Impact of the Recent Mouse P-Glycoprotein Structure for Structure-Based Ligand Design

Abstract

P-Glycoprotein (P-gp), a transmembrane, ATP-dependent drug efflux transporter, has attracted considerable interest both with respect to its role in tumour cell multi-drug resistance and in absorption-distribution and elimination of drugs. Although known since more than 30 years, the understanding of the molecular basis of drug/transporter interaction is still limited, which is mainly due to the lack of structural information available. However, within the past decade X-ray structures of several bacterial homologues as well as very recently also of mouse P-gp have become available. Within this review we give an overview on the current status of structural information available and on its impact for structure-based drug design.

1. Introduction

More than 30 years ago a membrane protein could be linked to the phenomenon of decreased uptake of vincristine in multidrug resistant tumour cells.[1] The protein was termed P-Glycoprotein (P-gp, ABCB1), as it shows a glycosylation site and seems to affect permeability of the cell membranes (P therefore accounts for permeability). P-gp is an ATP-driven, membrane bound protein transporting a wide variety of structurally and functionally diverse drugs out of tumor cells.[2] Only a few years later verapamil was identified as being able to reverse tumour cell drug resistance by blocking P-gp.[3] For the first time, the complex phenomenon of multidrug resistance was linked to a distinct protein, which could be targeted by drugs.[4] The verapamil induced restoration of cytotoxicity was also observed for anthracyclines and subsequently extended to other calcium channel blockers, such as benzothiazepines and 1,4-dihydropyridines.[5] Other therapeutically used drugs which also showed P-gp inhibitory activity comprised phenothiazines, quinine, tamoxifen and cyclosporine A.[6] However, 30 years and numerous clinical studies later there is still no compound on the market and there are serious concerns about the druggability of this ATP-dependent, transmembrane transport protein.[7] Within the past decade considerable progress has been made in unravelling the physiological function of P-gp and the other 47 human ABC-transporters (ABC accounts for ATP Binding Cassette).[7] P-gp and some of its analogs are expressed in the intestine, liver, kidney, and the blood-brain barrier and there is an overwhelming amount of data that clearly demonstrate their multiple involvement in drug-uptake, -disposition and -elimination[8] rendering them antitargets rather than classical targets suited for drug therapy.[9] One of the classic examples is the quinidine-digoxin interaction, where the P-gp inhibitor quinidine increased the digoxin absorption rate by 30%, the peak plasma concentration by 81%, and the plasma AUC by 77%.[10] Another example is the brain accumulation of a range of drugs (which normally do not enter the brain) observed in mdr1a double knockout mice.[11] Undoubtedly, P-pg is an important impediment for the entry of hydrophobic drugs into the brain. Recently also the breast cancer resistance protein (ABCG2) has been reported as playing a role in the brain uptake of a variety of compounds.[12] This physiological function of P-gp and ABCG2 at the blood-brain barrier challenges the medicinal chemists in two ways: (i) compounds which should not enter the brain should be designed as P-gp substrates; (ii) CNS active compounds must pass the blood-brain barrier and therefore should be poor substrates of P-gp. With this respect P-gp might now again be considered as target, as designing-in/designing-out substrate properties becomes a major task for optimising the pharmacokinetics and tissue distribution of drug candidates.[13] This is also emphasized in a very recent review of the International Transporter Consortium, which published guidelines how to include all these information on transporters in the drug development process.[14] At the beginning of the new millennium, the first X-ray structure of a bacterial homologue of P-gp was published,[15] and since last year the structure of the first murine ABC-transporter – mouse P-gp – is available.[16] The latter, for the very first time, showed an ABC-transporter complexed with a ligand. After a short overview on ligand-based studies we will outline the impact of these structural genomics attempts on our understanding of drug/transporter interaction and on consequences for structure-based inhibitor design.

