The molecular appearance of native TRPM7 channel complexes identified by high-resolution proteomics

The transient receptor potential melastatin-subfamily member 7 (TRPM7) is a ubiquitously expressed membrane protein consisting of ion channel and protein kinase domains. TRPM7 plays a fundamental role in the cellular uptake of divalent cations such as Zn2+, Mg2+, and Ca2+, and thus shapes cellular excitability, plasticity, and metabolic activity. The molecular appearance and operation of TRPM7 channels in native tissues have remained unresolved. Here, we investigated the subunit composition of endogenous TRPM7 channels in rodent brain by multi-epitope affinity purification and high-resolution quantitative mass spectrometry (MS) analysis. We found that native TRPM7 channels are high-molecular-weight multi-protein complexes that contain the putative metal transporter proteins CNNM1-4 and a small G-protein ADP-ribosylation factor-like protein 15 (ARL15). Heterologous reconstitution experiments confirmed the formation of TRPM7/CNNM/ARL15 ternary complexes and indicated that complex formation effectively and specifically impacts TRPM7 activity. These results open up new avenues towards a mechanistic understanding of the cellular regulation and function of TRPM7 channels.

The C-terminal α-kinase domain of TRPM7 acts in two ways: First, it autophosphorylates cytoplasmic residues of TRPM7, and second, it may target a variety of proteins with diverse cellular functions such as annexin A1, myosin II, eEF2-k, PLCγ2, STIM2, SMAD2, and RhoA (Runnels et al., 2001;Dorovkov and Ryazanov, 2004;Perraud et al., 2011;Clark et al., 2008;Romagnani et al., 2017;Voringer et al., 2020;Faouzi et al., 2017). In immune cells, the TRPM7 kinase domain has been reported to be clipped from the channel domain by caspases in response to Fas-receptor stimulation (Desai et al., 2012). In line with this observation, cleaved TRPM7 kinase was detected in several cell lines and shown to translocate to the nucleus, where it promotes histone phosphorylation (Krapivinsky et al., 2014).
The majority of the current knowledge about TRPM7 was derived from in vitro experiments with cultured cells, whereas insights into the operation of both channel and α-kinase activity of TRPM7 in native tissues are limited. We, therefore, investigated the molecular architecture of TRPM7 in rodent brain by using blue native polyacrylamide gel electrophoresis (BN-PAGE) and multi-epitope affinity purifications (ME-APs) in combination with high-resolution quantitative mass spectrometry (MS). These approaches showed that native TRPM7 channels are macromolecular complexes with an apparent size of ≧1.2 MDa and identified proteins CNNM1-4 and ADP-ribosylation factor-like protein 15 (ARL15) as complex constituents. Subsequent functional studies in Xenopus laevis oocytes and HEK293 cells suggested ARL15 and CNNM3 as hitherto unrecognised regulators of the TRPM7 ion channel and kinase activity, respectively.
Results ME-AP proteomic analyses of native TRPM7 channels TRPM7 channels assemble from four subunits (Fleig and Chubanov, 2014), each of which is about 1860 aa in length and comprises several distinct domains in its extended intracellular N-and C-termini in addition to a transmembrane channel domain ( Figure 1A). Unexpectedly, analysis by native gel electrophoresis (BN-PAGE) of TRPM7 channels either endogenous to HEK293 cells or exogenously expressed in these cells via transient transfection, elicited a molecular mass of at least 1.2 MDa considerably exceeding the molecular mass of ~850 kDa calculated for TRPM7 tetramers (Figure 1B,upper panel). To see whether this large molecular size is a peculiarity of HEK293 cells, we recapitulated the analysis for TRPM7 channels expressed in mouse brain using a recently developed technique that combines BN-PAGE with cryo-slicing and quantitative mass spectrometry (csBN-MS, Müller et al., 2019). In this approach, membrane fractions prepared from the entire mouse brain and solubilised with the mild detergent buffer CL-47 (Schwenk et al., 2016;Schwenk et al., 2012;Müller et al., 2010) are first separated on a native gel, which is subsequently embedded and cut into 300 µm gel slices using a cryo-microtome. In a second step, the protein content of each slice is analysed individually by nanoflow liquid chromatography tandem mass spectrometry (nanoLC-MS/MS), providing Kollewe, Chubanov, et  information on both the identity and amount of the proteins in each slice; noteworthy, protein amounts are determined with a dynamic range of up to four orders of magnitude (Müller et al., 2010;Schwenk et al., 2010;Bildl et al., 2012). As illustrated in Figure 1B, lower panel, csBN-MS analysis of mouse brain membranes detected the TRPM7 protein with an apparent molecular mass between 1.2 and 2.6 MDa, comparable to the results obtained from HEK293 cells ( Figure 1B, upper panel). Moreover, the Abundance-mass profile of TRPM7 obtained by cryo-slicing blue native mass spectrometry (csBN-MS) in a CL-47 solubilised membrane fraction from adult mouse brain (a total of 192 gel slices). Inset: Abundance of the indicated proteins in the mouse brain. Note the large apparent molecular mass of the native TRPM7 channel in both culture cells and mouse brain, markedly exceeding the mass calculated for tetrameric channel assemblies (about 850 kDa, red circles). (C) Table summarising the results of all anti-TRPM7 APs performed with the indicated antibodies on membrane fractions prepared from rodent brain and cultured HEK293 cells. Solubilisation conditions and specificity of purification of the listed proteins determined by comparison with stringent negative controls are colour-coded as given in the upper left; MW is indicated on the right. TUC refers to series of APs with target-unrelated control antibodies. Note that TRPM7 channels co-assemble with all CNNM family members and ADP-ribosylation factor-like protein 15 (ARL15) in the brain and HEK293 cells.
