tilt inwards, forming a cavity that is closed facing the lysosomal lumen (Fig. 2c). In this cavity, we observed a plate-shaped feature with an approximate volume of 2000 ų in the cryo-EM map (Fig. 2b). Docking attempts further confirm that cobalamin fits within the observed cryo-EM density, supporting its binding at this site. Cobalamin, in the base-on conformation with its cobalt ion coordinated by the dimethylbenzimidazole group, is positioned at the dimer interface of ABCD4 (Fig. 2a, b, d). Several residues, including Val60/Pro64/Tyr68 in TM1, Tyr286 in TM5, and Phe315/Ile318/Tyr319 in TM6 of protomer A, as well as Tyr68 in TM1, Tyr279/Tyr286 in TM5, and Phe315/Ile318/Tyr319/Ile321 in TM6 of protomer B, establish hydrophobic interactions with cobalamin (Fig. 2e, f). In addition, a number of hydrogen bonds link cobalamin to ABCD4. Notably, residues Asn93 in TM2 from protomer A as well as Asp278 in TM5, and Ser322 and Asp329 in TM6 from protomer B contribute to stabilizing the three side chains of cobalamin’s corrin macrocycle through hydrophilic contacts (Fig. 2e, f). Furthermore, Sequence alignment demonstrates that the residues Val60, Pro64, Asp278, Tyr279, Tyr286, Phe315, Tyr319, and Asp329 are highly conserved across various species (Supplementary Fig. 8), highlighting their essential role in cobalamin recognition. Structure of the substrate-bound ABCD4–LMBD1 and the substrate-binding cavity. a, b Side view of the overall structure of the cobalamin-bound ABCD4–LMBD1 complex. The two subunits of ABCD4 and LMBD1 are colored in light pink (protomer A), light blue (protomer B), and pale green, respectively. The six TMs of each subunit are numbered. The cobalamin (Cbl) molecule is displayed as yellow sticks, with its corresponding cryo-EM density shown at contour level 0.257. c Structural comparison of ABCD4 between the substrate-bound state (light pink/light blue) and the lumen-open (gray) state, viewed from the luminal side. The conformational changes are indicated by arrows. d Electrostatic surface representation of ABCD4 in substrate-bound state from the side of the membrane. The cobalamin molecule is shown by yellow sticks. e Interaction details of the cobalamin-binding site. Residues are represented as sticks; a dashed line represents hydrophilic interactions. f Schematic diagram of ligand interactions between ABCD4 and cobalamin. Polar interactions are shown as red dashed lines. g Relative ATPase activities of ABCD4–LMBD1 complexes with the indicated ABCD4 mutants upon addition of cobalamin. Relative activity reflects substrate-enhanced ATPase activity of ABCD4 or its variants. Data represent the mean of three independent experiments, and error bars indicate mean ± S.D. (n = 3). Source data are provided as a Source Data file. To validate our structural observations, we generated several ABCD4 variants and evaluated their activities using a cobalamin-enhanced ATPase assay (Supplementary Fig. 1c, d). Compared with the wild type, mutants that disrupt hydrophobic interactions with the corrin ring (F315A and P64W/I318W) or polar interactions (D278A/D329A) exhibited reduced substrate-enhanced ATPase activity (Fig. 2g). Furthermore, previous studies demonstrated that the pathogenic Tyr319Cys mutation abolishes both ATPase and cobalamin transport activities, whereas substituting Tyr319 with the aromatic residue Phe restores both functions in ABCD431. Additionally, replacing Asp329 with Lys significantly impairs cobalamin transport without affecting ATPase activity40. These findings support our structural observation that both hydrophobic contacts and a hydrogen-bonding network at the ABCD4 dimer interface are essential for efficient cobalamin engagement. Distinct from ABCD4, other members of the ABCD family, such as ABCD1 and ABCD3, contain two substrate-binding sites within the dimer, with each protomer coordinating one substrate molecule (Supplementary Fig. 9a). In ABCD1 and ABCD3, the substrate-binding sites are located on opposite sides of the dimer, whereas in ABCD4 the cobalamin-binding site is positioned at the center of the dimer interface (Supplementary Fig. 9b). Sequence alignment reveals that acyl-CoA–binding residues are conserved in ABCD1–ABCD3 but absent in ABCD4, whereas residues mediating cobalamin binding are distinctively conserved in ABCD4 (Supplementary Fig. 9c), underscoring its specialization for cobalamin recognition. The cytosol-open state ATP hydrolysis is proposed to trigger conformational changes and promote the separation of the two NBDs in ABCD133–37 and ABCD338. Given that wild-type ABCD4 exhibits higher ATPase activity32, which can be further enhanced by cobalamin31, we aimed to purify the wild-type ABCD4 complex with LMBD1 (Supplementary Fig. 1b, c). The