Cryo-EM of human P-glycoprotein reveals an intermediate occluded conformation during active drug transport

Abstract

P-glycoprotein (Pgp) is an important human multidrug transporter that contributes to pharmacokinetics and multidrug resistance. Despite decades of study, the conformation transition cycle of Pgp undergoing active drug transport is not defined, thus the precise relevance of all available Pgp structures to uninterrupted multidrug transport remains unclear. Here, we use cryo-EM of membrane-embedded human Pgp under continuous turnover conditions to analyze the conformational ensembles of Pgp transporting distinct substrates. These results delineate multiple conformations including inward-facing and closed conformations, highlighting the occluded conformation as a critical intermediate state between transporter closure and substrate release. A combination of structural, functional, and computational studies reveals the transmembrane helices 4 and 10 undergoing drastic rearrangement to coordinate substrate binding, occlusion, and release, and identifies a peripheral site involved in substrate capture and Pgp inhibition. Together, our results provide a set of snapshots of Pgp undergoing continuous drug transport, unveiling the intricate interplay between transporter dynamics and drug movement, and shed light on the mechanism of polyspecificity.

In this work, the authors determine the conformational ensemble of human P-glycoprotein during active drug transport using cryo-EM. This reveals key intermediate states and a peripheral substrate-binding site providing insights into P-glycoprotein’s transport mechanism.

Introduction

P-glycoprotein (Pgp) is an ATP binding cassette (ABC) transporter mostly expressed at blood-organ barriers and epithelial cells, and extrudes xenobiotic and toxic compounds out of the cell¹. Pgp contributes to multidrug resistance (MDR) in cancer cells² and to pharmacokinetics of drugs due to its localization at epithelial barrier sites³. Pgp displays extraordinary substrate polyspecificity, exporting highly diverse compounds such as verapamil and vinblastine, which are inhibitors of calcium channels and microtubule assembly, respectively^4,5. Inhibitors of Pgp have long been desired to be included in chemotherapeutic cocktails to reduce MDR in cancer cells, yet have largely been unsuccessful in clinical trials⁶. Interestingly, some Pgp inhibitors can be transported at low concentration⁷, while some Pgp substrates inhibit ATP hydrolysis at high concentration^5,8, leaving no defined line between Pgp substrates and inhibitors. Understanding the fundamental mechanism of Pgp-compound interaction is critical for prevention of MDR and for better design of drugs with improved oral uptake and pharmacokinetics⁶.

Pgp contains two transmembrane (TM) domains (TMDs) and two cytosolic nucleotide binding domains (NBDs) within a single polypeptide chain. According to the prevalent alternating access model, Pgp binds substrates in the inward-facing conformation, followed by a transition to the outward-facing conformation, leading to substrate release outside of the cell, all driven by the energy of ATP binding and hydrolysis. However, the Pgp transport cycle is likely more complex, apparent by the identification of numerous substrate binding sites⁹, inward kinking of TMs¹⁰, and an NBD closed state with no substrate binding pocket¹¹. In addition, NBD closure requires Pgp binding to both transport substrate and ATP¹², which have been proposed to be cooperative⁵. Pgp binds and hydrolyzes ATP in its two ATP sites in an asymmetric manner⁸. Structural studies of Pgp have primarily used orthologous genes and revealed several inward-facing conformations with a central cavity for accommodation of substrates and inhibitors^13–22. Recently, cryo-EM was used to determine the structures of human Pgp, though bound to a conformation-selective inhibitory antigen-binding fragment (Fab)^10,20 or containing catalytic-deficient mutations¹¹. Thus, the key conformational states of human Pgp undergoing active drug transport remain undefined, and the crucial questions regarding how highly diverse drug molecules are captured, moved, and released by human Pgp are yet to be answered.

Here, we utilize cryo-EM to dissect the conformational ensembles of nanodisc-embedded human Pgp under continuous transport of different clinically approved drugs. Our work delineates the major conformational states of actively functioning Pgp and reveals the structural basis of Pgp interaction with chemically distinct compounds, thus offering insights into the enigmatic polyspecificity of Pgp.

Results and discussion

Preparation and biochemical characterization of Pgp in nanodiscs

Purification of human Pgp was facilitated by selecting Pgp-expressing cells with high level of drug export activity (Supplementary Fig. 1a). The Pgp protein was extracted and purified using dodecyl-β-D-maltopyranoside (DDM) and cholesteryl hemisuccinate (CHS), and subsequently reconstituted into lipid nanodiscs showing expected size and shape in negative-stain EM (Supplementary Fig. 1b–d). The ATPase activity of Pgp in nanodiscs shows the basal ATPase of roughly 0.25 nmol/μg/min, which is comparable to previous studies of Pgp in lipid nanodiscs^8,10. This ATPase activity was stimulated by three structurally and chemically distinct substrates: vinblastine up to ~12.5 μM, verapamil up to ~25 μM, and rhodamine up to ~200 μM; further increasing vinblastine and verapamil concentrations led to pronounced decrease of Pgp activity, while rhodamine at high concentration caused only minor inhibition (Fig. 1a). The basal ATPase, maximal substrate stimulation, and degree of substrate inhibition for these substrates varies across studies^8,10,11, likely influenced by differences in membrane composition. However, the biphasic substrate stimulation of Pgp ATPase activity shown here is consistent with previous findings^5,8.

