Mg2+-dependent Interactions of ATP with the Cystathionine-β-Synthase (CBS) Domains of a Magnesium Transporter*

Background: CNNMs are evolutionarily conserved Mg2+ transporters that possess CBS domains. Results: CBS domains are essential for molecular function, and they bind to ATP in a Mg2+-dependent manner. Conclusion: The unique binding feature with ATP may contribute to Mg2+ transport by CNNMs. Significance: This is the first report showing the Mg2+-dependent and specific interactions of ATP with CBS domains. Ancient conserved domain protein/cyclin M (CNNM) family proteins are evolutionarily conserved Mg2+ transporters. However, their biochemical mechanism of action remains unknown. Here, we show the functional importance of the commonly conserved cystathionine-β-synthase (CBS) domains and reveal their unique binding ability to ATP. Deletion mutants of CNNM2 and CNNM4, lacking the CBS domains, are unable to promote Mg2+ efflux. Furthermore, the substitution of one amino acid residue in the CBS domains of CNNM2, which is associated with human hereditary hypomagnesemia, abrogates Mg2+ efflux. Binding analyses reveal that the CBS domains of CNNM2 bind directly to ATP and not AMP in a manner dependent on the presence of Mg2+, which is inhibited in a similar pattern by the disease-associated amino acid substitution. The requirement of Mg2+ for these interactions is a unique feature among CBS domains, which can be explained by the presence of highly electronegative surface potentials around the ATP binding site on CNNM2. These results demonstrate that the CBS domains play essential roles in Mg2+ efflux, probably through interactions with ATP. Interactions with ATP, which mostly forms complexes with Mg2+ in cells, may account for the rapid Mg2+ transport by CNNM family proteins.

Magnesium is an essential element involved in a wide variety of biological activities. The homeostasis of magnesium levels is strictly regulated by intestinal absorption and renal reabsorption, in which the epithelia function as a barrier that permits selective and regulated transport of Mg 2ϩ from apical to basolateral surfaces. Genomic analyses of familial cases of hypomagnesemia have identified key molecules directly involved in these processes (1,2). One of the least characterized members among these is the ancient conserved domain protein/cyclin M (CNNM) 3 family. CNNM family proteins commonly possess a domain that is highly conserved from bacteria to humans (3). In humans, the CNNM family consists of four members, CNNM1-4. These proteins show a multidomain structure, composed of a sequence motif present in the cyclin box, a cyclic nucleotide monophosphate binding domain, two tandem cystathionine-␤synthase (CBS) domains, and a DUF21 domain (3). Significant evidence has suggested the importance of CNNMs in Mg 2ϩ transport (4 -6). Our recent study revealed that CNNM4 localizes to the basolateral membrane of intestinal epithelial cells and extrudes intracellular Mg 2ϩ to the outside of the cells in exchange for Na ϩ , thereby mediating the transcellular transport of Mg 2ϩ through the epithelia (7). However, the functional importance of each domain or motif present in the CNNM family is poorly understood. The DUF21 and CBS domains are supposed to be important for the biological function, as stated by the studies on human hereditary diseases caused by mutations in the CNNM family gene. Several point mutations that cause Jalili syndrome, which is characterized by recessive amelogenesis imperfecta, occur in the DUF21 domain-encoding region in CNNM4 (8,9). In addition, a point mutation that causes familial dominant hypomagnesemia, a rare human disorder characterized by renal Mg 2ϩ wasting, occurs in the CBS domain-encoding region in CNNM2 (6).
