Structural insights into toxicant export mediated by ABCC2 in Arabidopsis thaliana Authors: Xuan Qiu, Zhisen Yang, Yongxiang Gao, Linfeng Sun, Xin Liu Published: 2025 ABSTRACT Plants are highly vulnerable to damage from environmental pollutants, making detoxification mechanisms essential for sustaining growth and development. ABCC2 in Arabidopsis thaliana (AtABCC2) plays a critical role in detoxification by exporting diverse toxic compounds. Here, we report the structures of AtABCC2 in three distinct states: substrate-free, bound to the substrate S-(2,4-dinitrophenyl)glutathione (DNP-GS), and bound to ATP. Both monomeric and dimeric forms of AtABCC2 are observed. Unlike other dimeric ABCC homologs, AtABCC2 features a dimer interface mediated by its transmembrane domains. DNP-GS occupies an amphipathic cavity formed by the transmembrane domains. ATP binding drives the conformational changes in each protomer which bring the transmembrane and nucleotide-binding domains closer together, transitioning the transporter from a cytosolic-facing to an occluded state. Together, these findings advance our understanding of the molecular basis of substrate binding and transport by AtABCC2, and shed light on plant detoxification mechanisms. Introduction Plant cells are highly vulnerable to various natural or biosynthetic toxic substances, such as heavy metals and xenobiotics, and have evolved different types of detoxification mechanisms to ensure their survival and normal growth1,2. Isolating the toxicants in cellular compartments, particularly the vacuoles, can minimize their toxic effects in the cytosol and protect the cells from damage3,4. Typically, the toxicants are conjugated to molecules such as glutathione sulfate, amino acids, or chelators like phytochelatins or metallothioneins, after which they are transported into the vacuoles or expelled from the cytosol1,3,5,6. The ATP-binding cassette subfamily C (ABCC) transporters play a crucial role in the detoxification process by mediating the export of the toxicants or their conjugates7,8. In Arabidopsis thaliana, there are 15 members of the ABCC subfamily9. Arabidopsis thaliana ABCC1 (AtABCC1) and ABCC2 (AtABCC2) were identified as detoxification transporters located on the vacuole membrane10,11. They are widely expressed across various plant tissues, including roots, stems, leaves, flowers, and buds12. Notably, AtABCC2 exhibits a significantly higher transport capacity compared to AtABCC111. AtABCC2 is known for its broad substrate specificity and is capable of mediating the transport of diverse compounds, including heavy metals, herbicides, and pesticides13,14. It facilitates the transport of glutathione conjugates or glucuronic acid conjugates, including heavy metals like cadmium-glutathione conjugates, the glutathione conjugate S-(2,4-dinitrophenyl)glutathione (DNP-GS), and the glucuronic acid conjugate 17β-estradiol-(β-D-glucuronide) (E217βG)15. DNP-GS is derived from 1-chloro-2,4-dinitrobenzene (CDNB), a compound which is predominantly used in the production of pesticides and exhibits a high cytotoxicity to plant cells16,17. Once absorbed into the cytosol by the plant cell, CDNB is conjugated to glutathione by the glutathione S-transferase (GST), and then exported into the vacuole by AtABCC215,17. In addition to exogenous toxic substances, AtABCC2 also plays a role in transporting various endogenous metabolites, such as chlorophyll metabolites18, β-D-glucopyranosyl abscisate (ABA-GE), oxidized glutathione (GSSG), and anthocyanins19–21. The ABCC transporters, also known as multidrug resistance proteins (MRPs), are a crucial group of membrane transporters utilizing the energy of ATP hydrolysis to transport substrates, and are conserved across various species, including yeasts, plants, and mammals12,22. Topologically, the ABCC subfamily is classified into “long” and “short” transporters based on the presence or absence of an additional transmembrane domain (TMD0) located before the two core TMDs forming the substrate-binding cavity23,24. The TMD0 is connected to the core transmembrane domains through a long peptide, known as the lasso motif or L0 linker9. Previous studies have shown that the N-terminal extension of ABCC transporters, consisting of TMD0 plus L0 linker (or L0 linker alone), plays a crucial role in the proper localization of the protein, its interactions with other regulatory factors, and the regulation of its transport activity25–33. Structures of the ABCC subfamily transporters in Saccharomyces cerevisiae and mammals have been reported34–42. However, the protein sequences vary across species, particularly in the TMD0 domain (Supplementary Figs. 1 and 2a, b). The mechanisms by which plant ABCC transporters export toxicants remain unclear. In this study, we characterize the ATPase activity of AtABCC2 under different compounds, and determine its structures using cryo-EM single-particle analysis under three distinct states: the substrate-free (apo) state, the substrate DNP-GS-bound state, and the ATP-bound state. Together, the structural and biochemical analysis of AtABCC2 provides insights into the mechanisms of substrate binding and transport, as well as the plant detoxification process. Results ATPase activity characterization of AtABCC2 We overexpressed the full-length wild-type (WT) AtABCC2 (AtABCC2WT) and a catalytic mutant, in which the conserved catalytic residue E1404 in the Walker B motif of the nucleotide-binding domain 2 (NBD2) was substituted by glutamine (designated AtABCC2E1404Q or AtABCC2EQ), in human embryonic kidney (HEK) 293F cells. The proteins were purified with the detergent digitonin. Under this condition, only one peak corresponding to the AtABCC2WT or AtABCC2EQ protein was observed in size-exclusion chromatography (Supplementary Fig. 3a, b). The peak fractions were collected and subjected to further biochemical and structural analysis (Supplementary Fig. 3a, b). We first examined the ATPase activity of AtABCC2 in the presence of varying ATP concentrations. The ATPase activity of AtABCC2WT increased with rising ATP concentrations. The Km and Vmax values were determined to be 0.38 ± 0.04 mM and 2,547 ± 73 nmol∙min−1∙mg−1 protein, respectively, using the Michaelis–Menten equation (Fig. 1a). In contrast, the ATPase activity of the AtABCC2E1404Q mutant was significantly reduced compared to that of the AtABCC2WT (Fig. 1a, b). We then tested the ATPase activity of the AtABCC2WT in the presence of serial concentrations of different compounds which have been previously identified as