---
pmid: '42497866'
title: Dynamic dimer-of-dimers architecture defines Mg(2+) transport in human CNNM4.
authors:
- Bai Z
- Zhou XE
- Lü W
- Du J
journal: Cell
year: '2026'
full_text_available: true
full_text_extraction_method: xml
pmcid: PMC13479726
doi: 10.1016/j.cell.2026.06.039
pubmed_publication_types:
- Journal Article
publication_type: PRIMARY_RESEARCH
---

# Dynamic dimer-of-dimers architecture defines Mg(2+) transport in human CNNM4.
**Authors:** Bai Z, Zhou XE, Lü W, Du J
**Journal:** Cell (2026)
**DOI:** [10.1016/j.cell.2026.06.039](https://doi.org/10.1016/j.cell.2026.06.039)
**PMC:** [PMC13479726](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13479726/)

## Abstract

1. Cell. 2026 Sep 17;189(19):6097-6110.e4. doi: 10.1016/j.cell.2026.06.039. Epub
2026 Jul 24.

Dynamic dimer-of-dimers architecture defines Mg(2+) transport in human CNNM4.

Bai Z(1), Zhou XE(1), Lü W(2), Du J(3).

Author information:
(1)Department of Molecular Biosciences, Northwestern University, Evanston, IL
60208, USA.
(2)Department of Molecular Biosciences, Northwestern University, Evanston, IL
60208, USA; Department of Pharmacology, Northwestern University, Chicago, IL
60611, USA; Chemistry of Life Processes Institute, Northwestern University,
Evanston, IL 60208, USA. Electronic address: wei.lu@northwestern.edu.
(3)Department of Molecular Biosciences, Northwestern University, Evanston, IL
60208, USA; Department of Pharmacology, Northwestern University, Chicago, IL
60611, USA; Chemistry of Life Processes Institute, Northwestern University,
Evanston, IL 60208, USA. Electronic address: juan.du@northwestern.edu.

Mg2+ is essential for all living organisms, yet its transport across mammalian
membranes remains poorly understood. Here, we present cryoelectron microscopy
(cryo-EM) structures of a full-length mammalian Mg2+ transporter on the plasma
membrane, human CNNM4, in outward-facing and occluded states, revealing an
unexpected tetrameric assembly organized as a dimer of asymmetric
dimers-distinct from the symmetric dimers in prokaryotic homologs and long
assumed for eukaryotic CNNMs. We show that Mg2+/ATP binding stabilizes the
dynamic intracellular domains and promotes tetramerization, while an acidic
patch binds additional Mg2+, potentially acting as a sensor to couple
cytoplasmic Mg2+ levels to transport activity. Within the transmembrane domain,
a key glutamate flips upon Na+ binding and destabilizes the Mg2+-binding site in
the outward-facing state, thereby promoting Mg2+/Na+ exchange. Together, these
findings establish a mechanistic framework for CNNM transport and regulation
that diverges from prokaryotic models and links CNNM function to human
physiology and disease.

Copyright © 2026 The Author(s). Published by Elsevier Inc. All rights reserved.

DOI: 10.1016/j.cell.2026.06.039
PMCID: PMC13479726
PMID: 42497866 [Indexed for MEDLINE]

Conflict of interest statement: Declaration of interests The authors declare no
competing interests.

## Full Text

IntroductionMagnesium (Mg2+) is an essential cation that supports fundamental cellular processes, including energy metabolism, nucleic acid synthesis, enzyme catalysis, and ion transport regulation1–4. Intracellular free Mg2+ concentration is tightly regulated in mammalian cells, and disruptions in Mg2+ homeostasis can lead to serious physiological consequences. Such disruptions often arise from defects in magnesium transport or systemic homeostatic control, resulting in hypomagnesemia, a condition associated with cardiovascular disorders, neurological and psychiatric disorders such as seizures and depression, as well as metabolic and endocrine disorders like type II diabetes and insulin resistance5–8.Despite its importance, the mechanisms of Mg2+ transport across the plasma membrane remain poorly studied, particularly in mammals. No full-length structures of mammalian Mg2+ transporters and channels have been reported9–12. Among them, the Cyclin and CBS Domain Magnesium Transport Mediator (CNNM) family—comprising CNNM1, 2, 3, and 4—has emerged as a central regulator of Mg2+ efflux in human and is evolutionarily conserved across prokaryotic and eukaryotic organisms13–15. In mammals, they mediate Mg2+ export and also interact with proteins such as Phosphatase of Regenerating Liver (PRL) proteins, thereby linking CNNM activity to cancer progression16,17. More recently, CNNMs have been shown to assemble with TRPM7 and TRPM618–20, two critical Mg2+-permeable channels at the plasma membrane, forming a remarkable transporter-channel supercomplex thought to be coordinate the bidirectional regulation of cellular Mg2+ homeostasis.Consistent with their molecular and physiological functions, genetic studies highlight the critical roles of CNNMs in human health and disease. Genetic ablation of CNNMs in mice disrupts intestinal Mg2+ extrusion, leading to hypomagnesemia and profound physiological abnormalities that range from developmental defects, infertility, seizures, and cognitive deficits to embryonic lethality21–31. Moreover, dysregulation of CNNMs has also been implicated in a broad spectrum of conditions, including hypertension32,33, schizophrenia34, cancer16,17,35,36, and liver disease37, underscoring their broad impact on human health. Of note, mutations in CNNM4 underlie Jalili syndrome, an autosomal recessive disorder characterized by cone-rod dystrophy and amelogenesis imperfecta38–43.What sets CNNMs apart from classical transporters is their multi-domain architecture13,14. Each CNNM subunit consists of four domains: an extracellular domain (ECD), a transmembrane domain (TMD), and two intracellular regulatory domains—the cystathionine-β-synthase (CBS)-pair (CBSP) domain, and the cyclic nucleotide-binding homology (CNBH) domain. In contrast, prokaryotic homologs such as CorB/CorC contain only the TMD and CBSP, but lack the N-terminal ECD and C-terminal CNBH44,45. Prior studies of eukaryotic CNNMs focused solely on isolated domains, showing that only MgATP engages CBSP in solution and drives its dimerization—a prerequisite for transport activity, whereas MgADP, MgAMP, or nucleotide alone do not46. The ECD also mediates dimerization, and disruption of its interface abolishes transport activity47. Consistent with these findings, structures of the prokaryotic homolog CorB/CorC indeed revealed a symmetric homodimeric achitecture44,45.While studies of isolated domains and prokaryotic homologs have shaped the prevailing view that CNNMs function as homodimers13,14, the absence of a full-length CNNM structure leaves fundamental questions unanswered. The oligomeric assembly and regulatory mechanisms of CNNMs remain unresolved, and it is unclear whether they indeed adopt symmetrical homodimers like their prokaryotic counterparts. The functional contribution of the eukaryotic-specific ECD and CNBH domains, the interdomain coordination across the full-length protein, and the structural basis of key transport states—such as outward-facing and occluded conformations—also remain undefined. Furthermore, how CNNMs integrate intracellular energy status and local Mg2+ levels to fine-tune transport activity is still unclear.Here, we present cryo-EM structures of full-length human CNNM4 in multiple conformational states. These structures reveal that CNNM4 does not assemble as a simple symmetric dimer but forms an asymmetric tetramer organized as a dimer-of-dimers. Tetramerization is MgATP-dependent, with the intracellular modules remaining highly dynamic otherwise. This distinctive oligomerization and asymmetric assembly couple structural dynamics to Mg2+ sensing and transport. Together, these findings define the structural principles of CNNM function, advancing our understanding of Mg2+ homeostasis and providing a mechanistic basis for linking CNNM dysfunction to human disease.

Introduction

Magnesium (Mg2+) is an essential cation that supports fundamental cellular processes, including energy metabolism, nucleic acid synthesis, enzyme catalysis, and ion transport regulation1–4. Intracellular free Mg2+ concentration is tightly regulated in mammalian cells, and disruptions in Mg2+ homeostasis can lead to serious physiological consequences. Such disruptions often arise from defects in magnesium transport or systemic homeostatic control, resulting in hypomagnesemia, a condition associated with cardiovascular disorders, neurological and psychiatric disorders such as seizures and depression, as well as metabolic and endocrine disorders like type II diabetes and insulin resistance5–8.

Despite its importance, the mechanisms of Mg2+ transport across the plasma membrane remain poorly studied, particularly in mammals. No full-length structures of mammalian Mg2+ transporters and channels have been reported9–12. Among them, the Cyclin and CBS Domain Magnesium Transport Mediator (CNNM) family—comprising CNNM1, 2, 3, and 4—has emerged as a central regulator of Mg2+ efflux in human and is evolutionarily conserved across prokaryotic and eukaryotic organisms13–15. In mammals, they mediate Mg2+ export and also interact with proteins such as Phosphatase of Regenerating Liver (PRL) proteins, thereby linking CNNM activity to cancer progression16,17. More recently, CNNMs have been shown to assemble with TRPM7 and TRPM618–20, two critical Mg2+-permeable channels at the plasma membrane, forming a remarkable transporter-channel supercomplex thought to be coordinate the bidirectional regulation of cellular Mg2+ homeostasis.

Consistent with their molecular and physiological functions, genetic studies highlight the critical roles of CNNMs in human health and disease. Genetic ablation of CNNMs in mice disrupts intestinal Mg2+ extrusion, leading to hypomagnesemia and profound physiological abnormalities that range from developmental defects, infertility, seizures, and cognitive deficits to embryonic lethality21–31. Moreover, dysregulation of CNNMs has also been implicated in a broad spectrum of conditions, including hypertension32,33, schizophrenia34, cancer16,17,35,36, and liver disease37, underscoring their broad impact on human health. Of note, mutations in CNNM4 underlie Jalili syndrome, an autosomal recessive disorder characterized by cone-rod dystrophy and amelogenesis imperfecta38–43.

What sets CNNMs apart from classical transporters is their multi-domain architecture13,14. Each CNNM subunit consists of four domains: an extracellular domain (ECD), a transmembrane domain (TMD), and two intracellular regulatory domains—the cystathionine-β-synthase (CBS)-pair (CBSP) domain, and the cyclic nucleotide-binding homology (CNBH) domain. In contrast, prokaryotic homologs such as CorB/CorC contain only the TMD and CBSP, but lack the N-terminal ECD and C-terminal CNBH44,45. Prior studies of eukaryotic CNNMs focused solely on isolated domains, showing that only MgATP engages CBSP in solution and drives its dimerization—a prerequisite for transport activity, whereas MgADP, MgAMP, or nucleotide alone do not46. The ECD also mediates dimerization, and disruption of its interface abolishes transport activity47. Consistent with these findings, structures of the prokaryotic homolog CorB/CorC indeed revealed a symmetric homodimeric achitecture44,45.

While studies of isolated domains and prokaryotic homologs have shaped the prevailing view that CNNMs function as homodimers13,14, the absence of a full-length CNNM structure leaves fundamental questions unanswered. The oligomeric assembly and regulatory mechanisms of CNNMs remain unresolved, and it is unclear whether they indeed adopt symmetrical homodimers like their prokaryotic counterparts. The functional contribution of the eukaryotic-specific ECD and CNBH domains, the interdomain coordination across the full-length protein, and the structural basis of key transport states—such as outward-facing and occluded conformations—also remain undefined. Furthermore, how CNNMs integrate intracellular energy status and local Mg2+ levels to fine-tune transport activity is still unclear.

Here, we present cryo-EM structures of full-length human CNNM4 in multiple conformational states. These structures reveal that CNNM4 does not assemble as a simple symmetric dimer but forms an asymmetric tetramer organized as a dimer-of-dimers. Tetramerization is MgATP-dependent, with the intracellular modules remaining highly dynamic otherwise. This distinctive oligomerization and asymmetric assembly couple structural dynamics to Mg2+ sensing and transport. Together, these findings define the structural principles of CNNM function, advancing our understanding of Mg2+ homeostasis and providing a mechanistic basis for linking CNNM dysfunction to human disease.