CBS domains are small protein modules (typically 60 amino acid residues in length) that usually exist as two tandem repeats, each of which is made up of highly conserved two ␣-helices and three ␤-strands arranged in a ␤-␣-␤-␤-␣ order (10). CBS domains were originally identified in human cystathionine-␤synthase (11) and subsequently found in thousands of cytosolic and membrane-associated proteins from all kingdoms of life (10). In addition to CNNMs, there is an example of a CBS domain-possessing Mg 2ϩ transporter, and in the case of bacterial MgtE, the CBS domains are proposed to be important for the regulation of Mg 2ϩ transporting activity (12,13). The phys-iological importance of CBS domains has been demonstrated by a number of observations that point mutations found in several hereditary diseases occur in the CBS domain-encoding region in the genes and disrupt their protein functions (14). Recent studies have shown that adenine nucleotides, such as ATP, ADP, or AMP, can bind to CBS domains of several proteins, such as AMP-activated protein kinase (AMPK), inosine-5Ј-monophosphate dehydrogenase (IMPDH), and chloride channel (ClC) family proteins (10,15). The functional consequences and physiological importance of interactions with adenine nucleotides have been characterized most extensively in the case of AMPK. The CBS domains of the regulatory ␥-subunit of AMPK have two discrete sites that can bind to AMP, ADP, and ATP (named as site 1 and site 3) (16,17). Structural and biochemical studies have shown that AMP or ADP binding, but not ATP binding, to site 3 promotes phosphorylation and inhibits the dephosphorylation of Thr-172 on the catalytic ␣-subunit, thereby maintaining AMPK in its most active form (17,18). Phosphorylated AMPK is further activated by AMP binding, but not by either ADP or ATP binding, to site 1 (17,18). Such intricate regulations of the activity are impaired by disease-associated amino acid substitutions found in patients with Wolff-Parkinson-White syndrome, indicating the physiological importance of interactions with adenine nucleotides (15,17).
In this study, we examined the functional importance of the CBS domains of CNNM family proteins. Binding analyses revealed that the CBS domains of CNNM2 bind to ATP, but not AMP, in a manner dependent on the presence of Mg 2ϩ . A mutation in the CBS domains of CNNM2 associated with hereditary hypomagnesemia and the equivalent mutation of CNNM4 completely abrogated their Mg 2ϩ efflux functions. Furthermore, this mutation also abrogated the binding of the CBS domains of CNNM2 to ATP. These results demonstrate the crucial importance of the CBS domains for Mg 2ϩ efflux function of CNNM family proteins probably via the interaction with ATP.

EXPERIMENTAL PROCEDURES
Materials-AMP sodium salt, ATP disodium salt hydrate, and the mouse and rabbit anti-FLAG antibodies were purchased from Sigma-Aldrich. [2, H]AMP (15 Ci/mmol) was purchased from Movarek, and [␣-32 P]ATP (3000 Ci/mmol) was purchased from Izotop. Magnesium Green-AM and rhodamine-phalloidin were purchased from Invitrogen.
Cell Culture and Transfection-HEK293 cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and antibiotics. Expression plasmids were transfected in each cell by using Lipofectamine 2000 (Invitrogen).
Mg 2ϩ Imaging Analyses-Mg 2ϩ imaging analyses with Magnesium Green were performed as described previously (7).
Preparation of Recombinant Proteins-Recombinant proteins of CBS domains of CNNM1-4, ClC-5, and IMPDH2 were expressed as GST fusion forms in Escherichia coli by using the expression constructs in pGEX-6p1 and were purified using glutathione-Sepharose beads (GE Healthcare). For surface plasmon resonance (SPR) analysis, the GST tag was cleaved off by digestion with PreScission protease (GE Healthcare).
