Crystal structure of the anion exchanger domain of human erythrocyte band 3

See allHide authors and affiliations

Science  06 Nov 2015:
Vol. 350, Issue 6261, pp. 680-684
DOI: 10.1126/science.aaa4335

Getting rid of carbon dioxide

In mammals, red blood cells deliver oxygen to tissues and remove carbon dioxide. Key to this essential process is a membrane protein called anion exchanger 1 (AE1) which transports bicarbonate (formed from carbon dioxide) out of red blood cells in exchange for chloride. This decreases the pH inside the blood cells, so that oxygen is released from hemoglobin and can diffuse into tissues. Arakawa et al. report the crystal structure of the transmembrane anion exchanger domain of AE1, which includes 14 transmembrane helices. The structure provides a basis for understanding the effects of mutations that lead to red blood cell diseases and also gives insight into the mechanism of ion transport.

Science, this issue p. 680


Anion exchanger 1 (AE1), also known as band 3 or SLC4A1, plays a key role in the removal of carbon dioxide from tissues by facilitating the exchange of chloride and bicarbonate across the plasma membrane of erythrocytes. An isoform of AE1 is also present in the kidney. Specific mutations in human AE1 cause several types of hereditary hemolytic anemias and/or distal renal tubular acidosis. Here we report the crystal structure of the band 3 anion exchanger domain (AE1CTD) at 3.5 angstroms. The structure is locked in an outward-facing open conformation by an inhibitor. Comparing this structure with a substrate-bound structure of the uracil transporter UraA in an inward-facing conformation allowed us to identify the anion-binding position in the AE1CTD, and to propose a possible transport mechanism that could explain why selected mutations lead to disease.

Efficient delivery of oxygen to tissues and the removal of carbon dioxide from the blood are fundamental for respiration. This is principally achieved via the red blood cells or erythrocytes. Anion exchanger 1 (AE1), also known as band 3 or SLC4A1, predominates in the erythrocyte ghost membrane and constitutes 30% of its protein (1). It plays a major role in gas transport. Carbon dioxide generated by metabolic processes in the tissues diffuses into the red blood cells. There it reacts with water in a reversible process catalyzed by carbonic anhydrase II to form HCO3 and protons (2). As the bicarbonate concentration in the erythrocyte increases, the anions are transported by AE1 out into the blood plasma in an electroneutral exchange for chloride ions. With ~106 AE1 molecules per cell (1) and each single protein transporting 4 × 104 to 5 × 104 ions per second (3, 4), transport is extremely fast, and ~90% of the CO2 is taken from the tissues to the lungs as the more soluble form of bicarbonate. Within the erythrocyte, the net result is the accumulation of protons from the hydration of CO2. As a result of the Bohr effect, the low pH causes hemoglobin to release oxygen, which can then diffuse into the tissues. When the blood reaches the lungs, the process is reversed and CO2 is exhaled.

Erythrocyte AE1 is the most abundant and widely studied anion exchanger, but AE1 and its close homologs AE2 and AE3 are found in diverse tissues (5, 6), playing important roles in the regulation of intracellular pH, cell volume, and membrane potential through the exchange of HCO3 and Cl. AE1 is highly expressed as an N-terminally truncated form in the kidney, where it is instrumental in the reabsorption of bicarbonate (7). Many morphological and anemic disorders in the erythrocytes, as well as distal renal tubular acidosis in the kidney, are caused by inherited mutations in AE1 (8).

Human erythrocyte AE1 is a 110-kD glycoprotein. It is built from two domains (9), a cytosolic N-terminal domain (residues 1 to 360) and an integral membrane domain (residues 361 to 911). Anion exchange is catalyzed by the C-terminal domain (1012). However, only the crystal structure of the cytosolic N-terminal domain (13)—which is important as an anchoring point for other proteins, including the scaffolding protein ankyrin (14, 15) and deoxyhemoglobin—has been determined (16, 17). A wealth of biochemical experiments, including cysteine scanning mutagenesis (1823) and N-glycan insertion mutagenesis (24, 25), have been used to derive topology models of AE1. The only three-dimensional structural information available to date for the transmembrane domain is from electron microscopy (26, 27), with the best maps at a resolution of 7.5 Å, calculated from two-dimensional crystals (28). As such, the membrane topology, substrate recognition, and anion-transport mechanism of this fundamental protein remain unclear, and diverse models have been proposed (2831). Here we report the crystal structure at 3.5 Å resolution of the membrane domain of human AE1, locked in an outward-facing open conformation by a covalently bound inhibitor, in complex with a Fab fragment from a monoclonal antibody.