ResultsMgATP drives CNNM4 conformational rearrangementTo determine the full-length structures of human CNNM4 and elucidate its mechanism of Mg2+ transport, we purified human CNNM4 and prepared cryo-EM samples in the absence of ligands (apo state), and in the presence of MgATP that activates CNNM4 (Figure S1A and S1B)46,48. Size-exclusion chromatography revealed a marked shift of CNNM4 in elution volume between the apo and MgATP-containing samples, indicating a conformational transition driven by MgATP, which was further captured in the two cryo-EM datasets (Figure S1A–S1C). In the apo dataset, CNNM4 exhibited pronounced conformational heterogeneity, particularly in its intracellular domains (ICDs), which limited structural analysis to low-resolution 2D class averages where only the ECD and TMD—but not the ICD—was discernible (Figure 1A). By contrast, the Mg/ATP dataset yielded a 3.48 Å reconstruction with well-defined density across the entire protein, highlighting the critical role of MgATP in stabilizing CNNM4 into a structurally ordered conformation (Figures 1A–1C, S2A). These findings suggest that, in the absence of MgATP, the intracellular domains fail to adopt a stable architecture—likely due to the lack of CBSP domain dimerization observed in previous crystal structures of the isolated domain—and that MgATP-driven stabilization of the cytoplasmic assembly is a prerequisite for transport activity49.CNNM4 is a tetramer of two asymmetric dimersThe MgATP-bound structure of human CNNM4 reveals an unusually elongated, flat-plate architecture, measuring approximately 216 Å in height and 114 Å in length, but only 48 Å in width (Figure 1B). This distinctive shape arises from CNNM4 assembling as a tetramer in a dimer-of-dimers configuration, with four subunits arranged side-by-side in a linear row (Figure 1B and 1C). Strikingly, this tetramer lacks the fourfold symmetry typically seen in homomeric assemblies of identical subunits50. Instead, it adopts a dimer-of-dimers arrangement with twofold symmetry, organized around an axis perpendicular to the membrane plane (Figure 1C). In addition, each CNNM4 dimer is itself asymmetric, in contrast to the symmetric arrangement expected for homodimers, as seen in the prokaryotic homologs CorB or CorC44,45. Structural overlays of the two subunits within each dimer show marked conformational differences in the intracellular domain, most prominently at the CNBH, with a high root-mean-square deviation (RMSD). This indicates that—even though they share the same amino acid sequence—each protomer adopts a distinct conformation, thereby breaking internal symmetry (Figure 1D). Moreover, we found that several reported disease-linked mutations are located within the interfaces that mediate the dimer-of-dimers assembly, implying that this higher-order configuration is both physiologically and pathologically relevant26,27,29,51.Each CNNM4 subunit comprised four domains arranged from top to bottom: an extracellular domain (ECD), a transmembrane domain (TMD), a cystathionine-β-synthase pair (CBSP) domain, and a cyclic nucleotide–binding homology (CNBH) domain, with a two-helix acidic helical bundle (AHB) linking the TMD and CBSP (Figure 1B and 1C). On the extracellular side, the ECD adopts an immunoglobulin-like fold and carries N-acetylglucosamine (NAG) modifications at three asparagine residues—N55, N122, and N127. Notably, this glycosylation pattern differs from that reported for the flatworm Echinococcus granulosus homolog47 and for mouse CNNM252 and CNNM353. Because glycosylation is essential for CNNM trafficking and localization to the plasma membrane, these findings suggest a conserved yet species- and isoform-specific role for glycosylation in regulating CNNM surface expression.The ECD caps the extracellular end of the TMD via an ECD–TMD linker that penetrates approximately one-third of the detergent micelle (Figures 1B and 1C), indicating it is membrane-embedded rather than solvent-exposed. This arrangement results in an unusually short TMD, in which each subunit contributes three α-helices, within a dimer, assemble into a complete transport unit (Figures 1B and 1C). Although the extracellular portion of the TMD is embedded within the membrane, the extracellular entrance of the transport pathway is free of lipid-like densities. This is consistent with the strongly negative electrostatic potential at the entrance, primarily conferred by D262, which likely prevents lipid occlusion of the pathway (Figure S1D). Immediately beneath the TMD lies the short AHB, followed by the CBS layer containing the MgATP-binding sites (Figures 1B and 1C)13,14. Each dimer harbors two such binding pockets arranged face-to-face, collectively accommodating two ATP molecules and eight Mg2+ ions. Four Mg2+ ions coordinate directly in ATP coordination44,45, whereas the remaining four—absent in prokaryotic homolog structures—may play a role in Mg2+ sensing in the eukaryotic system (Figure 5A and 5B) (see section “MgATP binding site and Mg2+ sensing”). At the base, the CNBH domain forms the lowest layer and is the only intracellular domain to mediate direct inter-dimer contacts (Figure 1B and 1C).Another distinctive feature of CNNM4 is domain swapping between the TMD and the intracellular domain. This domain crosstalk is initiated at AHB and leads to a reordering of subunits: from A–B–B’–A’ (with A/B and B’/A’ forming the two dimers) in the upper ECD and TMD layers to B–A–A’–B’ in the lower layers (Figure 1C).Functional partitioning of intra- and inter-dimer interactionsThe CNNM4 tetramer integrates extensive intra- and inter-dimer contacts across multiple layers, enabling long-range communication both within and between subunits. Intra-dimer interfaces—within each transport unit—are formed primarily by the ECD, TMD, and CBSP domains, whereas inter-dimer—between the two transport units—contacts are mediated distally through the ECD and CNBH domains (Figures 2A–2D). This functional partitioning highlights an evolutionary expansion from prokaryotic homologs, which lack the ECD and CNBH, toward a eukaryotic system in which these additional domains stabilize inter-dimer assembly and provide a platform for allosteric regulation between the two transport units. By contrast, the central TMD and CBS domains contribute little or no direct inter-dimer interaction, underscoring that the core transport function is confined to individual dimers. While intra-dimer interfaces have previously been shown to be essential for activity46,47,54, here we focus on the inter-dimer interfaces, which provide eukaryotic-specific mechanisms for stabilizing and regulating the tetramer.The ECD is unique in that it forms extensive contacts both within dimers and between dimers on the extracellular side (Figure 2A and 2B). Mutations that disrupt the ECD–ECD inter-dimer interface markedly impairs function, without decreasing CNNM4 surface expression (Figures S4A–S4C). In particular, the inter-dimer KRR triple mutant (K113A/R140A/R141A) markedly reduced transport activity in Mg2+ efflux assay performed as described previously22,44 (Figures 2E, 2F and S5). Cryo-EM analysis of this mutant (Figure S3A) confirmed that the ECD dimer–dimer interface was lost and the ECD–TMD linkers became disordered, which in turn distorted the tetrameric architecture (Figure 3A and 3B). Relative to wild type CNNM4, the mutant exhibited a pronounced shearing motion between dimers across all structural layers except the distal CNBH layer (Figures 3B). In particular, the two ECD dimers shifted apart by nearly 12 Å (Figure 3C), while the distal CNBH layer remained largely unaffected. Together, these structural and functional findings underscore the importance of the ECD dimer–dimer interface in maintaining tetrameric architecture, establish ECD-mediated dimerization and tetramerization as key regulators of Mg2+ transport (Figures 2E and 2F), and highlight that this inter-dimer interface—absent in prokaryotic homologs—represents a eukaryotic-specific mechanism that both joins two transport units into a functional tetramer and enables allosteric regulation between them.To understand how the ECD regulates transport activity in the TMD, we focused on the ECD–TMD linker—the sole connection between the extracellular domain and the transport core. This linker is highly unusual in containing a positively charged residue, R178, buried deep within the membrane interior where charged residues are normally excluded due to energetic penalties (Figure 2C). Notably, R178 forms a conformation-dependent hydrogen bond with the backbone carbonyl of L264 on helix S2—present in the outward-facing state but absent in the occluded state (see next section for details of these states). This conformational dependence suggests that the linker may act as a critical coupling element that transmits ECD signals to the transport core during the transport cycle. Supporting this, substituting R178 with alanine—despite preserving protein expression and surface localization—markedly decreased transport activity (Figures 2E and 2F).The CNBH domain resides at the opposite distal end of the CNNM4 tetramer and is the only cytosolic domain mediating inter-dimer contacts (Figures 1C, 2A and 2D). Unlike at the ECD, where the A/A’ subunits are positioned peripherally and the B/B’ subunits form the central inter-dimer interface, at the CNBH layer the arrangement is reversed: the A/A’ subunits are positioned centrally and engage in extensive interactions, while the B/B’ subunits are located at the periphery as a result of domain swapping. Moreover, unlike the ECD, where the four subunits are aligned side-by-side in a linear row, the CNBH domains adopt a nested configuration such that each CNBH interacts with a subunit from the adjacent dimer rather than its own. This unique arrangement creates a complex tetrameric interface involving three distinct contact regions: a central interface between subunits A and A’ mediated by an X-shaped helix–helix crossing (“X-helix”), and two lateral interfaces between A/B’ and B/A’ mediated by β-sheet interactions (Figure 2A and 2D). Moreover, whereas the CNBH domains of the A/A’ subunits connect to the CBSP layer only through the intervening linker, the B/B’ CNBH domains establish substantial additional contacts with the CBSP domains above via the X-helix, providing a potential pathway by which the CNBH layer could regulate the transport core.At the central interface, mutations impaired transport activity to varying degrees. A similar loss of function was observed with mutations disrupting the lateral interfaces (Figure 2E and 2F). Notably, all mutants exhibited elevated protein expression levels while maintaining proper membrane localization, consistent with a loss-of-function phenotype. Together, these findings demonstrate that CNNM4 function rely on an intricate network of inter-dimer contacts mediated by both ECD and CNBH domains.Outward-facing and occluded states of CNNM4The crystal structures of CorB and CorC, the prokaryotic homolog of CNNM4, were captured in an inward-facing state, with the translocation pathway open toward the cytoplasm but sealed at the extracellular side44,45. By contrast, in the MgATP-bound wild-type structure of CNNM4, the extracellular entrance is widened, with the Cα atoms of D262 on S2 ~13 Å apart, whereas the equivalent residues in CorC are closely apposed. On the cytoplasmic side, the Cα atoms of A208 on S1 are only 4.6 Å apart (compared with 16.8 Å in CorC), forming a sealed intracellular gate. These features define an outward-facing state not previously observed in the prokaryotic homologs44,45 (Figure 4A and 4B).To investigate the transport mechanism of CNNM4, we sought to stabilize it in an alternate conformation within its transport cycle. Because CNNM4 is a Mg2+/Na+ antiporter (exporting Mg2+ in exchange for Na+ influx), we first explored ion substitution by replacing Na+ with a monovalent cation (NMDG+) that does not bind CNNM4, reasoning that Na+ binding from the extracellular side favors the outward-facing state, and that removal of Na+ might destabilize this state. However, this substitution alone did not induce major structural changes, suggesting that ion exchange alone is insufficient to overcome the energetic barrier between states under the isolated conditions used for cryo-EM. We next introduced a D262C mutation at the extracellular vestibule, with the idea that the disulfide crosslinking might stabilize a closed extracellular gate and thereby favor an occluded or inward-facing state. Indeed, cell lysates of the D262C variant showed partial disulfide crosslinking on SDS-PAGE, consistent with CNNM4 naturally sampling this conformation under physiological conditions (Figure 4C). Moreover, the D262C mutant showed a marked loss of function in the Mg2+ efflux assay (Figures 4D and 4E), consistent with disulfide-mediated conformational trapping that likely restricts the ability of the transporter to sample the full set of states required to complete a transport cycle.Cyro-EM analysis of the D262C mutant, with Na+ replaced by K+—a monovalent cation that, like NMDG+, does not bind CNNM4—revealed two distinct particle populations: crosslinked and non-crosslinked, consistent with dynamic gating (Figures 4C, S2B, S3B). The non-crosslinked class adopted the same outward-facing conformation as wild-type CNNM4. By contrast, the crosslinked class showed a closed extracellular gate, and a modestly widened cytoplasmic opening compared to wild type, although still narrower than inward-facing state observed in prokaryotic homologs (Figure 4A, 4B, 4F–4H). This conformation, sealed on both sides, likely represents an occluded conformation that has not been previously observed in CNNMs or related homologs.Transitioning from the occluded to outward-facing state involves coordinated rearrangements across the entire complex (Figure 3I and 3J). Using the distal CNBH domain—which remains largely unchanged between states—as a reference for alignment, we observed asymmetric, multi-directional motions throughout the assembly. Within the TMD, the inner subunits shifted approximately as rigid bodies in opposite directions when viewed from the extracellular side, whereas the peripheral subunits underwent larger movements in a similar direction (Figure 3J, second panel from top; compare short vs. long arrows). This asymmetry compressed the extracellular vestibule while widening the cytoplasmic gate. To accommodate these differential TMD motions, the ECDs and CBSP domains rotated counterclockwise around the central twofold axis, with smaller displacements in the inner subunits but more substantial ones in the peripheral subunits (Figure 3I and 3J). These findings reveal finely tuned interlayer coordination during state transition and suggest that TMD gating could be regulated by signals sensed by the extracellular ECD and intracellular CBSP domains.MgATP binding site and Mg2+ sensingWhile previous studies of isolated CBSP domains in solution established that only MgATP—not MgADP, MgAMP, or nucleotide alone—promotes CBSP dimerization, a process essential for transport activity46—the molecular mechanism by which this binding drives transport has remained unclear in the context of full-length eukaryotic CNNM proteins.In our structure, MgATP binding induces tetramerization of CNNM4 into a dimer-of-dimers (Figure 1A). Notably, however, this tetramerization does not involve the CBSP layer, which remains self-contained and lacks inter-dimer contacts. Instead, as in prokaryotic homologs, MgATP promotes dimerization within the CBSP layer, where two ATP molecules bind at the dimer interface. Each ATP nestles its adenine ring into a hydrophobic pocket between the upper and lower CBS motifs of one subunit, while its triphosphate tail projects toward the partner subunit (Figure 5A and 5B). At the intradimer interface, the γ-phosphates of the paired ATP molecules approach to within ~4 Å, creating electrostatic repulsion that is resolved by two Mg2+ ions per ATP. One Mg2+ ion (Mg1) bridges the α and γ phosphates within a single ATP (Figure 5B), contributing to charge neutralization, as seen previously in prokaryotic and isolated human CBSP structures. Interestingly, our structure reveals a second Mg2+ ion (Mg2), located across the intradimer interface and directly coordinating the γ-phosphates of both ATP molecules (Figure 5B). This Mg2+ not only alleviates the electrostatic clash but also acts as a molecular “salt bridge” that cements the CBSP dimer. Importantly, this mechanism explains why ATP—but not AMP or ADP—stabilize functional dimerization: the γ-phosphate tail is indispensable for forming the Mg2+-mediated intersubunit bridge.Beyond these Mg2+ binding sites directly associated with ATP, our structure also revealed a cluster of Mg2+ ions—two per subunit, eight in total—bound to a highly acidic surface patch on the upper CBS motif, coordinated by evolutionally conserved residues E497, D498, E501, E502, E507, D513, D517, and N518 (Figures 5C and S6). Positioned at a key architectural junction, this acidic patch connects upward to the TMD via the AHB and downward to the CNBH domain through a flexible linker. To assess the functional role of this acidic patch, we introduced alanine substitutions at several key residues. These mutants exhibited markedly elevated surface expression, demonstrating that the reduced activity of these mutants reflects a genuine functional defect rather than reduced surface localization (Figure S4B and S4C), and Mg2+ efflux assays confirmed substantially reduced transport activity (Figure 5D and 5E). This contrasts with mutations in the MgATP binding site, which completely abolish transport activity46.Together, these findings suggest a dual role of Mg2+ in CNNM4 regulation: an interfacial Mg2+ salt bridge at the CBSP dimer interface that is absolutely required for transport, and additional Mg2+ to the acidic patch that allosterically tunes activity. We propose that this acidic patch, enriched in acidic residues, acts as a potential Mg2+-sensing module: depending on cytosolic Mg2+ levels, it could bind variable numbers of Mg2+ ions to modulate CNNM4 activity—upward, by directly influencing TMD conformation; and downward, by modulating the CNBH positioning, which feeds back to the TMD to regulate the transport cycle (Figure 2D–2F).Mechanism of Mg2+ transportIn the MgATP-bound wild-type structure of CNNM4, captured in an outward-facing conformation, each tetramer forms two Mg2+ translocation pathways, one per dimer. These pathways are lined by a mixture of negatively charged and polar residues, creating an environment suitable for divalent cation (Figures 6A and 6B). However, unlike the prokaryotic homolog CorC, captured in an inward-facing state with two bound Mg2+ within its single pathway44, we observed little to no densities corresponding to Mg2+ at the equivalent sites in either CNNM4 dimer, even though the surrounding residues are conserved in the CNNM family (with the exception of CNNM3) (Figure 6C).Comparison of the inward-facing CorC structure with the outward-facing CNNM4 structure provides key insight into how Mg2+ binding and release are coupled to conformational state. In CorC, each subunit binds one Mg2+ near the middle of the pathway in canonical octahedral geometry by a glutamate, an asparagine, two serines, and two backbone carbonyl oxygens (Figure 6D, right panel). This complete coordination cage explains the high Mg2+ affinity in the inward-facing state, consistent with its role in binding cytoplasmic Mg2+. By contrast, in CNNM4 the equivalent sites are disrupted. Most notably, the side chain of E284—a key acidic residue located in the central π-helical segment of S3—flips away from the binding site (Figure 6D, left panel). This conformational change is accompanied by a displacement of the adjacent backbone carbonyl of G283, another Mg2+-coordinating residue. Together, these rearrangements distort the coordination geometry and open the site toward the extracellular space in the outward-facing state, consistent with its role in Mg2+ release rather than binding.Interestingly, we observed three additional ion-like densities near each E284 in CNNM4. These densities are unlikely to represent Mg2+, given their relatively loose coordination geometry and close spatial proximity to one another (~4 Å), which would result in strong electrostatic repulsion between divalent cations. We therefore tentatively assign these densities to Na+ (Na1 and Na1’), consistent with CNNM4’s role as a Mg2+ exporter coupled to Na+ import and with the expectation that the outward-facing state binds Na+ with high affinity (Figure 6E).To test whether the observed densities indeed correspond to Na+, we first attempted to replace Na+ with N-methyl-D-glucamine (NMDG) throughout protein purification and cryo-EM data collection, with the expectation that these densities would either be abolished (as NMDG does not bind) or replaced by NMDG densities at the putative Na+ binding sites. However, under these conditions the protein became unstable and prone to aggregation, preventing the determination of cryo-EM structures at resolutions comparable to those obtained in the presence of Na+. We therefore adopted an alternative strategy by mutating E284 to alanine, reasoning that disruption of this coordinating residue should eliminate Na+ binding at these sites. The structure of the E284A mutant was determined at a resolution comparable to that of the wild type (Figure S3C), enabling a fair and direct structural comparison (Figure 6F). Consistent with our hypothesis, the putative Na+ densities are absent in the E284A structure, supporting their assignment as Na+ (Figure 6F).We propose that Na+ binding stabilizes the flipped conformation of E284, thereby weakening Mg2+ coordination and promoting Mg2+ release. In addition, a third putative ion density (Na2) is observed between the two E284 residues at the intradimer interface (Figure 6E). Although it lies beyond the typical range for direct Na+–sidechain interactions, it may represent a Na+ ion coordinating E284 via water molecules—unresolved at the current map resolution—which could further stabilize the flipped conformationTogether, these findings support a mechanistic cycle for CNNM4-medated Mg2+ transport. In the inward-facing state, Mg2+ binds with high affinity through complete coordination, as seen in CorC. Transitions through an occluded intermediate lead to the outward-facing state, where disrupted coordination, assisted by Na+ binding, lowers Mg2+ affinity and enables Mg2+ release, completing the transport cycle.