substrates, including DNP-GS, E217βG, PC2, GSSG, GSH, and ABA-GE (Fig. 1c and Supplementary Fig. 3c). The results showed that the addition of DNP-GS, E217βG, or PC2 significantly stimulated the ATPase activity of AtABCC2 in a concentration-dependent manner (Fig. 1c), with a maximum increase of about 60%, 53%, or 12%, respectively, as indicated by the fitted Vmax values using the Hill equation (Fig. 1d, e). In contrast, GSSG, GSH, or ABA-GE induced only a modest increase in the ATPase activity, with a maximum increase of about 5% relative to the basal activity observed in the absence of the compounds (Fig. 1d, e). The fitted EC50 values for DNP-GS, E217βG, and PC2 were 86.79 ± 28.13 μM, 78.91 ± 40.09 μM, and 2.42 ± 1.46 μM, respectively (Fig. 1d). Additionally, the presence of DNP-GS or E217βG resulted in an increase in Vmax and a decrease in Km for AtABCC2. The fitted Km value in the presence of DNP-GS or E217βG was 0.15 ± 0.03 mM and 0.18 ± 0.04 mM, respectively (Fig. 1f). These findings were consistent with previous reports, suggesting that AtABCC2 efficiently transports DNP-GS or E217βG11,15.Fig. 1Characterizations of the ATPase activity of AtABCC2.a ATPase activities of the AtABCC2WT (WT) and AtABCC2E1404Q (E1404Q) in detergent micelle in the presence of different concentrations of ATP. Data points were nonlinearly fitted using the Michaelis–Menten equation. Data are means of three independent measurements (n = 3) ± s.d. b ATPase activities of AtABCC2WT and AtABCC2E1404Q at 3 mM ATP. Independent experiments were repeated three times (n = 3). Significances were determined using a two-tailed unpaired t-test. ****P < 0.0001. Data are means ± s.d. c ATPase activity of AtABCC2WT in serial concentrations of different compounds. Data points were nonlinearly fitted using the Michaelis–Menten equation. Data are means of three independent measurements (n = 3) ± s.d. d Enzymatic kinetics of the ATPase activity of AtABCC2WT in different compounds. e ATPase activity of AtABCC2WT in different compounds. The concentration was 40 μM for each compound. Independent experiments were repeated three times (n = 3). Significances were determined using one-way ANOVA with Dunnett’s multiple comparisons test. ****P < 0.0001. Data are means ± s.d. f ATPase activities of AtABCC2WT in the presence of different concentrations of ATP in E217βG or DNP-GS, respectively. The concentration was 60 μM for E217βG or DNP-GS. Data points were nonlinearly fitted using the Michaelis–Menten equation. Data are means of three independent measurements (n = 3) ± s.d. Source data are provided as a Source Data file. Characterizations of the ATPase activity of AtABCC2. a ATPase activities of the AtABCC2WT (WT) and AtABCC2E1404Q (E1404Q) in detergent micelle in the presence of different concentrations of ATP. Data points were nonlinearly fitted using the Michaelis–Menten equation. Data are means of three independent measurements (n = 3) ± s.d. b ATPase activities of AtABCC2WT and AtABCC2E1404Q at 3 mM ATP. Independent experiments were repeated three times (n = 3). Significances were determined using a two-tailed unpaired t-test. ****P < 0.0001. Data are means ± s.d. c ATPase activity of AtABCC2WT in serial concentrations of different compounds. Data points were nonlinearly fitted using the Michaelis–Menten equation. Data are means of three independent measurements (n = 3) ± s.d. d Enzymatic kinetics of the ATPase activity of AtABCC2WT in different compounds. e ATPase activity of AtABCC2WT in different compounds. The concentration was 40 μM for each compound. Independent experiments were repeated three times (n = 3). Significances were determined using one-way ANOVA with Dunnett’s multiple comparisons test. ****P < 0.0001. Data are means ± s.d. f ATPase activities of AtABCC2WT in the presence of different concentrations of ATP in E217βG or DNP-GS, respectively. The concentration was 60 μM for E217βG or DNP-GS. Data points were nonlinearly fitted using the Michaelis–Menten equation. Data are means of three independent measurements (n = 3) ± s.d. Source data are provided as a Source Data file. Overall structure of the wild-type AtABCC2 in the apo state With the purified protein in a detergent micelle, we determined the structure of AtABCC2WT in the apo state using the cryo-EM single-particle analysis method (Supplementary Figs. 4a–h and 5a). Notably, during particle 2D and 3D classifications of the data processing steps, protein particles in both the monomeric and the dimeric form were observed, and reconstructed separately (Supplementary Fig. 4b–h). The monomeric form was the predominant state, taking up about 47% of the selected particles, while the dimeric form accounted for about 26% (Supplementary Fig. 5a). Finally, we obtained the EM density maps of AtABCC2WT in both the monomeric and the dimeric forms, with overall resolutions of 3.8 Å and 4.2 Å, respectively (Fig. 2a and Supplementary Figs. 4c, f, and 5a). The local resolution of the monomeric form was better for the TMDs of AtABCC2 than for the NBDs, enabling side chain assignment for the TMDs (Supplementary Figs. 4e, h, and 5b, c). Notably, the local resolution of the TMDs on the vacuolar side was lower than the rest, suggesting intrinsic flexibilities of this region. The structure model was built combining both manual building and the predicted model from AlphaFold243. As is typical for ABC transporters, the core structure of AtABCC2 consists of two TMDs (TMD1 and TMD2) and two NBDs (NBD1 and NBD2) (Fig. 2b, c). TMD1 is formed by transmembrane segments (TMs) TM6-TM11, whereas TMD2 consists of TM12-TM17 (Fig. 2b–d). The two TMDs are arranged in a typical domain-swapped manner of the type IV ABC transporter44: TM6-TM8, TM11, TM15, and TM16 form one structural unit (TM bundle 1, TMB1), whereas TM9, TM10, TM12-TM14, and TM17 form the other (TM bundle 2, TMB2). The two TMBs exhibit a “V”-shaped architecture, creating a large cavity facing the intracellular side (Fig. 2d, e). It suggests that the captured WT AtABCC2 structure in the absence of substrates represents an inward-facing state, which is typical for the ABC transporters adopting similar structures44–46.Fig. 2Overall structure of AtABCC2WT in the apo state.a Overview of the cryo-EM density map for the monomeric AtABCC2WT in the apo state. TMD1 and NBD1 are colored in blue. TMD2 and NBD2 are colored in orange. TMD0 is colored in light blue, and the lasso motif is colored in pink, respectively. The density of the detergent micelle is shown in gray. b Cartoon representation of the AtABCC2WT structure in the