resulting protein complex was incubated with cobalamin prior to cryo-EM grid preparation. We determined the structure at 2.98 Å resolution without imposing symmetry (Fig. 3a, b and Supplementary Figs. 10 and 11, and Supplementary Table 1). Similar to the above states, only one LMBD1 subunit was well-resolved in the cryo-EM map. The cytosol-open structure of the ABCD4–LMBD1 complex. a, b Cartoon structures representing ABCD4 in complex with LMBD1 in a cytosol-open state, viewed from the side of the membrane. The two subunits of ABCD4 and LMBD1 are colored in light pink (protomer A), light blue (protomer B), and pale green, respectively. The six TMs of each subunit are numbered. c Structural comparison of ABCD4 (light pink/light blue) and ABCD3 (light orange), viewed from the side of the membrane. d Structural comparison of ABCD4 dimers between the cytosol-open state (light pink/light blue) and the substrate-bound state (gray), viewed from the side of the membrane. The conformational change is indicated by arrows. e Structural comparison of ABCD4 monomers between the cytosol-open state (light pink) and the substrate-bound state (gray), viewed from the side of the membrane. The conformational change is indicated by the arrow. The distance between TM4-TM5 bundles and other TM bundles is shown by dash line. f An extra density (yellow) is observed in the cytosol-open state at the same position as cobalamin in the substrate-bound state. In this structure, dimerization of the NBDs was disrupted, with the complete separation of the Walker A and the signature motifs. An ADP molecule was identified bound to the Walker A motif of one NBD subunit (Supplementary Fig. 11b). The overall fold of ABCD4 in a cytosol-open conformation is similar to the cytosol-facing conformation of ABCD3 with an RMSD of 3.832 Å (Fig. 3c). TM1, TM2, TM3, and TM6 of one protomer interact with TM4 and TM5 of an adjacent protomer, creating a V-shaped cleft. Compared with the substate-bound state, the cytosol-open state exhibits a scissor-like movement, accompanied by a ~48 Å displacement of the interwoven TM4 and TM5 of one protomer away from the other helices (Fig. 3d, e). Although a notable conformational change occurs, the TM1-TM2, TM3-TM6, and TM4-TM5 pairs from both states largely retain their integrity, moving as rigid bodies and maintaining their overall conformations (Fig. 3d, e). The resulting conformation generates a wide cytosolic-facing opening, facilitating substrate release into the cytosol. Remarkably, we observed a large cryo-EM density at a position similar to that of cobalamin in the substrate-bound state (Fig. 3f). However, due to the limited local resolution, we were unable to unambiguously determine the precise conformation of cobalamin. These observations further support the notion that cobalamin is released from this site in the cytosol-open conformation. The binding of LMBD1 to ABCD4 Cryo-EM data processing revealed that LMBD1 adopts multiple conformations in the complex, especially in the lumen-open state (Supplementary Fig. 2a). Analysis of the four identified cryo-EM maps reveals that the interface between the cytosolic bundle of LMBD1 and the NBD of ABCD4 remains conserved, while the TMD of LMBD1 exhibits variable rotational orientations (Supplementary Fig. 3b). This binding flexibility may be due to the limited contacts of the membrane interface. Only the end of TM4 and the β-sheet between TM7 and TM8 of LMBD1 contribute two major hydrophobic interactions, while the loop between the cytosolic helix bundle inserts into the NBD of ABCD4, forming both hydrophobic and hydrophilic interactions (Fig. 1f, g and Supplementary Fig. 5). Similar flexibility of LMBD1 was also observed across different states. Superimposition of one protomer among these states shows that while the TMD of LMBD1 adopts distinct rotational positions, the interactions between ABCD4 and LMBD1 remain the same (Fig. 4a, b). The trafficking of ABCD4 relies on its interaction with LMBD1. a, b Structural comparison of the ABCD4 protomers between the lumen-open state (gray and pale cyan) and the substrate-bound (gray and yellow) state (a) and between the substrate-bound (gray and yellow) and the cytosol-open (gray and light blue) state (b), viewed from the side of the membrane. The conformational changes are indicated by arrows. c, d Co-IP assay was performed in HEK-293T cells using wild-type ABCD4 or its mutants and wild-type LMBD1-GFP or its mutants that were tagged with HA or Flag tags, respectively. Pulldowns were performed with anti-HA beads, and Western blots were performed to detect Flag-tagged proteins in the immunoprecipitation (IP) and input samples. Experiments were repeated three times with similar results. Mutants of