Fig. 1. Analysis of Pgp under continuous vinblastine transport.

a ATPase activity of Pgp in presence of the substrates rhodamine 123, verapamil, and vinblastine. Inset shows identical data but only up to substrate concentrations of 100 μM. ATPase was performed technical replicates in either duplicate or triplicate (n = 2–3), and data are presented as mean values ± standard deviation. b Cartoon showing the analysis of Pgp under continuous turnover conditions. Pgp sample is mixed with Mg-ATP and drug, generating a sample of Pgp that is continuously transporting drug and hydrolyzing ATP. This is rapidly applied to cryo-EM grid and plunge frozen. c Side view of overlay of two different inward-facing conformations of Pgp bound to vinblastine. d Bottom-up view of the NBDs. Distance measurements are from Walker A motif serine and signature motif serine with nucleotide binding site 1 (NBS1) from Ser429 to Ser1177 and NBS2 from Ser532 to Ser1072. e Cryo-EM maps of Pgp in different conformational states, including Vin-ATP in inward-facing state (Vin-ATP-IF1 and Vin-ATP-IF2), NBD closed state (Vin-ATP-CL2), and closed collapsed state (Vin-ADPVi-CC). The Vin-ADPVi-CC was acquired by the addition of sodium vanadate to trap the closed state. All models and cryo-EM maps are colored as TMD1-NBD1 in blue, TMD2-NBD2 in orange, and vinblastine in green. Source data are provided as a Source Data file.

Pgp transports vinblastine through multiple inward-facing and closed conformations

We used single-particle cryo-EM to characterize the conformational ensemble of Pgp undergoing continuous vinblastine transport (Vin-ATP; Fig. 1b). Pgp (~25 μM) with vinblastine (100 μM) was pre-warmed at 37 °C, and a large excess of Mg-ATP (~8 mM ATP) was added to initiate the reaction, followed by rapid application to cryo-EM grid and plunge freezing (See “Methods” for details). Pgp hydrolyzes ~30 ATP molecules per minute in the presence of 100 μM vinblastine (Fig. 1a), which is consistent with previous findings⁸. The short reaction time (within 15 s) allowed hydrolysis of ~7 ATP per Pgp, leading to ATP concentration reduction by only ~175 μM. Hence, the resulting cryo-EM images are snapshots of Pgp actively transporting vinblastine in the presence of saturating concentration of Mg-ATP.

Two-dimensional (2D) analyzes of cryo-EM particle images of Pgp showed clear structural features (Supplementary Fig. 2a, b), indicating good quality of EM data. However, routine three-dimensional (3D) classification starting with one single 3D density map did not effectively separate particle images according to different Pgp conformations. After extensive experimentation, we employed a strategy named “multiple models at multiple resolutions (M3R)” for robust and reproducible 3D classification (Supplementary Fig. 2c, see “Methods” for details). Throughout this study, all cryo-EM datasets of Pgp under different conditions were processed following the same overall workflow, so that all results can be directly compared.

3D classification of the cryo-EM particle images of Pgp undergoing vinblastine transport demonstrated a mix of inward-facing and closed conformations, as well as smaller variations within each of these conformations (Supplementary Fig. 2d). We obtained two final cryo-EM maps of Pgp in inward-facing conformations at 3.9- and 4.1-Å resolution, termed IF1 and IF2, respectively (Fig. 1c–e and Supplementary Figs. 3–4). The quality and resolution of these density maps are the best in the TMDs and decrease towards the NBDs, allowing for accurate model building in TM region and confident secondary structure placement in the NBDs. When superimposed based on the TMDs, the IF1 and IF2 structures show little difference within the membrane (Fig. 1c), but IF2 exhibits gradual inward movement moving down towards the NBDs (Fig. 1c). As a result, the nucleotide binding sites (NBS1: S429-S1177; NBS2: S532-S1072) in IF2 are ~4.6 Å closer than those in IF1 (Fig. 1d). These NBS distances are greater than those in the Pgp bound to an inhibitory Fab^10,20, and are within the broad range of NBS distances of early studies of Pgp using crystallography (Supplementary Fig. 5a).

Our study also generated density maps at ~6 Å resolution of Pgp in two closed conformations (termed CL1 and CL2) which are characterized by dimerized NBDs and clear separation of the TMDs in the region between the membrane and NBDs (Fig. 1e and Supplementary Fig. 2d). Such features are reminiscent of the occluded conformation reported for other ABC transporters^23,24, and distinct from the fully closed conformation of the ATP-bound catalytically deficient Pgp mutant that shows tightly dimerized NBDs and a collapsed central cavity¹¹. The lack of fully closed conformation from our cryo-EM analysis suggests that such a conformation is either highly transient or completely absent for Pgp transporting vinblastine. To test if Pgp can sample the fully closed conformation, we characterized the Pgp conformations in the presence of vinblastine, Mg-ATP, and sodium vanadate (Vin-ATPVi; Fig. S6). The resulting cryo-EM structure of vanadate-trapped Pgp at 3.9-Å resolution is nearly identical to the structure of fully closed ATP-bound Pgp mutant¹¹ (Fig. 1e and Supplementary Fig. 7). Furthermore, parallel analyzes were performed for Pgp with vinblastine in the absence of nucleotide (Vin-alone; Fig. S8) and Pgp in the apo form (Apo; Supplementary Figs. 9–10). Neither condition generated Pgp with tightly associated NBDs, supporting that NBD closure requires nucleotide¹². Moreover, comparison of the Vin-alone and Apo IF states with the VinATP IF states revealed additional density at the NBS in the VinATP cryo-EM maps, indicative of nucleotide bound, but the limited resolution precludes confident identification of these nucleotides. Together, our results demonstrate that vinblastine-transporting Pgp undergoes dynamic conformation transition cycles that contain multiple inward-facing and closed conformations.

Vinblastine binding to Pgp involves TM10 kinking

A strong vinblastine density was observed in the central cavity of Pgp in both IF states (Fig. 2a, b and Supplementary Fig. 11a, b), exhibiting comparable size and shape (Supplementary Figs. 3d and 4d). These densities closely resemble the corresponding density in the cryo-EM map of Pgp bound to vincristine, an anticancer drug similar to vinblastine²⁰. In the IF states of vinblastine-bound Pgp, TM10 is inward kinked, while TM4 is straight (Fig. 2a). This is in contrast to Pgp bound to vincristine and an inhibitory Fab, showing inward kinking of both TM4 and TM10²⁰. Interestingly, the position of the inward kinking of TM4 is instead occupied by a second vinblastine molecule in our cryo-EM maps (Fig. 2a, c), thus defining a peripheral substrate site (described in a later section). The drastic kinking of TM10 initiates at Pro866 and creates a hydrophobic pocket below the catharanthine moiety of vinblastine (Fig. 2a). This pocket is lined by several hydrophobic residues: M949, Y950, M68, M69, F336, L65, and I340 (Fig. 2d, e). The vindoline moiety of vinblastine is more centralized within the pocket and forms primarily hydrophobic contacts with some polar interactions at the bottom of the binding pocket involving Q946, Q347, Q990, and E875 (Fig. 2d, e). M986 further sits in the crevice between the catharanthine and vindoline moiety of vinblastine (Fig. 2e). In a molecular dynamics (MD) simulation, the vinblastine in the central cavity of Pgp shows a root mean square deviation (RMSD) of 3.4 Å with small vibrational movements (Supplementary Fig. 12a), consistent with a relatively stable binding pose.

