Recently, the crystal structure of a murine ABCB1 homologue has been published [16], but it still leaves us some way Keywords ABC transporter; ABCC1; ABCG2; homology modelling; MsbA; mul
Trang 1Multidrug efflux pumps: The structures of prokaryotic
ATP-binding cassette transporter efflux pumps and
implications for our understanding of eukaryotic
P-glycoproteins and homologues
Ian D Kerr1, Peter M Jones2and Anthony M George2
1 School of Biomedical Sciences, University of Nottingham, UK
2 Department of Medical and Molecular Biosciences, Institute for the Biotechnology of Infectious Diseases, University of Technology Sydney, Australia
Introduction
The spectre of multidrug resistance (MDR) haunts
many a clinical intervention Most pertinent to this
minireview is the resistance of leukaemias and many
solid tumours to anticancer chemotherapy [1] Encoded
within the human genome are three ATP-binding
cassette (ABC) transporters which have been shown to
be able to contribute towards the MDR phenotype
These three proteins, ABCB1 (P-glycoprotein), ABCC1
(multidrug resistance protein 1; MRP1) and ABCG2
(breast cancer resistance protein; BCRP) have been
the focus of numerous biochemical studies since their
original cloning and identification [2–4] Among the
advances have been the demonstration of multiple pharmacologically distinct binding sites for transported drugs, [5–7], the determination of kinetic parameters for ATPase activity [8–10], and the establishment of high-level expression systems amenable to purification and structural work [11–13]
In spite of this, the structure at high resolution of any of these three proteins remains unknown, although all three have been imaged using low- to medium-reso-lution electron microscopy (EM) [12,14,15] Recently, the crystal structure of a murine ABCB1 homologue has been published [16], but it still leaves us some way
Keywords
ABC transporter; ABCC1; ABCG2; homology
modelling; MsbA; multidrug pump;
P-glycoprotein; Sav1866; structure; transport
mechanism
Correspondence
I D Kerr, School of Biomedical Sciences,
University of Nottingham, Queen’s Medical
Centre, Nottingham NG7 2UH, UK
Tel: +44 115 8230122
E-mail: Ian.kerr@nottingham.ac.uk
(Received 22 July 2009, revised
1 October 2009, accepted 22 October 2009)
doi:10.1111/j.1742-4658.2009.07486.x
One of the Holy Grails of ATP-binding cassette transporter research is a structural understanding of drug binding and transport in a eukaryotic multidrug resistance pump These transporters are front-line mediators of drug resistance in cancers and represent an important therapeutic target in future chemotherapy Although there has been intensive biochemical research into the human multidrug pumps, their 3D structure at atomic resolution remains unknown The recent determination of the structure of
a mouse P-glycoprotein at subatomic resolution is complemented by struc-tures for a number of prokaryotic homologues These strucstruc-tures have pro-vided advances into our knowledge of the ATP-binding cassette exporter structure and mechanism, and have provided the template data for a num-ber of homology modelling studies designed to reconcile biochemical data
on these clinically important proteins
Abbreviations
ABC, ATP-binding cassette; CFTR, cystic fibrosis transmembrane conductance regulator; EM, electron microscopy; ICL, intracellular loop; MDR, multidrug resistance; NBD, nucleotide-binding domain; TM, transmembrane; TMD, transmembrane domain.
