|Home | About | Journals | Submit | Contact Us | Français|
The human breast cancer resistance protein (BCRP/ABCG2) is the second member of the G subfamily of the large ATP-binding cassette (ABC) transporter superfamily. BCRP was initially discovered in multidrug resistant breast cancer cell lines where it confers resistance to chemotherapeutic agents such as mitoxantrone, topotecan and methotrexate by extruding these compounds out of the cell. BCRP is capable of transporting non-chemotherapy drugs and xenobiotiocs as well, including nitrofurantoin, prazosin, glyburide, and 2-amino-1-methyl-6-phenylimidazo [4,5-b]pyridine. BCRP is frequently detected at high levels in stem cells, likely providing xenobiotic protection. BCRP is also highly expressed in normal human tissues including the small intestine, liver, brain endothelium, and placenta. Therefore, BCRP has been increasingly recognized for its important role in the absorption, elimination, and tissue distribution of drugs and xenobiotics. At present, little is known about the transport mechanism of BCRP, particularly how it recognizes and transports a large number of structurally and chemically unrelated drugs and xenobiotics. Here, we review current knowledge of structure and function of this medically important ABC efflux drug transporter.
The human breast cancer resistance protein (BCRP) is the second member of the G subfamily of the ATP-binding cassette (ABC) efflux transporter superfamily, and hence also designated as ABCG2. BCRP is an approximately 75 kDa polytopic plasma membrane protein first identified in 1998 in a multidrug resistant human breast cancer cell line MCF-7/AdrVp which does not express other known multidrug efflux transporters such as P-glycoprotein (P-gp) or the multidrug resistance protein 1 (MRP1) . Two almost identical proteins as BCRP with only a few amino acid differences were later discovered independently by other laboratories from mitoxantrone-resistant human cancer cell lines (so named as MXR)  and human placenta (so named as ABCP) . Despite the fact that there is low protein sequence identity in nucleotide binding domains (NBDs) (~20%) and essentially no protein sequence identity in membrane spanning domains (MSDs) between BCRP and P-gp or MRP1, transfection of cells with BCRP cDNA confirmed its ability to confer resistance to a variety of chemotherapeutic agents such as mitoxantrone and topotecan [1–2]. Like P-gp or MRP1, BCRP performs ATP hydrolysis-dependent efflux transport of a large number of structurally and chemically unrelated compounds that also include non-chemotherapy drugs and xenobiotics [4–6].
Various clinical studies have demonstrated BCRP expression and a possible role of the transporter in drug resistance in leukemia [7–9]. In addition, numerous studies have documented that BCRP is highly enriched in a population of primitive stem cells, the so-called side population (SP), of human bone marrow and other organs, and may provide xenobiotic protection for stem cells [10–14]. Similar to P-gp, BCRP is also highly expressed in organs important for the absorption (the small intestine), elimination (the liver and kidney), and distribution (the blood-brain and placental barriers) of drugs and xenobiotics , it is therefore increasingly recognized for its important role in drug disposition and tissue protection [4–5, 16–17] An unique feature is that BCRP expression in the mammary gland of mice, cows and humans was found to be strongly induced during lactation and responsible for the secretion and concentration of clinically and toxicologically important substances such as the dietary carcinogen 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP) and the chemotherapeutic agent topotecan into milk . BCRP is a highly polymorphic transporter with over 80 single-nucleotide polymorphisms in the BCRP gene . Of these, Q141K is most extensively studied and has been found to be associated with inter-individual variations in the pharmacokinetics, response or toxicity of drugs [20–22] as well as genetic diseases such as gout . The physiological roles of BCRP are likely to provide tissue protection against endogenous toxins or xenobiotics and regulate cellular homeostasis of physiologically important endogenous compounds such as heme, porphyrins, riboflavin, and estrogens [24–26]. Despite its significant medical importance, our understanding of the transport mechanism of BCRP is still limited. In this mini-review, we summarize current knowledge of structure and function of human BCRP.
BCRP transports a highly diverse range of substrates. The list of BCRP substrates has been rapidly expanding since its discovery. BCRP substrates include not only chemotherapeutic agents such as mitoxantrone, methotrexate, topotecan, irinotecan and its active analog SN-38, and tyrosine kinase inhibitors imatinib and gefitinib, but non-chemotherapy drugs such as prazosin, glyburide, nitrofurantoin, dipyridamole, statins, and cimetidine as well as non-therapeutic compounds such as the dietary flavonoids, porphyrins, estrone 3-sulfate (E1S), and the carcinogen PhIP. This has been extensively reviewed elsewhere [4–6, 16]. Chemical structures of representative BCRP substrates are shown in Fig. (1). There is a considerable overlap in substrate specificity between BCRP and other ABC transporters. For example, of the BCRP substrates listed in Fig. (1) glyburide, imatinib, methotrexate, mitoxantrone, prazosin, and SN-38 are also P-gp substrates, although some of them such as methotrexate are poor substrates of P-gp [27–29]. Notably, in addition to hydrophobic substrates such as mitoxantrone, BCRP can also transport hydrophilic conjugated organic anions, particularly the sulfated conjugates with high affinity, whereas P-gp generally transports hydrophobic compounds. The overlap in substrate specificity between BCRP and P-gp can lead to a synergistic effect of the transporters in limiting drug penetration across tissue barriers such as the blood-brain barrier [30–31]. The transport kinetic parameters (Km and Vmax) of BCRP for several hydrophilic substrates have been determined using plasma membrane vesicle transport assays and are summarized in Table 1. Except for methotrexate, most of the BCRP substrates have the affinity constants (Km) in a low μM range. For hydrophobic substrates, the apparent Km value of only one substrate, glyburide, was estimated using the cellular accumulation assay . Understanding transport kinetics is of important value in predicting the contribution of BCRP to the disposition of substrate drugs in vivo. Lipid membrane environment could affect transport kinetics of BCRP. For example, the addition of cholesterol dramatically increased Vmax of BCRP for methotrexate without affecting Km . At present, what determines the substrate specificity of BCRP remains elusive. To date, there appears to be only one structure-activity relationship (SAR) study for BCRP substrates, namely SN-38 and its analogs . These authors found that SN-38 analogues with high polarity were good BCRP substrates. The high polarity of SN-38 analogues appears to be attributed to the presence of hydroxyl or amino groups at positions 10 and 11 on the A ring of SN-38. Hydrogen bonds formed through a hydroxyl or amino group could contribute to SN-38 recognition by BCRP. Based on such SAR analysis, novel SN-38 analogues that are not BCRP substrates have been developed to bypass BCRP-mediated SN-38 resistance .