MgATP drives CNNM4 conformational rearrangementTo determine the full-length structures of human CNNM4 and elucidate its mechanism of Mg2+ transport, we purified human CNNM4 and prepared cryo-EM samples in the absence of ligands (apo state), and in the presence of MgATP that activates CNNM4 (Figure S1A and S1B)46,48. Size-exclusion chromatography revealed a marked shift of CNNM4 in elution volume between the apo and MgATP-containing samples, indicating a conformational transition driven by MgATP, which was further captured in the two cryo-EM datasets (Figure S1A–S1C). In the apo dataset, CNNM4 exhibited pronounced conformational heterogeneity, particularly in its intracellular domains (ICDs), which limited structural analysis to low-resolution 2D class averages where only the ECD and TMD—but not the ICD—was discernible (Figure 1A). By contrast, the Mg/ATP dataset yielded a 3.48 Å reconstruction with well-defined density across the entire protein, highlighting the critical role of MgATP in stabilizing CNNM4 into a structurally ordered conformation (Figures 1A–1C, S2A). These findings suggest that, in the absence of MgATP, the intracellular domains fail to adopt a stable architecture—likely due to the lack of CBSP domain dimerization observed in previous crystal structures of the isolated domain—and that MgATP-driven stabilization of the cytoplasmic assembly is a prerequisite for transport activity49.

MgATP drives CNNM4 conformational rearrangement

To determine the full-length structures of human CNNM4 and elucidate its mechanism of Mg2+ transport, we purified human CNNM4 and prepared cryo-EM samples in the absence of ligands (apo state), and in the presence of MgATP that activates CNNM4 (Figure S1A and S1B)46,48. Size-exclusion chromatography revealed a marked shift of CNNM4 in elution volume between the apo and MgATP-containing samples, indicating a conformational transition driven by MgATP, which was further captured in the two cryo-EM datasets (Figure S1A–S1C). In the apo dataset, CNNM4 exhibited pronounced conformational heterogeneity, particularly in its intracellular domains (ICDs), which limited structural analysis to low-resolution 2D class averages where only the ECD and TMD—but not the ICD—was discernible (Figure 1A). By contrast, the Mg/ATP dataset yielded a 3.48 Å reconstruction with well-defined density across the entire protein, highlighting the critical role of MgATP in stabilizing CNNM4 into a structurally ordered conformation (Figures 1A–1C, S2A). These findings suggest that, in the absence of MgATP, the intracellular domains fail to adopt a stable architecture—likely due to the lack of CBSP domain dimerization observed in previous crystal structures of the isolated domain—and that MgATP-driven stabilization of the cytoplasmic assembly is a prerequisite for transport activity49.

CNNM4 is a tetramer of two asymmetric dimersThe MgATP-bound structure of human CNNM4 reveals an unusually elongated, flat-plate architecture, measuring approximately 216 Å in height and 114 Å in length, but only 48 Å in width (Figure 1B). This distinctive shape arises from CNNM4 assembling as a tetramer in a dimer-of-dimers configuration, with four subunits arranged side-by-side in a linear row (Figure 1B and 1C). Strikingly, this tetramer lacks the fourfold symmetry typically seen in homomeric assemblies of identical subunits50. Instead, it adopts a dimer-of-dimers arrangement with twofold symmetry, organized around an axis perpendicular to the membrane plane (Figure 1C). In addition, each CNNM4 dimer is itself asymmetric, in contrast to the symmetric arrangement expected for homodimers, as seen in the prokaryotic homologs CorB or CorC44,45. Structural overlays of the two subunits within each dimer show marked conformational differences in the intracellular domain, most prominently at the CNBH, with a high root-mean-square deviation (RMSD). This indicates that—even though they share the same amino acid sequence—each protomer adopts a distinct conformation, thereby breaking internal symmetry (Figure 1D). Moreover, we found that several reported disease-linked mutations are located within the interfaces that mediate the dimer-of-dimers assembly, implying that this higher-order configuration is both physiologically and pathologically relevant26,27,29,51.Each CNNM4 subunit comprised four domains arranged from top to bottom: an extracellular domain (ECD), a transmembrane domain (TMD), a cystathionine-β-synthase pair (CBSP) domain, and a cyclic nucleotide–binding homology (CNBH) domain, with a two-helix acidic helical bundle (AHB) linking the TMD and CBSP (Figure 1B and 1C). On the extracellular side, the ECD adopts an immunoglobulin-like fold and carries N-acetylglucosamine (NAG) modifications at three asparagine residues—N55, N122, and N127. Notably, this glycosylation pattern differs from that reported for the flatworm Echinococcus granulosus homolog47 and for mouse CNNM252 and CNNM353. Because glycosylation is essential for CNNM trafficking and localization to the plasma membrane, these findings suggest a conserved yet species- and isoform-specific role for glycosylation in regulating CNNM surface expression.The ECD caps the extracellular end of the TMD via an ECD–TMD linker that penetrates approximately one-third of the detergent micelle (Figures 1B and 1C), indicating it is membrane-embedded rather than solvent-exposed. This arrangement results in an unusually short TMD, in which each subunit contributes three α-helices, within a dimer, assemble into a complete transport unit (Figures 1B and 1C). Although the extracellular portion of the TMD is embedded within the membrane, the extracellular entrance of the transport pathway is free of lipid-like densities. This is consistent with the strongly negative electrostatic potential at the entrance, primarily conferred by D262, which likely prevents lipid occlusion of the pathway (Figure S1D). Immediately beneath the TMD lies the short AHB, followed by the CBS layer containing the MgATP-binding sites (Figures 1B and 1C)13,14. Each dimer harbors two such binding pockets arranged face-to-face, collectively accommodating two ATP molecules and eight Mg2+ ions. Four Mg2+ ions coordinate directly in ATP coordination44,45, whereas the remaining four—absent in prokaryotic homolog structures—may play a role in Mg2+ sensing in the eukaryotic system (Figure 5A and 5B) (see section “MgATP binding site and Mg2+ sensing”). At the base, the CNBH domain forms the lowest layer and is the only intracellular domain to mediate direct inter-dimer contacts (Figure 1B and 1C).Another distinctive feature of CNNM4 is domain swapping between the TMD and the intracellular domain. This domain crosstalk is initiated at AHB and leads to a reordering of subunits: from A–B–B’–A’ (with A/B and B’/A’ forming the two dimers) in the upper ECD and TMD layers to B–A–A’–B’ in the lower layers (Figure 1C).

CNNM4 is a tetramer of two asymmetric dimers

The MgATP-bound structure of human CNNM4 reveals an unusually elongated, flat-plate architecture, measuring approximately 216 Å in height and 114 Å in length, but only 48 Å in width (Figure 1B). This distinctive shape arises from CNNM4 assembling as a tetramer in a dimer-of-dimers configuration, with four subunits arranged side-by-side in a linear row (Figure 1B and 1C). Strikingly, this tetramer lacks the fourfold symmetry typically seen in homomeric assemblies of identical subunits50. Instead, it adopts a dimer-of-dimers arrangement with twofold symmetry, organized around an axis perpendicular to the membrane plane (Figure 1C). In addition, each CNNM4 dimer is itself asymmetric, in contrast to the symmetric arrangement expected for homodimers, as seen in the prokaryotic homologs CorB or CorC44,45. Structural overlays of the two subunits within each dimer show marked conformational differences in the intracellular domain, most prominently at the CNBH, with a high root-mean-square deviation (RMSD). This indicates that—even though they share the same amino acid sequence—each protomer adopts a distinct conformation, thereby breaking internal symmetry (Figure 1D). Moreover, we found that several reported disease-linked mutations are located within the interfaces that mediate the dimer-of-dimers assembly, implying that this higher-order configuration is both physiologically and pathologically relevant26,27,29,51.

Each CNNM4 subunit comprised four domains arranged from top to bottom: an extracellular domain (ECD), a transmembrane domain (TMD), a cystathionine-β-synthase pair (CBSP) domain, and a cyclic nucleotide–binding homology (CNBH) domain, with a two-helix acidic helical bundle (AHB) linking the TMD and CBSP (Figure 1B and 1C). On the extracellular side, the ECD adopts an immunoglobulin-like fold and carries N-acetylglucosamine (NAG) modifications at three asparagine residues—N55, N122, and N127. Notably, this glycosylation pattern differs from that reported for the flatworm Echinococcus granulosus homolog47 and for mouse CNNM252 and CNNM353. Because glycosylation is essential for CNNM trafficking and localization to the plasma membrane, these findings suggest a conserved yet species- and isoform-specific role for glycosylation in regulating CNNM surface expression.

The ECD caps the extracellular end of the TMD via an ECD–TMD linker that penetrates approximately one-third of the detergent micelle (Figures 1B and 1C), indicating it is membrane-embedded rather than solvent-exposed. This arrangement results in an unusually short TMD, in which each subunit contributes three α-helices, within a dimer, assemble into a complete transport unit (Figures 1B and 1C). Although the extracellular portion of the TMD is embedded within the membrane, the extracellular entrance of the transport pathway is free of lipid-like densities. This is consistent with the strongly negative electrostatic potential at the entrance, primarily conferred by D262, which likely prevents lipid occlusion of the pathway (Figure S1D). Immediately beneath the TMD lies the short AHB, followed by the CBS layer containing the MgATP-binding sites (Figures 1B and 1C)13,14. Each dimer harbors two such binding pockets arranged face-to-face, collectively accommodating two ATP molecules and eight Mg2+ ions. Four Mg2+ ions coordinate directly in ATP coordination44,45, whereas the remaining four—absent in prokaryotic homolog structures—may play a role in Mg2+ sensing in the eukaryotic system (Figure 5A and 5B) (see section “MgATP binding site and Mg2+ sensing”). At the base, the CNBH domain forms the lowest layer and is the only intracellular domain to mediate direct inter-dimer contacts (Figure 1B and 1C).

Another distinctive feature of CNNM4 is domain swapping between the TMD and the intracellular domain. This domain crosstalk is initiated at AHB and leads to a reordering of subunits: from A–B–B’–A’ (with A/B and B’/A’ forming the two dimers) in the upper ECD and TMD layers to B–A–A’–B’ in the lower layers (Figure 1C).

Functional partitioning of intra- and inter-dimer interactionsThe CNNM4 tetramer integrates extensive intra- and inter-dimer contacts across multiple layers, enabling long-range communication both within and between subunits. Intra-dimer interfaces—within each transport unit—are formed primarily by the ECD, TMD, and CBSP domains, whereas inter-dimer—between the two transport units—contacts are mediated distally through the ECD and CNBH domains (Figures 2A–2D). This functional partitioning highlights an evolutionary expansion from prokaryotic homologs, which lack the ECD and CNBH, toward a eukaryotic system in which these additional domains stabilize inter-dimer assembly and provide a platform for allosteric regulation between the two transport units. By contrast, the central TMD and CBS domains contribute little or no direct inter-dimer interaction, underscoring that the core transport function is confined to individual dimers. While intra-dimer interfaces have previously been shown to be essential for activity46,47,54, here we focus on the inter-dimer interfaces, which provide eukaryotic-specific mechanisms for stabilizing and regulating the tetramer.The ECD is unique in that it forms extensive contacts both within dimers and between dimers on the extracellular side (Figure 2A and 2B). Mutations that disrupt the ECD–ECD inter-dimer interface markedly impairs function, without decreasing CNNM4 surface expression (Figures S4A–S4C). In particular, the inter-dimer KRR triple mutant (K113A/R140A/R141A) markedly reduced transport activity in Mg2+ efflux assay performed as described previously22,44 (Figures 2E, 2F and S5). Cryo-EM analysis of this mutant (Figure S3A) confirmed that the ECD dimer–dimer interface was lost and the ECD–TMD linkers became disordered, which in turn distorted the tetrameric architecture (Figure 3A and 3B). Relative to wild type CNNM4, the mutant exhibited a pronounced shearing motion between dimers across all structural layers except the distal CNBH layer (Figures 3B). In particular, the two ECD dimers shifted apart by nearly 12 Å (Figure 3C), while the distal CNBH layer remained largely unaffected. Together, these structural and functional findings underscore the importance of the ECD dimer–dimer interface in maintaining tetrameric architecture, establish ECD-mediated dimerization and tetramerization as key regulators of Mg2+ transport (Figures 2E and 2F), and highlight that this inter-dimer interface—absent in prokaryotic homologs—represents a eukaryotic-specific mechanism that both joins two transport units into a functional tetramer and enables allosteric regulation between them.To understand how the ECD regulates transport activity in the TMD, we focused on the ECD–TMD linker—the sole connection between the extracellular domain and the transport core. This linker is highly unusual in containing a positively charged residue, R178, buried deep within the membrane interior where charged residues are normally excluded due to energetic penalties (Figure 2C). Notably, R178 forms a conformation-dependent hydrogen bond with the backbone carbonyl of L264 on helix S2—present in the outward-facing state but absent in the occluded state (see next section for details of these states). This conformational dependence suggests that the linker may act as a critical coupling element that transmits ECD signals to the transport core during the transport cycle. Supporting this, substituting R178 with alanine—despite preserving protein expression and surface localization—markedly decreased transport activity (Figures 2E and 2F).The CNBH domain resides at the opposite distal end of the CNNM4 tetramer and is the only cytosolic domain mediating inter-dimer contacts (Figures 1C, 2A and 2D). Unlike at the ECD, where the A/A’ subunits are positioned peripherally and the B/B’ subunits form the central inter-dimer interface, at the CNBH layer the arrangement is reversed: the A/A’ subunits are positioned centrally and engage in extensive interactions, while the B/B’ subunits are located at the periphery as a result of domain swapping. Moreover, unlike the ECD, where the four subunits are aligned side-by-side in a linear row, the CNBH domains adopt a nested configuration such that each CNBH interacts with a subunit from the adjacent dimer rather than its own. This unique arrangement creates a complex tetrameric interface involving three distinct contact regions: a central interface between subunits A and A’ mediated by an X-shaped helix–helix crossing (“X-helix”), and two lateral interfaces between A/B’ and B/A’ mediated by β-sheet interactions (Figure 2A and 2D). Moreover, whereas the CNBH domains of the A/A’ subunits connect to the CBSP layer only through the intervening linker, the B/B’ CNBH domains establish substantial additional contacts with the CBSP domains above via the X-helix, providing a potential pathway by which the CNBH layer could regulate the transport core.At the central interface, mutations impaired transport activity to varying degrees. A similar loss of function was observed with mutations disrupting the lateral interfaces (Figure 2E and 2F). Notably, all mutants exhibited elevated protein expression levels while maintaining proper membrane localization, consistent with a loss-of-function phenotype. Together, these findings demonstrate that CNNM4 function rely on an intricate network of inter-dimer contacts mediated by both ECD and CNBH domains.

Functional partitioning of intra- and inter-dimer interactions

The CNNM4 tetramer integrates extensive intra- and inter-dimer contacts across multiple layers, enabling long-range communication both within and between subunits. Intra-dimer interfaces—within each transport unit—are formed primarily by the ECD, TMD, and CBSP domains, whereas inter-dimer—between the two transport units—contacts are mediated distally through the ECD and CNBH domains (Figures 2A–2D). This functional partitioning highlights an evolutionary expansion from prokaryotic homologs, which lack the ECD and CNBH, toward a eukaryotic system in which these additional domains stabilize inter-dimer assembly and provide a platform for allosteric regulation between the two transport units. By contrast, the central TMD and CBS domains contribute little or no direct inter-dimer interaction, underscoring that the core transport function is confined to individual dimers. While intra-dimer interfaces have previously been shown to be essential for activity46,47,54, here we focus on the inter-dimer interfaces, which provide eukaryotic-specific mechanisms for stabilizing and regulating the tetramer.

