To assess the evolutionary relationships among MGT10, MGR8, and MGR9, we compared their protein sequences with those of several well-characterized magnesium transporters. Phylogenetic analyses showed that MGT10 shared the closest evolutionary history with the E. coli CorA, whereas MGR8 and MGR9 clustered together with the CorC proteins of Thermus parvatiensis and E. coli ( Supplementary Figure S1A ). The magnesium transport function of CorA protein family members depends on the Gly-Met-Asn (GMN) motif located in the extracellular loop of the channel (Guskov et al., 2012; Ishijima et al., 2021). This motif could be found in the MGT10 sequence, whereas it was absent in MGR8 and MGR9 sequences ( Supplementary Figure S1B ). Previous work reported that MGR8 and MGR9 belong to a separate clade of the plant MGR family, which is most distant from CNNMs (Tang et al., 2022; Zhang et al., 2022). Our phylogenetic analyses confirmed that MGR8 and MGR9 cluster together in a MGR sub-family ( Figure 1 ). Interestingly, this cluster does not include CNNMs and MGR1–7 and does include the bacterial Mg²⁺ transporter CorC ( Figure 1 , Supplementary Figure S1A ). Structurally, CNNMs consist of an N-terminal extracellular domain, a transmembrane domain of unknown function (DUF21), a large cytosolic region containing a cystathionine-synthase (CBS) pair domain, and a putative cytosolic cyclic nucleotide–binding homology (CNBH) domain at the C-terminus. Our sequence alignment shows that while MGR8 and MGT9 also harbor the DUF21 and the CBS-pair domain, the C-terminus is distinct from the CNBH domain of CNNMs and shares a considerable number of identical amino acids with the C-terminal CorC-HlyC domain of bacterial CorC ( Supplementary Figure S1B ). Members of the CorC Mg²⁺ transporter family were shown to be Na⁺ dependent since depletion of Na⁺ resulted in loss of their Mg²⁺ transport activity (Yamazaki et al., 2013; Jin et al., 2022). The Asparagine residue Asn94 in the transmembrane domain of the CorC protein from Thermus parvatiensis was identified as important for Na⁺ sensitivity (Huang et al., 2021). Interestingly, both MGR8 and MGR9 also contain this residue in their transmembrane DUF21 domain ( Supplementary Figure S1B ), suggesting possible coordination of transport between Mg²⁺ and Na⁺ across the chloroplast inner envelope. Taken together, these data indicate that MGR8 and MGR9 may function as CorC-like transporters, whereas MGT10 shares the closest evolutionary history with the CorA channel.

Phylogenetic analysis of MGR8 and MGR9 homologs revealed that both proteins are distributed amongst land plants, algae, and cyanobacteria. Within the plant genus Brassicaceae, MGR8 and MGR9 proteins form two well-supported clades and all investigated species code for 2–3 copies each ( Supplementary Figure S2 ). However, species from Brassica, Eruca, Crambe, and Sinapis, only carry the MGR8-like protein, albeit in multiple distinct copies, resulting from gene duplication. There are 1–2 copies of MGR8- and MGR9-related proteins in other plant species and cyanobacteria species like Synechocystis sp., Synechoccocus elongatus, and Gloeobacter violaceus ( Figure 1 ). In diatoms (e.g., Thalassiosira pseudonana), we could find homologs for CNNMs and MGR1–7 but not for MGR8 and MGR9. These findings indicate that MGR8 and MGR9 share evolutionary histories but are not evenly distributed in nature.

To assess if the Mg²⁺ transport protein sequences of Arabidopsis MGR8, MGR9, and MGT10 are conserved in green algae, we have searched for homologs in the unicellular Chlamydomonas reinhardtii and the multicellular Volvox carteri. While there was no AtMGR8 or AtMGR9 homolog, we identified one MRS4 sequence in each green algae species as close homologs of AtMGT10 ( Supplementary Figure 3 ). MRS4 shares the highest sequence identity with MGT10 in the N-terminal long loop and in the transmembrane region ( Supplementary Figure S4 ). In addition, the protein sequences of both CrMRS4 and VcMRS4 contain the characteristic GMN motif.

Using heterologous expression in the Mg²⁺ uptake-deficient E. coli strain TM2 (ΔcorA ΔmgtA ΔyhiD), we assessed the magnesium transport activity of MGR8 and MGR9 and compared it with that of MGT10. Under standard conditions, the growth of TM2 requires the addition of at least 10 mM Mg²⁺ to the LB medium (Ishijima et al., 2015). TM2 cells expressing either MGR8 or MGR9 cDNA grew optimally in LB medium supplemented with 1 mM Mg²⁺ but not in the absence of added Mg²⁺, while the cells with an empty vector did not grow in either absence or presence of up to 1 mM Mg²⁺ ( Figures 2A, B ). Interestingly, TM2 cells expressing MGR9 could grow well in LB medium supplemented with 0.5 mM Mg²⁺, while MGR8-expressing cells failed to grow ( Figures 2A, B ). In addition, when 0.8 mM or less Mg²⁺ was added, the growth of cells expressing MGR8 but not of cells expressing MGR9 was reduced ( Supplementary Figures S5A, B ). At 10 mM Mg²⁺, the cells expressing MGR8 grew much faster and better than the MGR9-expressing cells and the cells with an empty vector ( Supplementary Figure S5C ). As previously reported by Ishijima and colleagues (2015; 2021), we observed that TM2 cells expressing MGT10 could readily grow in LB medium without Mg²⁺ supplementation ( Supplementary Figure S5D ). These results indicate that MGR8, MGR9, and MGT10 functionally complement the Mg²⁺ auxotrophy of the TM2 cells and that the expressed proteins are capable of transporting Mg²⁺ without any additional protein partners. The ability of TM2 cells expressing MGR9 to grow at lower external Mg²⁺ concentration than the cells expressing MGR8 indicates a broader concentration range at which MGR9 may be active.