Trang 2short of the goal of an atomic resolution (3 A˚ or
better) structure that could allow rational
interpreta-tion of the MDR phenomenon in ABC transporters
Although the secondary structure and
membrane-span-ning topology of the three proteins differ, a common
‘functional core’ may well exist because all three
com-prise a pair of nucleotide-binding domains (NBDs),
which are well conserved across the ABC transporter
family [17], and two transmembrane domains (TMDs)
with six membrane-spanning a helices (although as
dis-cussed later, ABCC1 contains an additional N-terminal
domain)
This functional core is represented in a number of
pro-karyotic ABC proteins including Sav1866, MsbA and
LmrA [18–20] Sav1866 and MsbA are now available at
medium to high resolution [21,22], and the third (LmrA)
may be a functional homologue of ABCB1 (see below)
[18] This prompts the main questions that this minireview
seeks to address: to what extent can the structural data
for Sav1866 and MsbA be used as templates for
MDR-type ABC exporters in general? Do these data map the
conformational states that a dynamic ABC transporter
adopts during its catalytic cycle? In this minireview,
there-fore, we use the high-resolution templates of prokaryotic
MDR exporters to look over the horizon to the
eukary-otic MDR pumps, as typified by ABCB1, ABCC1 and
ABCG2 In the final part of the minireview we ask to
what extent the recent structure of murine ABCB1A [16]
will replace the current homology models of ABCB1
Throughout our discussion it is worth recalling that the
prokaryotic ABC exporters mentioned are homodimers,
as is the eukaryotic MDR pump ABCG2, whereas
ABCB1 and ABCC1 comprise a single polypeptide
A crystallization of the recent history
of ABC exporters: when seeing is
believing
A brief account of the complete structures of ABC
exporters can be summarized as belonging to two
phases, pre-Sav1866 (2001-2006) and post-Sav1866 (2006-present) The first phase commenced with the publication of the first complete ABC crystal structure
in 2001, namely the MsbA lipid A half-transporter from Escherichia coli The bacterial homodimeric MsbA is a close homologue of human ABCB1, the eukaryotic MDR pump that continues to attract the most interest and study The first MsbA structure was followed by two further MsbAs reported by the same group in 2003 and 2005, and these were at higher reso-lution and in different orientations During the pre-Sav1866 stage, these MsbA exporter structures were called into question [23,24] because there were discrep-ancies between them and other structural and bio-chemical data on complete ABC transporters and isolated dimeric NBDs [25–30]
These anomalies were not fully understood until late
in 2006 when Kaspar Locher’s group published the structure of the Sav1866 protein from Staphylococ-cus aureus, solved as a homodimer at 3.0 A˚ resolution, with outward-facing TMDs and a canonical NBD dimer with ADP sandwiched between Walker-A and Signature motifs [21,31] To date, this new structure has proved to be ‘bullet-proof’, with a seemingly convincing tertiary scaffold The Chang group has since realized that his crystallographic data-processing package had led him to interpret the MsbA data incorrectly, result-ing in the retraction of these papers The reinterpreted data (summarized in Table 1) has been republished [22] and this is discussed below Much has been written sub-sequent to these retractions and the most pertinent comments appear in Petsko [32], with its cautionary reminder that structural data should be consistent with the majority of the available biochemical data
The six structures published for prokaryotic ABC exporters are summarized in Table 1 Resolution is highest for the ADP-bound form of Sav1866, and decreases to < 5 A˚ for the MsbA structures in the unliganded state (unliganded with respect to nucleotide substrate) For the majority of the MsbA structures,
Table 1 Structural data for bacterial homologues of eukaryotic ABCB multidrug pumps.
ABC, ATP-binding cassette; NBD, nucleotide-binding domain.
a
Soaked into ADP-containing crystals.bContains side chain atoms All other MsbA structures listed are Ca trace only.cAlthough the NBDs are closed in this conformation, there is no apposition of Walker-A motif with Signature motif.
Trang 3the level of resolution is not so great as to allow
unambiguous determination of side chain orientation
Simulated annealing approaches to ‘extend’ the
resolu-tion of MsbA structural data have been described
pre-viously [33] and such efforts may also be applied to
the revised data However, the important factors
perti-nent to comparisons with eukaryotic ABC MDR
pumps are: (a) the way in which the two half
trans-porters come together to form a dimeric unit, because
this reveals possible domain organizations for
eukary-otic MDR pumps; and (b) the assignment of secondary
structure and transmembrane topology
In the second of these factors the structures show
agreement They each describe six transmembrane
(TM) helices per TMD, followed a classical ABC
transporter NBD structure (Fig 1) Of the six TM
helices in each monomer, five are extended
signifi-cantly compared with the presumed membrane
thick-ness of 35 A˚ (only helix 1, which is preceded by a
perpendicular ‘elbow’ helix demarking the membrane
surface is not extended in this way) The other five
helices have an average length of 43–44 amino acids,
a span of almost 70 A˚, and thus make a substantial
contribution to the cytoplasmic structure of the
protein The five long helices provide three main con-tact points to the NBD The extension to TM6 has a direct covalent linkage into the NBD itself, whereas the linker regions between TM helices 2 and 3 and between TM helices 4 and 5 provide noncovalent interactions with sites on the NBD These latter two linker regions, called intracellular loops 1 and 2 (ICL1 and ICL2) can both be represented structurally
as a pair of antiparallel a helices, connected by an 8–
12 amino acid stretch (also a-helical in secondary structure and called ‘coupling helices’ 1 and 2) form-ing the bottom of the loops and makform-ing significant contacts with residues in the NBD [21]
The unexpected finding of the Sav1866 structure, seen later in the revised MsbA structures, is that the TMD of one ABC protomer contacts the NBD of the second protomer and vice versa (Fig 1) Specifi-cally, the second coupling helix, between TM helices