ABCD4 in (c) were generated: Mutant1 (D450A, H454A, and Y482A), Mutant2 (M42A, F84A, and F99A), Mutant3 (M42A, F84A, F99A, D450A, and Y482A) and the single mutants D450A, H454A, Y482A, and Q541A. Mutants of LMBD1 in (d) were generated: the triple mutants TPM1 (F160A/W391A/W394A), TPM2 (R259A/R264A/K270A), and the single mutants R259A, R264A, and K270A. Source data are provided as a Source Data file. e Subcellular localization of wild-type ABCD4 or its mutants, which were tagged with HA, co-expressed with LMBD1-GFP or empty vector (control, with wild-type ABCD4) in HEK-293T cells. The distribution of ABCD4 or its mutants, detected by immunofluorescence staining with an anti-HA antibody (red), was compared with the localization of lysosome marker LAMP1 (blue). Scale bar, 10 μm. f Quantification of Pearson correlation coefficient between HA and LAMP1 signals as shown in (e). Values were obtained from the regions of interest that represent ABCD4-HA expressing cells and quantified from ten randomly selected fields (mean ± S.E.M.; n = 10); ns, no significance; ∗∗∗∗p < 0.0001, by one-way ANOVA with Dunnett’s multiple comparison test (p < 0.0001, p = 0.0556, p < 0.0001 from left to right). Representative data from one of three independent experiments are shown. Source data are provided as a Source Data file. To validate the binding interface between LMBD1 and ABCD4, we generated several ABCD4 variants: mutant1 (D450A/H454A/Y482A), mutant2 (M42A/F84A/F99A), mutant3 (M42A/F84A/F99A/D450A/Y482A) and the single mutants D450A, H454A, Y482A and Q541A. The interactions between these variants and LMBD1 were assessed by co-immunoprecipitation (co-IP) assay. HEK-293T cells were co-transfected with LMBD1-GFP-Flag and different HA-tagged ABCD4 variants. Cells were lysed in CA-630 and subsequently subjected to co-IP using anti-HA antibody-conjugated beads. In line with a previous study25, LMBD1-GFP-Flag was co-immunoprecipitated with wild-type ABCD4-HA (Fig. 4c). Mutant2, D450A, and H454A retain the ability to bind LMBD1, although their interactions are weaker compared to the wild type. By contrast, mutant1, mutant3, Y482A and Q541A demonstrated significantly impaired binding to LMBD1 (Fig. 4c). We next performed complementary mutagenesis on LMBD1, generating two triple mutants TPM1 (F160A/W391A/W394A), TPM2 (R259A/R264A/K270A), and the single mutants R259A, R264A, and K270A. Co-IP assays showed that TPM1, R264A, and K270A retained binding to ABCD4 but with reduced interaction compared with wild-type LMBD1, whereas TPM2 and R259A exhibited markedly impaired binding (Fig. 4d). Since Met42, Phe84 and Phe99 in ABCD4 and Phe160, Tyr391, and Tyr394 in LMBD1 are responsible for interactions at the membrane interface (Fig. 1f and Supplementary Fig. 5a), whereas Tyr482 and Gln541 in ABCD4 and Arg259 in LMBD1 are involved in mediating interactions at the cytosolic interface (Fig. 1g and Supplementary Fig. 5b), it is tempting to speculate that the cytosolic interface plays a more critical role in the interaction between LMBD1 and ABCD4 than the membrane interface. The major role of LMBD1 is to mediate the transport of ABCD4 from the ER to the lysosomes25. To further validate the importance of LMBD1 binding in ABCD4 trafficking, we investigated the distribution of HA-tagged ABCD4 variants co-expressed with LMBD1-GFP-Flag using fluorescence microscopy. ABCD4 localizes to the lysosome when co-expressed with LMBD1, whereas its distribution does not overlap with lysosomes when co-expressed with the control plasmid (Fig. 4e, f). Mutant2, when co-expressed with LMBD1, displayed a punctate distribution that colocalized with the lysosomal marker LAMP1. However, mutant1, mutant3, Y482A and Q541A exhibited a reticular distribution pattern that did not overlap with LAMP1, indicating defective lysosomal localization (Fig. 4e, f and Supplementary Fig. 12a, b). Consistently, analysis of LMBD1 variants showed that TPM1 retained the ability to transport ABCD4 to lysosomes, whereas TPM2, R259A, R264A, and K270A impaired ABCD4 lysosomal localization (Supplementary Fig. 12c, d). Together with the co-IP results, these findings indicate that the cytosolic interface is critically important for the translocation of ABCD4 to lysosomes in conjunction with LMBD1. Discussion So far, only a few lysosomal transporters are known to require chaperones for lysosomal translocation, including ABCD4 and LAT1–4F2hc leucine transporter. ABCD4 acts as the cobalamin transporter, forming a complex with LMBD1 in the ER to facilitate its trafficking to the lysosome25,31. LAT1–4F2hc mediates the influx of essential amino acids, while LAPTM4b recruits LAT1–4F2hc to lysosomes for leucine uptake, promoting mTORC1 activation via the V-ATPase41–44. However, the