Molecular Modeling Correctly Predicts the Functional Importance of Phe594 in Transmembrane Helix 11 of the Multidrug Resistance Protein, MRP1 (ABCC1)*

The human ATP-binding cassette (ABC) transporter, multidrug resistance protein 1 (MRP1/ABCC1), confers resistance to a broad range of anti-cancer agents and transports a variety of organic anions. At present, essentially no structural data exists for MRP1 that might be used to elucidate its mechanism of transport. Consequently, we have applied a modeling strategy incorporating crystal and indirect structural data from other ABC transporters to construct a model of the transmembrane domains of the core region of MRP1 that includes the amino acid side chains. Three conserved Trp residues and one non-conserved Tyr residue, shown previously to be of functional importance (Koike, K., Oleschuk, C. J., Haimeur, A., Olsen, S. L., Deeley, R. G., and Cole, S. P. C. (2002) J. Biol. Chem. 277, 49495-49503), were found to line the “pore” in our model proximal to the membrane cytosol interface. A fifth aromatic residue (Phe594) was identified that, with the Trp and Tyr residues, completed a ring or “basket” of aromatic amino acids and, accordingly, we postulated that it would also be of functional importance. To test this idea, MRP1-Phe594 mutants were expressed in human embryonic kidney cells, and their properties were examined using membrane vesicles. Substitution of Phe594 with Ala substantially reduced or eliminated the transport of five organic anion substrates by MRP1 and abrogated the binding of leukotriene C4. On the other hand, the conservatively substituted F594W and F594Y mutants remained transport competent, although significant substrate- and substitution-specific changes were observed. These studies provide some structural insight into a possible substrate binding/transport site of MRP1 at the beginning of a putative substrate translocation pathway and demonstrate the usefulness of modeling for directing structure-function analyses of this transporter.

Chemotherapeutic approaches in cancer treatment are often hindered by the fact that tumor cells can become simultaneously resistant to a number of different anti-cancer agents. The multidrug resistance phenotype can often be attributed to the increased expression of one or more membrane proteins belonging to the ATP-binding cassette (ABC) 1 transporter superfamily, which can actively extrude anti-cancer drugs from the cell. Two of the best characterized examples of these transporters are the 170-kDa P-glycoprotein and the 190-kDa MRP1, encoded by the human ABCB1 and ABCC1 genes, respectively (1,2). Despite its shared ability with P-glycoprotein to confer resistance to a broad spectrum of anticancer agents, MRP1 is distinct from P-glycoprotein in that it can also transport organic anions, many of which are conjugated to glutathione, glucuronate, or sulfate (1).
The 1531-amino acid human MRP1 contains two NBDs and up to 17 TM helices distributed within three TMDs. This contrasts with P-glycoprotein, which contains just 12 TM helices in two TMDs. However, it has been shown that an NH 2 -terminally truncated "core" form of MRP1 in which the extra fivehelix TMD has been removed retains its ability to transport organic anions (3).
To understand the MRP1 transport mechanism, high-resolution structures of the protein at each phase of the transport cycle are required. However, given that MRP1 is a large membrane protein with both TM and cytosolic domains, the difficulties associated with crystallizing even one intermediate in the transport cycle are substantial. Although efforts continue to overcome the challenges associated with high level expression, purification, re-folding, and crystallization of MRP1, it is possible to model this transporter based on homologous structures.
To date, three full-length ABC transporter structures have been solved that might be used to model MRP1, although none of these contain the additional NH 2 -terminal TMD found in MRP1 and related proteins. Two of these are from Escherichia coli and one is from Vibrio cholera (4 -6). The first structure is the lipid A transporter MsbA from E. coli (EC-MsbA) (4). However, only the C-␣ coordinates at 4.5-Å resolution were released for this structure and, additionally, the coordinates for more than 100 residues were entirely absent from the published Protein Data Bank file. Furthermore, the validity of the EC-MsbA crystal as an in vivo representation of the molecule remains controversial, because there is little biochemical or structural evidence that supports the EC-MsbA crystal dimer. The second ABC structure solved was that of BtuCD, an E. coli vitamin B 12 transporter (5). This structure is of higher resolution than the EC-MsbA structure, and the majority of the atoms in the protein were resolved. Furthermore, the protein appears to have been crystallized in a physiologically plausible dimer. However, BtuCD is not suitable for a MRP1 modeling study because it is not homologous to any mammalian ABC proteins of medical interest. Most recently, a second MsbA structure from V. cholera has been released (VC-MsbA) (6). The overall topology of VC-MsbA is different from that of EC-MsbA and more consistent with available non-crystallographic experimental evidence. Indeed, using existing biochemical and biophysical data, we previously "derived" a likely conformation of P-glycoprotein that is comparable with the overall topology of VC-MsbA (7).