apo state. c A topological diagram of AtABCC2WT. Unresolved structural regions are indicated by dashed lines. d A cartoon representation of the core TMDs of AtABCC2WT. TMD1 is colored in blue, and TMD2 is colored in orange. e Surface electrostatic potential of AtABCC2WT. Negative and positive charges are colored red and blue, respectively. The cavity is indicated with yellow dashed lines. f Structure alignment of AtABCC2WT, hABCC2 (PDB: 8JX7), and hABCC3 (PDB: 8HVH) in the apo state. The TMD0, the lasso motif, and the remaining domains are colored as indicated, respectively. g Interactions between TMD0 and the core TMDs in AtABCC2WT. The interacting residues are shown as sticks. Hydrogen bonds are indicated by red dashed lines. EM density maps for the residues are shown at a contour level of 3.5. Overall structure of AtABCC2WT in the apo state. a Overview of the cryo-EM density map for the monomeric AtABCC2WT in the apo state. TMD1 and NBD1 are colored in blue. TMD2 and NBD2 are colored in orange. TMD0 is colored in light blue, and the lasso motif is colored in pink, respectively. The density of the detergent micelle is shown in gray. b Cartoon representation of the AtABCC2WT structure in the apo state. c A topological diagram of AtABCC2WT. Unresolved structural regions are indicated by dashed lines. d A cartoon representation of the core TMDs of AtABCC2WT. TMD1 is colored in blue, and TMD2 is colored in orange. e Surface electrostatic potential of AtABCC2WT. Negative and positive charges are colored red and blue, respectively. The cavity is indicated with yellow dashed lines. f Structure alignment of AtABCC2WT, hABCC2 (PDB: 8JX7), and hABCC3 (PDB: 8HVH) in the apo state. The TMD0, the lasso motif, and the remaining domains are colored as indicated, respectively. g Interactions between TMD0 and the core TMDs in AtABCC2WT. The interacting residues are shown as sticks. Hydrogen bonds are indicated by red dashed lines. EM density maps for the residues are shown at a contour level of 3.5. Like the “long” ABCC homologs24, the structure of AtABCC2 consisted of an additional TMD0 and a linker (L0) preceding the first transmembrane domain, TMD1. TMD0 has five transmembrane helices (TM1 to TM5) (Fig. 2b, c). The L0 linker contains three helical structures, with the last helix oriented almost parallel to the membrane surface and positioned on the exterior region of TMB1, close to TM6, TM7, TM11 and TM16 (Fig. 2b). Structural comparison revealed that, although the L0 linker of AtABCC2 aligned well with the ABCC homologs represented by the human ABCC2 (hABCC2)42,47 and ABCC3 (hABCC3)41, TMD0 exhibited a significant lateral shift (Fig. 2f). In hABCC2 and hABCC3, TMD0 is positioned laterally relative to TMB1, whereas in AtABCC2, TMD0 is located between the two TMBs (Fig. 2f). Human ABCB6 (hABCB6), which is a half-transporter belonging to the ABCB subfamily, also contains a TMD0 domain48. The TMD0 domains in hABCB6 and AtABCC2 show different localizations relative to the core TMBs as well (Supplementary Fig. 6a). This suggests that the extra N-terminal TMD0 domain displays structural variations among the ABC transporters. Based on sequence alignment results, the TMD0 exhibits low sequence conservation, thus possibly leading to the varied orientations observed in the structures (Supplementary Fig. 1a). In AtABCC2, TM1 and TM4 of TMD0 interact with TM15 and TM16 of TMB1, primarily through hydrophobic packing and hydrogen bonding (Fig. 2g). To assess the functional roles of TMD0 and L0 linker, we generated two truncated versions of AtABCC2: one with the TMD0 truncated (ΔTMD0) and the other with both the TMD0 and L0 truncated (ΔTMD0 + L0), and measured their ATPase activities. The truncated variants were well-folded as indicated by the results of the size-exclusion chromatography (Supplementary Fig. 6b). The ΔTMD0 mutant exhibited similar ATPase activity compared to the wild-type protein, with a fitted Vmax value of 2824 ± 73 nmol∙min−1∙mg−1 protein and a Km value of 0.74 ± 0.06 mM. In contrast, truncation of both the TMD0 and L0 linker resulted in a dramatic reduction in ATPase activity compared to the WT protein, with a fitted Vmax value of 2737 ± 354 nmol∙min−1∙mg−1 protein and a Km value of 2.25 ± 0.53 mM (Supplementary Fig. 6c, d). These results are consistent with previous studies, suggesting that while TMD0 has minimal impact on the transport and ATPase activity of ABCC transporters, the L0 linker is essential to substrate transport, likely by fine-tuning the conformational changes required for substrate translocation26,31,41. The dimeric form of AtABCC2WT exhibits a two-fold C2 symmetry (Supplementary Fig. 6e). The monomeric structure model can be docked well into the EM map in the dimeric state, supporting the same conformation for the monomeric protein (Supplementary Fig. 6f). The dimer interface of AtABCC2WT is formed by both the TMB regions and the NBDs of the two protomers (Supplementary Fig. 6f). In the TMB region, the interface is mediated by hydrophobic interactions between TM12 and TM13 from each protomer (Supplementary Fig. 6f). In the NBD region, NBD2 from one protomer packs to NBD2 from the other protomer (Supplementary Fig. 6f). Due to the lower local resolutions of the NBDs in the EM map, we were unable to analyze the interactions in this region in more detail. To confirm the observed monomeric and dimeric forms of AtABCC2WT observed in the cryo-EM sample, we carried out the mass photometry (MP) analysis of the purified protein in the detergent of digitonin. Consistent with the EM analysis, the MP data showed that AtABCC2WT exists in two major forms, the predominant one (approximately 67%) corresponding to the monomeric form with a molecular weight of 206 kDa which is close to the theoretical value of 182 kDa, and the other one (approximately 33%) corresponding to the dimeric form with a molecular weight of 407 kDa close to the theoretical value of 364 kDa (Supplementary Fig. 7a). Meanwhile, the ΔTMD0 mutant showed similar oligomeric patterns to the WT protein (Supplementary Fig. 7a), consistent with the structural observation that TMD0 is not involved in the dimerization of AtABCC2. To separate the two oligomeric species, we tried to purify the protein under other detergent conditions. In the detergent of lauryl maltose neopentyl glycol (LMNG) plus cholesteryl hemisuccinate (CHS), the AtABCC2WT protein displayed two peaks in size-exclusion chromatography (Supplementary Fig. 7b, c). Fractions of each peak were collected and subjected to MP analysis. The