molecular mechanism of chaperone-mediated targeting of these transporters to lysosomes remains unclear. In this study, our structures of the ABCD4–LMBD1 complex reveal the atomic details of their interactions, shedding light on the chaperone-dependent trafficking mechanism to the lysosome. LMBD1 interacts with ABCD4 through two primary interfaces: a cytosolic bundle that forms hydrophobic and polar interactions with the NBD of ABCD4, and a membrane interface where TM4 and β-strands of LMBD1 interact with the TMD of ABCD4 (Fig. 1f, g). These interactions help stabilize the ABCD4–LMBD1 complex, which further facilitates the proper localization of ABCD4 to the lysosome, as shown by co-IP assay and fluorescence microscopy (Fig. 4c–f and Supplementary Fig. 12). Mutations in ABCD4 at these interfaces, especially at the cytosolic interface, disrupt the interactions between ABCD4 and LMBD1, impairing ABCD4 trafficking and preventing its efficient delivery to the lysosome. The structures of the ABCD4–LMBD1 complex resemble those of ABC transporters containing an additional N-terminal transmembrane domain, known as TMD010. Structural comparisons with previously reported TMD0-containing proteins, including MRP145, Ycf146, SUR147, ABCB648, and ABCC349, reveal that the transmembrane segments of LMBD1 mimic the TMD0 domain found in these proteins, although the TMD of LMBD1 and TMD0s adopt distinct poses relative to the ABC core (Fig. 5a–f). Despite limited evidence, studies on SUR, TAP (ABCB2/3), and TAPL (ABCB9) provide insights into the function of TMD0. In SUR and TAP, TMD0 mediates interactions with the Kir6.2 tetramer and tapasin, forming potassium channels and peptide-loading complexes, respectively50,51. In TAPL, TMD0 deletion does not affect transport activity but causes mistargeting from the lysosome to the plasma membrane52. Thus, LMBD1 resembles TMD0 in TAPL, lacking transport activity itself but regulating ABCD4 lysosomal trafficking. Moreover, LMBD1, being highly glycosylated, may protect the non-glycosylated ABCD4 from lysosomal proteolysis through complex formation, as the high glycosylation of lysosomal membrane proteins within the lumen helps them evade degradation3,25,29,53. Structural comparison of ABCD4–LMBD1 with TMD0-containing ABC transporters and working model for the ABCD4-mediated cobalamin translocation. a Overall structure of MRP1 (PDB code: 5UJ9). The TMD0 and ABC core are colored in wheat and gray, respectively. b Overall structure of Ycf1 (PDB code: 7MPE). The TMD0 and ABC core are colored in light yellow and gray, respectively. c Overall structure of SUR1 (PDB code: 6PZB). The TMD0 and ABC core are colored in pale cyan and gray, respectively. d Overall structure of ABCB6 (PDB code: 7D7N). The TMD0 and ABC core are colored in light blue and gray, respectively. e Overall structure of ABCC3 (PDB code: 8HVH). The TMD0 and ABC core are colored in light pink and gray, respectively. f Structural comparison of ABCD4–LMBD1(pale green/gray) with MRP1 (wheat/gray), Ycf1 (light yellow/gray), SUR1 (pale cyan/gray), ABCB6 (light blue/gray) and ABCC3 (light pink/gray), viewed from the exterior side. g Diagram depicting the proposed transport cycle of ABCD4, inferred from the three resolved states. Initially, ABCD4 forms a dimer with ATP bound to its NBDs, and the translocation cavity faces the lysosomal lumen (lumen-open state). Upon cobalamin binding, the transporter shifts to a substrate-bound state, in which its transmembrane domains close to trap the substrate. ATP hydrolysis then triggers separation of the NBDs, inducing a conformational change that opens the cavity to the cytosol (cytosol-open state), and facilitating the substrate release into the cytosol. Finally, the release of ADP and rebinding of ATP resets ABCD4 to the lumen-open state, and ready for another transport cycle. ABCD4 and LMBD1 are colored in blue/pink and green, respectively. Our structures of the ABCD4–LMBD1 complex in three distinct conformational states, together with functional activity assays, allow us to propose a mechanistic model for ABCD4-mediated cobalamin transport (Fig. 5g). Initially, in the lumen-open state, ABCD4 exists as a dimer with ATP bound to its NBDs. At this stage, the membrane cavity opens toward the lysosomal lumen. Cobalamin binding triggers a conformational change that leads to the formation of the substrate-bound state. In this state, the transmembrane domains of ABCD4 close inward, creating a sealed cavity that traps cobalamin within the protein complex. Following ATP hydrolysis, the NBDs of ABCD4 separate, triggering conformational changes associated with substrate release. This hydrolytic event induces further conformational changes in the transporter, disrupting the dimer