A wide variety of compounds that can inhibit BCRP have been extensively reviewed [4–6, 16, 36]. BCRP inhibitors include, among others, fumitremorgin C (FTC), the FTC analogue Ko143, the acridone carboxamide derivative GF120918, anti-HIV protease inhibitors nelfinavir and ritonavir, the dietary flavonoids chrysin and biochanin A, the tyrosine kinase inhibitors gefitinib and imatinib [5, 16, 36], and herb extracts . Chemical structures of representative BCRP inhibitors are shown in Fig. (2). Some BCRP inhibitors such as Ko143 [38–39], GF120918 , gefitinib , and imatinib  are highly potent with IC50 values in a nM range. Of these, FTC and Ko143 are highly selective for BCRP with no or little inhibition for P-gp and MRP1 [38–39]. However, others such as cyclosporin A and anti-HIV protease inhibitors seem to be general inhibitors of ABC transporters. GF120819 is a potent inhibitor for both BCRP and P-gp.
SAR and quantitative SAR (QSAR) analyses for structurally related or unrelated BCRP inhibitors have been performed to understand BCRP-inhibitor interactions. This topic has been extensively reviewed elsewhere [43–46], and hence is only briefly discussed here. SAR analysis has been reported for taxane-based reversal agents , flavonoids [48–51], tamoxifen analogues , cyclin-dependent kinase inhibitors , tariquidar analogues , and FTC analogues . Structural features of these compounds contributing to BCRP inhibition were investigated in these studies. For example, hydroxyl groups at positions 3, 7, and 4′ of flavonoids are detrimental to BCRP inhibition, whereas the hydroxyl group at position 5 has a positive contribution . QSAR models have also been developed to identify pharmacophores in BCRP inhibitors which were analyzed either as structurally related compounds such as flavonoids [49–50] and tariquidar analogues  or as groups of structurally diverse compounds [45, 55–57]. Various modeling methods were used, including classical 2D-QSAR , comparative molecular field analysis (CoMFA) or comparative molecular similarity indices analysis (CoMSIA) , 3D-QSAR method based on VolSurf descriptors of molecular-interaction fields (MIFs) related to hydrophobic interaction forces, polarizability, and hydrogen bonding potential , classification methods based on molecular descriptors including logD and polarizability , and generation of chemical fragmentation codes coupled with regression analysis [45, 55]. According to the QSAR models reported so far [55, 57], some descriptors such as lipophilicity and polarizability appear to be significant determinants for BCRP inhibition. On the other hand, depending on compounds, some descriptors such as planar structure, amine bonded to carbon of a heterocyclic ring, and hydrogen bonding potential may also be important. QSAR models could seemingly help predict BCRP inhibition for certain inhibitors. For example, Saito et al.  selected 49 compounds representing 9 different classes of BCRP inhibitors and developed a QSAR model using chemical fragmentation codes and multiple linear regression analysis. This model well predicted BCRP inhibition by gefitinib, a compound that was not used in the original model development. Because of the structural diversity, none of the existing QSAR models can work for all BCRP inhibitors. Also, a QSAR model developed using one structural set of compounds cannot be used for another structural set of compounds. Even for compounds with similar chemical structures such as flavonoids, contradicting results were obtained in different laboratories with respect to which substitutions would have favorable or detrimental effects on BCRP inhibition. Note that all the current QSAR models were developed through indirect modeling based on chemical structures of inhibitors because a high resolution 3D structure of BCRP has not been available. Hence, exactly how inhibitors interact with BCRP is essentially not known. Moreover, owning to multiple drug binding sites in BCRP as discussed below, the inhibition profile used in QSAR analysis that was determined with one substrate may not be the same as that obtained with a different substrate.
An earlier study demonstrated that two different substrates such as mitoxantrone and rhodamine 123, topotecan or daunorubicin did not reciprocally inhibit the efflux of each other , suggesting that there are multiple drug binding sites in BCRP that are not or only partially overlapping. To prove drug binding in BCRP, direct drug binding interactions with BCRP has been investigated using photo-affinity labeling [59–62]. The prazosin analogue [125I]-IAAP is the photo-active compound that is most frequently used. Labeling of BCRP with [125I]-IAAP was achieved [59, 61–62] and could be prevented by the addition of other BCRP substrates or inhibitors in a concentration-dependent manner [61–63]. [125I]-IAA-rhodamine 123  and [3H]-azidopine  have also been successfully used to photo-label BCRP and again the labeling could be prevented by increasing concentrations of several known BCRP substrates. Since both [125I]-IAAP and [3H]-azidopine are BCRP substrates , their ability to photo-label BCRP suggests a direct interaction of the two compounds with the transporter. Photo-affinity labeling of purified BCRP combined with trypsin digestion and mass spectrometry analysis of peptide fragments may provide valuable information for the drug-binding sites in BCRP. At present, the residues potentially involved in interactions of the photo-active substrates with BCRP have not been identified. BCRP substrates or inhibitors can displace the binding of photo-affinity analogues, indicating that photo-affinity labeling studies could serve to screen BCRP substrates or inhibitors. Unfortunately, [3H]-azidopine is no longer commercially available.