The online version of this article includes the following figure supplement(s) for figure 1: determination of the total protein amount by signal integration over all slices showed that TRPM7 levels in the brain are rather low compared to other members of the TRP family of proteins. Thus, the abundance of TRPM7 is about one to three orders of magnitude below that obtained for TRPC4, TRPM3, or TRPV2 ( Figure 1B, lower right). Together, these results indicated that native TRPM7 complexes exceed the predicted molecular size of bare tetrameric assemblies in different cellular environments suggesting that the rather simplistic view on the molecular make-up of native TRPM7 channel complexes has to be revised.
To identify proteins that may co-assemble with TRPM7, we used affinity purifications with multiple antibodies targeting distinct epitopes of the TRPM7 protein ( Figure 1A, Figure 1-figure supplement 1) and evaluated the respective eluates of HEK293 cells and rodent brains by high-resolution quantitative MS analysis (ME-APs, Schwenk et al., 2016;Schwenk et al., 2012;Müller et al., 2010;Schwenk et al., 2010). HEK293 cells were selected because these cells are widely used for the functional assessment of endogenous and overexpressed TRPM7. The brain was chosen since TRPM7 plays a critical role in neurological injuries and synaptic and cognitive functions (Aarts et al., 2003;Sun et al., 2009;Liu et al., 2018). For these ME-APs, membrane fractions prepared either from whole brains of adult mice and rats or from WT HEK293 cells were solubilised with detergent buffers of mild (CL-47) or intermediate (CL-91) stringency (Schwenk et al., 2012;Müller et al., 2010;Schwenk et al., 2010) prior to TRPM7 purification. TRPM7 was also affinity-isolated from HEK293 cells transiently (over)-expressing C-terminally HA-tagged TRPM7 using an anti-HA antibody.
In all APs, TRPM7 could be reliably detected under both solubilisation conditions ( Figure 1C) with MS-identified peptides covering a large percentage of the primary sequence of TRPM7 in samples from mouse brain as well as from HEK293 cells (77% and 98%, respectively).
All other proteins identified in the ME-APs were evaluated for specificity and consistency of their co-purification with TRPM7 based on protein amounts determined by label-free quantification (see Materials and methods section). The specificity of co-purification was assessed by comparing protein amounts in APs targeting TRPM7 with protein amounts obtained with stringent negative controls. Thus, (i) APs with five different target-unrelated control (TUC) antibodies were used as negative controls for anti-TRPM7 APs from rodent brain, (ii) anti-TRPM7 APs from a TRPM7 -/-HEK293 cell line (Abiria et al., 2017) served as negative controls for anti-TRPM7 APs from WT HEK293 cells, and (iii) HEK293 cells heterologously expressing TRPM7-myc were used as negative Notes: Relative abundance refers to the amount of TRPM7 as a reference and was classified as follows: = when between 0.33-fold and 3.3-fold of reference, < when between 0.033-fold and 0.33-fold of reference, << when between 0.0033-fold and 0.033-fold of reference, and <<< when less than 0.0033-fold of the reference amount. Transmembrane proteins; cytoplasmic proteins. † Co-purified from HEK293 cells with anti-M7a (CL-47) and with anti-M7c (CL-91); ## co-purified with anti-M7c from rat brain membranes (CL-91); ### co-purified with anti-M7a from HEK293 cells (CL-47, CL-91). controls for anti-HA APs from HEK293 cells overexpressing TRPM7-HA. A protein was considered consistently co-purified if detected in APs with at least two antibodies under the same solubilisation condition. Together, these specificity and consistency criteria identified five proteins as high-confidence interaction partners of TRPM7: ARL15 and the cyclin M family proteins CNNM1-4, putative Mg 2+ transporters ( Figure 1C, Table 1). Neither of these proteins was detected in any of the negative controls. Moreover, they were not only consistently co-purified with several antibodies but with the exception of CNNM1 also from both rodent brain and HEK293 cells. Comparison of the degree of association under the two solubilisation conditions revealed that the interaction between TRPM7, ARL15, and CNNMs was weakened by the more stringent detergent CL-91 ( Figure 1C, Table 1).