4 and 5, exclusively contacts the NBD of the other Sav1866 molecule in the dimeric arrangement The contact surfaces on the NBDs include residues C-ter-minal to the Walker-A motif, and to a conserved motif (‘X-loop’) immediately N-terminal to the ABC transporter Signature sequence [21] Also of note, TM helices 4 and 5 are splayed away from one of the TMDs and form the majority of their interhelical contacts with TM helices from the opposite protomer This cross-protomer interaction clearly suggests a mechanism for co-operativity as exemplified by ABC transporters, and the direct contacts of the TMDs with the NBDs also hint at how transported sub-strate–TMD interactions could be communicated to the ATP-binding pockets in the NBDs Thus far, pro-karyotic ABC import systems do not show this cross-protomer interaction, and thus we should be cautious about the extent to which we interpret bacterial MDR structural data and apply it to eukaryotic ABC exporters
First, we need to understand the degree to which the prokaryotic ABC proteins can function as multidrug pumps (discussed below), and second, we need to vali-date some of the key structural findings The latter has been provided in a cross-linking study of cysteine-free human ABCB1 [34] in which it has been demonstrated that a cysteine residue introduced into the loop between TM helices 8 and 9 (equivalent to TM2 and TM3 in either Sav1866 protomer) can be cross-linked
to a cysteine residue introduced just C-terminal to the Walker-A motif of the opposite NBD This is convinc-ing supportconvinc-ing evidence that ABCB1, and possibly other eukaryotic ABCB proteins, deploy a similar
‘domain swapping’ and TMD–NBD interface to that observed in the Sav1866 structure
Fig 1 The cross-protomer interaction of Sav1866 The intracellular
portion of a Sav1866 homodimer is represented in cartoon fashion;
the two Sav1866 molecules are coloured yellow and red, and blue
and green Bound nucleotide is rendered in grey space-filling
repre-sentation The cross-protomer (‘domain swapping’) interaction is
illustrated by the intracellular loops of one TMD (blue) interacting
primarily with the NBD of the opposite protomer (yellow).
Trang 4Function of the prokaryotic ABC
exporters
For the bacterial exporters to be used as structural
templates for understanding eukaryotic MDR pumps
also requires demonstration that they are sufficiently
similar in terms of function This is pertinent because
the physiological relevance of Sav1866 is not clear,
and the function of MsbA is in the transport of lipid
A (a component of the lipolysaccharide outer
mem-brane) [20,35] Detailed characterization of their
func-tion has been undertaken and provides some evidence
that MsbA and Sav1866 can function as multidrug
pumps For Sav1866, the protein was characterized in
a Lactococcus lactis expression system [19] Inside-out
vesicles, intact cells and proteoliposomes containing
purified, reconstituted protein were used to determine
that the transport substrate specificity of Sav1866
includes the dye Hoescht33342 and ethidium bromide,
and ATPase activities further added verapamil and
vinblastine to the list of compounds with which
Sav1866 interacts [19] For MsbA, similar studies
argue that the protein is a functional homologue of
multidrug pumps – the protein can confer resistance to
ethidium bromide and transport this cation, as well as
another DNA-intercalating agent Hoescht 33342
Furthermore, membranes containing MsbA have an
ATPase activity that is stimulated by daunomycin, and
can interact with a further typical MDR substrate
azi-dopine [36] Intriguingly, not only can MsbA substitute
for LmrA in conferring multidrug resistance on E coli,
but LmrA can restore growth of a MsbA
temperature-sensitive mutant, suggesting functional
complementar-ity [36] Moreover, LmrA can substitute for human
ABCB1 in transfected tissue culture cells as a MDR
determinant [18] Thus, LmrA, MsbA and Sav1866,
irrespective of their physiological roles, all interact
with multiple substrates, many of which are also
sub-strates⁄ modulators of the human multidrug pumps
(Table 2) The ability to function across species
barri-ers also suggests that the study of other eukaryotic
ABC proteins might be advanced by the identification and characterization of bacterial homologues
Conformational transitions observed in prokaryotic MDR pumps
The structural data for MsbA (Table 1 and Fig 2) describe three different configurations of the trans-porter [22] An open, nucleotide-free state was observed for the E coli structure, in which the two NBDs are a significant distance apart (50 A˚; Fig 2A)
A second nucleotide-free state was observed for the Vibrio cholera MsbA in which the NBDs are now closed (Fig 2B, but not in the classical sandwich dimer, i.e there is no Walker-A motif⁄ Signature motif interaction) [22,27,30,31] This ‘closed apo’ structure can be arrived at from the ‘open apo’ structure by a rigid body closure, centred on a hinge in the extracellu-lar loops [22] Formation of the closed, nucleotide-bound structure (as observed in Salmonella typhi MsbA; Fig 2C) requires a further pair of motions to align the NBDs, thus forming the nucleotide sandwich dimer, and a concomitant retraction of TM1 and TM2 from TM3 and TM6, generating an outward facing configuration of the TMDs similar to that observed in Sav1866 [21,22,37] (Fig 2D) These three conforma-tional states are postulated to be intermediates in the functional cycle of MsbA – but their magnitude calls this into question
EPR spectroscopy has the power to give dynamic structural data on membrane proteins, determining inter-residue distances, residue accessibility and confor-mational transitions [38] For MsbA, several studies have attempted to determine the likelihood of the extreme conformations observed, and to verify the structures themselves [39–42] One potential limitation here is the low resolution of the MsbA data which means that accessibilities of residues have to be inferred from Ca positions, an imperfect science Resi-due accessibility studies of the Signature and His-loop regions [42] are only partially consistent with the wide
Table 2 Substrate interaction with the prokaryotic and eukaryotic MDR pumps n ⁄ r, not recorded.