interface and opening the cavity to the cytosol, resulting in the cytosol-open state. The movement of the transmembrane-helices, particularly the shift of TM4 and TM5, generates a large opening that facilitates the release of cobalamin into the cytosol. This step is essential for the active transport of cobalamin out of the lysosome. Overall, this cycle highlights the efficiency of ABCD4 as a solute cargo exporter, essential for the movement of cobalamin from the lysosome to the cytosol, which is a crucial step in cellular cobalamin homeostasis. Furthermore, the structural determination of the ABCD4–LMBD1 complex provides a framework for mapping disease-related mutations associated with cblF and cblJ disorders (Supplementary Fig. 13). Mutations located within the cobalamin-binding pocket may directly impair cobalamin transport, while those at the cytosolic interface between ABCD4 and LMBD1 may disrupt their interaction and lead to mislocalization of ABCD4. Thus, our structures offer a foundation for future biophysical, biochemical, and cellular investigations into the molecular basis of cobalamin defects. Methods Protein expression and purification The cDNA of human ABCD4 (Protein ID: AAH12815.1) was cloned into pEGBacMam with a N-terminal Strep-tag or pHL-sec with a N-terminal Flag-tag. The cDNA of human LMBD1 (Protein ID: AAH47073.1) was cloned into pEGBacMam with a C-terminal Flag-tag or pHL-sec with an N-terminal Strep-tag. Point mutations were introduced into the coding region of ABCD4 and LMBD1 by two-step PCR. The coding region of each plasmid was sequenced to ensure the integrity of the construct. For the ABCD4E549Q–LMBD1 complex, ABCD4E549Q and LMBD1 were co-expressed in HEK-293F cells using the transient transfection system. HEK-293F cells were cultured in 293F Hi-exp medium (OPM) at 37 °C in 8% CO₂ conditions. Cell transient transfection was performed when the cell density reached ~3 × 106 cells/ml. PEI (1.28 mg, Polyscience) was mixed with 480 µg total plasmids (240 µg each of pHL-sec-ABCD4E549Q and pHL-sec-LMBD1) in 36 ml fresh medium. After incubation for 20 min, the mixture was added to 400 ml of cells. After growing at 37 °C for 8 h, 20 ml of supplementary feeding medium (OPM) and 1.62 g glucose were added per 400 ml cells and the cells were cultured for an additional 36 h before harvesting. For the ABCD4WT–LMBD1 complex, ABCD4WT and LMBD1 were co-expressed in HEK-293S GnTI− cells (ATCC) using the baculovirus system. To generate baculovirus, 5 μg bacmid was transfected into 1 × 107 sf9 cells. After total 72 h, the P1 virus was harvested. For the P2 virus, 1 ml P1 was added to 50 ml 3 × 106 sf9 cells, and then 50 ml medium was added the next day. After total 72 h, the P2 virus was harvested. 800 ml of HEK-293S GnTI− cells at a density of 3 × 106 cells/ml were infected with 50 ml of P2 virus. After 8 h, 10 mM sodium butyrate was added, and the cells were cultured for an additional 36 h before harvesting. Protein purification was performed as previously described with a few modifications54. For the ABCD4E549Q–LMBD1 complex, cells were disrupted by sonication in buffer A (20 mM HEPES, pH 7.5, and 150 mM NaCl) supplemented with 1 mM phenylmethylsulfonyl fluoride and 5 μg/ml leupeptin. After low-speed centrifugation (4500 rpm, rotor: M-S4-400, RWD, 4 °C, 5 min), the resulting supernatant was incubated in buffer A with 1% (w/v) lauryl maltose neopentyl glycol (LMNG) (Anatrace) for 1 h at 4 °C. The lysate was then centrifuged at high speed (18,000 rpm, rotor: R30AT-0808, Eppendorf, 4 °C, 25 min) to remove insoluble components, and the supernatant was loaded onto a Strep-Tactin affinity column (IBA). After washing with two column volumes, the protein was eluted in 20 mM Hepes, pH 7.5, 150 mM NaCl, 50 mM biotin, 0.01% LMNG, and the eluate was then loaded onto the anti-Flag M2 affinity resin (Sigma-Aldrich). After two tandem washes, the protein was eluted in buffer A supplemented with 100 μg/ml 3×Flag peptide (Shanghai Apeptide Co., Ltd), 0.01% (w/v) LMNG and concentrated by a 30-kDa ultrafiltration centrifuge filter (Membrane Solutions LLC, UTPES150030N). The concentrated protein was purified by a Superose 6 Increase 10/300 size-exclusion chromatography column (GE Healthcare) in buffer A with 0.005% (w/v) glyco-diosgenin (GDN). Finally, the peak fractions were pooled and concentrated, then directly flash frozen in liquid nitrogen and stored at −80 °C for cryo-EM sample preparation. For ATPase activity measurement, 0.005% (w/v) glyco-diosgenin (GDN) was substituted by 0.01% LMNG (w/v), and 0.001% CHS (w/v). The ABCD4WT–LMBD1 complex was first purified using anti-Flag M2 affinity resin and then reconstituted into nanodiscs using a similar procedure as previously described55. Soy