Next, we verified the identified interactions between TRPM7, ARL15, and CNNM1-4 in co-expression experiments performed in TRPM7 -/-HEK293 cells ( Figure 2). Flag-tagged ARL15 and CNNM proteins could be specifically and robustly co-purified with HA-tagged TRPM7 in anti-HA APs when all three proteins were present, whereas the association was markedly less efficient when ARL15-Flag or CNNM-Flag were co-expressed with TRPM7-HA alone ( Figure 2, Figure 2-figure supplement 1). These results corroborated the ME-AP results from the rodent brain and strongly suggested the formation of ternary complexes containing TRPM7, ARL15, and CNNM proteins. Effects of CNNM3 and ARL15 on TRPM7 channel activity

MW (kDa)
To investigate if the assembly of TRPM7 with ARL15 and CNNM proteins modified TRPM7 function, we studied their effect(s) on TRPM7 currents by co-expression in X. laevis oocytes. This approach allows co-expression of defined protein ratios by cRNA injection and, therefore, is widely used for functional assessment of ion channel complexes, including functional interaction of TRPM7 with TRPM6 (Chubanov et al., 2018;Chubanov et al., 2004). The two-electrode voltage clamp (TEVC) measurement in Figure 3A illustrates a typical current-voltage (I-V) relationship of constitutively active TRPM7 channels characterised by steep outward rectification and very small inward currents over the whole range of negative membrane potentials (Nadler et al., 2001). Co-expression of TRPM7 and CNNM3, the most efficiently co-purified CNNM protein ( Figure 1C), neither changed the shape of the I-V relationship nor current amplitudes. In contrast, ARL15 effectively suppressed constitutive TRPM7 currents in a concentration-dependent manner, as deduced from experiments with increasing amounts of ARL15 ( Figure 3B and C). Oocytes co-expressing all three proteins TRPM7, CNNM3, and ARL15 did not exhibit TRPM7 currents, similar to the co-expression of TRPM7 and ARL15 ( Figure 3A). The suppressive effect was specific for TRPM7, as co-expressed ARL15 did not inhibit another TRP channel, TRPV1, in an analogous experiment ( Figure 3-figure supplement 1). Consistently, co-expression of TRPM7 with another ARL family member, ARL8A (Gillingham and Munro, 2007), did not affect TRPM7 currents (Figure 3-figure supplement 2). Next, we examined if the interference of ARL15 with the TRPM7 function was due to reduced expression levels or altered membrane localisation. Western blot analysis of oocytes injected with Trpm7 or Trpm7 and Arl15 cRNAs did not reveal any change in the expression level of TRPM7 protein ( Figure 3D). Using immunofluorescence staining with the anti-M7d antibody, we detected TRPM7 at the cell surface of oocytes injected with Trpm7 but not in uninjected oocytes ( Figure 3E). Notably, the TRPM7 signal was similarly detectable at the cell surface of oocytes co-expressing TRPM7 and ARL15 ( Figure 3E).
TRPM7 inward currents at negative membrane potentials are small, and, consequently, quantification of the comparably large outward currents is commonly used for functional assessment of the TRPM7 channel activity. Nevertheless, we asked whether TRPM7 inward currents could be equally suppressed by ARL15 (Figure 3-figure supplement 3A, B). This analysis revealed that ARL15 acted similarly on inward and outward TRPM7 currents, suggesting that ARL15 elicited a general block of the TRPM7 channel.
To obtain further insight into the functional interaction of ARL15 with TRPM7, we investigated whether the kinase activity of TRPM7 is necessary for the inhibitory effect of ARL15. To this end, we examined oocytes expressing a kinase-dead TRPM7 mutant (K1646R, Nadler et al., 2001;Runnels et al., 2002) and observed that the K1646R mutation did not change the sensitivity of TRPM7 for the inhibitory effect of ARL15 (Figure 3-figure supplement 3C).
Finally, we investigated whether ARL15 could also regulate TRPM7 channels in mammalian cells. Using the patch-clamp technique, we measured endogenous TRPM7 currents in HEK293 cells. Similar to previous reports Ferioli et al., 2017), removing intracellular Mg 2+ by using a pipette solution free of divalent cations induced endogenous TRPM7 currents (Figure 3figure supplement 4). Transient expression of ARL15 however caused a significant reduction of these TRPM7 currents (Figure 3-figure supplement 4).
Collectively, these results suggest that the inhibitory effect of ARL15 on TRPM7 currents is specific and concentration-dependent.

Impact of CNNM3 on TRPM7 Mg 2+ currents and kinase activity
Given the crucial role of TRPM7 and CNNM proteins in membrane Mg 2+ transport (Mittermeier et al., 2019;Schmitz et al., 2003;Funato and Miki, 2019), we asked whether CNNM3 would specifically affect TRPM7 Mg 2+ currents rather than exerting a general (i.e., ARL15-like) effect. To this end, we conducted TEVC measurements with TRPM7-expressing oocytes using external saline containing 3 mM Mg 2+ (instead of 3 mM Ba 2+ in Figure 3A), implying that at negative membrane potentials, the TRPM7 channel should primarily exhibit Mg 2+ currents under such experimental conditions (Nadler et al., 2001). TRPM7 expressing oocytes displayed characteristic TRPM7 currents with a very small inward Mg 2+ component, which was suppressed by co-expression of ARL15 ( Figure 4A   Next, we studied whether heterologous expression in mammalian cells would allow uncovering any functional effects of CNNM3 on TRPM7. We transiently transfected HEK293 cells with Trpm7 and Cnnm3 plasmid cDNAs (ratio 2:1) and performed patch-clamp measurements (Figure 4-figure supplement 1). TRPM7 currents were induced using the standard divalent cation-free internal solution and an external buffer containing 1 mM CaCl 2 and 2 mM MgCl 2 . When currents were developed, cells were exposed to mannitol-based saline containing 10 mM Mg 2+ . In accord with previous publications , the perfusion of TRPM7-expressing cells with 10 mM Mg 2+ led to a significant reduction of outward currents accompanied by a relatively modest decrease of inward currents ( Previously, we found that TRPM7 controls the uptake of Mg 2+ to maintain the cellular content of this mineral in resting cells (Mittermeier et al., 2019). To investigate whether CNNM3 modulates TRPM7-dependent Mg 2+ uptake, we employed inductively coupled plasma mass spectrometry (ICP-MS) to compare total amounts of magnesium in TRPM7 -/-HEK293 cells transfected with Trpm7, Cnnm3, or Trpm7 plus Cnnm3 cDNAs (Figure 4-figure supplement 2). Next, we normalised the levels of magnesium to cellular sulphur (a biomarker for the total protein content) and observed that transient expression of TRPM7 increased the cellular Mg content, whereas expression of CNNM3 did not change this parameter ( Figure 4-figure supplement 2). Importantly, we found that co-expression of TRPM7 with CNNM3 did not impact the ability of TRPM7 to regulate the cellular content of Mg 2+ (Figure 4-figure supplement 2). Hence, different experimental approaches did not reveal significant effects of CNNM3 on TRPM7 channel activity.
