ARL15 modulates magnesium homeostasis through N-glycosylation of CNNMs

Cyclin M (CNNM1-4) proteins maintain cellular and body magnesium (Mg2+) homeostasis. Using various biochemical approaches, we have identified members of the CNNM family as direct interacting partners of ADP-ribosylation factor-like GTPase 15 (ARL15), a small GTP-binding protein. ARL15 interacts with CNNMs at their carboxyl-terminal conserved cystathionine-β-synthase (CBS) domains. In silico modeling of the interaction between CNNM2 and ARL15 supports that the small GTPase specifically binds the CBS1 and CNBH domains. Immunocytochemical experiments demonstrate that CNNM2 and ARL15 co-localize in the kidney, with both proteins showing subcellular localization in the endoplasmic reticulum, Golgi apparatus and the plasma membrane. Most importantly, we found that ARL15 is required for forming complex N-glycosylation of CNNMs. Overexpression of ARL15 promotes complex N-glycosylation of CNNM3. Mg2+ uptake experiments with a stable isotope demonstrate that there is a significant increase of 25Mg2+ uptake upon knockdown of ARL15 in multiple kidney cancer cell lines. Altogether, our results establish ARL15 as a novel negative regulator of Mg2+ transport by promoting the complex N-glycosylation of CNNMs. Supplementary Information The online version contains supplementary material available at 10.1007/s00018-021-03832-8.


Introduction
Mg 2+ is an essential cation for all living organisms. Mg 2+ deficiency, therefore, causes organ dysfunction and is associated with a wide range of human diseases [1]. As co-factor of ATP, Mg 2+ is involved in over 600 enzymatic reactions and as such it plays important roles in a plethora of biological mechanisms including DNA transcription, protein synthesis and energy metabolism [1]. For example, decreased intracellular Mg 2+ levels suppress cell cycle progression and are consequently associated with cell growth impairment [2]. Recent studies therefore have focused on intracellular Mg 2+ concentration as a central mechanism for cellular metabolism and proliferation [3].
Among the various mechanisms of Mg 2+ sensing and transport, proteins of the Cyclin M (CNNM) family were shown to play an important role in the intracellular sensing of the Mg 2+ availability [4,5]. The CNNM family has 4 members in mammalians, which are highly conserved but have different tissue distribution [3][4][5][6]. CNNM1, CNNM2 and CNNM4 have highest expression levels in brain, kidney and intestine, respectively, whereas CNNM3 shows a ubiquitous expression pattern [7]. Genetic defects of CNNMs have been demonstrated to be causative for disease. Mutations in CNNM2 are causative for a multi-organ syndrome comprising intellectual disability, seizures and renal Mg 2+ wasting [OMIM: #616418] [8]. Patients with CNNM4 mutations suffer from Jalili syndrome, consisting of cone-rod dystrophy and amelogenesis imperfecta [OMIM: #217080] [9].
CNNMs interact with phosphatases of the regenerating liver (PRL-1-3) via the cystathionine-β-synthase (CBS) domain of CNNMs [10,11]. CNNMs and PRLs have been associated with progression of breast and colon cancers [11,12]. Although the exact function of CNNM proteins is under debate [13,14], it has been shown that engineered and naturally occurring mutations in CNNMs result in decreased Mg 2+ transport in both cell and animal models [8,11].
Recently, a large GWAS identified the ADP-ribosylation factor-like GTPase 15 (ARL15) locus to be associated with urinary Mg 2+ excretion [15]. ARL15 is structurally similar to Ras-related GTP-binding proteins, which regulate intracellular vesicle trafficking [16,17]. Within the kidney, ARL15 is highly expressed in the thick ascending limb (TAL) and distal convoluted tubule (DCT), like CNNM2, where Mg 2+ reabsorption is tightly regulated [15]. However, the exact function of ARL15 and the mechanism by which ARL15 regulates renal Mg 2+ handling are still unknown.
In this study, we identified CNNMs as direct interacting partners of ARL15 by binding to the CBS1 domain of CNNM2. We demonstrate that ARL15 is a negative regulator of cellular Mg 2+ transport through its modulation of CNNMs N-glycosylation and the inverse correlation of its expression to levels of intracellular Mg 2+ .

ARL15 is a new binding partner of CNNMs
To identify protein interactions of CNNMs, pull-down of mouse CNNM2 carboxyl (C)-terminal region was performed followed by mass spectrometry (Fig. 1a). Given that CNNM2 is known to regulate Mg 2+ reabsorption in the DCT segment of the nephron, DCT-enriched kidney lysates were used. Twenty-four interacting partners of CNNM2 were detected, including ARL15 (Fig. 1b). Since ARL15 has been shown to be involved in urinary Mg 2+ excretion [15], we decided to further explore this particular interacting partner. To confirm the interaction between ARL15 and CNNMs, BioID of ARL15 was performed and we identified 221 proteins as potential interacting partners of ARL15 and CNNM2, CNNM3 and CNNM4 were among them (Supp. Fig. 1). The Gene Ontology (GO) biological process overrepresentation analysis indicated that Mg 2+ homeostasis is one of the most highly enriched processes (Fig. 1c). Immunoprecipitation demonstrated that CNNM3 and ARL15 interact endogenously (Fig. 1d). To further confirm this association, co-immunoprecipitation assays were performed with overexpressed ARL15 and the four members of the CNNM family. ARL15 binding was observed to a similar extent for each CNNM protein (Fig. 1e). Endogenous PRL-1 and PRL-2, which are well described interacting partners of CNNMs that regulate Mg 2+ flux [11,12], also co-precipitated with CNNMs and ARL15 (Fig. 1e). The above data indicate that ARL15 is a novel interacting partner of the four members of the CNNM family of Mg 2+ modulators.