2. Ligand-Based Studies

2.1. Inhibitor Design

P-glycoprotein and its homologues belong to the large group of membrane-bound proteins, which lack considerable structural information. Thus, inhibitor-design had to rely on classical ligand-based approaches, such as quantitative structure-activity relationship (QSAR) studies and pharmacophore modelling. Especially verapamil analogues, triazines, acridonecarboxamides, phenothiazines, thioxanthenes, flavones, dihydropyridines, propafenones and cyclosporine derivatives have been extensively studied, and the results are summarized in several excellent reviews.[17, 18] Main results obtained from QSAR analyses clearly indicate the major importance of lipophilicity for high P-gp inhibitory activity. However, this might be due to the fact that the interactions of the compounds with the transporter most probably take place within the membrane bilayer rather than in the intracellular compartment. This gave also rise to the notation of P-gp as hydrophobic vacuum cleaner.[19] Interestingly, Wiese and co-authors could convincingly show that the hydrophobicity of the ligand should be treated as a space directed property rather than as an overall feature of the compounds.[20] For a series of ortho-, meta-, and para-substituted aryloxypropanolamines we could demonstrate that the hydrophobic moment of the compounds is a better predictor than calculated logP values.[21] Both studies point towards a distinct drug/protein interaction rather than an unspecific hydrophobic attachment to the transporter. Moreover, in the group of propafenone analogs a clear SAR could be deduced (Figure 1).[22] Briefly, highly active compounds should have a highly lipophilic, but small substituent at the nitrogen atom (best up to now is 4-xylylpiperazine), should be ortho-substituted at the central aromatic ring and should have electron donating substituents at the two aromatic rings, preferentially located in para-position.

Figure 1.

Summary of the results of structure–activity relationship studies on propafenone-type inhibitors of P-gp.

In addition to these extensive QSAR studies, also numerous pharmacophore models were published, both for substrates and for inhibitors.[23] Main features identified comprise hydrophobic, H-bond acceptor and positive ionisable. However, the models, although thoroughly validated and predictive in virtual screening runs, show only minor overlap in the spatial arrangement of the pharmacophoric features. This prompted several groups to postulate multiple binding sites at P-gp, which has been further evidenced by experimental data.[24] Finally, although consistent and statistically valid models were achieved, all these attempts could only minor contribute to the understanding of the function and the molecular basis of ligand-polyspecificity of P-gp and related transporters.

2.2. Prediction of Substrate Properties

As outlined above, P-gp is expressed in several organs, such as kidney, liver, intestine and also at the blood brain barrier (BBB). P-gp substrates therefore show poor oral absorption, enhanced renal and biliary excretion and usually do not enter the brain.[8, 25] Furthermore, they are likely to be affected by the MDR phenotype and are thus not suitable as anticancer agents. This prompted the development of in silico models for predicting P-gp substrate properties of compounds of interest.[26] Models developed relied both on simple rule-based classifications and on more advanced methods such as support vector machines and artificial neural networks. First rule-based methods came up more than 10 years ago when Seelig postulated that substrate recognition is particularly based on one of two specific hydrogen bonding patterns. Her analysis suggests that substrates contain either two hydrogen bonding features in a spatial separation of approximately 2.5 Å or three hydrogen bonding features with a spatial separation of the outer two features of approximately 4.6 Å.[27, 28] Later on, Didziapetris et al. formulated the “rule of four”, which states that compounds with the number of hydrogen bond acceptors in a molecule (N + O)≥8, and a molecular weight (MW)>400 Da and most acidic pK_a>4 are likely to be ABCB1 substrates whereas compounds with (N + O)≤4, MW<400, and most basic pK_a<8 probably are non-substrates.[29] Even simpler than the rule of four is the “Gombar-Polli Molecular E-state (MolES) Rule”, which states that molecules with MolES>110 seem to be substrates and those with MolES<48 seem to be non-substrates.[30] Finally, using only four simple ADMET descriptors (molecular weight, logP, positive ionizable and negative ionizable) Gleeson et al. could demonstrate that neutral or basic molecules showing a MW>400 and a logP value>4 are more likely to be transported by ABCB1 than acidic or zwitterionic compounds.[31]