We found that AlCl₃ inhibited the growth of TM2 cells expressing MGT10 ( Supplementary Figure S5D ) as previously reported (Ishijima et al., 2015), likely due to Al³⁺ uptake into the cells (Ishijima et al., 2018). In contrast, no growth inhibition of TM2 cells expressing either MGR8 or MGR9 was observed with 1 mM AlCl₃ in combination with 1 mM Mg²⁺ ( Figures 2C, D ). These results indicate that the Mg²⁺ transport activity of MGR8 and MGR9 is not inhibited by Al³⁺ and that they do not transport Al³⁺ into the E. coli cells under these conditions. Taken together, we propose that MGR8 and MGR9 transport Mg²⁺ by a different mechanism than MGT10.

Human CNNMs contain evolutionarily conserved residues, whose mutations cause hypomagnesemia and associated congenital diseases (Stuiver et al., 2011; Giménez-Mascarell et al., 2019; Chen Y.S. et al., 2020). Disease-causing mutations include E357K from the DUF21 domain of CNNM2, R407L, P409L, and T495I from the ATP-binding site within the CBS domain of CNNM4 (Chen Y.S. et al., 2020), demonstrating the importance of these four residues for CNNM transport activity. In addition, the residues corresponding to G466, E471, and D472 of MGR8 are conserved near the end of the CBS domain in the CNNM family (Zhang et al., 2022). To test whether these seven residues ( Supplementary Figure S1B ) are important for the Mg²⁺ transport activity of MGR8, we introduced point mutations (E261K, R388L, P390A, P390L, G466A, T469A, T469I, E471A, and D472A) and expressed the constructs in TM2 cells. The TM2 cells expressing the MGR8 G466A mutant did not grow in LB medium supplemented with 1 mM Mg²⁺ ( Figure 2E ). All other mutants grew similarly to the cells expressing the wild type MGR8 ( Figures 2E, F ). These results indicate that, among the mutated residues conserved in CNNMs, only G466 in the CBS domain is critical for the Mg²⁺ transport activity of MGR8. Based on these findings and the notion that all mutated residues are also conserved in MGR9 ( Supplementary Figure S1B ), we postulate that MGR8 and MGR9 transport Mg²⁺ by a different mechanism than CNNMs.

The tissue-specific expression pattern of the MGT10 gene was previously analysed by GUS-staining (Drummond et al., 2006; Sun et al., 2017). Recently, the more sensitive quantitative RT-PCR method was used to analyze the expression of MGR8 and MGR9 genes (Zhang et al., 2022). In our quantitative RT-PCR analysis we included all three transporter genes and investigated their tissue expression pattern in Arabidopsis plants grown at 0.75 mM Mg²⁺ (standard conditions). MGR8 exhibited high expression in all vegetative organs and lower expression in flowers ( Figure 3A ). MGT10 was highest expressed in mature organs, in contrast to MGR9 which was predominantly expressed in seedlings. Notably, expression levels of MGR8 and MGT10 genes were highest in leaves.

Considering the observed activity of MGR9 in TM2 cells under low Mg²⁺ conditions, we postulated the possible involvement of MGR9 in Arabidopsis growth at such levels. Accordingly, we evaluated gene expression in response to three different Mg²⁺ concentrations. Plants were grown for two weeks at 0.75 mM Mg²⁺ and then transferred for 3–4 weeks at either no (0 mM), standard (0.75 mM), or high (3 mM) Mg²⁺. At 0 mM Mg²⁺, expression of the MGR9 gene in leaves was upregulated 20-fold, while expression of MGR8 and MGT10 was downregulated as compared to standard conditions ( Figure 3B ). MGR9 expression was also upregulated at 3 mM Mg²⁺ (2-fold), while the expression of the other two genes was unaltered relative to 0.75 mM Mg²⁺. These results indicate that MGR8 and MGT10 are the mainly expressed Mg²⁺ transporters in the leaves of Arabidopsis plants grown at standard Mg²⁺ concentration, whereas MGR9 expression is elevated in conditions of deficiency or excess Mg²⁺.

To further investigate the gene expression patterns of the three magnesium transporters under Mg²⁺-deficient conditions, we obtained two independent homozygous T-DNA insertion knockout mutants for MGR8 (mgr8-1 and mgr8-2) and MGR9 (mgr9-1 and mgr9-2), and a heterozygous knockdown mutant for MGT10 (mgt10) ( Supplementary Figure S6 ). We speculate that these mutant plants might compensate for the loss of one transporter by upregulating the expression of the other magnesium transporter genes. The expression of MGR9 was similar in mgr8 and mgt10 mutants grown hydroponically at 0.75 mM Mg²⁺, whereas MGR8 was more expressed in mgt10 than in mgr9 ( Supplementary Figure S7 ). When grown at 0 mM Mg²⁺, MGR9 was upregulated 2-fold in mgr8 and slightly upregulated in mgt10, whereas MGT10 and MGR8 expression was downregulated or unaltered in mgr9 and mgt10 mutants relative to 0.75 mM Mg²⁺ ( Figure 3C ). These results strengthen the evidence for a role of MGR9 in Mg²⁺-deficient conditions.