The ECD is unique in that it forms extensive contacts both within dimers and between dimers on the extracellular side (Figure 2A and 2B). Mutations that disrupt the ECD–ECD inter-dimer interface markedly impairs function, without decreasing CNNM4 surface expression (Figures S4A–S4C). In particular, the inter-dimer KRR triple mutant (K113A/R140A/R141A) markedly reduced transport activity in Mg2+ efflux assay performed as described previously22,44 (Figures 2E, 2F and S5). Cryo-EM analysis of this mutant (Figure S3A) confirmed that the ECD dimer–dimer interface was lost and the ECD–TMD linkers became disordered, which in turn distorted the tetrameric architecture (Figure 3A and 3B). Relative to wild type CNNM4, the mutant exhibited a pronounced shearing motion between dimers across all structural layers except the distal CNBH layer (Figures 3B). In particular, the two ECD dimers shifted apart by nearly 12 Å (Figure 3C), while the distal CNBH layer remained largely unaffected. Together, these structural and functional findings underscore the importance of the ECD dimer–dimer interface in maintaining tetrameric architecture, establish ECD-mediated dimerization and tetramerization as key regulators of Mg2+ transport (Figures 2E and 2F), and highlight that this inter-dimer interface—absent in prokaryotic homologs—represents a eukaryotic-specific mechanism that both joins two transport units into a functional tetramer and enables allosteric regulation between them.

To understand how the ECD regulates transport activity in the TMD, we focused on the ECD–TMD linker—the sole connection between the extracellular domain and the transport core. This linker is highly unusual in containing a positively charged residue, R178, buried deep within the membrane interior where charged residues are normally excluded due to energetic penalties (Figure 2C). Notably, R178 forms a conformation-dependent hydrogen bond with the backbone carbonyl of L264 on helix S2—present in the outward-facing state but absent in the occluded state (see next section for details of these states). This conformational dependence suggests that the linker may act as a critical coupling element that transmits ECD signals to the transport core during the transport cycle. Supporting this, substituting R178 with alanine—despite preserving protein expression and surface localization—markedly decreased transport activity (Figures 2E and 2F).

The CNBH domain resides at the opposite distal end of the CNNM4 tetramer and is the only cytosolic domain mediating inter-dimer contacts (Figures 1C, 2A and 2D). Unlike at the ECD, where the A/A’ subunits are positioned peripherally and the B/B’ subunits form the central inter-dimer interface, at the CNBH layer the arrangement is reversed: the A/A’ subunits are positioned centrally and engage in extensive interactions, while the B/B’ subunits are located at the periphery as a result of domain swapping. Moreover, unlike the ECD, where the four subunits are aligned side-by-side in a linear row, the CNBH domains adopt a nested configuration such that each CNBH interacts with a subunit from the adjacent dimer rather than its own. This unique arrangement creates a complex tetrameric interface involving three distinct contact regions: a central interface between subunits A and A’ mediated by an X-shaped helix–helix crossing (“X-helix”), and two lateral interfaces between A/B’ and B/A’ mediated by β-sheet interactions (Figure 2A and 2D). Moreover, whereas the CNBH domains of the A/A’ subunits connect to the CBSP layer only through the intervening linker, the B/B’ CNBH domains establish substantial additional contacts with the CBSP domains above via the X-helix, providing a potential pathway by which the CNBH layer could regulate the transport core.

At the central interface, mutations impaired transport activity to varying degrees. A similar loss of function was observed with mutations disrupting the lateral interfaces (Figure 2E and 2F). Notably, all mutants exhibited elevated protein expression levels while maintaining proper membrane localization, consistent with a loss-of-function phenotype. Together, these findings demonstrate that CNNM4 function rely on an intricate network of inter-dimer contacts mediated by both ECD and CNBH domains.

Outward-facing and occluded states of CNNM4The crystal structures of CorB and CorC, the prokaryotic homolog of CNNM4, were captured in an inward-facing state, with the translocation pathway open toward the cytoplasm but sealed at the extracellular side44,45. By contrast, in the MgATP-bound wild-type structure of CNNM4, the extracellular entrance is widened, with the Cα atoms of D262 on S2 ~13 Å apart, whereas the equivalent residues in CorC are closely apposed. On the cytoplasmic side, the Cα atoms of A208 on S1 are only 4.6 Å apart (compared with 16.8 Å in CorC), forming a sealed intracellular gate. These features define an outward-facing state not previously observed in the prokaryotic homologs44,45 (Figure 4A and 4B).To investigate the transport mechanism of CNNM4, we sought to stabilize it in an alternate conformation within its transport cycle. Because CNNM4 is a Mg2+/Na+ antiporter (exporting Mg2+ in exchange for Na+ influx), we first explored ion substitution by replacing Na+ with a monovalent cation (NMDG+) that does not bind CNNM4, reasoning that Na+ binding from the extracellular side favors the outward-facing state, and that removal of Na+ might destabilize this state. However, this substitution alone did not induce major structural changes, suggesting that ion exchange alone is insufficient to overcome the energetic barrier between states under the isolated conditions used for cryo-EM. We next introduced a D262C mutation at the extracellular vestibule, with the idea that the disulfide crosslinking might stabilize a closed extracellular gate and thereby favor an occluded or inward-facing state. Indeed, cell lysates of the D262C variant showed partial disulfide crosslinking on SDS-PAGE, consistent with CNNM4 naturally sampling this conformation under physiological conditions (Figure 4C). Moreover, the D262C mutant showed a marked loss of function in the Mg2+ efflux assay (Figures 4D and 4E), consistent with disulfide-mediated conformational trapping that likely restricts the ability of the transporter to sample the full set of states required to complete a transport cycle.Cyro-EM analysis of the D262C mutant, with Na+ replaced by K+—a monovalent cation that, like NMDG+, does not bind CNNM4—revealed two distinct particle populations: crosslinked and non-crosslinked, consistent with dynamic gating (Figures 4C, S2B, S3B). The non-crosslinked class adopted the same outward-facing conformation as wild-type CNNM4. By contrast, the crosslinked class showed a closed extracellular gate, and a modestly widened cytoplasmic opening compared to wild type, although still narrower than inward-facing state observed in prokaryotic homologs (Figure 4A, 4B, 4F–4H). This conformation, sealed on both sides, likely represents an occluded conformation that has not been previously observed in CNNMs or related homologs.Transitioning from the occluded to outward-facing state involves coordinated rearrangements across the entire complex (Figure 3I and 3J). Using the distal CNBH domain—which remains largely unchanged between states—as a reference for alignment, we observed asymmetric, multi-directional motions throughout the assembly. Within the TMD, the inner subunits shifted approximately as rigid bodies in opposite directions when viewed from the extracellular side, whereas the peripheral subunits underwent larger movements in a similar direction (Figure 3J, second panel from top; compare short vs. long arrows). This asymmetry compressed the extracellular vestibule while widening the cytoplasmic gate. To accommodate these differential TMD motions, the ECDs and CBSP domains rotated counterclockwise around the central twofold axis, with smaller displacements in the inner subunits but more substantial ones in the peripheral subunits (Figure 3I and 3J). These findings reveal finely tuned interlayer coordination during state transition and suggest that TMD gating could be regulated by signals sensed by the extracellular ECD and intracellular CBSP domains.

Outward-facing and occluded states of CNNM4

The crystal structures of CorB and CorC, the prokaryotic homolog of CNNM4, were captured in an inward-facing state, with the translocation pathway open toward the cytoplasm but sealed at the extracellular side44,45. By contrast, in the MgATP-bound wild-type structure of CNNM4, the extracellular entrance is widened, with the Cα atoms of D262 on S2 ~13 Å apart, whereas the equivalent residues in CorC are closely apposed. On the cytoplasmic side, the Cα atoms of A208 on S1 are only 4.6 Å apart (compared with 16.8 Å in CorC), forming a sealed intracellular gate. These features define an outward-facing state not previously observed in the prokaryotic homologs44,45 (Figure 4A and 4B).

To investigate the transport mechanism of CNNM4, we sought to stabilize it in an alternate conformation within its transport cycle. Because CNNM4 is a Mg2+/Na+ antiporter (exporting Mg2+ in exchange for Na+ influx), we first explored ion substitution by replacing Na+ with a monovalent cation (NMDG+) that does not bind CNNM4, reasoning that Na+ binding from the extracellular side favors the outward-facing state, and that removal of Na+ might destabilize this state. However, this substitution alone did not induce major structural changes, suggesting that ion exchange alone is insufficient to overcome the energetic barrier between states under the isolated conditions used for cryo-EM. We next introduced a D262C mutation at the extracellular vestibule, with the idea that the disulfide crosslinking might stabilize a closed extracellular gate and thereby favor an occluded or inward-facing state. Indeed, cell lysates of the D262C variant showed partial disulfide crosslinking on SDS-PAGE, consistent with CNNM4 naturally sampling this conformation under physiological conditions (Figure 4C). Moreover, the D262C mutant showed a marked loss of function in the Mg2+ efflux assay (Figures 4D and 4E), consistent with disulfide-mediated conformational trapping that likely restricts the ability of the transporter to sample the full set of states required to complete a transport cycle.

Cyro-EM analysis of the D262C mutant, with Na+ replaced by K+—a monovalent cation that, like NMDG+, does not bind CNNM4—revealed two distinct particle populations: crosslinked and non-crosslinked, consistent with dynamic gating (Figures 4C, S2B, S3B). The non-crosslinked class adopted the same outward-facing conformation as wild-type CNNM4. By contrast, the crosslinked class showed a closed extracellular gate, and a modestly widened cytoplasmic opening compared to wild type, although still narrower than inward-facing state observed in prokaryotic homologs (Figure 4A, 4B, 4F–4H). This conformation, sealed on both sides, likely represents an occluded conformation that has not been previously observed in CNNMs or related homologs.

Transitioning from the occluded to outward-facing state involves coordinated rearrangements across the entire complex (Figure 3I and 3J). Using the distal CNBH domain—which remains largely unchanged between states—as a reference for alignment, we observed asymmetric, multi-directional motions throughout the assembly. Within the TMD, the inner subunits shifted approximately as rigid bodies in opposite directions when viewed from the extracellular side, whereas the peripheral subunits underwent larger movements in a similar direction (Figure 3J, second panel from top; compare short vs. long arrows). This asymmetry compressed the extracellular vestibule while widening the cytoplasmic gate. To accommodate these differential TMD motions, the ECDs and CBSP domains rotated counterclockwise around the central twofold axis, with smaller displacements in the inner subunits but more substantial ones in the peripheral subunits (Figure 3I and 3J). These findings reveal finely tuned interlayer coordination during state transition and suggest that TMD gating could be regulated by signals sensed by the extracellular ECD and intracellular CBSP domains.

MgATP binding site and Mg2+ sensingWhile previous studies of isolated CBSP domains in solution established that only MgATP—not MgADP, MgAMP, or nucleotide alone—promotes CBSP dimerization, a process essential for transport activity46—the molecular mechanism by which this binding drives transport has remained unclear in the context of full-length eukaryotic CNNM proteins.In our structure, MgATP binding induces tetramerization of CNNM4 into a dimer-of-dimers (Figure 1A). Notably, however, this tetramerization does not involve the CBSP layer, which remains self-contained and lacks inter-dimer contacts. Instead, as in prokaryotic homologs, MgATP promotes dimerization within the CBSP layer, where two ATP molecules bind at the dimer interface. Each ATP nestles its adenine ring into a hydrophobic pocket between the upper and lower CBS motifs of one subunit, while its triphosphate tail projects toward the partner subunit (Figure 5A and 5B). At the intradimer interface, the γ-phosphates of the paired ATP molecules approach to within ~4 Å, creating electrostatic repulsion that is resolved by two Mg2+ ions per ATP. One Mg2+ ion (Mg1) bridges the α and γ phosphates within a single ATP (Figure 5B), contributing to charge neutralization, as seen previously in prokaryotic and isolated human CBSP structures. Interestingly, our structure reveals a second Mg2+ ion (Mg2), located across the intradimer interface and directly coordinating the γ-phosphates of both ATP molecules (Figure 5B). This Mg2+ not only alleviates the electrostatic clash but also acts as a molecular “salt bridge” that cements the CBSP dimer. Importantly, this mechanism explains why ATP—but not AMP or ADP—stabilize functional dimerization: the γ-phosphate tail is indispensable for forming the Mg2+-mediated intersubunit bridge.Beyond these Mg2+ binding sites directly associated with ATP, our structure also revealed a cluster of Mg2+ ions—two per subunit, eight in total—bound to a highly acidic surface patch on the upper CBS motif, coordinated by evolutionally conserved residues E497, D498, E501, E502, E507, D513, D517, and N518 (Figures 5C and S6). Positioned at a key architectural junction, this acidic patch connects upward to the TMD via the AHB and downward to the CNBH domain through a flexible linker. To assess the functional role of this acidic patch, we introduced alanine substitutions at several key residues. These mutants exhibited markedly elevated surface expression, demonstrating that the reduced activity of these mutants reflects a genuine functional defect rather than reduced surface localization (Figure S4B and S4C), and Mg2+ efflux assays confirmed substantially reduced transport activity (Figure 5D and 5E). This contrasts with mutations in the MgATP binding site, which completely abolish transport activity46.Together, these findings suggest a dual role of Mg2+ in CNNM4 regulation: an interfacial Mg2+ salt bridge at the CBSP dimer interface that is absolutely required for transport, and additional Mg2+ to the acidic patch that allosterically tunes activity. We propose that this acidic patch, enriched in acidic residues, acts as a potential Mg2+-sensing module: depending on cytosolic Mg2+ levels, it could bind variable numbers of Mg2+ ions to modulate CNNM4 activity—upward, by directly influencing TMD conformation; and downward, by modulating the CNBH positioning, which feeds back to the TMD to regulate the transport cycle (Figure 2D–2F).

MgATP binding site and Mg2+ sensing

While previous studies of isolated CBSP domains in solution established that only MgATP—not MgADP, MgAMP, or nucleotide alone—promotes CBSP dimerization, a process essential for transport activity46—the molecular mechanism by which this binding drives transport has remained unclear in the context of full-length eukaryotic CNNM proteins.

In our structure, MgATP binding induces tetramerization of CNNM4 into a dimer-of-dimers (Figure 1A). Notably, however, this tetramerization does not involve the CBSP layer, which remains self-contained and lacks inter-dimer contacts. Instead, as in prokaryotic homologs, MgATP promotes dimerization within the CBSP layer, where two ATP molecules bind at the dimer interface. Each ATP nestles its adenine ring into a hydrophobic pocket between the upper and lower CBS motifs of one subunit, while its triphosphate tail projects toward the partner subunit (Figure 5A and 5B). At the intradimer interface, the γ-phosphates of the paired ATP molecules approach to within ~4 Å, creating electrostatic repulsion that is resolved by two Mg2+ ions per ATP. One Mg2+ ion (Mg1) bridges the α and γ phosphates within a single ATP (Figure 5B), contributing to charge neutralization, as seen previously in prokaryotic and isolated human CBSP structures. Interestingly, our structure reveals a second Mg2+ ion (Mg2), located across the intradimer interface and directly coordinating the γ-phosphates of both ATP molecules (Figure 5B). This Mg2+ not only alleviates the electrostatic clash but also acts as a molecular “salt bridge” that cements the CBSP dimer. Importantly, this mechanism explains why ATP—but not AMP or ADP—stabilize functional dimerization: the γ-phosphate tail is indispensable for forming the Mg2+-mediated intersubunit bridge.

Beyond these Mg2+ binding sites directly associated with ATP, our structure also revealed a cluster of Mg2+ ions—two per subunit, eight in total—bound to a highly acidic surface patch on the upper CBS motif, coordinated by evolutionally conserved residues E497, D498, E501, E502, E507, D513, D517, and N518 (Figures 5C and S6). Positioned at a key architectural junction, this acidic patch connects upward to the TMD via the AHB and downward to the CNBH domain through a flexible linker. To assess the functional role of this acidic patch, we introduced alanine substitutions at several key residues. These mutants exhibited markedly elevated surface expression, demonstrating that the reduced activity of these mutants reflects a genuine functional defect rather than reduced surface localization (Figure S4B and S4C), and Mg2+ efflux assays confirmed substantially reduced transport activity (Figure 5D and 5E). This contrasts with mutations in the MgATP binding site, which completely abolish transport activity46.