Trang 5‘open apo’ structure because two of the five Signature
sequence residues and one of the His-loop residues are
buried according to EPR quenching data [42], but
rather exposed in the E coli structure [22] The ‘closed
apo’ structure only partially remedies this conflict
Moreover, EPR data in multiple configurations of
MsbA argue that the major conformational changes
occur upon nucleotide hydrolysis, suggesting that the
difference seen between the two ‘apo’ structures is a
reflection of crystallographic conditions rather than
being physiological
Furthermore, Dong et al [43] investigated the
struc-ture of MsbA in liposomes and mapped conformational
changes during the ATPase cycle by EPR analysis of
112 spin-labelled mutants trapped in four intermediate
states, including apo and AMP–PNP bound Notably,
this study found that residues in the N-terminal half of
TM helix 6, (residues 284-296), show very low
accessibil-ity to the aqueous phase in all stages of the transport cycle examined The accessibility data are in excellent agreement with cysteine mutagenesis studies of the equivalent region in ABCB1 [44] (residues 331-343), but are harder to reconcile with EM images of ABCB1 showing a 5–6 nm diameter, 5 nm deep aqueous cham-ber within the membrane open to the cell exterior [45] (although this conformation was obtained under nucleo-tide-free conditions for EM) The inaccessibility of the N-terminal half of TM6 to the solvent is even more puzzling given that in the MsbA accessibility studies [43], C-terminal regions of TM helix 6 (residues 300 and 303) located near the middle of the membrane, are accessible to the bulk solvent in all phases of the trans-port cycle Clearly, the accessibility data are at odds with the MsbA crystal structures (and even the Sav1866 structure), and full reconciliation to experimental data might only be explained by the trapping of other TMD
D C
Fig 2 Conformational states of MDR type ABC exporters The structures of ‘open apo’ MsbA (A), ‘closed apo’ MsbA (B), nucleo-tide-bound MsbA (C) and nucleonucleo-tide-bound Sav1866 (D) are shown as Ca traces with the two monomers in red and blue, respec-tively The bound nucleotide is rendered as green space-filling in the lower two panels.
Trang 6configurations that MsbA⁄ Sav1866 adopt during the
translocation cycle
Lastly, spin–spin distance data obtained on
deter-gent or liposome-embedded MsbA [39] identified both
the magnitude and sign of interdomain distance
changes occurring in the transition from the
nucleo-tide-free to the ADP⁄ Vi-trapped states of MsbA With
all five pairs of residues (located along the axis of the
protein perpendicular to the membrane) the sign of
the distance change was the same as that observed in
the three structural states of Chang, and the
magni-tudes of distance changes for three of the four pairs of
residues for which data were obtained in liposomes
correlates well with the change in distance observed
going from the ‘closed apo’ V cholerae structure to
the vanadate-trapped Salmonella typhi structure
[22,39] Perhaps most pertinently among these data,
the distance between residues within the NBDs
changes by 30 A˚ according to EPR data, this is incom-patible with a transition from the ‘open apo’ structure which would be accompanied by a 50 A˚ distance change In summary, it remains unclear whether either
of the nucleotide-free states of MsbA is physiologically relevant, and the extent of the conformational transi-tions seen remains questionable Indeed, recent com-mentaries have addressed whether the Sav1866 structures could be compatible with these elaborate TMD and NBD movements [21,46]
To what extent can the structures of prokaryotic ABC exporter proteins be used as homology models for
eukaryotic members of the family? Homology modelling is a process that generates a 3D map of a target protein, built against a template of a
Table 3 Sequence identity comparisons of human and prokaryotic multidrug pump nucleotide-binding domains (NBD) The NBDs are defined for this purpose as encompassing residues from 10 N-terminal to the conserved aromatic reside of the A-loop to 10-residues C-terminal to the conserved histidine of the His-loop [17].