Since TRPM7 contains a C-terminal kinase domain, we studied whether CNNM3 might modulate the TRPM7 kinase moiety ( Figure 5 and Figure 5-figure supplement 1). To assess the activity of the TRPM7 kinase, we relied on the anti-(p)Ser1511 M7 antibody, which specifically recognises the known autophosphorylation site (Ser1511) of mouse TRPM7 (Romagnani et al., 2017). To verify that autophosphorylation of Ser1511 is dynamic, and changes of the TRPM7 kinase activity could therefore be visualised by the anti-(p)Ser1511 M7 antibody we treated HEK293 cells transiently overexpressing TRPM7 with TG100-115, a drug-like TRPM7 kinase inhibitor (Song et al., 2017). We observed that the exposure of living cells to TG100-115 led to suppression of (p)Ser1511 TRPM7 immunoreactivity in a dose-dependent fashion ( Figure 5-figure supplement 1A). Moreover, the inhibitory effect of TG100-115 was time-dependent and could be detected 10 min after application of TG100-115 left. Two independent batches of injected oocytes (n = 8-16) were examined. *p < 0.05; ****p < 0.0001 (ANOVA). (B) Left panel: Representative I-V relationships of TRPM7 currents measured in oocytes expressing TRPM7 or co-expressing TRPM7 with ARL15 at the indicated ratios of injected cRNAs. Right panel: Current amplitudes (mean ± SEM) at +80 mV in measurements shown on the left. Two independent batches of injected oocytes (n = 5-7) were examined. *p < 0.05; ****p < 0.0001 (ANOVA). (C) Western blot analysis of ARL15 expression using the anti-Myc antibody in total lysates of oocytes injected with Trpm7 or Trpm7 and Arl15 cRNAs (ratios 200:1, 20:1, and 10:1). Representative results are shown for two independent experiments. Anti-Na + /K + -ATPase antibody was used for loading controls. (D) Western blot analysis of TRPM7 expression using the anti-M7d antibody in total lysates of oocytes injected with Trpm7 or Trpm7 and Arl15 cRNAs (ratio 10:1). Anti-Na + /K + ATPase antibody was used for loading controls. Representative results are shown for two independent experiments. (E) Immunofluorescence staining of un-injected oocytes (control) or oocytes injected with Trpm7 (TRPM7) or Trpm7 and Arl15 cRNAs (TRPM7+ ARL15, ratio 10:1) using anti-M7d antibody and anti-mouse antibody conjugated with Alexa Fluor 488. Confocal images of Alexa Fluor 488 fluorescence (Alexa488) and overlays of Alexa488 with differential interference contrast images (overlay) are depicted for two independent oocytes per image; scale bars, 50 μm. The diagrams depict fluorescence intensity acquired along the green bars shown in overlay images.
The stars indicate the cell surface of two oocytes. Typical examples of two independent experiments (n = 10 oocytes) are shown.
The online version of this article includes the following figure supplement(s) for figure 3:       Figure 5-figure supplement 1B). Furthermore, we found that wash-out of TG100-115 by fresh cell culture medium caused a fast recovery of the (p)Ser1511 TRPM7 signal ( Figure 5-figure supplement  1C). Hence, detection of (p)Ser1511 TRPM7 levels seems a reliable means to monitor the TRPM7 kinase activity. Accordingly, we investigated whether co-expression of ARL15 could modulate TRPM7 kinase activity and found no changes in (p)Ser1511 TRPM7 immunoreactivity ( Figure 5). Co-expression of CNNM3 however caused a significant reduction of the (p)Ser1511 TRPM7 signal ( Figure 5), suggesting that CNNM3 functions as a negative regulator of the TRPM7 kinase.

Identification of new phosphorylation sites in the TRPM7 protein
In addition to subunit assembly, the MS data provided further insight into the post-translational modification(s) of the TRPM7 protein. Thus, TRPM7 purified either from rodent brain or from transfected HEK293 cells showed very similar patterns of serine and threonine phosphorylation, reflected by matching MS/MS spectra of peptides harbouring phosphorylation sites ( Figure 6A, Figure 6-figure supplement 1, Supplementary file 2 to Figure 6). Out of the nine shared phospho-sites, four have not been reported for TRPM7 in native tissue before (S1300, S1360, T1466, and S1567; Supplementary file 2 to Figure 6). An additional 26 phosphorylated serine and threonine residues could be assigned to TRPM7 isolated from HEK 293 cells, presumably based on the higher amounts of TRPM7 available for analysis from heterologous (over)-expression material; 22 of these 26 sites match sites previously reported for TRPM7 endogenously or heterologously expressed in cell lines, and four sites were newly detected (S1208, S1480, S1496, S1853; Supplementary file 2 to Figure 6). Most of the identified phosphorylation sites were found to cluster within the C-terminal cytoplasmic domain of TRPM7.