ARL15 interacts with CNNM2 cytoplasmic region
The topology for CNNMs consists of a signal peptide, three transmembrane domains (TM) and an intracellular C-terminus containing the Bateman module (which in turn is formed by two consecutive CBS domains, CBS1 and CBS2) and a cyclic nucleotide monophosphate-binding homology domain (CNBH) domain) [7]. To determine which domain of CNNM2 is important for the interaction with ARL15, truncated CNNM2 proteins were co-immunoprecipitated with ARL15 (Fig. 2a). The presence of the CBS domains is essential for the binding of ARL15 to CNNM2. The interaction of the proteins decreased by 36% when the CBS2 domain is removed (Fig. 2b, c). No interaction with the transmembrane domain was observed. The interaction Fig. 1 ARL15 interacts with CNNMs. a Immunoblot of GST-mCNNM2 c-tail pGex and empty pGex was used as control. The C-tail of mCNNM2 is purified and bound to glutathione beads. The DCT-enriched fraction from 8 parvalbumin GFP mice were added to the beads. b List of binding proteins to mCnnm2 C-tail (emPAI value). c Representative pathways related to metal ion transport and Golgi-trafficking identified using gene ontology biological process overrepresentation analysis with ARL15 interacting partners. d Endogenous CNNM3 was immunoprecipitated from HEK293 lysates and endogenous ARL15 co-immunoprecipitated with it. IgG antibody of the same subclass as CNNM3 antibody was used as negative control. e Anti-FLAG beads were used to immunoprecipitate overexpressed FLAG-tagged CNNM1-4 proteins. The blot shows that ARL15 interacts with CNNM1-4, as well as endogenous PRL-1 and PRL-2. emPAI Exponentially modified protein abundance index, TCL Total cell lysate, IP Immunoprecipitation, GST Glutathione-S-transferase, ARL15 ADP ribosylation factor-like GTPase 15, CNNM Cyclin M ◂ between ARL15 and the Bateman module of CNNM2 was confirmed by gel filtration chromatography (Fig. 2d,  To determine whether the interaction between ARL15 and CNNMs is dependent on Mg 2+ , ARL15 and CNNM were immunoprecipitated in the presence or absence of Mg 2+ . Results showed that the presence of 1 mM Mg 2+ in the lysis buffer does not affect the association between ARL15 and CNNM2 compared to the condition of the lysis buffer without Mg 2+ (Supp. Fig. 2A). In line with this, ARL15 and CNNM3 binding was similar in both conditions (Supp. Fig. 2B).

Docking models of the CNNM-ARL15 interaction
To assess the role of the various cytoplasmic CNNM2 cytosolic domains as targets, ARL15 docking computations were carried out. The coordinates of the flexible linker joining the Bateman module with the CNBH domain were modelled onto the available X-ray diffraction data of the CNNM2 cytosolic domains. Figure 3 displays the domain arrangement of CNNM2 cytosolic domains (Fig. 3a) as well as the electrostatic potential on their surface (Fig. 3b). Notably, the linkers joining the Bateman domain to the CNBH display a stretch of basic residues. Best score solutions displayed ARL15 interacting with both CNNM2 monomers (Supp. Movie 1). ARL15 was partly inserted into the cleft between the Bateman and the CNBH domains (Fig. 3c, d). In each of the two symmetric solutions, the total buried solvent accessible surface per partner was 2175.7 Å 2 -1830.8 Å 2 (932.6 Å 2 , from CBS1, 262.3 Å 2 from the linker and 344.9 Å 2 from the CNBH domain) with one of the two CNNM2 monomers and 344.9 Å 2 (190.6 Å 2 from CBS1* and 154.3 Å 2 from CNBH*) for the interaction with the second CNNM2 monomer. Such values are typical of rather stable complexes, and in the upper limit of non-obligate interactions [18]. In addition, the complex displays complementarity of charges at the rim of the interface, as shown in Fig. 3d.
As pointed out, the CBS1 domains from complementary Bateman modules in the CNNM2 dimer contacted with ARL15, in particular that of the first monomer with a surface region defined between α−helix A, the β1-β2 hairpin loop and α-helix H1 in one of them, but also contacting the second monomer. Figure 3c shows the lowest energy solution found in different simulations. Other solutions with low energies corresponded to the hydrophobic surface of H0 helix pair, which interacts with the lipid bilayer, and spots of the CNBH domains in which the absence of coordinates from the nearby unstructured regions disclosed an artifactual interaction site, otherwise concealed. Notably, Brownian Dynamics computations (Supp. Fig. 3a, Supp. Table1) yielded results very similar to those displayed herein (Supp. Fig. 3). Thus, the interaction model herein is compatible with a 1:1 CNNM2:ARL15 stoichiometry. Other sets of solutions targeted the same region, but displayed different orientations of ARL15 towards CNNM2 surface, and were not found in all simulations. The result in Fig. 3, however, was hardly dependent on the modelled conformation of the linker residues 525 to 546 (Fig. 3d). Nevertheless, in most simulations several positive residues of the linker mapped near the interface. As CNNM3 has been reported to bind ARL15 with an affinity higher than that for CNNM2, HEX docking simulations were also performed using a CNNM3 model (Supp. Fig. 3). Notably, ARL15 targeted the same site in CNNM3 as does in CNNM2, though the orientation differs by a slight rotation of ARL15.
Recently, ARL15 was modelled using two different methods: template-based threading and ab initio modelling, while we applied template-driven simulated annealing [19]. Our ARL15 model is in line with this publication.