Together, these findings suggest a dual role of Mg2+ in CNNM4 regulation: an interfacial Mg2+ salt bridge at the CBSP dimer interface that is absolutely required for transport, and additional Mg2+ to the acidic patch that allosterically tunes activity. We propose that this acidic patch, enriched in acidic residues, acts as a potential Mg2+-sensing module: depending on cytosolic Mg2+ levels, it could bind variable numbers of Mg2+ ions to modulate CNNM4 activity—upward, by directly influencing TMD conformation; and downward, by modulating the CNBH positioning, which feeds back to the TMD to regulate the transport cycle (Figure 2D–2F).

Mechanism of Mg2+ transportIn the MgATP-bound wild-type structure of CNNM4, captured in an outward-facing conformation, each tetramer forms two Mg2+ translocation pathways, one per dimer. These pathways are lined by a mixture of negatively charged and polar residues, creating an environment suitable for divalent cation (Figures 6A and 6B). However, unlike the prokaryotic homolog CorC, captured in an inward-facing state with two bound Mg2+ within its single pathway44, we observed little to no densities corresponding to Mg2+ at the equivalent sites in either CNNM4 dimer, even though the surrounding residues are conserved in the CNNM family (with the exception of CNNM3) (Figure 6C).Comparison of the inward-facing CorC structure with the outward-facing CNNM4 structure provides key insight into how Mg2+ binding and release are coupled to conformational state. In CorC, each subunit binds one Mg2+ near the middle of the pathway in canonical octahedral geometry by a glutamate, an asparagine, two serines, and two backbone carbonyl oxygens (Figure 6D, right panel). This complete coordination cage explains the high Mg2+ affinity in the inward-facing state, consistent with its role in binding cytoplasmic Mg2+. By contrast, in CNNM4 the equivalent sites are disrupted. Most notably, the side chain of E284—a key acidic residue located in the central π-helical segment of S3—flips away from the binding site (Figure 6D, left panel). This conformational change is accompanied by a displacement of the adjacent backbone carbonyl of G283, another Mg2+-coordinating residue. Together, these rearrangements distort the coordination geometry and open the site toward the extracellular space in the outward-facing state, consistent with its role in Mg2+ release rather than binding.Interestingly, we observed three additional ion-like densities near each E284 in CNNM4. These densities are unlikely to represent Mg2+, given their relatively loose coordination geometry and close spatial proximity to one another (~4 Å), which would result in strong electrostatic repulsion between divalent cations. We therefore tentatively assign these densities to Na+ (Na1 and Na1’), consistent with CNNM4’s role as a Mg2+ exporter coupled to Na+ import and with the expectation that the outward-facing state binds Na+ with high affinity (Figure 6E).To test whether the observed densities indeed correspond to Na+, we first attempted to replace Na+ with N-methyl-D-glucamine (NMDG) throughout protein purification and cryo-EM data collection, with the expectation that these densities would either be abolished (as NMDG does not bind) or replaced by NMDG densities at the putative Na+ binding sites. However, under these conditions the protein became unstable and prone to aggregation, preventing the determination of cryo-EM structures at resolutions comparable to those obtained in the presence of Na+. We therefore adopted an alternative strategy by mutating E284 to alanine, reasoning that disruption of this coordinating residue should eliminate Na+ binding at these sites. The structure of the E284A mutant was determined at a resolution comparable to that of the wild type (Figure S3C), enabling a fair and direct structural comparison (Figure 6F). Consistent with our hypothesis, the putative Na+ densities are absent in the E284A structure, supporting their assignment as Na+ (Figure 6F).We propose that Na+ binding stabilizes the flipped conformation of E284, thereby weakening Mg2+ coordination and promoting Mg2+ release. In addition, a third putative ion density (Na2) is observed between the two E284 residues at the intradimer interface (Figure 6E). Although it lies beyond the typical range for direct Na+–sidechain interactions, it may represent a Na+ ion coordinating E284 via water molecules—unresolved at the current map resolution—which could further stabilize the flipped conformationTogether, these findings support a mechanistic cycle for CNNM4-medated Mg2+ transport. In the inward-facing state, Mg2+ binds with high affinity through complete coordination, as seen in CorC. Transitions through an occluded intermediate lead to the outward-facing state, where disrupted coordination, assisted by Na+ binding, lowers Mg2+ affinity and enables Mg2+ release, completing the transport cycle.

Mechanism of Mg2+ transport

In the MgATP-bound wild-type structure of CNNM4, captured in an outward-facing conformation, each tetramer forms two Mg2+ translocation pathways, one per dimer. These pathways are lined by a mixture of negatively charged and polar residues, creating an environment suitable for divalent cation (Figures 6A and 6B). However, unlike the prokaryotic homolog CorC, captured in an inward-facing state with two bound Mg2+ within its single pathway44, we observed little to no densities corresponding to Mg2+ at the equivalent sites in either CNNM4 dimer, even though the surrounding residues are conserved in the CNNM family (with the exception of CNNM3) (Figure 6C).

Comparison of the inward-facing CorC structure with the outward-facing CNNM4 structure provides key insight into how Mg2+ binding and release are coupled to conformational state. In CorC, each subunit binds one Mg2+ near the middle of the pathway in canonical octahedral geometry by a glutamate, an asparagine, two serines, and two backbone carbonyl oxygens (Figure 6D, right panel). This complete coordination cage explains the high Mg2+ affinity in the inward-facing state, consistent with its role in binding cytoplasmic Mg2+. By contrast, in CNNM4 the equivalent sites are disrupted. Most notably, the side chain of E284—a key acidic residue located in the central π-helical segment of S3—flips away from the binding site (Figure 6D, left panel). This conformational change is accompanied by a displacement of the adjacent backbone carbonyl of G283, another Mg2+-coordinating residue. Together, these rearrangements distort the coordination geometry and open the site toward the extracellular space in the outward-facing state, consistent with its role in Mg2+ release rather than binding.

Interestingly, we observed three additional ion-like densities near each E284 in CNNM4. These densities are unlikely to represent Mg2+, given their relatively loose coordination geometry and close spatial proximity to one another (~4 Å), which would result in strong electrostatic repulsion between divalent cations. We therefore tentatively assign these densities to Na+ (Na1 and Na1’), consistent with CNNM4’s role as a Mg2+ exporter coupled to Na+ import and with the expectation that the outward-facing state binds Na+ with high affinity (Figure 6E).

To test whether the observed densities indeed correspond to Na+, we first attempted to replace Na+ with N-methyl-D-glucamine (NMDG) throughout protein purification and cryo-EM data collection, with the expectation that these densities would either be abolished (as NMDG does not bind) or replaced by NMDG densities at the putative Na+ binding sites. However, under these conditions the protein became unstable and prone to aggregation, preventing the determination of cryo-EM structures at resolutions comparable to those obtained in the presence of Na+. We therefore adopted an alternative strategy by mutating E284 to alanine, reasoning that disruption of this coordinating residue should eliminate Na+ binding at these sites. The structure of the E284A mutant was determined at a resolution comparable to that of the wild type (Figure S3C), enabling a fair and direct structural comparison (Figure 6F). Consistent with our hypothesis, the putative Na+ densities are absent in the E284A structure, supporting their assignment as Na+ (Figure 6F).

We propose that Na+ binding stabilizes the flipped conformation of E284, thereby weakening Mg2+ coordination and promoting Mg2+ release. In addition, a third putative ion density (Na2) is observed between the two E284 residues at the intradimer interface (Figure 6E). Although it lies beyond the typical range for direct Na+–sidechain interactions, it may represent a Na+ ion coordinating E284 via water molecules—unresolved at the current map resolution—which could further stabilize the flipped conformation

Together, these findings support a mechanistic cycle for CNNM4-medated Mg2+ transport. In the inward-facing state, Mg2+ binds with high affinity through complete coordination, as seen in CorC. Transitions through an occluded intermediate lead to the outward-facing state, where disrupted coordination, assisted by Na+ binding, lowers Mg2+ affinity and enables Mg2+ release, completing the transport cycle.

DiscussionOur structural and functional study reveals the architecture of full-length CNNM4, a eukaryotic Mg2+/Na+ antiporter, and establishes a molecular basis for how it mediates Mg2+/Na+ exchange and Mg2+ sensing. CNNM4 assembles as a multidomain tetramer organized as a dimer of dimers, with subunit non-equivalence among otherwise four identical polypeptides. This architecture diverges fundamentally from the homodimeric prokaryotic homologs: while superficially resembling paired prokaryotic dimers, CNNM4 incorporates extensive domain swapping within and between dimers, creating interfaces that integrate the two transporter units into an indivisible regulatory assembly. The emergence of inter-dimer interfaces in both the extracellular ECD and intracellular CNBH domain—features absent in prokaryotic homologs—indicates that eukaryotic CNNMs have not simply duplicated prokaryotic transporters but represent an evolutionary innovation, adding layers of regulation that may couple transport activity to cellular signaling. Supporting this, mutations in these impair transport and are linked to human disease phenotypes, underscoring the physiological importance of these regulatory interfaces.A key feature of this evolutionary innovation is the dynamic nature of the intracellular domains, which form a stable tetramer only upon MgATP binding to the CBSP domain—a process essential for transport activity. Adjacent to the ATP-binding cavity, we further identify an evolutionally conserved acidic patch that accommodates additional Mg2+ ions. Functional analysis shows that this patch tunes, but does not abolish, transport, suggesting its role as a Mg2+ sensor. Together with the MgATP-binding site, this patch likely couples CNNM4 output to cytosolic Mg2+ levels and metabolic status, enabling graded activity states according to cellular demand.While CNNM4 adopts a distinctive dimer-of-asymmetric-dimers architecture, higher-order oligomerization is also observed in other transport systems. For example, the mitochondrial calcium uniporter (MCU) forms a symmetric tetrameric pore that can further dimerize through its N-terminal domains, giving rise to V-shaped dimers of tetramers in the holocomplex55,56. In addition, although direct structural evidence remains limited, dimer-of-dimers or higher-order assemblies have been proposed for several solute carrier (SLC) transporter families, including the human dopamine transporter (DAT) in the SLC6 family57, as well as organic cation transporters OCT1 and OCT2 in the SLC22 family58,59. In this broader context, the asymmetric dimer-of-dimers organization of CNNM4 stands out by introducing additional layers of regulation through extensive domain swapping and multiple inter-dimer interfaces spanning both extracellular and intracellular domains. These features likely enhance structural stability and enable cooperative regulation across transporter subunits, distinguishing CNNM4 from previously described higher-order assemblies. Together with earlier biochemical and microscopy studies, our structures support the emerging view that higher-order organization is likely a common and functionally relevant feature of many membrane transporters, facilitating coordinated regulation, stabilizing transport pathways, and ensuring robust integration within the dynamic lipid bilayer.We captured CNNM4 in two distinct conformational states—outward-facing and occluded—previously unseen in either the CNNM family or their prokaryotic homologs. Transition between these states involves asymmetric yet coordinated motions across all four structural layers, reflecting intricate interdomain coupling. A key transducer revealed by our structures is the ECD–TMD linker, which extends across nearly one-third of the membrane bilayer and harbors an unusual membrane-buried arginine residue. Substitution of this residue abolishes transport activity, highlighting its role in coupling extracellular cues to TMD motions. These findings suggest a “string-puppet” mechanism, in which the ECD dimer directs the TMD dynamics through the connecting linkers. Comparison of our outward-facing and occluded CNNM4 structures with the inward-facing conformation of the prokaryotic homolog provides a framework for the transport cycle. At the conserved Mg2+-binding site within the TMD, Mg2+ coordination observed in the inward-facing stat—consistent with Mg2+ loading—is disrupted in the outward-facing state by Na+-facilitated flipping of a conserved glutamate (E284). This rearrangement reduces Mg2+ affinity and promotes release—consistent with Mg2+ unloading—implicating Na+ as both a counter-ion and a structural cofactor in the transport cycle. We further demonstrate that replacing Na+ with K+ is required to trap the occluded conformation in a crosslinking-sensitive mutant, suggesting that monovalent cations modulate conformational equilibrium. This finding raises the possibility that CNNM4 responds dynamically to extracellular Na+ and/or intracellular K+ levels, adding an additional regulatory layer beyond Mg2+ sensing.CNNMs are implicated in magnesium homeostasis in humans and animals. Mutations in CNNMs are directly related to various human diseases14. Our structures may provide novel insights on some CNNMs’ disease mutations. Hypomagnesemia-associated CNNM2 mutations E122K26 (E95 in CNNM4) and Y189C29 (Y148 in CNNM4) are located at or near the dimer–dimer interface in the ECD. S795L27 (S699 in CNNM4) lies close to the dimer–dimer interface within the CNBH domain. D335N27 (D262 in CNNM4) corresponds to the position we used for crosslinking to achieve the occluded conformation. The CNNM3 mutation Y628D51 (F698 in CNNM4), which has been linked to schizophrenia, is also located near the dimer–dimer interface in the CNBH domain. We also provide information on the CNNM4 D498E60 mutation associated with Jalili syndrome, which may be involved in binding additional magnesium.Together, our findings reveal how CNNM4 integrates an unexpected asymmetric tetrameric architecture, dynamic intracellular domains stabilized by MgATP, an acidic Mg2+-sensing patch, and Na+-facilitated cargo release to achieve regulated Mg2+/Na+ exchange. These insights establish CNNM4 as a paradigm for how eukaryotic transporters evolve additional layers of cooperative regulation absent in prokaryotic homologs. Capturing an inward-facing state will be essential to complete the transport cycle and fully define how intra- and inter-dimer communication drives allosteric control.Limitations of the studyIn this study, we present cryo-EM structures of full-length human CNNM4 in outward-facing and occluded states in a dimer-of-dimers assembly. While the comparison of these structures with the inward-facing bacterial homolog structure provides mechanistic insights into how CNNM4 binds and releases Mg2+, the bacterial homologs lack the two key regulatory domains and do not form the dimer-of-dimers architecture (and therefore lack domain swapping). As a result, the insights from such comparisons are inherently limited. To fully understand the transport mechanism—and how the regulatory domains contribute to the function of the Mg2+ transporter—we will ultimately need structures of CNNM4 in all key functional states, particularly in the inward-facing state.

Our structural and functional study reveals the architecture of full-length CNNM4, a eukaryotic Mg2+/Na+ antiporter, and establishes a molecular basis for how it mediates Mg2+/Na+ exchange and Mg2+ sensing. CNNM4 assembles as a multidomain tetramer organized as a dimer of dimers, with subunit non-equivalence among otherwise four identical polypeptides. This architecture diverges fundamentally from the homodimeric prokaryotic homologs: while superficially resembling paired prokaryotic dimers, CNNM4 incorporates extensive domain swapping within and between dimers, creating interfaces that integrate the two transporter units into an indivisible regulatory assembly. The emergence of inter-dimer interfaces in both the extracellular ECD and intracellular CNBH domain—features absent in prokaryotic homologs—indicates that eukaryotic CNNMs have not simply duplicated prokaryotic transporters but represent an evolutionary innovation, adding layers of regulation that may couple transport activity to cellular signaling. Supporting this, mutations in these impair transport and are linked to human disease phenotypes, underscoring the physiological importance of these regulatory interfaces.

A key feature of this evolutionary innovation is the dynamic nature of the intracellular domains, which form a stable tetramer only upon MgATP binding to the CBSP domain—a process essential for transport activity. Adjacent to the ATP-binding cavity, we further identify an evolutionally conserved acidic patch that accommodates additional Mg2+ ions. Functional analysis shows that this patch tunes, but does not abolish, transport, suggesting its role as a Mg2+ sensor. Together with the MgATP-binding site, this patch likely couples CNNM4 output to cytosolic Mg2+ levels and metabolic status, enabling graded activity states according to cellular demand.

