Three-dimensional Reconstruction of the Saccharomyces cerevisiae Multidrug Resistance Protein Pdr5p*

Pdr5p, the major multidrug exporter inSaccharomyces cerevisiae, is a member of the ATP-binding cassette (ABC) superfamily. Pdr5p shares similar mechanisms of substrate recognition and transport with the human MDR1-Pgp, despite an inverted topology of transmembrane and ATP-binding domains. The hexahistidine-tagged Pdr5p multidrug transporter was highly overexpressed in yeast strains where other ABC genes have been deleted. After solubilization and purification, the 160-kDa recombinant Pdr5p has been reconstituted into a lipid bilayer. Controlled detergent removal from Pdr5p-lipid-detergent micelles allowed the production of peculiar square-shaped particles coexisting with liposomes and proteoliposomes. These particles having 11 nm in side were well suited for single particle analysis by electron microscopy. From such analysis, a computed volume has been determined at 25-Å resolution, giving insight into the structural organization of Pdr5p. Comparison with the reported structures of different bacterial ABC transporters was consistent with a dimeric organization of Pdr5p in the square particles. Each monomer was composed of three subregions corresponding to a membrane region of about 50 Å in height that joins two well separated protruding stalks of about 40 Å in height, ending each one with a cytoplasmic nucleotide-binding domain (NBD) lobe of about 50–60 Å in diameter. The three-dimensional reconstruction of Pdr5p revealed a close arrangement and a structural asymmetric organization of the two NBDs that appeared oriented perpendicularly within a monomer. The existence of different angular positions of the NBDs, with respect to the stalks, suggest rotational movements during the catalytic cycle.

Pdr5p, the major multidrug exporter in Saccharomyces cerevisiae, is a member of the ATP-binding cassette (ABC) superfamily. Pdr5p shares similar mechanisms of substrate recognition and transport with the human MDR1-Pgp, despite an inverted topology of transmembrane and ATP-binding domains. The hexahistidinetagged Pdr5p multidrug transporter was highly overexpressed in yeast strains where other ABC genes have been deleted. After solubilization and purification, the 160-kDa recombinant Pdr5p has been reconstituted into a lipid bilayer. Controlled detergent removal from Pdr5p-lipid-detergent micelles allowed the production of peculiar square-shaped particles coexisting with liposomes and proteoliposomes. These particles having 11 nm in side were well suited for single particle analysis by electron microscopy. From such analysis, a computed volume has been determined at 25-Å resolution, giving insight into the structural organization of Pdr5p. Comparison with the reported structures of different bacterial ABC transporters was consistent with a dimeric organization of Pdr5p in the square particles. Each monomer was composed of three subregions corresponding to a membrane region of about 50 Å in height that joins two well separated protruding stalks of about 40 Å in height, ending each one with a cytoplasmic nucleotidebinding domain (NBD) lobe of about 50 -60 Å in diameter. The three-dimensional reconstruction of Pdr5p revealed a close arrangement and a structural asymmetric organization of the two NBDs that appeared oriented perpendicularly within a monomer. The existence of different angular positions of the NBDs, with respect to the stalks, suggest rotational movements during the catalytic cycle.
Drug resistance is a crucial clinical problem in the treatment of human cancers and infection of bacterial or fungal origin (1)(2)(3)(4)(5). The most important resistance mechanism, ubiquitous from bacteria to man, which leads to multidrug resistance (MDR) 1 is the overexpression of membrane-associated transporters that extrude drugs out of the cell. The best characterized and the clinically most important MDR transporters are members of the ATP-binding cassette (ABC) superfamily such as the human P-glycoprotein (Pgp) and the MDR-associated proteins (6 -9).