ARL15 and CNNM co-localize in the perinuclear region and in kidney DCTs
The subcellular localization for the ARL15-CNNM complex was determined by immunocytochemistry in HEK293 cells. Overexpression of the mCherry-tagged Golgi-marker Beta-1,4-galactosyltransferase 1 (B4GALT1) showed colocalization with ARL15 and CNNM2 in HEK293 cells, suggesting that the function of ARL15 is exerted within the perinuclear region containing the endoplasmic reticulum (ER) and the Golgi apparatus (Fig. 4a). Similar results were obtained with ARL15 and CNNM3, confirming that the proteins predominantly co-localize in the Golgi-apparatus (Fig. 4b). Since CNNM2 has been implicated in Mg 2+ transport in the DCT segment of the kidney [7], we investigated whether ARL15 co-localizes with CNNM2 in this segment of the nephron (Fig. 4c). CNNM2 staining was concomitant ◂ with ARL15 staining in mouse kidney sections. In SK-RC-39 kidney cancer cells, similarly to HEK293, wild-type CNNM3 was co-localized with ARL15; however, the colocalization with CNNM3 N-glycosylation mutant (N73A) was significantly decreased (Fig. 4d, Supp. Fig. 4a). Wild -type CNNM3 and ARL15 co-localize partially with ER (Pearson correlation of 0.72 and 0.46, respectively) and Golgi (Pearson correlation of 0.65 and 0.61, respectively) and also showed co-localization at the plasma membrane ( Fig. 4d, Supp. Fig. 4b). CNNM3 N73A mutant was mostly trapped in the perinuclear region and did not localize to plasma membrane (Fig. 4d, Supp. Fig. 4b). N-glycosylation of CNNM2 has been shown to regulate CNNM2 membrane trafficking [7] and it appears that it plays a similar role with CNNM3. One of the ARL15-interacting partners identified using BioID was ribophorin I protein (RPN1) (Supp. Fig. 1). RPN1 is part of the oligosaccharyltransferase complex that is involved in N-glycosylation and is found in the ER [20]. To confirm the interaction, co-immunoprecipitation assays with overexpressed CNNM3, ARL15 and RPN1 were performed (Supp. Fig. 5). CNNM3 interacted with RPN1 and ARL15, corroborating the localization of the three proteins to the ER as well (Fig. 4d). Our results demonstrate that ARL15 is localized in the ER and the Golgi-system and one of the enriched GO biological processes identified in ARL15 BioID was "Protein N-linked glycosylation via asparagine" (Fig. 1a), which suggests that ARL15 is involved in the N-glycosylation of CNNMs.

CNNM3 N-glycosylation is modulated by ARL15 and Mg 2+
As common post-translational protein modification, N-glycosylation plays important roles to regulate protein functions. Indeed, CNNM2 has been previously shown to have an N-glycosylation site close to its amino (N)-terminus, and it is necessary to stabilize CNNM2 on plasma membrane [7]. Using NetNGlyc [21], N-glycosylation sites were predicted in CNNM1, CNNM3 and CNNM4. The confirmed site in CNNM2 and the predicted sites in the three remaining CNNM family members were analogous (Supp. Fig. 6), all located in the extracellular N-terminal domain. The only other predicted CNNM3 N-glycosylation site was asparagine 395, which is located in the cytoplasmic CBS2 domain, which cannot be glycosylated due to its cytoplasmic localization. To confirm that CNNM3 does indeed have an N-glycosylation site similar to CNNM2, the predicted asparagine 73 (N73) was mutated to alanine (N73A mutant). Compared to wild-type (WT) CNNM3, the N73A mutant migrated faster, which is indicative of the loss of N-glycosylation [22] (Fig. 5a). Endo H and PNGase F were used to characterize the type of glycans attached to CNNM3 (Fig. 5b). When WT CNNM3 was treated with Endo H, which cleaves high mannose and hybrid oligosaccharides from N-linked glycoproteins, the lower band of the doublet shifted further. Cleavage with PNGase F shifted all bands of CNNM3 to a lower molecular mass, indicating that the top band represents the hybrid/complex glycoform (Fig. 5a). To verify that N73 is the only site of N-glycosylation, a lectin gel-shift assay was performed, where concanavalin A lectin is added to the top part of the resolving gel and results in a retardation of N-glycosylated proteins during migration [23]. Wild-type CNNM3 resulted in a smear, indicating a retardation of N-glycosylated protein, whereas the N-glycosylation mutant N73A ran without a smear, indicating that this is the only site of N-glycosylation (Fig. 5c). To assess the effect of ARL15 on CNNM3 N-glycosylation, four different kidney cancer cell lines were used: Caki-1 and RCC4 as well as ACHN and SK-RC-39, belonging to clear cell renal cell carcinoma and papillary renal cell carcinoma, respectively. ARL15 was stably overexpressed in these four cell lines and the pattern of endogenous CNNM3 bands was evaluated. In all four cell lines, overexpression of ARL15 resulted in the increase of hybrid/complex glycoform and a decrease in oligomannose glycoform of CNNM3 (Fig. 5d). Interestingly, the effect of ARL15 expression on CNNM3 glycosylation could be modulated by the Mg 2+ concentration in the growth media (Fig. 5e). When overexpressing ARL15 in SK-RC-39 cells, Mg 2+ -deficient medium resulted in an increase of the oligomannose CNNM3 glycoform. Additionally, CRISPR knockout of ARL15 resulted in a decrease of complex glycoform of CNNM3 as can be seen in the last two lanes (Fig. 5e). Neither overexpression nor knockout of ARL15 resulted in changes to overall protein N-glycosylation (Supp. Fig. 7), indicating that the effect is specific to CNNMs.