While CNNM4 adopts a distinctive dimer-of-asymmetric-dimers architecture, higher-order oligomerization is also observed in other transport systems. For example, the mitochondrial calcium uniporter (MCU) forms a symmetric tetrameric pore that can further dimerize through its N-terminal domains, giving rise to V-shaped dimers of tetramers in the holocomplex55,56. In addition, although direct structural evidence remains limited, dimer-of-dimers or higher-order assemblies have been proposed for several solute carrier (SLC) transporter families, including the human dopamine transporter (DAT) in the SLC6 family57, as well as organic cation transporters OCT1 and OCT2 in the SLC22 family58,59. In this broader context, the asymmetric dimer-of-dimers organization of CNNM4 stands out by introducing additional layers of regulation through extensive domain swapping and multiple inter-dimer interfaces spanning both extracellular and intracellular domains. These features likely enhance structural stability and enable cooperative regulation across transporter subunits, distinguishing CNNM4 from previously described higher-order assemblies. Together with earlier biochemical and microscopy studies, our structures support the emerging view that higher-order organization is likely a common and functionally relevant feature of many membrane transporters, facilitating coordinated regulation, stabilizing transport pathways, and ensuring robust integration within the dynamic lipid bilayer.

We captured CNNM4 in two distinct conformational states—outward-facing and occluded—previously unseen in either the CNNM family or their prokaryotic homologs. Transition between these states involves asymmetric yet coordinated motions across all four structural layers, reflecting intricate interdomain coupling. A key transducer revealed by our structures is the ECD–TMD linker, which extends across nearly one-third of the membrane bilayer and harbors an unusual membrane-buried arginine residue. Substitution of this residue abolishes transport activity, highlighting its role in coupling extracellular cues to TMD motions. These findings suggest a “string-puppet” mechanism, in which the ECD dimer directs the TMD dynamics through the connecting linkers. Comparison of our outward-facing and occluded CNNM4 structures with the inward-facing conformation of the prokaryotic homolog provides a framework for the transport cycle. At the conserved Mg2+-binding site within the TMD, Mg2+ coordination observed in the inward-facing stat—consistent with Mg2+ loading—is disrupted in the outward-facing state by Na+-facilitated flipping of a conserved glutamate (E284). This rearrangement reduces Mg2+ affinity and promotes release—consistent with Mg2+ unloading—implicating Na+ as both a counter-ion and a structural cofactor in the transport cycle. We further demonstrate that replacing Na+ with K+ is required to trap the occluded conformation in a crosslinking-sensitive mutant, suggesting that monovalent cations modulate conformational equilibrium. This finding raises the possibility that CNNM4 responds dynamically to extracellular Na+ and/or intracellular K+ levels, adding an additional regulatory layer beyond Mg2+ sensing.

CNNMs are implicated in magnesium homeostasis in humans and animals. Mutations in CNNMs are directly related to various human diseases14. Our structures may provide novel insights on some CNNMs’ disease mutations. Hypomagnesemia-associated CNNM2 mutations E122K26 (E95 in CNNM4) and Y189C29 (Y148 in CNNM4) are located at or near the dimer–dimer interface in the ECD. S795L27 (S699 in CNNM4) lies close to the dimer–dimer interface within the CNBH domain. D335N27 (D262 in CNNM4) corresponds to the position we used for crosslinking to achieve the occluded conformation. The CNNM3 mutation Y628D51 (F698 in CNNM4), which has been linked to schizophrenia, is also located near the dimer–dimer interface in the CNBH domain. We also provide information on the CNNM4 D498E60 mutation associated with Jalili syndrome, which may be involved in binding additional magnesium.

Together, our findings reveal how CNNM4 integrates an unexpected asymmetric tetrameric architecture, dynamic intracellular domains stabilized by MgATP, an acidic Mg2+-sensing patch, and Na+-facilitated cargo release to achieve regulated Mg2+/Na+ exchange. These insights establish CNNM4 as a paradigm for how eukaryotic transporters evolve additional layers of cooperative regulation absent in prokaryotic homologs. Capturing an inward-facing state will be essential to complete the transport cycle and fully define how intra- and inter-dimer communication drives allosteric control.

Limitations of the studyIn this study, we present cryo-EM structures of full-length human CNNM4 in outward-facing and occluded states in a dimer-of-dimers assembly. While the comparison of these structures with the inward-facing bacterial homolog structure provides mechanistic insights into how CNNM4 binds and releases Mg2+, the bacterial homologs lack the two key regulatory domains and do not form the dimer-of-dimers architecture (and therefore lack domain swapping). As a result, the insights from such comparisons are inherently limited. To fully understand the transport mechanism—and how the regulatory domains contribute to the function of the Mg2+ transporter—we will ultimately need structures of CNNM4 in all key functional states, particularly in the inward-facing state.

Limitations of the study

In this study, we present cryo-EM structures of full-length human CNNM4 in outward-facing and occluded states in a dimer-of-dimers assembly. While the comparison of these structures with the inward-facing bacterial homolog structure provides mechanistic insights into how CNNM4 binds and releases Mg2+, the bacterial homologs lack the two key regulatory domains and do not form the dimer-of-dimers architecture (and therefore lack domain swapping). As a result, the insights from such comparisons are inherently limited. To fully understand the transport mechanism—and how the regulatory domains contribute to the function of the Mg2+ transporter—we will ultimately need structures of CNNM4 in all key functional states, particularly in the inward-facing state.

Resource availabilityLead contactRequests for further information and resources should be directed to and will be fulfilled by the lead contact, Juan Du (juan.du@northwestern.edu).Materials availabilityThis study did not generate any new, unique reagents. Plasmids generated in this study are available upon request to lead contact.Data and code availabilityThe Cryo-EM density maps have been deposited at the EMDB (Electron Microscopy Data Bank) under accession codes EMD-72694 (WT–CNNM4_Mg/ATP), EMD-72696 (D262C–CNNM4_Mg/ATP), EMD-75682 (E284A–CNNM4_Mg/ATP), EMD-72697 (K113A/R140A/R141A–CNNM4_Mg/ATP). The atomic coordinates have been deposited in the Protein Data Bank (PDB) under accession codes 9Y9D (WT–CNNM4_Mg/ATP), 9Y9F (D262C–CNNM4_Mg/ATP), 11GQ (E284A–CNNM4_Mg/ATP), 9Y9G (K113A/R140A/R141A–CNNM4_Mg/ATP).

Resource availability

Lead contactRequests for further information and resources should be directed to and will be fulfilled by the lead contact, Juan Du (juan.du@northwestern.edu).

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Juan Du (juan.du@northwestern.edu).

Materials availabilityThis study did not generate any new, unique reagents. Plasmids generated in this study are available upon request to lead contact.

Materials availability

This study did not generate any new, unique reagents. Plasmids generated in this study are available upon request to lead contact.

Data and code availabilityThe Cryo-EM density maps have been deposited at the EMDB (Electron Microscopy Data Bank) under accession codes EMD-72694 (WT–CNNM4_Mg/ATP), EMD-72696 (D262C–CNNM4_Mg/ATP), EMD-75682 (E284A–CNNM4_Mg/ATP), EMD-72697 (K113A/R140A/R141A–CNNM4_Mg/ATP). The atomic coordinates have been deposited in the Protein Data Bank (PDB) under accession codes 9Y9D (WT–CNNM4_Mg/ATP), 9Y9F (D262C–CNNM4_Mg/ATP), 11GQ (E284A–CNNM4_Mg/ATP), 9Y9G (K113A/R140A/R141A–CNNM4_Mg/ATP).

Data and code availability

The Cryo-EM density maps have been deposited at the EMDB (Electron Microscopy Data Bank) under accession codes EMD-72694 (WT–CNNM4_Mg/ATP), EMD-72696 (D262C–CNNM4_Mg/ATP), EMD-75682 (E284A–CNNM4_Mg/ATP), EMD-72697 (K113A/R140A/R141A–CNNM4_Mg/ATP). The atomic coordinates have been deposited in the Protein Data Bank (PDB) under accession codes 9Y9D (WT–CNNM4_Mg/ATP), 9Y9F (D262C–CNNM4_Mg/ATP), 11GQ (E284A–CNNM4_Mg/ATP), 9Y9G (K113A/R140A/R141A–CNNM4_Mg/ATP).

STAR★METHODSEXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILSBacterial strainsEscherichia coli strain DH10Bac was cultured at 37°C in LB medium with 50 μg/mL kanamycin, 7 μg/mL gentamicin, and 10 μg/mL tetracycline for extraction of recombinant bacmid.Cell linesSpodoptera frugiperda (Sf9) insect cells (Thermo Fisher Scientific, Cat# 12659017; RRID:CVCL_0549) were grown in Sf9-900 II SFM (Thermo Fisher Scientific, Cat# 10902187) at 27°C in a shaker incubator.HEK293T cells were maintained in Freestyle 293 Expression medium in suspension culture (Thermo Fisher Scientific, Cat#12338026, RRID:CVCL_2737) at 37°C in a shaker incubator.HEK293 cells (ATCC, Cat# CRL-1573, RRID:CVCL_0045) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM, Thermo Fisher Scientific, Cat# 10566016) supplemented with 5% (v/v) fetal bovine serum (FBS) at 37°C in a humidified incubator with 5% CO2.METHOD DETAILSCNNM4 expression and purificationGene encoding full-length human CNNM4 (UniProtKB accession numbers Q6P4Q7) was synthesized by Genescript and was subcloned into a pEG-BacMam vector with a His tag, twin-strep tag and alpha-tag at C-terminus61,62. Site-directed mutagenesis was performed by using Q5 Site-Directed Mutagenesis (NEB) or QuikChange II Site-Directed Mutagenesis (Qiagen) kit and the provided protocols, and confirmed by whole plasmid sequencing and Sanger sequencing (Plasmidsaurus/Eurofins). For baculovirus production, CNNM4 in the BacMam vector was transformed into DH10Bac cells, followed by P1 and P2 baculovirus generation in Sf9 cells. P2 viruses (8%) were used to infect HEK293T cells grown in Freestyle 293 Expression medium in suspension culture (Thermo Fisher Scientific). Infected cells were incubated for an initial 8–10 h at 37°C before addition of 10 mM sodium butyrate. Cells were then moved to a 30°C incubator and allowed to grow for another 48 h with vigorous shaking. At 58 h post-infection, cells were collected by centrifugation at 5,000 r.p.m., at 4°C for 30 min. Cell pellets were washed with buffer containing 150 mM NaCl and 20 mM Tris pH 8.0 (TBS buffer) and stored at −80°C. Cell pellets from a 800-ml culture were thawed on ice and resuspended in TBS buffer containing 1 mM PMSF, 0.8 μM aprotinin, 2 μg/mL leupeptin, 2 mM pepstatin A (all protease inhibitors) plus 2% GDN detergent (Anatrace). Protein was extracted from the membrane by whole-cell solubilization for 1h at 4°C with rotation. The solubilized protein was loaded with 5–10 ml of Strep-Tactin resin (IBA) and washed with TBS buffer supplemented with 0.02% GDN. Protein was eluted with TBS buffer supplemented with 0.02% GDN and 10mM Desthiobiotin. The eluent was concentrated to 500 μl and further purified by size-exclusion chromatography in TBS buffer containing 0.02% GDN. Peak fractions containing CNNM4 were pooled and concentrated to 6–10 mg/ml for grids freezing.Electron microscopy sample preparation and data acquisitionFreshly purified CNNM4 protein in GDN detergent was mixed with 5 mM MgCl2 and 2 mM ATP before grid preparation. A 2.5-μl aliquot of the sample was applied to a glow-discharged Quantifoil holey carbon grid (gold, 2/1-μm size/hole space, 300-mesh), blotted for 1.5 s at 100% humidity using Vitrobot, and then plunge-frozen in liquid ethane cooled by liquid nitrogen. The grids were loaded into an FEI Titan Krios transmission electron microscope operating at 300 kV with a nominal magnification of 105,000X and an energy filter (20-eV slit width) with a binned pixel size of 0.828 Å, or FEI Glacios transmission electron microscope operating at 200kV with a nominal magnification of 130,000X and an energy filter (10-eV slit width) with a pixel size of 0.91Å. Each Krios video was dose-fractionated to 50 frames with a total dose of 58 e−/Å2. Image acquisition was facilitated using SerialEM, and the nominal defocus range was set from −0.1 μm to −1.6 μm. Each Glacios video was captured with a total dose of 40 e−/Å2 by employing EPU, with the nominal defocus range from −0.6 μm to −1.3 μm.Cryo-electron microscopy data analysis procedureThe detailed workflow for the data processing procedure is summarized in Figures S1 and S4. In general, the raw super-resolution tif or eer video files for each dataset were motion-corrected using MotionCor263 or RELION5.064. The per-micrograph defocus values were estimated using ctffind465. Particle picking was performed using topaz66 and/or reference-based template match. Junk particles were removed by two-dimensional (2D) classification and heterogeneous refinement using CryoSPARC67. Selected good particles were then used to generate an initial 3D model by ab initio reconstruction followed by homogeneous refinement with C2 symmetry. Multiple rounds of CTF refinement and Bayesian polishing were performed in RELION to further improve the map resolution. To further improve the map quality, we performed focused classification by subtracting ICD signals from the particles. For all datasets, the gold standard Fourier shell correlation (FSC) 0.143 criteria was used to provide the map resolution estimation. The cryo-EM maps were visualized using UCSF ChimeraX/Chimera68,69. PDB structures are visualized using UCSF ChimeraX/Chimera or PyMOL70.Model building and structure refinementFor the WT CNNM4/MgATP model, the original monomer model of human CNNM4 (UniProt:Q6P4Q7) from AlphaFold Protein Structure Database71 was docked into the cryo-EM map using UCSF Chimera69. The D262C CNNM4/MgATP, E284A CNNM4/MgATP and K113A/R140A/R141A CNNM4/MgATP models were built by docking the WT CNNM4/MgATP model into the respective density maps. All the models were manually rebuilt and adjusted to fit into the density map in Coot (Coot v0.89)72 and subjected to real-space refinement in Phenix73 against the density maps. The model statistics were calculated using MolProbity in Phenix. Structural figures were prepared in ChimeraX68 and PyMOL70. The final data and refinement statistics are provided in Table S1.Mg2+ efflux assay and statistical analysesMg2+ efflux activity was judged by magnesium efflux assay following published methods22,44. HEK293T cells were plated in Poly-L-Lysine (Sigma-Aldrich P4707) coated coverslips (Fisher Scientific™ Cat.No.1251000) in 24-well plates, human CNNM4-mcherry (wildtype or mutants) or control constructs were transformed by lipofectamine 3000 (Invitrogen™ L3000008) and cultured with growth medium supplemented with 40 mM Mg2+. 24 hours posttransfection, cells were used for imaging. Live-cell Mg2+ fluorescence imaging was performed using a Prime BSI Express sCMOS camera (Teledyne Photometrics) controlled by the open-source Micro-Manager platform (v2.0.3, https://micro-manager.org/). Camera exposure and LED illumination were synchronized using an Arduino-based TTL pulse generator operated through a custom Python GUI (source code available at: https://github.com/junuk861/arduino-ttl-pulse-gui). Illumination was provided by the 470-nm LED channel of the CoolLED pE-800 Fura system (LightBridge v1.3.0.0) through a 470/40x excitation filter (450–490 nm). Emitted fluorescence was transmitted through a 495LP dichroic mirror and collected with a 525/50m emission filter (500–550 nm). Cells were incubated with Mg2+-loading buffer (78.1 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl2, 40 mM MgCl2, 5.5 mM glucose, 5.5 mM HEPES-KOH, pH 7.4) containing 2 μM Magnesium Green-AM (Invitrogen M3735) for 45 min at 37 °C. Cells were imaged every 20 sec. 1 mins after start imaging, the extracellular solution was then replaced with Mg2+-free buffer (MgCl2 substituted with 60 mM NaCl). Data were quantified with ImageJ, and the fluorescence intensity of each cell at each time point was normalized to the corresponding value at 0 s. Statistical analyses were performed using GraphPad Prism (version 10.6.1). Data are shown as means ± SD, and statistical significance was determined using One-Way ANOVA test.Surface protein labeling and normalization of Mg2+ efflux activity.Surface protein labeling experiments were performed using the Pierce™ Cell Surface Protein Isolation Kit (Thermo Fisher Scientific). Surface and total protein levels of wild-type and mutant human CNNM4 were assessed by in-gel fluorescence of the C-terminal mCherry tag. Actin was used as a loading control and was detected by western blotting with an HRP-conjugated anti-beta actin antibody (Proteintech, HRP-60008).Normalization of Mg2+ efflux activity was performed in two steps. First, surface expression levels of each CNNM4 mutant were normalized to actin and then divided by the corresponding actin-normalized value of wild-type CNNM4. Ratios obtained from three independent experiments were averaged to yield a mean scaling factor for each mutant (M). Second, Mg2+ efflux time-course data were corrected using this scaling factor. For a given mutant, the raw efflux value at a specific time point was defined as Eraw, and the corresponding value from the negative control (cells transfected with the parent mCherry plasmid lacking CNNM4) at the same time point was defined as E0. The normalized efflux value was calculated as:

Enorm=E0+Eraw-E0M
This normalization rescales the CNNM4-dependent component of Mg2+ efflux assuming a linear relationship between surface expression level and transport activity. The normalized value replaced Eraw at each time point to generate the corrected time-course curve.Qualitative assessment of CNNM4 membrane localization by confocal microscopy.HEK293T cells expressing wild-type or mutant human CNNM4 tagged with a C-temrinal mCherry were fixed in PBS containing 3.7% formaldehyde and imaged using a Nikon SoRa spinning disk confocal microscope equipped with a 60x objective. Confocal imaging was performed to qualitatively confirm protein expression and plasma membrane localization, rather than for quantitative analysis. Alexa Fluor 488–conjugated phalloidin (Invitrogen™ A12379) was used to label F-actin.QUANTIFICATION AND STATISTICAL ANALYSISStatistical analyses were performed using GraphPad Prism v10.6.1 (GraphPad Software, LLC). Relative quantification methods and tools are described in each section of the methods and/or figure legends.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILSBacterial strainsEscherichia coli strain DH10Bac was cultured at 37°C in LB medium with 50 μg/mL kanamycin, 7 μg/mL gentamicin, and 10 μg/mL tetracycline for extraction of recombinant bacmid.Cell linesSpodoptera frugiperda (Sf9) insect cells (Thermo Fisher Scientific, Cat# 12659017; RRID:CVCL_0549) were grown in Sf9-900 II SFM (Thermo Fisher Scientific, Cat# 10902187) at 27°C in a shaker incubator.HEK293T cells were maintained in Freestyle 293 Expression medium in suspension culture (Thermo Fisher Scientific, Cat#12338026, RRID:CVCL_2737) at 37°C in a shaker incubator.HEK293 cells (ATCC, Cat# CRL-1573, RRID:CVCL_0045) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM, Thermo Fisher Scientific, Cat# 10566016) supplemented with 5% (v/v) fetal bovine serum (FBS) at 37°C in a humidified incubator with 5% CO2.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Bacterial strainsEscherichia coli strain DH10Bac was cultured at 37°C in LB medium with 50 μg/mL kanamycin, 7 μg/mL gentamicin, and 10 μg/mL tetracycline for extraction of recombinant bacmid.

Bacterial strains

Escherichia coli strain DH10Bac was cultured at 37°C in LB medium with 50 μg/mL kanamycin, 7 μg/mL gentamicin, and 10 μg/mL tetracycline for extraction of recombinant bacmid.

Cell linesSpodoptera frugiperda (Sf9) insect cells (Thermo Fisher Scientific, Cat# 12659017; RRID:CVCL_0549) were grown in Sf9-900 II SFM (Thermo Fisher Scientific, Cat# 10902187) at 27°C in a shaker incubator.HEK293T cells were maintained in Freestyle 293 Expression medium in suspension culture (Thermo Fisher Scientific, Cat#12338026, RRID:CVCL_2737) at 37°C in a shaker incubator.HEK293 cells (ATCC, Cat# CRL-1573, RRID:CVCL_0045) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM, Thermo Fisher Scientific, Cat# 10566016) supplemented with 5% (v/v) fetal bovine serum (FBS) at 37°C in a humidified incubator with 5% CO2.

Spodoptera frugiperda (Sf9) insect cells (Thermo Fisher Scientific, Cat# 12659017; RRID:CVCL_0549) were grown in Sf9-900 II SFM (Thermo Fisher Scientific, Cat# 10902187) at 27°C in a shaker incubator.

HEK293T cells were maintained in Freestyle 293 Expression medium in suspension culture (Thermo Fisher Scientific, Cat#12338026, RRID:CVCL_2737) at 37°C in a shaker incubator.

HEK293 cells (ATCC, Cat# CRL-1573, RRID:CVCL_0045) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM, Thermo Fisher Scientific, Cat# 10566016) supplemented with 5% (v/v) fetal bovine serum (FBS) at 37°C in a humidified incubator with 5% CO2.

METHOD DETAILSCNNM4 expression and purificationGene encoding full-length human CNNM4 (UniProtKB accession numbers Q6P4Q7) was synthesized by Genescript and was subcloned into a pEG-BacMam vector with a His tag, twin-strep tag and alpha-tag at C-terminus61,62. Site-directed mutagenesis was performed by using Q5 Site-Directed Mutagenesis (NEB) or QuikChange II Site-Directed Mutagenesis (Qiagen) kit and the provided protocols, and confirmed by whole plasmid sequencing and Sanger sequencing (Plasmidsaurus/Eurofins). For baculovirus production, CNNM4 in the BacMam vector was transformed into DH10Bac cells, followed by P1 and P2 baculovirus generation in Sf9 cells. P2 viruses (8%) were used to infect HEK293T cells grown in Freestyle 293 Expression medium in suspension culture (Thermo Fisher Scientific). Infected cells were incubated for an initial 8–10 h at 37°C before addition of 10 mM sodium butyrate. Cells were then moved to a 30°C incubator and allowed to grow for another 48 h with vigorous shaking. At 58 h post-infection, cells were collected by centrifugation at 5,000 r.p.m., at 4°C for 30 min. Cell pellets were washed with buffer containing 150 mM NaCl and 20 mM Tris pH 8.0 (TBS buffer) and stored at −80°C. Cell pellets from a 800-ml culture were thawed on ice and resuspended in TBS buffer containing 1 mM PMSF, 0.8 μM aprotinin, 2 μg/mL leupeptin, 2 mM pepstatin A (all protease inhibitors) plus 2% GDN detergent (Anatrace). Protein was extracted from the membrane by whole-cell solubilization for 1h at 4°C with rotation. The solubilized protein was loaded with 5–10 ml of Strep-Tactin resin (IBA) and washed with TBS buffer supplemented with 0.02% GDN. Protein was eluted with TBS buffer supplemented with 0.02% GDN and 10mM Desthiobiotin. The eluent was concentrated to 500 μl and further purified by size-exclusion chromatography in TBS buffer containing 0.02% GDN. Peak fractions containing CNNM4 were pooled and concentrated to 6–10 mg/ml for grids freezing.Electron microscopy sample preparation and data acquisitionFreshly purified CNNM4 protein in GDN detergent was mixed with 5 mM MgCl2 and 2 mM ATP before grid preparation. A 2.5-μl aliquot of the sample was applied to a glow-discharged Quantifoil holey carbon grid (gold, 2/1-μm size/hole space, 300-mesh), blotted for 1.5 s at 100% humidity using Vitrobot, and then plunge-frozen in liquid ethane cooled by liquid nitrogen. The grids were loaded into an FEI Titan Krios transmission electron microscope operating at 300 kV with a nominal magnification of 105,000X and an energy filter (20-eV slit width) with a binned pixel size of 0.828 Å, or FEI Glacios transmission electron microscope operating at 200kV with a nominal magnification of 130,000X and an energy filter (10-eV slit width) with a pixel size of 0.91Å. Each Krios video was dose-fractionated to 50 frames with a total dose of 58 e−/Å2. Image acquisition was facilitated using SerialEM, and the nominal defocus range was set from −0.1 μm to −1.6 μm. Each Glacios video was captured with a total dose of 40 e−/Å2 by employing EPU, with the nominal defocus range from −0.6 μm to −1.3 μm.Cryo-electron microscopy data analysis procedureThe detailed workflow for the data processing procedure is summarized in Figures S1 and S4. In general, the raw super-resolution tif or eer video files for each dataset were motion-corrected using MotionCor263 or RELION5.064. The per-micrograph defocus values were estimated using ctffind465. Particle picking was performed using topaz66 and/or reference-based template match. Junk particles were removed by two-dimensional (2D) classification and heterogeneous refinement using CryoSPARC67. Selected good particles were then used to generate an initial 3D model by ab initio reconstruction followed by homogeneous refinement with C2 symmetry. Multiple rounds of CTF refinement and Bayesian polishing were performed in RELION to further improve the map resolution. To further improve the map quality, we performed focused classification by subtracting ICD signals from the particles. For all datasets, the gold standard Fourier shell correlation (FSC) 0.143 criteria was used to provide the map resolution estimation. The cryo-EM maps were visualized using UCSF ChimeraX/Chimera68,69. PDB structures are visualized using UCSF ChimeraX/Chimera or PyMOL70.Model building and structure refinementFor the WT CNNM4/MgATP model, the original monomer model of human CNNM4 (UniProt:Q6P4Q7) from AlphaFold Protein Structure Database71 was docked into the cryo-EM map using UCSF Chimera69. The D262C CNNM4/MgATP, E284A CNNM4/MgATP and K113A/R140A/R141A CNNM4/MgATP models were built by docking the WT CNNM4/MgATP model into the respective density maps. All the models were manually rebuilt and adjusted to fit into the density map in Coot (Coot v0.89)72 and subjected to real-space refinement in Phenix73 against the density maps. The model statistics were calculated using MolProbity in Phenix. Structural figures were prepared in ChimeraX68 and PyMOL70. The final data and refinement statistics are provided in Table S1.Mg2+ efflux assay and statistical analysesMg2+ efflux activity was judged by magnesium efflux assay following published methods22,44. HEK293T cells were plated in Poly-L-Lysine (Sigma-Aldrich P4707) coated coverslips (Fisher Scientific™ Cat.No.1251000) in 24-well plates, human CNNM4-mcherry (wildtype or mutants) or control constructs were transformed by lipofectamine 3000 (Invitrogen™ L3000008) and cultured with growth medium supplemented with 40 mM Mg2+. 24 hours posttransfection, cells were used for imaging. Live-cell Mg2+ fluorescence imaging was performed using a Prime BSI Express sCMOS camera (Teledyne Photometrics) controlled by the open-source Micro-Manager platform (v2.0.3, https://micro-manager.org/). Camera exposure and LED illumination were synchronized using an Arduino-based TTL pulse generator operated through a custom Python GUI (source code available at: https://github.com/junuk861/arduino-ttl-pulse-gui). Illumination was provided by the 470-nm LED channel of the CoolLED pE-800 Fura system (LightBridge v1.3.0.0) through a 470/40x excitation filter (450–490 nm). Emitted fluorescence was transmitted through a 495LP dichroic mirror and collected with a 525/50m emission filter (500–550 nm). Cells were incubated with Mg2+-loading buffer (78.1 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl2, 40 mM MgCl2, 5.5 mM glucose, 5.5 mM HEPES-KOH, pH 7.4) containing 2 μM Magnesium Green-AM (Invitrogen M3735) for 45 min at 37 °C. Cells were imaged every 20 sec. 1 mins after start imaging, the extracellular solution was then replaced with Mg2+-free buffer (MgCl2 substituted with 60 mM NaCl). Data were quantified with ImageJ, and the fluorescence intensity of each cell at each time point was normalized to the corresponding value at 0 s. Statistical analyses were performed using GraphPad Prism (version 10.6.1). Data are shown as means ± SD, and statistical significance was determined using One-Way ANOVA test.Surface protein labeling and normalization of Mg2+ efflux activity.Surface protein labeling experiments were performed using the Pierce™ Cell Surface Protein Isolation Kit (Thermo Fisher Scientific). Surface and total protein levels of wild-type and mutant human CNNM4 were assessed by in-gel fluorescence of the C-terminal mCherry tag. Actin was used as a loading control and was detected by western blotting with an HRP-conjugated anti-beta actin antibody (Proteintech, HRP-60008).Normalization of Mg2+ efflux activity was performed in two steps. First, surface expression levels of each CNNM4 mutant were normalized to actin and then divided by the corresponding actin-normalized value of wild-type CNNM4. Ratios obtained from three independent experiments were averaged to yield a mean scaling factor for each mutant (M). Second, Mg2+ efflux time-course data were corrected using this scaling factor. For a given mutant, the raw efflux value at a specific time point was defined as Eraw, and the corresponding value from the negative control (cells transfected with the parent mCherry plasmid lacking CNNM4) at the same time point was defined as E0. The normalized efflux value was calculated as:

Enorm=E0+Eraw-E0M
This normalization rescales the CNNM4-dependent component of Mg2+ efflux assuming a linear relationship between surface expression level and transport activity. The normalized value replaced Eraw at each time point to generate the corrected time-course curve.Qualitative assessment of CNNM4 membrane localization by confocal microscopy.HEK293T cells expressing wild-type or mutant human CNNM4 tagged with a C-temrinal mCherry were fixed in PBS containing 3.7% formaldehyde and imaged using a Nikon SoRa spinning disk confocal microscope equipped with a 60x objective. Confocal imaging was performed to qualitatively confirm protein expression and plasma membrane localization, rather than for quantitative analysis. Alexa Fluor 488–conjugated phalloidin (Invitrogen™ A12379) was used to label F-actin.

METHOD DETAILS

CNNM4 expression and purificationGene encoding full-length human CNNM4 (UniProtKB accession numbers Q6P4Q7) was synthesized by Genescript and was subcloned into a pEG-BacMam vector with a His tag, twin-strep tag and alpha-tag at C-terminus61,62. Site-directed mutagenesis was performed by using Q5 Site-Directed Mutagenesis (NEB) or QuikChange II Site-Directed Mutagenesis (Qiagen) kit and the provided protocols, and confirmed by whole plasmid sequencing and Sanger sequencing (Plasmidsaurus/Eurofins). For baculovirus production, CNNM4 in the BacMam vector was transformed into DH10Bac cells, followed by P1 and P2 baculovirus generation in Sf9 cells. P2 viruses (8%) were used to infect HEK293T cells grown in Freestyle 293 Expression medium in suspension culture (Thermo Fisher Scientific). Infected cells were incubated for an initial 8–10 h at 37°C before addition of 10 mM sodium butyrate. Cells were then moved to a 30°C incubator and allowed to grow for another 48 h with vigorous shaking. At 58 h post-infection, cells were collected by centrifugation at 5,000 r.p.m., at 4°C for 30 min. Cell pellets were washed with buffer containing 150 mM NaCl and 20 mM Tris pH 8.0 (TBS buffer) and stored at −80°C. Cell pellets from a 800-ml culture were thawed on ice and resuspended in TBS buffer containing 1 mM PMSF, 0.8 μM aprotinin, 2 μg/mL leupeptin, 2 mM pepstatin A (all protease inhibitors) plus 2% GDN detergent (Anatrace). Protein was extracted from the membrane by whole-cell solubilization for 1h at 4°C with rotation. The solubilized protein was loaded with 5–10 ml of Strep-Tactin resin (IBA) and washed with TBS buffer supplemented with 0.02% GDN. Protein was eluted with TBS buffer supplemented with 0.02% GDN and 10mM Desthiobiotin. The eluent was concentrated to 500 μl and further purified by size-exclusion chromatography in TBS buffer containing 0.02% GDN. Peak fractions containing CNNM4 were pooled and concentrated to 6–10 mg/ml for grids freezing.