The C-terminal anion exchanger domain of AE1 (AE1CTD), with the N terminus cleaved by trypsin, was purified directly from white ghost membranes of human erythrocytes and treated with H2DIDS (4,4-diisothiocyanatodihydro-stilbene-2,2-disulfonic acid), a disulfonic stilbene derivative that irreversibly inhibits anion exchangers by covalently binding to the protein and blocking the transport cycle (32, 33). Two steps were required to obtain well-diffracting crystals. First, the transporter was deglycosylated with N-glycosidase F. Second, the protein was cocrystallized with a monoclonal antibody Fab fragment that binds tightly to a conformational epitope of AE1CTD. This antibody was selected from a panel of antibodies raised by the inoculation of mice with AE1CTD-displaying baculovirus (34). The structure was solved using MIRAS (multiple isomorphous replacement with anomalous scattering), in combination with noncrystallographic symmetry averaging, and refined at a resolution of 3.5 Å to an R factor of 27.4% and a free R factor of 29.0% (tables S1 and S2) (32). The final model contains all residues from 381 to 887, apart from three loop regions (553 to 567, 640 to 649, and 742 to 753). The asymmetric unit of the crystal contains two dimers of AE1CTD, with one Fab fragment bound on the outer side of each protomer (figs. S1 and S2).

Each protomer comprises 14 transmembrane segments (TMs) (Fig. 1) and has dimensions of about 60 × 60 × 50 Å. The lengths of the individual TMs vary from 18 to 42 Å. Like many other secondary transporters, the structure is built from two repeats inverted in the plane of the membrane (fig. S3). In AE1CTD, the repeat is made of seven TMs, with the third TM of each repeat only partially spanning the membrane before the helical structure unravels. Although it is difficult to superpose all TMs from the inverted repeat as a unit, if the repeat is treated as a two-component module, TMs 1 to 4 can be superposed on TMs 8 to 11 [root mean square deviation (RMSD) of 2.1 Å for 62 out of 103 Cα atoms (supplementary materials)] and TMs 5 to 7 onto TMs 12 to 14 (RMSD of 2.1 Å for 53 out of 100 Cα atoms) (fig. S3B).

Fig. 1 Structure of AE1CTD in an outward-facing conformation.

(A) View of the structure in the plane of the membrane (left) and from the extracellular side of the membrane (right). The cartoon representation of the structure has been colored in rainbow order from blue at the N terminus to red at the C terminus. The H2DIDS is shown as spheres (C, yellow; S, dark yellow; O, red; N, blue). Numbers indicate TMs. Six short helices on the membrane surface are shown as H1 to H6. OUT, extracellular side IN; intracellular side. (B) Structure of the dimer, viewed from the extracellular side. The color coding of the left monomer is the same as in (A). The right monomer is shown in gray.

The inverted repeat units intertwine to form two structural domains separated by a cleft on the extracellular side of the protein (Fig. 1A). We define these structural domains as the core (TMs 1 to 4 and 8 to 11) and the gate (TMs 5 to 7 and 12 to 14), following the convention of the uracil transporter UraA (35, 36). Within the core domain, the N termini of the two half-helices (TMs 3 and 10) face one another at a distance of ~10 Å, creating the appearance of a continuous helix (Fig. 2 and fig. S2B).

Fig. 2 Structural framework of AE1CTD and comparison with UraA.

(A) Topology and (B) overall structure of AE1CTD, viewed in the plane of the membrane. TM 3 is shown in cyan and TM 10 is shown in orange. The other transmembrane helices of the core domain are colored yellow, and those of the gate domain are blue. (C) The core and gate domains of AE1. The left panel shows the core domain viewed from the gate domain; the right panel shows the gate domain viewed from the core domain. Coloring is as in (A). (D) As in (C), but for UraA (Protein Data Bank accession code 3QE7). The two proteins were aligned on their respective core domains.