There are only a few equilibrium or kinetic binding studies for substrates or inhibitors of BCRP [64–66]. Clark et al. examined the kinetics of association and dissociation of [3H]-daunomycin with the BCRP mutant R482G which can effectively transport the drug  using plasma membranes isolated from insect cells, and the ability of several other drugs to displace [3H]-daunomycin binding . They showed that the binding affinity (Kd) of [3H]-daunomycin to BCRP was approximately 0.6 μM. Doxorubicin, prazosin and non-radioactive daunomycin nearly completely displaced [3H]-daunomycin binding, whereas mitoxantrone and Hoechst 33342 only had a partial displacement. Rhodamine 123 and methotrexate had no any effects on [3H]-daunomycin binding. These results again suggest that there are multiple drug binding sites in BCRP with doxorubicin, prazosin and daunomycin binding clustered in one region and rhodamine 123 and methotrexate binding in different regions. The binding sites for mitoxantrone and Hoechst 33342 may partially overlap with the binding sites for doxorubicin, prazosin and daunomycin. The drug binding study monitored by quenching of intrinsic fluorescence of purified BCRP by Pozza et al. revealed that there seem to be two binding sites for each substrate or inhibitor in BCRP, with one high affinity (Kd in a low μM range) and one low affinity sites . The inhibitor GF120918 did not displace mitoxantrone binding, suggesting distinct binding sites for substrate and inhibitor.
For ABC proteins, the transport process is supposed to be coupled with ATP binding and hydrolysis which provide the energy for substrate translocation. ATP hydrolysis (ATPase) activity of BCRP has been reported in various studies . ATP hydrolysis by BCRP was confirmed by photo-labeling of BCRP under hydrolytic conditions with 8-azido [α-32P]ATP in the presence of vanadate and Mg2+ or Co2+ [69–70]. Most of the studies used BCRP-enriched insect cell plasma membranes, although plasma membranes from other expression systems such as HEK cells, Pichia pastoris, and Lactococcus lactis or purified BCRP were also utilized. In general, BCRP exhibits basal ATPase activity that is considerably lower than that of P-gp, but much greater than that of MRP1. Unlike P-gp, whose basal ATPase activity can be strongly stimulated several-fold by its substrates, basal ATPase activity of BCRP could only be stimulated up to 2-fold by its substrates such as prazosin [33, 70–73]. It has been suggested that this insensitivity of drug-stimulation of BCRP ATPase activity may reflect the presence of endogenous substrates or a partially uncoupled form of BCRP . Note that, basal AT-Pase activity of MRP1 could not be strongly stimulated by its substrates either . The addition of cholesterol can significantly potentiate ATPase activity of BCRP and its substrate-stimulation characteristics . How ATP binding and/or hydrolysis is coupled with drug transport in BCRP is not known. Only one study showed that ATP binding alone could convert the drug binding site from a high to a low affinity state , suggesting that it is ATP binding, not ATP hydrolysis, that appears to initiate drug transport by BCRP.
A unique feature that distinguishes BCRP from other ABC proteins such as P-gp and MRP1 is in its basic structure. P-gp or MRP1 possesses two repeated halves which contain two MSDs each followed by one NBD. In contrast, BCRP contains only one NBD preceding one MSD [1–3]. It has been widely accepted that a functional ABC transporter requires two MSDs and two NBDs which form a central substrate translocation pathway. Thus, BCRP may function as a homodimer with evidence provided in several earlier studies [76–78]. The earlier study by Kage et al. revealed that BCRP migrated as a 70 kDa band on SDS-PAGE under reducing condition, but as a 140 kDa complex in the absence of reducing agents, suggesting that BCRP could form a homodimer bridged by disulfide bonds . Litman et al.  also observed a molecular mass shift of BCRP from 72 kDa to 180 kDa after treatment with chemical cross-linking agents. Bhatia et al.  showed that chimeric fusion proteins containing two BCRP monomers fused with or without a linker peptide were properly targeted to the plasma membrane and retained drug transport activity. Several recent studies demonstrated that BCRP may exist as a higher order oligomer [55, 66, 73]. Using sucrose density gradient sedimentation and non-denaturing gel electrophoresis, Xu et al.  provided evidence that detergent-solubilized BCRP was capable of formation of homotetramers. Likewise, the electron microscopy (EM) analysis of BCRP protein particles in detergent solutions identified an octameric state that was organized as a tetramer of dimers . The recent cryo-EM analysis of 2D crystals revealed that BCRP could form a tetrameric complex (two BCRP dimers) in one unit cell of the projection maps . These data suggest that BCRP can form higher order oligomeric states in vitro. Most recently, BCRP with GFP/YFP attached at its N-terminus was expressed in HEK cells and, dimer/oligomer formation of BCRP was determined by measuring the FRET efficiencies between GFP and YFP in intact cells using fluorescence resonance energy transfer (FRET) microscopy . The FRET efficiencies of GFP/YFP-tagged BCRP were significantly higher than the GFP/YFP pair control, suggesting that BCRP likely forms a homodimer or homooligomer in vivo in intact cells . At present, the role of oligomerization in BCRP function is not clear. It has been suggested that the function of BCRP could be regulated by the dynamic association/dissociation of BCRP monomers in the dimeric/oligomeric complex via protein-protein interactions . Therefore, disruption of the contact interfaces between BCRP monomers by small peptides or small molecules, thus preventing the formation of active higher order oligomers, may represent a promising strategy for inhibition of BCRP. Studies performed to identify residues or regions in BCRP potentially involved in oligomerization are discussed below.
