Membrane Topology and Intracellular Processing of Cyclin M2 (CNNM2)*

Background: Mutations in CNNM2 cause severe dominant hypomagnesemia. Results: Structure of CNNM2 consists of an extracellular N terminus and intracellular C terminus containing CBS domains, which are affected by the identified mutations. Conclusion: CNNM2 is intensively processed before being expressed in its final structure at the plasma membrane. Significance: CNNM2 structure analysis will aid to elucidate CNNM2 function in renal magnesium transport. Recently, mutations in the cyclin M2 (CNNM2) gene were identified to be causative for severe hypomagnesemia. In kidney, CNNM2 is a basolaterally expressed protein with predominant expression in the distal convoluted tubule. Transcellular magnesium (Mg2+) reabsorption in the distal convoluted tubule represents the final step before Mg2+ is excreted into the urine, thus fine-tuning its final excretion via a tightly regulated mechanism. The present study aims to get insight in the structure of CNNM2 and to characterize its post-translational modifications. Here, membrane topology studies using intramolecular epitopes and immunocytochemistry showed that CNNM2 has an extracellular N terminus and an intracellular C terminus. This suggests that one of the predicted transmembrane regions might be re-entrant. By homology modeling, we demonstrated that the loss-of-function mutation as found in patients disturbs the potential ATP binding by the intracellular cystathionine β-synthase domains. In addition, the cellular processing pathway of CNNM2 was exposed in detail. In the endoplasmic reticulum, the signal peptidase complex cleaves off a large N-terminal signal peptide of about 64 amino acids. Mutagenesis screening showed that CNNM2 is glycosylated at residue Asn-112, stabilizing CNNM2 on the plasma membrane. Interestingly, co-immunoprecipitation studies evidenced that CNNM2a forms heterodimers with the smaller isoform CNNM2b. These new findings on CNNM2 structure and processing may aid to elucidate the physiological role of CNNM2 in Mg2+ reabsorption in the kidney.

ant 2 (Cnnm2v2, NM_199076), which lacks exon 6. CNNM2 expressed in Xenopus laevis oocytes and in the Mg 2ϩ -deficient Salmonella enterica strain MM281 has been suggested to be involved in Mg 2ϩ transport (6,7). However, electrophysiological analysis using mammalian cells expressing CNNM2 showed Mg 2ϩ -sensitive Na ϩ currents instead of Mg 2ϩ currents (3). This last finding could indicate that instead of transporting Mg 2ϩ itself, CNNM2 might be involved in a Mg 2ϩ -sensing mechanism that, in turn, could regulate renal Mg 2ϩ uptake.
At present, no studies have been reported showing the structure of CNNM2 or any of the other CNNM family members. The aim of this study was to elucidate the membrane topology of CNNM2 and to provide a molecular characterization of the intracellular processing mechanisms.
Total RNA was isolated using TRIzol total RNA isolation agent (Invitrogen). Obtained RNA was treated with DNase (Promega) to remove genomic DNA. Subsequently, reverse transcription of the RNA by M-MLV reverse transcriptase (Invitrogen) was performed for 1 h at 37°C. Gene expression levels were determined by quantitative real-time PCR on a Bio-Rad analyzer and normalized for Gapdh expression levels. Primer sequences are provided in supplemental Table 1.
DNA Constructs-Mouse Cnnm2 was cloned into the pCINeo HA IRES GFP vector as described previously (3). To obtain a C-terminal vesicular stomatitis virus epitope, an XbaI restriction site was introduced before the stop codon using the QuikChange site-directed mutagenesis kit (Agilent, Amstelveen, The Netherlands) according to the manufacturer's protocol. Subsequently, oligonucleotides (forward, 5Ј-GAAT-CGCTTGGGGAAGTAGCTCGAGAATTCCGCCCC-3Ј, reverse, 5Ј-GGGGCGGAATTCTCGAGCTACTTCCCCAAG-CGATTC-3Ј) encoding the vesicular stomatitis virus epitope were hybridized and ligated into the pCINEO HA-CNNM2 IRES GFP vector using the XbaI and XhoI restriction sites. CNNM2 mutations were inserted in the construct using the QuikChange site-directed mutagenesis kit according to the manufacturer's protocol. Human CNNM4 full-length cDNA clone (GenBank TM accession number BC063295) was amplified using Phusion polymerase (Finnzymes, Vantaa, Finland) and primers (forward, 5Ј-CGGCTAGCGCCACCATGGCG-CCGGTGG-3Ј, reverse, 5Ј-CGACCGGTCTATGCGTAGT-CTGGCACGTCGTATGGGTAGATGGCATTCTCGTGG-GAG-3Ј) to introduce an in-frame HA tag before the stop codon and AgeI and NheI restriction sites. The amplicons were cloned into the pCINeo IRES GFP expression vector using the AgeI and NheI restriction sites. All constructs were verified by sequence analysis.