ABCB1
NBD1
ABCB1 NBD2
ABCC1 NBD1
ABCC1 NBD2
ABCG2
ABCB1
NBD1
100
ABCB1
NBD2
ABCC1
NBD1
ABCC1
NBD2
Table 4 Sequence identity comparisons of human and prokaryotic multidrug pump transmembrane domains (TMD) The TMDs are defined for this purpose as being from the first amino acid of the first predicted transmembrane (TM) helix to the final residue of the last predicted helix (although the additional five TM helices at the N-terminus of ABCC1 are ignored for this exercise).
ABCB1
TMD1
ABCB1 TMD2
ABCC1 TMD1
ABCC1 TMD2
ABCG2
ABCB1
TMD1
100
ABCB1
TMD2
ABCC1
TMD1
ABCC1
TMD2
Trang 7close homologue, whose X-ray structure is known and
which has mutual sequence similarity [47] Homology
modelling, in essence ‘structural mimicry’, is
particu-larly appropriate for membrane proteins for which
there is a scarcity of high-resolution structures Several
algorithms are available to generate a homology
model, including modeller [48], insight ii [49], and
internet-based servers such as swiss-model [50] and
what if [51] In general, homology modelling
approaches follow a four-step process involving:
tem-plate selection, sequence alignment, model building,
and model optimization and validation
Template selection is made using similarity search
algorithms such as blast or psi-blast from the RCSB
Protein Data Bank (PDB) An accurate alignment
using a program such as clustalw requires a degree
of manual adjustment of the two sequences in order to
accommodate unmatched gaps and insertions for
which there is no equivalent template sequence Many
ABC exporters have the conserved architectural
scaf-fold of two TMDs in 6 + 6 helical bundles, and two
NBDs in a ‘head-to-tail’ configuration However, when
selecting templates for eukaryotic MDR homology
modelling a cautionary tale emerges from sequence
comparisons (Tables 3 and 4) providing the percentage
amino acid identity across the NBD and the TMD to
prokaryotic ABC exporters The data demonstrate that
ABCB1 is much more similar to the prokaryotic ABC
exporters than ABCC1 and ABCG2 In particular,
ABCG2 shows barely any homology to the bacterial
species, particularly in the TMDs, but also in the
NBDs the percentage sequence identity is in the low
20s This is discussed further below
Model building is the easiest of the four stages,
rely-ing as it does essentially on a default task of the
soft-ware, provided that the target–template alignment is
matched accurately modeller, for example, extracts
the distance and dihedral angle restraints from the
alignment then combines these restraints with
CHARMM energy terms to generate the target 3D
model with proper stereochemistry [47] Multiple
struc-tures are usually generated and one of these is
sub-jected to validation of the restraints and backbone
angles using programs such as what if; or the best
‘raw’ structure is optimised by short energy
minimiza-tion runs of the order of 2 ns, using a molecular
dynamics package such as gromacs [52]
Homology models of ABC exporters began
appear-ing durappear-ing the pre-Sav1866 period and, because of the
apparent congruence of the MsbA structure with the
‘generic’ ABC transporter architecture, MsbA was
used as the template for all homology models, whose
target ABC proteins were: MsbA itself [33], ABCB1
[53–56], ABCC1 [57], BmrA [58] and LmrA [59,59a] Unfortunately, much of this early work amounted to very little following the retraction of all three MsbA structures In retrospect, the first MsbA structure was
a poor template choice because its resolution was low at only 4.8 A˚ with the Ca backbone electron density map lacking side chain definition; and the NBDs were nondimeric at 50 A˚ apart with an incor-rect tail-to-tail orientation Nevertheless, who could blame those with the requisite skills from building homology models of their favourite ABC transport-ers? Of the pre-Sav1866 homology models, only one was rendered correctly [56]; and this was achieved by rotating the NBDs of MsbA⁄ ABCB1 150 relative
to the cognate TMDs, generating a P-glycoprotein homology model with a consensus NBD–NBD inter-face and outward-facing TMD helical bundles that bears some resemblance to the Sav1866 TMD tertiary structure This model was also broadly consistent with cross-linking data [60] and low-resolution EM images of ABCB1 [14,61]