ARL15 affects Mg 2+ flux and ATP production
To determine whether ARL15 regulates CNNM-dependent Mg 2+ transport, 25 Mg 2+ uptake was studied. Kidney cancer cells were stably transduced to either overexpress or produce a CRISPR knockout of ARL15. ARL15 overexpression decreased 25 Mg 2+ uptake, while ARL15 knockout significantly increased 25 Mg 2+ uptake more than 3-fold in RCC4 and SK-RC-39 cells (Fig. 6a). Since multiple enzymes involved in ATP production use Mg 2+ as a co-factor and ATP is bound to intracellular Mg 2+ to form Mg-ATP [1], we next assessed the impact of ARL15 expression on ATP production. In RCC4 and SK-RC-39 cells, overexpressing ARL15 resulted in a significant decrease in ATP production (Fig. 6b).
Endogenous cell surface expression of CNNM3 was further confirmed using cell surface biotinylation assay in SK-RC-39 cells with overexpression or CRISPR-mediated knockout of ARL15. The complex N-glycosylated CNNM3 glycoform migrates at approximately 90kDa, whereas oligomannose glycoform of CNNM3 migrates at around 75kDa. Although the plasma membrane expression of CNNM3 was not altered in the presence or absence of ARL15 (Supp. Fig. 8), mainly the oligomannose CNNM3 glycoform was observed at the plasma membrane in the ARL15 knockout cells.
For lentivirus production, HEK293T/17 cells were transfected with a 4:2:1 ratio of lentiviral construct of interest:PAX2:VSV-G. 24 h later, the medium was changed. 48 h later, the medium was collected and filtered through a 0.45 μm filter. To infect the cells, medium was substituted with HEK293T/17 supernatant containing virus with 8 μg/ml polybrene. 48 h later, the medium was changed to fresh medium containing a selection reagent and the selection was carried out for 7 days.
Cell lines used for experiments were obtained from ATCC.

Site-directed mutagenesis
To mutate asparagine 73 (N73) that was predicted to be the site of N-glycosylation of CNNM3 into alanine (A), pcDNA3.1-CNNM3-mCherry or pDONR221-CNNM3 carrying the wild-type sequence without a stop codon was amplified using a single primer PCR. Amplification was carried out with KAPA HiFi polymerase using GC buffer (Kapa Biosystems, Wilmington, MA, USA). The following primer was used: 5'-GGC CCG GGC TTC GCCgcCAG CTC TTG GTC CTG GGT GG-3', indicating the nucleotides necessary to introduce the mutation in lower-case letters. After amplification, the original plasmid was digested with DpnI (New England Biolabs) for 2 h at 37 °C, which was followed by transformation of chemically competent DH5α E. coli. The construct was verified by sequencing.

Pull-down assay
A pull-down assay was performed to determine which proteins of mice DCT interact with CNNM2. pGEX-mCNNM2 c-tail-GST was inserted in bacteria to be able to multiply the protein. Afterwards, bacteria were lysed with lysis buffer (150mM NaCl, 5mM EGTA, Triton 1% (v/v), 1mg/ml pepstatin, 1mM PMSF, 5mg/ml leupeptin, 5mg/ml aproptin, 50mM Tris/HCl, pH 7.5). Glutathione beads (Thermo Fisher, Rockford, IL, USA) were used to purify the samples to obtain only Cnnm2 c-tail. The beads were washed with pull-down buffer (20mM Tris-HCL pH 7.4, 140mM NaCl, 1 mM CaCl 2 , 0.2% (v/v) triton-x-100, 0.2% (v/v) NP-40, 1:1000 pepstatin, 1:1000 Aprotin, 1:400 Leupeptin, 1:100 PMSF) and bacterial lysate was added to the beads followed by an incubation of 3 h at 4 °C. After incubation, beads were transferred into a new tube and washed with pull-down buffer. For kidney tissue, kidneys of 8 PV-GFP mice were dissected and minced in small pieces. The DCT was selected and dissolved in pull-down buffer followed by homogenizing of the mixture using ultra-turrax. After 30 m of incubation on ice, aliquots were taken and centrifuged. Supernatant was added to the beads followed by an incubation overnight rotating at 4 °C. After incubation, beads were washed with pull-down buffer, Laemmli and DTT were added, and Western blot was performed. Migration was stopped when the sample was 1 cm in the running gel. Then samples were cut out and measured with mass spectrometry.