Several studies suggest that dimer/oligomer formation of BCRP may involve intermolecular disulfide bonds formed between extracellular Cys residues [78, 82–85]. These studies showed that mutations of Cys603 in the extracellular loop connecting transmembrane (TM) segments 5 and 6 resulted in the appearance of BCRP monomers on non-denaturing polyacrylamide gel under non-reducing conditions, whereas wild-type BCRP formed only dimeric complex under the same conditions. Mutations of other Cys residues Cys592 and Cys608 in the same extracellular loop resulted in a mixed population of monomeric and dimeric BCRP  or only dimeric BCRP  under non-reducing conditions. These data appear to support that Cys603 is involved in intermolecular disulfide bond formation that is responsible for dimer/oligomer formation of BCRP, but Cys509 and Cys608 are not. Nevertheless, several lines of evidence indicate that intermolecular disulfide bond formation involving Cys603 is likely not the sole mechanism of BCRP oligomerization. Various studies have shown that mutations of Cys603 alone do not affect expression and function of BCRP at all [80, 82–83]. Additionally, the studies by Liu et al.  and Shigeta et al.  have suggested that Cys592 and/or Cys608 are also potentially involved in intermolecular disulfide bond formation. Importantly, Shigeta et al. showed that the triple-mutant C592S/C603S/C608S retained substantial resistance to mitoxantrone, topotecan, and SN-38 . Because intermolecular disulfide bonds via the three extracellular Cys residues are not expected to be formed in the triple-mutant, these authors concluded that BCRP could possibly function as a dimer or an oligomer formed through non-covalent protein-protein interactions. The recent study using FRET and chemical cross-linking also revealed that Ala substitution of Cys603 had no any effect on dimer/oligomer formation of BCRP in vivo in intact cells . Thus, all these studies appear to support the conclusion that, besides intermolecular disulfide bonds formed by Cys603, intermolecular disulfide bonds formed by other Cys residues and/or non-covalent protein-protein interactions could also have a crucial contribution to BCRP oligomerization. Xu et al.  showed that the recombinant fragment containing TM5-loop-TM6 was capable of forming homododecamer, suggesting that this region is important for BCRP oligomerization. Phosphorylation of BCRP by the Pim-1 kinase has been shown to promote its oligomerization and drug resistance in human prostate cancer cells; however, the underlying mechanism is unknown . Taken together, despite extensive studies, it is still not clear as to whether the minimal functional unit of BCRP in the plasma membrane is a homodimer or a homooligomer and the effect of the dimeric or oligomeric state of BCRP on drug binding and transport. This requires further investigation into the mechanisms by which BCRP forms homodimers or homooligomers.
Knowledge in membrane topology of BCRP is important for understanding the structural basis of BCRP action and for homology modeling of the transporter. Hydropathy analysis based on amino acid sequence predicted that BCRP contains one NBD (residues ~1 – 396) followed by one MSD (residues ~397 – 655), and the MSD contains 6 TM α-helices [1–3, 89–90]. The membrane topology of BCRP has recently been determined experimentally [72, 91]. In total of 20 hemagglutinin (HA) tags were inserted in various locations throughout BCRP, and the insertion mutants were expressed in HEK cells by transient transfection. Polarity of the tags with respect to the plasma membrane was determined by immunofluorescence in the presence and absence of membrane-permeabilizing detergent. The study confirmed 6 TM α-helices in BCRP; however, there are major differences between the experimental and the computer-predicted topology models as follows. As shown in Fig. (3) the residues 431 – 450 predicted to be part of TM2 were found to be in an extracellular loop connecting TM1 and TM2. The residues 566 – 583 predicted in the extracellular loop between TM5 and TM6 were determined to form TM5. The residues 460 and 462 predicted in an intracellular loop connecting TM2 and TM3 were now found in the TM2. The computer-predicted TM1, TM3, TM4 and TM6 are in agreement with the experimental data. Such a change in the membrane topology would be expected to alter the composition and shape of the putative substrate translocation pathway which is primarily composed of multiple TM α-helices. Although this experimental topology model awaits further verification by other biochemical and structural experiments, it appears to be consistent with existing biochemical data. For example, Asp590 which is the only N-linked glycosylation site in BCRP [92–93] is located in the extracellular loop connecting TM5 and TM6 Fig. (3). Cys603 responsible for intermolecular disulfide bond formation is also in the extracellular loop, and Arg482 which is crucial for substrate specificity and overall transport activity of BCRP is located in TM3 near the cytosolic membrane interface Fig. (3).