SPR Analyses-The binding of Mg 2ϩ and ATP to the CBS domains of CNNM2 was analyzed using Biacore T200 instrument with an S series CM5 sensor chip (GE Healthcare). GSTcleaved CBS-WT and CBS-T568I were immobilized onto the sensor chip by using a standard amine coupling protocol. Binding analyses were performed in running buffer (50 mM HEPES-KOH (pH 7.0), 140 mM KCl, 1 mM 2-mercaptoethanol, and 0.05% Tween 20) at 25°C in the presence of 0, 0.5, or 10 mM MgCl 2 . For analyses of Mg 2ϩ binding, MgCl 2 was injected as an analyte in running buffer without MgCl 2 . SPR data were analyzed using the Biacore T200 Evaluation software and fit to a 1:1 binding model using the following equation, R ϭ (R max ϫ C)/ (K d ϩ C), where R is the observed response, R max is the maximal binding capacity, C is the analyte concentration, and K d is the dissociation constant. For the injection of ATP with buffer switching, the dual injection mode was applied, according to the manufacturer's instructions. Observed responses were normalized to R max for each experiment and are represented as normalized response unit (RU). For normalization of sensorgrams for ATP injection without significant responses, R max of SPR responses to ATP injection in the presence of Mg 2ϩ 10 mM from the CBS-WT surface was used.
Filter Binding Assays-Filter binding assays were performed in 10-l reaction mixtures containing reaction buffer (50 mM HEPES-KOH (pH 7.0), 140 mM KCl, 1 mM 2-mercaptoethanol), 50 M recombinant proteins, and various concentrations of [␣-32 P]ATP (500 cpm/pmol) or [2, H]AMP (60 cpm/pmol) in the presence or absence of 10 mM MgCl 2 . The reaction systems were incubated for 20 min at 25°C and immediately filtered through nitrocellulose membranes (25-mm diameter, 0.22-m pore size; Millipore), which were prewashed with reaction buffer and washed with 1 ml of ice-cold reaction buffer. Filters were dried, and the radioactivity was determined by scintillation counting using a LS-6500 scintillation counter (Beckman). Nonspecific binding was determined by control assays with GST and was subtracted. The data were analyzed using the GraphPad Prism software (version 4.0) and fit to a 1:1 binding model using the same equation used in SPR analyses, where R was substituted with the amount of bound adenine nucleotides, C was substituted with the concentration of adenine nucleotides, and R max was substituted with estimated maximum binding of adenine nucleotides. The data were normalized to the percentage of estimated maximum binding of ATP and are represented as the mean Ϯ S.E. of three independent experiments. For competition assays, recombinant proteins were incubated with 5 M [␣-32 P]ATP (5000 cpm/pmol) and reaction buffer in the presence of increasing concentrations of ATP or AMP and were subsequently processed as described above.
Structural Analyses-Homology models for the CBS domains of CNNM1, CNNM2, and CNNM4 in complex with ATP were built using Spanner (20) based on the structures of the CBS domains of CorC (a bacterial homolog of CNNMs) from Bordetella parapertussis (Protein Data Bank (PDB) ID 3JTF), and that of IMPDH2 was built using Spanner based on the structure of and IMPDH from Pseudomonas aeruginosa (PDB ID 4DQW) (21). ATP was placed into the predicted binding sites, based on the similar structures in complex with ATP found from GIRAF (22). Electrostatic surface potentials were calculated for these modeled structures and the crystal structure of the CBS domains of human ClC-5 in complex with ATP (PDB ID 2J9L) (19) using the Adaptive Poisson-Boltzmann Solver software (23), mapped on the solventaccessible surface, and visualized in PyMOL.

Indispensable Role of the CBS Domains in Mg 2ϩ
Efflux-In our previous study, we reported that CNNM4 can stimulate Mg 2ϩ efflux when expressed in HEK293 cells (7). To examine whether the other CNNM family proteins also have the ability to stimulate Mg 2ϩ efflux, we transfected HEK293 cells with CNNM1-FLAG, CNNM2-FLAG, CNNM3-FLAG, or CNNM4-FLAG (Fig. 1A, lower panel) and then, performed imaging analyses with Magnesium Green, a fluorescent indicator for Mg 2ϩ . Cells were first loaded with Mg 2ϩ by bathing them in a solution containing 40 mM Mg 2ϩ and then exposed to a Mg 2ϩ -free solution to promote Mg 2ϩ efflux artificially. The intensity of fluorescent signals in the cells expressing CNNM4-FLAG rapidly decreased immediately after Mg 2ϩ depletion, whereas only a very subtle decrease was observed in empty vector-transfected cells (Fig. 1A, top), thereby confirming our previous finding (7). A significant level of Mg 2ϩ efflux was also observed in cells expressing CNNM2-FLAG. In contrast, only a very subtle level of Mg 2ϩ efflux was observed in cells expressing CNNM1-FLAG, and no significant Mg 2ϩ efflux was observed in the cells expressing CNNM3-FLAG. These data indicate that CNNM2 can also potently stimulate Mg 2ϩ efflux, similar to the efflux induced by CNNM4, when expressed in HEK293 cells.
