This study used the Renewable Industrial Products from Rapeseed (RIPR) diversity population of inbred lines of Brassica napus genotypes (Thomas et al., 2016). These were developed from the ERANET-ASSYST consortium diversity population (Bus et al., 2011, 2014; Körber et al., 2012, 2015) with further lines included. A subset of 383 genotypes were selected, comprising 160 winter-, 127 spring-, and seven semi-winter-oilseed rape (OSR), 35 swede, 15 winter fodder, and 39 exotic/unspecified habits. These were previously characterized for leaf mineral concentrations by inductively coupled plasma-mass spectrometry (ICP-MS) of polytunnel-grown plants sampled at the rosette stage (typically 6–8 true leaves showing; Thomas et al., 2016). The full leaf mineral dataset is available at the Brassica Information Portal (BIP¹; The Earlham Institute, Norwich, United Kingdom) and at doi: 10.5281/zenodo.59937.

The leaf concentrations of 21 mineral elements including Ca and Mg in the RIPR diversity population were previously determined by Thomas et al. (2016). Leaf Ca concentrations varied over threefold across the population, from 5,838 mg kg^-1 to 18,752 mg kg^-1. Leaf Mg concentrations were of a similar order of magnitude and varied over twofold, from 5,118 mg kg^-1 to 13,429 mg kg^-1. The frequency distribution of these two elements approximated a normal distribution (Supplementary Figure 6). Leaf Ca and Mg concentrations were among the highest, positively correlated elements measured across genotypes and tissues, with an r-value of 0.87 (P < 0.001). Leaf Ca and Mg concentrations varied between crop type, with higher concentrations of both elements in leaves of spring and semi-winter OSR than in winter OSR, winter fodder, and swede types.

To identify candidate loci, SNPs and GEMs were used separately in analyses. A total of 1,295 and eight SNPs were found to be significantly associated with B. napus leaf Ca concentration after FDR and Bonferroni corrections, respectively. After removing SNPs with low second allele frequency, this was reduced to 247 and five SNPs, respectively, across all chromosomes. Visually determined association peaks on Manhattan plots were observed on chromosomes A3, A6, A7, A10, C2, C3, and C9 (Figure 1A). The most well defined peak was located on chromosome A10 and contained four out of the five SNPs above the Bonferroni corrected significance threshold (P = 0.05). The fifth SNP above this threshold fell in a peak on chromosome C9, in a region known to share sequence homology with parts of chromosome A10 (Chalhoub et al., 2014). A total of 5,557 and 141 GEMs were identified as significantly associated with leaf Ca concentration after FDR and Bonferroni corrections, respectively (Figure 1B). Notable peaks were observed on chromosomes A2 and C2. Single, associated GEMs were found at similar locations to SNP association peaks on chromosomes A3 and C2. The A. thaliana ortholog of B. napus genes corresponding to both these GEMs is At5g10140, which encodes FLOWERING LOCUS C (FLC), a transcription factor important for controlling flowering time (Michaels and Amasino, 1999). A further associated GEM was found in a region of chromosome A10, close to a SNP peak associated with leaf Ca concentration. Other single GEMs associated with leaf Ca concentration were observed on chromosomes A5, A6, C4, and C6. B. napus genes corresponding to these GEMs on chromosomes A5 and C4 are orthologous to A. thaliana At2g45660, which encodes SUPPRESSOR OF OVEREXPRESSION OF CO 1 (SOC1), another flowering time regulator (Lee et al., 2000).