The overall fold of AE1 is very similar to the structure of UraA (35), although they have only 12% sequence identity, as aligned using the structures (fig. S4). This structural similarity has been previously suggested on the basis of threading combined with mutagenesis experiments (29). UraA is also made of two seven-TM repeat units that form two domains. As shown in Fig. 2, the core domains of AE1CTD and UraA are similar and can be superposed with a RMSD of 1.8 Å for 145 of 268 Cα atoms (1.9 Å for 131 out of 190 Cα atoms when only the TM segments are considered; fig. S5). The gate domains of the two proteins also have the same topology, but it is more difficult to superpose these domains (RMSD of 2.0 Å for only 55 out of 166 Cα atoms) (Fig. 2, C and D, and fig. S5). This is in part because the relative positions of the three pairs of helices (TMs 5 and 12, TMs 6 and 7, and TMs 13 and 14) are different between AE1 and UraA. Given that the gate domains are directly involved in substrate binding, as discussed below, the structural variation could reflect the different substrates of the two proteins.

AE1CTD is a physiological dimer (37), but dimerization is not necessary for transport (38, 39). The dimer in the crystal consists of two AE1CTD monomers with 1092 Å2 of surface area buried at the interface (32). The monomers are related by a twofold axis parallel to the membrane normal, consistent with dimer formation in the erythrocyte membrane (Fig. 1B). There is no obvious difference between the two dimers as observed in the crystal at the current resolution. The interaction between subunits is formed exclusively through residues on the gate domains, including TMs 5, 6, and 7 (Fig. 1B).

The Fab fragments bind solely to the core domain of each AE1CTD (figs. S1 and S2). The interactions between AE1CTD and the Fab are identical for the four molecules of the asymmetric unit. The predominant interactions are between the heavy chain of the Fab and the C-terminal end of TM 3, the following loop, and the loop before TM 8 (fig. S1). The light chain of the Fab interacts with the end of TM 1 and the following loop.

H2DIDS covalently crosslinks Lys539 and Lys851 (40); a nonprotein density between them, consistent with the inhibitor, is evident in the electron density maps (fig. S6). However, part of the H2DIDS density is poorly defined, and it is possible either that the link may have been damaged by radiation or that the crosslinking may not have been carried out to completion (fig. S6, A and B). The H2DIDS molecule spans TMs 5 and 13 of the gate domain (Fig. 2A) at the entrance of a large cavity 15 Å wide, 7 Å long, and 11 Å deep, on the extracellular side of the protein between the core and gate domains (Fig. 3A and figs. S6 to S8). The cavity is formed by TMs 1, 3, 5, 13, and 14 and has predominantly hydrophobic walls, with polar residues near the entrance and at the bottom.

Fig. 3 Outward- and inward-facing cavities and substrate binding sites in AE1 and UraA.

(A) Surface representations of the outward-facing structure of the substrate-free AE1CTD complex with H2DIDS (left) and the inward-facing structure of UraA with uracil (right). These views are in the plane of the membrane. The surfaces are semitransparent and slabbed to show the positions of the transmembrane helices, which are colored as in Fig. 2. H2DIDS and uracil are shown in magenta in the left and right panels, respectively. (B) Comparison of the putative anion binding site in AE1 (left) with the uracil binding site in UraA (right). The coloring is as in (A). Glu241 and Glu290 of uracil correspond to Glu681 and Arg730 of AE1, respectively. The dark blue spheres represent the amide nitrogens of the depicted residues, and the cyan sphere represents the Cα atom of Gly463.