One of the main problems of all these studies is the fact that data sets available are rather small and sometimes also inconsistent.[32] Few years ago the group of Gottesman published a comprehensive study where they correlated the cellular toxicity of 1400 selected compounds from the NCI60 screen with the mRNA levels of the 48 human ABC-transporter over the range of 60 human tumour cell lines.[33] An inverse correlation between transporter mRNA levels and compound toxicity indicates that a compound is a substrate for the respective transporter. Based both on this data set as well as on a set of 259 compounds compiled from the literature we used simple ADME-type descriptors (such as logP, number of rotatable bonds, number of H-bond donors and acceptors), van der Waals surface area descriptors[34] and 2D autocorrelation vectors as input matrix for several classification algorithms. When comparing binary QSAR and support vector machines, the latter gave more robust models with total accuracies in the range of 80%. Generally, the prediction of non-substrates performs better than those for substrates.[35] Very recently, based on a large data set provided by Boehringer Ingelheim, we developed a method for classifying rules (RuleFit) based on simple, interpretable physicochemical descriptors.[36] Interpretation of the best performing model indicates that P-gp substrates show a higher number of H-bond acceptors, more rotatable bonds and higher logP values than non-substrates. Although these features are quite general, the respective models showed a sensitivity of 81% and a specificity of 98% for an external test set. Thus, RuleFit modelling might be a versatile tool for establishing predictive and interpretable classification models also for other ABC-transporter.

3. ABC Transporter Structures

3.1. Topology of P-Glycoprotein

P-Glycoprotein is a pseudosymmetrical heterodimer where each monomer consists of a transmembrane (TM) as well as a nucleotide binding (NB) domain (Figure 2c). As the latter is responsible for the ATP-binding and hydrolysis it shows high sequence similarity throughout the ABC-transporter family. In contrast, the TM domains, which comprise 2×6 TM helices, are responsible for the respective substrate profile of the ABC-transporters and thus show only low sequence similarities among different transporters. The six helices of each TM domain (TMD) are connected by three extracellular and 2 intracellular loops. In addition, the intracellular loops comprise coupling helices which are responsible for the TMD-NB domain (NBD) interaction.

Figure 2.

ABCB1 architecture: a) wing-like helix-arrangement of P-gp on the basis of the ADP-bound 2HYD structure; b) helix-arrangement on the basis of the apo-structure 3G5U; c) visualization of the different domains of the N-terminal half of P-gp.

The principle of the topology described above is consistent throughout the whole family of human ABC-transporters. However, there are differences concerning the TMD and NBD arrangements. The ABCC transporter subfamily for instance possesses a third TMD at the N-terminus (TMD0) comprised of five helices, which is directly connected with TMD1.[37, 38] Members of the ABCG subfamily, half transporters that undergo homodimerization to gain full functionality,[39] show an inverse topology with the NBD at the N-terminus and the TMD at the C-terminal end.[40]

As could be shown by cryo-electron microscopy and biochemical experiments, where P-gp was trapped in different states of the catalytic cycle (using the non-hydrolysable ATP analog AMP-PNP and ADP-V_i), P-gp undergoes large conformational changes during the catalytic cycle.[41] The mechanism of the energy driven drug transport, rendering the high-affinity into a low-affinity binding site, is currently hypothesized in two different ways (extensively reviewed in[42]): The ATP switch model interprets the NBD dimerisation as the power stroke that is needed for altering the affinity for the substrates. Upon ATP binding the substrate is released and the subsequent hydrolysis of both ATP molecules results in the regeneration of the initial apo state, where another drug molecule can bind.[25] The second theory favours the sequential occlusion and hydrolysis of ATP molecules, where the occlusion of one ATP molecule is sufficient for the conformational change resulting in drug release (Figure 3).[43] Therefore, only 1 mol ATP/mol P-gp is sufficient for drug transport, whereas hydrolysis of in total two ATP molecules is needed to recover the transporter.[44–46]

Figure 3.

Occlusion-induced switch model.