The SNP and GEM associations for leaf Mg concentration were similar to those described for leaf Ca concentration. A total of 1,012 and one significant SNP(s) were found after FDR and Bonferroni corrections, respectively. After removing SNPs with low second allele frequency, this was reduced to 166 and zero SNPs, respectively, across all chromosomes. Of these 166 SNPs, 86 were identical to SNPs identified as significantly associated with leaf Ca concentration after FDR correction and removal of SNPs with low second allele frequency, indicating the potential of similar mechanisms to partly regulate accumulation. Visually determined association peaks largely co-localized with those associated with leaf Ca concentration, specifically on chromosomes A3, A7, A10, C2, C3, and C9 (Figure 1C). The most well defined peak was again on chromosome A10, and as before, a region on chromosome C9 with sequence homology to this region also contained associated SNPs. An association peak on chromosome A2 was also particularly well defined, containing 15 SNPs above the FDR corrected significance threshold. A total of 12,973 and 1,489 GEMs were identified as significantly associated with leaf Mg concentration after FDR and Bonferroni corrections, respectively, across all chromosomes (Figure 1D). Of these, 5,160 and 131 were also significantly associated with leaf Ca concentration after FDR and Bonferroni corrections, respectively. Notable peaks were again observed on chromosomes A2 and C2. The most highly associated GEMs on A3 and C2 were identical to those associated with leaf Ca concentration which correspond to A. thaliana FLC, and an associated GEM on C4 is identical to the GEM associated with leaf Ca concentration corresponding to SOC1.

Linkage disequilibrium describes the non-random association of alleles at different loci (Slatkin, 2008). Genes located physically near to each are generally inherited together, and hence are often in very strong LD. It is therefore feasible that any number of genes within LD of SNPs significantly associated with a trait may be controlling such associations. Based on associative transcriptomics results, seven loci were focussed on for the identification of candidate genes. These comprised regions of chromosomes A2, A3, A5, A6, A10, C2, and C4. A total of 17 B. napus candidate genes orthologous to 15 A. thaliana genes are summarized in Table 2. Four candidate genes were selected based on direct GEM hits as described above. These are Cab002472.4 and BnaC02g00490D (on chromosomes A3 and C2, respectively) encoding orthologs of A. thaliana At5g10140 (FLC), and Cab025356.1 and Bo4g024850.1 (on chromosomes A5 and C4, respectively) encoding orthologs of A. thaliana At2g45660 (SOC1). One and two candidate genes on chromosomes A2 and C2, respectively are orthologous to A. thaliana MAGNESIUM TRANSPORTER 7 (MGT7/MRS2-7). This was previously characterized in Arabidopsis thaliana as an Mg transporter important for Mg uptake at low external concentrations (Gebert et al., 2009). A further candidate gene on chromosome A10 is orthologous to A. thaliana AUTOINHIBITED CA²⁺-ATPASE, ISOFORM 8 (ACA8). This was previously characterized as a plasma membrane-localized Ca²⁺ transporting ATPase in A. thaliana (Bonza et al., 2000) and was identified as Ca responsive in B. rapa (Graham et al., 2014). The functions of the remaining nine candidate genes were selected based on sequence homology and annotations of A. thaliana orthologs and are either uncharacterized, or have not previously been experimentally shown to be involved in plant Ca or Mg accumulation (Table 2). These and At2g45660 (SOC1) were used for the selection of A. thaliana mutants.

Shoot Ca concentrations in a preliminary A. thaliana phenotyping experiment varied threefold between individual plants, from 8,684 to 26,387 mg kg^-1 dry weight (DW; Supplementary Table 3). Much of this variation was observed within genotypes, with the largest variation observed in lines At5g7320.1 and At5g08670.1. Four mutant lines had significantly lower mean shoot Ca concentrations than wild type plants. These were At2g13610.2 (P < 0.05), At5g07320.2 (P < 0.01), At5g08670.1 (P < 0.05) and At5g48650.2 (P < 0.01; Figure 2A). Leaf Mg concentrations varied less, with twofold variation from 8,189 to 16,186 mg kg^-1 DW observed between individual plants. Much of this variation was between genotypes. Two mutant lines had lower mean shoot Mg concentration than wild type plants. These were At2g13610.1 (P < 0.05) and At5g48650.2 (P < 0.05; Figure 2B). Based on these data, lines At2g13610.2, At5g07320.2, and At5g48650.2 were selected for characterization of tissue specific leaf Ca and Mg concentration.