In the present study we have modeled MRP1 using existing structural data of other ABC proteins and then designed experiments to assess the validity of our models. In previous studies, we showed that Ala substitution of the Trp residues at positions 445, 553, 1198, and 1246, which are highly conserved among members of the "C" family of human ABC transporters, substantially reduced or eliminated the transport activity of MRP1 (8,9). Moreover, Phe substitution of a non-conserved Tyr at position 1243 also resulted in decreased conjugated organic anion transport activity and drug resistance (10). Our models of the 12 TM helices that form the two core TMDs of MRP1 both predict Trp 553 (TM10), Trp 1198 (TM16), Tyr 1243 (TM17), and Trp 1246 (TM17) to be located proximal to the membrane-cytosol interface and the Trp and Tyr side chains to be directed in toward a chamber through which substrates are presumed to be translocated. A fifth aromatic residue, Phe 594 , is located in the inner leaflet of TM11 and also faces into the chamber. Based on its position close to the four aforementioned polar aromatic residues, we hypothesized that it too would be important for MRP1 transport activity. To test this hypothesis, we used site-directed mutagenesis to replace MRP1-Phe 594 with Ala, Trp, and Tyr and examined the transport and organic anion binding properties of these mutants. Our results demonstrate that Phe 594 , like Trp 553 , Trp 1198 , Tyr 1243 , and Trp 1246 , is important for the transport activity and substrate specificity of MRP1 and support the existence of an "aromatic basket" situated at the beginning of a substrate translocation pathway.

EXPERIMENTAL PROCEDURES
Materials- [6,   . Similarly, conserved TMDs homologous to EC-MsbA were also identified for human P-glycoprotein (gi 2506118) (ABCB1) and VC-MsbA (gi 15641880). The NH 2 -and COOH-terminal core TMD sequences from each of these proteins were extracted and aligned separately using CLUSTAL W (12). In regions of low local sequence alignment, e.g. helices 6 and 1 from MRP1 and P-glycoprotein, respectively, the alignment was optimized using the secondary structure prediction algorithms DAS (13), Tmpred (14), and HMMTOP (15). This gave an alignment for MRP1 with P-glycoprotein and VC-MsbA (see Table S1 in the supplemental data located in the on-line version of this article). To construct the model of the MRP1 TMDs based on the TMDs of EC-MsbA in a closed conformation, the alignment file of MRP1 to P-glycoprotein was read into MODELLER (16) using our previously constructed model of P-glycoprotein as the template (7). To construct the model of the MRP1 TMDs based on VC-MsbA TMDs, the alignment file of MRP1 to VC-MsbA was read into MODELLER (16) using the structure of VC-MsbA (Protein Data Bank code 1PF4; Ref. 6) as the template. Thus the rotamers of the side chains of the EC-MsbA/P-glycoprotein-based MRP1 model are largely derived from rotamer libraries and energy minimization, whereas those of the VC-MsbA-based MRP1 model are derived from a crystal structure. Five models from a hundred were selected based on potential energy values calculated by MODELLER and were then minimized via ϳ50 iterations of steepest descents energy minimization in the GROMACS 3.0 molecular dynamics package (17). These five models were then evaluated for stereochemical quality using PROCHECK (18) and WHAT-IF (19), and the highest scoring model was used in the modeling study.