Fig. 2. Vinblastine binding to Pgp.

a Orthogonal side views of the Pgp transmembrane helices with vinblastine binding to central cavity and peripheral site. The structural model of Pgp in apo is gray, and the IF1 conformation model is colored. Cryo-EM densities of vinblastine are colored as light green in central cavity and dark green in peripheral site. Right panel shows only TM4, TM5, and TM10. b, c Top-down view of IF1 showing slices of model and cryo-EM map at the height of central cavity (b) and peripheral site (c), as indicated in (a). d, e Zoomed-in view of the vinblastine binding site with interacting side chains shown as sticks. e is rotated 180° from (d). f Diagram of vinblastine binding to Pgp. g In-cell transport assay of Pgp transporting rhodamine 123. Flow cytometry histograms show accumulation of rhodamine in cells that express WT or mutant Pgp. Gating strategy for (g) is provided in the Source Data file. WT is defined as wild-type, and Zos is Zosuquidar.

Pgp transports verapamil through inward-facing, occluded, and closed conformations

To investigate how Pgp transports diverse compounds, we conducted parallel cryo-EM analysis of Pgp under active verapamil export (Ver-ATP; Supplementary Fig. 13). Verapamil is distinct from vinblastine in size, structure, and chemical property, and led to higher ATPase activity of Pgp with less substrate-induced inhibition (Fig. 1a). We obtained two inward-facing cryo-EM maps, termed IF1 and IF2 (Fig. 3a, b), at overall resolutions of 4.0 Å (Supplementary Fig. 14) and 4.4 Å (Supplementary Fig. 15), respectively. IF1 is slightly more open than IF2, similar to the two IF conformations of Pgp transporting vinblastine (Fig. 1c, d). Notably, unlike Pgp bound to vinblastine, Pgp with verapamil in the IF states exhibits no kinking in TM10 (Fig. 3), indicating differential interactions between TM10 and distinct substrates.

Fig. 3. Conformational ensemble of Pgp transporting verapamil.

a Cryo-EM maps of Pgp transporting verapamil in different conformational states, including two inward-facing states (Ver-ATP-IF1 and Ver-ATP-IF2), occluded state (Ver-ATP-OC), and collapsed closed state (Ver-ATP-CC). b Ribbon models for the maps in (a). TM10 and TM4 are colored red and cyan, respectively. c Ribbon model showing only TM4 (cyan), TM6 (blue), TM10 (red), and TM12 (orange), rotated 120° with respect to (b). Verapamil cryo-EM density is shown pink. Positions of TM4 and TM10 kinking at Pro223 and Pro866 in the occluded conformation are shown, respectively.

To better understand the correlation between IF conformational dynamics and Pgp transport function, we compared the NBS distances in all IF states from this study with all previous structures of Pgp (Supplementary Fig. 5). NBS2 is always closer compared to NBS1, showing consistent NBD asymmetry, which is defined as the difference between NBS1 and NBS2 distances. There is no clear hinge region that drives the asymmetry; instead, we observed a gradual increase in asymmetry progressing from the distal end of TMs to the NBDs (Supplementary Fig. 16f). 3D variability analysis (3DVA) revealed two major components of conformational variation consistently present in all datasets: NBD rotation (Supplementary Fig. 16a) and NBD translational opening and closing (Supplementary Fig. 16b). Analysis of 3DVA in “cluster” viewing mode generated ten separate reconstructions (Supplementary Fig. 16c) per dataset. Interestingly, this analysis showed that vinblastine binding reduced the NBD distances compared to the apo state, while the continuous transport conditions slightly increased the NBD distances. The latter is likely due to the fact that the 3D reconstructions captured during drug transport are snapshots of Pgp undergoing continuous conformational transition. Presumably, the Pgp proteins with smaller NBD distances are more ready to close, and those with greater NBD distances take a longer time to close, thus becoming more long-lived and better represented in the cryo-EM data. This analysis showed more variation in NBD asymmetry in the two datasets without nucleotide (Apo and Vin-alone), indicating that the conformation of NBDs is more restricted under active substrate transport with nucleotide present (Supplementary Fig. 16d).

Our analysis of verapamil-transporting Pgp generated two cryo-EM maps characterized with closed NBDs, representing the occluded state (Ver-ATP-OC) and collapsed closed state (Ver-ATP-CC) at 4.4- and 4.1-Å resolution, respectively (Supplementary Fig. 17–18). Although the lower resolution did not allow modeling of most side chains, the well-defined secondary structural features enabled reliable modeling of the overall secondary structure. While the latter exhibits a completely collapsed central cavity, the TMDs in the former remain separate within the membrane, thus occluding the central cavity from both the cytosol and extracellular space. Inside the occluded cavity, there is a clear density consistent with the size and shape of a verapamil molecule (Fig. 4g, h). This density is located slightly higher than the verapamil in the IF conformations (Figs. 3c and 4f), indicating an intermediate position between verapamil binding and release. The most distinct feature in the TM region of the occluded state is the inward kinking of TM4 and TM10 (Fig. 3c), both initiating at a proline (Pro223 and Pro866) and extending to the space immediately below the verapamil in the central cavity. Despite lower resolution of the TM region in this map, the secondary structure features strongly support that these TMs are kinked inwards. The coupling helices (CH), CH2 connecting TM4-5 and CH4 connecting TM10-11, bridge the conformational changes of the TMDs and NBDs. Thus, the inward kinking of TM4 and TM10 and NBD dimerization are coupled together (Supplementary Fig. 19a). The occluded conformation has been observed for other ABC transporters such as McjD and PCAT1^23,24, in which the TMs bowed out to form a substantially larger occluded cavity. While the occluded state is likely a common intermediate between the inward-facing and fully closed states, different ABC transporters can use distinct TM rearrangements to fulfill conformational transition and confer substrate specificity. Further, these structures are distinct between recent mouse Pgp structures with covalently cross-linked substrates, which resemble the closed, collapsed state with a bulge in TMs to accommodate the substrate²¹.