This topology structure of BCRP was compared with that of mouse P-gp observed in crystal structures . Except for two regions, the overall topology structures of BCRP and P-gp were surprisingly similar . First, the large intracellular loop connecting TM2 and TM3 of P-gp does not exist in the corresponding region of BCRP. Second, the extracellular loop between TM5 and TM6 of BCRP is significantly longer than the corresponding region in P-gp. Notably, the amino acid identity in MSDs between BCRP and P-gp is only ~ 4% based on pairwise alignment ; however, according to the experimental topology model of BCRP, the amino acid identity in MSDs between BCRP and the first half of mouse P-gp is increased to 18% (without gaps). Likewise, the amino acid identity in MSDs between BCRP and the bacterial ABC half transporter Sav1866 or the lipid A transporter MsbA was also found to be much improved .
High yield expression of human BCRP with a His tag attached has been achieved in baculovirus-infected insect cells [73, 95] and the yeast Pichia pastoris , facilitating purification of BCRP by Ni2+-affinity chromatography for structural analysis [61, 66, 72]. BCRP could be solubilized from membranes using detergents such as 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonate (CHAPS) , Foscholine derivatives , or n-dodecyl-β-D-maltoside (DDM) . Purified BCRP in detergent solution was active in drug binding  or ATP hydrolysis [66, 72], and suitable for electron microscopy (EM) analysis of protein particles  or 2D crystals .
At present, there are only two structural studies for BCRP [61, 72]. McDevitt et al.  reported the first EM analysis of single protein particles of BCRP purified from insect cells. These analyses illustrated that BCRP protein particles in a detergent solution formed a higher order oligomeric complex that was organized as a tetramer of BCRP dimers. A 3D structure was reconstructed at 18 Å resolution with a maximum diameter of 180 Å and an overall peak height of 145 Å . At such a low resolution, only the overall shape and oligomeric state of the BCRP complex could be visualized. Most recently, Rosenberg et al.  determined the first projection structures of BCRP purified from Pichia pastoris by cryo-EM of well-diffracting 2D crystals. The 2D crystals showed a p121 symmetry and the projection structures were determined to 5 Å resolution. At this relatively high resolution, ring-shaped high density features in the projection maps were clearly visualized, probably representing TM α-helices (and NBDs). The dimension of the asymmetric unit cell of the crystals indicated that there were 4 BCRP monomers (two BCRP dimers) in one unit cell, indicating the existence of an oligomeric complex of BCRP in the 2D crystals. Importantly, the unit cell dimension of the crystals grown in the presence of mitoxantrone was significantly smaller than that in the absence of mitoxantrone in the dimension b, suggesting that BCRP has undergone a significant conformational change upon mitoxantrone binding. Thus, BCRP appears to have a more closed configuration in the presence of mitoxantrone. This is the first experimental evidence showing substrate binding induced conformational changes in BCRP, which has implications for transport mechanism of the transporter. The number of TM α-helices, and their orientations in the membrane and assignment in the BCRP protein sequence cannot be resolved from the projection structures. This will require the calculation of a 3D map at a high resolution.
To further illustrate drug binding and transport mechanism of BCRP, homology models have been developed in different laboratories. The first model of BCRP was constructed using the crystal structure of MsbA from Vibrio cholera (VcMsbA)  as the template for TM segments. Unfortunately, this crystal structure has now been retracted. Two additional models of BCRP were developed using the crystal structure of Sav1866  as the template [89, 97]. Although certain features in the models seem to be consistent with experimental data, for example, showing multiple drug binding sites in a large central cavity, these studies were solely based on the computer-predicted BCRP topology. However, we now know that the topology structure of BCRP based on computer prediction is significantly different from the experimental topology structure. Here, we discuss the homology models that were developed based on the experimental topology structure of BCRP .
First of all, the sequence alignment for TM segments between BCRP and the templates (MsbA, the first and second halves of mouse P-gp, and Sav1866) was refined manually by comparing the experimentally determined TM segments of BCRP with those of the templates observed in crystal structures . The template used in modeling of the NBD was the ATP-binding subunit of the E. coli maltose transporter MalK, whose crystal structures were resolved at atomic resolutions . Next, the dimeric NBD-MSD structure of BCRP was modeled as a whole in a single step, with no loss of information in the interactions between monomers and between protein domains within the same monomer. The templates were edited to reflect the same domain organization of BCRP, by “cutting” the linker regions between the MSD and the NBD of the templates and changing the order of the MSD and the NBD. The model building was done by the MODELLER package .
With the same method as described above, three homology models of BCRP representing different conformational states have been generated. The first model using the MsbA structure as template (PDB code 3B5W)  represents the substrate-unbound nucleotide-free inward-facing open apo conformation Fig. (4A). The second model using the mouse P-gp structure as template (PDB code 3G60)  represents the substrate-bound nucleotide-free inward-facing closed apo conformation Fig. (4B). The third model using the Sav1866 structure as template (PDB code 2HYD)  represents the nucleotide-bound outward-facing conformation Fig. (4C). The first two models have been published . Ward et al.  reported crystal structures of MsbA trapped in different conformations, two similar nucleotide-bound outward-facing conformations and two different nucleotide-free inward-facing apo conformations. TM segments in the closed apo form of MsbA are much more compact than in the open apo form. These authors proposed that substrate binding to the open apo form promotes the closure of TMs, which in turn, sends a signal to the NBDs, allowing the formation of the ATP sandwich in the outward-facing conformation upon nucleotide binding. The results of homology modeling for BCRP as well as the finding that the projection structures of BCRP display a more closed conformation in the presence of mitoxantrone than in the absence of the drug appear to be consistent with the structural data of MsbA.