Knowing a protein’s subcellular localization is a pivotal element in unravelling its functional role within the cell. Sun and colleagues (2017) localized MGT10 to the chloroplast envelope by western blot analysis. Recently, MGR8 and MGR9 were also localized to the chloroplast envelope by using a GFP-fluorescence approach (Zhang et al., 2022). In our study, we raised an antibody against an MGR8-specific C-terminal peptide ( Supplementary Figure S1B ). The generated antibody detected a band at a relative molecular weight (Mr) of 65 kDa in total protein extracts from wild type (WT) leaves and in mutant lines mgr9-1 and mgr9-2, but not in mgr8-1 and mgr8-2. The Mr is in good agreement with the theoretical MW of 65.65 kDa for the protein lacking the chloroplast transit peptide (amino acids 1-71, Supplementary Figure S1B ). Immunoblot analyses of purified chloroplasts membrane subfractions, thylakoid- and envelope membranes confirmed an envelope location of the MGR8 protein ( Figure 3D and Supplementary Figure S8 ).

Thylakoid stacking, also referred to as overall grana size, is stabilized by Mg²⁺ ions. We, therefore, examined the chloroplast ultrastructure and thylakoid stacking in three single mutant lines using transmission electron microscopy (TEM). Regular WT-like chloroplast and thylakoid morphology were observed in 6-week-old mgr8 and mgr9 mutants ( Figures 4A–C ). Interestingly, it was reported that the homozygous double mutant mgr8mgr9 had impaired chloroplast development (Zhang et al., 2022). This double mutant had no thylakoid stacks in two-week-old plants and short and inflated thylakoid stacks in 3-week-old plants.

The mgt10 mutant had a peculiar leaf phenotype with yellow vein leaves ( Supplementary Figure S9 ), in agreement with previously published results (Liang et al., 2017; Sun et al., 2017). The green interveinal regions contained chloroplasts with regular grana ( Figure 4D ), whereas the yellow vein regions contained normal chloroplasts with regular grana and peculiar plastids with macro-grana ( Figures 4E, F ). Macro-grana are unusually wide grana consisting of a high number of stacked thylakoid lamellae and are associated with an overall underdeveloped stroma thylakoid membrane system. Macro-grana were especially abundant in the bundle sheath cells, although they could be also observed in palisade and spongy parenchyma cells sampled from the vein region. Taken together, these results in combination with the results from Zhang and colleagues (2022) suggest that both families are required for regular thylakoid stacking and chloroplast development. However, the opposite pattern in grana size, i.e., smaller in mgr8mgr9 (Zhang et al., 2022) and larger in the peculiar plastids observed in mgt10 ( Figure 4F , Sun et al., 2017), implies differential roles of the two families of transporters.

To understand the role of MGR8, MGR9, and MGT10 in photosynthetic reactions in the thylakoid membrane, we grew plants at three different concentrations of Mg²⁺ (0, 0.75, and 3 mM) and measured the slow kinetics of chlorophyll fluorescence induction in fluctuating light. In addition to the single mutants, we investigated the impact of a double mutant of the two most expressed transporters on photosynthetic performance. We crossed the mgr8-2 and mgt10 mutant lines, resulting in the mgt10mgr8-2 double mutant. All mutant lines grew like WT and displayed similar shoot and root weight at all three Mg²⁺ concentrations ( Supplementary Figure S9 ). We first determined the maximum quantum yield of photosystem II (PSII) photochemistry (F_v/F_m). Our results show that the F_v/F_m yield was slightly but significantly lower in both mgr9 lines and mgt10 at 0 mM Mg²⁺, indicating a reduced maximum photosynthetic efficiency in terms of electron transport ( Supplementary Table S1 ).

In plants grown at 0 and 0.75 mM Mg²⁺, on transition from low to high light, NPQ was induced slower in all mutants as compared to WT ( Figures 5A, B ). The steady-state NPQ was the lowest in mgt10 and mgt10mgr8-2, intermediate in mgr8, and the least affected in the mgr9 mutants. Following the transition from high to low light, NPQ relaxed similarly in all genotypes. The electron transport through both photosystems, indicated by Y(II) and Y(I), was like WT in all mutants and conditions ( Figures 5A, B ). In parallel experiments, we recorded electrochromic shift (ECS) kinetics at the end of each low-to-high light and high-to-low light transition to estimate the total PMF and its partitioning. The PMF size was alike WT at the end of each transition ( Supplementary Figure S10A ). The ΔpH, also known as the H⁺ concentration gradient, in transition from low to high light significantly decreased in all mutants grown at 0 and 0.75 mM Mg²⁺ ( Figures 5D, E ), and was like WT in transition from high to low light ( Supplementary Figure S10B ). These data indicate that MGR8, MGR9, and MGT10 are involved in building up the pH gradient to rapidly activate NPQ without largely affecting the electron transport through photosystems and overall PMF size. Plants grown at 3 mM Mg²⁺ did not differ in any of the studied parameters ( Figures 5C, F , Supplementary Figures S10A, B ).