3.2. Available X-Ray Structures

The availability of high resolution structures of targets is essential for understanding the molecular basis of their function and obviously also for performing structure-based design studies. While the entries in the protein data bank (PDB) are rising exponentially, the structure determination of membrane proteins is still problematic and only relatively few structures have been resolved up to now. Thus, the X-ray structures of E. coli MsbA (PDB code: 1JSQ, resolution: 4.5 Å), a lipid A transporter, raised a lot of interest in the ABC-transporter field.[15] However, even higher attention provided the retraction of these structures in 2006, which was due to an error in the data processing.[47]

3.2.1. Sav1866 and Domain Swapping

In 2006 Dawson et al. published the X-ray structure of the multi-drug transporter Sav1866 of Staphylococcus aureus in complex with ADP[48] (PDB code: 2HYD, resolution: 3.00 Å).[49] In contrast to P-gp, Sav1866 is a half-transporter. As already mentioned with the ABCG subfamily such transporters consist of two identical monomers which have to homodimerize to yield a functional transporter unit, comprising two TMDs and two NBDs.

This bacterial ABC transporter brought up an interesting arrangement of the transmembrane helices which has not been expected and which finally gave rise to the retraction of the MsbA structures.[47] Instead of having the TMDs on separated halves, building the sides of the large inner cavity (as originally proposed by the retracted MsbA structures), two TM helices (TMH) of one TMD cross the portal thus interacting with the opposite NBD (Figure 2). The parts of the protein that consist of TMH1 and TMH2 from one monomer and TMH3-TMH6 from the other represent a wing like structure (Figure 2a).

The close proximity of the NBDs in the Sav1866 structure was consistent with the electron microscope structure of ABCB1 previously published by Lee et al.,[50] which is also in agreement with the “nucleotide-sandwich dimer” in MJ0796.[51] In addition cross-linking experiments with P-gp showed that TMH 5 and 8 as well as 2 and 11 are located closely together.[52–54] Other cross-linking experiments suggested that the area enclosed by both TMDs of P-glycoprotein is funnel-shaped, with a wide opening at the extracellular side.[55] All this data are consistent with the SAV1866 structure.[56] Half a year later the same group crystallized Sav1866 in complex with the non-hydrolysable ATP-analog AMP.PNP (PDB code: 2ONJ, resolution: 3.40 Å)[57] in a slightly worse resolution. The comparison with the ADP bound state (2HYD) showed no significant differences among these structures.[57] This suggests that 2HYD most likely also resembles the energized, ATP-bound state. However when comparing the nucleotide bound X-ray structures that are open to the extracellular space, with the catalytic cycle depicted in Figure 3, it can be noticed that these states show low affinity for substrates and thus represent a state where the major conformational change already has occurred. The high affinity state therefore seems to be represented by the apo, inverted V-shape like state.[58, 59]

3.2.2. MsbA – the Corrected Structures

Ward et al. fulfilled the need of the nucleotide-free ABC transporter structure and published four different X-ray structures of MsbA in 2007,[60] two nucleotide bound and two in the absence of a nucleotide. One of the two apo structures was captured in a cytoplasmic-facing open state (PDB code: 3B5W, E. coli, resolution: 5.30 Å), with the two NBDs located ~50 Å apart from each other. The other apo structure also represented a cytoplasmic-facing, but closed conformation of MsbA (PDB code: 3B5X, V. cholerae, resolution: 5.50 Å). As mentioned before, electron microscopy experiments suggested that the NBDs of P-glycoprotein are located close together even in a nucleotide-free state.[50] Moreover, when considering the large hydrophobic pocket of the apo-structure, it might be questioned whether it would be filled with lipid molecules or with water. The closure of this pocket and the associated displacement of these molecules anyway seems rather unlikely.[52] Therefore the MsbA-apo-closed conformation is better qualified to be taken as a homology modelling template for P-gp than the open conformation. On the other hand, this structure was resolved at a resolution of 5.50 Å only representing the Cα-trace, which renders the modelling process quite difficult.