Calcium concentrations varied over eightfold between tissues and pooled tissue samples, ranging from 4,998 mg kg^-1 DW in stems to 40,536 mg kg^-1 DW in cauline leaves (Figures 3A–D and Supplementary Table 4). Cauline and rosette leaf Ca concentrations were similar, ranging from 32,966 mg kg^-1 DW in rosette leaves to 40,536 mg kg^-1 DW in cauline leaves. Mean silique Ca concentrations were lower in lines At2g13610.2, At5g07320.2, and At5g48650.2 than wild type plants (P < 0.01; Figure 3B). Mean stem Ca concentrations were lower in lines At5g07320.2 and At5g48650.2 than wild type plants (P < 0.05; Figure 3C). Mean cauline leaf Ca concentration was lower in line At5g48650.2 than wild type plants (P < 0.05; Figure 3D). Mg concentrations varied over ninefold between tissues and pooled samples, ranging from 2,608 mg kg^-1 DW in stems to 23,999 mg kg^-1 DW in rosette leaves (Figures 4A–D). Cauline leaf and rosette leaf Mg concentrations had a similar range, from 18,026 to 21,328 mg kg^-1 DW and from 19,240 to 23,999 mg kg^-1 DW, respectively. Lines At2g13610.2 and At5g48650.2 had lower mean silique Mg than wild type plants (P < 0.05; Figure 4B). Line At5g48650.2 also had lower mean stem Mg than wild type plants. Finally, and comparable with results from cauline leaf Ca concentration analysis, mean cauline leaf Mg concentration was lower in line At5g48650.2 than in wild type plants (P < 0.05; Figure 4D).

In summary, data from A. thaliana experiments identified four and two mutant lines with lower shoot Ca and Mg concentrations than wild type plants, respectively, and three of these mutations have tissue specific phenotypes. The main tissue specific effects were observed in silique tissue, with lower silique Ca concentrations in all three mutant lines investigated in the tissue partitioning experiment.

Leaf Ca concentration was highly associated with loci on chromosomes A3, A6, A7, A10, C2, C3, and C9 (Figure 1A). Similar loci were associated with leaf Mg concentration, specifically in regions of chromosomes A3, A7, A10, C2, C3, and C9 (Figure 1C). The most highly associated SNP for leaf Ca concentration was located on chromosome A10 (Figure 1A). This co-localizes with associated markers on C9 and markers on A10 and C9 for leaf Mg concentration. Co-localization of association peaks and associated markers for both mineral elements is unsurprising, as leaf Ca and Mg concentration data used in this study were very highly correlated (r = 0.87, P < 0.001; Thomas et al., 2016) and may reflect the relative lack of selectivity between these and other group II elements during accumulation within the plant (White, 2001). Such correlations between shoot Ca and Mg concentration have been previously shown in B. oleracea (Broadley et al., 2008) and a number of other angiosperm species (Broadley et al., 2004). Bus et al. (2014) previously investigated the genetic control of shoot ionome traits across 505 lines of B. napus using 3,910 SNPs in association analyses. Results showed two associations at a locus on chromosome C9 for shoot Ca and Mg concentration with a further association on chromosome C7 for shoot Ca concentration. The detection of an association locus on C9 is consistent with co-localized associations identified in this study. These results are also consistent with earlier findings by Broadley et al. (2008) who identified significant QTL for shoot Ca and Mg in B. oleracea on chromosomes C2, C6, C7, C8, and C9. Together, these results indicate the importance of loci on chromosomes A10 and C9 for Ca and Mg accumulation. The QTL identified for shoot Mg in B. oleracea on chromosome C2 by Broadley et al. (2008) is also consistent with findings in the present study that a locus on C2 is highly associated with leaf Mg concentration in B. napus. However, further work is required to confirm whether the loci are in close proximity to one another. To our knowledge, the remaining loci identified in this study have not been previously identified as important QTL for leaf Ca and Mg concentration in Brassica spp.