CNNM4 expression and purification

Gene encoding full-length human CNNM4 (UniProtKB accession numbers Q6P4Q7) was synthesized by Genescript and was subcloned into a pEG-BacMam vector with a His tag, twin-strep tag and alpha-tag at C-terminus61,62. Site-directed mutagenesis was performed by using Q5 Site-Directed Mutagenesis (NEB) or QuikChange II Site-Directed Mutagenesis (Qiagen) kit and the provided protocols, and confirmed by whole plasmid sequencing and Sanger sequencing (Plasmidsaurus/Eurofins). For baculovirus production, CNNM4 in the BacMam vector was transformed into DH10Bac cells, followed by P1 and P2 baculovirus generation in Sf9 cells. P2 viruses (8%) were used to infect HEK293T cells grown in Freestyle 293 Expression medium in suspension culture (Thermo Fisher Scientific). Infected cells were incubated for an initial 8–10 h at 37°C before addition of 10 mM sodium butyrate. Cells were then moved to a 30°C incubator and allowed to grow for another 48 h with vigorous shaking. At 58 h post-infection, cells were collected by centrifugation at 5,000 r.p.m., at 4°C for 30 min. Cell pellets were washed with buffer containing 150 mM NaCl and 20 mM Tris pH 8.0 (TBS buffer) and stored at −80°C. Cell pellets from a 800-ml culture were thawed on ice and resuspended in TBS buffer containing 1 mM PMSF, 0.8 μM aprotinin, 2 μg/mL leupeptin, 2 mM pepstatin A (all protease inhibitors) plus 2% GDN detergent (Anatrace). Protein was extracted from the membrane by whole-cell solubilization for 1h at 4°C with rotation. The solubilized protein was loaded with 5–10 ml of Strep-Tactin resin (IBA) and washed with TBS buffer supplemented with 0.02% GDN. Protein was eluted with TBS buffer supplemented with 0.02% GDN and 10mM Desthiobiotin. The eluent was concentrated to 500 μl and further purified by size-exclusion chromatography in TBS buffer containing 0.02% GDN. Peak fractions containing CNNM4 were pooled and concentrated to 6–10 mg/ml for grids freezing.

Electron microscopy sample preparation and data acquisitionFreshly purified CNNM4 protein in GDN detergent was mixed with 5 mM MgCl2 and 2 mM ATP before grid preparation. A 2.5-μl aliquot of the sample was applied to a glow-discharged Quantifoil holey carbon grid (gold, 2/1-μm size/hole space, 300-mesh), blotted for 1.5 s at 100% humidity using Vitrobot, and then plunge-frozen in liquid ethane cooled by liquid nitrogen. The grids were loaded into an FEI Titan Krios transmission electron microscope operating at 300 kV with a nominal magnification of 105,000X and an energy filter (20-eV slit width) with a binned pixel size of 0.828 Å, or FEI Glacios transmission electron microscope operating at 200kV with a nominal magnification of 130,000X and an energy filter (10-eV slit width) with a pixel size of 0.91Å. Each Krios video was dose-fractionated to 50 frames with a total dose of 58 e−/Å2. Image acquisition was facilitated using SerialEM, and the nominal defocus range was set from −0.1 μm to −1.6 μm. Each Glacios video was captured with a total dose of 40 e−/Å2 by employing EPU, with the nominal defocus range from −0.6 μm to −1.3 μm.

Electron microscopy sample preparation and data acquisition

Freshly purified CNNM4 protein in GDN detergent was mixed with 5 mM MgCl2 and 2 mM ATP before grid preparation. A 2.5-μl aliquot of the sample was applied to a glow-discharged Quantifoil holey carbon grid (gold, 2/1-μm size/hole space, 300-mesh), blotted for 1.5 s at 100% humidity using Vitrobot, and then plunge-frozen in liquid ethane cooled by liquid nitrogen. The grids were loaded into an FEI Titan Krios transmission electron microscope operating at 300 kV with a nominal magnification of 105,000X and an energy filter (20-eV slit width) with a binned pixel size of 0.828 Å, or FEI Glacios transmission electron microscope operating at 200kV with a nominal magnification of 130,000X and an energy filter (10-eV slit width) with a pixel size of 0.91Å. Each Krios video was dose-fractionated to 50 frames with a total dose of 58 e−/Å2. Image acquisition was facilitated using SerialEM, and the nominal defocus range was set from −0.1 μm to −1.6 μm. Each Glacios video was captured with a total dose of 40 e−/Å2 by employing EPU, with the nominal defocus range from −0.6 μm to −1.3 μm.

Cryo-electron microscopy data analysis procedureThe detailed workflow for the data processing procedure is summarized in Figures S1 and S4. In general, the raw super-resolution tif or eer video files for each dataset were motion-corrected using MotionCor263 or RELION5.064. The per-micrograph defocus values were estimated using ctffind465. Particle picking was performed using topaz66 and/or reference-based template match. Junk particles were removed by two-dimensional (2D) classification and heterogeneous refinement using CryoSPARC67. Selected good particles were then used to generate an initial 3D model by ab initio reconstruction followed by homogeneous refinement with C2 symmetry. Multiple rounds of CTF refinement and Bayesian polishing were performed in RELION to further improve the map resolution. To further improve the map quality, we performed focused classification by subtracting ICD signals from the particles. For all datasets, the gold standard Fourier shell correlation (FSC) 0.143 criteria was used to provide the map resolution estimation. The cryo-EM maps were visualized using UCSF ChimeraX/Chimera68,69. PDB structures are visualized using UCSF ChimeraX/Chimera or PyMOL70.

Cryo-electron microscopy data analysis procedure

The detailed workflow for the data processing procedure is summarized in Figures S1 and S4. In general, the raw super-resolution tif or eer video files for each dataset were motion-corrected using MotionCor263 or RELION5.064. The per-micrograph defocus values were estimated using ctffind465. Particle picking was performed using topaz66 and/or reference-based template match. Junk particles were removed by two-dimensional (2D) classification and heterogeneous refinement using CryoSPARC67. Selected good particles were then used to generate an initial 3D model by ab initio reconstruction followed by homogeneous refinement with C2 symmetry. Multiple rounds of CTF refinement and Bayesian polishing were performed in RELION to further improve the map resolution. To further improve the map quality, we performed focused classification by subtracting ICD signals from the particles. For all datasets, the gold standard Fourier shell correlation (FSC) 0.143 criteria was used to provide the map resolution estimation. The cryo-EM maps were visualized using UCSF ChimeraX/Chimera68,69. PDB structures are visualized using UCSF ChimeraX/Chimera or PyMOL70.

Model building and structure refinementFor the WT CNNM4/MgATP model, the original monomer model of human CNNM4 (UniProt:Q6P4Q7) from AlphaFold Protein Structure Database71 was docked into the cryo-EM map using UCSF Chimera69. The D262C CNNM4/MgATP, E284A CNNM4/MgATP and K113A/R140A/R141A CNNM4/MgATP models were built by docking the WT CNNM4/MgATP model into the respective density maps. All the models were manually rebuilt and adjusted to fit into the density map in Coot (Coot v0.89)72 and subjected to real-space refinement in Phenix73 against the density maps. The model statistics were calculated using MolProbity in Phenix. Structural figures were prepared in ChimeraX68 and PyMOL70. The final data and refinement statistics are provided in Table S1.

Model building and structure refinement

For the WT CNNM4/MgATP model, the original monomer model of human CNNM4 (UniProt:Q6P4Q7) from AlphaFold Protein Structure Database71 was docked into the cryo-EM map using UCSF Chimera69. The D262C CNNM4/MgATP, E284A CNNM4/MgATP and K113A/R140A/R141A CNNM4/MgATP models were built by docking the WT CNNM4/MgATP model into the respective density maps. All the models were manually rebuilt and adjusted to fit into the density map in Coot (Coot v0.89)72 and subjected to real-space refinement in Phenix73 against the density maps. The model statistics were calculated using MolProbity in Phenix. Structural figures were prepared in ChimeraX68 and PyMOL70. The final data and refinement statistics are provided in Table S1.

Mg2+ efflux assay and statistical analysesMg2+ efflux activity was judged by magnesium efflux assay following published methods22,44. HEK293T cells were plated in Poly-L-Lysine (Sigma-Aldrich P4707) coated coverslips (Fisher Scientific™ Cat.No.1251000) in 24-well plates, human CNNM4-mcherry (wildtype or mutants) or control constructs were transformed by lipofectamine 3000 (Invitrogen™ L3000008) and cultured with growth medium supplemented with 40 mM Mg2+. 24 hours posttransfection, cells were used for imaging. Live-cell Mg2+ fluorescence imaging was performed using a Prime BSI Express sCMOS camera (Teledyne Photometrics) controlled by the open-source Micro-Manager platform (v2.0.3, https://micro-manager.org/). Camera exposure and LED illumination were synchronized using an Arduino-based TTL pulse generator operated through a custom Python GUI (source code available at: https://github.com/junuk861/arduino-ttl-pulse-gui). Illumination was provided by the 470-nm LED channel of the CoolLED pE-800 Fura system (LightBridge v1.3.0.0) through a 470/40x excitation filter (450–490 nm). Emitted fluorescence was transmitted through a 495LP dichroic mirror and collected with a 525/50m emission filter (500–550 nm). Cells were incubated with Mg2+-loading buffer (78.1 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl2, 40 mM MgCl2, 5.5 mM glucose, 5.5 mM HEPES-KOH, pH 7.4) containing 2 μM Magnesium Green-AM (Invitrogen M3735) for 45 min at 37 °C. Cells were imaged every 20 sec. 1 mins after start imaging, the extracellular solution was then replaced with Mg2+-free buffer (MgCl2 substituted with 60 mM NaCl). Data were quantified with ImageJ, and the fluorescence intensity of each cell at each time point was normalized to the corresponding value at 0 s. Statistical analyses were performed using GraphPad Prism (version 10.6.1). Data are shown as means ± SD, and statistical significance was determined using One-Way ANOVA test.

Mg2+ efflux assay and statistical analyses

Mg2+ efflux activity was judged by magnesium efflux assay following published methods22,44. HEK293T cells were plated in Poly-L-Lysine (Sigma-Aldrich P4707) coated coverslips (Fisher Scientific™ Cat.No.1251000) in 24-well plates, human CNNM4-mcherry (wildtype or mutants) or control constructs were transformed by lipofectamine 3000 (Invitrogen™ L3000008) and cultured with growth medium supplemented with 40 mM Mg2+. 24 hours posttransfection, cells were used for imaging. Live-cell Mg2+ fluorescence imaging was performed using a Prime BSI Express sCMOS camera (Teledyne Photometrics) controlled by the open-source Micro-Manager platform (v2.0.3, https://micro-manager.org/). Camera exposure and LED illumination were synchronized using an Arduino-based TTL pulse generator operated through a custom Python GUI (source code available at: https://github.com/junuk861/arduino-ttl-pulse-gui). Illumination was provided by the 470-nm LED channel of the CoolLED pE-800 Fura system (LightBridge v1.3.0.0) through a 470/40x excitation filter (450–490 nm). Emitted fluorescence was transmitted through a 495LP dichroic mirror and collected with a 525/50m emission filter (500–550 nm). Cells were incubated with Mg2+-loading buffer (78.1 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl2, 40 mM MgCl2, 5.5 mM glucose, 5.5 mM HEPES-KOH, pH 7.4) containing 2 μM Magnesium Green-AM (Invitrogen M3735) for 45 min at 37 °C. Cells were imaged every 20 sec. 1 mins after start imaging, the extracellular solution was then replaced with Mg2+-free buffer (MgCl2 substituted with 60 mM NaCl). Data were quantified with ImageJ, and the fluorescence intensity of each cell at each time point was normalized to the corresponding value at 0 s. Statistical analyses were performed using GraphPad Prism (version 10.6.1). Data are shown as means ± SD, and statistical significance was determined using One-Way ANOVA test.

Surface protein labeling and normalization of Mg2+ efflux activity.Surface protein labeling experiments were performed using the Pierce™ Cell Surface Protein Isolation Kit (Thermo Fisher Scientific). Surface and total protein levels of wild-type and mutant human CNNM4 were assessed by in-gel fluorescence of the C-terminal mCherry tag. Actin was used as a loading control and was detected by western blotting with an HRP-conjugated anti-beta actin antibody (Proteintech, HRP-60008).Normalization of Mg2+ efflux activity was performed in two steps. First, surface expression levels of each CNNM4 mutant were normalized to actin and then divided by the corresponding actin-normalized value of wild-type CNNM4. Ratios obtained from three independent experiments were averaged to yield a mean scaling factor for each mutant (M). Second, Mg2+ efflux time-course data were corrected using this scaling factor. For a given mutant, the raw efflux value at a specific time point was defined as Eraw, and the corresponding value from the negative control (cells transfected with the parent mCherry plasmid lacking CNNM4) at the same time point was defined as E0. The normalized efflux value was calculated as:

Enorm=E0+Eraw-E0M
This normalization rescales the CNNM4-dependent component of Mg2+ efflux assuming a linear relationship between surface expression level and transport activity. The normalized value replaced Eraw at each time point to generate the corrected time-course curve.

Surface protein labeling and normalization of Mg2+ efflux activity.

Surface protein labeling experiments were performed using the Pierce™ Cell Surface Protein Isolation Kit (Thermo Fisher Scientific). Surface and total protein levels of wild-type and mutant human CNNM4 were assessed by in-gel fluorescence of the C-terminal mCherry tag. Actin was used as a loading control and was detected by western blotting with an HRP-conjugated anti-beta actin antibody (Proteintech, HRP-60008).

Normalization of Mg2+ efflux activity was performed in two steps. First, surface expression levels of each CNNM4 mutant were normalized to actin and then divided by the corresponding actin-normalized value of wild-type CNNM4. Ratios obtained from three independent experiments were averaged to yield a mean scaling factor for each mutant (M). Second, Mg2+ efflux time-course data were corrected using this scaling factor. For a given mutant, the raw efflux value at a specific time point was defined as Eraw, and the corresponding value from the negative control (cells transfected with the parent mCherry plasmid lacking CNNM4) at the same time point was defined as E0. The normalized efflux value was calculated as:

Enorm=E0+Eraw-E0M

This normalization rescales the CNNM4-dependent component of Mg2+ efflux assuming a linear relationship between surface expression level and transport activity. The normalized value replaced Eraw at each time point to generate the corrected time-course curve.

Qualitative assessment of CNNM4 membrane localization by confocal microscopy.HEK293T cells expressing wild-type or mutant human CNNM4 tagged with a C-temrinal mCherry were fixed in PBS containing 3.7% formaldehyde and imaged using a Nikon SoRa spinning disk confocal microscope equipped with a 60x objective. Confocal imaging was performed to qualitatively confirm protein expression and plasma membrane localization, rather than for quantitative analysis. Alexa Fluor 488–conjugated phalloidin (Invitrogen™ A12379) was used to label F-actin.

Qualitative assessment of CNNM4 membrane localization by confocal microscopy.

HEK293T cells expressing wild-type or mutant human CNNM4 tagged with a C-temrinal mCherry were fixed in PBS containing 3.7% formaldehyde and imaged using a Nikon SoRa spinning disk confocal microscope equipped with a 60x objective. Confocal imaging was performed to qualitatively confirm protein expression and plasma membrane localization, rather than for quantitative analysis. Alexa Fluor 488–conjugated phalloidin (Invitrogen™ A12379) was used to label F-actin.

QUANTIFICATION AND STATISTICAL ANALYSISStatistical analyses were performed using GraphPad Prism v10.6.1 (GraphPad Software, LLC). Relative quantification methods and tools are described in each section of the methods and/or figure legends.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses were performed using GraphPad Prism v10.6.1 (GraphPad Software, LLC). Relative quantification methods and tools are described in each section of the methods and/or figure legends.

Supplementary Material123TableS2. Source Data for Figure 2, related to Figure 2.4TableS3. Source Data for Figure 4, related to Figure 4.5TableS4. Source Data for Figure 5, related to Figure 5.6TableS5. Source Data for Figure S4, related to Figures 2, 4, 5.7TableS6. Source Data for Figure S5, related to Figures 2, 4, 5.

Supplementary Material

TableS2. Source Data for Figure 2, related to Figure 2.

TableS3. Source Data for Figure 4, related to Figure 4.

TableS4. Source Data for Figure 5, related to Figure 5.

TableS5. Source Data for Figure S4, related to Figures 2, 4, 5.

TableS6. Source Data for Figure S5, related to Figures 2, 4, 5.