A field experiment was conducted with seven rice genotypes differing in their tolerance mechanism to Fe toxicity ( Table 1 ; Rajonandraina et al., 2023). Plants were grown under irrigated conditions at Sambaina, Manjakandriana district in the Central Highlands of Madagascar (18°53’ S, 47°47’ E) during the wet season (December to May) in 2020-2021. The site’s soil is a Gleysol with clay loam texture, pH (H₂O) 4.5, pH (KCl) 4.1, organic carbon 62 g kg^-1, cation exchange capacity 2.4 cmolc kg^-1 and total Fe 57 g kg^-1 of which 10 g kg^-1 is easily-soluble on soil reduction (Rakotoson et al., 2019).

A split-plot design was implemented with eight fertilizer treatments containing all genotypes and four replications. The following treatments were applied: control (i.e. no fertilizer), Mg, NK, NKMg, NP, NPK, NPMg and NPKMg at transplanting at the following dose as in a previous study carried out in the same field (Rajonandraina et al., 2023); 50 kg N ha^-1 as urea, 20 kg P ha^-1 as triple superphosphate, 20 kg K ha^-1 as potassium sulphate, and 26 kg Mg ha^-1 as kieserite. The size of a subplot, which contained one genotype, was 1.6 m² (0.8 m × 2 m). Subplots were randomized in each plot, which was further randomized in blocks of replicates. Four subplots were prepared for each genotype and treatment. Seedlings were transplanted at 21-d old with single plant per hill and a spacing of 20 cm × 20 cm between hills.

Sampling and measurement were done at different stages of plant growth, namely, tillering, booting, flowering and maturity. Leaf symptoms induced by Fe toxicity were visually scored as leaf bronzing score (LBS) across the whole plant canopy of each subplot as a percentage of leaf area affected on a scale from 0 (no symptoms) to 10 (100% of the leaf area affected) according to Wu et al. (2014). LBS was scored for each time point except at the maturity because of difficulties in differentiation with plant senescence. Plants were collected from two randomly selected hills per subplot, and each plant was dissected into youngest leaf (YL), middle leaves (ML), old leaves (OL) and stem/leaf sheath (ST). Flag leaves (FL) and panicles were additionally sampled at the flowering and maturing stage, respectively. The tissues were oven-dried at 60°C for 48 h until the samples were completely dry and weighed. The samples from two plants from a subplot were pooled prior to element analyses and considered as one replicate. For tissue Fe analysis, oven-dried samples were ground to a powder using a mixer mill (Retsch ZM 200, 0.2-mm sieve).

Fe concentration in dry tissue was determined as described previously (Hartmann and Asch, 2018). Briefly, Fe was extracted from pulverized dry tissues using 500 mM sodium dithionite solution, followed by colorimetric assay using 2,2’-dipyridyl. The concentration of Fe in the extract was determined by using a standard curve made with serially diluted Fe standard solution. For the measurement of other mineral elements (Ca, K, Mg), dry samples were acid-digested as reported previously (Ueda and Wissuwa, 2022; Rajonandraina et al., 2023) and measured by an inductively coupled plasma mass spectrometry (NexION 350, PerkinElmer) and inductively coupled plasma atomic emission spectrometer (ICPE-9000, Shimadzu) for field and hydroponic samples, respectively, using a standard curve made with serially diluted standard solution.

The content of malondialdehyde (MDA) was quantified according to a previous report (Hodges et al., 1999). Briefly, 0.1% trichloroacetic acid was added to pulverized frozen samples and mixed vigorously. After centrifuging the mixture at 14,000 x g for 10 min at 4°C, 90 µL of the resultant supernatant was mixed either with solution I (0.01% 2,6-di-tert-butyl-4-methylphenol in 20% trichloroacetic acid) or solution II (solution I added with 0.65% 2-thiobarbituric acid) and heated at 95°C for 30 min. The resultant mixture was briefly centrifuged and the absorbance of the supernatant was recorded at 440, 532 and 600 nm using a microplate reader (Multiscan GO, Thermo Scientific).

Total RNA was extracted from frozen pulverized samples using RNeasy Plant Mini Kit (Qiagen) according to manufacturer’s instructions. cDNA was synthesized using the PrimeScript RT Master Mix (Perfect Real Time) kit (Takara). Quantitative RT-PCR was conducted using the CFX96 Real-Time PCR Detection System (Bio-Rad) and TB Green Premix Ex Taq II (Tli RNaseH Plus) (Takara). The expression of target genes was normalized against an internal reference gene OsC3H38 (Höller et al., 2015).