first peak predominantly (approximately 83%) had a molecular weight of 425 kDa corresponding to the dimeric form (Supplementary Fig. 7d), while the second peak predominantly (approximately 80%) had a molecular weight of 248 kDa corresponding to the monomeric form (Supplementary Fig. 7e). We characterized the ATPase activity of the two species. The results showed that the ATPase activity of the monomeric and dimeric forms had no significant differences either in the absence or presence of the substrate DNP-GS (Supplementary Fig. 7f). Structure of the AtABCC2EQ mutant in the apo state We then determined the cryo-EM structure of AtABCC2EQ mutant in the apo state (Supplementary Fig. 8a–f). During data processing, we also observed particles in both the monomeric and dimeric forms (Supplementary Fig. 8f). However, in contrast to the AtABCC2WT, the dimeric form was predominant (Supplementary Fig. 8f). The mass photometry results are consistent, showing a higher ratio of the dimer than the monomer in the AtABCC2EQ protein solution (Supplementary Fig. 7a). Due to the limited number of particles in the monomeric form, we were unable to reconstruct a high-resolution EM map (Supplementary Fig. 8f). The EM map of the dimeric form was reconstructed with a C2 symmetry. Finally, we obtained a map of the dimeric AtABCC2EQ at an overall resolution of 3.5 Å (Fig. 3a and Supplementary Fig. 8c–e). Compared to the EM map of the dimeric AtABCC2WT, the local resolutions of the transmembrane region and the NBDs involved in the dimer interface were higher, allowing accurate residue assignment (Supplementary Fig. 8e, g). Like the AtABCC2WT, the dimer interface of AtABCC2EQ was formed by TMB2 and NBD2 from both protomers (Fig. 3b, c). TM12 and TM13 from the two protomers mediated the dimer formation mainly through hydrophobic packing (Fig. 3c and Supplementary Fig. 9a). Compared with the dimeric AtABCC2WT, TM12 and TM13 of each protomer in the dimeric AtABCC2EQ rotated about 10° (Supplementary Fig. 6g). Consequently, the two NBD2 domains of each protomer were closer than in the AtABCC2WT structure, leading to a different NBD interface (Supplementary Fig. 6g). In the dimer interface of AtABCC2EQ, two clear EM densities corresponding to lipid molecules were identified, which can be docked well with phosphatidylcholines, each localized between TM14 from one protomer and TM13 from the other (Supplementary Fig. 10a). The hydrocarbon tails of the lipid molecule pack to the hydrophobic residues in TM13 and TM14 (Supplementary Fig. 10a). Thus, these lipid molecules contribute to the dimer formation of AtABCC2 by mediating additional contacts between the two protomers.Fig. 3Structure of AtABCC2EQ mutant in the apo state.a Overview of the cryo-EM density map for the dimeric AtABCC2EQ in the apo state. TMD1 and NBD1 of protomer-1 are colored in blue. TMD2 and NBD2 of protomer-1 are colored in orange. TMD0 and the lasso motif of protomer-1 are colored in light blue or pink, respectively. Protomer-2 is colored in dark gray. The density of the micelle is shown in gray. b Cartoon representation of the AtABCC2EQ structure in the apo state. Protomer-1 is colored in a rainbow. Protomer-2 is colored in dark gray. c Dimer interface of the AtABCC2EQ structure in the apo state. The interfaces in the TMD and NBD domains are shown by two zoom-in images, respectively. Hydrogen bonds are indicated by red dotted lines. EM density maps for the residues are shown at a contour level of 4.0. d Structure alignments of AtABCC2WT (colored in slate) and AtABCC2EQ (colored in aquamarine) in the apo state. Magnified views of the comparisons of NBDs and TMBs are shown. Shift of NBD2 is indicated by the red arrow. Structure of AtABCC2EQ mutant in the apo state. a Overview of the cryo-EM density map for the dimeric AtABCC2EQ in the apo state. TMD1 and NBD1 of protomer-1 are colored in blue. TMD2 and NBD2 of protomer-1 are colored in orange. TMD0 and the lasso motif of protomer-1 are colored in light blue or pink, respectively. Protomer-2 is colored in dark gray. The density of the micelle is shown in gray. b Cartoon representation of the AtABCC2EQ structure in the apo state. Protomer-1 is colored in a rainbow. Protomer-2 is colored in dark gray. c Dimer interface of the AtABCC2EQ structure in the apo state. The interfaces in the TMD and NBD domains are shown by two zoom-in images, respectively. Hydrogen bonds are indicated by red dotted lines. EM density maps for the residues are shown at a contour level of 4.0. d Structure alignments of AtABCC2WT (colored in slate) and AtABCC2EQ (colored in aquamarine) in the apo state. Magnified views of the comparisons of NBDs and TMBs are shown. Shift of NBD2 is indicated by the red arrow. Structure alignment of the monomeric AtABCC2WT structure and the protomer in the dimeric AtABCC2EQ structure revealed a root-mean-square deviation (RMSD) of 2.6 Å (1026 Cα atoms aligned) (Fig. 3d). While TMD0, L0 linker, TMB1, and NBD1 of the two protomers align well, TMB2 and NBD2 exhibit a rotational shift (Fig. 3d). Consequently, the two TMBs and NBDs are closer in the AtABCC2EQ structure compared to the AtABCC2WT structure. While the vacuolar termini of the transmembrane segments in TMB2 remain unchanged, the intracellular parts of TMB2, including TM9, TM10, TM12-TM14, and TM17, shift toward TMB1, resulting in a more compact, intracellularly-facing cavity (Fig. 3d). This configuration of AtABCC2EQ may represent an intermediate state in the substrate transport cycle. Recently, dimeric structures have also been reported for the yeast glutathione-conjugated heavy metal transporter Ycf1 and the human ABCC1 (hABCC1)39,40. Like AtABCC2, lipid or detergent molecules were identified to be involved in the dimer formation of Ycf1 and hABCC139,40. However, the dimer interface of Ycf1 or hABCC1 is formed mainly by the TMD0 of one protomer and the core TMDs of the other, distinct from that of AtABCC2 (Supplementary Fig. 10b). The buried surface area (BSA) analysis revealed that Ycf1 has the largest BSA value (approximately 1599 Å2) which is consistent with its more compact dimer organization observed in the structure, while AtABCC2 and hABCC1 have comparable BSA values (1058 Å2 and 1002 Å2, respectively) (Supplementary Fig. 10c–e). DNP-GS binding site of AtABCC2EQ The ATPase activity analysis supports that DNP-GS conjugate is a transport substrate of AtABCC2. To investigate the toxicant recognition mechanism, we tried to get a DNP-GS-bound structure by adding DNP-GS to the purified AtABCC2EQ protein and subjecting