To investigate the importance of the CBS domains of CNNM family proteins in their Mg 2ϩ efflux functions, we constructed deletion mutants of CNNM2 and CNNM4 lacking CBS domains, called CNNM2-⌬CBS and CNNM4-⌬CBS, respectively. Similar Mg 2ϩ imaging analyses showed that these ⌬CBS mutants could not stimulate Mg 2ϩ efflux (Fig. 1B). We performed confocal microscopy and confirmed that these ⌬CBS mutants, like WT, were localized mostly to the plasma mem-brane (Fig. 1C). Collectively, these results indicate the importance of the CBS domains in Mg 2ϩ efflux functions.
Disease-associated Amino Acid Substitution in the CBS Domains Completely Abrogates Mg 2ϩ Efflux-Mutations in the CNNM2 gene have recently been reported to be associated with human familial dominant hypomagnesemia (6). Two different mutations have been shown to occur in patients, and one of the mutations is an amino acid substitution of the threonine 568 residue with isoleucine, which is located in the CBS domains of CNNM2 (6). As shown in Fig. 2A, this threonine residue is completely conserved among CNNM family members across various species, ranging from bacteria to humans (indicated with the asterisk). To assess the importance of this mutation in the Mg 2ϩ efflux function, we constructed the amino acid substitution mutants CNNM2-T568I and CNNM4-T495I, which is an equivalent mutant of CNNM4, and performed Mg 2ϩ -imaging analyses with Magnesium Green, as described previously. Almost no Mg 2ϩ efflux was observed in the cells expressing CNNM2-T568I or CNNM4-T495I (Fig. 2B). These mutant proteins, like WT proteins, were also localized mostly to the plasma membrane (Fig. 2C). These results confirm the importance of the highly conserved and disease-associated threonine residue in Mg 2ϩ efflux functions, further supporting the role of the CBS domains. ATP Binds to the CBS Domains of CNNM2 in a Mg 2ϩ -dependent Manner-Adenine nucleotides have been shown to interact with the CBS domains of several proteins and to regulate their activities (10,15). To investigate the possibility that adenine nucleotides can also interact with the CBS domains of CNNM family proteins, we performed binding analyses using SPR. Recombinant proteins of the CBS domains of wild-type CNNM2 (CBS-WT) were immobilized on the surface of a sensor chip, and various concentrations of ATP were injected over the SPR sensor surface as an analyte by using the running buffer that does not contain any divalent cation. However, no significant response was observed (data not shown). Because there are examples of proteins for which binding to ATP depends on the presence of Mg 2ϩ (24 -27), we determined to examine the interaction in the presence of Mg 2ϩ . Before ATP binding analyses, we checked SPR responses using Mg 2ϩ itself as an analyte. The results in Fig. 3A show a gradual and dose-dependent increase of responses, indicating the occurrence of direct interaction between Mg 2ϩ and the CNNM2 CBS domains. However, this increase of the SPR response never reached plateau even at 10 mM, and thus, it was impossible to determine the K d value for the interaction within the physiological range (0.5-1 mM). We then performed SPR analyses to examine the interaction of CBS-WT with ATP by setting the concentrations of Mg 2ϩ to 0.5 mM and 10 mM. As shown in Fig. 3B, significant and dose-dependent SPR responses were observed, and the levels of  the responses were similar, indicating that ATP can bind to the CNNM2 CBS domains under both conditions. Thus, we determined to set the concentration of Mg 2ϩ in the running buffer as 10 mM in the following experiments. First, we quantitatively compared the ATP binding ability of CBS-WT and CBS-T568I. In the absence of Mg 2ϩ , we observed no SPR response (Fig. 4A).