Fig. 4. Verapamil binding to Pgp.

a Cross-sectional side view of the IF2 model showing non-proteinaceous density in the central cavity. b Zoomed-in view of verapamil binding site showing only verapamil EM density. c, d Verapamil binding in central cavity with interacting side chains shown as sticks. d is rotated 180° relative to c, e Diagram showing verapamil binding to Pgp in IF2 state. f Overlay of verapamil models in inward-facing (gray) and occluded (pink) conformations with vinblastine model in IF1 conformation (green). g Cross-sectional side view of occluded closed state showing non-proteinaceous density. h Superimposition of the model and EM density of the verapamil molecule in the occluded Pgp.

The collapsed closed state (Ver-ATP-CC; Fig. 3) represents the conformation after substrate release and is similar (r.m.s.d 0.77 Å) to the post-hydrolysis collapsed closed structure with ADP-vanadate (Vin-ADPVi-CC; Fig. 1e). Despite its lower resolution, the VerATP-CC cryo-EM map shows a similar structure as Vin-ADPVi-CC, supports the important role of such conformation in the transport cycle of Pgp. Compared to the occluded state (Ver-ATP-OC), the collapsed closed state (Ver-ATP-CC) exhibits minimal changes in the outer eight TMs (TMs 1–3, 5, 7–9, 11) (Supplementary Fig. 19b, c), but drastic rearrangements in the inner four TMs (TMs 4, 6, 10, 12) (Fig. 3b, c and Supplementary Fig. 19d, e). The inward kinking of TM4 and TM10 characteristic in the occluded state is straightened back and bowed outwards (Fig. 3c and Supplementary Fig. 19d, e), opening space for TM6 and TM12 to migrate inwards and collapse the central cavity (Supplemental Video 1). In sum, our analysis of verapamil-transporting Pgp (Ver-ATP) defined all key conformational states of actively functioning Pgp: two IF states for substrate binding, occluded state for substrate accommodation upon NBD dimerization, and a collapsed closed state after substrate release.

Verapamil binds to Pgp with distinct features of interactions

A clear non-proteinaceous density inside the TMDs of Pgp in both IF states matches the shape and size of one verapamil molecule (Fig. 4a, b). Since this density is stronger in the IF2 state, further discussion focuses on IF2 state. The surfaces of the central cavity involved in verapamil and vinblastine binding largely overlap (Fig. 4f), though with some notable differences. Met986 is positioned in the crevice between the vindoline and catharathine moieties in vinblastine (Fig. 2d, e), but this crevice is absent for verapamil, likely making the role of Met986 less pronounced when compared to vinblastine (Fig. 4c, d). In addition, the catharathine moiety of vinblastine extends upwards towards Y953 and M949 (Fig. 4f), while verapamil is further away from this position (Fig. 4c, d). The verapamil and vinblastine molecules within the central cavity are positioned near the bottom of the M-site but do not fully extend into it. Portions of these molecules extend into the upper region of the R-site, as previously described²⁵. MD analysis of verapamil resulted in RMSD of 4.0 Å (Supplementary Fig. 12e Supplementary Data 5), which is higher than that for the bound vinblastine (3.4), indicating less conformational constraint for verapamil. It is conceivable that, due to less rigid binding to Pgp, verapamil represents more efficient transport substrates exhibiting higher stimulation of ATPase activity and less Pgp inhibition (Fig. 1a), which agrees with previous studies^21,26. Further, it is possible that verapamil, having a smaller size and more rotatable bonds, plays a role in its higher efficiency as a Pgp substrate^27,28. The MD simulation of the Ver-ATP-OC state showed a highly dynamic verapamil molecule with RMSD of 7.0 Å (Supplementary Fig. 12f and Supplementary Data 6), supporting the notion that the substrate becomes less stable upon Pgp transition from the inward-facing to the occluded conformation.

Pgp captures substrates using a peripheral site

Our cryo-EM structures of vinblastine or verapamil transporting Pgp (Vin-ATP and Ver-ATP) in the IF states revealed an additional substrate molecule near TM4-5 below the central cavity, defining a peripheral substrate binding site. The cryo-EM density at this site is slightly weaker compared to the central cavity. This site involves mainly small residues (G226, A229, A233, A302, and G346) surrounded by larger hydrophobic residues (F343, I306, F303, I229), forming a shallow hydrophobic pocket (Fig. 2d, e). This binding site is distinct from the “vestibule” described previously for inhibitors, which is near TM12 and TM9²⁰. Further, this site is close to the proposed R-site²⁵, but more towards the TM4-TM6 opening. Substrate binding to this pocket sterically blocks the inward kinking of TM4 seen in the occluded conformation, presumably causing Pgp inhibition by substrates at high concentration (Fig. 1a). In the peripheral site of vinblastine-transporting Pgp, the vinblastine density in the IF1 state is mostly complete and sticks out between TM4 and TM6 (Fig. 2c and Supplementary Fig. 11a), whereas this density in the IF2 state is less complete (Supplementary Fig. 11c) and correlated with smaller NBD opening (Fig. 1c). In comparison, the verapamil density at the peripheral site is less defined (Supplementary Fig. 11c), consistent with less rigid Pgp interaction and lower substrate-induced inhibition (Fig. 1a). Both TM4 and TM6 contain a glycine (G226 and G346) at the level near the top of the peripheral vinblastine (Fig. 2d, e) which may increase mobility of these TMs and facilitate substrate entry.

