ATP binding cassette transporters are ubiquitous integral membrane proteins that actively transport ligands across biological membranes, a process critical for most aspects of cell physiology. These proteins are important clinically and economically. Their dysfunction underlies a number of human genetic diseases, and the ability of some to pump cytotoxic molecules from cells confers resistance to antibiotics, herbicides, and chemotherapeutic drugs. Recent structure analyses interpreted in light of a large body of biochemistry has resulted in the ATP-switch model for function in which the paired nucleotide binding domains switch between an ATP-dependent closed conformation and a nucleotide-free, open conformation to drive the translocation of ligand.
ABC Transporters are Ubiquitous and Clinically Important
ATP binding cassette (ABC) transporters are found in all species and many species have lots. There are, for example, 28 in Saccharomyces,58 in Caenorhabditis, 51 in Drosophila,129 in Arabadopsis,and the 69 ABC transporters in E. coli account for almost 5% of its genomic coding capacity (27). Typically, they transport ligands across cellular lipid membranes, which is critical for most aspects of cell physiology, including the uptake of nutrients and elimination of waste products, energy generation, and cell signalling.
There are 48 ABC transporters in humans and mutations in many are at the root of genetic disorders including a bleeding disorder (2) and a number of eye (39) and liver diseases (20), all of which are caused by the failure to export a specific ligand across a lipid bilayer (Table 1⇓; Refs. 10, 14). The normal function of some human ABC transporters is to secrete from cells cytotoxic compounds (dietary cytotoxics and therapeutic drugs). These transporters (P-glycoprotein, BCRP, and MRP1) are highly expressed in the gut, liver and kidneys where they restrict the bioavailability of administered drugs. P-glycoprotein and BCRP in particular are also expressed in the epithelia of sensitive tissues [for example, the brain and placenta (22, 50)] and in stem cells (66), where they perform a barrier function (48). There are also a few atypical ABC proteins. In this category, I have included the cystic fibrosis transmembrane regulator (CFTR) and sulphonyl urea receptor (SUR) because, although they have the requisite domains to ostensibly function as transporters (see below), CFTR is a channel for chloride ions (55) and SUR is a regulator of a potassium channel to which it is directly coupled in a hetero-octameric complex (19). These proteins share this category with others that are derived from the energy-converting domains, the nucleotide binding domains (NBDs), but which have evolved to couple the energy released by binding and hydrolysis of ATP to processes other than transbilayer transport. These include proteins involved in chromatin organization (13), DNA repair, telomere maintenance (15, 44), and mRNA trafficking through the nuclear pore (23). The molecular mechanism described below for ATP catalysis by the NBDs and its coupling to transport is also likely to describe the conformational changes involved in these processes.
ABC Transporters Have Four Core Domains
To achieve export [import appears to require an additional periplasmic binding domain (PBP)], ABC transporters require a minimum of four domains (FIGURE 1⇓). Two transmembrane domains (TMDs) form the ligand binding sites and provide specificity, and two NBDs bind and hydrolyze ATP to drive the translocation of the bound ligand. The NBDs, but not the TMDs, are homologous throughout the family and have several characteristic motifs including the Walker A and B motifs common to many nucleotide binding proteins and others like the ABC signature, stacking aromatic D, H, and Q loops, which are unique to the family. The four domains can be encoded on different polypeptides, as in the vitamin B12 importer BtuCD from E. coli,or can be fused together in any combination (27), including as a single four-domain protein such as found in P-glycoprotein.
The ATP Catalytic Cycle
A considerable body of biochemical and genetic studies has been amassed on many different ABC transporters from different species. Depending on the ligand, the host organism, and also protein function (CFTR, for example, allows careful kinetic study of how the function of the NBDs controls the opening and closing of the chloride channel formed by the TMDs), different systems have been able to provide distinct insights into how these proteins work. Mammalian cells can be selected for overexpression of P-glycoprotein by cultivation in the presence of cytotoxic drugs. This facilitated purification of the protein and also characterization of its ATP catalytic cycle more than 10 years ago (51). Both nucleotide binding domains were shown to bind ATP (1), and both are essential for function as mutation of either eliminated drug transport from cells (4) and ATPase activity (32), suggesting that the two sites were coupled allosterically. Hydrolysis of ATP was shown not to involve transient phosphorylation of the protein, and measurable catalysis was inhibited by orthovanadate [Vi(57)], which occupies the position of the liberated γ-phosphate in a ternary complex with P-glycoprotein and ADP, mimicking the transition state intermediate. These initial data were interpreted as evidence for a two-step mechanism with each catalytic event coupled to distinct stages (51). However, structural data has necessitated reinterpretation of the biochemistry data, resulting in a fundamental change to the perceived transport model. First, the structures of isolated NBDs were solved [the structure of HisP was reported in 1998 (18)], then dimers of NBDs in different physiological conformations, and finally of ABC transporters complete with NBDs and TMDs.