There are several unique features in the models. First, the intracellular entry of the two inward-facing apo forms Fig. (4A and 4B) is large enough to allow access of a bulk of BCRP substrates from either the inner lipid leaflet of the plasma membrane or cytoplasm. Second, the extracellular loop 1 connecting TM1 and TM2 forms intermolecular contacts with the extracellular loop 3 between TM5 and TM6, thus possibly stabilizing the dimeric structure. Third, there is not a coupling helix 1 in the intracellular loop connecting TM 2 and TM3, which is present in all of the available ABC efflux transporter structures and could play a crucial role in the communications between the NBD and the TMD of the same monomer. Residues 285 – 380 in the putative linker region connecting the NBD to the MSD of BCRP could not be modeled due to the lack of appropriate templates. This linker region in BCRP is much longer than those in other ABC transporters, part of which may serve as the coupling helix 1. As revealed in crystal structures of Sav1866, MsbA, and P-gp, there is a coupling helix 2 that crosses over and associates with the NBD of the opposite monomer. Such a coupling helix 2 also exists in the BCRP models in the intracellular loop connecting TM4 and TM5 which may define the relative orientation of individual domains and allow flexibility in the transporter. Forth, comparing the open and closed inward-facing apo forms Fig. (4A and 4B), there is a rotation of ~30 °C for TM4 and TM5 of BCRP which subtend a narrower angle to the rest of the MSD in the closed apo form. Similar changes in the angle of TM4 and TM5 may also occur in other ABC transporters such as Sav1866 and P-gp as pointed out by Kos and Ford . It is worth noting that it is TM4 and TM5 that mediate the coupling helix 2 to cross over the opposing NBD. The rotation and change in the angle of TM4 and TM5 also force TMs 1 – 3 and 6 to similarly change their positions, resulting in a more closed intracellular entry in the closed apo structure as compared with the open apo structure. Such a domain rearrangement would alter the shape of the substrate-binding cavity which is formed primarily by residues in the TMs by making it more compact in the presence of substrate binding. Fifth, in the outward-facing form of BCRP, there are even more drastic conformational changes, with the intracellular entry completely closed and a wide open V-shaped gap in the extracellular side extending to the lipid bilayer Fig. (4C). This V-shaped gap may allow the release of substrates from BCRP after ATP binding/hydrolysis. Because the intracellular entry in the outward-facing form is completely closed, the two NBDs of the BCRP dimer form the so-called ATP sandwich. Finally, comparing the membrane topology of P-gp and BCRP, the extracellular loop between TM5 and TM6 in BCRP is much larger than the corresponding region in P-gp . Recent studies suggest that this extracellular loop may be critical in modulating substrate binding  as well as stability and ubiquitin-mediated degradation of BCRP . In our models, this extracellular loop closely contacts with the extracellular loop between TM1 and TM2. Such contacts may help stabilize the BCRP dimer through non-covalent interactions. Consistent with this, Xu et al.  demonstrated that the TM5-loop-TM6 fragment may play a critical role in BCRP oligomerization. A monoclonal antibody 5D3 that recognizes a conformational sensitive epitope in this extracellular loop has been highly useful for biochemical characterization of BCRP in terms of interactions of the transporter with substrates or inhibitors [63, 102, 104] or conformational changes induced by point mutations .
Do the homology models interpret existing biochemical data? First of all, docking calculations of various BCRP substrates to the closed apo form did suggest the existence of multiple substrate binding sites in the central cavity which is primarily formed by TM α-helices . Second, Arg482 has been extensively analyzed by site-directed mutagenesis and found to be crucial for substrate specificity and transport activity [106–109]. In the homology models, Arg482 in TM3 is located in the central cavity close to the cytosolic membrane interface with the side chain pointing towards the drug translocation pathway. Docking calculations indicated that Arg482 may directly interact with mitoxantrone and Hoechst33342, but not with prazosin and SN-38. This is consistent with previous studies showing that resistance to mitoxantrone was increased, but resistance to SN-38 or efflux of prazosin was not significantly affected, by the mutations of Arg482 [107, 109]. This also appears to be in good agreement with the studies showing that prazosin binds to a pharmacologically distinct site relative to other BCRP substrates such as mitoxantrone and Hoechst33342  and the binding of a prazosin derivative to BCRP was relatively unaffected by the mutations of Arg482 . Thus, the homology models could be used to interpret existing data. Further biochemical studies are needed to validate these models. If validated, these models may be valuable for the development of pharmacophores of BCRP substrates or inhibitors.
Naturally occurring variants of BCRP caused by single nucleotide polymorphisms in the coding region of BCRP gene may have significant physiological and pharmacological relevance. A large body of data is available [19, 110] and summarized in Table 2. The most important variant is Q141K, which occurs in Japanese and Chinese populations at high allele frequencies (30 – 60%) and in Caucasians and African-American populations at relatively low allele frequencies (5 – 10%) [111–112]. Several studies consistently revealed that Q141K had a lower protein expression level than wild-type BCRP in both transfected cells and human tissues [113–115]. The transport activity of Q141K in vivo would be expected to be decreased compared with wild-type BCRP owing to its lower level of protein expression. Indeed, human subjects carrying the Q141 variant often had higher plasma levels of BCRP substrate drugs than the subjects carrying wild-type BCRP [116–118]. A recent study has revealed that Q141K undergoes increased lysosomal and proteasomal degradations than wild-type BCRP, possibly explaining the lower level of protein expression of the variant . A systematic study of 18 natural variants of BCRP expressed in insect cells showed that the variants Q126stop, F208S, S248P, E334stop, and S441N were defective in porphyrin transport, whereas F489L displayed approximately 10% of the transport activity of wild-type BCRP . These are rare variants with allele frequencies generally less than 2% [111–112, 120–121].