The identification of yeast genes sharing homology with the mammalian drug resistance genes provided interesting possibilities for genetic and molecular manipulations (10,11). The yeast Saccharomyces cerevisiae genome project revealed the existence of 31 distinct genes encoding ABC proteins, several of which are implicated in multidrug resistance (5,12,13). In the yeast S. cerevisiae, a phenotype resembling the mammalian multidrug resistance phenotype and known as pleiotropic drug resistance (PDR), confers resistance to most currently available classes of clinically and agriculturally important fungicides and also to many antibiotics and herbicides (14 -19). The first and, by now, the best characterized yeast PDR transporter is the PDR 5 gene product. The protein Pdr5p has been shown to share nucleotide triphosphatases activities, as well as substrates and modulators, with the human MDR1-Pgp (13, 17, 18, 20 -23). The predicted topography of Pdr5p comprises two hydrophobic domains, each composed of six trans-membrane segments (TMS 6 ), and two cytoplasmic nucleotide-binding domains (NBD), corresponding to a named (TMS 6 -NBD) 2 "full transporter" (24,25). However, the disposition of the two hydrophobic domains and of the two hydrophilic NBDs of Pdr5p mirrors that of the major eucaryotic ABC proteins. Each half-Pdr5p starts with an NH 2 -terminal NBD followed by the first TMS 6 tract, whereas in Pgp, the TMS 6 tracts precede the nucleotide binding domains. Thus, despite similar mechanisms of substrate recognition and transport, the significance of such domain inversion in yeast ABC transporters is unknown. This is mainly related to the lack of structural information on yeast ABC transporters as compared with mammalian full-transporters (26 -30) or bacterial "half-transporters" consisting of one TMS 6 domain connected to one NBD and assembled as a (TMS 6 -NBD) 2 homodimer (31)(32)(33).
Here, we report the first three-dimensional reconstruction of an ABC transporter from S. cerevisiae. The overexpression of Pdr5p in the yeast pdr1-3 mutant allowed the production of high amounts of the transporter (12,16). The overexpressed His-tagged Pdr5p plasma membrane protein of 160 kDa was solubilized and purified through Ni-NTA chromatography. Controlled detergent removal from a lipid-detergent-Pdr5p micellar solution led to the production of square-shaped particles of about 11 nm in side that have been analyzed by electron microscopy and single particle techniques. A three-dimensional structure of Pdr5p has been computed at 25-Å resolution from negatively stained lipid-reconstituted samples. This volume showed a dimeric organization of Pdr5p, in which each monomer protruded out of the membrane through two stalks ending each one by a cytoplasmic NBD lobe. Comparison with the reported structures of bacterial ABC transporters (31,32) showed a common cone-shaped organization that leaves a large chamber between the two stalks bearing the NBDs. The present work also revealed that the two NBDs are spatially close and asymmetrically organized, appearing oriented perpendicularly in the Pdr5p monomer. Such an observation supports the hypothesis of a functional asymmetry of ABC transporters (34,35) and suggests catalytic mechanisms in which the NBDs can move around the major axe of the stalk.

EXPERIMENTAL PROCEDURES
Materials-Phospholipids of the highest purity were purchased from Avantis Polar Lipids, n-dodecyl-␤-D-maltopyranoside (DDM) was from Sigma and Bio-Beads SM2 (20 -50 mesh) were from Bio-Rad. Ni-NTA columns were from Qiagen. All other reagents were of analytical grade.
The PDR5-6HIS allele was constructed by PCR retrieval on the PDR3.3deltaK plasmid of 615 bp of the PDR5 gene spanning from the internal MluI site (ϩ3939/ATG) to the last codon before the STOP, using the primer sequences 5Ј-GAACGCGTCTGCAGCTGGCCAGT and 5Ј-CGCCAGGTTAACTTAGTGATGGTGATGGTGATGTTTCTTGGAG-AGTTTACCGTTCTTTTT, to which the restriction site AvrII, the STOP codon, and the His 6 -tag had been included. The MluI/AvrII PCR fragment was used for replacement of the original 807-bp MluI/AvrII fragment of PDR5. This replacement resulted in the loss of PDR5 terminator from position ϩ4531(STOP) to 4745 and yielded pAD-YE/ PDR5-6HIS in which the His 6 -tag substituted the last 3B4-terminal PDR5 codon. The 6.8-kb KpnI/BamHI fragment containing the PDR5-6HIS allele was cloned into pSK to give pAD-SK/PDR5-6HIS. The AD12345678 strain was transformed with the KpnI/SpeI-cleaved pAD-SK/PDR5-6HIS plasmid. Yeast transformants were selected onto 0.05 g/ml cycloheximide, and recombination in the yeast genome of the PDR5-6HIS allele was checked by PCR.