Mouse Cnnm2 was cloned into pcDNA3 (Invitrogen) as described previously (3). To obtain splice variant 2 with a 66-nucleotide deletion corresponding to exon 6, site-directed mutagenesis (QuikChange, Agilent) was used, according to the two-stage PCR mutagenesis protocol described by Wang and Malcolm (11). The same protocol was used to delete the C-terminal tag and to insert internal HA tags with an additional serine residue (NH 2 -YPYDVPDYAS-COOH) (12) after residues Thr-24, Gly-210, or Pro-744 or to replace the C-terminal HA tag with a FLAG sequence. Mutagenesis primer sequences are available upon request. To obtain an N-terminal HA-tagged expression clone, the ORF in pFLCI-mCnnm2WT (3) was amplified using Expand polymerase (Roche Applied Science) using forward primer 5Ј-ATGGGGTACCATGGCTTACCC-ATATGATGTTCCAGATTACGCTATTGGCTGTGGCGC-TTGTG-3Ј containing an in-frame HA tag after the Kozak sequence and reverse primer 5Ј-ATCTTCTAGAGGTCAGA-CCAAGTCCAGGAG-3Ј and subsequently cloned in pcDNA3 as described previously (13). The complete ORFs of all expression constructs were sequence-verified.
Immunocytochemistry-COS-7 cells were seeded on glass coverslips and transfected using Lipofectamine 2000 according to the manufacturer's protocol with constructs expressing epitope-tagged CNNM2 or CNNM4. After 2 days of growth in medium supplemented with 20 mM MgCl 2 , the cells were rinsed with PBS and fixed for 10 min with 4% (w/v) methanolfree formaldehyde solution (Thermo Scientific) in PBS. In a control experiment (supplemental Fig. S1, lower panels), living cells were incubated with primary antibody rat anti-HA (Roche Applied Science, high affinity 3F10, 1:250) for 60 min at 4°C before formaldehyde fixation. After rinses in PBS, some of the cells were incubated for 10 min with 0.2% (v/v) Triton X-100 in PBS for permeabilization. After PBS rinses, the cells were incubated with blocking buffer (5% (v/v) donkey serum (Sigma) in PBS) for 1 h and immunolabeled with the primary antibodies rat anti-HA (1:250), mouse anti-FLAG M2 (Sigma, 1:250), rabbit anti-calnexin (C5C9, Cell Signaling Technology, 1:50), or rabbit anti-CNNM2 (Abcam, Ab111351, 1:50) in blocking buffer for 90 min. After rinses in PBS, cells were incubated with Alexa Fluor 488-conjugated donkey anti-rabbit or anti-mouse IgG or Alexa Fluor 594-conjugated donkey anti-rat IgG secondary antibodies (Invitrogen, 1:500) and DAPI (2 g/ml) for 45 min. After PBS and ethanol rinses, the cells were mounted on slides using ProLong Gold antifade reagent (Invitrogen). Fluorescence microscopy was performed with a Leica DMIRE2 inverted microscope, and images were taken with the Openlab software.
Homology Modeling-A homology model was built using the modeling script in the WHATIF and YASARA Twinset with standard parameters (14,15). We used the structure of a cystathionine B-synthase (CBS) domain pair structure of the Mg 2ϩ and Co 2ϩ efflux protein CorC from Bordetella parapertussis (Protein Data Bank (PDB) file 3jtf) as a template for modeling.
Cell Surface Biotinylation, PNGase Treatment, and Immunoblotting-Biotinylation experiments were performed as described previously (16). In short, HEK293 cells were transfected for 48 h, and cell surfaces were biotinylated at 4°C for 1 h by adding sulfo-NHS-LC-LC-biotin. Subsequently, protein lysates were incubated overnight with NeutrAvidin-agarose beads (Pierce).