The nucleotide-bound MsbA structures represented complexes with AMP.PNP (PDB code: 3B5Y, S. typhimurium, resolution: 4.50 Å) and ADP·V_i (PDB code: 3B5Z, S. typhimurium, resolution: 4.20 Å). At this resolution the complexes are identical, showing a RMSD of < 0.65 Å between the Cα positions. Although the resolution of these MsbA X-ray structures are rather low and thus insufficient for a detailed investigation of drug-transporter interactions, they provided substantial insight into the catalytic cycle of ABC transporters and further confirmed the domain swapping topology suggested by the Sav1866 structures.

3.2.3. Mouse P-Glycoprotein

The first mammalian X-ray structure of an ABC transporter was published last year by Aller et al.[16] The publication comprised three nucleotide-free structures of murine P-glycoprotein, without a ligand (PDB code: 3G5U, resolution: 3.80 Å) and in complex with two enantiomeric cyclic peptide P-gp inhibors (CPPI, RRR- and SSS-QZ59; PDB codes: 3G60/3G61, resolution: 4.40 Å/4.35 Å).

Also in this case, the two transmembrane halves, where each of it consists of TMH 1–3 and TMH 6 of one monomer and TMH 4–5 from the other monomer, form an inverted V-shape structure (Figure 2b). For the very first time also co-crystals with inhibitors were available and provided new insights into possible binding areas. The interactions between the protein and the QZ59 isomers have recently been reviewed by Gutmann et al.[61] Even though both enantiomers showed distinct binding regions, the interacting TM helices were almost the same. Considering all residues within 4 Å of the bound CPPIs, both stereoisomers showed contacts with TMH5, TMH6, TMH7 and TMH12. While the RRR-isomer also interacted with TMH11, the SSS-enantiomer showed interactions with TMH 1.

When superposing the Cα positions of the three crystal structures (3G5U: apo, 3G60: co-crystal with QZ59-RRR, 3G61: co-crystal with QZ59-SSS) the RMSD calculated amounts to 0.56 Å, with 0.61 Å difference between the apo and the complexed structure. This information suggests that the binding of these large inhibitors hardly affects the protein structure, with at least almost no backbone movement involved. However, the behaviour of the side chains remains unclear. Therefore, we expanded the superposition by all atoms. The calculated RMSD in this case was raised to a value of 0.76 Å, which suggests that the positions of the protein atoms are highly similar among these structures. This fact is surprising as the polyspecificity of P-gp was always connected with its flexibility, Loo et al. even proposed that the protein binds its ligands via an induced fit mechanism.[62] This assumption should have resulted in a higher RMSD between the crystal structures of the complexes in comparison to the apo structure.

Regarding the amino acid residues that line the drug binding pocket, only one residue (Phe974) showed an average RMSD of > 2 Å between the three crystal structures. This Phe974 is part of TMH12 and directly extends into the binding cavity (Figure 4a). As can be seen in Figure 4b the different rotations of this residue are essential for the different binding of the CPPIs. The pose of QZ59-RRR (dark green) would not be possible with the Phe974 rotamers of the protein structures of 3G61 or 3G5U. In case of the isomer QZ59-SSS, two molecules can be seen in the structure.[16] While for the lower positioned molecule the rotation of Phe974 should have no influence, the upper one would be clearly inhibited by the 3G5U and 3G60 rotamers.

Figure 4.

a) Different rotamers of F974 of the different murine X-ray structures. 3G5U (blue), 3G60 (light green), 3G61 (dark green); b) together with co-crystallized QZ59 isomers. QZ59-RRR (light green), QZ50-SSS (dark green).

The murine P-gp structure shares highest similarity with the closed-apo MsbA structure (PDB code: 3B5X) and therefore is also consistent with electron microscope investigations mentioned before.[50] However, the MsbA structure is more opened on the top of the protein, whereas the NBDs lie closer together than with P-gp.

