The Structure of a Cyanobacterial Bicarbonate Transport Protein, CmpA*

Cyanobacteria, blue-green algae, are the most abundant autotrophs in aquatic environments and form the base of the food chain by fixing carbon and nitrogen into cellular biomass. To compensate for the low selectivity of Rubisco for CO2 over O2, cyanobacteria have developed highly efficient CO2-concentrating machinery of which the ABC transport system CmpABCD from Synechocystis PCC 6803 is one component. Here, we have described the structure of the bicarbonate-binding protein CmpA in the absence and presence of bicarbonate and carbonic acid. CmpA is highly homologous to the nitrate transport protein NrtA. CmpA binds carbonic acid at the entrance to the ligand-binding pocket, whereas bicarbonate binds in nearly an identical location compared with nitrate binding to NrtA. Unexpectedly, bicarbonate binding is accompanied by a metal ion, identified as Ca2+ via inductively coupled plasma optical emission spectrometry. The binding of bicarbonate and metal appears to be highly cooperative and suggests that CmpA may co-transport bicarbonate and calcium or that calcium acts a cofactor in bicarbonate transport.

Cyanobacteria are the most abundant microorganisms in aquatic environments and play a key role in the global carbon cycle (1). It is estimated that these photosynthetic microbes are responsible for ϳ50% of carbon fixation in the oceans. Over their 2.7-billion-year existence, cyanobacteria had to adapt to a changing gaseous environment where the levels of CO 2 declined and O 2 increased (2). Because O 2 can compete with CO 2 for binding to the carbon-fixing enzyme Rubisco (3), cyanobacteria evolved the most effective CO 2 -concentrating mechanism (CCM) 2 that allows them to concentrate CO 2 levels around Rubisco up to 1000-fold. The CCM involves the import and accumulation of inorganic carbon as HCO 3 Ϫ in the cytoplasm and subsequent conversion to CO 2 in the protein microcompartment called the "carboxysome" via carbonic anhydrase.
One component of this CCM machinery in Synechocystis PCC 6803 is the cmpABCD operon that encodes a high affinity bicarbonate ABC transporter that is induced under low CO 2 conditions (4). This transporter is composed of four polypeptides, a high affinity solute-binding lipoprotein (CmpA), an integral membrane permease (CmpB), a cytoplasmic ATPase (CmpD), and an ATPase/solute-binding fusion protein (CmpC) that regulates transport (Fig. 1). The CmpABCD transporter is the highest affinity bicarbonate transporter of cyanobacteria (4). This affinity is predominantly conferred by the binding of bicarbonate to CmpA (K d ϭ 5 M) (5). CmpA is anchored to the periplasmic face of the cytoplasmic membrane via a lipid anchor attached to a conserved cysteine (6). The closest known homologue of CmpA is NrtA, the solute-binding protein of the nitrate-specific NrtABCD transporter that is 48% identical and 61% similar in amino acid sequence (7). We recently published an analysis (8) of the 1.6-Å structure of NrtA complexed with nitrate to elucidate the molecular determinants of nitrate specificity. From this structure, it seemed likely that the nitrate versus bicarbonate specificity was mainly due to the replacement of a lysine in the nitrate coordination sphere in NrtA with a glutamate in CmpA.
To better compare and contrast these two important ABC transport systems, the x-ray structure of CmpA has been determined in three different states, complexed with H 2 CO 3 (carbonic acid) at pH 5.0, bound with HCO 3 Ϫ (bicarbonate) at pH 8.0, and in the absence of ligands at pH 8.0. The "C-clamp" structure of CmpA is remarkably homologous to that of NrtA (8) with ligands binding in the cleft between the two domains. At pH 5.0, nearly all of the bicarbonate is in the fully protonated, carbonic acid form. Under these conditions, the ligand is not bound deep inside the cleft, as was observed in NrtA, but rather at the entrance to the ligandbinding region. At pH 8.0, almost all of the dissolved inorganic carbon is in the bicarbonate form. In this case, the ligand is found deep inside the cleft of the C-clamp and, unexpectedly, is bound concomitantly with a calcium ion. Indeed, bicarbonate binds to CmpA if (and only if) calcium is also present. These and other results suggest that calcium and bicarbonate bind in a strongly cooperative manner and CmpA may transport calcium and bicarbonate simultaneously or that calcium acts as a cofactor in the bicarbonate transport.