Dimeric NBD Structures Hint at the Molecular Mechanism
Purified NBDs isolated from their cognate TMDs are generally monomeric. Those that are homodimeric within the transport complex in vivo readily dimerise to hydrolyse ATP, but the association is transient (21, 41, 43, 64). Structures of NBDs in their catalytic conformation have therefore been difficult to achieve. However, structures have been determined for the NBD from the hemolysin exporter HlyB (65) and MJ0796 (52) in the presence of ATP, but only after mutation of the NBD to abrogate hydrolysis. These “closed NBD dimer” structures show the NBDs closed around two molecules of ATP at the interface (FIGURE 2A⇓). The presence of two molecules of ATP is significant because it implies that the binding pockets act in concert and that the energy derived from ATP binding and the formation of the closed NBD dimer is used in a single step of the transport cycle. Similar structures have been observed with GlcV (60), MalK (7), and the nontransporter ABC NBDs Rad50 (15) and MutS (44); and with the conserved motifs of the NBDs involved in important protein-ATP or protein-protein contacts, it is clear that this is a physiological state (FIGURE 2B⇓). There is also considerable evidence from studies with P-glycoprotein and CFTR to suggest that, in vivo, the TMDs of P-glycoprotein share an interface at the intracellular face of the membrane (34, 54) and that the NBDs interact directly (31, 40). The nucleotide binding pockets are therefore composite sites with both NBDs contributing to each site. In the dimeric state ATP is bound predominantly by the core subdomain, and the Walker A and B motifs, the stacking aromatic (as it has become known for its π-π interaction with the adenosine ring of the ATP), and the His-loop histidine all contact the bound nucleotide. The Q-loop glutamine also contacts the γ-phosphate of ATP via a water molecule. The Q-loop motif has been extended in Figure 2B⇓ where it is colored blue to illustrate that it originates from a helix on the top surface of the NBD where it interdigitates with the intracellular loops from the TMDs (see below). It also extends COOH terminally to connect the core subdomain of the NBD to the α-helical subdomain. The α-helical subdomain contains the ABC signature that contacts the second ATP (colored gray in FIGURE 2B⇓), which is bound predominantly by the core subdomain of the associated blue NBD. Finally, the D-loop of the core subdomain may form a hydrogen bond with the back bone of the Walker A motif of the blue NBD.
Conformational Change at the NBDs
The position of the Q-loop is important because it is in a position to couple the ligand-binding sites within the TMDs to the ATP-binding sites of the NBDs. It also links the two subdomains of the NBD and therefore the two composite ATP-binding sites offering a molecular explanation for the observed allostery between the ATP catalytic sites (8, 28, 43, 64). The Q-loop has been suggested to move as ATP is bound consistent with a coupling role; however, this conclusion is drawn from comparison of monomeric nucleotide-free and dimeric ATP-bound NBDs (52, 65). Only in the case of MalK, the NBD of the maltose transporter from E. coli , are nucleotide-free NBDs dimeric in solution (7). This is because MalK has an additional regulatory domain that maintains an interface in the nucleotide-free form (FIGURE 3⇓). Comparison of these two MalK structures suggests that the domains make a major conformational change as ATP is bound, likened to the action of tweezers. However, in common with the other structures of isolated NBDs, this interpretation does not take into account the influence of the TMDs on the extent of the conformational change at the NBDs. In BtuCD and HI1470/1, where each domain is on a separate polypeptide, the complex remains intact in the absence of ATP and, given that isolated NBDs only interact transiently during ATP hydrolysis, maintenance of the transport complex is mediated by NBD-TMD contacts.
The precise nature of the open and closed NBD dimer conformations, therefore, awaits structure determination of the same full-size transporter in different physiological conformations in the presence and absence of ATP. However, circumstantial but persuasive corroboration that the Q-loop is the site of major conformational change and coupling between the energy-converting domains and the ligand-binding moiety of ABC proteins is found in the SMC proteins in which the Q-loop is the site of insertion of large-coiled coil domains, which extend from the NBDs to bind chromosomes (13).