polar lipid (Avanti, 541602C-100mg) in chloroform was dried in vacuum overnight to remove the solvent. The lipid layer was resuspended in buffer A to reach a final concentration of 10 mg·ml−1 and sonicated in a water bath at room temperature for 30 min. Membrane scaffold protein E3D1 (MSPE3D1) was expressed and purified from Escherichia coli as previously described56. Purified ABCD4WT–LMBD1 protein was mixed with MSPE3D1 and soy polar lipids. The molar ratio of ABCD4WT–LMBD1 to MSPE3D1 to total lipids was 1:4:300. The mixture was rotated for 1 h and 50 mg Bio-Beads SM-2 (Bio-Rad, 1523920) was added per 1 ml mixture. After 2 h, the old beads were replaced by new ones and the mixture was continuously rotated overnight at 4 °C. The mixture was loaded onto a Strep-Tactin affinity column (IBA) to separate assembled ABCD4WT-LMBD1 nanodiscs from the empty nanodiscs. After two tandem washes, the protein was eluted in buffer A supplemented with 50 mM biotin, and concentrated by using Amicon Centrifugal Filters. The concentrated protein was purified by a Superose 6 Increase 10/300 size-exclusion chromatography column in buffer A. Finally, the peak fractions of assembled ABCD4WT-LMBD1 nanodiscs were pooled, concentrated and stored at −80 °C for cryo-EM sample preparation. EM sample preparation and imaging A volume of 3 μl of the protein complex sample at ~10 mg·ml−1 was added to Quantifoil R1.2/1.3 400 mesh Au holey carbon grids or Quantifoil R1.2/1.3 300 mesh Cu holey carbon grids (Quantifoil), blotted using a Vitrobot Mark IV (FEI) and frozen in liquid ethane. To obtain substrate-bound and cytosol-open states, cobalamin was added into the sample at a final concentration of 666 μM before grid freezing. The grids were imaged in a 300 kV Titan Krios (FEI). For the complex in lumen-open and substrate-bound states, raw movie stacks were collected with a Gatan K3 Summit direct electron detector (Gatan) at 0.824 Å per pixel and a nominal defocus range of −1.2 to −1.8 μm. Each stack of 40 frames was exposed for 2.16 s, with an exposure time of 0.054 s per frame. The total dose for each stack was ~50 electrons per Å2. For the complex in cytosol-open, raw movie stacks were collected with a Falcon 4i camera at 0.738 Å per pixel and a nominal defocus range of −0.8 to −1.8 μm. The exposure time for each micrograph was 4 s with a total dose of ∼60 electrons per Å2. Imaging processing For the complex in lumen-open and substrate-bound states, dark-subtracted images were first normalized by gain reference (0.824 Å per pixel). Drift correction, CTF, Blob or template picking and particle extraction were performed and estimated using CryoSPARC57. Subsequent two-dimensional classification, multi-class ab initio modeling, heterogeneous three-dimensional (3D) refinement, and nonuniform refinement were performed in v4.4.0. All the masks for the refinement were automatically generated by CryoSPARC during the nonuniform refinement and local refinement. Notably, we identified four distinct structural classes in the lumen-open states (Supplementary Fig. 2a). The first class revealed an ABCD4 dimer associated with a single LMBD1 molecule on one side, while the second class showed an ABCD4 dimer bound to two LMBD1 molecules on both sides. The remaining two classes also exhibited 2:1 and 2:2 ABCD4:LMBD1 stoichiometries, respectively, but featured additional cryo-EM densities surrounding the NBD of ABCD4 (Supplementary Fig. 2a). Structural analysis suggests that the additional densities likely correspond to the NBDs of neighboring ABCD4 molecules, with the observed interactions potentially mediated by disulfide bonds between the C-terminal cysteine residues of the two ABCD4 protomers (Supplementary Fig. 3a). However, The TM region densities of LMBD1 in all classes were weaker than those of ABCD4, suggesting that LMBD1 may adopt multiple conformations upon binding to ABCD4. To enhance structural resolution, a focused mask was applied to the transmembrane regions of a single ABCD4 protomer along with LMBD1, followed by 3D classification without image alignment in RELION58. This approach yielded four similar conformations, among which the highest-resolution map enabled atomic model building and represents the lumen-open state. For the complex in the cytosol-open state, dark-subtracted images were first normalized by gain reference (0.738 Å per pixel). Drift correction and CTF were performed and estimated using CryoSPARC. Auto picking for all datasets was performed with crYOLO-v 1.9.359 using the general model with the particle threshold of 0.1. Particle extraction was performed in CryoSPARC v3.3.0. Subsequent two-dimensional classification, multi-class ab initio modeling, heterogeneous three-dimensional (3D) refinement, nonuniform