MATERIALS AND METHODS
Cloning of cmpA from Synechocystis PCC 6803-The solutebinding domain of CmpA (residues 27-452) from Synechocystis PCC 6803 was cloned from genomic DNA. The cmpA gene was PCR-amplified with Platinum Pfx DNA polymerase (Invitrogen) according to the manufacturer's instructions and standard cycling conditions. The forward primer 5Ј-GGGAA-TTCCATATGGCCGGCAATCCCCCCGAT-3Ј included an NdeI restriction site, and the reverse primer 5Ј-CCGC-TCGAGTTAGACTTTTTTGATTGCCAAACTTTGCAG-3Ј included an XhoI restriction site for cloning into pET-28a. The PCR product was purified with the QIAquick PCR purification kit (Qiagen) followed by digestion with both NdeI and XhoI at 37°C overnight. The gene was separated from digestion byproducts on a 1.0% agarose gel and purified by the QIAquick Gel purification kit (Qiagen). The purified cmpA fragment was then ligated into the expression vector pET-28a (Novagen). The pET-28a (Novagen) vector was previously modified such that the thrombin cleavage site was replaced with a recombinant tobacco etch virus (rTEV) cleavage site. Escherichia coli DH5␣ cells were transformed with the ligation mixture and then plated onto LB medium supplemented with 30 g/ml kanamycin. Individual colonies were selected and cultured overnight, and plasmid DNA was extracted with the QIAprep Spin Miniprep kit (Qiagen). Positive clones were sequenced by Lark Technologies (Houston, TX).
Protein Expression-For protein expression, E. coli Rosetta(DE3)pLysS cells (Novagen) were transformed with the pET28a-cmpA plasmid and plated onto LB medium supplemented with 30 g/ml kanamycin. After ϳ16 h growth at 37°C, the colonies were harvested from the plates and used for the inoculation of 2 ϫ 4-liter baffled flasks containing TB (Terrific Broth) medium supplemented with 30 g/ml kanamycin and 30 g/ml chloramphenicol. The cells were grown at 37°C with aeration to an A 600 of ϳ0.4, at which time the temperature was lowered to 22.5°C for the remainder of the experiment. Thirty minutes after lowering the temperature, protein expression was induced by the addition of 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside. The cells were allowed to grow for an additional 16 h before harvesting by centrifugation. The cell paste was frozen in liquid nitrogen and stored at Ϫ80°C.
Expression of the Selenomethionine-labeled Protein in E. coli-Selenomethionine-substituted protein was expressed via the methionine inhibitory pathway (9). Rosetta(DE3)pLysS (Novagen) cells were transformed with the pET28a-CmpA plasmid and plated onto LB medium supplemented with kanamycin. After ϳ16 h growth at 37°C, the colonies were harvested from the plates and used for the inoculation of 2 ϫ 4liter baffled flasks containing M9 (42 mM Na 2 HPO 4 , 136 mM KH 2 PO 4 , 58 mM NaCl, 53.5 mM NHyCl) medium supplemented with 0.015 mM thiamine, 30 g/ml kanamycin, and 30 g/ml chloramphenicol. Cultures were grown at 37°C to an A 600 of ϳ0.4 before the temperature was adjusted to 22.5°C for the remainder of the growth. Cultures were grown to an A 600 of ϳ0.8 before each flask was supplemented with 200 mg each of L-lysine, L-threonine, and L-phenylalanine and 100 mg each of L-leucine, L-isoleucine, L-valine, and L-selenomethionine. After 30 additional min of growth, the cells were induced with 1 mM isopropyl 1-thio-␤-D-galactopyranoside and allowed to grow for 16 h. Cultures were harvested by centrifugation, and the cell paste was frozen in liquid nitrogen for storage at Ϫ80°C.