4. Homology Models for Structure-Based Design

Before the publication of the mouse P-gp structure last year the structural investigations on human P-glycoprotein heavily relied on homology models based on bacterial transporters. As Kerr et al. recently stated in a review,[63] the high resolution structures of ABC transporters and therefore the templates for homology modelling can be divided into two phases, the pre-Sav1866 (2001–2006) and the post-Sav1866 (2006 – present) phase. The same scheme can be applied with homology models, since some of them unfortunately relied on the retracted MsbA structure (Table 1). Nevertheless, since the beginning of the post-Sav1866 phase a considerable number of new homology models has been published.

Table 1. List of templates that were used for homology models of P-glycoprotein.

Template	Organsim	Sequence Identitiy/Similarity[a]	Co-crystal[b]	PDB Code	Resolution [Å]	Reference	Homology Models
MsbA	E. coli	36%/57%	Apo-open[c]	1JSQ	4.50	[15] Retracted	[52, 64, 65]
MsbA	V. cholerae	33%/55%	Apo-closed[d]	1PF4	3.80	[66] Retracted	[67]
MsbA	S. typhimurium	37%/57%	ADP·Vi	1Z2R	4.20	[68] Retracted
Sav1866	S. aureus	34%/52%	ADP	2HYD	3.00	[48]	[69–74]
MsbA	E. coli	36%/57%	Apo-open	3B5W	5.30	[60]	[75]
MsbA	V. cholerae	33%/55%	Apo-closed	3B5X	5.50	[60]	[72]
MsbA	S. typhimurium	37%/57%	AMP-PNP	3B5Y	4.50	[60]
MsbA	S. typhimurium	37%/57%	ADP·Vi	3B5Z	4.20	[60]
MsbA	S. typhimurium	37%/57%	AMP-PNP	3B60	3.70	[60]	[72]
MalK	E. coli	31%/50%	Apo-semi open	1Q1B	2.80	[73]	[73]
MalK	E. coli	31%/50%	Apo-open	1Q1E	2.90	[73]	[73]
ABCB1	M. musculus	87%/93%	Apo-closed	3G5U	3.80	[16]	[76, 77]
ABCB1	M. musculus	87%/93%	QZ59-RRR	3G60	4.40	[16]
ABCB1	M. musculus	87%/93%	QZ59-SSS	3G61	4.35	[16]	[78]

^[a]Sequence Identity/Similarity with human P-glycoprotein[79].

^[b]Co-crystallized molecules (Nucleotides, drugs or apo).

^[c]Apo-open describes the nucleotide-free protein with the NBDs far apart.

^[d]Apo-closed describes the nucleotide-free protein with NBDs that lie close together

Table 1 gives an overview on homology models of P-gp and the templates they are based on. So far most homology models rely on the 2HYD structure, since this is the best resolved ABC-transporter structure available. In addition, they fulfill most of the structural restraints obtained by cross-linking studies.[69, 70, 72] Models on basis of the MsbA structures were mainly used for exploring the conformational changes during the catalytic cycle or for performing docking studies. However, these structures unfortunately possess resolutions far from being suitable for docking experiments, with some templates only showing Cα atoms.

It should be stressed that also the homology models based on the retracted MsbA structures could fulfil a substantial amount of biochemical data.[52, 65] Also photoaffinity labelling data obtained with benzophenone-analogous propafenone derivatives could convincingly be mapped onto a homology model of P-gp based on the retracted MsbA structure.[67] This stresses the importance of a careful validation of the X-ray structures and the need for high resolution (< 2 Å) structures. Especially in case of these highly flexible, highly promiscuous membrane transporters cysteine cross link studies and ligand photoaffinity labelling could be interpreted in several ways and thus might lead to convincing, sound hypotheses even when based on partially wrong assumptions on the structure of the protein.