Vector Construction and Site-directed Mutagenesis-Mutations in MRP1 were generated using the QuikChange™ site-directed mutagenesis kit (Stratagene). The template for mutagenesis of Phe 594 was prepared by cloning a 1.9-kb BamHI/SphI fragment (containing nucleotides 841-2701 encoding amino acids 281-900) from pcDNA3.1(Ϫ)-MRP1 K into pBluescriptSK(ϩ) (Stratagene) into which an SphI site had been introduced into the multiple cloning site (9). Phe 594 substitutions were generated in the pBluescriptSK(ϩ) plasmid above according to the manufacturer's instructions with the following mutagenic primers (substituted nucleotides are underlined): F594A, 5Ј-G TTC AAC ATC CTC CGG GCT CCC CTG AAC ATT CTC C-3Ј; F594W, 5Ј-C TTG TTC AAC ATC CTC CGC TGG CCC CTG AAC ATT CTC CCC-3Ј; and F594Y, 5Ј-G TTC AAC ATC CTC CGC TAT CCC CTG AAC ATT CTC C-3Ј. After confirming all mutations by sequencing or diagnostic restriction enzyme digests, the 1-kb BamHI/Bsu36I fragments were subcloned back into pcDNA3.1(Ϫ)-MRP1 K , and the fragments in the full-length constructs were sequenced once again.

Transfection of MRP1 Expression Vectors and Determination of MRP1 Levels in Membrane
Vesicles-Wild-type and mutant pcDNA3.1(Ϫ)-MRP1 K expression vectors were transfected into SV40transformed human embryonic kidney cells (HEK293T), and membrane vesicles were prepared as described previously (8,9). Empty vector and vector containing the wild-type cDNAs were included as controls in all experiments. Levels of wild-type and mutant MRP1 proteins were determined by immunoblotting of membrane vesicles essentially as described (20). MRP1 was detected with monoclonal antibody QCRL-1, and relative levels of MRP1 protein expression were estimated by densitometric analysis using a ChemiImager™ 4000 (Alpha Innotech, San Leandro, CA). To confirm equal protein loading of gels, the immunoblot was also stained with Amido Black.
MRP1-mediated Transport of Organic Anions by Membrane Vesicles-ATP-dependent transport of the 3 H-labeled MRP1 substrates LTC 4 , E 2 17␤G, and E 1 3SO 4 by the membrane vesicles was measured using a rapid filtration technique as described previously (8,21). All data were corrected for the amount of 3 H-labeled substrate that remained bound to the filter, which was usually Ͻ10% of the total radioactivity. Transport in the presence of AMP was subtracted from transport in the presence of ATP to determine ATP-dependent uptake. All transport assays were carried out in triplicate, and results are expressed as means Ϯ S.D.
Apigenin-stimulated [ 3 H]GSH uptake was also measured by rapid filtration with membrane vesicles (20 g protein) incubated at 37°C for 20 min in a 60-l reaction volume with 100 M [ 3 H]GSH (120 nCi per reaction) and apigenin added to 30 M (22). To minimize GSH catabolism by ␥-glutamyltranspeptidase during transport, membranes were preincubated in 0.5 mM acivicin for 10 min at 37°C prior to measuring [ 3 H]GSH uptake. [ 3 H]MTX uptake was also performed at 37°C for 20 min in a 50-l total reaction volume containing membrane vesicles (10 g protein) and 100 M MTX (250 nCi per reaction) (9).
Photolabeling of MRP1 with [ 3 H]LTC 4 -Membrane proteins were photolabeled with [ 3 H]LTC 4 essentially as described (21). Briefly, vesicles prepared from transfected cells (20 g protein in 40 l) were incubated with [ 3 H]LTC 4 (0.1 Ci; 200 nM) at room temperature for 30 min and then frozen in liquid nitrogen. Samples were then alternately irradiated at 302 nm for 1 min using a CL-1000 ultraviolet cross-linker (DiaMed, Mississauga, Ontario, Canada) and snap-frozen in liquid N 2 10 times. Radiolabeled proteins (20 g) were resolved by SDS-PAGE and processed for autoradiography. Relative levels of photolabeling were estimated by densitometry as before (9).