Sequence library was prepared with HiSeq Standard mRNA Library Prep kit using total RNA. The resultant library was sequenced with the HiSeqX ten instrument (Illumina) and 150 bp paired-end reads were generated. Quality of the raw reads were analyzed by the FastQC software (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were trimmed by trimmomatic software (Bolger et al., 2014) using the following options; LEADING:3 TRAILING:3 SLIDINGWINDOW:4:20 MINLEN:25. Gene expression was quantified as reported previously (Pertea et al., 2016). Briefly, obtained reads were mapped to Nipponbare reference transcript (IRGSP-1.0) (Sakai et al., 2013) using HISAT2 software (Kim et al., 2015) with –min-intronlen 20 –max-intronlen 10000 options. Expression of genes was quantified using the Stringtie software (Pertea et al., 2015), yielding transcript per million (TPM) values as well as raw count data.

Differentially expressed genes were analyzed by the DESeq2 software (Love et al., 2014) using count data. PCA plot was created by plotPCA function in DESeq2 after transforming the data by the vst function.

Cis-element enrichment analysis was performed as reported previously (Ashrafuzzaman et al., 2020). The promoter element (defined as upstream 1,000-bp from the start codon) for genes significantly affected by excess Fe or combination of excess Fe and Mg was obtained at Ensembl Plants Biomart website (Kinsella et al., 2011). The enrichment analysis was performed by the AME software (McLeay and Bailey, 2010), using the database of Arabidopsis cis-elements (O’Malley et al., 2016) as the reference. The promoter sequences of all the 18,960 genes expressed in leaves (i.e. average TPM >2) were used as control.

Statistical analyses were performed with the R program (Version 4.2.0; https://www.R-project.org/) with significance level set at P = 0.05. In the field experiment, treatment effects were analyzed with a mixed model ANOVA using the lmer function in the lme4 package (Bates et al., 2015). When evaluating the effect of treatment on all genotypes, treatment was considered as a fixed effect, while genotype, block (replicate), and the interaction between block and treatment factor were considered as random effects, whereas growth stage replaced genotype as random effect when evaluating the treatment effect on each genotype across growth stages. To analyze the effect of Mg, samples from NPK and NP treatments were pooled and considered as -Mg sample group, and samples from NPKMg and NPMg treatments were pooled and considered as +Mg sample group. Similarly, samples from NP and NPMg were pooled and considered as -K sample group, and samples from NPK and NPKMg were pooled and considered as +K sample group. Each of NP, NPMg, NPK, and NPKMg treatment contained 4 replicates of 7 genotypes, thus the sample size was 56 for each of -Mg, +Mg, -K and +K treatment.

In the hydroponic experiment, one-way ANOVA was used to analyze differences among different treatments. Subsequently, Tukey-Kramer post-hoc test was implemented to compare values of each treatment.

Plant responses to an imbalance in the availability of a nutrient are dependent on the status of other mineral elements. For instance, the strength of P deficiency responses is affected by the availability of N and Fe (Balzergue et al., 2017; Hu et al., 2019; Ueda et al., 2020). Fe toxicity is observed at tissue Fe concentrations ranging from 300 to 2000 ppm (Dobermann and Fairhurst, 2000) and this wide range of critical tissue concentrations is likely due to interactions between excess Fe and deficiencies of other nutrients such as P, K, Ca, Zn and Mg. Such deficiencies may be exacerbated by excess Fe-induced changes in the rhizosphere coupled with unfavorable soil properties such as low cation exchange capacity and pH buffer power (Becker and Asch, 2005; Kirk et al., 2022). Increased availability of these elements consistently mitigates symptoms caused by Fe toxicity (Benckiser et al., 1984; Sahrawat, 2004). For instance, an application of dolomite, which is a mixture of Mg and Ca, effectively mitigated Fe toxicity-triggered reductions in growth and yield of rice in an acid Fe toxic soil in Sri Lanka (Suriyagoda et al., 2017). Similarly, the ratio of available Fe to other divalent cations (i.e. Ca, Mg, Mn) were more closely associated with the degree of yield reduction than Fe concentration alone in Fe toxic acid sulphate soils in Thailand (Moore et al., 1990). These studies would suggest the benefit of increased supply of divalent cations lies in lowering excess Fe uptake, possibly by direct competition, or increasing their supply offsets any negative effects of their deficiency on Fe toxicity responses. Our data suggest that in some genotypes Mg supply helps Fe exclusion mechanisms ( Figure 3 ), which could be explained by multiple factors. Tadano (1975) reported that reduced Mg supply lowered the Fe exclusion capacity of roots and increased the ratio of Fe translocated to shoots in culture solution, suggesting that Mg decreases Fe uptake or excess root-to-shoot translocation or both. Additionally, increased Fe concentration in stems by Mg treatment ( Figure 2C ) may form another layer of Fe-exclusion mechanism, as suggested previously (Engel et al., 2012). These effects of Mg on Fe uptake, root-to-shoot translocation and retention in stems could be regarded as a general function of Mg under Fe toxicity, as seen by a negative correlation between leaf Fe and Mg concentrations in a pot study using Fe toxic soils (Genon et al., 1994).