To directly investigate the importance of the peripheral substrate site, single point mutations of A233R and I299R were generated to block substrate binding to the peripheral site without affecting the central cavity. Consistent with previous finding²⁹, these mutations did not affect protein folding or expression level (Supplementary Fig. 11e–g). In the cell-based rhodamine transport assay, the inhibitor zosuquidar or the catalytic deficient mutation (E1201Q) caused Pgp inhibition, and the A233R and I299R mutants exhibited pronounced reduction of transport activity (Fig. 2g). This supports the involvement of the peripheral site in substrate transport. MD simulations were performed using the structures of apo Pgp (Supplementary Fig. 12h and Supplementary Data 8) and Pgp containing vinblastine in only the central cavity, only in the peripheral site, or in both sites (Supplementary Fig. 12 and Supplementary Data 1–4). Samples with vinblastine bound to the peripheral site were analyzed with manually straightened TM10 to remove any influence it would have on the dynamics of the peripheral site-bound vinblastine molecule. Through the simulations, TM10 stayed straight when both central cavity and peripheral site were occupied by vinblastine, but slightly kinked inward when only the peripheral site was bound or neither site was bound. When present only in the central cavity, vinblastine is relatively stable with minor vibrational movement (RMSD 3.4 Å), whereas the vinblastine only in the peripheral site showed higher mobility (RMSD 5.3 Å) and an upward movement (Supplementary Fig. 12a, b and Supplementary Data 1–2). Notably, when bound simultaneously to both sites, the two vinblastine molecules were stabilized with RMSD values of 2.3 and 3.1 Å, respectively (Supplementary Fig. 12c and Supplementary Data 3), and vinblastine bound to peripheral site exhibited weaker binding in the calculated binding energy (Supplementary Fig. 12c and Supplementary Data 3). Interestingly, the simulation with a kinked TM10 displayed higher RMSD compared to that with a straight TM10, yet the weaker binding energy at the peripheral site was maintained (Supplementary Fig. 12d and Supplementary Data 4). Further, these results were consistent with the simulation of Pgp in POPG nanodiscs (Supplementary Fig. 12g and Supplementary Data 7). In sum, our results suggest that Pgp substrates interact with the peripheral site with weak affinity before moving into the central cavity, and that some substrates can stably occupy both sites at the same time to hinder the conformational transition required for substrate export.

Comparison with previously published Pgp structures

Recent cryo-EM structures of Pgp provide valuable insights into its interactions with various inhibitors and substrates^{10,11,18–21}. Unlike the unrestrained sample described in this study, previous Pgp structures were stabilized using techniques such as inhibitory Fab binding^10,18,20,21, cross-linking²¹, or catalytically deficient mutation¹¹. Structures with inhibitory Fab bound (10, 18, 20, 21; PDBs 8Y6I, 7A65, 7A69, 7A6C, 7A6E, 7A6F, 6QEX, 6QEE, 6FN1, 6FN4) consistently show kinked TM4 and TM10, which resemble our VerATP-OC conformation, albeit without dimerized NBDs. Notably, even the Fab-bound apo state (20; PDB 7A65) displays inward kinked TMs, contrasting with the straight TM4 and TM10 observed in the crystal structures (17; PDB 4Q9H) and our cryo-EM structures of apo Pgp. The ligand binding positions in the central cavity observed in our structures align well with those in previous structures (Supplementary Fig. 20a–c), and our cryo-EM maps with verapamil and vinblastine show no evidence of their extension into the “vestibule,” as seen with inhibitor-bound states (20; Supplementary Fig. 20d). Furthermore, the peripheral site identified in our structures is not present in the cryo-EM structure of Pgp bound to three molecules of elacridar and an inhibitor Fab³⁰ (Supplementary Fig. 20e). In fact, the vinblastine molecule in the peripheral site (next to the straight TM4) overlaps with the kinked TM4 in the previous structures (Supplementary Fig. 11d), suggesting that substrate binding to the peripheral site and TM4 kinking are mutually exclusive.

The first NBD closed structure of Pgp was determined using E-to-Q mutant that can bind but not hydrolyze ATP¹¹, closely resembling our ADP-Vi and VerATP-CC states of wild-type Pgp. More recently, the structures of Pgp trapped with E-to-Q mutation or ADP-Vi, along with covalently linked substrates in the exit tunnel, show a comparable closed conformation but with TM1 dilation to accommodate the substrate²¹. With outward-bowed TM4 and TM10, these structures are distinct from our VerATP-OC state, likely representing a later state after substrate exits the central cavity. Our cryo-EM analysis of actively transporting Pgp did not generate a structure with substrate positioned in the exit tunnel, supporting that the substrate release is a highly transient process and challenging to capture during uninterrupted functional cycle.

Pgp transport mechanism

In this study, we established a workflow to investigate the structural dynamics of Pgp undergoing continuous drug transport and generated a nearly complete molecular movie of the actively functioning Pgp, enabling us to propose an up-to-date model for the Pgp transport mechanism (Fig. 5).

Fig. 5. Model for Pgp transporter cycle.

All conformational states show TMD1-NBD1 as blue and TMD2-NBD2 as orange. Drug molecules are represented as green rectangles, ATP molecules are represented as yellow triangles, and ADP and Pi are represented in red. See text for description of proposed transporter cycle of Pgp. Conformational states include (1) apo, (2) inward-facing with substrate bound to peripheral site, (3) inward-facing with substrate bound to central cavity, (3i) inward-facing inhibited with substrate bound to both central and peripheral site, (4) occluded, (5) outward-facing, and (6) collapsed closed.