BioID
BioID was performed as previously described [28]. Briefly, cleared lysates from ARL15 A86L expressing Flp-In™ T-REx™ HeLa cells were incubated with streptavidin beads (17-5113-01, 5 ml, GE Healthcare). Trypsin digestion on beads was performed Fig. 4 ARL15 and CNNMs co-localize in the Golgi system. a Immunocytochemistry of HEK293 overexpressing ARL15, CNNM2 and mCherry-tagged Golgi-apparatus marker, B4GALT1. Nuclei are stained with DAPI. b Immunocytochemistry of HEK293 overexpressing ARL15, CNNM3 and mCherry-tagged Golgi-apparatus marker, B4GALT1. Nuclei are stained with DAPI. c Mouse kidneys were permeabilized and co-immunostained with anti-ARL15 and anti-CNNM2. Merged pictures stained for ARL15 in green, CNNM2 in red and DAPI (nuclei) in blue. Bar in figure A and B represents 5 μm, bar in figure C represents 50 μm. Representative image presented of three independent experiments with three replicates. G, Glomerulus, PT, Proximal tubule, DCT, Distal convoluted tubule. d SK-RC-39 cells were co-transfected with ARL15-mClover3, CNNM3-mCherry (WT or N73A N-glycosylation mutant) and pmTurquoise2-ER or -Golgi. Plasma membrane localization is indicated with arrows, colocalization in the Golgi with triangles and co-localization in the ER with asterisks. ◂ and peptides were collected in water, dried in a SpeedVac and pellets were resuspended in 5% formic acid prior to their injection into the mass spectrometers. Digested peptides from HEK293 and HeLa cells were, respectively, injected into LTQ-Orbitrap Velos and Q Exactive (Thermo Fisher) onto a 75 μm i.d. × 150 mm Self-Pack C18 column installed in the Easy-nLC 1000 system (Proxeon Biosystems) coupled to a Nanospray Flex Ion Source, at the IRCM Proteomics core facility.

ARL15 model
A model of the structure of ARL15 was generated by simulated annealing as implemented in MODELLER [31] using as templates the X-ray diffraction coordinates of the complex between human ARL2 and BART (pdb code 3DOE chain A; 36.3 identity) complex between murine ARL2 and PDEδ (pdb code 1KSG chain A; 36.3 identity) murine ARF6 (pdb code 6BBQ; 35.67% identity) human ARF6 (pdb code 2A5G; 35.67% identity). Results were monitored in UCSF Chimera [32]. A single structure-displaying the lowest zDOPE [33] score (− 0.58) and an RMSD value of 3.54 Å with respect to the closest template-out of 100 results was selected. This structure was subjected energy minimization using the Amber 14SB force field [34]. A final check was performed with Coot [35] before deposition and the Mod-elArchive server.

CNNM2 models
To generate alternative linker conformations, in addition to completing sidechains from a SAXS model, simulated annealing computations were carried out using MODEL-LER [31]. 100 conformations were generated, arranged according to their zDOPE [33] scores and tested for collisions with other subunits with UCSF Chimera [32]. At the end, three different conformations were tested. A model of the cytosolic domains of CNNM3 was also generated using the full precursor sequence (NCBI accession code 060093.3) and the XRD coordinates of the CNNM2 and CNNM4 cytosolic domains, obtained from corresponding crystals of the independent cytosolic domains and from SAXS data of the entire cytosolic region of CNNM4 [36]. The dimeric assembly of the complementary Bateman modules used in the model corresponds to the flat (ATP-bound and/or PRL-bound like) conformation of the CBS module (association of two Bateman domains) [36,37]. The modeled region showed a 53.6 % sequence identity with respect to the template. Best model showed 6.1 Á RMSD with respect to the template, and a zDOPE value of 0.39. This positive value is mainly due to the model including flexible loop and linker regions.

Docking models
Docking computations were carried out with benchtop HEX [38] and compared with Brownian dynamics (BD) computations (see Supplemental methods). Hex computations included shape evaluation, in vacuo electrostatics and Decoys As a Reference State (DARS) potentials [39]. In every simulation, 20000 structures were generated, clustered using a RMSD cut-off value of 3 Å and arranged according to their energies. First 100 were selected for analysis. Postprocessing consisted in a short OPLS energy minimization as implemented in the software.

Brownian dynamics
Docking modeling was complemented with Brownian Dynamics computations using the WebSDA server [40] to test if results could be reproduced by other methods. The force-field grids used in BD included all electrostatics and desolvation [41,42] grids. Charges were added with PDB-2PQR [43].
200 trajectories of BD computations were run in docking mode. A total of 500 non-redundant complexes were recorded along the computations. Supplemental table YD displays the summary of the clustering analysis of the distinct complexes found. First cluster corresponds to binding to CNBH domains at regions nearby missing structure, so they are considered artefactual. Cluster 2, however, showed the highest population and-on average-displayed lowest energies, and the lowest dispersion of conformations, according to RMSD values with respect to the representative structure. Supplemental figure ZD illustrates the consistency between the BD computations and docking simulations performed with Hex (see main text).