To test whether the observed reduced NPQ and altered thylakoid ultrastructure could be a result of altered magnesium homeostasis, we measured the mineral content in shoots and isolated chloroplasts using inductively coupled plasma optical emission spectrometry (ICP-OES) and inductively coupled plasma mass spectrometry (ICP-MS), respectively. When grown in the absence of Mg²⁺, all mutants had a slightly but significantly higher Mg²⁺ content in the shoots compared to WT ( Supplementary Figure S11 ). At standard Mg²⁺ concentration (0.75 mM), shoots of all single mutants had less whereas the double mutant mgt10mgr8-2 had a WT-like Mg²⁺ content. At 3 mM Mg²⁺, all mutant shoots had WT levels of Mg²⁺ except for mgt10 which had a significantly higher content. In chloroplasts isolated from plants grown at standard Mg²⁺ concentration, mgr8-2 and mgt10 had a Mg²⁺ content reduced by 38%, mgr9 by 21%, whereas mgt10mgr8-2 displayed a 17% increase relative to WT ( Figure 6A ). The chloroplast Na⁺ and K⁺ contents in most single mutant lines were similar to those in the WT ( Figures 6B, C ). Nevertheless, mgt10mgr8-2 contained significantly higher levels of both Na⁺ and K⁺, suggesting that the double loss of MGR10 and MGR8 also impacts other ion transporters.

To investigate whether the observed differences in the chloroplast Mg²⁺ content have impacted RuBisCO activity in CO₂ fixation, we have measured net photosynthesis at the growth light and at a higher light intensity. As shown in Supplementary Figure S12 , this activity was not significantly different among the genotypes at neither light intensity, in line with the WT-like growth and biomass data ( Supplementary Figure S9 ).

We identified CrMRS4 and VcMRS4 proteins as close homologs of AtMGT10 ( Supplementary Figures S3 , S4 ). To explore the function of MRS4, we characterized a Chlamydomonas knock-out mutant from the CLiP library (mrs4) and complemented this mutant with the VcMRS4 gene (mrs4::MRS4) ( Supplementary Figure S13 ). The knockout mutant grew poorly in light on minimum TP medium, but supplementation with Mg²⁺ or complementation with VcMRS4 considerably improved its growth ( Figure 7A ). Growth of mrs4 in TAP medium in darkness did not impact growth ( Figure 7A ), indicating that MRS4 is involved in autotrophic growth.

To further investigate the cause of the reduced growth of mrs4, we examined various parameters of photosynthetic reactions in the thylakoid membrane. The knockout mutant mrs4 had a significantly reduced F_v/F_m relative to WT due to a higher F₀. These results indicate that mrs4 had a higher proportion of closed PSII centers already in the dark-adapted state ( Figures 7B, C ). In addition, mrs4 had reduced electron transport rates throughout the range of tested light intensities, indicating enhanced light sensitivity ( Figure 7D ). When exposed to high light (2500 µmol photons m^-2 s^-1), the knockout mutant was able to induce NPQ in the first minute of illumination, but it decreased to 0 after 3 min. The NPQ in the supplemented and complemented lines were higher than in mrs4 and more stable over time, but still lower than WT ( Figure 7E ). In addition, the PSII activity, indicated by Y(II), was lower in all mutants ( Figure 7F ). The total PMF across the thylakoid membrane was slightly but significantly higher in mrs4 and in mrs4::MRS4 at two time points in high light (660 µmol photons m^-2 s^-1), and like WT in the supplemented line ( Figure 7G ). The mrs4 mutant PMF was dominated by ΔpH, whereas the complemented line and the Mg²⁺ supplemented mutant had intermediate ΔpH levels between WT and mutant ( Figure 7H , Supplementary Figure S14 ). The observed high ΔpH in the mrs4 mutant is likely to cause lower PSII activity due to photoinhibition. Notably, despite the high ΔpH, the NPQ levels in the mutant were low, indicating a reduced capacity for photoprotection ( Figures 7D, F ). Taken together, these results suggest that the mrs4 mutant has an enhanced sensitivity to light due to photoinhibition and malfunctioning of the PSII complex.

To test if the observed lower PSII activity is linked to an altered thylakoid structure, WT and mrs4 morphology were analyzed using TEM. Chlamydomonas cultures grown in TAP medium in darkness revealed no major differences between WT and mrs4 ( Supplementary Figure S15A ). In addition, a similar overall cell ultrastructure was observed in WT and the mrs4 mutant after 3 h or 4 days of illumination at 100 µmol photons m^-2 s^-1 on minimal TP medium. However, inside the chloroplast, the mrs4 thylakoids were more stacked and consisted of more layers ( Supplementary Figures S15A, B ).