the complex to cryo-EM analysis (Supplementary Fig. 11a–f). Like the apo AtABCC2EQ sample, particles in both the monomeric and dimeric forms were present (Supplementary Fig. 11e). We obtained EM maps of the dimer and monomer at an overall resolution of 3.5 Å and 4.1 Å, respectively (Supplementary Fig. 11b–e). The two maps can be superimposed well, suggesting almost identical structures (inset panel of Supplementary Fig. 11e). The EM density corresponding to one DNP-GS molecule was clearly identified within the transmembrane cavity (Fig. 4a, b). Electrostatic potential and hydrophobic surface analysis revealed that the DNP-GS binding interface was amphipathic, with one region highly positive in electrostatic potential and another predominantly hydrophobic (Fig. 4c, d). Residues involved in DNP-GS binding were identified (Fig. 4e). Specifically, R1198 forms ionic interactions with the carboxyl group of the glutathione part of DNP-GS and the nitro group of the dinitrophenyl part of DNP-GS (Fig. 4e). The carboxyl group of the glutathione part is also close to the side chains of Q415, N1140, and R1141 (Fig. 4e). Additionally, the phenyl group of the dinitrophenyl part packs closely to the side chains of F1034, M1037, and L1199. F361 is also positioned closely to the glutathione part of DNP-GS (Fig. 4e). To confirm the roles of these residues in DNP-GS binding, we mutated each residue to alanine and tested the ATPase activity of the mutants in the presence of DNP-GS. The results showed that all mutants exhibited decreased ATPase activities compared to the wild-type AtABCC2, with the R1198A mutant displaying the most significant reduction (Fig. 4f). This suggests that both ionic and hydrophobic interactions are crucial for the binding of DNP-GS to AtABCC2, and both the glutathione and dinitrophenyl parts of DNP-GS are involved in the substrate recognition.Fig. 4Structure of AtABCC2EQ in complex with DNP-GS.a Overview of the cryo-EM density map for the AtABCC2EQ in the DNP-GS-bound state. b Cartoon representation of the AtABCC2EQ structure in the DNP-GS-bound state. The EM density of DNP-GS is shown on the right. c Binding of DNP-GS to an amphipathic pocket in AtABCC2EQ. Section views of the surface electrostatic potential are presented, with DNP-GS shown in sticks. d Section view of the hydrophobic surface of the substrate-binding pocket. DNP-GS is shown in sticks. Hydrophilic and hydrophobic regions are colored cyan and yellow, respectively. e Coordination of DNP-GS by AtABCC2EQ. Side chains of residues close to DNP-GS are shown in sticks. TMD1 is colored in blue, and TMD2 is colored in orange. Hydrogen bonds are indicated by red dotted lines. f ATPase activities of AtABCC2EQ and mutants in the presence of 3 mM ATP upon addition of DNP-GS. Data are means of three independent measurements (n = 3) ± s.d. ****P < 0.0001 (one-way ANOVA with Dunnett’s multiple comparisons test). Source data are provided as a Source Data file. g Structure alignments of apo AtABCC2EQ (colored in aquamarine) and DNP-GS bound AtABCC2EQ (colored in pink). h Structure comparisons of TMBs of apo AtABCC2EQ and DNP-GS-bound AtABCC2EQ. The conformational change of TMB1 is indicated by the red arrow. i Structure comparisons of NBDs of apo AtABCC2EQ and DNP-GS bound AtABCC2EQ. The conformational change of NBD1 is indicated by the red arrow. Structure of AtABCC2EQ in complex with DNP-GS. a Overview of the cryo-EM density map for the AtABCC2EQ in the DNP-GS-bound state. b Cartoon representation of the AtABCC2EQ structure in the DNP-GS-bound state. The EM density of DNP-GS is shown on the right. c Binding of DNP-GS to an amphipathic pocket in AtABCC2EQ. Section views of the surface electrostatic potential are presented, with DNP-GS shown in sticks. d Section view of the hydrophobic surface of the substrate-binding pocket. DNP-GS is shown in sticks. Hydrophilic and hydrophobic regions are colored cyan and yellow, respectively. e Coordination of DNP-GS by AtABCC2EQ. Side chains of residues close to DNP-GS are shown in sticks. TMD1 is colored in blue, and TMD2 is colored in orange. Hydrogen bonds are indicated by red dotted lines. f ATPase activities of AtABCC2EQ and mutants in the presence of 3 mM ATP upon addition of DNP-GS. Data are means of three independent measurements (n = 3) ± s.d. ****P < 0.0001 (one-way ANOVA with Dunnett’s multiple comparisons test). Source data are provided as a Source Data file. g Structure alignments of apo AtABCC2EQ (colored in aquamarine) and DNP-GS bound AtABCC2EQ (colored in pink). h Structure comparisons of TMBs of apo AtABCC2EQ and DNP-GS-bound AtABCC2EQ. The conformational change of TMB1 is indicated by the red arrow. i Structure comparisons of NBDs of apo AtABCC2EQ and DNP-GS bound AtABCC2EQ. The conformational change of NBD1 is indicated by the red arrow. The DNP-GS-bound AtABCC2EQ structure adopts a similar inward-facing conformation to that of the apo-state AtABCC2EQ structure (Fig. 4g). Superimposing the two dimeric structures revealed an RMSD of 1.2 Å (1427 Cα atoms aligned). The dimer interface, formed by TMB2 and NBD2, remains unchanged in the two structures (Fig. 4g). Upon DNP-GS binding, the cytosolic parts of the TMs in TMB1 undergo a slight shift toward TMB2, resulting in a partial closure of the substrate-binding cavity. Consequently, NBD1 moves closer to NBD2 (Fig. 4h, i). Substrate-bound structures have been determined for several ABCC homologs, including the E217βG-bound hABCC3 structure, the LTC4-bound bABCC1 structure, and the different ligand-bound human MRP2 and MRP4 structures34,37,41,47. Structural alignments of the DNP-GS-bound state of AtABCC2EQ, the LTC4-bound bABCC1, and the E217βG-bound hABCC3 structures reveal an RMSD of 2.4 Å (over 780 Cα atoms) and 3.4 Å (over 790 Cα atoms), respectively (Supplementary Fig. 12a). While hABCC3 binds two E217βG molecules asymmetrically in its tapered cavity, both bABCC1 and AtABCC2 bind only one substrate molecule. In all cases, the substrate molecules bind to a highly conserved site within the corresponding transporters (Supplementary Fig. 12a, b). Notably, E217βG, LTC4, and DNP-GS are all amphiphilic molecules and fit well into the amphipathic cavity (Supplementary Fig. 12c). Comparison of the bABCC1 and AtABCC2 structures revealed highly conserved residues in this binding site, including the positively charged arginine residues and the aromatic residues (Supplementary Figs. 12b and 9c, d). The relatively large binding pocket could accommodate substrates of various sizes, which may contribute to the broad