Glycosidase and tunicamycin treatment
20 µg of protein from SK-RC-39 cells was treated with N-glycosidase F (PNGase F) and Endoglycosidase H (Endo H) (New England Biolabs) according to the manufacturer's instructions. Briefly: protein was denatured at 100 ℃ for 10 m, 1 unit/µl of a glycosidase was added with appropriate buffers and digested at 37 ℃ for 1 h. SK-RC-39 cells grown in DMEM were treated with tunicamycin (Sigma-Aldrich) at 1 ug/ml for 8 h to inhibit N-glycosylation.

Co-Immunoprecipitation
HEK293 or HeLa cells were seeded and transfected in Petri dishes, 24 to 48 h after transfection cells were lysed with lysis buffer. 12 h before collecting the transfected HEK293 cells, carbobenzoxy-Leu-Leu-leucinal (MG132) (Sigma-Aldrich, St. Louis, MO, USA) dissolved in Dimethyl sulfoxide (DMSO) were treated to the HEK293 cells. BCA assay was performed to determine the protein concentration of the lysates. Input samples were taken to be able to check transfection efficiency by performing western blot. For co-immunoprecipitation samples, 30 µl/sample protein A/G plus agarose beads (Santa Cruz Biotechnology, CA, USA) were washed with PBS. And 1.5 µl/sample in mouse raised antibody for human HA (Sigma-Aldrich, St. Louis, MO, USA) was added to the beads. For negative control, no antibody was added to the beads. Samples were then incubated for 2 h rotated at 4 °C. Unbound antibodies were removed by washing with lysis buffer. Then, lysates with the same amount of proteins (as calculated with the BCA test) were added to the beads and incubated overnight rotated at 4 °C. After incubation, lysates and beads were washed with PBS, Laemmli and DTT were added, and western blot was performed. For FLAG-tagged protein immunoprecipitation, for each sample, 20 µg of ANTI-FLAG® M2 affinity agarose gel (Sigma-Aldrich, St. Louis, MO, US) was pre-blocked with 5% (w/v) BSA and 300 µg of total cell lysate (TCL) was pre-cleared with 20 µg of Protein A/G agarose mix, for 2H at 4°C on a rotator. TCL supernatant was collected and incubated with pre-blocked FLAG beads for 2 h at 4 °C on a rotator.

CRISPR-Cas9 experiments
sgRNAs targeting ARL15 were cloned into lentiCRIS-PRv2 plasmid (Addgene #52961) [45] using Golden-Gate Assembly with the Esp3I endonuclease following the standard protocol [46]. For lentivirus production, HEK293T/17 was transfected using Lipofectamine 2000 (Invitrogen, Breda, The Netherlands) with 4:2:1 ratio of lentiCRISPRv2:PAX2:VSV-G. 24 h later, the medium was changed. 48 h later, the medium was collected and filtered through a 0.45 μm filter. To transduce the cells, medium was substituted with HEK293T/17 supernatant containing virus with 8 μg/ml polybrene. 48 h later, the medium was changed to fresh media containing puromycin and the selection was carried out for 7 days. ATP production 5,000 kidney cancer cells were plated per well in a 96-well plate and ATP production was measured after growing overnight. CellTiter-Glo 2 kit (Promega) using manufacturer's instructions.

Mg 2+ uptake
Cells were seeded in Petri dishes and 24 h after transfection, cells were re-seeded into 12-well plates. Two days after transfection, cells were washed once with basic uptake buffer without Mg 2+ (125 mM NaCl, 5 mM KCl, 0.5 mM CaCl 2 , 0.5 mM Na 2 HPO 4 , 0.5 mM Na 2 SO 4 , 15 mM HEPES/ NaOH, pH 7.5), followed by an incubation for 0 or 15 m at 37 °C with basic uptake buffer containing 1 mM 25 Mg 2+ (Cortecnet, Voisins Le Bretonneux, France). After incubation, the basic uptake buffer containing 25 Mg 2+ was removed and cells were washed with ice cold PBS. Then, cells were lysed with 1 ml nitric acid which was then diluted in Milli-Q until nitric acid (Sigma Aldrich, St. Louis, MO, USA) concentration of 10%. Samples were analysed with ICP-MS (inductively coupled plasma mass spectrometry).

Cell surface biotinylation
Petri dishes were coated with poly-l-lysine before seeding and transfection. Two days after transfection, the cells were washed with PBS-CM (100 ml 10x PBS, 1 mM MgCl 2 , 0.5 mM CaCl 2 , pH 8.0 adjusted with NaOH, Milli-Q until total volume of 1000 ml) followed by adding 0.5 mg/ml Sulfo-NHS-LC-LC-biotin (Thermo fisher scientific, Rockford, USA). After 30 min incubation, cells were washed with 0.1% (w/v) PBS-CM BSA and PBS. Cells were then lysed with lysis buffer, and protein concentration was measured with a BCA test. Input samples were taken to be able to check transfection efficiency by performing Western blot. Then, lysates with the same amount of proteins (as calculated with the BCA test) were added with the neutravidin agarose beads (Thermo fisher scientific, Rockford, USA) and incubated overnight rotated at 4 °C. After incubation, protein lysates were washed with lysis buffer and Laemmli and DTT buffer were added and incubated at 37 °C for 30 min before loading the protein samples on SDS-PAGE.