The appearance in late 2006 of the S aureus Sav1866 half-transporter ushered in new homology models of other ABC transporters, namely ABCB1 [62,63], LmrA [64], ABCG2 [65,66], ABCC1 [67], ABCC4 and C5 [68,69], as well as the possibly bidirec-tional plant auxin transporter ABCB4 [70] Among this Sav1866-based group, the two ABCG2 models presented problems during their construction and sub-sequent interpretive analyses of cross-linking data and ligand docking, chiefly because of the NBD–TMD reverse domain order, the lack of conserved structural motifs ICD2 and ICD3, and shorter TMD helices and low sequence identity with Sav1866 (Tables 3 and 4; 6–23%) Despite these limitations, one of these ABCG2 studies [65] reported blind docking calcula-tions on their homology model to discern clearly dis-tinct but neighbouring TMD-binding sites for rhodamine, doxorubicin and prazosin; although it should be stressed that the bound ligands, if correctly docked, would be positioned at or near low-affinity binding sites because the ABCG2 homology model was constructed with the TMDs in the outward-facing conformation The authors of both ABCG2 studies acknowledged the limitations of the models and that further refinement and authentication were required [65,66] The ABCB1 and ABCC1 models were much better matched to the Sav1866 template because their primary sequences align more closely In the case of ABCC1, the model was built without including the ABCC subfamily-specific N-terminal five-helix TMD0 domain [67] ABCB1 was rendered as three homology models representing the three catalytic states of closed
Trang 8(ATP bound), semi-open and open apo or ADP
bound These models were generated by a ‘cut and
paste’ approach, using the Sav1866 nucleotide-bound
NBD-ICLs or the MalK nucleotide-free NBDs, and
Sav1866 for the ABCB1 TMDs, which were
subse-quently refined and energy minimized The authors of
all of these post-Sav1866 models contend that they
that are generally consistent with a raft of cysteine
cross-linking studies and spin-labelling and EPR
stud-ies [62,63] With respect to the ABCB1 cross-linking
data for residues within the TM segments, if the
crite-rion for correlation is that the length of any successful
cross-linker (plus 6 A˚ for the cysteine side chains) falls
within 4 A˚ of any distance for the residue pair from
the three different modelled conformations in O’Mara
& Tieleman [62], then 28⁄ 52 results fit (see
Supplemen-tary Table 2 in O’Mara & Tieleman [62]) Given the
latitude of the correlation criterion, this fit is probably
better described as ‘fair’ rather than ‘good’
Plainly, homology modelling uses a crystal structure
template to generate a ‘look-alike’ ABC transporter
For the NBDs, the model building can be very
accu-rate because the sequences are highly conserved (in the
order of 50%) and the NBD tertiary folds of
tem-plate and target superimpose very closely with small
root mean square deviations However, there is only
low sequence similarity in the TMDs ( 15%) A
sec-ond caution is that no matter how accurate a
homol-ogy model can be rendered using a seemingly reliable
template such as Sav1866, homology models suffer
from the same interpretive limitations as static crystal
structures in representing ‘snapshots’ of a multistep
transport mechanism In general, homology models are
comparable with medium-resolution structures and
would not usually be of sufficient quality to be used
for structure-based design directly, although there is
scope for using X-ray scattering, cross-linking data
and MD simulations to improve the models [13,71,72]
Homology modelling of ABCB1 is
consistent with regions showing
correlated evolution
The homology models of ABCB1, as discussed above,
can be validated against biochemical data We have
attempted a validation against bioinformatics data
using the principle of residue co-evolution, i.e the
extent to which evolution of a residue i in a given
protein is coupled to evolution of a residue j Although
iand j may be close together in the 3D structure (and
thus their co-evolution would be expected on structural
grounds), it has also been determined that co-evolution
of pairs of residues at distant sites is indicative of an
allosteric communication between the two sites, as recently explored for the cystic fibrosis transmembrane conductance regulator (CFTR) [73] Many methods are available to determine which regions of a protein are subject to co-evolutionary constraint, and descrip-tion of these is beyond the current review (but see refs [74–78]) For ABCB1, blast analysis [79] and muscle sequence alignment [80] enabled the generation of a multiple sequence alignment of over 150 ABCB type sequences from eukaryotic organisms Analysis of this alignment using several algorithms [74–78,81] enabled identification of regions in the primary sequence of ABCB1 that are co-evolving with other regions When the highest scoring (i.e most likely to be co-evolving) regions are mapped onto the nucleotide-bound model
of O’Mara & Tieleman [62] (Fig 3), it is striking to observe that the majority of these map to key domain– domain interfaces For example, ICL2, which is in direct contact with NBD2 [21], is co-evolving with at least three other regions of ABCB1 Two of these are located in the a-helical region of NBD2, explaining mutagenesis data for ABCB1, as previously described [82], whereas the third is located in TM helix 12 pro-viding an intriguing co-evolutionary perspective on the allosteric communication between NBD1 and TMD2 Further analysis of the data provides many stimulating opportunities for functional analysis of other ABCB1 mutant isoforms
Is the structural basis for interdomain communication observed for several ABC proteins likely to be preserved across the whole family?