Protein Purification-Native and selenomethionine-substituted CmpA were purified in an identical manner. Approximately 10 g of frozen cells were thawed in 50 ml of cold buffer A (25 mM NaH 2 PO 4 , 300 mM NaCl, 10 mM imidazole, pH 8.0) with one Complete EDTA-free protease inhibitor tablet (Roche Applied Science). Cells were lysed on ice by four cycles of sonication (45 s) separated by 3 min of cooling. Cellular debris was removed by centrifugation at 4°C for 30 min at 30,000 ϫ g. The clarified lysate was loaded onto a 5-ml HiTrap chelating HP cartridge column (GE Healthcare) charged with Ni 2ϩ and preequilibrated with buffer A. After loading, the column was washed with ϳ40 ml of 90% buffer A/10% buffer B (25 mM NaH 2 PO 4 , 300 mM NaCl, 300 mM imidazole at pH 8.0) followed by gradient elution of the protein from 10 to 100% buffer B. Protein-containing fractions were pooled based on SDS-PAGE and dialyzed against 20 mM HEPES (pH 7.5) with 200 mM NaCl overnight at 4°C. For removal of the His tag, rTEV protease was added to dialyzed CmpA at a molar ratio of rTEV:CmpA of ϳ1:50 CmpA, and the mixture was incubated for 4 h at ϳ23°C and then 16 h at 4°C. The rTEV protease contained an N-terminal His 6 tag, which allowed the separation of cleaved CmpA from both His-tagged CmpA and rTEV via Ni 2ϩ affinity chromatography, as described above. Cleaved CmpA was dialyzed overnight at 4°C against 20 mM HEPES, pH 7.5, with 20 mM NaCl and then loaded onto a 5-ml HiTrap Q HP anion exchange cartridge (GE Healthcare) pre-equilibrated with the dialysis buffer. After loading, the column was washed with HEPES buffer until the A280 fell below ϳ0.1 OD, and then the FIGURE 1. Schematic representation of the assembled CmpABCD bicarbonate transporter. CmpA is tethered to the periplasmic membrane by a flexible linker and captures bicarbonate in the periplasm for delivery to the transmembrane complex created by CmpB. In many ABC transporters, the transmembrane pore is created by a dimer of two transmembrane-spanning polypeptides. CmpC and CmpD are ATPases that couple ATP hydrolysis to bicarbonate transport through the pore. CmpC is unique in that it contains a C-terminal solute-binding domain homologous to CmpA. SBP, solute-binding protein.
protein was eluted with a gradient of 20 -500 mM NaCl in 20 mM HEPES, pH 7.5. Protein-containing fractions were pooled based on SDS-PAGE and dialyzed against 20 mM HEPES (pH 7.5) with 100 mM NaCl. For crystallization, CmpA was concentrated to ϳ22 mg/ml based on an extinction coefficient of ϳ1.78 ml/cm⅐mg as calculated by the program Protean (DNAstar, Inc., Madison, WI).
Crystallization of Native and Selenomethionine-labeled CmpA-In the first set of crystallization experiments, 25 mM Na 2 CO 3 was added to CmpA prior to crystallization. Potential crystallization conditions were screened using the Hampton Screen at both room temperature and 4°C via the hanging drop method of vapor diffusion. Single crystals were observed at room temperature from 2 M ammonium sulfate. Crystallization conditions were refined, and large single crystals of CmpA (native and selenomethionine-substituted) complexed with carbonate were obtained by streak seeding of the hanging drop crystals into batch plates containing 11 mg/ml CmpA, 50 mM succinate, pH 5.0, 1.1-1.5 M ammonium sulfate, and 12 mM Na 2 CO 3 . Four additional structures of CmpA were determined at pH 8.0 using the following combinations of additives, NaHCO 3 , EGTAϩNaHCO 3 , EGTA, and no additives. These crystals were obtained by streak seeding the Na 2 CO 3 -containing native crystals into batch plates containing 11 mg/ml CmpA, 50 mM HEPPS, pH 8.0, 1.4 -1.6 M ammonium sulfate, and 5 mM EGTA and/or 15 mM NaHCO 3 where applicable. Crystals grew to dimensions of ϳ0.5 ϫ 0.3 ϫ 0.15 mm in 2-3 weeks. All of the various CmpA crystals were triclinic, with similar unit cell dimensions as noted in Tables 1 and 2. The solvent content of the crystals was ϳ50%, with one molecule in the asymmetric unit.