Much attention has also been devoted to elucidating the structure and function of BCRP by site-directed mutagenesis of particular residues of interest, and the data is summarized in Table 2. There are basically four categories of non-natural BCRP mutants reported thus far as discussed below. First, mutations do not affect plasma membrane expression, but alter substrate specificity and/or overall transport activity. In this category, Arg482 has been most extensively characterized. Arg482 is an important determinant of substrate specificity and transport activity. For example, wild-type BCRP does not transport daunorubicin, rhodamine 123, and Lyso-Tracker Green; however, the mutants R482T and R482G do . Therefore, for these substrates, the mutations of Arg482 result in “gain-of-function”. On the other hand, mitoxantrone, BODIPY-prazosin, and Hoechst 33342 are substrates of both wild-type BCRP and the two mutants [107–108]. Methotrexate is a substrate only for wild-type BCRP . Mutations of Arg482 generally increased resistance to mitoxantrone, but did not have much effect on resistance to SN-38 . Thus, the effect of Arg482 on transport activity is substrate-dependent. Although Arg482 is likely located in the substrate translocation pathway, exactly how mutations of Arg482 change substrate specificity and transport activity is still unknown. Alqawi et al.  showed that wild-type BCRP was more intensely photo-labeled with a photo-active analog of rhodamine 123 than R482T even though wild-type BCRP does not transport rhodamine 123. The positive charge of Arg482 does not seem to be a key determinant because the substitution of Arg482 with Lys of the same charge resulted in a complete loss of drug resistance .
Ala substitution of the basic residue His457 within TM2 has been shown to increase transport activity and drug resistance several-fold and caused significant conformational change . Docking calculations suggest that His457 may be directly involved in substrate binding. Miwa et al.  reported that amino acid substitutions of Glu446 resulted in complete loss of drug resistance to SN-38 and mitoxantrone. These authors also found that mutations of Asn557 tended to selectively decrease resistance to mitoxantrone and/or SN-38. Similarly, Leu substitution of Phe431 retained full transport activity for porphyrin, but not for methotrexate . According to the membrane topology of BCRP Fig. (1), Phe431, Glu446 and Asn557 are located in the extracellular loop 1 connecting TM1 and TM2 and the intracellular loop 2 connecting TM4 and TM5, respectively. How can amino acid substitutions of residues in the cellular loops between TM α-helices affect transport activity remains to be determined. Mutation of Glu211 adjacent to the Walker B motif of the NBD also belongs to this category. Hou et al.  demonstrated that Gln substitution of Glu211 had no effect on protein expression, but completely abolished ATP-dependent transport of methotrexate and ATPase activity. This confirms that Glu211 directly adjacent to Asp210 within the Walker B motif may activate the water molecule to attack the bound Mg·ATP in the ATP hydrolytic cycle as previously suggested for other ABC transporter . This capability would be lost by mutation at Glu211, thus leading to inactivation of BCRP.
In the second category, mutations affect biogenesis with decreased stability, lower expression and/or altered subcellular distribution of BCRP. Arg383 in the linker region has been shown to be crucial for biogenesis of BCRP. Mutations of Arg383 resulted in a significant decrease in the protein level, partial retention in the endoplasm reticulum, and altered glycosylation . The lower levels of Arg383 mutants may be caused by rapid degradation by the proteasome, and the treatment with mitoxantrone assisted in protein maturation . Likewise, Ala substitution of Arg426 in the extracellular loop 1 also caused rapid degradation, leading to lower levels of expression of intact BCRP and transport activity (unpublished results).
Gly406 and Gly410 in the GXXXG dimerization motif in TM1 of BCRP have been investigated for their impact on potential dimerization [97, 127]. Single or double Leu substitutions of Gly406 and Gly410 resulted in lower expression possibly caused by increased degradation, as well as impaired or complete loss of ATP hydrolysis and substrate transport . Some of these mutants were partially retained in the endoplasm reticulum. Triple mutations of Gly406 and Gly410 with Thr402 which is just adjacent to the GXXXG motif resulted in an even more reduced level of expression . These results suggest that mutations within or near the GXXXG motif affect proper folding of BCRP. Since the GXXXG motif and Thr residues in TMs are known to be important for interhelical interactions [128–129], the GXXXG motif and Thr402 in BCRP were suggested to play a role in dimerization [97, 127]. In a Sav1866-based model, the GXXXG motif and Thr402 in TM1 interact with TMs 5 and 6 of the opposite BCRP monomer, contributing to dimerization . However, the double or triple mutants of Thr402, Gly406 and Gly410 could still form dimers by chemical cross-linking [97, 127]. Thus, these data do not conclusively support the role of the GXXXG motif and Thr402 in BCRP dimerization. In the homology models shown in Fig. (2), TM1 interacts with TM2 and/or TM3 of the same BCRP monomer. The exact interaction partners of the GXXXG motif and its role in BCRP dimerization remain to be determined.
Leu or Glu substitution of Gly553 led to rapid degradation and retention in the endoplasm reticulum of BCRP , suggesting that Gly553 is critical for plasma membrane targeting and proper folding of the transporter. In the models shown in Fig. (2), Gly553 is located in the intracellular loop connecting TM4 and TM5. This intracellular loop contains a coupling helix interacting with the NBD of the opposite BCRP monomer. Whether mutations of Gly553 affect the communication between the NBD and the MSD and thus the activity and overall integrity of BCRP remains to be investigated.