With the mouse P-gp structure published last year a high number of homology models for human P-gp are expected to be published this and the following years. The access to an X-ray structure of P-glycoprotein, although not in perfect resolution, represents a huge step forward for structure-based studies on this transporter. Although bacterial homologues share sequence identities with ABCB1 of about 35% (Table 1), one has to bear in mind that this incidence relies on the high conservation of the NBDs (> 50% sequence identity). The sequence in the TM domains possess only about 20% identity[79] and is therefore in the so called “twilight-zone” concerning homology modelling.[80]

5. Docking Studies

5.1. Binding Sites

Docking is a prevalent tool to identify the binding mode of drugs in the target protein and to use this information for identifying new hits in structure-based virtual screening runs. Concerning ABC transporters in general and P-glycoprotein in particular, we face the problem that hardly any binding sites for known P-gp ligands have been identified unambiguously. Current it seems common sense that there is a large binding cavity in the transmembrane region[81] which comprises distinct active sites. However, due to the polyspecificity of P-gp, there is still only limited knowledge on concrete interaction sites. Furthermore, cysteine-scanning mutagenesis studies showed that the protein is able to bind at least two different molecules simultaneously.[82] A more detailed characterisation of concrete binding sites for distinct substrates utilised techniques like cysteine and arginine scanning, photoaffinity labelling, or hypothesis driven mutagenesis (reviewed in References[24, 42, 76, 83]). This led to the characterization of the interaction regions of Rhodamine 123 and Hoechst 33342, named R- and the H-site,[84, 85] together with a regulatory site, which binds prazosin/progesterone.[86] Nevertheless, the release of the P-gp/CPPI-complexes presented another step forward in elucidating drug/P-gp interactions. Since the co-crystallized enantiomers showed distinct binding patterns, this information raised the assumption of stereoselectivity of P-gp in its ligand binding quality.[16] Stereoselectivity has also been shown for flupentixol[87] and propafenone derivatives. However, this has to be taken with a grain of salt and there are also ample reports on equal activity of enantiomes. Thus, as for niguldipine and verapamil both enantiomers showed equivalent activities,[88, 89] the distomers with respect to cardiovascular activity were used for clinical studies.

5.2. Docking

As the resolution of the hitherto available templates used for constructing protein homology models is quite low, only very few docking studies have been conducted so far. In a recently published paper, Pajeva et al.[90] docked quinazolinones into a homology model of human P-gp based on 3G61, which is in complex with SSS-QZ59. The binding site they used was defined by the co-crystallized ligands and was extended by 14 Å. The results suggested interaction with TM helices 5, 6 and 11 and were further confirmed by a pharmacophore model.

Becker et al. performed docking studies of the P-glycoprotein modulators colchicine, rhodamine B, verapamil and vinblastine into a homology model based on the closed-apo MsbA structure 3B5X.[91] The binding site was defined as a 30 ų cube which covered the complete central cavity. The resultant poses predicted that all ligands were able to interact with residues that were experimentally identified as important for ligand binding, strongly involving TM helices 5, 6, 7, 11 and 12. However, none of the drugs was able to contact every identified residue, which favours the hypothesis of distinct interactions sites forming one binding cavity.

Based on our extensive data from SAR studies on propafenones, we selected a small set of compounds for docking into a homology model based on 3G5U (mouse P-gp without QZ59 isomer).[77] The structure of the apo protein was chosen due to the better resolution compared to those of the complexes. As pointed out already, the structures do not seem to differ to a large extent, except for the amino acid residue Phe974. This residue corresponds to the hP-gp residue Phe978 in the alignment proposed by Aller et al.[16] and does not extend into the binding pocket in the homology model. Assuming a similar binding mode of the propafenone derivatives, poses were prioritized on basis of common scaffold clustering and protein-ligand interaction fingerprints. Interestingly, our results proposed similar interacting regions as have been identified for quinazolinones, involving especially TM helices 5, 6, 7, 8 and 12.[77] This might indicate a common binding region for these two compound classes.

