Multidrug transporters: recent insights from cryo-electron microscopy-derived atomic structures and animal models

Overall structure

ABC transporters in general have a symmetric or pseudosymmetric structure, with two canonical TMDs connected by a flexible linker and two nucleotide-binding domains (NBDs)²⁶. The TMDs are involved in substrate recognition and translocation, whereas the NBDs bind and hydrolyze ATP. The structure can be a single polypeptide chain, as in the case of P-gp and MRP1 (Figures 1A and 1B, respectively), or it could be a homodimer, as is the case for ABCG2 (Figure 1C). Additional features further divide the ABC transporters into sub-families. For example, MRP1, which belongs to the ABCC family, has an additional TMD in the N-terminus (TMD0). Clearly, the TMD0 is not involved in drug–substrate recognition or translocation (Figure 1B). P-gp consists of two TMDs and two functional NBDs. ABCG2, a homodimer, has two identical TMDs and NBDs, and the intracellular loops connecting the TMD to the NBD are shorter than in P-gp or MRP1, resulting in shorter distances between the NBDs as well as between the NBDs and the cell membrane (Figure 1C).

Figure 1.

The overall structure of (A) taxol-bound P-glycoprotein (P-gp), (B) leukotriene C4-bound multidrug resistance protein 1 (MRP1), and (C) estrone-3-sulfate-bound ABCG2. Cartoon representation of the different transporters on the top panels and cartoon and surface representation of the binding cavity as observed from the cytosolic region in the bottom panels. TMD0 of MRP1 is colored in yellow, the N-terminal half of P-gp and MRP1 (transmembrane domain [TMD] 1 and nucleotide-binding domain [NBD] 1) are colored in green, and the corresponding C-terminal halves (TMD2 and NBD2) are colored in cyan. Each monomer of the homodimer of ABCG2 is colored in green or cyan. Ligands bound in the transmembrane region are shown in ball and stick format (gray, carbon; red, oxygen; blue, nitrogen).

The structure of mouse P-gp (mP-gp; 87% identical to human P-gp) has been solved by different groups using X-ray crystallography^27–30. In general, the flexibility of the mP-gp molecule has yielded structures with lower resolution^27–29, while shortening the linker region has increased the resolution by decreasing the mobility of the protein³⁰. Human P-gp seems to have a shorter distance between the NBDs when compared to its mouse counterpart, and only cryo-EM has yielded structural information^20,23. In contrast, MRP1 and ABCG2 structures were unknown until their cryo-EM structures were solved, except for certain specific domains^31,32.

The distance between the NBDs changes during the ATP-binding and hydrolysis cycle, resulting in a change in the conformation of the molecule from inward-facing to outward-facing. The ATP-dependent transport cycle of P-gp is summarized in Figure 2. Briefly, substrates bind in the transmembrane region of P-gp, directly from the membrane (Figure 2A, 2B). ATP binds to the NBDs in the cytosol, favoring NBD dimerization. Conformational changes occur upon ATP binding that bring the NBDs closer together and prevent the release of substrates to the cytosol (Figure 2C). Changes in the distance between the NBDs are translated to conformational changes in the TMDs that allow the translocation of molecules from the cytosol to outside the cell (Figure 2D). ATP hydrolysis occurs at the interface of the NBDs, when they are in closest proximity to each other (Figure 2E). After ATP hydrolysis, the molecule is reset to the inward-open conformation.

Figure 2. Proposed P-glycoprotein (P-gp) transport cycle.

Substrates are presented as blue diamonds. ATP and ADP are presented in black and grey ovals. (A) P-gp in the ligand-free, inward-open conformation. (B) Ligand binds to the transmembrane region. (C) ATP binds to the nucleotide-binding domains (NBDs) and favors their dimerization (outward-occluded conformation). (D) The substrate is effluxed (outward-open conformation) and ATP is hydrolyzed to ADP and inorganic phosphate. Inorganic phosphate and ADP are released (E), and the molecule is reset to the inward-open conformation.

ABCG2 as well as P-gp structures have been solved in the presence of the 5D3- and UIC2-Fab antibodies, respectively. They both include extracellular loops that link the transmembrane helices (TMHs) together, known to be important in terms of blocking the back-influx of substrates as well as for antibody recognition^14,19,21. While two molecules of 5D3 bind symmetrically to ABCG2, a single UIC2 binds asymmetrically to P-gp. Binding of 5D3 prevents the formation of the outward-open conformation. The binding of the antibodies provides structural insights into the extracellular loops in the transporters, with flexibility constrained by the binding. Additionally, increasing the size of the transporters by way of the antibodies has facilitated the resolution of the structure by cryo-EM.