Mutagenesis studies of mouse AE1 demonstrate that the protein is still capable of Cl self-exchange when the lysine residues interacting with H2DIDS are mutated (41), suggesting that the inhibitor does not bind in the same place as the substrate does. Under the H2DIDS molecule, the cavity opens out slightly (18 Å) where the N termini of TMs 3 and 10 meet (Fig. 3). This region has been suggested as a cation selectivity filter (23). In UraA, uracil binds in the space between the positive dipoles TMs 3 and 10. The uracil interacts with two glutamate residues, one on TM 8 (Glu241) and the other on TM 10 (Glu290) (Fig. 3B) (35). In AE1CTD, Glu681 and the positively charged Arg730 take the positions of these two glutamates on TMs 8 and 10, respectively (Fig. 3B). Both of these residues are conserved in AE1, and anion exchange is lost if either is mutated (19, 4244). Although there is no apparent density indicating a substrate, and substrate binding could be blocked by bound H2DIDS, a bicarbonate ion can be placed between the positive dipoles of TMs 3 and 10. The negatively charged bicarbonate could interact with the positively charged Arg730 and could hydrogen-bond to Glu681. The anions may also interact directly with the amide protons at the N termini of TMs 3 and 10 (Fig. 3B). Consistent with anion binding in this space, mutagenesis studies indicate that only a conservative mutation to a threonine (30) or cysteine (29) is tolerated at Ser465 at the N terminus of TM 3. Mutation to the larger isoleucine (29), asparagine, or aspartic acid (30) abolishes transport. The arrangement of an anion between the positively charged dipoles of half-helices (TMs 3 and 10 in AE1), with a negatively charged residue nearby (Glu681 in AE1), is similar to the selectivity filter of the CLC chloride transporter, although the topology of the two proteins is completely different (fig. S9) (45).

AE1 exhibits modes of transport other than the physiological 1:1 exchange of bicarbonate and chloride ions. It can also conduct anions (46) or cotransport protons to drive the uptake of divalent sulfate (47) or chloride ions (48). These ions can easily be accommodated in the basic cavity (figs. S7 and S8). Glu681 on the translocation pathway is a potential proton acceptor during cotransport of H+ and SO42–, because modification to an alcohol (Glu681OH) or mutation to a glutamine leads to a highly proton-independent SO42–/Clexchange (42, 49, 50).

Specific mutations in the AE1CTD domain (fig. S10) are related to red cell diseases such as spherocytosis (51), stomatocytosis (52), and Southeast Asian ovalocytosis (SAO) (53). Some of these mutations, particularly those leading to spherocytosis, cause misfolding of the protein, whereas others exhibit abnormal transport kinetics. Some of the mutations in the transport domain are listed in table S3 and shown in Fig. 4 and figs. S11 and S12. An increase in membrane permeability to monovalent cations, caused by mutations in AE1, has been observed in a number of human pedigrees with dominantly inherited hemolytic anemia (22, 44, 52). These mutations mostly occur on the cytoplasmic half of the core domain (Fig. 4). They include mutation of Arg730, the putative bicarbonate-binding residue, to Cys, as well as two other mutations on the half-helix TM 10 (Ser731 to Pro and His734 to Arg). The deletion of residues 400 to 408, which leads to SAO, also causes a cation leak in intact red cells (54). These residues reside at the TM 1 N terminus at the interface with TM 7 of the gate domain (Fig. 4 and fig. S12), and the mutations in the residues would presumably alter the structure of the core domain, as well as the interaction between AE1CTD and the N-terminal cytoplasmic domain. It is interesting that this deletion may confer protection against cerebral malaria (54). Fewer mutations have been reported in the gate domain.

Fig. 4 Some mutations that are reported to cause diseases, plotted on the structure of AE1CTD.

The core domain is shown in tan and the gate domain is shown in gray. The view is from the cytoplasmic side of the membrane. The deletion mutation of residues 400 to 408, which causes SAO, is shown in yellow. Point mutations are shown by spheres, with coloring according to conservation among human SLC4 family transporters. The level of conservation was analyzed with the assistance of the ConSurf server ( (62) using seven homologs, including AE1 to -4, electrogenic sodium bicarbonate cotransporters 1 and 2 (NBCe1 and -2), electroneutral sodium bicarbonate cotransporter (NBCn1), sodium-driven chloride bicarbonate exchanger (NDCBE), sodium-dependent bicarbonate transporter (NCBE), and sodium bicarbonate transporter–like protein 11 (BTR1). Colors (in rainbow order) represent degree of conservation, with blue indicating the most highly conserved and red indicating the most poorly conserved residues among these homologues. R730 is only conserved among the electrogenic family members. The details of mutations are described in table S3 and figs. S11 and S12. The position of uracil in the UraA structure, corresponding to the possible anion binding site in AE1, is shown with magenta spheres.