Protein lysates were denatured in Laemmli containing 100 mM DTT for 30 min at 37°C and subsequently subjected to SDS-PAGE. Then, immunoblots were incubated with mouse anti-HA (Cell Signaling Technology) primary antibodies and peroxidase conjugated sheep anti-mouse secondary antibodies (Sigma). PNGase F (New England Biolabs) was added to the protein lysates in Laemmli-DTT buffer and incubated at 37°C for 1 h before loading the protein samples on gel.
Co-immunoprecipitation-COS-7 cells were cultured and transfected as described above with expression vectors for HAand FLAG-tagged wild type CNNM2 or CNNM4 isoforms. 48 h after transfection, cells were scraped from dishes in ice-cold 50 mM Tris (pH 8.0) followed by centrifugation at 5000 ϫ g. The pellets were lysed in lysis buffer (150 mM NaCl; 0.1% (v/v) Nonidet P-40; 50 mM Tris, pH 7.5; and Complete protease inhibitor mixture (Roche Applied Science)) by repeated passage through a 26-gauge needle. Protein concentration measurement was performed using a BCA assay (Pierce). The next incubation steps were all done under rotary agitation at 4°C. Protein aliquots of 100 g in lysis buffer were precleared by incubation with protein G-Sepharose (Sigma-Aldrich) for 2 h after which the precleared lysate was incubated for 18 h with 2.5 g of anti-HA antibody (Roche Applied Science, high affinity 3F10). The antibody-lysate mixture was added to 50 l of fresh protein G-Sepharose beads and incubated for 1 h. The beads were collected by centrifugation at 1000 ϫ g for 1 min at 4°C and washed four times with lysis buffer. The proteins were separated from the beads by boiling for 5 min in 1ϫ Laemmli sample buffer and detected by immunoblotting using a mouse monoclonal antibody against FLAG M2 (Sigma-Aldrich, 1:1000).
In Vitro Transcription/Translation-The cDNA template for in vitro transcription contained the Cnnm2-HA sequence downstream of a T7 promoter sequence in a pT7Ts vector. The coupled in vitro transcription/translation reaction was incubated at 30°C for 90 min. The reaction mixture contained 2 g of cDNA template, 2 l of reaction buffer, 1 l of T7 RNA polymerase, 1 l of amino acid mix without methionine, 3 l of [ 35 S]Met (1 Ci/l), RNasin, and 25 l of rabbit reticulocyte lysate (Promega). The reaction was performed with or without signal peptidase inhibitor MeoSuc-Ala-Ala-Pro-Val-chloromethylketone (AAPV-CMK, Sigma). The protein mixture was separated on an 8% (w/v) SDS-PAGE gel. The gel was fixed in a mixture of acetic acid, methanol, and H 2 O (20:10:70) and incubated in 20% (w/v) 2,5-diphenyloxazole in dimethyl sulfoxide (DMSO) to intensify the radioactive signal. The proteins were visualized using CL-XPosure films (Pierce, Etten-Leur, The Netherlands).
Statistical Analysis-In all experiments, data are expressed as means Ϯ S.E. Statistical significance was determined using an unpaired Student's t test, and differences with p Ͻ 0.05 were regarded as statistically significant.