Structural Data from Full-Size Transporters
At present, there are three complete structures in the database following the recent retraction of three distinct structures of MsbA (6). The domain arrangement for the remaining three full-length ABC transporters, is shown in Figure 4A⇓. All are of bacterial origin; two are homologous PBP-dependent importers, BtuCD from E.coli , which transports vitamin B12 (30), and HI1470/1 from Haemophilus influenzae,which transports a metal-chelate (45), and the third is a drug exporter, Sav1866 from Staphlococcus aureus (9). The four core domains of the importers are on separate polypeptides and form a complex of homodimeric TMDs and homodimeric NBDs. Sav1866 is a “half transporter” with one NBD and one TMD per polypeptide, which functions as a homodimer. The homologous TMDs of the vitamin B12 and metal-chelate importers, BtuC and HI1471, respectively, and the non-homologous TMD of Sav1866, cross the membrane as α-helices, but the similarity ends there. BtuC and HI1471 span the membrane 10 times per domain, whereas Sav1866 spans the membrane 6 times per domain. The intracellular loops (ICLs) of BtuC and HI1471 are short with the NBDs close to the inner leaflet of the membrane, whereas the ICLs of Sav1866 are much longer, extending 40 Å into the cytoplasm as helical extensions of the transmembrane helices.
The structures of BtuCD and HI1470/1 are similar but not identical. Essentially, they are both crystallized in the absence of nucleotide (BtuCD has tetra-vanadate in both nucleotide binding pockets, but the NBDs are not in the closed NBD conformation expected when ATP is bound). At the centre of the TMD-TMD interface, an aqueous chamber is evident, which is open at the extracellular face of the protein but tapers to close toward the intracellular face. The structure of BtuF (the PBP for BtuCD) in complex with vitamin B12 can be docked onto the BtuCD structure in a plausible, in silico experiment (5), suggesting that the conformation of the transport complex is in an outward configuration appropriate for binding of the BtuF-B12 complex from the periplasm. The TMDs of HI1470/1 also describe an aqueous chamber in what would be the plane of the membrane, but, in contrast to BtuCD, the chamber is open at the intracellular face of the protein, suggesting an inward-facing aspect for release of the ligand intracellularly. These two structures may therefore represent good models of the conformational changes expected in the TMDs during the transport process. However, the very subtle differences at the NBDs, with neither structure in the closed NBD conformation, means that hypotheses of how conformational changes in the TMDs are controlled by the ATP catalytic cycle, and vice versa, must remain speculative.
The domain organization of the importers is not shared with the multidrug resistance (MDR) class of exporter. The three structural models each have a plausible TMD-TMD interface and a NBD-NBD interface capable (or within a small conformational change) of coordinating nucleotide. Indeed, Sav1866 has ADP and BtuCD has tetra-vanadate mimicking nucleotide in an appropriate position. However, comparison of the interdomain architecture of the PBP-dependent importers with Sav1866 suggests that they may not be compatible. The TMD-TMD interface of Sav1866 is rotated with respect to the NBD-NBD interface, such that each TMD contacts both NBDs. This is readily illustrated by the footprints that the ICLs of one TMD make on the surface of the NBD dimer ( FIGURE 4C⇑). In Sav1866, ICL1 of the red TMD spans the composite ATP-binding site formed by the core subdomain of the gold NBD and the α-helical subdomain of the slate blue NBD (the footprint is shown in green). ICL2 of the red TMD of Sav1866 makes contact exclusively with the slate blue NBD and not to the orange NBD to which it is fused in a single polypeptide. In the BtuCD and HI1470/1 structures, each TMD contacts only one NBD (illustrated in FIGURE 4C⇑ by the patchwork coloring of the surface of the NBD-NBD dimer of all residues within 4 Å of the ICLs of the blue TMD). Assessing the veracity of each of these distinct arrangements with the current information is difficult. There is no biochemical evidence that the TMDs of Sav1866 make contact with an NBD other than the one to which it is attached covalently. Also, when the different domains of the MDR class P-glycoprotein were expressed as individual proteins, NBD1 only interacted with TMD1, and NBD2 only interacted with TMD2 (33), suggesting that, if the TMDs interact with both NBDs during the transport cycle, it may only be transient and perhaps only within the complete, four-domain transport complex.