Protein Purification-The recombinant strains were grown overnight at 30°C in 5.8% glucose, 2% yeast extract, pH 5.2, until late exponential phase (300 million cells/ml). Enriched plasma membrane fraction (including the acidic precipitation of contaminating mitochondrial membranes) were prepared as described in Goffeau and Dufour (37) and stored before use at Ϫ70°C. Plasma membranes were solubilized by DDM (protein/detergent ratio ϭ 1.2 w/w) in a 10 mM Tris buffer, pH 7.5, during 15 min at room temperature under stirring. After centrifugation the solubilized membrane proteins were loaded onto a Ni-NTA resin column pre-equilibrated in 0.1% DDM, 10 mM Tris, pH 7.5. After 2-h incubation at 4°C for His-tagged protein binding, the column was washed extensively with 10 mM imidazole buffer, pH 7.5, 0.1% DDM. Purified protein was eluted with 250 mM imidazole buffer, pH 7.5, 0.1% DDM. The purified protein eluted at a concentration of about 0.5 mg of protein/ml, and samples were stored at Ϫ70°C. The protein content was assayed according to Bradford (38) using bovine serum albumin as a standard. Proteins were analyzed on 10% SDS-PAGE gels, according to the method of Laemmli (39), and the gels were stained with Coomassie Blue R-250 or silver nitrate. Proteins were transferred onto nitrocellulose membrane, and Pdr5-6HISp was detected by chemiluminescence using the INDIA HisProbe-HRP (Pierce). UTPase activity was measured at pH 6.3 as described in Decottignies et al. (36).
Protein Reconstitution-Purified His-Pdr5p was mixed with phos-pholipids in the same buffer used for protein purification. Different lipidic compositions, including L-␣-dimyristoylphosphatidylcholine, dioleoylphophatidylcholine, Escherichia coli lipids, and egg phosphatidylcholine with or without 10% negatively charged phospholipids were analyzed. The best results were obtained using L-␣-dimyristoylphosphatidylcholine at lipid to protein ratios ranging from 0.5 to 1 w/w. After 1-h incubation at 25°C, allowing micellar equilibration, detergent was removed by addition of 5 mg Bio-Beads per 50 l and aliquots taken as a function of time for electron microscopy analysis (32). Electron Microscopy-Structures produced during protein reconstitution experiments were analyzed by negative staining using 1% uranyl acetate, after sample adsorption onto glow-discharged 300-mesh carbon-coated grids. For three-dimensional reconstruction, untilted electron micrographs and tilted pairs (0 and 45°) were recorded onto a Philips CM120 electron microscope at an accelerating voltage of 120 kV and nominal magnification of ϫ35,000, using a low dose system and Ϫ1 m as the defocus value. All micrographs were recorded on a 1024 ϫ 1024 pixel Gatan ssCCD camera resulting in a pixel size of 5.78 Å/pixel.
Image Analysis-A total number of 2 ϫ 910 projections from tilted pairs and 767 from untilted images were windowed and centered by using X-MIPP software (40) and PSPC algorithm (41) before classification. Self-organizing mapping based on the Kohonen neural network (42) was performed over untilted images of tilted pairs. Multivariate statistical analysis (43, 44) of dictionary vectors was used to perform the map segmentation and to identify the homogeneous groups of projections. The average image of each group was computed and a rotational analysis was performed before three-dimensional reconstruction of each homogeneous class. Volume computing, merging, and refinement (45,46) were performed using SPIDER software (47). Volume symmetry was analyzed by searching the maximum correlation coefficient value between the refined volume and itself when rotated at different angles. Final volume was computed after symmetry applying, refinement, and filtering at the resolution estimated by the Fourier ring correlation method (48,49). Volume rendering was performed using ETDIPS v2 software (50).