RESULTS
Expression Profile of CNNM Family-To examine the transcript expression levels for CNNM1-4 in various tissues, a mouse tissue cDNA panel was constructed, and Cnnm mRNA expressions were quantified by real-time PCR analysis. Gapdh expression was used to normalize values. Cnnm3v1 and Cnnm3v2 (containing an alternative last exon) show a ubiquitous expression pattern with highest expression in kidney, brain, lung, spleen, and heart ( Fig. 1, D and E). Cnnm1 expression is predominantly found in brain and testis (Fig. 1A), whereas Cnnm4 shows highest expression along the gastrointestinal tract (Fig. 1F). Interestingly, Cnnm2 shows a ubiquitous expression pattern with highest expression in kidney, lung, spleen, and testis (Fig. 1, B and C). CNNM2A and CNNM2B (encoded by Cnnm2v1 and Cnnm2v2, lacking exon 6) share a similar pattern; only minor differences between the tissue expression levels were measured. Because CNNM2 has been described to be causative for hypomagnesemia due to excessive renal Mg 2ϩ wasting (3), its kidney localization was investigated in detail. Using immunohistochemistry, co-expression of CNNM2 with markers of different distal nephron segments was examined: the Na-K-Cl cotransporter NKCC2 (thick ascending limb of Henle's loop), the thiazide-sensitive Na-Cl cotransporter NCC (DCT), calbin-din-D 28k (CNT and cortical collecting duct), and aquaporin-2 (collecting duct). CNNM2 was expressed on the basolateral side of the cells (Fig. 2, A-D, green), opposed to the luminal expression of the marker proteins (Fig. 2, A-D, red). CNNM2 immunopositive tubules were negative for aquaporin-2 staining, but did co-stain with NCC-positive tubules. Only a part of the tubules positive for calbindin-D 28k showed CNNM2 staining. Although most of NKCC2-positive tubules are negative for CNNM2, only a few NKCC2-positive tubules show faint CNNM2 staining. These results indicate that CNNM2 is predominantly expressed in DCT. mRNA expression levels measured by real-time PCR on GFP-sorted DCT and CNT tubules showed an equivalent pattern; expression of Cnnm2v1 and Cnnm2v2 was high in DCT and to a lesser extent also in the CNT (Fig. 2E). As a control for DCT and CNT specificity of the mRNA, we examined the expression of the transcripts for the marker proteins TRPM6 and TRPV5, respectively (Fig. 2E). These and previous results (3) indicate that CNNM2 is predominantly expressed basolaterally in DCT and also, but in lower amounts, present in thick ascending limb of Henle's loop and CNT (Fig. 2F).
CNNM2 Has an Intracellular C Terminus Containing CBS Domains-The consensus topology prediction for the plasma membrane protein CNNM2 indicates five potential transmem-brane (TM) domains and an extracellular C terminus (3), which would locate the CBS domains (see below) in an unprecedented, extracellular position. We decided to investigate the topology in detail using an intramolecular epitope strategy by expression of various epitope-tagged CNNM2 proteins (Fig.  3A) in COS-7 cells with subsequent immunocytochemistry. A similar approach was used successfully to aid in the topology determination of other plasma membrane transporters such as multidrug resistance proteins (17) and an iron transporter (18). Immunocytochemistry of cells overexpressing CNNM2a proteins with HA epitopes at positions 211 (HA211), 745 (HA745), or at the extreme C terminus revealed these proteins to be targeted to the plasma membrane of the cells (Fig. 3B, upper panels), similar to the results obtained in MDCK-C7 cells (3). Importantly, C-terminally HA-tagged CNNM2b showed a localization that is undistinguishable from that of CNNM2a, indicating that the absence of the amino acids translated from exon 6 does not influence sorting to the plasma membrane.
To determine whether the epitopes are located on the inside or the outside of the cells, formaldehyde-fixed cells, both intact and permeabilized, were immunostained for the HA epitope. The HA tag at position 211 could be detected on intact cells (Fig. 3B, lower panels), indicating extracellular localization. On the contrary, detection of HA745 or the epitope at the C termi- nus required cell permeabilization, suggesting that these latter two are located intracellularly. Similar results were obtained when living cells were incubated with anti-HA antibodies prior to fixation (supplemental Fig. S1).
As an alternative approach to confirm that the C terminus is located intracellularly, cells that expressed HA211 were immunostained with Abcam antibody 111351, raised against amino acids 590 -640, located just after the pair of CBS domains. Also here, the epitope could only be detected after permeabilization (Fig. 3C), indicative for an intracellular position of the C terminus that importantly includes the two CBS domains.
To investigate whether the intracellular localization of the C terminus is shared with other members of the CNNM family, the same permeabilization assay was performed on cells transfected with C-terminally HA-tagged CNNM4 (Fig. 3D). Again, the plasma membrane staining was only observed after permeabilization, indicating that also for CNNM4, the C terminus resides inside the cell. In conclusion, the CNNM2 protein has a membrane topology with its C terminus inside of the cell.
CNNM2 N Terminus Contains a Large Signal Peptide and a Glycosylation Site-We observed that the HA epitope at the extreme N terminus or at position 25 of CNNM2a was not detected at the plasma membrane, but rather in distinct perinuclear structures (supplemental Fig. S2), which suggested that the N terminus might contain a signal peptide and, therefore, ultimately results in cleavage of the HA epitope. To exclude incorrect processing of the recombinant protein, a construct was expressed with the HA epitope at position 25 of CNNM2a and a FLAG epitope at the C terminus. We observed that the FLAG epitope was correctly targeted at the plasma membrane, but the HA epitope was retained in perinuclear structures (Fig.  3E). Colocalization studies with an antibody against calnexin (supplemental Fig. S2) showed that these perinuclear structures are part of the endoplasmic reticulum (ER), which is in line with the presence of a signal peptide.