Statistical analysis
Results are expressed as mean ± standard error of the mean (SEM). Biotinylation results of experiments are statistically analyzed by performing one-way ANOVA followed by Tukey as post-test. 25 Mg 2+ uptake results with CNNM2 and ARL15 are statistically analyzed by performing a two-way ANOVA followed by Tukey as post-test. Differences with p < 0.05 were regarded as statistically significant.

Discussion
In this study, we identified ARL15 as a novel interacting partner of CNNMs. ARL15 binds to CNNMs in the ER and regulates their complex N-glycosylation in the Golgi system (Fig. 7). Our data suggest that the complex N-glycosylation of CNNMs reduces their activity at the plasma membrane, Fig. 7 Summary of the effect of ARL15 CNNM complex on Mg 2+ flux. In the presence of ARL15, it interacts with CNNMs in the ER and Golgi, resulting in their complex N-glycosylation, which in turn decreases Mg 2+ uptake. In the absence of ARL15, CNNMs are found in less complex glycoforms and that results in increased Mg 2+ uptake. ER endoplasmic reticulum, ARL15 ADP ribosylation factor-like GTPase 15, CNNM cyclin M. as reflected in a reduced 25 Mg 2+ uptake in renal carcinoma cells. As such, ARL15 may affect cellular Mg 2+ homeostasis and energy metabolism.
ARL15 is a member of the superfamily of ARF-like (ARL) proteins, which have been functionally characterized as small GTPases. Although ARL15 has been relatively scarcely studied to date, many studies have demonstrated that other ARL and ARF proteins are involved in vesicle trafficking [47]. Our immunocytochemistry in HEK293 cells demonstrated that ARL15 is predominantly localized in the plasma membrane and the perinuclear region (Fig. 4), which is in line with previous reports of ARL15 in 3T3-L1 preadipocytes [48]. Moreover, proximity labeling of ARL15 identified many Golgi-specific proteins (Supp. Fig. 1), such as golgins (GOLGA3, GOLGA4, GOLGB1), Golgi SNAP receptor complex member 1 (GOSR1), and Golgi scaffolding protein GOPC, that also interacts with another member of the ARF family ARFRP1 [49]. However, the localization with RPN1, a protein of the oligosaccharyltransferase complex, suggests that the interaction between ARL15 and CNNMs is already formed in the ER. Indeed, several ARF proteins, e.g. ARF1 and ARF3, have been shown to regulate ER-to-Golgi trafficking [50].
The main finding of our study is that ARL15 binds CNNMs and modulates their activity via N-glycosylation. Biosynthesis of glycoproteins commences in the ER, where a pre-assembled oligosaccharide is bound onto the nascent protein by oligosaccharyltransferases. After passing the ER, a series of reactions further assemble the glycan in the medial-Golgi and maturation takes place in the trans-Golgi. Three different types of N-glycans are commonly present on proteins: oligomannose, hybrid and complex N-glycans. Our results demonstrate that ARL15 stimulates the formation of complex N-glycans on CNNMs. Although ARL15 does not possess the enzymatic capacity to change the glycosylation per se, we have identified several proteins of the oligosaccharyltransferase complex in our BioID experiments (e.g. RPN1, STT3A and STT3B) (Fig. 1). Moreover, protein N-glycosylation was among the enriched GO-terms in our analysis. Additionally, LMAN1 was identified in the BioID experiments. LMAN1 is located in the ER-Golgi intermediate compartment (ERGIC) and it participates in ER-to-Golgi trafficking of proteins by binding oligomannose N-glycan moieties of N-glycosylated cargo [51]. We hypothesize that trans-Golgi trafficking via ARL15 is an essential step in the complex glycosylation of CNNMs. A similar mechanism has been postulated for ARFGEF1, which regulates the N-glycosylation and trafficking of integrins [52].
Multiple solute carrier proteins were identified in the ARL15 BioID. These are transmembrane proteins which are N-glycosylated. For example, the importance of N-glycan glycoforms in its localization has been shown for mouse Slc12a2 [53]. N-glycosylation of another putative ARL15 interacting partner SLC26A3 has been implicated in its plasma membrane expression and activity [54].
The N-glycosylation of CNNMs is essential for their plasma membrane expression. CNNMs contain a single N-glycosylation site in the extracellular N-terminal region [7]. Mutation of the Asn73 residue to Ala in CNNM3 completely abrogated its N-glycosylation (Fig. 5a) and its membrane expression (Fig. 4d). Similar results were previously obtained for Asn112 residue in CNNM2 [7]. In this study, we demonstrated that not only the presence of, but also the composition of the glycan is essential for the protein function. ARL15 overexpression resulted in a complex glycosylation of CNNM3, which was accompanied by a significant reduction of 25 Mg 2+ uptake. Vice versa, ARL15 downregulation resulted in oligomannose glycoform and increased 25 Mg 2+ uptake. Altogether, we propose a model in which the composition of the N-glycan affects CNNM activity (Summarized in Fig. 7).