RESULTS AND DISCUSSION
The plasma membrane preparation, from recombinant yeast cells, contains high amounts of Pdr5p similar to those of Pma1p, the major constitutive component in wild type yeast plasma membranes in which Pdr5p is undetectable. It was verified that the Pdr5-H6p-enriched plasma membranes contained a novel UTPase activity (about 0.8 mol⅐min Ϫ1 ⅐mg Ϫ1 ), which was previously shown to be a specific marker for Pdr5p (12). The solubilization of the membranes by DDM and the protein purification through a Ni-NTA column allowed us to obtain samples containing a major polypeptide of 160 kDa (Fig.  1A), which corresponds to the expected molecular weight of Pdr5-H6p, cleared out from the major Pma1p contaminant.
When analyzed by electron microscopy, the purified preparation only shows small and large (about 11 ϫ 11 nm) micellar particles with no aggregates observable. In addition, due to the presence of high detergent concentration, negative staining of these preparations produces noisy images, precluding detailed single particle analysis. Addition of phospholipids to the micellar purified protein, followed by the addition of Bio-Beads for detergent removal, led to a partial reconstitution of Pdr5p into proteoliposomes. Interestingly, after 1-h treatment with Bio-Beads, which decreases the detergent concentration to below its critical micellar concentration (data not shown; see Ref. 32), the proteins non-incorporated into proteoliposomes appear as clearly visible square particles having 11 nm in side and suitable for single particle analysis (Fig. 1B). The partial reconstitution of Pdr5p into proteoliposomes is dependent upon the lipid composition and the best results, in terms of the number of individual particles non-incorporated into proteoliposomes, have been obtained with dimyristoylphosphatidylcholine. Longer incubation with Bio-Beads to remove residual detergent leads to the stacking of the square-shaped particles into larger globular lipid-protein aggregates.
A gallery of representative images windowed from negatives is shown in Fig. 1C. Self-organizing mapping and multivariable statistical analysis on these images result in the identification of three major classes representing 91% of the total population. The first class (56% of images) is composed by 4-fold symmetrical projections presenting four circular stain excluding regions, of about 5 nm in diameter, arranged around a central stain-penetrating area ( Fig. 2A). The second class (20% of images) corresponds to 2-fold rectangular shape views of Pdr5p having about 11 nm ϫ 13 nm (Fig. 2B). Finally, the third class (15% of images) is a 4-fold cross pattern presenting four pearshaped stain-excluding regions, about 5 ϫ 4 nm, that join at the center of the particle (Fig. 2C). Comparison of the non-symmetrical volumes, computed from the corresponding tilted images belonging to each class, demonstrates that the three classes corresponded to different projections of the same object (Table  I). Thus, projections belonging to the second class (Fig. 2B) correspond to a volume perpendicularly oriented ( Ϸ 90°, Ϸ 0°) to that computed from the projections of the first class ( Fig. 2A). The volume from the third class correlates well with an orientation intermediate between those of the two other classes.
The existence of two perpendicular volumes allowed us to compute a merged reconstruction for refinement and symmetry analysis. The reconstruction obtained after two cycles of refinement from the merged volume presents a maximum correlation value (0.75), with itself when rotated Ϸ 180°, ϭ Ϸ 0°, demonstrating the existence of a 2-fold symmetry. Combination of the whole image set, including images belonging to the third class, allows us to compute a final 25-Å resolution threedimensional reconstruction after 2-fold symmetry imposition (Fig. 3A). This final volume reveals the presence of four 140-Å-long components, each one composed of three domains corresponding, from top to bottom, to a near globular domain (50-Å diameter and 60-Å height) compatible in dimensions with a NBD, a 40-Å stalk and, finally, a 50-Å height domain compatible in dimensions with a lipid bilayer. According to the predicted (TMS 6 -NBD) 2 structural organization of the Pdr5p monomer, it is concluded that the final reconstruction represents a dimer of this full ABC transporter with the four globular domains corresponding to the four NBD domains. The membrane domains provide contacts between the four hydrophobic components. As these contacts are not equivalent, the interpretation of a dimeric structure, in which the two dimers are separated by the region of lowest stain penetration, is favored (see Fig. 3C).