Signal peptide prediction software (SignalP (19)) predicts a long N-terminal signal peptide of 64 amino acids. Examination of the N-terminally HA-tagged CNNM2 construct by immunoblotting confirmed that the CNNM2 N terminus is cleaved (Fig.  4A, upper panel) because the HA tag was undetectable. Mutation of the predicted cleavage region ( 62 GCTA 65 ) to hydrophobic residues 62 LLLV 65 marred this cleavage event, evidenced by detection of an HA-positive band in the mutant (Fig. 4A, upper  panel, right). To identify the exact cleavage site, constructs were generated in which the Cys-63/Ala-65 and the Gly-62/Thr-64 positions were mutated to hydrophobic amino acids. However, none of these two mutations were able to impair the cleavage of CNNM2 (Fig. 4A, upper panel).
To confirm that the signal peptidase complex (SPC, a protein complex expressed on the ER membrane known to be involved in signal peptide cleavage) facilitates cleavage, an in vitro protein translation assay in the presence of SPC inhibitor AAPV-CMK was performed. SPC inhibition by AAPV-CMK prevented signal peptide cleavage, as shown by a higher molecular weight on gel (Fig. 4B). Together, we conclude that SPC is responsible for CNNM2 signal peptide cleavage.
In eukaryotic cells, trafficking of membrane proteins to the cell surface is often regulated by glycosylation of asparagine residues present in extracellular domains. To determine the CNNM2 glycosylation profile, four potential candidate asparagine residues were mutated to alanine, namely Asn-112, Asn-327, Asn-527, and Asn-591 (based on prediction software Net-NGlyc 1.0, Center for Biological Sequence Analysis (CBS)). Subsequently, the presence of N-linked glycans in the mutants was discerned by immunoblotting the proteins after treatment with PNGase F, an enzyme causing N-glycan cleavage. The glycosylated wild type CNNM2 runs at ϳ105 kDa, but PNGase F treatment reduced its molecular mass to ϳ96 kDa (Fig. 5A). Of all the candidates, only residue Asn-112 was identified as a glycosylation target in CNNM2 because the N112A mutation resulted in a molecular mass of 96 kDa (Fig. 5A). PNGase F treatment did not cause a further reduction of the molecular mass, indicating that no N-glycan is present in this mutant. To determine the effect of the N-glycan at position 112 on membrane expression of the CNNM2 protein, cell surface biotinylation studies of the N112A mutant were performed (Fig. 5B,  left panel). A reduction of 90% in plasma membrane expression of the CNNM2-N112A mutant was observed when compared with the wild type CNNM2 protein (Fig. 5B, right panel), suggesting that Asn-112 glycosylation is necessary for CNNM2 membrane stability. Taken together, the CNNM2 extracellular N terminus provides important sites for post-translational modifications necessary for proper plasma membrane expression.
C-terminal CBS Domains Are Disrupted by T568I Mutation-One of the mutations recently described to be causative for hypomagnesemia is located in the CBS domains within the C terminus (3). Therefore, emphasis was placed on determining the structure of the intracellular CBS domains (Fig. 3A) that have been associated with ATP binding, among other functions (20). Here, we provide a homology model of the CNNM2 CBS domains based on the B. parapertussis Mg 2ϩ /Co 2ϩ CorC protein (Fig. 6A). Analysis of this model showed conservation of the potential ATP-binding site in the CBS dimer of CNNM2 (Fig. 6A). Interestingly, the T568I mutation, identified in a Czech family to be causative for hypomagnesemia (3), is lining the ATP-binding site in the CBS domain (Fig.  6B). In our model, mutation of the threonine into the bigger isoleucine residue causes steric bumps with the ATP molecule. Besides that, Thr-568 also forms a hydrogen bond that stabilizes the position of the Glu-570 and Asp-571 residues (Fig. 6B). These two residues form important hydrogen bonds and ionic interactions to the ATP molecule and residues lining the ATP-binding pocket, respectively. According to the model, the T568I mutation of the patient will also affect the position of the Glu-570 and Asp-571 residues and could, therefore, severely affect ATP binding. In conclusion, this model might explain why the T568I mutation of the patient is causative for nonfunctional CNNM2.