The TMD–NBD ‘transmission interface’ features the structurally conserved two short coupling helices that nevertheless share little or no sequence similarity among the different transporters [83] The coupling helices are deployed roughly parallel to the membrane and ‘fit’ into grooves in the tops of the NBDs, in the manner of a ball and socket joint Despite this con-served interface, Sav1866 is the only structure in which the coupling helices are domain swapped – that is, the coupling helix from TMD1–NBD1 interacts with NBD2 and vice versa; and this effect is supported by experimental cross-linking and genetic data for the eukaryotic drug exporters ABCB1 [34], Yor1p [84] and the chloride channel CFTR [85,86] Domain swapping
of the coupling helices does not occur in any of the ABC importer structures and, if this clear distinction between importers and exporters is maintained in future solved ABC transporter structures, it could inform about the mechanics of translocation for which
Trang 9the TMDs need to adopt alternately inward- and
out-ward-facing conformations for the import or export of
allocrites
Intriguingly, only ABC exporters contain the
con-served short X-loop motif (consensus TEVGERG) that
is located just N-terminal to the Signature sequence
and that appears to be involved in cross-linking the
ICLs to one another Thus the X-loop’s chief function
could be to enable the mechanical domain swapping of
the ICL helices for ABC exporters An increasing
number of recent studies of naturally occurring or
artificially swapped domains has widened the range of
functions of domain swapping to include mechanistic
considerations For example, interdomain contacts
between the coupling helices and NBDs of CFTR
comprise aromatic clusters important for stabilization
of the interfaces and also involve the Q-loops and X-loops that are in close proximity to the ATP-binding sites [85,86] The aromatic clusters within the ICLs of CFTR are almost certainly involved in effecting inter-domain communication between the NBDs and TMDs, and such a cluster is found in Sav1866 at the interface of ICL2 and the NBD, but whether this holds true for ABC transporters generally remains to be seen These examples of differences between the TMD–NBD interface might therefore only pertain to the mechanistic coupling involved in import versus export among ABC transporters, that is, between sub-families, and it is likely that the structural basis for interdomain communication is preserved across the prokarya and eukarya kingdoms within the ABC fam-ily, but has evolved to meet the needs of specific functions
An allosteric model of ABC exporter function
A simple, modified allosteric model for membrane pumps was proposed by Jardetsky in 1966 [87] To function as a pump, a membrane protein need only meet three structural conditions, it must: (a) contain a cavity in the interior large enough to admit the solute; (b) be able to assume inward- and outward-facing con-figurations such that the cavity is alternately open to one side of the membrane; and (c) contain a binding site for the transported species within the cavity, the affinity of which is different in the two configurations
In this model, pumps for different molecules need dif-fer only in the specificity of binding sites, and the same pump molecule could be adapted to translocate more than one molecular species For prokaryotic ABC importers we are already seeing these multiple confor-mations at higher resolution [25,88–90] From such structures, and despite structurally unrelated TMD folds, a unified alternating access model for ABC importers and exporters, based on the Jardetsky allo-steric model, has been proposed [91a] and developed further by comparative analysis of several full-length ABC structures [83]
Despite the obvious appeal of this model, there remain several unanswered questions regarding sub-strate transport through MDR-type ABC exporters If the NBDs are directly coupled mechanically to open-ing of the TMDs then what is the magnitude of domain separation required to enable access of trans-port substrate (drug)? Does this vary according to the size of the substrate, and whether it is strongly parti-tioned into the inner leaflet of the membrane (as is likely for many MDR transporter substrates) [91]?
Fig 3 Co-evolving residues of ABCB family members map to
domain interfaces in ABCB1 homology models Co-evolution
analy-sis of ABCB sequences was performed using tools at
http://coevo-lution.gersteinlab.org/coevolution/ and residues identified by
multiple analyses are superimposed onto a structural model of
ABCB1 [62] Regions are coloured as follows: red,177–186
(C-termi-nal to the coupling helix of ICL1, TMD1); blue, 145–155 (N-termi(C-termi-nal
to the coupling helix of ICL1, TMD1); purple, 807–819 (C-terminal
to the coupling helix ICL1, TMD2); orange, 895–915 (ICL2, TMD2);
green, 255–268 (ICL2, TMD1); yellow, 465–475 (Gln loop, NBD1);
cyan, 540–545 (Signature–Walker-B, NBD1); pink, 1220–1229 (His
loop, NBD2 in the lower foreground); salmon, 1120–1140 (Gln loop
and C-terminal helix, NBD2 in the background).