In the presence of Mg 2ϩ , significant and concentration-depen-dent SPR responses were observed from the CBS-WT surface but not from the CBS-T568I surface (Fig. 4B), indicating the occurrence of the Mg 2ϩ -dependent interaction between the CNNM2 CBS domains and ATP, which is abolished by the disease-associated mutation. As one adenine nucleotide typically binds to two tandem CBS domains (10), we fitted the steadystate plot of the CBS-WT surface response in the presence of Mg 2ϩ to a 1:1 binding model (Fig. 4C) and estimated K d of 141 Ϯ 12 M (represented as the mean Ϯ S.E. from three independent experiments), which is within the range of those reported for other CBS domains (15).
To further confirm the Mg 2ϩ dependence of ATP binding to the CBS domains of CNNM2, we performed SPR analysis with sequential changes of Mg 2ϩ concentration in the running buffer. ATP was first injected in the running buffer without Mg 2ϩ and then in a buffer containing Mg 2ϩ , and changes in SPR responses were observed. As shown in Fig. 4D, significant SPR responses were observed with CBS-WT just after the injection of Mg 2ϩ , and the amplitude of the responses increased in a manner dependent on the ATP concentration. In contrast, there was no significant SPR response observed from CBS-T568I even after the Mg 2ϩ injection. Thus, ATP binding to the CBS domains of CNNM2 requires the presence of Mg 2ϩ , indicating that the CBS domains of CNNM2 bind to ATP in a Mg 2ϩ -dependent manner.
The Specificity and Uniqueness of ATP Binding to the CBS Domains of CNNM2-To further ascertain the Mg 2ϩ dependence of ATP binding to the CBS domains of CNNM2, we performed filter binding assays using [␣-32 P]ATP as a ligand to detect the amount of ATP bound to proteins trapped on the filter. The results of assays in the presence of 10 mM Mg 2ϩ showed a clear saturated binding curve with apparent K d of 159 Ϯ 28 M, whereas no binding was observed in the absence of Mg 2ϩ (Fig. 5A and Table 1). 10 mM Mg 2ϩ is much more than the normal cytosolic concentration of Mg 2ϩ , and thus, we also performed filter binding assays using the same concentrations of Mg 2ϩ as those of ATP. As shown in Fig. 5A, CBS-WT showed similar levels of ATP binding signals compared with the experiments using constant 10 mM Mg 2ϩ , when the concentrations of ATP and Mg 2ϩ were set to 200, 400, and 800 M, which are far above the reported K d value between ATP and Mg 2ϩ (ϳ78 M) (28). As most cytosolic ATP forms complexes with Mg 2ϩ and the free cytosolic Mg 2ϩ level is ϳ0.5-1 mM, we concluded that the interaction between ATP and CBS-WT is physiological. In contrast, the affinity of CBS-T568I to ATP was significantly lower than that of CBS-WT (Fig. 5B), and the estimated K d value was Ͼ1000 M (Table 1). These data are consistent with the results shown in Fig. 4 and confirm the notion that the CBS domains of CNNM2 bind specifically to ATP in a Mg 2ϩdependent manner.