High Resolution X-ray Data Collection-Both the native and selenomethionine-substituted protein crystals were flash-frozen in the same manner. Briefly, crystals were harvested from the batch plates and soaked for several hours in a synthetic mother liquor composed of 100 mM succinate (pH 5.0) or 100 mM HEPPS (pH 8.0), 1.5-1.7 M ammonium sulfate, and 300 mM NaCl. The additives EGTA, NaHCO 3 , or Na 2 CO 3 were included in all synthetic mother liquor and cryoprotectant solutions, when applicable, at 5 mM above the concentrations used for crystallization. Crystals were serially transferred to a final cryoprotectant solution containing  100 mM succinate, pH 5.0, or 100 mM HEPPS, pH 8.0, 300 mM NaCl, 1.5-1.7 M ammonium sulfate, 21-25% ethylene glycol, and the appropriate amount of additive. The crystals were then flash-cooled by submersion in liquid nitrogen. X-ray data sets from the selenomethionine-substituted crystals and native CmpA complexed with Na 2 CO 3 were collected on a "SBC3" charge-coupled detector at the Structural Biology Center 19-BM beamline (Advanced Photon Source, Argonne National Laboratory, Argonne, IL). The x-ray data were processed with HKL2000 and scaled with SCALEPACK (10). Relevant x-ray data collection statistics are summarized in Table 1. X-ray data on the CmpA crystals grown in the presence of additional NaHCO 3 were collected in-house using CuK ␣ radiation from a Bruker-nonius FR591 x-ray generator operated at 45 kV and 90 mA. The x-ray data were collected and processed using a Bruker Proteum charge-coupled detector system. X-ray data on the remaining three complexes (EGTA, EGTA, and NaHCO 3 added, and no additives) were collected by Dr. Matthew Benning at Bruker AXS, Inc. in Madison, WI. The data were collected on a Microstar generator (running at 2.7 kW) with Helios optics. The detector was a Pt-135 charge-coupled device. Data reduction was performed using the programs SAINT and SAD-ABS from the Proteum2 software suite (Bruker AXS Inc., 2006).
X-ray Structural Analyses-The structure of CmpA was solved via multiwavelength anomalous dispersion phasing with x-ray data collected from the selenomethioninesubstituted protein crystals ( Table  2). The software package SOLVE (11) was used to determine and refine the positions of the selenium atoms, and the program RESOLVE (12) was then used to perform solvent flattening and initial protein model building. RESOLVE successfully built 356 of 452 residues for CmpA, and the remaining residues were added manually using the graphics package O (13). Alternate cycles of maximum likelihood refinement with crystallography NMR software and manual model building reduced the R-factor to Ͻ20% for all five models. Relevant data collection and refinement statistics are summarized in Table 2. All five models yielded similar Ramachandran analyses; 90% of residues were in the most favored regions, 9.1% in the additionally allowed regions, 0.9% in the generously allowed regions, and 0% in the disallowed regions. In all five models, three residues are in the generously allowed regions of the backbone conformation, Asn-150, Glu-270, and Tyr-319. However, the high resolution electron density for these residues is unambiguous and consistent with the model. The peptidic carbonyl oxygen of Asn-150 hydrogen bonds to Asn-152 or Glu-270, and both are involved in Ca 2ϩ binding. Similarly, Glu-270 is directly involved in Ca 2ϩ binding and is in a random coil region connecting two ␤-strands that line the back of the solute-binding pocket. Tyr-319 is part of a surface loop that connects two ␣-helices. Methodology for Metals Determination-Determination of the calcium and nickel content in the protein was performed using inductively coupled plasma optical emission spectrometry (ICP-OES). A Thermo Jarrell-Ash IRIS-AP spectrometer was used. Calcium measurements were taken at wavelengths of 393.366 and 396.847 nm, and nickel content was measured at 231.604 nm. The sample was run after 10-fold dilution with de-ionized water. Duplicate reagent blanks were also run. Instrument detection limits for both calcium and nickel were Ͻ1 ng/ml. A semiquantitative scan for other elements present was performed using inductively coupled plasma mass spectrometry (ICP-MS). A ThermoElemental PQ Excel ICP-MS instrument was used. This determination revealed only the presence of iron (ϳ3.5 g/ml) and zinc (ϳ0.2 g/ml) at appreciable concentrations.