substrate specificity of the ABCC family transporters. Conformational changes of AtABCC2EQ upon ATP binding The binding of ATP to ABC transporters has been shown to facilitate the transport and subsequent release of substrates49. We added ATP to the purified AtABCC2EQ protein sample, and then performed cryo-EM analysis (Supplementary Fig. 13a–e). Similar to the apo-state and DNP-GS-bound samples, both monomeric and dimeric forms were present in the ATP-bound AtABCC2EQ particles, with the dimeric form being the predominant one (Supplementary Fig. 13e). Due to limited particle numbers, the EM map of the monomeric form was only determined at a low resolution (Supplementary Fig. 13e), precluding further analysis. Finally, we obtained an EM map of the ATP-bound dimeric structure at an overall resolution of 3.4 Å (Fig. 5a and Supplementary Fig. 13b–d). The local resolutions of both the TMDs and NBDs were significantly improved (Supplementary Fig. 13d, f), especially for the NBDs, suggesting that ATP binding stabilizes the NBDs. The density corresponding to the ATP molecule was clearly observed between the two NBDs (Supplementary Fig. 13f). Compared to the apo-state AtABCC2EQ structure, the dimer interface, as well as the TMD0 and L0 linker, remains unchanged (Fig. 5b, c). However, large conformational changes are observed between the two halves of the transporter. NBD1 and TMB1 move towards their counterparts (Fig. 5b–d). Structural analysis revealed that ATP binds to the cleft between the Walker A motif of one NBD and the ABC signature motif of the opposite NBD, triggering dimerization of two NBDs (Fig. 5e and Supplementary Fig. 9b, e). Both the observed conformational changes and the ATP-binding site are highly conserved in ABC transporters44,45. Hydrogen bonds and salt bridges are formed between ATP and the backbones or side chains of several key residues, including V738 and S741 of NBD1, as well as Y1251, G1281, K1282, S1283, S1284, and Q1324 of NBD2 (Fig. 5e). The conformational changes of the NBDs are transmitted to the TMBs via the linker helices located between the TMBs and NBDs (Fig. 5b–d). Consequently, the two TMBs pack closely on the intracellular side, sealing the substrate-binding cavity (Fig. 5f). In this state, the binding pocket is occluded on both sides of the membrane (Fig. 5f), representing an intermediate conformation of AtABCC2 following substrate release. Structural alignment of the monomeric apo-state and ATP-bound AtABCC2EQ structures revealed that while TMB2 aligns well in the two structures, the intracellular parts of TMB1 move toward TMB2. Large shifts are observed in TM6-TM8, TM11, TM15, and TM16 (Fig. 5g). Notably, TM11 undergoes a large conformational change, breaking from a long helix in the apo state into two helical segments. Two proline residues, P569 and P574, are located in the discontinuous region, possibly contributing to the helix breaking (Fig. 5h). On the cytosolic side of the TMBs, intense hydrogen bonds form between the two TMBs upon closure of the substrate-binding pocket, including interactions between S409 of TM8 and K1022 of TM14, S1093 of TM15 and N1018 of TM14, R589 of TM11, N401 of TM8 and D475 of TM9 (Supplementary Fig. 14a). On the vacuolar side, bulky hydrophobic residues seal the binding pocket, including L323 and L327 of TM6, F557 of TM11, and L1187 of TM17 (Supplementary Fig. 14b). Taken together, these two gates regulate the substrate entry from the cytosolic side and exit into the vacuolar side.Fig. 5Conformational changes of AtABCC2EQ upon ATP binding.a Overview of the cryo-EM density map for the AtABCC2EQ in the ATP-bound state. TMD1 and NBD1 of protomer-1 are colored in blue. TMD2 and NBD2 of protomer-1 are colored in orange. TMD0 and the lasso motif of protomer-1 are colored in purple or pink, respectively. Protomer-2 is colored in cyan. b Structure alignments of AtABCC2EQ in the apo state (colored aquamarine) and ATP-bound state (colored deep salmon). c Structure alignments of the monomeric structures of AtABCC2EQ in the apo state and the ATP-bound state. TMD0 is indicated by black dashed lines. d Zoom-in view of structure comparisons of the NBD domains of AtABCC2EQ in the apo state and the ATP-bound state. e Coordination of ATP and Mg2+ by AtABCC2EQ. ATP molecule and interacting residues are shown as sticks in the zoom-in views. Hydrogen bonds and salt bridges are shown as red dashed lines. f Surface electrostatic potential of the ATP-bound AtABCC2EQ. Negative and positive charges are colored red and blue, respectively. The substrate-binding cavity is indicated by red dashed lines. g Comparison of the TMBs of the apo and the ATP-bound AtABCC2EQ structures. h Comparison of the TMB1 during structure transitions between the apo state and the ATP-bound state of AtABCC2EQ. The zoom-in view of TM11 during structure transitions is indicated by black dashed lines. The conformational change of TM11 is indicated by the blue arrow. Conformational changes of AtABCC2EQ upon ATP binding. a Overview of the cryo-EM density map for the AtABCC2EQ in the ATP-bound state. TMD1 and NBD1 of protomer-1 are colored in blue. TMD2 and NBD2 of protomer-1 are colored in orange. TMD0 and the lasso motif of protomer-1 are colored in purple or pink, respectively. Protomer-2 is colored in cyan. b Structure alignments of AtABCC2EQ in the apo state (colored aquamarine) and ATP-bound state (colored deep salmon). c Structure alignments of the monomeric structures of AtABCC2EQ in the apo state and the ATP-bound state. TMD0 is indicated by black dashed lines. d Zoom-in view of structure comparisons of the NBD domains of AtABCC2EQ in the apo state and the ATP-bound state. e Coordination of ATP and Mg2+ by AtABCC2EQ. ATP molecule and interacting residues are shown as sticks in the zoom-in views. Hydrogen bonds and salt bridges are shown as red dashed lines. f Surface electrostatic potential of the ATP-bound AtABCC2EQ. Negative and positive charges are colored red and blue, respectively. The substrate-binding cavity is indicated by red dashed lines. g Comparison of the TMBs of the apo and the ATP-bound AtABCC2EQ structures. h Comparison of the TMB1 during structure transitions between the apo state and the ATP-bound state of AtABCC2EQ. The zoom-in view of TM11 during structure transitions is indicated by black dashed lines. The conformational change of TM11 is indicated by the blue arrow. Discussion In this study, we showed that AtABCC2, a key player in plant detoxification, specifically recognizes toxic compounds such as the glutathione conjugate