CNNM2a and CNNM2b Form Multimers-In GenBank, two splice variants exist that are conserved between human and mouse, the first being full length, the second lacking exon 6. Interestingly, splice variant 2 was consistently found in conjunction with that of splice variant 1 in all tissues examined, including kidney (Fig. 1). Further examination of potential interactions between CNNM2a and CNNM2b was performed by studying the dimerization of CNNM2 isoforms. Interestingly, besides the CNNM2 monomer (with an expected molecular mass of 105 kDa) (5), additional complexes were visualized in the immunoblot at ϳ200 kDa (Fig. 7A). This molecular mass matches the mass of a hypothetical CNNM2 dimer. To confirm CNNM2 oligomerization events, co-immunoprecipitation assays of CNNM2 isoforms were performed. CNNM2a-HA and CNNM2b-HA were each able to co-immunoprecipitate both CNNM2a and C-terminally FLAG-tagged CNNM2b. CNNM4-HA only weakly co-immunoprecipitated, although it was highly expressed, indicating the specificity of these interactions (Fig. 7B). These data show that CNNM2a and CNNM2b are able to form specific homo-and heteromultimeric complexes.

DISCUSSION
Mutations in CNNM2 have, recently, been described to be causative for a severe form of hypomagnesemia (3). Initially described as nuclear proteins involved in cell cycle regulation (5), the family of CNNM proteins has now been shown to reside at the plasma membrane (3,6). Although their physiological role is still poorly understood, all members are likely involved in electrolyte homeostasis. CNNM1 has been identified as a cytosolic copper chaperone, CNNM2 was shown to mediate Mg 2ϩsensitive Na ϩ currents in HEK293 cells, and Cnnm2, Cnnm3, and Cnnm4 have been linked with serum Mg 2ϩ levels (3,21,22). Furthermore, Cnnm2 has been associated with coronary artery disease and hypertension (23)(24)(25). Finally, mutations in Cnnm4 were shown to be causative for recessive cone-rod dystrophy with amelogenesis imperfecta (26,27). In the present study, we aimed to elucidate the CNNM2 protein topology and characterize its post-translational modifications.
Although Gómez García et al. (28) have been working on structural elucidation of the CBS domains of CNNM4, complete structural information for any of the CNNM proteins is presently unknown. Using immunocytochemical methods, we examined the membrane topology of CNNM2. Based on our results showing an extracellular N terminus and an intracellular C terminus (Fig. 3), we propose a structure comparable with that of the glutamate receptor (GluR) ion channels (29), containing three full membrane-spanning regions and an additional re-entrant loop. In silico analysis of the CNNM2 protein sequence showed that of the four hydrophobic regions, the second one is the shortest and the least hydrophobic, indicating that the second hydrophobic region might not be completely membrane-spanning, but instead forms a re-entrant loop (Fig. 8).
Similar to the glutamate receptor family, CNNM2 contains an N-terminal signal peptide that is cleaved in the ER. Here, the SPC seems to be the protease complex responsible for CNNM2 cleavage (Fig. 4B). The human SPC is a complex of five proteins expressed on the ER membrane and recruited post-translationally to the signal recognition particle to perform signal peptide cleavage (30). Our results do not identify the exact position of the cleavage site. Signal peptide recognition is known to be dependent on the residues at positions Ϫ1 and Ϫ3 of the actual cleavage site (31). Nevertheless, conversion of the Ϫ1 and Ϫ3 position of the predicted site (Gly-62/Thr-64) to hydrophobic residues did not prevent cleavage. Likewise, the cleavage was also not impaired when the alternative cleavage site (Cys-63/ Ala-65) was mutated. Only mutation of both sites prevented CNNM2 cleavage, suggesting that the SPC can cleave CNNM2 at the preferential site between Thr-64 and Ala-65, but also at an alternate site between Ala-65 and Ala-66. Thus, CNNM2 contains a remarkable lengthy signal peptide of ϳ64 amino acids. Long signal peptides have been associated with multiple functions, but might serve as a mechanism to achieve proper membrane insertion of the protein (32). This also suggests that the first predicted transmembrane domain (Fig. 3A) actually represents the hydrophobic core of the signal peptide that, after cleavage, will remain at the ER membrane (Fig. 8).