RESULTS
Modeling MRP1 Based on MsbA-Sequence similarities as identified by BLAST searches indicate that the core region of MRP1, composed of amino acids 204 -1531 (TM6 to TM17), is homologous with MsbA and, thus, by extension, P-glycoprotein. We have previously presented an alterative model of P-glyco-protein that attempts to reconcile the crystal data with the available biochemical evidence (7). The TMDs of this model are based on the EC-MsbA TMDs in a closed conformation as inferred from three-dimensional EM data (23), and the NBDs are based on the MJ0796 NBD dimer crystal (24). In this way, we were able to satisfy both the EM and experimental data currently available for P-glycoprotein. To determine whether the structure derived for P-glycoprotein might also be of use in designing experiments to test structure-function relationships of MRP1, we modeled the TMDs of MRP1 in the closed conformation (Fig. 1, A and B). Additionally, while this manuscript was in preparation the VC-MsbA structure was released (6), so we have also modeled MRP1 using the VC-MsbA crystal as a template (Fig. 1, C and D). The intra-TMD helix packing is the same in both models; however, they differ in that the TMDs of the VC-MsbA-based model remain slightly tilted in the splayed V conformation.
In our models of MRP1, several aromatic residues line the putative translocation pathway and are located proximal to the bilayer-cytosol interface (Fig. 1), where the substrate might enter the chamber. In earlier studies, we demonstrated that four of these residues (Trp 553 , Trp 1198 , Tyr 1243 , and Trp 1246 ) were important for MRP1 transport activity and substrate specificity (8 -10). Our models of MRP1 identified a fifth aromatic residue, Phe 594 , proximal to the Trp and Tyr residues that we hypothesized would also be important for MRP1 transport activity. Consequently, both non-conservative (Ala) and conservative (Tyr and Trp) substitutions of Phe 594 were generated by site-directed mutagenesis, and the properties of these MRP1 mutants were investigated.
LTC 4 Transport and Photo-labeling Is Eliminated by Ala Substitution of Phe 594 -As shown in Fig. 2A, all three mutants generated (F594A, F594W, and F594Y) were expressed at levels 60 -100% of those of wild-type MRP1 in transfected HEK cells. This indicates that none of the Phe 594 substitutions had a major effect on the biogenesis or stability of MRP1. LTC 4 up-take assays were performed with membrane vesicles prepared from the transfected cells to determine the relative levels of uptake of this substrate by the Phe 594 mutants (Fig. 2B). After correcting for differences in MRP1 protein expression levels, ATP-dependent LTC 4 uptake by the F594A mutant was reduced by more than 90%, whereas uptake by the F594W and F594Y mutants was comparable with that by wild-type MRP1.
To further investigate the loss of LTC 4 transport by the F594A mutant, protein-labeling experiments were carried out with this intrinsically photoactivatable arachidonic acid derivative. As shown in Fig. 2C, a radiolabeled 190  labeled membrane vesicles prepared from cells expressing wild-type MRP1 as expected. In contrast, no such band is detectable in photolabeled vesicles from cells expressing comparable levels of the F594A mutant. This indicates that this non-conservative substitution of Phe594 abrogates photolabeling by LTC4 and, hence, binding of this substrate to MRP1, a finding that is consistent with the complete loss of LTC 4 transport activity by the F594A mutant. The conservatively substituted Phe 594 mutants F594W and F594Y could still be photolabeled by [ 3 H]LTC 4 , although photolabeling was reduced, by ϳ50% when corrected for differences in MRP1 protein expression. The reason for this decrease is not known at present, but it may be that the larger, more hydrophilic side chains of Tyr and Trp cause a moderate decrease in LTC 4 affinity.
Transport of Other Organic Anions Is Variously Affected by Substitutions of Phe 594 -To determine whether the Phe 594 substitutions affected the transport of other MRP1 substrates, vesicular uptake of E 2 17␤G, E 1 3SO 4 , MTX, and GSH by membrane vesicles enriched for the Phe 594 mutant proteins was determined (Fig. 3). The uptake levels by the Ala-substituted Phe 594 mutant were reduced by Ͼ50% for E 2 17␤G (Fig. 3A) and MTX (Fig. 3B) and by Ͼ90% for E 1 3SO 4 (Fig. 3C) and GSH (Fig. 3D).