Before substrate binding, Pgp resides in the IF states (state 1). The substrate presides in the inner leaflet of the membrane as demonstrated in several studies^31–33. Substrate initially moves laterally in the inner leaflet and binds Pgp at the peripheral site near TM4 and TM5 (state 2) and subsequently migrates up into the central cavity (state 3), which induces inward kinking of both TM4 and TM10. Some substrates can simultaneously occupy the central cavity and the peripheral site, and the two substrate molecules stabilize each other (step 3i), sterically blocking the inward kinking of TM4 and causing substrate-induced inhibition. In contrast, productive inward kinking of TM4 and TM10 promotes NBD dimerization and facilitates the transition to the occluded conformation (state 4). The substrate is released in a hypothetical outward-facing state of Pgp (state 5), which is very short-lived, evident from its absence from our cryo-EM analyzes. This state may be similar to the collapsed closed state, except with more flexible TM conformation at the level of the outer membrane leaflet, allowing the substrate to be squeezed out³⁴, reminiscent of a peristaltic pump²⁰, as described above. Following substrate release, Pgp proceeds to the collapsed closed state (state 6). ATP hydrolysis by the NBDs allows dissociation of the NBD dimer and resets Pgp for the next transport cycle.

Insights into Pgp polyspecificity

Our results reveal several key factors that can contribute to the polyspecificity of Pgp. First, the different IF conformations of Pgp with a range of degrees of opening allow accommodation of molecules with various sizes and shapes. Second, binding of different substrates to the central cavity involves common residues such as F983, F336, and L65, but the interactions are not substrate specific. This is reminiscent to ABCG2, another major human ABC multidrug transporter, which utilizes simple π-π interactions between two F439 residues and the ring structure in different substrates^35,36. Such non-specific substrate interactions in Pgp and ABCG2 are in sharp contrast with substrate-specific contacts in highly specific ABC transporters^37–40. Third, peripheral substrate site is a nonspecific, shallow hydrophobic pocket that appears to bind many compounds with relatively low affinity. Fourth, in either the peripheral or central cavity, no binding pose seems to be fully stabilized, suggesting that highly stable interactions are unfavorable for efficient transport, and forming weak, non-specific binding interactions is inherently promiscuous. This is evident by more efficient substrates such as verapamil (this study) or taxol¹⁰ having weaker and less defined EM densities, and the less efficiently transported substrates such as vinblastine (this study), vincristine²⁰, or Pgp inhibitors^18,20 showing more defined EM densities and better resolved structural features. It is further supported by MD simulations in which interactions with aromatic residues stabilize ligand poses and reduce conformational transitions²⁵.

Methods

Protein expression and purification

Human Pgp gene (UniProt: P08183) with C-terminal maltose binding protein (Pgp-MBP) was expressed in Expi293F cells (Thermo Fisher Scientific) in a stable cell line selected for high Pgp expression. Stable cell line was generated by plating cells in 95% Freestyle 293 Expression Medium (Gibco) and 5% fetal bovine serum at 37 °C on a 6-well plate. Transfection was performed on adherent cells by the addition of 1 μg of pFastBacMam-Pgp-MBP plasmid with 3 μg of linear polyethylenimine and incubating for 24 h. The next day, media was replaced with the addition of 2 ng/μL of puromycin to select for cells with high Pgp expression and genetic incorporation. This killed most of cells, with the remaining cells reestablishing the original cell density within 2–3 weeks from initial selection. This generated a polyclonal cell population and was immediately scaled up for large-scale protein expression, making it unlikely for mutations to arise or to become dominant. Nevertheless, RNA sequencing of selected cells was performed (described below) to confirm the correct Pgp sequence without any mutations. Cells were scaled up in suspension culture grown in Freestyle 293 Expression Medium (Gibco) and 2 ng/μL puromycin to cell density of 3–4 × 10⁶ cells per mL and harvested by centrifugation at 1000 × g for 10 min, then flash frozen in liquid nitrogen and stored at −80 °C.

For purification of Pgp-MBP, a cell pellet was resuspended in buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 20% glycerol, 0.5 mM tris(2-carboxyethyl)phosphine (TCEP), 2% dodecyl-β-D-maltopyranoside (DDM), 0.2% cholesteryl hemisuccinate (CHS), and protease inhibitors (5 μg/ml Aprotinin, 2 μg/ml Leupeptin, and 2 μM Pepstatin A) and dounce homogenized. This mixture was incubated for 2 h at 4 °C with nutating. Insoluble material was pelleted by ultracentrifugation at 40,000 × g for 1 h at 4 °C. The resulting supernatant was applied to 3 mL of amylose affinity resin and incubated by nutating for 30 min. Supernatant was removed by gravity flow, and resin was washed with buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 20% glycerol, 0.5 mM TCEP, 0.1% DDM, 0.02% CHS. Nanodisc reconstitution was performed with protein bound on-resin by adding Membrane Scaffold Protein 1D1 (MSP1D1) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG) at a 1:60 molar ratio to the 3 ml of resin to a final volume of 10 mL containing final concentrations of 50 mM Tris (pH 7.4), 150 mM NaCl, 30 mM sodium cholate, 50 μM MSP1D1, and 3 mM POPG. Following an incubation of 2 h at 4 °C, 6 g of biobeads were added and incubated overnight at 4 °C with nutating. The next day, the on-resin nanodisc reconstitution protein was washed with buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 20% glycerol, and 0.5 mM TCEP. Elution was performed with identical buffer as the wash buffer with the addition of 10 mM maltose. Resultant elution was concentrated to <1 mL and injected into Superose 6 gel filtration column with buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, and 0.1 mM TCEP. Fractions containing Pgp protein were concentrated to 3.5–5 mg/mL and used for preparing cryo-EM grids. The construct used for ATPase and cryo-EM experiments contained a C-MBP tag with 11-residue linker. This construct is fully functional, as shown by successful puromycin selection in Expi293F cells, high substrate-stimulated ATPase activity (Fig. 1a), no density in our cryo-EM maps for MBP suggesting high flexibility, and previous studies demonstrating that C-terminal tags do not affect Pgp function including large tags (>50 kDa) such as tandem mVenus-mCerulean¹².