refinement, CTF refinement and local refinement of the best class were performed in CryoSPARC v3.3.0. All the masks for the refinement were automatically generated by CryoSPARC during the nonuniform refinement. Model building The initial models were built based on 6JBJ for ABCD4 and the AlphaFold2 predicted structure for LMBD1, and then manually adjusted and refined using COOT60. Due to the limited resolution, residues 1–12, 357–383 and 605–606 of ABCD4, residues 1–13, 125–132, 163–184, 228–232, 334–348, 444–464, and 516–540 of LMBD1 in lumen-open and substrate-bound states, and residues 1–12, 357–383 and 605–606 of ABCD4, residues 1–11, 125–133, 163–184, 334–348, 442–463 and 514–540 of LMBD1 in cytosol-open were not built. The models were refined in real space using PHENIX61. For cross-validation, MolProbity62 was used to validate the final model. Local resolutions were estimated using CryoSPARC. Structure figures were generated using COOT60, PyMOL (http://www.pymol.org) and ChimeraX63. Chemical cross-linking assay Purified ABCD4–LMBD1 complex were incubated with the crosslinker BS3 at a final concentration of 1 mM for 5, 15, 30, 60 min, or at final concentrations of 0.1, 1, 3, or 5 mM for 30 min at room temperature. The reactions were quenched by the addition of 1 M Tris (pH 8.0). Samples were then mixed with 5× SDS loading buffer and subjected to SDS–PAGE. Gels were subsequently stained with Coomassie Brilliant Blue, and Western blot analysis was performed to assess the stoichiometry of the ABCD4–LMBD1 complex. ATPase activity assay The ATPase activity was measured by quantitating inorganic phosphate using the Malachite Green Phosphate Assay Kit (Sigma-Aldrich). 0.5 µM purified protein was incubated in 20 μL reaction buffer containing 150 mM NaCl, 20 mM Hepes (pH 7.5), 0.01% (w/v) LMNG/0.001% (w/v) CHS, 5 mM MgCl2, 1 mM cobalamin, and 5 mM ATP. After 60 min incubation at 37 °C, 80 μL reagent was mixed with 20 μL reaction product in 96-well plates and incubated for 15 min at room temperature. Absorbance at 620 nm was measured using a BioTek plate reader. and phosphate concentrations were calculated from a standard curve generated with known phosphate concentrations. Background phosphate levels, determined from protein-free control reactions, were subtracted from all measurements. For the cobalamin-enhanced ATPase assay, WT and E549Q mutants were assayed in the presence or absence of cobalamin under the same reaction conditions. For the relative ATPase activity analysis, the mean activity of the WT group was defined as 100%, and the activities of other groups were normalized to the WT mean to obtain relative activity values. All data were analyzed using GraphPad Prism 8. Each experiment was performed using three independent biological replicates, and for each biological replicate, measurements were made in three technical replicates. Data were presented as mean ± standard deviation (SD). Co-immunoprecipitation To examine the interaction between ABCD4 and LMBD1, pcDNA3.1-ABCD4-HA (or its mutants) and pcDNA3.1-LMBD1-GFP-Flag (or its mutants) were co-transfected into HEK-293T cells. All of the following procedures were carried out at 4 °C. Cells were harvested 36–48 h after transfection, the culture medium was discarded, and cells were washed with PBS. To lyse the cells, NETN lysis buffer (20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5% CA-630 (I3021, Sigma-Aldrich)) supplemented with 1 mM PMSF and 1% Protease Inhibitor Cocktail (New Cell & Molecular Biotech) was added and incubated for 1 hour. The cell lysis was then centrifuged for 15 min and 80 μl of the resulting supernatant was collected and mixed with 5× SDS loading buffer. The sample was incubated in a metal bath at 65 °C for 10 min and labeled as the input group. The protein loading amounts were kept consistent across all input groups using a BCA assay. The remaining supernatant was incubated overnight at 4 °C with pre-equilibrated anti-HA magnetic beads (MCE, HY-K0201) in buffer B (20 mM HEPES, pH 7.5, 150 mM NaCl, 0.02% DDM) under gentle rotation. After incubation, the supernatant was removed, and the beads were washed three times with buffer B. For elution, 30 μl of 1x SDS loading buffer was added to the beads, followed by thorough mixing and heating in a metal bath at 95 °C for 5 min. The resulting supernatant was separated using a magnetic rack and labeled as the bound group. Both input and bound samples were subsequently subjected to SDS–PAGE. After electrophoresis, the proteins were transferred to polyvinylidene fluoride filters (IPVH00010, Sigma-Aldrich), which were then incubated overnight at 4 °C with one of the following antibodies dissolved in 5% (w/v) BSA: rabbit anti-HA (3724S, CST, 1:1000), rabbit