The CBS domain is evolutionarily conserved and present in the proteome of archaebacteria, prokaryotes, and eukaryotes, and there are a number of reports stating the binding of CBS domains to ATP (10). However, there have been no reports, to date, showing such clear Mg 2ϩ dependence of this interaction as in the case of CNNM2. To examine whether the Mg 2ϩ -dependent interaction is a unique feature of the CBS domains of CNNM2, we also performed ATP binding assays using two other typical CBS domains derived from ClC-5 and IMPDH2, both of which are reported to bind to ATP (15,19), in the presence or absence of Mg 2ϩ . As shown in Fig. 5, C and D (Table 1). Therefore, the affinities of the CBS domains of these proteins for ATP were stronger in the absence of Mg 2ϩ than in the presence of Mg 2ϩ . These results support the notion that Mg 2ϩ dependence of the CBS domains of CNNM2 for binding to ATP is a unique feature among the CBS domains. To investigate whether this unique feature is conserved among CNNM family proteins, we also performed filter binding assays using CBS domains derived from CNNM1, CNNM3, and CNNM4. In the presence of Mg 2ϩ , all of these proteins, except for CNNM3, showed saturable binding with ATP (Fig. 5E). K d values were 43.4 Ϯ 8.9 M for CNNM4 and 915 Ϯ 389 M for CNNM1 (Table 1). Interestingly, these K d values inversely correlate with the capacities for Mg 2ϩ efflux shown in Fig. 1A. In to that of ATP for 20 min at 25°C. Samples were filtered through a nitrocellulose membrane, and the amount of ATP bound to each filter was quantitated by performing liquid scintillation counting. Nonspecific binding was determined by performing control assays with GST and was subtracted. The data were normalized to the percentage of ATP bound at saturation and are represented as the means Ϯ S.E. (error bars) of three independent experiments. B, filter binding assays using GST-CBS-WT and GST-CBS-T568I in the presence of 10 mM Mg 2ϩ were performed as in A. C and D, filter binding assays using the CBS domains of ClC-5 (C) and IMPDH2 (D) in the presence or absence of 10 mM Mg 2ϩ were performed as in A. E, filter binding assays using the CBS domains of CNNM1-4 in the presence or absence of 10 mM Mg 2ϩ were performed as in A. F, competition assays were performed to examine the binding of GST-CBS-WT to ATP or AMP. ATP or AMP (0.4, 0.8, 1.6, or 3.2 mM) was added to reactions containing 5 M [␣-32 P]ATP, 10 mM Mg 2ϩ , and GST-CBS-WT, and the assays were performed as in A. The data are represented as a percentage of specific binding in the absence of any competitor, and the means Ϯ S.E. (error bars) of three independent experiments are shown. the absence of Mg 2ϩ , no binding was observed from all of the proteins (Fig. 5E). These results confirm the Mg 2ϩ dependence of the interaction of the CNNM CBS domains with ATP and suggest its importance in Mg 2ϩ efflux. In general, CBS domains that bind to ATP can also bind to other adenine nucleotides, such as AMP (15). To examine whether the CBS domains of CNNM2 bind to AMP, we performed filter binding assays using [ 3 H]AMP as a ligand in the presence of Mg 2ϩ . Saturable binding curves were obtained from ClC-5 and IMPDH2 with K d values of 241 Ϯ 56 M and 141 Ϯ 58 M, respectively (Table 1), demonstrating their interactions with AMP as reported previously (15,19). In contrast, AMP binding to CBS-WT of CNNM2 was not saturable, and we could not determine the K d value (data not shown). Therefore, we performed competition assays using nonradiolabeled ATP or AMP to serve as a competitor against [␣-32 P]ATP (5 M). Although nonradiolabeled ATP effectively competed with [␣-32 P]ATP, even 3.2 mM AMP (the maximum concentration we tested) could only outcompete 24% of [␣-32 P]ATP, indicating that the IC 50 value of AMP binding to the CBS domains of CNNM2 is significantly Ͼ3.2 mM (Fig. 5F). Therefore, AMP does not virtually bind to the CBS domains of CNNM2. Collectively, the interactions between ATP and the CNNM2 CBS domains require the terminal phosphate groups of ATP, which are a unique feature of CNNM2. Because Mg 2ϩ is known to form complexes with ATP through interactions with the phosphate groups of ATP, these results suggest that Mg-ATP is the true ligand for the CNNM2 CBS domains.