RESULTS
The overall molecular structure of CmpA with carbonic acid is shown in Fig. 2. CmpA is an ␣-␤ protein and belongs to the periplasmic-binding protein superfamily (14). Similar to the related nitrate-binding protein NrtA, CmpA is composed of two domains (I and II) organized in a C-clamp shape. The core of each domain consists of a five-stranded mixed ␤-sheet flanked by ␣-helices. Domain I, which is slightly larger than domain II, is composed of three different segments of the polypeptide chain, 54 -150, 267-321, and 350 -426. This domain contains ␤1-␤4 and ␤10 arranged as a mixed ␤-sheet, flanked by ␣1-␣4, ␣8 -␣10, and ␣12-␣14. Similarly, domain II is composed of several portions of the polypeptide chain, 151-266, 322-349, and 427-452. The fold of domain II is similar to that of domain I with ␤5-␤9 forming a mixed ␤-sheet flanked by ␣4 -␣7, ␣11, and ␣15. In addition, domain II contains two antiparallel ␤-strands, 11 and 12, that are adjacent to the core motif. The two domains of CmpA are connected by random coil elements and flanked on either side of the solute-binding cleft by ␣-helices.
Based on the order of the ␤-strands in each domain (21354), CmpA belongs to class II of the periplasmic-binding protein superfamily (14), which also includes the oxyanion-binding proteins specific for sulfate, phosphate, and molybdate (15)(16)(17)(18)(19). However, CmpA, similar to NrtA, is much larger (ϳ100 amino acids) than these oxyanionbinding proteins. The last 100 residues of CmpA are mostly in an ␣-helical conformation with the exception of the 2-stranded antiparallel ␤-sheet in the C-terminal domain. These terminal residues form ␣-helices that wrap around the back of the structure and cradle the two ␣/␤ domains. As observed in NrtA, the surface charge distribution on the face opposite to the ligandbinding cleft of CmpA is almost entirely composed of acidic residues. Other solute-binding proteins (e.g. zinc, iron, sulfate, and phosphate) have a similar charge distribution and, during our analysis of the NrtA structure, led us to suggest that such a negative charge may facilitate delivery of solute to the transmembrane pore by limiting unproductive interactions of the solute-binding protein with the phospholipid membrane.
The major difference between the CmpA and NrtA structures is that the CmpA clamp is in a more open conformation than that of NrtA. This suggests that CmpA, at pH 5.0 (R work ϭ 17.5%, R free ϭ 19.2%), is in the unliganded form. However, electron density consistent with a CO 3 molecule was observed near, but not in, the solute-binding cleft (Figs. 2-4). At pH 5.0, nearly all of the solute is in the protonated, carbonic acid form. Therefore, it appears that carbonic acid does not bind in the expected ligand-binding pocket and, in turn, does not induce closure of the C-clamp. Interestingly, the carbonyl oxygen of carbonic acid is within 2.6 Å of the N ⑀ atom of Trp-245 but does not hydrogen bond with any other residue in the protein. Bicarbonate is the predominant carbon species in the aquatic environment of Synechocystis and binds with a dissociation constant of ϳ5 M. Therefore, it is unlikely that this carbonic acid/CmpA interaction occurs in nature. However, it is also possible that the position of carbonic acid represents an initial interaction between bicarbonate and the exterior loops of the C-terminal domain to facilitate delivery of solute to the binding pocket. This kind of "preloading" of ligand to exterior loops of the transport protein was suggested in the case of the zinc-binding protein ZnuA (20).