6. Conclusions – a Personal View

Structure-based drug design with low-resolution X-ray structures has to be done very cautiously. Using protein homology models on basis of these structures is even more risky.[92] After almost 30 years of “structural blindness”, topped by the retraction of five X-ray structures, the recently published structure of mouse P-gp raised considerable hopes that structure-based drug design to elucidate the molecular basis of transporter/ligand binding becomes possible. However, docking studies performed so far kept the protein rigid. The problem of considering flexible receptors in docking experiments is to find the right balance between the accuracy of the method and the computational cost (reviewed in References[93, 94]). In case of small, defined binding sites with a co-crystallized ligand which is structurally similar to the compounds to be docked, rigid receptor docking would be appropriate. This is definitely not the case for P-gp, so we strongly need to question whether it is adequate to keep the protein rigid, although it is known that P-glycoprotein is highly flexible and, in addition, the resolution of the X-ray structure is as low as 3.80 Å! Furthermore, the X-ray structures of mouse P-gp, although for the first time co-crystallized with inhibitors, represent only singly snapshots of the transporter on its way through the transport cycle. Thus, there are still many drawbacks and pitfalls, which render docking/scoring approaches quite risky. On our personal opinion structure-based design approaches in this area might be useful for creating hypotheses, but definitely not for identifying new hit structures. Intense studies on the structures available, including molecular dynamics simulations of the different templates embedded in phospholipid bilayers, as well as careful validation of the homology models using the plethora of information available from cysteine and alanine scanning as well as cross-linking studies and mutagenesis will be necessary to provide accurate starting structures for docking experiments. The latter, due to the polyspecificity of the transporter, definitely will have to rely on information from ligand-based structure-activity relationship studies, pharmacophore modelling and photoaffinity labeling to be able to provide sound binding hypotheses for selected ligands and to aid in a deeper understanding of the molecular basis of transporter/drug interaction. However, even if the binding mode of several ligands could be resolved, there is still a long way to go for explaining the polyspecificity of the transporter on one side, and the presence of clear structure-activity relationships within structurally related compound series on the other side. As outlined in the introductory section, P-gp inhibitors show clear SAR, but this is observed for numerous scaffolds. Finally, there is experimental evidence that some of the compounds described as inhibitors are stimulating the ATPase activity of P-gp, thus indicating that they are substrates (propafenones,[95, 96] verapamil,[97] cyclosporine[98]). This adds another layer of complexity, especially when considering the fact that within the structurally analogous series of propafenone analogs some compounds inhibited the ATPase activity, some stimulated it in low concentrations and inhibited it in high concentrations and some compounds showed only ATPase stimulation, but no inhibition.[96] To explain these data on a structural basis, intense studies on the dynamics of the transporter combined with detailed investigations of the coupling of ATP-binding to TMD movement need to be performed. Thus, there is definitely still a long way to go for full understanding of the structure and function of this important drug efflux pump and its 47 human homologues.

Biographies

Freya Klepsch received a Master’s Degree in Biotechnology with a specialization in Chemistry of Active Substances from the University of Applied Sciences, FH Campus Wien, in 2008 under the supervision of Prof. Ernst Urban. Now she is PhD student in the pharmacoinformatics research group at the Department of Medicinal Chemistry of the University of Vienna. Her work focuses on the evaluation of binding modes of small molecules in ABC transporters.

Gerhard Ecker heads the Pharmacoinformatics Research Group at the Department of Medicinal Chemistry, University of Vienna. He also coordinates the research focus “Computational Life Sciences” of the Faculty of Life Sciences. He studied pharmacy and received his doctorate in natural sciences from the University of Vienna under the supervision of Wilhelm Fleischhacker and Christian Noe. After his postdoctoral training at the group of J. Seydel in Borstel (Germany) he was appointed associate professor for medicinal chemistry at the University of Vienna in 1998 and full professor for pharmacoinformatics in 2009. He has published more than 90 articles related to SAR and QSAR studies on P-glycoprotein and his main scientific interests include pharmacoinformatic approaches to target drug efflux pumps, in silico high throughput screening methods for promiscuous targets and antitargets, and nonlinear methods in drug design. He is currently vice-president of the Austrian Pharmaceutical Society and President of the European Federation for Medicinal Chemistry. He also coordinates the EUROPIN PhD Programme in Pharmacoinformatics.