Even though Fe exclusion mechanisms are strengthened, some pieces of evidence suggested tissue-tolerance mechanisms conferred by Mg supply. It was suggested that the concentration ratio of Fe to Mg + Ca in flag leaves, rather than the concentration of Fe per se, is a good indicator for leaf bronzing under Fe toxicity (Sylla, 1994). This implies that Ca and Mg may confer at least some tissue tolerance. Our observations in field and hydroponic conditions both indicated attenuated leaf bronzing formation by Mg supply without a concomitant reduction in Fe concentration (X265 in Figures 3 – 5 ). A previous study also showed that Mg deficiency did not have a substantial effect on leaf Fe concentration but did induce genetic responses similar to those observed under Fe toxicity (Kobayashi et al., 2018). This suggests that Mg supply may also support tissue-based tolerance mechanisms. The hypothesis that increased Mg supply confers tissue tolerance is also supported by genotypic differences in the benefit provided by Mg treatment. Our data show that the includer genotypes benefitted to a lesser extent from Mg fertilization than the excluder genotypes ( Figure 3A ). This implies that Mg supply may provide tissue tolerance to plants that already possess exclusion mechanisms, i.e., an additive effect. However, the associated mechanisms underlying increased tissue tolerance conferred by Mg supply might be different in field and hydroponic studies, since we did not observe significant increase in Mg concentrations in young leaves at the booting stage in the field study. It implies that altered rhizospheric chemical processes or secondary effects on the uptake of other nutrients ( Figure S2 ), rather than an increase in foliar Mg concentrations, could explain this phenomenon, at least at this growth stage.

Contrary to the field study, additional Mg supply increased tissue Mg concentrations in hydroponic study ( Figure 5 ), which clearly suggests that tissue tolerance is conferred by increased tissue Mg concentrations. Deficiency of Mg disrupts Fe storage in leaf cell vacuoles (Kobayashi et al., 2018) and additional Mg supply might increase storage of Fe in vacuoles that is otherwise kept in cytosol or other compartments in a toxic state. Interestingly, in our hydroponic experiment, the reduction in leaf bronzing was not accompanied by a decrease of MDA, a biomarker for oxidative stress. These data indicate that the leaves of plants supplied with excess Mg perceive a similar extent of oxidative stress but form less leaf bronzing, suggesting that Mg specifically suppresses the process of cell death caused by oxidative stress. Thus, the underlying mechanism of Mg effects observed in our study is not via reduced oxidative stress as previously suggested (Tränkner et al., 2016; Hauer-Jákli and Tränkner, 2019). This is further supported by the gene expression data showing that most of the genes affected by Mg were not related to oxidative stress ( Figures 6E, F ).