After cleavage in the ER, CNNM2 is further processed by glycosylation in the Golgi apparatus. N-Glycosylation is known to facilitate plasma membrane trafficking for several membrane proteins (33). Here, Asn-112 was identified as the residue that is glycosylated and implicated in CNNM2 membrane stability (Fig. 5, A and B). This can be interpreted as an additional argument showing that the N terminus is located on the extracellular side of the membrane. Interestingly, our immunoblotting data showed a shift of the molecular weight of the CNNM2 protein when the alternative residue Asn-327 was mutated. However, PNGase F treatment still lowers the molecular weight of the N327A mutant, indicating that the protein is not deglycosylated (Fig. 5A). Because Asn-327 is predicted to be located in the second TM domain, we suspect that mutation of this residue instigates an altered protein folding that runs different on gel, explaining the observed shift.
Once at the plasma membrane, the functional CNNM2 unit likely represents a multimer. Interestingly, we showed that both CNNM2a and CNNM2b can form homomultimers as well as heteromultimers (Fig. 7B). This interaction is specific, as evi- denced by the fact that CNNM2a and CNNM2b hardly dimerize with CNNM4, which shares the highest sequence homology with CNNM2 (96.7% similarity and 81.4% identity at the protein level) (5).
In magnesium research, CNNM2 function is heavily debated. CNNM2 was first reported as a Mg 2ϩ transporter (6, 7), but recent findings questioned the ability of CNNM2 to transport Mg 2ϩ and proposed a Mg 2ϩ -sensing function (3). We, therefore, examined whether our topological findings could further elucidate the function of CNNM2. We reasoned that although CNNM2 might contain a re-entrant region, it is unlikely that this region forms a pore for Mg 2ϩ transport. The re-entrant loop does not include negatively charged residues that could provide Mg 2ϩ specificity. Furthermore, the protein topology of only three transmembrane segments would be small in comparison with known eukaryotic Mg 2ϩ transporters such as TRPM6/7 (tetramers of six TM proteins) (2) and SLC41A1 (10/11 TM domains) (34). Thus, our topological findings strengthen the previously reported patch clamp data (3), suggesting that CNNM2 might be a Mg 2ϩ sensor indirectly regulating Mg 2ϩ transport.
To further substantiate this hypothesis, we focused on the CBS domains during analysis of the CNNM2 structure. CBS domains are found in 100 -200 mammalian proteins and play a role in a variety of functions such as adenosine phosphate binding and ionic strength sensing (35,36). The CBS domains in CNNM2 are highly homologous to those found in the B. parapertussis Co 2ϩ /Mg 2ϩ protein CorC, of which the structure has been crystallized recently (PDB 3jtf). As shown by homology modeling based on the CorC structure, the CBS domains in the CNNM2 C terminus contain a potential ATP-binding domain (Fig. 4B). ATP binding in CBS domains has been reported to activate or further inhibit protein function in eukaryotic CBS domain-containing proteins (20).
Our model does not allow discrimination between ATP and Mg-ATP binding, but we expect that the CBS domains of CNNM2 are also able to bind Mg-ATP. This theory is strengthened by the recently elucidated crystal structure of the bacterial Mg 2ϩ transporter MgtE in which its CBS domains are involved in Mg 2ϩ binding (PDB 2yvy (37,38). As the majority of intracellular Mg 2ϩ is ATP bound, the intracellular Mg-ATP concentrations are representative for the cellular Mg 2ϩ content. The suggestion that CNNM2 binds Mg-ATP implies that rather than ATP being an energy source for active transport, the Mg-ATP binding would provide a Mg 2ϩ -sensing mechanism. Given that the CNNM2 T568I loss-of-function mutation found in patients (3) is located at the region predicted to form the ATPbinding pocket (Fig. 6B), the (Mg-)ATP binding is likely of major importance for CNNM2 function. Because the T568I mutation is specifically disturbing the ATP-binding pocket, Mg 2ϩ alone will possibly not be able to activate CNNM2.
In conclusion, we propose a protein topology of CNNM2 consisting of three membrane-spanning domains and a re-entrant loop (Fig. 8). CNNM2 is extensively post-translationally modified by signal peptide cleavage and glycosylation before being functional in multimers at the plasma membrane. Rather than transporting Mg 2ϩ itself, we hypothesize that CNNM2 would sense intracellular Mg 2ϩ concentrations and regulate other Mg 2ϩ transporters. Future research should focus on the identification of CNNM2 protein partners to further elucidate its role in Mg 2ϩ homeostasis.