To determine the structure of CmpA with its authentic ligand, bicarbonate, the pH of the crystallization medium was increased to pH 8.0 (R work ϭ 17.9%, R free ϭ 19.3%). At this pH, nearly all of the dissolved CO 2 is in the bicarbonate form. It is important to note that, at this pH, bicarbonate is observed in the ligand-binding cleft whether or not it was added to the crystallization mother liquor. In both cases, bicarbonate is wedged in the binding cleft created by the juxtaposition of the two globular domains of CmpA and causes the C-clamp to close over the bound ligands (Figs. 2-4). Unexpectedly, a metal ion is also observed adjacent to the bicarbonate and bound by several residues from both the N-and C-terminal domains. There is no previous evidence suggesting that bicarbonate binding and transport by CmpA is metal-dependent. The metal ion is coordinated by five amino acid side chains and one bicarbonate oxygen in a distorted octahedral arrangement. The axial ligands are provided by side chain oxygens of Glu-270   5 Å), and a carboxyl oxygen (2.5 Å) from bicarbonate. The octahedral geometry of the metal coordination is completed by side chain oxygens from Glu-70 and Gln-198, which are within 2.3 and 2.4 Å of the metal, respectively. The average nonbonded distance (2.4 Å) immediately eliminated a water molecule at the center of the coordination sphere and was instead highly suggestive of a metal.
Both the chemistry and geometry of the metal-oxygen coordination sphere initially suggested that the bound metal was either Ca 2ϩ or Na ϩ (21,22). It is important to note that this putative metal was not observed in the structure at pH 5.0, and the nature of the metal/protein interactions are unlikely to be affected by the shift from pH 5.0 to 8.0. To ascertain which is the more likely metal bound to CmpA, both metals were modeled into the density and unrestrained B-value refinement was performed in crystallography NMR software. Ca 2ϩ refined to a B-value of 17.1 Å 2 , nearly identical to the average B-value of the surrounding atoms (16.1 Å 2 ). In contrast, when Na ϩ was refined at this position, the resulting B-value was, as expected, a below average value of 6.3 Å 2 . The contention that the bound metal was a Ca 2ϩ ion was further supported by analysis performed using the Metalloprotein Data Base and browser (Scripps Research Institute). Although both Na ϩ and Ca 2ϩ are observed to have similar ligand-metal distances, it is not as common to find Na ϩ ions coordinated by six ligands, and when it does occur, the ligating moieties are primarily water molecules. In contrast, Ca 2ϩ is most commonly observed octahedrally coordinated by oxygens contributed from side chains such as Glu, Asp, Asn, and Gln. Indeed, ICP-OES revealed that the only significant metal in the CmpA sample was calcium at a ratio of 0.6 Ϯ 0.04 mol for every 1 mol of protein. Because the purification and crystallization buffers did not contain Ca 2ϩ , the source of the metal is unclear. It is possible that CmpA acquired the Ca 2ϩ from E. coli during expression and carried it along through the purification process. Indeed, this has been observed in a number of high affinity metal-binding ABC transporters (e.g. the zinc-binding ZnuA (20) and the iron-binding FutA1. 3 Bicarbonate is bound adjacent to Ca 2ϩ in a more solventexposed position in the binding pocket. The negative charge of bicarbonate is polarized at the O1 atom and is counterbalanced by the Ca 2ϩ that is 2.5 Å away. The O3 atom of bicarbonate is positioned 2.5 Å from the carboxylate group of Glu271, suggesting that O3 is protonated. The O3 atom is also within hydrogen-bonding distance (2.8 Å) of the N ⑀1 of Trp-99. The carbonyl oxygen of bicarbonate, labeled O2, that points toward the opening of the binding cleft is of the appropriate hydrogenbonding geometry and within 2.7 Å of the side chain hydroxyl of Thr-192 and proximal to two water molecules.
To further test the role of calcium on bicarbonate binding, 10 mM EGTA was added to the CmpA crystallization medium at pH 8.0. This structure was refined to a final resolution to 1.35 Å, with R work and R free values of 17.9 and 19.3%, respectively ( Table 2). In this model, the C-clamp of CmpA is in the "open" conformation and is devoid of both divalent metal and bicar-bonate (data not shown). This structure of apoCmpA at pH 8.0 is essentially identical to that of CmpA complexed with carbonic acid, with a root mean square deviation of 0.1 Å for all atoms. That EGTA alone stripped away both Ca 2ϩ and bicarbonate suggests that solute binding is metal-dependent. Again, without adding either bicarbonate or calcium to the crystallization mother liquor, bicarbonate and metal are found in the binding pocket, but both are removed by the addition of EGTA.