To investigate if the decrease in photoprotective capacity, increased ΔpH and altered grana stacking could be a result of altered mineral content in the cell, we measured the Mg²⁺/Na⁺, K⁺/Na⁺, and Ca²⁺/Na⁺ ratios of mrs4 and WT using ICP-MS. There were no significant differences in the mineral content ratios for cells grown on TAP in darkness ( Supplementary Figure S16 ). However, we found that mrs4 cells grown at 100 µmol photons m^-2 s^-1 in minimal TP medium had significantly elevated Mg²⁺/Na⁺ and Ca²⁺/Na⁺ ratios as compared to WT ( Supplementary Figure S16 ). These results indicate that mrs4 accumulated higher levels of Mg²⁺ and Ca²⁺ inside the cell. An excess of Mg²⁺ could explain the increased grana size, as it was proposed for MGT10 in Arabidopsis (Sun et al., 2017). Taken together with the observed altered thylakoid ultrastructure and associated reactions, our data suggests that MRS4 ensures an optimal Mg²⁺ concentration in the chloroplast for photosynthesis and cell growth.

Arabidopsis thaliana wild type (Columbia-0) and the T-DNA insertion lines mgr8-1 (SALK_074964), mgr8-2 (SALK_007335), mgr9-1 (SALK_061515), and mgr9-2 (SALK_087652) were obtained from the SALK collection (Alonso et al., 2003), and have been described by Zhang et al. (2022). The mgt10 (GABI_764F12) was obtained from the GABI-KAT collection (Rosso et al., 2003) and is a heterozygous knockdown mutant. Genotyping of the T-DNA insertion lines was done by PCR/RT-PCR with gene-specific primers ( Supplementary Table S2 ). The double mgt10mgr8-2 mutant was obtained by crossing mgt10 and mgr8-2. Wild type (WT) plants and mutants were grown hydroponically for 6-7 weeks in a growth chamber (CLF PlantMaster; Plant Climatics, Wertingen, Germany) using a daily cycle of 16 h of light (120 µmol photons m⁻² s⁻¹) at 21°C and 8 h of dark at 19°C at a relative humidity of 70%. The nutrient solution was prepared as described (Gibeaut et al., 1997; Conn et al., 2013) with the modification that MgSO₄ was used at the concentration of 0 (low Mg²⁺), 0.75 (standard Mg²⁺) and 3 mM (high Mg²⁺).

Wild type Chlamydomonas reinhardtii (strain CC-4533) and the mrs4 mutant (strain LMJ.RY0402.244553) were obtained from the CLiP library of the Chlamydomonas Resource Centre (https://www.chlamylibrary.org/). The strains were maintained in darkness on agar plates (1.2% w/v) prepared with Tris-Acetate-Phosphate (TAP) containing 0.4 mM MgSO₄. For genotyping, DNA was extracted, and PCR was performed with the appropriate combination of primers ( Supplementary Table S2 ). For the spot tests, a loopful of culture was transferred to TAP medium and grown in darkness for 2 days. The cells were then resuspended to a density of 5 x 10⁵ cells mL^-1 in TAP which was enriched or not with 2 or 5 mM MgSO₄, spotted at different dilutions on TAP agar plates containing corresponding MgSO₄ concentration, and allowed to grow in darkness at 20°C. From the same inoculum, cells were resuspended in minimal Tris-Phosphate (TP) medium which was enriched or not with 2 or 5 mM MgSO₄, spotted on TP plates, and allowed to grow at 20°C using a daily cycle of 16 h of light (100 µmol photons m⁻² s⁻¹) and 8 h of dark. For photosynthetic analyses, strains were grown in liquid TP medium at an initial density of 1x 10⁶ cells mL^-1 in a volume of 50 ml for 4 days using a daily cycle of 16 h of light (100 µmol photons m⁻² s⁻¹) and 8 h of dark.

Whole genome sequence data from a representative set of plant species were downloaded from Phytozome v13 (https://phytozome-next.jgi.doe.gov/). Bacterial data were downloaded from NCBI (https://www.ncbi.nlm.nih.gov/) and sequences of the diatom Thalassiosira pseudonana from the JGI Genome portal (https://genome.jgi.doe.gov/portal/). The amino acid sequences were then queried using BLASTp v2.2.28+ (Camacho et al., 2009) and MAFFTv6.843b (Katoh and Toh, 2008) to identify homologs of the Arabidopsis MGR9 (At1g55930) and MGR8 (At3g13070) (https://github.com/topel-research-group/misc/blob/master/blast_and_align.py). Phylogenetic analyses of the identified homologs were then performed using MrBayes v3.2.6 (Ronquist et al., 2012) for 1 million generations (Brassicacese analysis) and 2 million generations (Plants, bacteria, diatom, green algae and human sequences), respectively, after which the Average Standard Deviation (AvgStdDev) was significantly low, and the analyses were assumed to have converged. The first 25% of the tree samples were discarded and the remaining trees were summarized in majority consensus trees (see Figure 1 , Supplementary Figure S2 ).

The following protein sequences were used for constructing the phylogenetic tree in Supplementary Figure S1A : AtMGT10 (NP_568424.1), EcCorA (P0ABI4), EcCorC (P0AE78), TpCorC (WP_060384576.1), AtMGR8 (NP_187914.1), AtMGR9 (NP_187914.1), hCNNM2 (Q9H8M5) and hCNNM4 (Q6P4Q7). For the construction of the phylogenetic tree of CrMRS4 (Cre50g761497), amino acid sequences for Mg²⁺ transporters were downloaded from the plant membrane protein database (http://aramemnon.uni-koeln.de/) for Arabidopsis thaliana, and from Phytozome v13 (https://phytozome-next.jgi.doe.gov/) for Chlamydomonas reinhardtii (v6.1) and Volvox carteri (v2.1). The sequences were aligned using ClustalW in MEGA11 (Tamura et al., 2021) and the phylogenetic trees were generated using the neighbor-joining method with 500 bootstraps and the default settings of MEGA11.