Verification of Pgp mRNA sequence

Sequencing of Pgp mRNA was performed by extraction of total mRNA using the NEB Monarch Total RNA Miniprep Kit following the manufacturer’s instructions. Total RNA was converted to cDNA using NEB Protoscript II First Strand cDNA synthesis kit. Pgp cDNA was amplified using NEB Q5 DNA polymerase, and PCR product was subjected to DNA sequencing, which confirmed that there were no mutations present in the Pgp gene.

EM sample preparation and data collection

Negatively stained EM grids were prepared with glow-discharged copper EM grids with 1.5% (w/v) uranyl formate. These were imaged on a Tecnai T12 electron microscopy (FEI) operated at 120 kV. Cryo-EM grids were frozen on Mark IV Vitrobot (Thermo Fisher Scientific) blot with force of zero and blot time of 3 s. A 3 μL volume of sample was loaded onto Quantifoil Au R1.2/1.3 holy carbon grids. Samples under continuous hydrolysis conditions were generated by first prewarming two samples to 37 °C. The first is Pgp (20–30 μM) premixed with substrate (either 100 μM vinblastine or 100 μM verapamil). The second is a sample containing pre-mixed 47 mM ATP with 60 mM MgCl₂. Next, 0.6 μl of Mg-ATP was added to a pre-warmed (37 °C) centrifuge tube, and 3 μL of Pgp-substrate was added, mixed by pipetting, and 3 μL of this mix was immediately applied to cryo-EM grid, blotted, and plunge frozen. Final concentrations were 83.3 μM vinblastine or verapamil, 7.8 mM ATP, 10 mM MgCl₂. This process was performed such that the time between mixing Pgp-substrate with Mg-ATP and plunge freezing was within 10–15 s. The ADP-vanadate sample was generated by incubating Pgp protein with 100 μM vinblastine, 1.5 mM ATP, 1.8 mM MgCl₂, and 2 mM sodium orthovanadate for 10–20 min at room temperature before freezing grids. Cryo-EM images were collected on a Titan Krios electron microscope (Thermo Fisher Scientific).

Electron microscopy image processing

Negative-stain EM and cryo-EM images were initially selected using SamViewer and a semi-automated procedure implemented in Simplified Application Managing Utilities for EM Labs (SAMUEL)⁴¹. Two-dimensional (2D) classification was performed with “samclasscas.py,” “samtree2dv3.py,” or 2D classification in RELION-3.0⁴². The beam-induced motion of dose-fractionated movies was corrected using MotionCor2⁴³, and defocus values were calculated using CTFFIND4⁴⁴. The 3D classification and refinement were all performed using RELION-3.0⁴². Standard 3D classification starting with one initial model was not effective or reproducible in separating the particles into distinct Pgp conformations, possibly due to small protein size, multiple conformations with small variation, and similar 2D views from distinct conformations in certain orientations. After experimenting with various strategies, robust cryo-EM particle sorting was achieved using a “multiple models at multiple resolutions (M3R)” strategy, in which two cryo-EM maps of Pgp, with the NBDs well separated or tightly associated, were each low-pass filtered to three different resolutions (7, 15 and 30 Å) and used as six initial models for 3D classification. The two cryo-EM maps used as an initial reference were obtained from the Vinblastine-ADP-Vi dataset, as shown in Supplementary Fig. 2C. The classes with closed NBDs and those with open NBDs were then processed separately. All downstream 3D classifications used a single initial model from the previous iteration of 3D classification and were low-pass filtered to 40 Å during binned 3D classification (binned by 3 × 3 pixels) and 30 Å for unbinned 3D classification. Following multiple rounds of 3D classification, 3D refinement was performed using a mask around the entire molecule, including the nanodisc density, and included solvent_correct_fsc in the relion refinement command. For all steps described above, particles from the last 5 cycles within selected classes were included for the next iteration. Next, a mask was constructed around Pgp, excluding the nanodisc, and particles were subjected to no-alignment 3D classification. Classes were then selected and subjected to a local refinement to generate the final cryo-EM map. All refinements were independently refined two half datasets. The overall resolution was estimated based on gold-standard Fourier shell correlation (FSC) of 0.143. ResMap was used to estimate the local resolution using the two half data maps⁴⁵.

3D variability analysis

3D variability analysis (3DVA) was performed using cryoSPARC v4.0.3. The particle stack after the first round of global refinement (before no-alignment classification) was used (shown with asterisk in Supplementary Fig. 2, 8, 9, and 12), with a mask around Pgp excluding the lipid nanodisc. Three eigenvectors were calculated with a 5 Å low-pass filter, and 3DVA outputs were visualized with 3D variability display tool in cryoSPARC using 10 clusters with intermediate mode, which performs a separate 3D reconstruction on each cluster of particles. NBD distance calculations to determine the range of motion present in 3D variability analysis were done by rigid-body fitting the individual NBDs (from Ver-ATP-IF2 model) into each 3D reconstruction using the “fit to map” feature in UCSF Chimera.

Model building and refinement

To build the Pgp bound to 100 μM vinblastine with Mg-ATP, the Pgp bound to vincristine and MRK16 fab (PDB 7A69) was used as an initial model after removing vincristine and MRK16, and manually adjusting the degree of NBD separation. To build the ADP-vanadate model, the Pgp EQ mutant bound to ATP (PDB 6c0v) was used as the initial model. All refinements were performed using phenix real-space refine⁴⁶ and manually adjusted in COOT⁴⁷. All models were initially built with all side chains present. Next, each model was manually scanned to determine if enough density is present to allow for confident side chain placement using a generally aggressive strategy if connected density is present, even at low contour. Using this strategy, most side chains were removed for the NBDs in all models. Models built in the lower resolution cryo-EM maps (Apo, VerATP-OC, VerATP-IF2, VerATP-CC) also had most side chains removed in the TM region, while those in higher resolution maps (VinATP-IF1, VinATP-IF2, ADP-Vi) had most side chains modeled in the TM region. For the datasets with nucleotide present, the modeled species was based on the expected state. For the IF conformations, ATP was modeled, while for the ADP-Vi-CC state, ADP-Vi was modeled. The resolution in the NBD region was not high enough to unambiguously assign the nucleotide state.