The structure of AE1CTD reported here is in an outward-facing conformation (Figs. 1 and 3). AE1CTD is reported to undergo large conformational changes during transport, in line with the alternating access mechanism (55, 56). By comparing the outward-facing AE1CTD structure with a substrate-bound, inward-facing UraA structure (35), we can gain some insight into the transport mechanism (Fig. 3A and fig. S13).

The core domains of these two proteins, including the uracil binding site in UraA, are very similar (Fig. 3B and fig. S5). The primary difference between the two structures is in the relative positions of the gate and core domains (Fig. 2, C and D, and fig. S5). The rotation of one domain against another as the transporter moves from outward- to inward-facing, as shown in fig. S5, is similar to the predominant movements observed for other families of transporters, including the LeuT family transporters, to which there is also some structural similarity (fig. S14) (36, 5761). Figure S13 shows a possible transport mechanism for AE1. Starting from an outward-facing open conformation of the protein, chloride at a high concentration in plasma binds to the anion-binding site. This causes some local conformational changes of the core domain, enabling it to rotate against the gate domain to form the inward-facing structure. At this point, chloride diffuses out and is replaced by bicarbonate to reverse the process. This is consistent with various results from kinetic studies that indicate that chloride and bicarbonate ions share the same binding site (4). The structure of the human AE1CTD and the proposed transport mechanism provide a framework with which to understand the many mutations in the protein that lead to diseases.

Supplementary Materials

Materials and Methods

Figs. S1 to S15

Tables S1 to S3

References (6384)


  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: We thank the beamline scientists at Diamond Light Source, European Synchrotron Radiation Facility and SPring-8 for help with data collection; L. Bruce at the Bristol Institute for Transfusion Sciences; National Health Service Blood and Transplant; and A. Toye at the University of Bristol for useful discussion. S. Weyand contributed to the refinement of the Fab fragment, and R. Suno at Kyoto University contributed to the preparation of the figures. The project was funded by the Biotechnology and Biological Sciences Research Council (grants BB/G023425/1 and BB/D019516/1 to S.I.); the ERATO Human Receptor Crystallography Project of JST (to S.I.); the Targeted Proteins Research Program of the Ministry of Education, Culture, Sports, Science, and Technology, Japan (MEXT) (to S.I.); a Grant-in-Aid for Scientific Research (B) (grants 20370035 and 23370049 to T.K.); MEXT’s Platform for Drug Discovery, Informatics, and Structural Life Science (to T.K. and S.I.); and the European Union’s European Drug Initiative for Channels and Transporters (grant 201924 to S.I. and A.C.). The authors are grateful for the use of the Membrane Protein Laboratory, funded by the Wellcome Trust (grant WT089809) at the Diamond Light Source. The authors declare no competing interests. The project was conceived by S.I., D.K., and N.H. H.H., H.K., and N.H. purified the AE1CTD protein. H.I. and T.Ha. raised the antibody using the AE1CTD-expressing baculovirus as an antigen. T.A., T.Hi., C.I.-S., and T.M. screened the antibody. T.K.-Y., T.K., T.A., Y.Ab., and M.I. crystallized the AE1CTD protein. T.K., Y.Al., A.D.C., and S.I. collected diffraction data. The structure was solved and refined by T.A., Y.Al. and A.D.C. The manuscript was written by T.K., A.D.C., Y.Al., and S.I. The coordinates and structure factors have been deposited with the Protein Data Bank (4YZF, band 3 with Fab4201; 5a16, Fab4201). T.A., T.K., T.K.-Y., T.Hi., T.M., N.N., S.I., T.Ha., H.I., Y.M., and N.H. are authors on a patent filed by JST Agency (JP 2013-103881 A) that covers the monoclonal antibody to human band 3.
View Abstract

Stay Connected to Science

Navigate This Article