The promoters of Fe-inducible genes were enriched with cis-elements such as WRKY, NAC and Homeobox-binding motifs as previously suggested (Kakei et al., 2021) ( Figure 7A ; Table S3 ). The binding sites for WRKY transcription factors were also enriched among the genes that were induced by the ++Fe++Mg treatment, suggesting that WRKY family proteins may not play pivotal roles in Mg-dependent alleviation of symptoms ( Figure 7A ; Table S3 ). In Arabidopsis, half of Mg deficiency-responsive genes are also responsive to ABA (Tanoi and Kobayashi, 2015), suggesting a close link between ABA and Mg status. Interestingly, some members of NAC transcription factor family that likely mediate the effect of Mg are induced by ABA and play central roles in the ABA signal transduction pathway (Todaka et al., 2012). Our results also show that many Mg-responsive genes (i.e. genes that exhibit the expression patterns of Cont <++Fe++Mg < ++Fe or Cont > ++Fe++Mg > ++Fe) are ABA, salinity or drought-related genes; Among 70 genes whose expression pattern follows Cont < ++Fe++Mg < ++Fe in leaf, OsEBP89, bHLH035, ONAC60, MYBR1, PIP1;3, and RAV2 are known to be involved in drought or salinity stress response and tolerance (Fang et al., 2014; Duan et al., 2016; Yin et al., 2017; Chen et al., 2018; Liu et al., 2020; Zhang et al., 2020). These genes also contained some NAC transcription factors (i.e. NAC15, ONAC60, NAC58, NAC121 and NAC103). Further investigations are necessary to identify the key gene(s) mediating the effect of Mg and elucidate the relationship between Mg, ABA and the causal gene under Fe toxicity conditions.

Optimal Mg supply is known to counteract several soil-related stresses. For instance, Mg confers tolerance to aluminum (Al) stress in various plant species (Rengel et al., 2015). In rice, Mg alleviates Al stress partly by reducing Al uptake and oxidative stress (Pandey et al., 2013), although the physiological pathways affected by Mg may vary in different plant species (Silva et al., 2001a). Mg also confers tolerance of salinity stress in rice (Chen et al., 2017). In both cases, Mg uptake, rather than the contact of roots with high Mg concentration, is the key factor, as shown by the attenuated tolerance in mutant lines lacking functional MGT1 which is a plasmamembrane-localized major Mg transporter (Chen et al., 2012; Chen et al., 2017).

Contrary to the effect of Mg, additional Ca application in our hydroponic study did not significantly affect leaf bronzing or Fe accumulation. This could be partly due to the similarity of the ionic radii of Fe²⁺ (77 pm) and Mg²⁺ (72 pm) compared with Ca²⁺ (100 pm). Similarity in radii may mean that Mg²⁺ competes with Fe²⁺ better than Ca²⁺ for binding with other molecules and mitigate the deleterious effects of Fe to confer tissue tolerance. Similar Mg-specific effects were also observed under Al stress. A previous investigation in soybean suggested that Al toxicity was specifically mitigated by Mg²⁺ but not by similar concentration of Ca²⁺ (Silva et al., 2001a; Silva et al., 2001b), consistent with our observation that the positive effect of Mg²⁺ is not simply due to the ionic strengths of competing divalent cations. Interestingly, the expression of MGT1 is induced both in roots and leaves under Fe toxicity ( Table S1 ). Thus, it is possible that upregulation of MGT1 and accompanying increase of Mg uptake might be an intrinsic mechanism that plants possess to cope with Fe toxicity, as well as other edaphic stresses such as salinity and Al toxicity. MGT1 is found within a QTL explaining Al stress tolerance (Yamaji et al., 2009). Intriguingly, a QTL for Fe toxicity-induced leaf bronzing (qFETOX1-2) was previously found close to the position of MGT1 (37.7 Mb on chromosome 1) (Wu et al., 2014), which is also close to a QTL for foliar Mg concentration (Norton et al., 2010). Whether allelic variation exists at MGT1 locus, and if that is the cause of genotypic variation of leaf bronzing, needs further investigation.

We have shown in field and hydroponic experiments that increasing the supply of Mg to rice roots improves Fe toxicity tolerance via exclusion- and tissue tolerance-based mechanisms. Evidence from transcriptome analysis indicated the involvement of several genes expressed in shoot tissue that potentially mediate the Mg effect, and of particular relevance seem to be the altered expression levels of genes containing the binding sites for NAC transcription factors. Previous studies showed more than three-fold variation in leaf Mg concentration among diverse rice accessions grown under control conditions (Chen and Ma, 2013; Yang et al., 2018), suggesting scope for improvement of the Mg status in rice plants through a breeding approach. Whether utilizing potential allelic variation of MGT1, as well as increased foliar Mg concentration effectively reduces Fe toxicity in the field need to be investigated further.