Transmembrane domains

In the case of MRP1 and P-gp, TMD1 and TMD2 contain six TMHs each, arranged in two pseudosymmetric bundles. Bundle one contains the first three and the sixth TMHs of TMD1 and the fourth and fifth TMHs of TMD2, while bundle two contains the remaining TMHs. In the case of ABCG2, each bundle of six TMHs is composed of a monomer. A large transmembrane vestibule is located at the interface of the two bundles, opening to the cytoplasm and penetrating halfway into the lipid bilayer.

TMD0, present in MRP1, has been resolved at a lower resolution when compared to the other TMDs, indicating some flexibility. Interestingly, deletion of TMD0 does not affect transport or ATP hydrolysis³³. Although it may be important for interaction with other proteins, its functional partners have not yet been discovered^21,33,34. The TMD0 structure consists of five TMHs followed by an interfacial domain (lasso motif). Unlike TMD0, the lasso motif is highly conserved and essential for MRP1 folding and function, as it folds against TMH7, TMH15, and TMH16 at the membrane–cytosol interface. The cystic fibrosis transmembrane regulator (CFTR, ABCC7) has a lasso motif that resembles the motif in MRP1, indicating its common importance for folding and trafficking of some of the members of this sub-family³⁵.

Substrate recognition

Substrates are believed to enter the drug-binding pocket of P-gp or ABCG2 directly from the membrane after partitioning into the inner leaflet of the lipid bilayer or directly from the cytoplasm³⁶. The helices in the TMD of P-gp can undergo rotational and translational movements to accommodate substrate entry³⁰. In contrast, substrates for MRP1 can enter only through the cytoplasm²¹.

The changes in the TMHs during the ATP hydrolysis cycle—rotation, translocation, and unwinding—are also reflected in the translocation pathway, supporting the idea of an induced fit model in which the substrate creates its own binding site by interacting with different residues from different TMHs within the translocation pathway, utilizing a number of different interactions including hydrophobic, aromatic, van der Waals, and hydrogen bonds³⁰. The presence of ligands inside the binding site provides invaluable information (simplified in Figure 2B). For example, MRP1 substrate binding has revealed local as well as global conformation changes²¹. The two transmembrane bundles rotate, moving the NBDs 12 Å closer in the presence of LTC₄. This conformational change also explains the increased ATP hydrolysis, given that the NBDs are closer and better aligned. LTC₄ interacts through hydrogen bonds and van der Waals interactions with both transmembrane bundles. The binding site has a positively charged region that coordinates the GSH moiety (P-pocket) and a large hydrophobic area that interacts with the lipid tail (H-pocket)²⁴. The P-pocket is formed by residues of both transmembrane bundles, whereas the H-pocket is formed by residues from transmembrane bundle 2. Most of LTC₄ is buried inside the cavity, except for a carboxylate that points towards the cytoplasm. The H-pocket is larger than the LTC₄ lipid tail, allowing the binding site to accommodate a variety of substrates. In addition, there are a few polar side chains that allow interaction with polar groups from otherwise hydrophobic substrates. Many residues undergo local rearrangement to accommodate the LTC₄ binding. The structure of MRP1 in the presence of ATP in the closed conformation reveals that LTC₄ is effluxed out of the protein prior to ATP hydrolysis²⁴. It also reveals movement of amino acids in the transmembrane region that forces the substrate out of the binding site by weakening the binding interaction and increasing steric hindrance²³.

P-gp transports hydrophobic or amphipathic compounds and some weakly positively charged substrates, in contrast to MRP1, which transports mostly organic acids¹. The translocation pathway in P-gp is mostly hydrophobic with some acidic patches, whereas the one in MRP1 is mostly basic. The structure of intermediate conformations of human P-gp bound to the substrate Taxol and the atomic structure of chimeric mouse–human P-gp bound to the inhibitor zosuquidar has helped in understanding the adaptability, plasticity, and polyspecificity of the binding site of this transporter, supporting the model of an induced fit mechanism for ligand binding^19,20. Particularly, two molecules of zosuquidar are found in the binding pocket of mouse P-gp, occupying most, but not all, of the empty space¹⁹. Similar to MRP1, residues from eight out of 12 TMHs interact with the ligands²⁰. Interestingly, two TMHs, TMH4 and TMH10, seem to play an important role in ligand binding to P-gp by alternating from a straight to bent to straight helix as the ligand binds and is translocated through the transmembrane region²⁰. The helix breakers in these helices seem to be essential to block the release of the substrate to the cytosol, generating a closed gate in the bound conformation as well as a continuum helix upon ATP binding, likely favoring the unidirectional movement of the substrate. The movement of these helices as well as TMH6 and TMH12 also blocks the lateral gate (portal) through which ligands are presumed to enter from the membrane. Structures of the ATP hydrolysis-deficient E-Q mutant of chimeric P-gp bound to zosuquidar in the absence or presence of ATP render a molecule with the two NBDs separated, suggesting that zosuquidar inhibits P-gp function by preventing the dimerization of the NBDs^19,20. Double electron-electron resonance (DEER) spectroscopy has been used to study the energy landscape of the interaction between stimulators of ATP hydrolysis and inhibitors, and it further supports the prevention of dimerization of NBDs by binding of inhibitors in the transmembrane region³⁷.