DNP-GS. The structures of AtABCC2 in three different states, including the apo state, substrate-bound state, and ATP-bound state, provide insights into the conserved transport cycle of ABC transporters49, particularly within the ABCC subfamily. We propose a transport cycle of toxicant by AtABCC2, which follows the basic mechanisms proposed for the ABCC transporters38,42,50. In the apo-form, AtABCC2 structure adopts an inward-open conformation with the two transmembrane domains forming a large conical cavity, which is amphipathic in electrostatic potential. The two NBD domains are separated in this state. Upon substrate binding, the amphiphilic toxic DNP-GS fits into the pocket and induces partial closure of the TM bundles. The binding of ATP molecules triggered further conformational changes starting from the NBDs. The two separate NBD domains move closer together, dimerizing and forming critical interactions. The changes in NBDs are then transmitted to the TMDs, leading to the closure of the substrate-binding cavity on the cytosolic side and opening on the vacuolar side for the substrate release. The outward-facing conformation of AtABCC2 following substrate release, however, remains to be captured. Once substrate release is complete, the transmembrane domain returns to an occluded state, as observed in the ATP-bound state of this study. Finally, following ATP hydrolysis, the two NBD domains undergo depolymerization and separate, resetting the structure to its resting state, preparing for the next transport cycle. Though highly conserved in the transport mechanism, the structures of AtABCC2 reveal features in the relative position of the N-terminal TMD0 domain to the TMDs, and the dimer formation compared to other ABCC homologs. While the TMD0 domain may not be directly involved in the regulation of the transport process, it may regulate the correct membrane localization or the interaction with other factors in cells25. In human ABCC2 (or MRP2), the TMD0 region has been reported to be required for the correct apical membrane localization of the transporter27. In human ABCC8 (Sulfonylurea Receptor 1, SUR1) or ABCC9 (SUR2), the TMD0 domain mediates the interaction with the inward rectifier potassium channel subunit (Kir6.1 or Kir6.2), and together form the ATP-sensitive potassium channel (KATP)51–54. The physiological role of TMD0 in the function of AtABCC2 remains to be characterized. Additionally, the relationship between the oligomeric state of AtABCC2 and its physiological regulation requires further investigation. The dimeric and monomeric forms of AtABCC2 showed little difference in the ATPase activity in the absence or presence of substrate (Supplementary Fig. 7f). For yeast Ycf1, the dimeric form is stabilized by proteolytic cleavage and displays lower ATPase activity compared to the monomeric form39. For hABCC1, while the dimer has lower ATPase activity than the monomer in the absence of the substrate cGAMP, the dimeric form has more enhanced ATPase activity by cGAMP and is suggested to be the active form of substrate transport46. Our determined structure of AtABCC2 reveals a distinct dimerization mechanism of ABCC transporters. Whether the different oligomeric states of AtABCC2 are involved in the regulation of its transport activity remains to be further investigated. Notably, lipid or detergent molecules are observed in the dimer interface of both AtABCC2, Ycf139, and hABCC140, thus indicating a potential mechanism for regulating the oligomeric states of ABCC transporters by lipid molecules or membrane compositions. AtABCC2 is capable of cellular detoxification by transporting a variety of conjugated substrates. The substrate-binding features within the structure of AtABCC2 may vary due to the different conjugated chemical groups. Our DNP-GS-bound structure of AtABCC2 provides insights into the substrate recognition of the glutathione conjugates. How AtABCC2 recognizes different modified substrates remains an intriguing question for future research. In conclusion, this study provides a detailed molecular understanding of toxicant export mediated by AtABCC2 and advances our knowledge of the vacuolar detoxification in plants, which may inform future agricultural and environmental applications, particularly in enhancing plant resistance to toxic compounds. Methods Protein expression and purification of AtABCC2 The DNA sequence of full-length WT AtABCC2, E1404Q mutant, and two truncated mutants ΔTMD0 (with amino acid residues 1–202 deleted) and ΔTMD0 + L0 (with amino acid residues 1–265 deleted) were subcloned into a pCAG vector (Invitrogen) with an N-terminal Strep tag and Flag tag. All AtABCC2 mutants were generated in the same way. A list of the primers used is provided in Supplementary Data 1. The HEK293F cells (Thermo Fisher Scientific, catalog number: A14635) were transfected with plasmids at a density of 2.5 × 106 cells mL−1. For 1 liter cell culture, 1.5 mg plasmids were premixed with 4 mg linear polyethylenimines (PEIs) (Polysciences) in 45 mL fresh medium for 15 min. Then the mixture was added into the cell, followed by a 15 min incubation. After 12 h, 10 mM sodium butyrate was added to the cells cultured at 30 °C. 48 h after transfection, the cells were collected by centrifugation at 2,100 × g for 10 min. For protein purified in detergent digitonin and subjected to further biochemical and EM analysis, the cell pellet was resuspended in buffer containing 25 mM HEPES-HCl, pH 7.4, and 150 mM NaCl, supplemented with 1.5% (w/v) n-dodecyl-β-D-maltoside (DDM; Anatrace), 1 mM phenylmethylsulfonyl fluoride (PMSF), aprotinin (1.3 mg mL−1), and pepstatin (0.7 mg mL−1). The cell pellet was then incubated for 2 h at 4 °C by gentle agitation. After detergent extraction, the supernatant was collected by centrifugation at 200,000 × g for 1 h and incubated with the anti-FLAG M2 affinity gel (Sigma-Aldrich) at 4 °C for 30 min. The resin was washed with 40 mL buffer containing 25 mM HEPES-HCl, pH 7.4, 150 mM NaCl, and 0.06% (w/v) digitonin (Apollo Scientific). Protein was eluted with the wash buffer plus 200 μg mL−1 FLAG peptides, which was then concentrated with a 100 kDa cut-off Centricon (Millipore). Finally, the sample was subjected to size-exclusion chromatography using the Superose-6 Increase column (GE Healthcare) in the buffer containing 25 mM HEPES-HCl, pH 7.4, 150 mM NaCl, and 0.03% (w/v) digitonin (Apollo Scientific). Peak fractions were collected and concentrated using a 100 kDa cut-off Centricon (Millipore) to ~16 mg mL−1 for further analysis. For protein purified in detergent lauryl maltose neopentyl glycol (LMNG) plus cholesteryl hemisuccinate (CHS), the cell pellet resuspended buffer containing 25 mM HEPES-HCl, pH 7.4, and 150 mM NaCl, supplemented with 1% (w/v) LMNG plus 0.2% CHS (Anatrace), 1 mM phenylmethylsulfonyl fluoride (PMSF), aprotinin (1.3 mg mL−1), and pepstatin (0.7 mg mL−1). The wash buffer contains 25 mM HEPES-HCl, pH 7.4, and 150 mM NaCl, supplemented with 0.01% (w/v) LMNG and 0.002% CHS (Anatrace). The buffer of gel filtration contains 25 mM HEPES-HCl, pH 7.4, and 150 mM NaCl, supplemented with 0.005% (w/v) LMNG and 0.001% CHS (Anatrace). The other purification steps are the same as described above. ATP hydrolysis measurements The ATPase activity was measured using the ATPase Colorimetric Assay Kit (Innova Biosciences). 