Finally, we asked whether measuring TRPM7 channel activity by TEVC would reveal any functional consequences of TRPM7 phosphorylation. We introduced phosphomimetic mutations in a subset of identified phospho-sites (S1208D, S1360D, S1480D, S1496D, and S1567D) and found that three TRPM7 mutants (S1208D, S1496D, and S1567D) displayed enhanced current amplitudes ( Figure 6B and C), whereas their expression levels were similar to WT TRPM7 ( Figure 6D). These findings suggest that phosphorylation of TRPM7 may represent a new regulatory mechanism reminiscent of the situation with TRPM8 (Rivera et al., 2021). To substantiate this notion further, it will be interesting to carry out a systematic functional analysis of the surprisingly extensive phosphorylation profile of TRPM7 ( Figure 6).

Discussion
In the present study, we investigated the molecular appearance and subunit composition of TRPM7 as present in the cell membrane(s) of the rodent brain. We show that TRPM7 forms macromolecular complexes by assembling with CNNM proteins 1-4 and ARL15. Moreover, functional expression in heterologous expression systems showed that ARL15 strongly affects TRPM7 channel function, while CNNM3 appears to act as a negative regulator of TRPM7 kinase activity.
The online version of this article includes the following figure supplement(s) for figure 4:   molecular mass of TRPM7 tetramers (~850 kDa) and suggesting that the TRPM7 channel kinase is predominantly embedded in a large macromolecular complex. Compared to other native TRP channels, such as TRPC4, TRPM3, and TRPV2, the expression level of TRPM7 was found to be up to three orders of magnitude lower, thus classifying TRPM7 as a very low-abundant protein in the rodent brain and indicating that comprehensive determination of the TRPM7 complexome is technically challenging. The unbiased ME-AP approach paired with stringent negative controls nevertheless allowed for the identification of high-confidence interaction partners based on their specific and consistent co-purification with TRPM7. Consequently, five proteins were found to assemble with native TRPM7, including four members of the CNNM gene family encoding putative Mg 2+ transporters CNNM1-4 and a small G-protein ARL15. The fact that we did not detect all the interactors seen in mouse brain also in APs from rat brain is most likely due to the low abundance of endogenous TRPM7 (~50% less TRPM7 compared to APs from mouse brain). The interaction of TRPM7 with ARL15 and CNNM proteins was successfully confirmed in heterologous expression experiments. We also noted that previous proteome-wide interactome screens in cultured cells suggested an association of ARL15 with TRPM7 (Huttlin et al., 2017;Huttlin et al., 2021), in line with our results. To obtain first insight into a possible functional impact of ARL15 and CNNM3, the most prominent interaction partners of TRPM7 in our experimental settings, we measured the channel activity of TRPM7 expressed in Xenopus oocytes and HEK293 cells. We found that co-expression of TRPM7 with CNNM3 did not lead to significant changes in TRPM7 currents applying a broad range of experimental conditions. Consistently, we observed that the ability of TRPM7 to increase cellular Mg levels was not affected by CNNM3. However, CNNM3 appears to act as a negative regulator of the TRPM7 kinase activity, resembling the action of the drug-like kinase inhibitor TG100-115. Collectively, these results suggest that CNNM3 may represent the first known protein acting as a physiological modulator of the TRPM7 kinase activity.
In contrast to CNNM3, co-expression of TRPM7 with ARL15 in oocytes, but not with the closely related small G-protein ARL8A, caused robust suppression of TRPM7 currents regardless of the experimental conditions applied. Of note, transient expression of ARL15 in HEK 293 cells resulted in inhibition of endogenous TRPM7 currents, reinforcing our conclusion that ARL15 acts as a potent and specific negative regulator of the TRPM7 channel.
The CNNM (Cyclin M; CorC) gene family encodes highly conserved metal transporter proteins identified in all branches of living organisms, ranging from prokaryotes to humans (Funato and Miki, 2019;Giménez-Mascarell et al., 2019). There are four family members in mammals, CNNM1-4, widely expressed in the body and abundantly present in the brain (Funato and Miki, 2019;Giménez-Mascarell et al., 2019). The genetic inactivation of Cnnm4 in mice leads to systemic Mg 2+ deficiency (Yamazaki et al., 2013). In humans, point mutations in CNNM2 cause hypomagnesemia (Stuiver et al., 2011), while mutations in CNNM4 are associated with Jalili syndrome (Parry et al., 2009). Functional expression studies proposed that CNNMs operate as Na + /Mg 2+ exchangers responsible for the efflux of cytosolic Mg 2+ from the cell (Funato and Miki, 2019;Giménez-Mascarell et al., 2019). In contrast to this view, other investigators proposed that CNNM proteins indirectly regulate the influx of Mg 2+ into the cell . Recently resolved crystal structures of two prokaryotic CNNM-like proteins revealed that CNNMs form dimers and that each monomer contains three transmembrane helices harbouring Mg 2+ and Na + binding sites consistent with the suggested Na + -coupled Mg 2+ transport function of CNNMs (Huang et al., 2021;Chen et al., 2021). While the majority of CNNM proteins in a cell is not bound to TRPM7, the direct association identified in this study suggests a new concept implying that two transporting mechanisms, TRPM7-mediated influx of divalent cations (Zn 2+ , Mg 2+ , and Ca 2+ ) and CNNM-dependent Na + /Mg 2+ exchange, can be physically coupled under TRPM7, ARL15-Myc, and CNNM3-Myc, respectively. Representative results are shown from three independent experiments. (B) Quantification of (p) Ser1511 TRPM7 levels in Western blot experiments (n = 3) shown in (A). A relative band density for each sample was obtained by dividing the (p)Ser1511 signal (upper panel) by the corresponding anti-M7d value (middle panel). The relative density of Sample 2 (TRPM7) was set as a 1.0 to calculate changes in (p)Ser1511 TRPM7 (mean ± standard error of the mean [SEM]) caused by co-transfection of Arl15 or Cnnm3 as outlined in the bar graph. ns, not significant; *p ≤ 0.05, **p ≤ 0.01 significant to the control (ANOVA).