The Arabidopsis sequences that encode for MGR8 (stock no. GPSO-0186) and MGR9 (BRC no. pda13388) were obtained from the Arabidopsis Biological Resource Center (http://abrc.osu.edu/) and the RIKEN BRC through the National BioResource Project of the MEXT/AMED, Japan, respectively. The MGR8 and MGR9 proteins contain N-terminal extensions. The ChloroP1.1 algorithms predicted a chloroplast leader and transit peptide cleavage sites at amino acid 71 of MGR8 and amino acid 72 of MGR9, respectively ( Supplementary Figure S1B ). The cDNA fragments that encode MGR8 from F72 to the C-terminal Q661 residue and MGR9 from L73 to the C-terminal E653 residue were subcloned in frame at the NcoI site of the pTV118N vector (TakaraBio, Japan) for the E. coli complementation assay (Ishijima et al., 2015). To generate various mutants, inverse PCR-based mutagenesis was performed using appropriate primer sets ( Supplementary Table S2 ), and the obtained mutant cDNAs were subcloned into the pTV118N vector. All plasmid sequences were confirmed by DNA sequencing.

The plasmids containing the MGR8 and MGR9 wild type and mutant cDNAs were transformed into E. coli mutant TM2 (ΔcorA ΔmgtA ΔyhiD) cells (Ishijima et al., 2015). TM2 transformed with an empty pTV118N vector was used as a negative control. Culture growth was monitored at 37°C with an ODBOX-c OD-Monitor (TAITEC, Japan).

Total RNA was isolated from two-week-old seedlings and plant tissues of 6-week-old plants with an E.Z.N.A. R6827-01 Plant RNA kit (Omega Bio-Tek, GA, USA) and residual DNA was removed with E1091 DNAse (Omega Bio-Tek). cDNA was synthesized using 500 ng of total RNA through iScript cDNA Synthesis Kit (Bio-Rad, Hercules, CA, USA). Quantitative real-time PCR analyses were conducted with a SsoAdvanced Universal SYBR Green Supermix on a CFX96 Touch Thermal Cycler (Bio-Rad). Fifty ng of cDNA was used as qPCR template in 10 µl reactions. Amplifications were done in a two-step PCR with the following conditions: initial denaturation for 2 min at 95°C, followed by 40 cycles of denaturation for 5 s at 95°C, annealing for 30 s at 60°C, and extension for 10 s at 72°C. After amplification, melt-curve analyses were performed for all primers. Gene-specific primers used were ordered from Bio-Rad ( Supplementary Table S3 ). ΔCq method (2^-ΔCq) was used to calculate relative expression using PEX4 and ACTIN8 as the reference genes.

Total protein extracts were prepared from 100 mg snap frozen leaves ground in 500 μL PEB buffer (AS08300, Agrisera, Umeå, Sweden) and centrifuged to remove insolubilized material. Intact chloroplasts were prepared from 50 g fresh leaves ground in 50 mM HEPES-KOH (pH 7.8), 300 mM sorbitol, 2 mM Na₂EDTA, and 5 mM ascorbic acid, centrifuged and loaded on Percoll 40/75% gradients. Intact chloroplasts were collected following centrifugation of the gradients at the 40%-75% interface, washed, and resuspended in 50 mM HEPES-KOH (pH 7.8), 300 mM sorbitol, and 2 mM Na₂EDTA. Intactness was verified by microscopy and the number of chloroplasts was counted using a hemocytometer. Thylakoid and envelope membranes were prepared from lysed chloroplasts, and centrifuged at 3000 g and 150,000 g, respectively. Protein concentration was determined using a Bradford assay (Bio-Rad). Equal amounts of protein (8 μg) were separated on 4-15% (w/v) SDS-PAGE gels (Bio-Rad) and transferred to polyvinylidene difluoride (PVDF) membranes. Nonspecific bindings on the membranes were blocked with 5% (w/v) milk. An antibody was generated by Agrisera in rabbit against the EHVLADNSKKQQ C-terminal peptide of MGR8 ( Supplementary Figure S1B ). The rabbit anti-Tic40 and Lhcb1 antibodies were purchased from Agrisera (AS10709, AS01004). Following incubation with a goat anti-rabbit secondary antibody (Bio-Rad), the chemiluminescent signal on the blots was detected using Clarity and Clarity Max ECL substrates (Bio-Rad).