Multiple roles for N-glycosylation of plasma membrane proteins have been described, ranging from effects on membrane trafficking, membrane stability and protein degradation [55,56]. In our experiments, neither ARL15 overexpression nor ARL15 downregulation affected the plasma membrane localization of CNNM3 (Supp. Fig. 7), suggesting that membrane trafficking of CNNM3 is not affected by ARL15. Consequently, changes in N-glycosylation may directly affect the activity of CNNM3 at the plasma membrane. It has been extensively described that N-glycans interact with glycan-binding proteins in the extracellular spaces, such as lectins. Lectin-glycan interactions are involved in many biological cellular processes, such as apoptosis, differentiation as well as regulation of membrane transport [57]. Indeed, binding of lectins and other glycan-binding proteins has been shown to regulate TRPV5-mediated Ca 2+ transport [58,59]. Whether the binding of lectins explains the changes in CNNM3 activity remains to be defined.
The interaction model that we developed is compatible with a 1:1 CNNM2:ARL15 stoichiometry. ARL15 interacts with the Bateman CBS1 domains, in particular with a surface region defined between α−helix A, the β1-β2 hairpin loop and α-helix H1. Indeed, co-immunoprecipitation experiments confirmed that the CBS domains are essential for ARL15 binding (Fig. 2b). The interaction also involves the CNBH domains 3C, thereby explaining the larger affinity observed with the full-length constructs. Overall, the interaction displays high surface complementarity and presents a core of hydrophobic and polar residues surrounded by charged ones. Interestingly, the negative-charged residues present in the CNNM2 docking site is complementarity to the positive-charged surface of ARL15 (Fig. 3d). The interaction of ARL15 with CNNM3 cytosolic domain seems to use the same region. The more negative charge as compared to CNNM2 may explain the larger affinity toward positively charged ARL15. Notably, no overlap is observed between the binding site of ARL15 and PRL, which mostly binds CBS2 by its long loop region comprising β5 and β6 [37] (Supp. Movie 1).
The identification of ARL15 as a novel regulator of CNNM Mg 2+ transport activity is of particular interest in the DCT segment of the kidney. In the DCT, transient receptor potential cation channel subfamily M member 6 (TRPM6) facilitates apical Mg 2+ transport [60]. Basolateral Mg 2+ extrusion in the DCT is regulated by CNNM2. Consequently, mutations in CNNM2 and TRPM6 have been shown to cause renal Mg 2+ wasting in patients [8,61,62]. Interestingly, the ARL15 locus was recently associated with urinary Mg 2+ wasting in a GWAS [15]. In the same study, TRPM6 channel activity was shown to be significantly increased in the presence of ARL15 [15]. Our results demonstrate that ARL15 also binds CNNM2, suggesting that both the apical and basolateral Mg 2+ transport mechanisms are simultaneously regulated. These findings explain how ARL15 determines urinary Mg 2+ excretion.
Interestingly, overexpression of ARL15 resulted in decreased ATP levels (Fig. 6b). Intracellular Mg 2+ and ATP levels are closely associated, as ATP must be bound to this cation to be biologically active [1]. Indeed, previous studies have shown that PRL-2 knockdown and decreased intracellular Mg 2+ levels, reduce intracellular ATP levels and regulate cellular metabolism [3,63]. Given that ARL15 overexpression reduced CNNM-mediated Mg 2+ uptake, the expression of ARL15 may indirectly regulate cellular metabolism. Indeed, ARL15 has been associated with a wide range of metabolic parameters and diseases in GWAS, including adiponectin, HDL, diabetes mellitus and body shape [64][65][66][67][68][69]. As these studies did not analyze Mg 2+ status as a modifying factor, it cannot be excluded that ARL15 has additional functions that explain these associations. Indeed, ARL15 has been demonstrated to modify the insulin-signaling pathway in myotubes [70].
Overall, our work establishes complex N-glycosylation of CNNMs as an essential process to regulate their activity. This crucial post-translational modification promoted by ARL15 on CNNMs adds to recent mechanisms of CNNMs modulation, such as their circadian rhythm expression [71] and their interactions with the PRL family by a magnesiumsensitive mechanism [3,11]. The increasing complexity of CNNMs regulation provides a dynamic system to ensure the correct levels of intracellular Mg 2+ during metabolic changes and cell requirements.
Availability of data and materials ARL15 model is available at Mod-elArchive server https:// doi. org/ 10. 5452/ ma-c4ded. CNNM3 model is available at ModelArchive server https:// doi. org/ 10. 5452/ ma-v6la8. The ARL15 BioID proteomics data is deposited to MassIVE under accession number MSV000087061 and the BirA* control dataset is deposited under accession number MSV000084477.

Conflict of interest
The authors declare that they have no conflict of interest.
Ethics approval The animals were housed at the Radboud Institute for Molecular Life Sciences, Nijmegen, the Netherlands, according the institutional guidelines. The study was ethically approved by the Local Ethical Committee of the Radboud University Nijmegen (RU DEC 2015-0112) and by the Dutch Central Commission for Animal Experiments (CCD, AVD103002016382).
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.