ATPase activity assay

ATPase hydrolysis was measured by monitoring the release of inorganic phosphate. 1 μg of Pgp-MBP reconstituted in MSP1D1-POPG nanodiscs was used per assay. All assays were incubated for 30 min at 37 °C in 50 μL reaction volume containing 50 μM Tris (pH 7.4), 150 μM NaCl, 5 mM ATP, 6 mM MgCl₂. The reaction was stopped after 30 min reactions by addition of 50 μL of 12% SDS, and the color was developed by the addition of 100 μL of 2% ammonium molybdate and 12% ascorbic acid in 1 M HCl. The absorbance was measured at 850 nm using a microplate reader. The amount of hydrolyzed phosphate was determined by comparison to standard curve of potassium phosphate (KH₂PO₄).

In-cell transport assay

Pgp mutants were generated using NEB Q5 Site-Directed Mutagenesis Kit and cloned into pSB-bi plasmid for generation of a stable cell line using the Sleeping Beauty Transposon system⁴⁸. Cell culture of 8 × 10⁵ adherent Expi293F cells in a 2 mL volume was grown and transfected with 3 μg PEI, 0.95 μg of pSB-bi-Pgp-GFP mutant, and 50 ng of transposase (SB100x). This was incubated at 37 °C for 24 h in 95% Freestyle and 5% FBS. The next day, the media was replaced with the addition of 2 ng/μL puromycin was added to select for cells with pSB incorporation. Cells were scaled up to 20 mL suspension culture grown in Freestyle media with 2 ng/μL puromycin. Analysis of protein quality was monitored by fluorescence size exclusion chromatography and SDS-PAGE. For this analysis, 1 mL of cell culture containing ~2 × 10⁶ cells was solubilized by incubating in buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 20% glycerol, 0.5 mM TCEP, 2% DDM, and 0.2% CHS for 1 h at 4 °C. This was centrifuged, and the supernatant was used for FSEC or SDS-PAGE analysis. In addition, protein expression was measured using flow cytometry as described below.

In-cell transport assay was performed using Rhodamine 123 as substrate. 1 mL of cell culture containing ~2 × 10⁶ cells was harvested and centrifuged at 500 × g for 5 min. Cell pellet was washed twice with PBS, then resuspended in 500 μL freestyle media. Next, 1 μM Rhodamine 123, with or without 20 μM Zosuquidar, was added and incubated for 1 h at 37 °C with gentle shaking. After incubation, cells were washed with PBS twice and analyzed with Attune NxT Flow Cytometer (Thermo Fisher Scientific) with the Attune NxT Fluorescent Protein Filter Kit. Analysis of Pgp-GFP protein expression was performed in reaction with no Rhodamine 123 added and detection of GFP fluorescence using BL1 channel with the 510/10 BP filter. Detection of Rhodamine 123 was performed using the BL2 channel with 540/30 BP filter. The gating strategy is provided in Source Data.

Molecular dynamics

All models for molecular dynamics simulations were separated into two chains with TM1-6 and NBD1 in chain A and TM7-12 and NBD2 in chain B without the flexible linker between the two halves. The disordered region between TM1 and TM2 was modeled using Modeler. VinATP MD simulation was done with the IF1 model and without the peripheral vinblastine molecule. Both simulations with peripheral vinblastine present (with and without the central cavity vinblastine) were done with TM10 manually straightened to allow full motion of peripheral vinblastine to investigate if it stably binds the peripheral binding site. Straightening of TM10 was performed in coot using geometry restraints for alpha helices. Ver-ATP MD simulation was done with IF2 model. VinATP-IF1 and VerATP-IF2 were chosen because these showed the best-defined density for their respective ligands compared to the other IF state. These models were embedded in a membrane containing palmitoyloleoylphosphatidylcholine (POPC) and cholesterol bilayer at a molar ratio of 80:20, respectively, using the CHARM-GUI⁴⁹. The system was then solvated with TIP3 waters and neutralized with 0.15 M NaCl. The simulation cell for all was within 1 Å of 96 × 96 × 188 Å, and electrostatics were calculated with particle-mesh Ewald method. The MD simulations were performed with Gromacs 2021.2 program using CHARMM36 all-atom force field. The system was subjected to energy minimization and equilibration at 310 K and at constant pressure and temperature. The parameter files of the MD simulation were obtained from CHARM-GUI website, and analysis of the MD simulation was done in PyMol (The PyMOL Molecular Graphics System, Version 2.5.2, Schrödinger, LLC) and VMD⁵⁰. RMSD values were calculated against their initial position after alignment of the transmembrane region. Binding energy calculations for vinblastine and verapamil bound to Pgp were estimated using gmx_MMPBSA version 1.6.4⁵¹ using all frames for the calculation. We also performed a simulation of Pgp embedded in POPG nanodiscs for comparison to Pgp embedded in POPC-Cholesterol bilayer. This was performed with identical conditions except that the simulation cell was 169 × 169 × 169 Å. Input files for each simulation are provided in Supplementary Files.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.L. conceived the project. A.C. performed molecular cloning, protein expression, protein purification, nanodisc reconstitution, ATPase assays, transport assays, molecular dynamics simulations, negative-stain EM, sample vitrification, cryo-EM data collection and processing, and model building. A.C. and M.L. analyzed cryo-EM data and wrote the manuscript.

Peer review

Peer review information

Nature Communications thanks Ricardo Ferreira, Takeshi Murata, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The cryo-EM density maps generated in this study have been deposited in EMDataBank under accession codes 40226, 40340, 40341, 40294, 40295, 40342, 40343, 40292, 40259, 40258, 40227, 40326, 40293. The atomic coordinates have been deposited in the Protein Data Bank under accession codes 8GMG, 8SB9, 8SBA, 8SB7, 8SA1, 8SA0, 8GMJ, 8SB8. Accession codes for both cryo-EM maps and atomic coordinates are summarized in Supplementary Table 1. Source Data is provided with this paper as a Source Data file. Source data are provided with this paper.

Competing interests

All authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-58561-4.