To assign in more detail each region of the Pdr5p molecule in the three-dimensional reconstruction obtained in negative staining, it has been compared with our recent cryo-electron microscopy three-dimensional reconstruction structure of the YvcC homodimer from Bacillus subtilis (32) and to the x-ray crystallographic data on the MsbA homodimer from E. coli (31). As depicted in Fig. 3, A-C, the dimensions and shape of the three-dimensional reconstruction of the Pdr5p dimer are similar to a dimer of YvcC homodimers, while the x-ray structure of MsbA matches well into the volume reconstruction of a Pdr5p monomer (see also Ref. 32 for a fit of MsbA into the volume reconstruction of YvcC). This allows us to conclude that these bacterial and Pdr5p have a common structure and that a monomer of the yeast ABC transporter protrudes out of the membrane through two well separated stalks to a height of about 4 nm, each one ending by a cytoplasmic lobe which correlates with a NBD domain. In addition, the three-dimensional reconstructions of Pdr5p and YvcC reveal a close arrangement of the NBDs, appearing slightly disconnected in Pdr5p due to the use of negative staining.
It is also interesting to compare our data with structural features reported from electron microscopy analysis of single molecules and different two-dimensional crystals of hamster and mouse Pgp (26,27,30). Single particle analysis, in negative staining, of these detergent-solubilized ABC transporters has been interpreted assuming a monomeric state of these ABC transporters. However, all reported structures approximate a cylinder of about 100 Å in diameter, which in fact corresponds to the size of a Pdr5p dimer or to the size of a dimer of an YvcC homodimer. On the other hand, the size and shape of Pgp monomers, deduced from projection maps of two-dimensional crystals at 25-Å resolution, varied with the origin of Pgp, the method used for the production of two-dimensional crystals as well as with the method of specimen embedding (28,30). Interestingly, the projection structure of the mouse Pgp was compared with the projection structure deduced from the X-Ray model of MsbA in the "open" conformation, with well separated nucleotide-binding domains as they appear in the three-dimensional crystal (31), or in the "closed" conformation, with the nucleotide-binding domains brought into contact by modifying the MsbA co-ordinate file (30). crystallized in the lipid bilayer in a closed conformation with the two NBDs and the two stalks in close contact. In contrast, our data on Pdr5p, as well as on YvcC, demonstrate that the NBDs can be in contact, or only slightly disconnected, while the stalks are spatially separated. This infers that the two NBDs can be in contact while preserving an open V-shaped structure of the transporter without the need for large movements of the trans-membrane segments to bring the two stalks in close contact. In this context, it should be stressed that the model proposed by Lee et al. (30) does not take into account that (i) in the reported MsbA open structure, the disordered NBDs are not resolved accurately and are expected to be much closer than 50 Å, and (ii) a projection structure in negative staining does not give any information about the organization of the transmembrane domains. Accordingly, it was reported that, in frozen-hydrated two-dimensional crystals, the hamster Pgp monomer had an elliptical shape, 91 ϫ 60 Å, larger than the dimensions, 70 ϫ 70 Å, observed in negatively stained twodimensional crystals (28) and more compatible with the dimensions of MsbA in the open conformation.
Another important feature of the Pdr5p structure is that the orientations of the four NBD lobes in the Pdr5p dimer reconstruction are not the same. As shown by the arrows in Fig. 3A, two contiguous NBD lobes appear to be oriented perpendicularly. This observation correlates with the clear asymmetry of the NBDs observed in the YvcC homodimer (Fig. 3B). Functional asymmetry has been proposed for the NBDs of other transporters including Pgp (51,52), CFTR (34), and Ste6 (53). Noteworthy, a transport model of a two-cylinder engine has been proposed (35). In such mechanistic model, a (TMS 6 -NBD) 2 structure, corresponding to a full transporter such as Pdr5 or to an homodimer of a half-transporter such as YvcC, would work co-operatively by using each (TMS 6 -NBD) alternatively to couple drug transport to ATP hydrolysis. The ATP-bound state is FIG. 2. Average images corresponding to the three major classes observed after classification. A, average image of the major class, showing a 4-fold symmetry and a central stain-penetrating region. Images corresponding to this class were used to compute the subvolume 1. B, average image of a second class having a 2-fold symmetry. Images corresponding to this class were used to compute the subvolume 2. C, average image computed from the projections of the third class that has been used to compute the third subvolume. As in the first class, a 4-fold symmetry was detected. Scale bar ϭ 5 nm.