It should be noted that CmpA, compared with some other solute-binding proteins, undergoes a relatively small conformational change upon solute binding. The root mean square deviation between the C-␣ backbones of the apo-and bicarbonate-Ca 2ϩ -complexed forms of CmpA is only ϳ0.3 Å. The largest differences between these two states are found in residues 189 -194 that comprise a surface loop proximal to the binding cavity (Fig. 4a). This loop includes Thr-192 that hydrogen bonds with bicarbonate and moves 1.3 Å into the ligand-binding cleft upon solute binding. This movement of Thr-192 creates a hydrogenbonding donor to the O3 of bicarbonate (Fig. 4b).
To further examine this metal-bicarbonate interaction, CmpA crystals were grown at pH 8.0 in the presence of 10 mM EGTA and an additional 20 mM NaHCO 3 . The resulting 1.35-Å structure of CmpA refined with an R work and R free of 20.9 and 22.2%, respectively. Unexpectedly, CmpA was observed in the "closed" conformation with both Ca 2ϩ and bicarbonate in the binding cleft. The coordinates from this structure and those of CmpA with bicarbonate and Ca 2ϩ , obtained without crystallization additives, are nearly identical, with a root mean square deviation of 0.2 Å over all 402 residues. Together, these results suggest that metal and bicarbonate bind to CmpA in a strongly cooperative manner.

DISCUSSION
CmpA is a member of the periplasmic binding protein superfamily and consists of two ␣/␤ domains organized as a C-clamp with solute occupying the cleft created between the two domains. Unlike some other members of this family, CmpA does not undergo a dramatic conformational change upon solute binding. This suggests that the binding cavity is relatively rigid and that solute binding does not require extensive rearrangement of the ligating residues.
The closest homologue of CmpA is the nitrate-binding protein NrtA that shares 48 and 61% amino acid sequence identity and similarity, respectively. We previously reported the 1.6-Å crystal structure of NrtA from Synechococystis PCC 6803 complexed with nitrate (8). The structures of CmpA and NrtA can be superimposed with a root mean square deviation of 1.1 Å for 344 C-␣ atoms (Fig. 4a). The most significant difference in the backbone of these proteins lies in residues 220 -226 (see loop noted with a star in Fig. 4a). In the structure of NrtA, this hydrophobic loop blocks solvent access to the nitrate-binding cavity. In contrast, the solute-binding cavity of CmpA remains slightly more open with bicarbonate partially exposed to solvent. The residues in CmpA that contact the bound bicarbonate are essentially those predicted from sequence homology between CmpA and NrtA (8). There are, however, some differences in how CmpA and NrtA bind their respective ligands. Nitrate and bicarbonate occupy the same general position, but nitrate is buried ϳ1.5 Å deeper in the NrtA-binding cleft and partially occupies the space taken by the Ca 2ϩ atom in CmpA. In both structures, the negative charge on bicarbonate and nitrate is polarized at the O1 oxygen of each anion, but this charge is stabilized differently in each protein. In NrtA, the N ⑀ of Lys-269 is located within 2.8 Å of the O1 oxygen, stabilizing the negative charge (Fig. 4c). In CmpA, this residue is replaced by Glu-271 (Fig. 4b). In our analysis of the NrtA crystal structure, we predicted that the exchange of Lys for Glu was necessary to provide a hydrogen bonding acceptor to the protonated oxygen of bicarbonate. This is, in fact, the case in CmpA with Glu-271 also contributing to the Ca 2ϩ coordination sphere. Similarly, the N ⑀ of lysine in NrtA is replaced by Ca 2ϩ , which interacts with the negative charge on the O1 oxygen of the polarized bicarbonate. To accommodate Ca 2ϩ in CmpA, several other compensatory changes occurred compared with NrtA. In CmpA, Glu-271, Gln-155, and Gln-198 shape the Ca 2ϩ coordination sphere, whereas in NrtA, these residues are replaced by Asn-152 and His-269, respectively, and are hydrogen bonding donors for the O1 and O2 oxygens of nitrate. Only the equivalent of Glu-70 (Glu-73 in NrtA) is present in NrtA. The side chain of Glu-73 is 2.8 Å from the N ⑀ of Lys-269 and likely aids in the positioning of the lysine side chain.