In contrast to MRP1 and P-gp, in which the binding cavity is wide and flexible, the cavity in ABCG2 is delimited by TMH1, TMH2, and TMH5 from both monomers and seems to be narrower, which likely sterically restricts substrates that can bind in the binding pocket (Figure 1C, bottom). The cavity opens to the cytoplasm and the inner leaflet of the lipid bilayer, and it can accommodate flat, polycyclic, and hydrophobic substrates¹⁴. In addition, there is a second cavity that is accessible only upon conformational changes in the first one. It has less hydrophobic surface area, which might lead to lower substrate affinity, favoring the release of the substrate on the extracellular side during transport. Cholesterol¹⁴, the substrate estrone-3-sulfate (E₁S)²⁵, as well as two different inhibitors, MB136 (a tariquidar derivative) and MZ29 (a Ko143 derivative)²², have all been shown to bind to the first cavity. The structure of ABCG2 bound to cholesterol and MZ29 revealed two molecules in the binding cavity, whereas for E₁S or MB136 only a single molecule was bound. Interestingly, the lower density found for the latter compounds suggests that these molecules may have more than one way of interacting with the binding site. Specific mutation of the amino acids found to interact with E₁S provides insights into the role of each residue for the binding of the substrate²⁵.

Although the structures of ABCG2 and P-gp with substrates and inhibitors bound in the transmembrane region have been solved, it is still not completely clear what exactly makes a molecule become an inhibitor or a stimulator of ATPase activity. Clearly, higher affinity, as a result of more local interactions and better coverage of the binding cavity, seems to be the driving force that distinguishes between a stimulator and an inhibitor by locking the transporters in a conformation in which the two NBDs can no longer dimerize^20,22,25.

NBD–TMD interface and NBD1–TMD2 linker

In general, the interfaces of the transporters do not seem to change significantly upon ATP binding; they move as rigid motifs accompanying NBD movement. In the case of ABCG2, the linker, which is an extension of TMH1, is much shorter and highly charged, yet the structure of this region remains unknown, suggesting high flexibility.

When the structure is a single polypeptide chain, such as in the case of P-gp and MRP1, the NBD1 and the TMD2 are connected by a flexible linker. Even though this linker is essential for proper function, it generally appears unstructured in most of the ABC transporter structures resolved to date^{19–21,23,24}.

Nucleotide-binding domains

In all three transporters, when ATP is bound, the NBDs dimerize and the TMDs open to the extracellular region, referred to as the inward-closed or outward-facing conformation. In the inward-open conformation, the two NBDs are separated from each other; however, during the transport cycle, they dimerize in a “head-to-tail” configuration to form two combined ATPase sites, with the Walker A/B motifs forming one NBD and the signature motif of the other. Interestingly, ABCG2 does not have an A-loop²². The adenine is stacked by an aromatic residue, as is found in other transporters. Rather, van der Waals interactions and salt bridges compensate for the binding²⁵. ATP hydrolysis is catalyzed via a general base mechanism in which a glutamate residue polarizes the hydrolytic water for inline attack of the ATP-γ-phosphate. Mutations to the highly conserved Walker B catalytic glutamate residue render P-gp, MRP1, and ABCG2 capable of binding ATP but not hydrolyzing it. However, the binding of transport substrate is not altered³⁸. These E-Q mutants have been informative in solving the ATP-bound closed conformations of several ABC transporters^23–25,39.