The helical bundles of the TMDs of the PBP-dependent importers and Sav1866 are also distinct. In BtuCD and HI1470/1, the 10 α-helices of each TMD form a single bundle (FIGURE 4A⇑), and although the two bundles share an interface in the membrane, they presumably fold and cross the membrane independently of each other. As illustrated by a different view of Sav1866 in Figure 4B⇑, there are also two helical bundles in the TMDs of Sav1866, but these are composite bundles formed by the interaction of two TM helices from one TMD (helices 4 and 5 ) with four TM helices from the other (helices 1′, 2′,3′, and 6′). Such a conformation for Sav1866, very different from the two examples of the PBP-dependent importers, will have to be tested to confirm its physiological veracity.
Could the PBP-Importers and the Drug Exporters Have Distinct Transport Mechanisms?
Sav1866 is crystallized with two molecules of ADP at the NBD-NBD interface. However, the NBD dimer interface of the Sav1866 structure closely resembles the ATP-bound conformations of bacterial NBDs crystallized in absence of their cognate TMDs (9). It is difficult to see how ADP can exit the Sav1866 structure unless there is further conformational change. Although this could, perhaps, be induced by binding of ligand, it may be better to consider Sav1866 as an ATP-bound model, but is it feasible that the protein moves between the Sav1866-like conformation (mimicking the ATP-bound state) to conformations more consistent with the two structures of the PBP-dependent importers? This would involve an extremely large change at the NBD-TMD interface, perhaps requiring uncoupling of the NBD from the TMD before reengagement. At much lower resolution but working with a full-size ABC transporter, evidence for large-scale conformational changes in the TMDs following ATP binding, and again after hydrolysis, has been obtained for P-glycoprotein (47). This evidence would seem to imply conformational change at the NBDs on the scale suggested by the MalK studies described above, but equally may be due to subtle changes in conformation at the NBDs transduced over a large distance (the intracellular loops of Sav1866, which is homologous with P-glycoprotein extend at least 40 Å from the lipid bilayer). In truth, there is insufficient data to describe, in detail with confidence, conformational changes that occur during the transport cycle, but at present it would be best to regard the MDR class of ABC exporters and the PBP-dependent importers as having different molecular mechanisms for ligand translocation through the membrane. The function of ABC transporters may be best thought of as the coupling of two distinct cycles: an ATP catalytic cycle at the NBDs and a ligand transport cycle at the TMDs. The conformational changes at the NBDs, switching between a nucleotide-free open conformation and an ATP-bound closed conformation, are likely to be conserved throughout the family. However, if the three available structures are physiologically relevant, the different classes of TMDs are likely to have evolved different mechanisms to couple the conformational changes of the NBD dimer to ligand translocation.
Mechanism of Action of an MDR Exporter
Extensive biochemical studies of the ATP catalytic cycle linked to pharmacological studies of drug affinity make P-glycoprotein arguably the best characterized ABC transporter. These data, now interpreted in light of the structural data described above, are the basis for the ATP-switch model to explain the mechanism of drug transport via P-glycoprotein (12). Essentially, transporters must cycle between high-and low-affinity states for ligand on different sides of the membrane. The ATP switch mechanism describes how these states are coupled to the ATP catalytic cycle in a way that is consistent with the structural data available (FIGURE 5⇓).
Ligand binds to the TMDs in the high-affinity open NBD dimer conformation, inducing increased affinity for ATP.
In the nucleotide-free form, ABC transporters have a high affinity for ligand [this has been clearly demonstrated for drug transporters P-glycoprotein and LmrA (38, 49, 58)]. Binding of ligand to the TMDs causes a conformational change in the NBDs (24, 29, 35, 36, 42, 53), transduced via the ICLs of the TMDs, which interdigitate with the NBDs. The close proximity of the ICLs with the Q-loop and Walker A motifs suggests that the binding of ligand could directly influence the affinity of the NBDs for ATP, as suggested by structural studies of some crystals of HlyB and LolD with unusual conformations of the Walker A motif, which would preclude binding of ATP (59, 65), although as a mechanism this is by no means clear. Evidence for an increase in affinity for ATP in the presence of ligand has been difficult to measure and is largely indirect. For example, two-point mutations in TMDs of the maltose importer from E. coli uncouple ATP catalysis from the binding of its PBP-maltose complex and cause a 30-fold increase in affinity for ATP (36). An increase in affinity for ATP following ligand binding has also been measured for the DNA repair enzyme MutS (25) and for P-glycoprotein, albeit subtly (46).