Electronegative Potentials around the ATP Binding Site of the CBS Domains of CNNMs-There are several examples of proteins that require Mg 2ϩ to bind to ATP. In the cases of sarcoplasmic reticulum Ca 2ϩ -ATPase and 6-hydroxymethyl-7,8-dihydropterin pyrophosphokinase (26,27), Mg 2ϩ neutralizes the electrostatic repulsion force between the electronegative surface potentials around the ATP binding site and negatively charged phosphate groups of ATP. Therefore, we compared the electrostatic surface potentials of the ATP binding sites between CNNM2, ClC-5, and IMPDH2. The electrostatic surface potentials were calculated for homology models of the CBS domains of CNNM2 and IMPDH2 based on the crystal structures of their bacterial homologs (21) and the previously solved crystal structure of the CBS domains of ClC-5 in complex with ATP (19). As shown in Fig. 6A, CNNM2 showed highly electronegative surface potentials around the ATP binding site, whereas both ClC-5 and IMPDH2 showed highly electropositive surface potentials, which would be suitable for the electrostatic interactions with the phosphate groups of ATP. This clear difference in the surface potentials presumably explains the unique requirement of Mg 2ϩ for the interactions between ATP and the CNNM2 CBS domains. We also calculated the electrostatic surface potentials for homology models of the CBS domains of CNNM1 and CNNM4. As shown in Fig. 6B, both CNNM1 and CNNM4 showed similar electronegative potentials around the ATP binding site as did CNNM2, which is consistent with their unique requirement of Mg 2ϩ for the interactions with ATP.

DISCUSSION
In this study, we have shown that the CBS domains of CNNM2 bind to ATP in a Mg 2ϩ -dependent manner, which presumably contributes to the Mg 2ϩ efflux function of CNNM2. The possibility of ATP binding to the CBS domains of CNNM2 was theoretically predicted in a previous report by using a homology model, which was based on the structural data of B. parapertussis CorC (29). In this model, the evolutionary conserved threonine (Thr-568) is located at the region predicted to form the ATP binding pocket and also forms a hydrogen bond that stabilizes the position of the Glu-570 and Asp-571 residues. The substitution of Thr-568 into the bigger isoleucine residue was predicted to cause steric bumps with the ATP molecule and also alter the position of the Glu-570 and Asp-571 residues. Therefore, the substitution was proposed to severely affect ATP binding. In this study, we confirmed these predictions experimentally by performing binding analyses; ATP was determined to bind to the CBS domains of CNNM2, and this binding was abrogated by T568I mutation. Moreover, we further demonstrated the Mg 2ϩ dependence of this binding, which was not accounted for in the predicted model. To the best of our knowledge, Mg 2ϩ dependence for binding to ATP has never been reported to occur in other CBS domains. Comparative analyses of electrostatic surface potentials of the CBS domains revealed the presence of highly electronegative potentials around the ATP binding site of the CBS domains of CNNM2, FIGURE 6. Comparison of electrostatic surface potential maps of the ATP binding sites. A, the electrostatic surface potential maps of the entire CBS domains (upper) and magnified images around the ATP binding site (lower) are visualized with colors ranging from blue to red (from ϩ2 kT/e to Ϫ2 kT/e). ATP is represented as sticks in atomic coloring (blue, nitrogen; green, carbon; orange, phosphorus; and red, oxygen). To calculate electrostatic surface potentials, the following structures of CBS domains in complex with ATP were used: a homology model based on the crystal structure of CorC from B. parapertussis (PDB ID 3JTF) for CNNM2, the crystal structure of human ClC-5 (PDB ID 2J9L) for ClC-5, and a homology model based on the crystal structure of IMPDH from P. aeruginosa (PDB ID 4DQW) for IMPDH2. B, the electrostatic surface potential maps of the entire CBS domains (upper) and magnified images around the ATP binding site (lower) of CNNM1 and CNNM4 were created and visualized as CNNM2 in A.
which may explain the unique requirement of Mg 2ϩ in the case of CNNM2.