Cryo-EM studies performed on purified P-gp in detergent micelles complexed with and without UIC2-Fab at different stages of the ATP hydrolysis cycle have provided key information on the movement of the NBDs during this process. Structures have been reported for the unbound state (apo), ATP-bound before hydrolysis, ADP-V_i-bound post hydrolysis, and ADP-bound after P_i release^18,19,23. The apo form was found to be present in the inward-facing conformation as well as in an outward-facing conformation. The ATP-bound pre-hydrolysis sample yielded a mixture of closed and open conformations in a different proportion compared to that observed in the apo structure¹⁸. ATP-bound after hydrolysis and trapped in the transition state with V_i replacing the outgoing P_i yielded mostly a closed structure with the two NBDs in the dimerized conformation¹⁸. Lastly, the post-hydrolysis and after release of the P_i states yielded structures in mostly the open conformation, suggesting that the presence of the ADP prevents the closing of the structure¹⁸. The structure of the ATP-bound outward-facing conformation suggests that the substrate is effluxed from the drug-binding cavity before ATP is hydrolyzed and that the energy from the ATP hydrolysis is essential for restoring the inward-facing conformation²³. This proposed mechanism is based on the atomic structure of the ATP-bound E-Q mutant transporter²³. This needs further validation by obtaining the high-resolution structure of the ATP-bound wild-type transporter in the presence and absence of a transport substrate.

In P-gp, both NBDs are highly homologous and capable of hydrolyzing ATP; however, they depend on each other for catalysis. A recent crystal structure shows preferential binding of ATP NBD1 in the shorter linker mP-gp, supporting the hypothesis of asymmetric binding³⁰. Only one ATP hydrolysis can occur at a time; inactivation of one of the sites inactivates ATP hydrolysis. How the two ATP-binding sites alternate remains unknown. DEER spectroscopy was used to study spin-labeled mutants in a variety of conditions⁴⁰. Distance distributions between each different pair of labels agree with a model that fluctuates between an inward-facing and an outward-facing conformation, with extensive conformational changes during the transition state. In this state, the distribution between intracellular labels was homogeneous, whereas the distribution in the extracellular region was more variable, highlighting a reconfiguration of the TMDs. Labels on the NBDs showed structural asymmetry and supported the alternating site hydrolysis model, in which two-stage hydrolysis is required to translocate drugs outside the cell⁴⁰. In contrast, MRP1 has NBD1 with significantly lower ATPase activity compared to NBD2^21,35. Cryo-EM structures of MRP1 in the apo, ligand-bound, and ATP-bound conformations show three different distances between the NBDs. Transmembrane bundle 1 and NBD1 as well as transmembrane bundle 2 and NBD2 move, bringing the NBDs closer together. A substrate can bind to the protein only before ATP binds, and binding of the substrate brings the NBDs closer together, resulting in enhanced ATPase activity. On the other hand, ABCG2 has shorter intracellular loops, resulting in shorter distances between the NBDs, which remain in contact despite the absence of nucleotide.

Based on the proposed mechanism described in Figure 2D, the substrate is released prior to ATP hydrolysis, suggesting that ATP hydrolysis is not needed for the transport function of P-gp but rather to reset the molecule to the inward-open conformation (Figure 2E). The residues that are found to interact with the ligands collapse, leaving no room for the substrate to be bound. In order to be released, the substrate has to be pushed towards the extracellular region, to a region of the protein where the affinity of the substrate is decreased. Additional high-resolution structures of wild-type transporters in different conformations including ATP- and ligand-bound will help to understand the role of ATP hydrolysis in the transport cycle.

The binding of ATP to the NBDs is accompanied by changes in the transmembrane region that result in a completely blocked binding site and translocation pathway. In contrast to the bacterial homodimer multidrug transporter (Sav1886), these transporters do not show a much wider outward-open conformation upon ATP binding⁴¹. The structure instead corresponds to a narrow translocation pathway for MRP1, P-gp, and ABCG2 that is protected from the lipid bilayer^23–25,39.

Membrane environment

The importance of a lipidic environment for transporter function has been previously shown^30,42,43. Cryo-EM structures of P-gp and ABCG2 have identified ordered cholesterol and phospholipid molecules directly interacting with the transmembrane region^20,22, suggesting that the lipids in the membrane could modulate the conformational changes associated with binding of substrates and inhibitors. In addition, it has been demonstrated that P-gp inhibitors no longer inhibit ATP hydrolysis when the protein is in detergent micelles⁴⁴, whereas the inhibition is recovered when the protein is reconstituted in nanodiscs or proteoliposomes. Also, it has been shown for the MsbA transporter that the distance between the NBDs is larger when the protein is in detergent micelles⁴⁵. These findings highlight the importance of using a membrane environment for structural studies^42,43 and show that purified protein in a detergent micelle environment is not suitable for studying the interaction of substrates and high-affinity modulators with P-gp. As all structural studies with MRP1 have been done in detergent micelles^21,24, it will be of interest to see if the same observations hold true for MRP1 reconstituted in nanodiscs.