Trang 10How does the interior cavity differ for different
sub-strates? What are the repulsive forces that drive the
TMDs and⁄ or NBDs apart and how is the extent of
domain separation controlled? What are the attractive
forces that bring the domains back into contact – is it
possible that electrostatic attraction across a
solvent-filled gap is sufficient to enable NBD re-association in
a timely and specific manner? As discussed above, the
Sav1866 structure is conformationally constrained by
the intertwined TMD ‘wings’ and domain-swapped
ICL–NBDs, prompting the authors to suggest that the
two subunits are unlikely to move independently and
their maximum separation during the transport cycle is
therefore limited [21] A detailed mechanistic
descrip-tion of substrate translocadescrip-tion through the TMDs of
MDR-type ABC exporters and its allosteric linkage to
ATP binding and hydrolysis within the NBDs will
require their structural characterization in multiple
states, including bound nucleotides and drug substrate
The power of molecular dynamics will also be central
to this challenge
The Holy Grail? A structure for a
eukaryotic MDR pump
Recently, the structure of mouse ABCB1a has been
described by Aller et al., [16] resolved to a resolution of
3.8 A˚ At first glance the structure seems to tick all the
boxes with regard to a structural understanding of
mul-tidrug binding The structure is comparable in terms of
the fold and the domain–domain interactions to the
structures of MsbA and Sav1866 Furthermore, a cavity
is contributed by both TMDs, and is sufficiently large
to accommodate a cyclic peptide drug molecule, with
stereospecificity However, a number of concerns arise
from close inspection of the structure Of most
rele-vance to the current discussion are the resolution, the
completeness of the structure, the spatial separation of
the NBDs and the drug-bound state First, the
resolu-tion is at best 3.8 A˚, which is considerably lower than
the Sav1866 structure The exact orientation of many
side chain residues will be difficult to determine at this
resolution and the very high B-factors in the structure
are a reflection of this uncertainty Second, the
struc-ture does not address one of the major topological
dis-tinctions between a prokaryotic MDR homologue and
eukaryotic ABCB MDR pumps, namely the presence
of a linker domain between the two halves of the
transporter The mouse ABCB1a structure is missing
the 56 amino acids between the end of the first NBD
and the start of the second TMD (the first 32 residues
are also unresolved) The missing linker region means
that no light can be shed on this important region –
phosphorylation of which influences the potency of several transported substrates to increase the ATPase activity implying a role in TMD–NBD communication [92], and that the spatial separation of the NBDs may not reflect the separation(s) observed physiologically Finally, with relevance to this minireview series, the drug-bound state has been determined with stereoi-somers of a cyclic peptide (related to MDR reversal agents from blue–green algae; dendromamides) [93] These are poorly characterized in terms of their inter-action with any of the eukaryotic MDR pumps, unlike the compounds listed in Table 2 Until the structure of ABCB1 with drug bound reaches the quality of the bacterial resistance nodulation division multidrug pumps [94], it seems likely that we will continue to rely on computational approaches (homology model-ling and drug docking) in order to elucidate aspects of both the structure of eukaryotic MDR pumps, and their interaction with a multitude of chemically dis-tinct compounds
Acknowledgements
P M Jones is supported by Cure Cancer Australia and UTS IBID fellowships
References
1 Bates SE (2003) Solving the problem of multidrug resistance: ABC transporters in clinical oncology In ABC Proteins: From Bacteria to Man (Holland IB, Cole SPC, Kuchler K & Higgins CF eds), pp 359–391 Academic Press, New York, NY
2 Chen CJ, Chin JE, Ueda K, Clark DP, Pastan I, Got-tesman MM & Roninson IB (1986) Internal duplica-tion and homology with bacterial transport proteins in the mdr1 (P-glycoprotein) gene from multidrug-resis-tant human cells Cell 47, 381–389
3 Cole SP, Bhardwaj G, Gerlach JH, Mackie JE, Grant
CE, Almquist KC, Stewart AJ, Kurz EU, Duncan AM
& Deeley RG (1992) Overexpression of a transporter gene in a multidrug-resistant human lung cancer cell line Science 258, 1650–1654
4 Doyle LA, Yang W, Abruzzo LV, Krogmann T, Gao
Y, Rishi AK & Ross DD (1998) A multidrug resistance transporter from human MCF-7 breast cancer cells Proc Natl Acad Sci USA 95, 15665–15670
5 Clark R, Kerr ID & Callaghan R (2006) Multiple drug binding sites on the R482G isoform of the ABCG2 transporter Br J Pharmacol 149, 506–515
6 Martin C, Berridge G, Higgins CF, Mistry P, Charlton
P & Callaghan R (2000) Communication between mul-tiple drug binding sites on P-glycoprotein Mol Phar-macol 58, 624–632