The online version of this article includes the following figure supplement(s) for figure 5:  ARL15 is a member of the ARF gene family of small G-proteins (Gillingham and Munro, 2007). A common feature of ARFs is their ability to bind and regulate effector proteins in a GTP-dependent manner (Gillingham and Munro, 2007). GDP-and GTP-bound states of ARFs are controlled by GTPase-activating proteins (GAP) in conjunction with GTP exchange factors (GEF) (Gillingham and 0001  0051  0101  0151  0201  0251  0301  0351  0401  0451  0501  0551  0601  0651  0701  0751  0801  0851  0901  0951  1001  1051  1101  1151  1201  1251  1301  1351  1401  1451  1501  1551  1601  1651  1701  1751  1801  1851 phospho-sites (heterologous) phospho-sites (rat brain and heterologous) A B C D Figure 6. Identification of transient receptor potential melastatin-subfamily member 7 (TRPM7) phospho-sites and functional assessment of phosphomimetic TRPM7 mutants. (A) Coverage of the primary sequence of TRPM7 and phosphorylation sites as identified by mass spectrometry (MS) analyses of affinity purifications (APs) from transfected HEK293 cells and rodent brain. Peptides identified by MS are in red; those accessible to but not identified in tandem mass spectrometry (MS/MS) analyses are in black, and peptides not accessible to the MS/MS analyses used are given in grey. Blue boxes indicate phospho-sites identified in the brain and transfected HEK293 cells; those uniquely seen in heterologous expressions are boxed in yellow. Colour coding of hallmark domains is as in Figure 1A; S1-S6 helices of TRPM7 are underlined. (B, C) Two-electrode voltage clamp (TEVC) measurements of phosphomimetic TRPM7 mutants performed and analysed as explained in Figure 3A. ( Munro, 2007). The best-characterised ARFs are involved in membrane trafficking, phospholipid metabolism and remodelling of the cytoskeleton (Gillingham and Munro, 2007). While genomewide association studies have linked ARL15 to systemic Mg 2+ homeostasis and energy metabolism in humans (Corre et al., 2018;Richards et al., 2009), the particular functional role and corresponding GAP, GEF, and effector proteins of ARL15 remain to be established. To this end, the strong effect of ARL15 in suppressing TRPM7 currents observed in our study may suggest that TRPM7 serves as a specific effector protein of ARL15. The significance of this modulatory effect for native TRPM7 in the rodent brain, however, remains to be shown. In some TRPM7-APs from HEK293 cells, we detected TRPM6, a genetically related channel, and two proteins representing the gene family of phosphatase of regenerating liver 1 and 3 (also entitled protein tyrosine phosphatases type 4A1 and 3, TP4A1 and 3) ( Table 1). The Mg 2+ transporter protein TRPM6 has been described to physically and functionally interact with TRPM7 Ferioli et al., 2017;Chubanov et al., 2016). In the present study, TRPM6, even though detected, could not be consistently co-purified with multiple anti-TRPM7 antibodies, likely because TRPM6 is expressed at very low levels in the brain and HEK293 cells. Nevertheless, a previous study reporting that heterologously expressed ARL15 positively modulates TRPM6 (Corre et al., 2018) might suggest an overlap between the TRPM6 and TRPM7 interactomes.
Interestingly, a recent interactome screen based on lentiviral overexpression of tagged proteins in HEK293 and HTC116 cells revealed that TP4A1 and TP4A2 also interact with ARL15 and CNNMs (Huttlin et al., 2017;Huttlin et al., 2021). Furthermore, a hypothesis-driven search for interaction partners of CNNMs has shown that TP4A proteins assemble with CNNMs and that such interactions shape Mg 2+ efflux from cells (Funato et al., 2014;Hardy et al., 2015;Gulerez et al., 2016;Kostantin et al., 2016;Zhang et al., 2017;Giménez-Mascarell et al., 2017). These findings are commensurate with our observation that TP4A1 and TP4A3 could be found in TRPM7 APs at low amounts.
Hence, based on the present analysis of native TRPM7 complexes in conjunction with earlier interactome experiments and functional expression studies, it is tempting to speculate that TRPM7/ ARL15/CNNMs/TP4As form a protein network orchestrating transport of divalent cations across the cell membrane.