1 µg purified protein was added to the reaction buffer containing 25 mM HEPES-KOH (pH 7.4), 150 mM KCl, 10 mM MgCl2, and 0.01% (w/v) DDM. Each substrate was diluted to serial concentrations, added to the mixture, and incubated for 20 min on ice. Next, ATP was added to the solution at a final concentration of 3 mM to start the reaction at 37 °C for 10 min, and the amount of inorganic phosphate released was measured quantitatively with the ATPase Colorimetric Assay Kit by determining the light absorbance at 650 nm wavelength with the BioTek Synergy H1 Multimode Reader. S-(2,4-dinitrophenyl)glutathione (DNP-GS), 17β-estradiol-β-D-glucuronide (E217βG), used for the measurement of the concentration-dependent ATPase activity, were dissolved in DMSO and then serially diluted to the desired final concentrations in 1% (v/v) DMSO. β-D-glucopyranosyl abscisate (ABA-GE), GSH, GSSG, phytochelatin 2 (PC2) were dissolved in aqueous solution, and the final concentration is 60 μM. Statistical analysis was done using GraphPad Prism 10. Kinetics of the ATPase activities of AtABCC2 under serial concentrations of ligands were fitted with the Hill equation, with the Hill coefficient value fixed to 1. Mass photometry Mass photometry measurements were carried out on the TwoMP instrument (Refeyn Inc.). The glass cover slip (1.5H thickness, 24 × 50 mm, VWR) was cleaned consecutively with isopropanol and water, then dried using a stream of clean nitrogen. Resuspend the molecular weight standard in its matching buffer to calibrate the particle-image contrast caused by scattered light and convert it to molecular weight units. AtABCC2WT, ΔTMD0, and AtABCC2EQ were diluted with the buffer used in size-exclusion chromatography to a final concentration of 1 μM. 1 μL of 1 μM protein solution was mixed with 15 μL buffer in the silicon gasket wells to 62.5 nM (the final concentration in the wells), and measurements were collected for 60 s. For each sample, the data was fitted with the Gaussian equation. Sample preparation and cryo-EM data acquisition The wild-type and E1404Q mutant AtABCC2 were concentrated to 16 mg mL−1 for cryo-EM sample preparation, respectively. After centrifugation at 11,000 × g for 3 min, 4 μL of the protein sample was applied to the glow-discharged holey carbon grids (Quantifoil Au R1.2/1.3, 300 mesh). The grid was blotted using Vitrobot Mark IV (Thermo Fisher Scientific) operated at 8 °C and 100% humidity with a blotting time of 3 s. To prepare the complex sample, 10 mM ATP dissolved in purification buffer, or 800 μM DNP-GS dissolved in 2% DMSO, was incubated with apo-form AtABCC2E1404Q for 30 min at 4 °C before freezing grids. Micrographs were acquired on a Titan Krios (Thermo Fisher Scientific) electron microscope operating at 300 kV, equipped with the BioQuantum energy filter and a K3 direct electron detector (Gatan). Image stacks were recorded using EPU in the super-resolution mode with defocus values varied from −1.5 to −2.3 μm, an exposure time of 3 s, and dose-fractionated to 32 frames with a total dose of 50 e− Å−2. Cryo-EM data processing Motion correction and dose weighting were performed using the RELION 3.1 implementation of MotionCor255. Defocus values were estimated by CTFFIND456. For apo AtABCC2WT, a total of 1,177,569 particles were automatically picked from 2123 micrographs and then subjected to 2D classification by using cryoSPARC (v3.2.0)57. In total, 471,775 particles from the best classes were put into ab initio reconstruction and non-uniform refinement, yielding a reconstruction monomeric map at an average resolution of 3.8 Å and a reconstruction dimeric map at an average resolution of 4.2 Å. For apo AtABCC2E1404Q, a total of 842,465 particles were automatically picked from 2205 micrographs and then subjected to 2D classification by using cryoSPARC (v.3.2.0). In total, 693,524 particles from the best classes were put into ab initio reconstruction and non-uniform refinement, yielding a reconstruction map at an average resolution of 3.5 Å. For the DNP-GS-bound AtABCC2E1404Q, a total of 1,270,918 particles were automatically picked from 2170 micrographs and then subjected to heterogeneous refinement and ab initio reconstruction. In total, 926,096 particles from the best classes were applied to ab initio reconstruction and non-uniform refinement. After multi-round refinement, yielding a final dimeric reconstruction map at an average resolution of 3.5 Å and a monomeric map at 4.1 Å. For the ATP-bound AtABCC2E1404Q, a total of 1,177,569 particles were automatically picked from 3027 micrographs and subjected to further 2D classification. In total, 1,116,364 particles from the best classes were applied to ab initio reconstruction and heterogeneous refinement, yielding a map with an overall resolution of 3.4 Å. Model building and refinement The final local refinement map was used for model building in Coot58,59. Using UCSF Chimera60,61, the initial rigid body model of AtABCC2 predicted by AlphaFold262 was docked into the construction map. Then, model building and refinement were performed manually by Coot. Structure refinements were carried out by PHENIX63 in real space. Overfitting of the model was monitored by refining the model in one of the two independent maps from the gold-standard refinement approach and testing the refined model against the other map. EM map or structure figures are prepared using PyMol or UCSF ChimeraX. The statistics pertaining to the 3D reconstruction and model refinement can be found in Supplementary Table 1. 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 Supplementary Data 1 Transparent Peer Review file Supplementary Information Reporting Summary Supplementary Data 1 Transparent Peer Review file Source data Source Data Source Data