So far, the biochemical importance of the interaction between CNNM2 and ATP remains unknown. Notably, AMPK binds to AMP or ADP with higher affinity than that of ATP (in the presence of Mg 2ϩ , similar to the physiological condition) and is regulated in response to changes in the intracellular AMP/ATP or ADP/ATP ratio, enabling AMPK to respond dynamically to the energetic state of cells (17,18). Thus, the CBS domains of AMPK are believed to function as a cellular energy sensor (30,31). However, in the case of CNNM2, the CBS domains have a unique feature such that they selectively bind to ATP with a relatively low K d of ϳ150 M, which is considerably lower than the ATP concentrations typically found in cells (1-10 mM) (32), and virtually does not bind to AMP. Therefore, the CBS domains of CNNM2 are considered to constitutively bind to ATP irrespective of the energetic state of cells. Another unique feature of the CBS domains of CNNM2 is the strict Mg 2ϩ dependence of its binding ability to ATP. In mammalian cells, the majority of intracellular Mg 2ϩ is present in the form of complexes with various intracellular molecules, such as phospholipids, proteins, and nucleic acids (28,33). Among these Mg 2ϩ -binding biomolecules, ATP represents a major component because of its abundance (1-10 mM) and relatively high Mg 2ϩ binding affinity (K d of ϳ78 M); hence, most ATP is considered to exist as Mg-ATP within the cells (28). Taking this into account, we hypothesize that the CBS domains of CNNM2 may utilize their ATP binding ability for efficiently recruiting Mg 2ϩ . Indeed, intracellular Mg-ATP (1-10 mM) is much more abundant than free Mg 2ϩ (0.5-1 mM) (28), and the size of Mg-ATP is much larger than that of Mg 2ϩ , which would be beneficial for molecular recognition. Once Mg 2ϩ is dissociated from Mg-ATP, it is supplied as a substrate for efflux, and Mg 2ϩ -free ATP dissociates from the CBS domains due to the loss of affinity. Then, another Mg-ATP can immediately occupy the binding site. Repeating this sequential process, Mg 2ϩ can be efficiently supplied to facilitate Mg 2ϩ efflux via CNNM2. This hypothesis should be tested in the future by much more detailed studies, such as structural analyses.
The T568I mutation of CNNM2 associated with human hereditary hypomagnesemia abrogated not only its ATP binding ability but also its Mg 2ϩ efflux function, suggesting the physiological importance of Mg 2ϩ efflux by CNNM2 in the regulation of magnesium homeostasis at the organismal level. In a previous study, we showed that CNNM4 localizes to the basolateral membrane of intestinal epithelial cells and mediates transcellular Mg 2ϩ transport across epithelia by extruding intracellular Mg 2ϩ to the outside of cells (7). In addition, CNNM4-knock-out mice exhibited mild hypomagnesemia due to the intestinal malabsorption of magnesium. Considering these similarities between CNNM2 and CNNM4, we speculate that CNNM2 also mediates transcellular Mg 2ϩ transport across epithelia, like CNNM4 does. In support of this possibility, two previous papers have reported a strong expression and localization of CNNM2 at the basolateral membrane of distal convoluted tubule cells in the kidney (6,29), where renal reabsorption of magnesium by transcellular Mg 2ϩ transport is known to occur. CNNM2 presumably plays an important role in the renal reabsorption of magnesium by promoting Mg 2ϩ efflux from the distal convoluted tubule cells to the inner parts of the body, which should be the focus of future studies on CNNM2.