Anti-TRPM7 2C7 mouse monoclonal antibody (anti-M7d, Figure 1A, Figure 1-figure supplement 1) was produced by Eurogentec (Belgium) as follows. The nucleotide sequence coding for His 6 -tag followed by a cleavage site sequence for TEV protease and the amino acids 1501-1863 (kinase domain, KD) of mouse TRPM7 protein was synthesised in vitro and cloned into the prokaryotic expression vector pT7. The resulting expression construct pT7-His 6 -Trpm7-KD was verified by sequencing and transformed in Escherichia coli (BL21 DE3 pLysS). Next, the transformed E. coli strain was amplified in LB medium at 25°C; 1 mM IPTG was used for induction of the His 6 -TRPM7-KD protein expression. The harvested cell pellet was disrupted by sonication. His 6 -TRPM7-KD was identified in the soluble fraction of the lysate. His 6 -TRPM7 was purified on an Ni Sepharose 6 Fast Flow column on an AKTA Avant 25 (GE Healthcare) using an imidazole gradient of 20-500 mM. The fraction containing His 6 -TRPM7-KD was dialysed against a Tris buffer (0.5 mM EDTA, 1 mM DTT, and 50 mM Tris HCl pH 7.5). His 6 -TRPM7-KD was subjected to TEV protease (New England Biolabs) digestion according to the manufacturer's instructions. Subsequently, non-digested His 6 -TRPM7-KD and His 6 -tagged fragments were removed using an Ni-Sepharose 6 Fast Flow column. The flow-through containing the cleaved TRPM7-KD was concentrated to 0.5 mg/ml in the Tris buffer and stored at -80°C. SDS-PAGE was used to verify the removal of the His 6 -tag.
The standard mouse monoclonal antibody production program of Eurogentec (Belgium) was conducted to immunise four mice using the TRPM7-KD protein and to produce a library of hybridomas. ELISA and Western blot were used to screen the hybridomas and to perform a clonal selection. Two hybridoma clones, 2C7 and 4F9 (isotypes G1;K), were selected based on the antibody quality released in the culture medium. Both clones were propagated, and the corresponding cell culture media were collected for large-scale purification of the IgG fraction using Protein G affinity chromatography. The IgG fractions from 2C7 (0.8 mg/ml) and 4F9 (1.4 mg/ml) were dialysed in PBS and stored at -80°C. The specificity of the 2C7 and 4F9 IgGs (dilution 1:1000) was verified by Western blot analysis of HEK293T cells overexpressing the TRPM6 and TRPM7 proteins (Figure 1-figure supplement 1). The 2C7 antibody detected the mouse or human TRPM7, but not the mouse or human TRPM6 (Figure 1-figure  supplement 1). In contrast, the 4F9 antibody detected only the mouse TRPM7 (Figure 1-figure  supplement 1). Consequently, the 2C7 antibody (anti-M7d) was used in the present study.
Cultured cells were harvested in phosphate buffer saline with protease inhibitors, collected by centrifugation (10 min, 500× g) and resuspended in homogenisation buffer. After sonication (2 × 5 pulses, duty 50, output 2 [Branson Sonifier 250]), membranes were pelleted for 20 min at 125,000× g and resuspended in 20 mM Tris/HCl pH 7.4. Protein concentration was determined with the Bio-Rad Protein Assay kit according to the manufacturer's instructions.
Immunoprecipitation: Membranes were resuspended in ComplexioLyte CL-47 or CL-91 solubilisation buffer (Logopharm) with added 1 mM EDTA/EGTA and protease inhibitors at a protein to detergent ratio of 1:8 and incubated for 30 min on ice. Solubilised protein was cleared by centrifugation (10 min, 125,000× g, 4°C) and incubated with antibodies cross-linked to Dynabeads (Invitrogen) by overhead rotation for 2 hr on ice. After two short washing steps with ComplexioLyte CL-47 dilution buffer (Logopharm), the captured protein was eluted in Laemmli buffer with dithiothreitol added after elution. Eluted proteins were separated by SDS-PAGE. For MS/MS analysis silver-stained (Heukeshoven and Dernick, 1988) protein lanes were cut-out, split at 50 kDa and pieces individually subjected to standard in-gel tryptic digestion (Pandey and Mann, 2000). For chemiluminescence detection, proteins were Western blotted onto PVDF membranes and probed with the following antibodies: anti-HA (11867423001, Roche), anti-Flag (F3165, Sigma), anti-βActin (bs-0061R, Bioss Inc).

Complexome profiling
The size distribution of solubilised native TRPM7-associated complexes was investigated using the high-resolution csBN-MS technique detailed in Faouzi et al., 2017. Briefly, membranes isolated from adult mouse brain were solubilised with ComplexioLyte CL-47 (salt replaced by 750 mM aminocaproic acid), concentrated by ultracentrifugation into a 20%/50% sucrose cushion, supplied with 0.125% Coomassie G250 Blue and run overnight on a hyperbolic 1-13% polyacrylamide gel. The region of interest was excised from the lane, proteins fixed in 30% ethanol/15% acetic acid and the gel piece embedded in tissue embedding media (Leica). After careful mounting on a cryo-holder, 0.3 mm slices were harvested, rinsed, and subjected to in-gel tryptic digestion as described (Faouzi et al., 2017).