The effects of the more conservative Trp and Tyr substitutions of Phe 594 on vesicular transport were more variable. For some substrates, uptake by the F594W mutant was up to 1.5-fold higher than that by wild-type MRP1 (GSH) and substantially higher than that by the F594Y mutant (E 1 3SO 4 and GSH). For other substrates the opposite effects were observed, i.e. vesicular uptake by the F594Y mutant was higher than that by wild-type MRP1 (E 2 17␤G and MTX) and even higher than that by the F594W mutant (E 2 17␤G and MTX). For example, E 2 17␤G uptake by F594W was ϳ30% of wild-type MRP1, whereas vesicular uptake of this glucuronide conjugate by F594Y was ϳ1.4-fold higher than that by wild-type MRP1 (Fig.  3A). In contrast, GSH uptake by the F594Y mutant was just 15% of that of wild-type MRP1, whereas uptake of this tripep-tide by the F594W mutant was ϳ1.6-fold higher than that of wild-type MRP1. Overall, substitution of an Ala residue in place of Phe 594 resulted in a general loss of MRP1 transport activity, whereas Tyr or Trp substitutions of Phe 594 tended to have a more moderate and selective effect on the transport activity and substrate specificity of MRP1.

DISCUSSION
Although a wealth of functional data exists for MRP1, the structural characterization of the protein has been relatively limited (25). We therefore generated models for the core TMDs of this ABC transporter that are each predicted to contain six TM helices. We first considered modeling this core region of MRP1 using the EC-MsbA dimer as has been done for Pglycoprotein (26). However, models of both P-glycoprotein and MRP1 built using EC-MsbA are problematic. EC-MsbA resembles an upside-down V, with the arms of the V splayed apart by as much as 40 Å. Models of MRP1 and P-glycoprotein based on EC-MsbA would suggest that the NBDs are separated by ϳ30 Å, precluding any possibility for inter-NBD communication, which is a well established requirement for the functionality of other ABC transporters, including P-glycoprotein (27,28). It has been suggested that a plausible representation of P-glycoprotein can be derived directly from the EC-MsbA crystal structure by simply closing the arms of the splayed V (29). However, this would bring the NBDs together in a conformation in which their "active transport" signature sequences are pointing away from the active sites and away from the opposite NBD. Four independent studies have concluded that the signature sequence of one NBD contacts the ATP active site of the other NBD (the "nucleotide sandwich") (5,24,30,31). Therefore, in order for the EC-MsbA crystal dimer to represent one snapshot of MRP1 and P-glycoprotein during their transport cycles, not only would its arms need to close and thus displace all of the lipid molecules that occupy the cleft, but each NBD would also have to undergo a rotation of ϳ120 o . Although not impossible, this seems unlikely. In addition, none of the four EM structures of P-glycoprotein to date, nor the single EM structure of MRP1, show the splayed V as being one of the conformations of these proteins (23,29,(32)(33)(34). Finally, the vast majority of biochemical cross-linking data reported for P-glycoprotein do not agree with a splayed V conformation (31,35,36). Thus, although it is possible that EC-MsbA represents a snapshot of an intermediate of the MRP1 and P-glycoprotein transport processes, in the absence of substantiating biochemical data we chose to analyze alternative models in this study.
We have presented two models of the core TMDs of MRP1 (TM6 to TM17) that include the amino acid side chains. One is based on our previous models of P-glycoprotein that were derived from the EC-MsbA TMDs in a closed configuration as discussed previously (7), and the other is based on the recently published structure of VC-MsbA (6) (Fig. 1). Both models are very similar and differ only in that the TMDs of the VC-MsbAbased model are slightly splayed apart from one another. However, the relatively small differences between the two models mean that we cannot exclude one or the other as possible snapshots for the MRP1 TMDs. The EM structures presently available for MRP1 are not yet of sufficient resolution to make meaningful comparisons (25), whereas EM structures for Pglycoprotein favor a model in which the TMDs are approximately perpendicular to the plane of the bilayer (23).