anti-Flag (14793S, CST, 1:1000), and GAPDH monoclonal antibody (60004-1-Ig, proteintech, 1:5000). Bound antibodies were visualized by chemiluminescence using High sensitive Plus ECL luminescence reagent (Sangon biotech) after incubation for 1 h at room temperature with HRP-conjugated Goat Anti-Rabbit IgG(H + L) (SA00001-2, proteintech,1:5000) or HRP-conjugated Goat Anti-Mouse IgG(H + L) (SA00001-1, proteintech,1:5000). The images were scanned using an MINICHEMI (SINSAGE) and analyzed using Image J. Immunofluorescence staining and confocal microscopy HEK-293T cells were cultured in high-glucose DMEM (E600003, Sangon) supplemented with 10% fetal bovine serum (AUS-01S-02, Cellbox) and 1% penicillin–streptomycin (P1400, Solarbio). Cells were seeded onto glass coverslips in 6-well plates at a density of 2 × 10⁵ cells per well. After 24 h, cells were co-transfected with 1 μg each of pcDNA3.1-ABCD4-HA or its mutants and pcDNA3.1-LMBD1-GFP-Flag using PEI. At 48 h post-transfection, cells were fixed with 4% paraformaldehyde (G1101, Servicebio) for 30 min at room temperature and washed three times with PBS. Permeabilization was performed using 0.05% PBST (BS084, Biosharp) for 1 min, followed by blocking in a mixture of goat serum (SL038, Solarbio) and 3% skim milk in PBS for 1 h at room temperature. Cells were incubated for 2 h at room temperature with the following primary antibodies diluted in blocking buffer: rabbit anti-HA (3724S, CST, 1:1000) and mouse anti-LAMP1 (ab278043, Abcam, 1:10). Cells were washed with PBS for 3 × 10 min and incubated for 1 h at room temperature in the dark with secondary antibodies: Alexa Fluor™ 405-conjugated goat anti-rabbit IgG (A-31553, Thermo Fisher, 1:200) and Alexa Fluor™ 594-conjugated goat anti-mouse IgG (A-11012, Thermo Fisher, 1:1000). Cells were then washed and mounted in antifade mounting medium (HY-K1042, MCE). Fluorescence images were acquired using a Leica laser-scanning confocal microscope and LAS X imaging software. To quantify colocalization between HA and LAMP1 signals, the ‘Coloc 2’ plugin in Fiji software was used with threshold regression set to Costes, PSF = 3.0, and 10 Costes randomizations. Pearson correlation coefficients (above threshold) were calculated within manually defined regions of interest (ROIs) and analyzed using GraphPad Prism 8. To avoid interference from overlapping or aggregated cells, ROIs were selected to exclude cell clusters, ensuring more representative quantification. Ten image fields were analyzed per experimental group. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Source data Source Data Acknowledgements We thank D. Li and X. Li from the Core Facility of Wuhan University and D. Stoddard from the UT Southwestern Medical Center Cryo-EM Facility for their assistance with cryo-EM data collection. We thank X. Li and P. Schmiege at the UT Southwestern Medical Center for editing the manuscript and helpful discussion. Author contributions T.L. designed the research. Q.L., X.L., M.Z., and T.L. performed the biochemical and structural experiments. Q.L., X.L., M.Z., K.Z., and Y.W. performed functional analysis. Q.L., X.L., M.Z., and T.L. analyzed the data and contributed to manuscript preparation. T.L. wrote the manuscript. T.L. and M.Z. supervised the project. Peer review Peer review information Nature Communications thanks Thomas Tomasiak and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available. Funding This work was supported by National Natural Science Foundation of China grants (32571430 to T.L.) and the Fundamental Research Funds for the Central Universities (2042025fg0008 to T.L.). Data availability The 3D cryo-EM density maps have been deposited in the Electron Microscopy Data Bank under the accession numbers EMD-69835, EMD-69836, and EMD-69837. Atomic coordinates for the atomic model have been deposited in the Protein Data Bank under the accession numbers 24VC, 24VD, and 24VE. All other data are available from the corresponding authors upon request. Source data are provided with this paper. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Qiwei Liu, Xingfan Li. Contributor Information Mi Zhou, Email: mizhou@whu.edu.cn. Tao Long, Email: taolong@whu.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-74552-5. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Source Data Data Availability Statement The 3D cryo-EM density maps have been deposited in the Electron Microscopy Data Bank under the accession numbers EMD-69835, EMD-69836, and EMD-69837. Atomic coordinates for the atomic model have been deposited in the Protein Data Bank under the accession numbers 24VC, 24VD, and 24VE. All other data are available from the corresponding authors upon request. Source data are provided with this paper.