Ala substitution of Cys592 and Cys608 has been shown to impair plasma membrane targeting and function of BCRP; however, the double mutation (C592A/C608A) partially restored plasma membrane expression . When the three Cys residues Cys592, Cys603 and Cys608 were mutated to Ala simultaneously, the expression level of the triple-mutant in HEK cells was drastically decreased with no measurable activity . In contrast, Liu et al.  showed that mutations of the three Cys residues did not affect BCRP expression and transport activity in Sf9 insect cells. The recent study by Shigeta et al.  also demonstrated that the triple-mutant C592S/C603S/C608S in PA317 cells retained substantial activity with some alteration in substrate recognition, although its level of expression was significantly decreased and its subcellular distribution was altered. The study by Wakabayashi et al. provided evidence that BCRP lacking the disulfide bond formed by Cys592 and Cys608 is misfolded and the misfolded protein undergoes ubiquitin-mediated protein degradation in proteasomes . Overall, these results, although not all consistent, appear to suggest that these Cys residues are important for structural integrity and proper folding of BCRP. Consistent with this, Cys-less BCRP can be well expressed in Sf9 insect cells, but lacks transport activity . This precludes the use of Cys-less BCRP for structure-function analyses.
In the third category, mutations cause alterations in chemical modifications such as N-linked glycosylation or disulfide bond formation in BCRP. The only N-linked glycosylation site in BCRP is Asn596 which is located in the extracellular loop 3 connecting TM5 and TM6 Fig. (3). Gln substitution of Asn596, but not the other two Asn residues (Asn418 and Asn557), produced a smaller molecular weight BCRP band on SDS-PAGE due to the lack of glycosylation [92–93]. The lack of glycosylation did not appear to affect plasma membrane targeting and transport activity of BCRP; however, glycosylation may stabilize BCRP because disruption of N-linked glycosylation has been shown to enhance ubiquitin-mediated proteasomal degradation of the transporter . Several studies have elucidated the role of Cys603 in dimer/oligomer formation of BCRP [82–84]. Ala substitution of Cys603 led to migration of BCRP bands to a size corresponding to BCRP monomers, but had no effect on plasma membrane expression and function, suggesting that Cys603 participates in intermolecular disulfide bond formation. An earlier mutagenesis study  suggests that Cys592 and Cys608 could form an intramolecular disulfide bond; however, the recent studies [85–86] indicate that Cys592 and Cys608 may participate in intermolecular disulfide bond formation. Therefore, contrasting data exist in the present literature with respect to the role of the three extracellular Cys residues in intra- or intermolecular disulfide bond formation.
In the fourth category, mutations do not have major effects on both plasma membrane expression and function of BCRP. Such mutants include K473A and H630X, suggesting that these residues are likely not critical for expression and function of BCRP.
Much progress has been made in understanding the in vitro and in vivo function of BCRP, such as the identification of an increasing number of substrates and inhibitors as well as the role of BCRP in the absorption, excretion and distribution of substrate drugs, xenobiotics, and endogenous substrates. BCRP is also highly expressed in the side population of stem cells, likely providing a protective function for stem cells. Although BCRP has long been recognized to play an important role in multidrug resistance of cancer cells, it has not been possible to circumvent clinical drug resistance in cancer patients by the use of BCRP inhibitors. This awaits the further development of more potent and highly selective BCRP inhibitors and the appropriate design of clinical trials in the future. In addition, we believe that a successful translation of scientific findings about BCRP into the clinic would require a full understanding of how BCRP acts to transport drugs and xenobiotics. For example, how BCRP recognizes its substrates and what determines its broad substrate specificity? How a substrate is transported by BCRP and how transport of substrate is coupled with ATP hydrolysis? Is the minimal functional unit of BCRP a homodimer or homooligomer, and how the dimeric or oligomeric state of BCRP might affect its activity? How BCRP monomers are organized to form a functional homodimer or higher order oligomer? At present, we virtually do not know any answers to these questions, and studies to address these questions are desperately needed. Mutagenesis studies in recent years have begun to identify amino acid residues, particularly those in the membrane-spanning domain, that are critical in determining substrate specificity, transport activity, and proper folding of BCRP. Homology models have also been developed to interpret the existing data and may provide guidance for further biochemical studies. Further biochemical studies, such as site-direct mutagenesis of amino acid residues in critical protein domains, identification of residues in the putative drug binding sites by photo-affinity labeling and mass spectroscopy analysis, and mapping the contact interface between BCRP monomers in a dimer or oligomer and between the MSD and the NBD, in turn could help refine and validate these homology models. Computational methods such as 3D-QSAR and molecular dynamic simulations using biochemical data as constraints may further refine the models, and help predict conformational changes associated with the transport process and characterize drug interactions in the putative drug binding cavity. In the absence of high resolution 3D structures of BCRP, the refined models may help rational design of substrate drugs or inhibitors that are highly selective and potent for BCRP. Nevertheless, in order to gain a clear understanding of transport mechanism, the ultimate goal is to resolve the structure of BCRP at atomic resolutions that is supported by most of the biochemical data. This is a big challenge. However, with the 3D structures of P-gp  and 2D structures of BCRP  resolved at relatively high resolutions, the possibility to achieve this goal is high.
We acknowledge financial support from the National Institutes of Health grant GM073715 (to Q.M). M.F.R. was supported by a VIP fellowship and in part by a grant 081406/Z/06/Z from the Wellcome Trust.
CONFLICT OF INTEREST
The authors state no conflict of interest and have received no payment in preparation of this manuscript.