Having established the overall plausibility of our MRP1 core TMD models, we noted that several aromatic residues (Trp 553 , Trp 1198 , Trp 1246 , and Tyr 1243 ), which were demonstrated previously by site-directed mutagenesis to be functionally important for MRP1 activity and specificity (8 -10), lined the putative pore or substrate translocation pathway proximal to the cytosol-membrane bilayer interface. These residues, together with a Phe at position 594 in TM11, appeared to form an aromatic basket or "net." Consistent with our proposal that Phe 594 would be important for MRP1 transport function, the non-conservative Ala substitution of this aromatic residue abrogated transport and binding of LTC 4 by MRP1 and markedly reduced transport of both conjugated (E 2 17␤G and E 1 3SO 4 ), and unconjugated (MTX and GSH) organic anion substrates (Figs. 2 and 3). On the other hand, conservative substitutions of Phe 594 with either Tyr or Trp had little effect on LTC 4 transport but, in some cases, caused some significant changes in the transport of at least two of the four other MRP1 organic anion substrates tested. Thus, although the aromatic properties of the residue at position 594 are critical for retaining overall activity (i.e. F594 (wild-type), F594W, and F594Y are active but F594A is not), the addition of hydrogen bonding capacity to the amino acid side chain (as in F594W and F594Y) also influences substrate specificity, as shown previously for the polar aromatic residues at positions 553, 1198, 1243, and 1246.
The substrate-selective differences among the transport activities of the non-conservatively and conservatively substituted MRP1-Phe 594 mutants described here and the polar aromatic residues tested previously suggest that these residues may be interacting directly with substrate (8 -10). Statistically, Trp and Tyr residues are present most of the time at a lipid/ water environment interface. Thus, in our model the substrates might come in direct contact with one or more of the Trp 553 , Trp 1198 , Trp 1246 , and Tyr 1243 and Phe 594 residues that form the aromatic basket at the cytosolic beginning of a putative substrate translocation pathway. Alternatively, the basket might simply serve as a gate or size filter that prevents the passage of substrates through the chamber or translocation pathway until the necessary signaling processes (e.g. substrate recognition and/or ATP binding and/or hydrolysis) occur that allow transport. Rotamerization of these aromatic basket residues in such a way that their rings are perpendicular to the bilayer rather than being in the plane of the bilayer would open a large volume of space to the substrates. Evidence for such a mechanism has been described previously for the homotetrameric influenza virus M2 proton channel in which four Trp indole side chains at the cytoplasmic end of the pore, one from the TM helix of each of the four subunits, act in concert to form a gate that opens and closes the channel in response to changes in pH detected by His residues elsewhere in the protein (37).
In our study, no attempts were made to model the NBDs and the first TMD that precedes the core structure of MRP1 (TM 1-5). Based on sequence homology, it seems probable that the MRP1 NBDs share a nucleotide sandwich dimer arrangement and that there is cross-talk between the NBDs and TMDs as for other ABC proteins. However, experimental testing is required to verify this hypothesis, particularly given the relatively greater functional asymmetry of the NBD sequences of MRP1 (38,39). No prediction can be made for the first NH 2 -proximal TMD, as there is currently no structural data available for this region of MRP1 or any other related protein.
Finally, it is important to remain cognizant of the limitations of the modeling for the regions of MRP1 presented here. The amino acid sequence identity of the core MRP1 TMDs to both P-glycoprotein and MsbA is ϳ20%. Thus, alternative alignments to the one we have employed here are clearly possible. Although the global structure of the MRP1 TMD models are preserved in these alternative alignments (i.e. residues belonging to a certain helix remain a part of that helix), it is possible to shift residues by one to two amino acids in regions of low local sequence identity. The consequence of such a shift can move a residue from lining the pore to pointing away from the pore, or vice versa. We conclude that, in the absence of a high resolution crystal structure that is more homologous to MRP1, modeling the transporter protein according to a given alignment and testing the resulting structure experimentally represents a starting point for directing further structure-function analyses.