The fixation and embedding of 1 x 2 mm Arabidopsis leaf pieces for TEM were carried out as described (Böszörményi et al., 2020). In the case of WT and mgt10, TEM samples close to leaf veins were also fixed. Fixation and embedding of Chlamydomonas cells were done following the protocol of Skepper (2000), with gentle centrifugation following each step to collect the cells. Briefly, the Chlamydomonas cells of the liquid media were centrifuged at 1000 g for 5 min at room temperature, washed with 10 mL fresh TAP/TP medium, then centrifuged again as described above. The pellet was then resuspended in 1 mL of 2.5% glutaraldehyde in 50 mM HEPES (pH 7.4) and fixed for 1 h at room temperature on a tube rotator. Cells were then centrifuged at 5000 g for 3 min at room temperature. The pellet was washed for 3 x 5 min with ddH₂O, with centrifugation with the above parameters after each washing step. Samples were postfixed for 1-2 hrs at 4°C in 1 mL 1% OsO₄, 1.5% (w/v) K₃[Fe(CN)₆], and 2 mM CaCl₂. After centrifugation as above, samples were washed 4 x 5 min with ddH₂O. Samples were then stained for 1-2 h at room temperature in 1 mL Bulk Stain solution (2% uranyl acetate in 0.05 M maleate buffer, pH 5.5). After staining, samples were again centrifuged as above and then washed 3 x 5min with ddH₂O. After dehydration in graded ethanol series, samples were transferred to acetonitrile and then embedded in epoxy resin (Quetol 651, nonenyl succinic anhydride - NSA, methyl-5-norbornene-2,3-dicarboxylic anhydride - NMA, and catalyst dimethylbenzylamine - BDMA). The resin-embedded Chlamydomonas pellet samples were polymerized in Eppendorf tubes that were then cut into smaller pieces which were then re-embedded in Durcupan ACM epoxy resin (Fluka Chemie AG) in blocks suitable for ultrathin sectioning. A Reichert Jung ULTRACUT E microtome was used for ultrathin (70 nm) sectioning of all resin-embedded samples. After a 5-min staining with 5% uranyl acetate dissolved in methanol, and a subsequent 5-min-long treatment with Reynold’s lead citrate solution, the copper grids holding the sections were investigated using a JEOL JEM 1011 (JEOL Ltd., Japan) at 80 kV accelerating voltage as described (Dukic et al., 2019). An Olympus Morada CCD camera (Olympus Optical Co. Ltd., Japan) was used to take digital images. At least 25 randomly chosen cell sections were studied for each sample and representative images for all samples were chosen for the preparation of Figure 4 and Supplementary Figure S15 .

Fast kinetics of Chl a induction in Arabidopsis leaves were recorded on 30 min dark-adapted plants by applying saturating red actinic light of 3,600 µmol photons m⁻² s⁻¹ for 1 s using a Handy-PEA (Hansatech, UK) fluorometer. Initial F₀ and F_m fluorescence values were determined by the saturating pulse. The maximal quantum yield of PSII photochemistry (F_v/F_m) was calculated as (F_m-F₀)/F_m using Hansatech PEA Plus v1.10 software according to Strasser et al. (2004). The slow kinetics of Chl fluorescence induction and P700 oxidation-reduction of Arabidopsis were simultaneously recorded with a Dual-PAM-100 equipped with DUAL-DB and DUAL-E emitter-detector module (Walz, Effeltrich, Germany). The kinetics were recorded on attached leaves of 30 min dark-adapted plants exposed to fluctuating red actinic light as follows: 10 min in low light (70 µmol photons m⁻² s ⁻¹) followed by 3 min in high light (650 µmol photons m⁻² s⁻¹) and then again 3 min in low light. Saturating red pulses of 5,000 µmol photons m⁻² s⁻¹ and 800 ms duration were applied for determination of the maximal fluorescence yield in the dark state (F_m) and during the period with actinic light (F_m′). NPQ and Y(II) were calculated based on changes in Chl fluorescence as (F_m–F_m ^′)/F_m ^′ and (F_m ^′–F)/F_m ^′, respectively, according to Genty et al. (1989). Y(I) was calculated from absorbance changes at 830 nm according to Klughammer and Schreiber (1994).

To measure PSII activity in Chlamydomonas, cells grown in TP in the light were resuspended to 30 µg Chl mL^-1 and incubated with agitation at 50 rpm in darkness for 15 min. Kinetics of Chl fluorescence induction were recorded using DUAL-PAM-100 during actinic illumination at 2,500 μmol m^-2 s^-1 for 8 min followed by 4 min of dark relaxation. Saturating red pulses of 3,000 µmol photons m⁻² s⁻¹ were applied to the samples in a cuvette under continuous stirring for determination of F_m and F_m ^′. The maximal photochemical efficiency of PSII (F_v/F_m), NPQ and Y(II) were calculated as above. Rapid light response curves of electron transport rate were recorded using photosynthetically active radiation increasing stepwise from 26 to 2,015 μmol photons m^-2 s^-1.

For imaging analysis of Chl fluorescence using FluorCam 800 MF (Photon System Instruments, Drasow, Czech Republic), 1 mL of Chlamydomonas cells containing 30 µg Chl were dark-adapted, transferred onto a 12-well cell culture plate (Nunc, Thermo Fisher Scientific, USA) and analyzed using default quenching 1 protocol with actinic white light at 100 µmol photons m⁻² s⁻¹ exposure for 6 min followed by 2 min in darkness. Saturating flashes of white light at 3,000 μmol photons m⁻² s⁻¹ were applied after every min of light/dark with a shutter speed of 10 μs and sensitivity of 50%.