Label-free quantification of proteins was carried out as described in Bildl et al., 2012;Müller et al., 2016. Peptide signal intensities (peak volumes, PVs) from FT full scans were determined, and offline mass calibrated using MaxQuant v1.6.3 (http://www. maxquant. org). Then, peptide PV elution times were pairwise aligned using LOESS regression (reference times dynamically calculated from the median peptide elution times overall aligned datasets). Finally, PVs were assigned to peptides based on their m/z and elution time (±1 min/2-3 ppm, as obtained directly or indirectly from MS/MS-based identification) using in-house developed software. PV tables were then used to calculate protein abundance ratios in AP versus control ( Figure 1C), the abundance norm value ( Figure 1B, lower right) as an estimate for molecular abundance (both described in Schwenk et al., 2010), and csBN-MS abundance profiles ( Figure 1B, lower left) as detailed in Müller et al., 2016. The latter were smoothed by sliding, averaging over a window of 5. Slice numbers were converted to apparent complex molecular weights by the sigmoidal fitting of (log(MW)) versus slice number of the observed profile peak maximum of mitochondrial marker protein complexes (Schägger and Pfeiffer, 2000).
Heterologous expression of TRPM7, CNNM3, ARL15, and ARL8A in X. laevis oocytes TEVC measurements: X. laevis females were obtained from NASCO (Fort Atkinson, WI) and kept at the Core Facility Animal Models (CAM) of the Biomedical Center (BMC) of LMU Munich, Germany (Az:4.3.2-5682/LMU/BMC/CAM) in accordance with the EU Animal Welfare Act. To obtain oocytes, frogs were deeply anaesthetised in MS222 and killed by decapitation. Surgically extracted ovary lobes were dissociated by 2.5 hr incubation (RT) with gentle shaking in ND96 solution (96 mM NaCl, 2 mM KCl, 1 mM CaCl 2 , 1 mM MgCl 2 , 5 mM HEPES, pH 7.4) containing 2 mg/ml collagenase (Nordmark) and subsequently defolliculated by washing (15 min) with Ca 2+ -free ND96. Stage V-VI oocytes were then selected and kept in ND96 containing 5 µg/ml gentamicin until further use.
The injected oocytes were kept in ND96 solution, supplemented with 5 μg/ml gentamicin at 16°C. TEVC measurements were performed 3 days after injection at room temperature (RT) in Ca 2+ /Mg 2+free ND96 containing 3.0 mM BaCl 2 instead of CaCl 2 and MgCl 2 using a TURBO TEC-05X amplifier (npi 7.2. Data are presented as means ± standard error of the mean (means ± SEM). Statistical comparisons (Prism 8.4.0) were made using one-way ANOVA or a two-tailed t-test, as indicated in the figure legends. Significance was accepted at p ≤ 0.05.

Determination of cellular Mg contents
The total content of Mg in TRPM7 -/-HEK293T cells (Abiria et al., 2017) was determined by ICP-MS in ALS Scandinavia (Sweden) as reported previously (Mittermeier et al., 2019) with several modifications. The cells were cultured in DMEM (Merck) supplemented with 10% FBS, 100 µg/ml streptomycin, 100 U/ml penicillin, and 10 mM MgCl 2 (all from Thermo Fisher Scientific) in a humidified cell culture incubator (Heraeus, Thermo Fisher Scientific) at 37°C and 5% CO 2 . To conduct ICP-MS experiments, TRPM7 -/-HEK293T cells were plated in 10 cm 2 dishes at ~50% confluence in standard DMEM (without additional 10 mM Mg 2+ ) and transiently transfected with 20 µg Trpm7, 10 µg Cnnm3, or 20 µg Trpm7 plus 10 µg Cnnm3 plasmid cDNAs using Lipofectamine 2000 reagent (Thermo Fisher Scientific). After 24 hr, the cells were washed with serum-free DMEM, mechanically detached, and cell suspensions collected in 10 ml plastic tubes. After centrifugation (3 min, 1000 rpm), the medium was removed, and the cell pellet was resuspended in 5 ml PBS and passed to a fresh 10 ml tube. The cell suspension was centrifuged (3 min, 3500 rpm), the supernatant removed, and the cell pellet frozen at -20°C. Cell pellets were analysed by ICP-MS in ALS Scandinavia (Sweden). The experiment was repeated five times. Elementary Mg levels were normalised to elementary contents of sulphur (S) and represented as mean ± SEM. Data were compared by one-way ANOVA (Prism 8.4.0). Significance was accepted at p ≤ 0.05.

Acknowledgements
VC, TG, SZ, US, and BF were supported by the Deutsche Forschungsgemeinschaft (German Research Foundation, DFG), TRR 152 (P02, P14 and P15). BF and US were supported by the DFG under Germany's Excellence Strategy (CIBSS-EXC2189 project ID: 390939984) and Project-ID 403222702 -SFB 1381. AN was supported by the DFG Project-ID 335447717 -SFB 1328 (P15). TG and AN were supported by Research Training Group 2338 (DFG). We thank Veit Flockerzi for anti-TRPC1/3 antibodies, David Clapham for TRPM7 -/-HEK293T cells, Carsten Schmitz for HEK293T-REx cells stably expressing TRPM7, and Ilia Rodushkin for the support in ICP-MS. We thank Joanna Zaisserer, Lisa Pleninger, Yves Haufe, Monika Haberland, and Anna Erbacher for their technical assistance. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.