TABLE I Euler angles , , and and cross-correlation coefficients (CCC) computed between the different independent volumes
from each major class of Pdr5 projections Volume 1, computed from the major class, has been selected as reference. Maximum cross-correlation coefficient and its associated Euler angles between volumes 1 and 2 demonstrated that they correspond to reconstructions computed from projections of the same object oriented in perpendicular directions. Euler angles observed for cross-correlation between volumes 1 and 3 show that the last one is slightly tilted in relation to the first one.  3. Comparison of Pdr5p volume with YvcC reconstruction and x-ray crystallographic data of MsbA. A, three-dimensional reconstruction of a Pdr5p dimer at 25-Å resolution in negative staining. Three regions are clearly differentiated: the first region corresponding to the lowest part of the volume has been attributed to the membrane embedded domains; the second region corresponds to four protruding segments that have been attributed to the stalks domains; the third region consists of four lobes that have been attributed to the NBDs. Arrows show the different orientations of the NBDs. B, three-dimensional reconstruction of a dimer of YvcC, as determined previously by cryo-electron microscopy (31). The three regions detected in the Pdr5p reconstruction are also visible in this volume being assigned to the membrane, stalk, and NBD regions (32). The different orientations of the NBDs are shown by arrows. C, docking of MsbA structure (31) in a Pdr5p monomer. thought to be associated with a high affinity drug-binding site, while the ADP-bound state would be associated with a low affinity site and the ADP-P i -bound state with an occluded drugbinding site. Moreover, differential interaction of nucleotides at the two NBD transporters has been recently described in the cystic fibrosis trans-membrane conductance regulator (54). According to the two-cylinder engine model, the two NBDs are never equivalent and appear asymmetric in the snapshot structure. In this framework, the clear asymmetry and the different orientations of the NBDs observed in the three-dimensional reconstruction of Pdr5 and YvcC suggest mechanisms in which the NBDs can move around the major axes of the stalks during the catalytic cycle. Such conformational changes could be coupled to a rotation of the trans-membrane segments as suggested from hydrogen/deuterium exchange kinetics and limited proteolysis on MDR-associated protein (MRP)-1 (55). Such a hypothetical mechanism would not necessarily imply large movements of the trans-membrane segments going from an open to a closed state bringing the two stalks in close contact.
A last point that deserves some comments is that Pdr5p tends to dimerize upon membrane reconstitution by detergent removal, as already observed for YvcC homodimers (Ref. 32; Fig. 3B). The appearance of dimers may simply reflect favorable protein-protein interactions during reconstitution rather than reflecting a native dimer. However, the possibility of an interaction between the first cytolosic loops (CL1) of two neighboring Pdr5p molecules has been reported recently and could possibly have implications for dimerization of this ABC transporter (56). In addition, a dimeric organization of different ABC transporters in native membranes has been proposed based on radiation inactivation and cross-linking experiments, as well as on the use of fusion proteins or monoclonal antibodies (57)(58)(59)(60)(61).
In summary, the present study reports the first three-dimensional reconstruction, at 25-Å resolution, of Pdr5p, an ABC transporter from S. cerevisiae. The dimensions and shape of the full-transporter Pdr5p compare well with the three-dimensional reconstruction of the B. subtilis half-transporter YvcC (32) and with the x-ray data on the E. coli half-transporter MsbA (31). This suggests a common architecture of these ABC transporters, which could be extended to other MDRs such as the mammalian P-glycoproteins. The most important features of our three-dimensional reconstruction are 1) the close arrangement of the two NBDs while the stalks are spatially separated and 2) the different orientations of the two NBDs. Our data suggest mechanisms involving significant rotational movements of the NBDs during the catalytic cycle. Threedimensional reconstructions in the presence of different substrates and/or inhibitors may provide more detailed understanding of the conformational changes associated with the pumping mechanisms of ABC transporters.