ATP binding induces formation of the closed NBD dimer, which in turn induces a large conformational change in the TMDs sufficient to translocate ligand.
Structural data show two molecules of ATP at the NBD dimer interface, implying that they act in concert at a single step. Formation of the NBD dimer closed around the ATP has been calculated to generate a significant amount of free energy (52), and several distinct measurements suggest that this available energy is used to reconfigure the TMDs (24, 35, 36, 53). The extent of this conformational change is large and can be visualized at low resolution for P-glycoprotein (47). This may involve breaking interactions between TM α-helices and formation of new contacts with different partners. Not surprisingly, this rearrangement can alter both the position and affinity of the ligand binding site. In P-glycoprotein, binding of nonhydrolysable analogs of ATP is sufficient to reduce the affinity for the drug vinblastine and so must effect transport of this drug at this stage (38). In CFTR, ATP binding in the absence of hydrolysis is sufficient to open the chloride channel (61), and in MutS conformational changes associated with ATP binding allows the repair complex to form (11, 25).
ATP hydrolysis initiates dissolution of the closed NBD dimer.
The transient nature of isolated NBD dimers induced by ATP suggests that hydrolysis is an inexorable consequence of closed NBD dimer formation. The mechanism of hydrolysis is contentious with base catalysis proposed for LolD (52) and substrate-assisted catalysis proposed for HlyD (63). Whichever mechanism, it is clear that hydrolysis destabilizes the closed NBD dimer. For some ABC transporters like P-glycoprotein, hydrolysis of both ATPs is necessary for completion of the transport cycle (4), and the ATPs are hydrolysed nonsimultaneously (16, 17, 51, 56). The two ATP binding pockets of P-glycoprotein have been shown to be biochemically asymmetric, but it is not yet clear whether this is a cause of nonsimultaneous hydrolysis or the result of it (56). In other ABC transporters, such as in MRP1 and CFTR, hydrolysis of only one ATP may be sufficient to drive the protein through the conformational cycle (3, 62). The post-hydrolytic state can be shunted into a stable conformation once phosphate has left the complex, as it can be replaced by vanadate. This form of the protein, which may mimic the ADP.Pi state, has a conformation distinct from the ATP-bound and nucleotide-free forms (34, 47) and may be coupled to transport from different drug binding pockets in Pgp (26, 49), although for vinblastine the protein remains in the low-affinity state (37).
Pi then ADP is released to complete the transport cycle and restore the protein to a high-affinity state for ligand.
Following hydrolysis, the liberated phosphate is released from the protein. Again two mechanisms have been postulated for this based on the analysis of the structures of HlyB [via exit channels (65)] and LolD [following electrostatic repulsion of the two NBDs (52)]. The quick release of phosphate, which allows the protein to be trapped with ADP and vanadate, correlates well with the restoration of high-affinity vinblastine binding to P-glycoprotein, indicating that the conformational change associated with phosphate release is also coupled to the ligand binding sites (38). Affinity of ABC tranporters for ADP is low, and ADP is unable to stabilize the dimeric interaction of isolated NBDs (which is one reason for doubting the absolute veracity of the ADP bound structure of Sav1866); however, at least for MutS, it appears that ADP remains bound until DNA binding induces an exchange for ATP (11).
The ATP-switch model is the product of biochemical data interpreted in light of recent advances in structure determination for several ABC transporters. The model is divided into four steps, or four conformational changes; the first associated with the binding of ligand, then three that make use of the free energy available from protein-ATP and protein-protein interactions associated with ATP binding, ATP hydrolysis, and ADP.Pi release. Any of the latter three steps could be used to transport ligand depending on the nature of the ligand binding site, so P-glycoprotein could potentially translocate different drugs at different stages of the cycle. It is also important to appreciate that each step is closely coupled to the next; for example, formation of the closed NBD dimer is likely to be inexorably linked to ATP hydrolysis, just as the transport of vinblastine by P-glycoprotein appears to be.
Although some important details remain to be elucidated (for example, the precise nature of the ligand binding sites), this general mechanism of the ATP switch allows minor modifications to explain the idiosyncratic behavior of different ABC transporters with different classes of TMD or ABC NBDs coupled to functions other than transport.
- © 2007 Int. Union Physiol. Sci./Am. Physiol. Soc.