ECS measurements were carried out using the Dual-PAM-100 system equipped with a P515/535 module. Arabidopsis plants were first 30 min dark-adapted and then exposed to fluctuating red light as described in section 4.8. ECS was measured at the end of each low-to-high light and high-to-low light transition. PMF size was calculated as the difference between the ECS signal in light and the minimum value of the ECS signal after the light was turned off. Calculation of PMF partitioning to ΔpH and ΔΨ was performed using the steady-state time point of the ECS signal in darkness (Cruz et al., 2001). Before each ECS measurement, three saturating 50-μs actinic red flashes of 200,000 μmol photons m⁻² s⁻¹ were applied to determine the ECS_ST; subsequently, the ECS_ST amplitude was used to normalize the ECS signal before the calculation of PMF size and partitioning values.

For ECS in Chlamydomonas, cells grown in TAP in darkness were resuspended at 30 µg Chl mL^-1 and incubated with agitation at 50 rpm in darkness for 15 min before being layered on a glass slide and exposed to actinic red light for 2 and 7 min. The light was switched off and decay kinetics were measured to calculate ECSt, ECSst, PMF size and partitioning to ΔpH and ΔΨ as for Arabidopsis.

Plant shoot samples were collected directly from the hydroponic system after 6-7 weeks, washed twice with double distilled water (ddH₂O), and then dried in the oven for 3 days at 80°C. The dried sample was crushed to a fine powder using a ceramic mortar and pestle. A 100 mg subsample was digested with 70% (v/v) HNO₃ and 30% (v/v) H₂O₂ in a ratio of 3:1 (Chen A. et al., 2020) at 240°C and 200 bars for 15 min in a pressurized microwave oven (Milestones SRL, Italy). After dilution to 3.5% acid, the element concentrations were determined using ICP-OES (5100; Agilent Technologies) and data were processed using Agilent ICP Expert software (Hansen et al., 2009). Similarly, a subsample of chloroplasts purified on a 40%/75% Percoll gradient and adjusted to contain 10⁹ chloroplasts mL^-1 were digested and analyzed using ICP-MS (7900; Agilent Technologies) and data were processed using Agilent Masshunter software, as the low element concentrations required a more sensitive detector than the one used for plant tissues.

To determine the mineral content in Chlamydomonas, cell colonies maintained on 1.2% (w/v) agar TAP plates were used to inoculate liquid TAP medium for growth in darkness and shaking for three days. Early log phase cultures were harvested and washed with either TAP or TP medium followed by resuspension in the corresponding medium to reach 0.5 x10⁵ cells mL^-1. The cells were grown either in TAP in darkness or TP under continuous illumination (100 μmol photons m^-2 s^-1) and shaking for four days. Samples (in triplicate) containing 2 x10⁷ cells (~800 mg) were harvested and washed twice in 5 mM HEPES (pH 7.0) and 2 mM EDTA before collection and air-drying of the cell pellet. Pelletized cells were digested in screw-capped Teflon vials on a hot plate at elevated temperatures (25-75°C) using a 5:2 mixture of HNO₃ (68-70% v/v) and H₂O₂ (30% v/v)). Samples were then dissolved in 2% HNO₃ and analyzed for cation ratios (Mg²⁺/Na⁺, K⁺/Na⁺, and Ca²⁺/Na⁺) on an ICP-MS (ICAP-Q; ThermoScientific). Elemental ratios were quantified using a series of externally calibrated standards (Inorganic Ventures Co.) and analytical uncertainties were estimated at ~5-10% (1σ) based on repeat measurements of the calibration standards.

Net photosynthesis rates in terms of CO₂ fixation were determined using Li-COR portable Photosynthesis System Li-6400 (Lincoln, Nebraska, USA). Plants were first light-adapted for at least 1 h in the growth chamber and then exposed in the gas exchange chamber to broad spectrum light generated with red, green, and blue LEDs at 120 and 700 µmol photons m⁻² s⁻¹ in atmospheric CO₂ (440 µmol mol⁻¹) for approximately 5 min or until a steady state was reached. Data were normalized to leaf area determined using ImageJ.

The Chlamydomonas MRS4 gene was incomplete in the v5.5 genome (Cre50.g761497), and trials to obtain it by TAIL-PCR have failed. For complementation experiments of mrs4, we have chosen to clone its homolog in Volvox carteri (Vocar.0040s0086) using the pipeline described by Mackinder et al. (2017). Briefly, the open reading frame was amplified from Volvox genomic DNA by PCR from two fragments (2-3 kbp each) using Phusion High-Fidelity DNA polymerase (New England BioLabs) and primers given in Supplementary Table S2 . The PCR products were purified from agarose gels, re-assembled, and cloned in-frame with a C-terminal Venus-3×FLAG in pLM005 by Gibson assembly (New England BioLabs). The construct was linearized by EcoRV-HF and verified by sequencing. For transformation of the mrs4 mutant, cells were cultured in TAP liquid medium at 23°C under light at 150 µmol photons m⁻² s⁻¹ to 1-2 x 10⁶ cells mL^-1. The cells were washed and suspended at 2 x 10⁸ cells mL^-1 in MAX Efficiency Transformation reagent (Invitrogen) together with the linearized plasmid and by electroporation using a NEPA21 electroporator (NEPA GENE). The transformants were plated on TP agar supplemented with 25 μg mL^-1 and grown under light (150 μmol photons m⁻² s⁻¹) for 7 days. Those transformants showing better growth than the mrs4 mutant were screened by PCR with primers given in Supplementary Table S2 to verify the correct insertion of VcMRS4.