Finding a metal co-ligand in the CmpA crystal structure was quite unexpected. Initially, either Ca 2ϩ or Na ϩ were possible candidates for the bound metal because of the number of acidic residues in the coordination sphere and the average metal-oxygen distance (2.4 Å). This was clarified by ICP-OES analysis that identified calcium as the major metal ion associated with the protein with 0.6 Ϯ 0.04 mol of Ca 2ϩ bound/mol of CmpA. Because none of the purification or crystallization reagents contained significant amounts of calcium, it is possible that CmpA acquired Ca 2ϩ during expression in E. coli and carried it through subsequent purification steps. This has been observed for some periplasmic metal-binding proteins (e.g. ZnuA, FutA1) that bind their ligand with a K d of Ͻ10 Ϫ6 M. Concomitant binding of both metal and anion has been observed with other periplasmic-binding proteins. Synergistic Fe 3ϩ binding with phosphate or CO 3 Ϫ has been observed for both the Haemophilus influenzae and Neisseria gonorrhoeae periplasmic ferric-binding proteins (23,24). The crystal structures of these proteins complexed with Fe 3ϩ and phosphate show that the anion not only contributes to the coordination sphere of the metal but makes multiple hydrogen bonds and electrostatic interactions with the protein as well (25,26). In these proteins, the anion increases the ferric-binding capacity of the ferricbinding protein and binds synergistically with the ferric ion (27).
A similar phenomenon occurs with the concomitant binding of calcium and bicarbonate in CmpA. The effects of adding EGTA and bicarbonate on ligand binding are complicated but suggest that metal and bicarbonate bind in a strongly cooperative manner. The fact that both metal and bicarbonate are found in the binding pocket at pH 8.0, even though neither were added during purification and crystallization, indicate that these ligands, at least together, bind with very high affinity. When EGTA is added to crystals grown in the absence of exogenously added bicarbonate and metal, EGTA is able to chelate the trace metal in solution. This alone is apparently sufficient to eliminate both bicarbonate and calcium from the solute-binding pocket. Further, when the crystals are grown at pH 5.0, the dissolved CO 2 is nearly all in the carbonic acid form and unable to bind to the solute-binding pocket. Under these conditions, metal is also unable to bind despite the fact that none of the ligating protein residues are expected to change to a protonation state when the pH is decreased to 5.0. These results strongly suggest that removal of either metal or bicarbonate causes concomitant release of the other ligand. This is further supported by the fact that exogenously added bicarbonate prevents the release of metal even when EGTA is added to the crystallization medium. This result suggests that relatively high concentrations of bicarbonate push the equilibrium to the bound state and effectively prevent metal release from CmpA. Taken together, it seems plausible that Ca 2ϩ and bicarbonate binding and release are highly cooperative.
Although Ca 2ϩ is an essential metal ion in a variety of biochemical processes, the uptake and regulation of Ca 2ϩ in bacteria has not been studied nearly as well as it has in eukaryotes. In cyanobacteria, Ca 2ϩ is an essential part of the reaction centers of photosynthetic proteins and may also be a signal for differentiation during heterocyst formation (28,29). In the case of CmpA, calcium may be used to either increase the binding affinity or selectivity of bicarbonate over the other anions such as nitrate. In this way, calcium could be a "coenzyme" that facilitates transport without being an important transported nutrient itself. It is interesting to note that only the fresh water strains of Synechocystis have the cmp operon and that fresh water has lower levels of bicarbonate and calcium (0.96 and 0.38 mmol/kg, respectively) compared with ocean water (2.39 and 10.3 mmol/kg, respectively) (30,31). Although it is tempting to speculate that the calcium bicarbonate complex is transported into the cell, there is currently no evidence that suggests that CmpABCD is involved in such a process. Indeed, there is some evidence that calcium is transported in bacteria, however, via a number of genes unrelated to the cmpABCD operon (32). Further studies are clearly necessary to understand the role of calcium in the facilitated transport of bicarbonate.