Gene-expression marker analyses associated markers corresponding to FLC and SOC1 with both leaf Ca and Mg concentrations. In A. thaliana, FLC is a repressor of flowering (Michaels and Amasino, 1999) and it has been previously shown that FLC transcript concentration correlates with vernalization requirements (Sheldon et al., 2000). Expression levels of SOC1 also correlate with flowering time in A. thaliana; in lines which flower later, SOC1 expression is very low (Lee et al., 2000). It is thought that SOC1 expression is repressed by FLC, indicating the tight regulatory links between these genes and flowering time. Leaf Ca and Mg concentration data used in this study were obtained from analysis of plants in the RIPR panel (Thomas et al., 2016) which includes a large number of spring and winter B. napus varieties. Thomas et al. (2016) observed differences in leaf Ca and Mg concentrations between these types, with higher mean concentrations of both leaf Ca and Mg in spring OSR compared to winter OSR and winter fodder types. Since winter OSR types are generally considered to have longer vernalization requirements than spring types, it is possible that the association of FLC and SOC1 with leaf Ca and Mg concentration observed in this study was a result of differences in vernalization requirement between these groups rather than direct genetic control of Ca and Mg uptake. It is worth noting that the association of GEMs with a trait does not indicate the causative polymorphism/s, only genes in which expression level is associated with variation in the trait. The causative polymorphism/s may lie in the promotor sequence of such genes, or localize somewhere upstream in the pathway. Hence, in the case of the flowering time genes identified here, it is unclear whether or not the observed associations with leaf Ca and Mg concentration are directly caused by changes in expression of FLC and SOC1. Despite this, their expression appears to be a suitable marker for the concentrations of these elements in B. napus. Further to this, the concentrations of a number of other mineral elements measured in the study of Thomas et al. (2016) were found to vary between crop types with typically different flowering times and vernalization requirements. Most notably, leaf concentrations of Mo, Na, P, and S were higher in spring OSR than winter OSR types. This suggests that flowering time, or the upstream mechanisms leading to changes in flowering time, has an effect on the concentrations of a number of nutrients in B. napus, though the pathway/s that lead to these differences remain unclear.

Identification of high-resolution loci influencing leaf Ca and Mg concentrations enabled locus-specific exploration of the Brassica pan-transcriptomes and other genome resources for candidate genes within LD of SNPs. LD is especially relevant to the efficacy of associative transcriptomics in the absence of a marker in a trait-controlling gene. LD decays relatively quickly in B. napus (Ecke et al., 2010; Harper et al., 2012), and this helps to reduce the number of possibilities when searching for candidate genes. However, in this study, typically hundreds of genes were still within previously estimated LD decay (around 1–2 cM on average; Ecke et al., 2010) of most candidate loci. Fortunately, well annotated browsers of Brassica A and C genomes are available at Ensembl Plants (Kersey et al., 2016), which enabled rapid identification of nearby genes in the reference sequences with links to functional annotation of A. thaliana orthologs.

Most notable genes identified using this workflow include an ortholog of A. thaliana ACA8 near markers associated with leaf Ca on chromosome A10 and two orthologs of A. thaliana MGT7 near markers associated with leaf Mg on chromosomes A2 and C2. ACA8 encodes a Ca²⁺ transporting ATP-ase localized to the plasma membrane (Bonza et al., 2000). A B. rapa ortholog of A. thaliana ACA8 was previously identified under an eQTL hot spot on chromosome A3 (Graham et al., 2014). The eQTL associated with this gene was defined as Ca-responsive, i.e., the direction of the eQTL changed under high Ca supply. The A. thaliana ortholog of ACA8 was further investigated in silico in the same study using publically available phenotypic data at the PIIMs database (Baxter et al., 2007). This led to the identification of ACA8 T-DNA knockout mutants with greater shoot Ca concentrations than control plants in over 50% of mutant samples, indicating the ability of this gene to influence Ca accumulation in Brassica. MGT7 is a member of the MGT/MRS2 Mg transport family. This was previously characterized as a key transporter for Mg uptake at low external Mg concentrations by Gebert et al. (2009). Arabidopsis thaliana T-DNA knockout mutants were severely retarded in development when grown at low external Mg concentrations, but were visually unaffected when grown at higher external Mg concentrations. Both ACA8 and MGT7 are very promising candidate genes for the control of Ca and Mg accumulation in B. napus. The presence of these genes within LD of highly associated SNPs demonstrates the effectiveness of associative transcriptomics in candidate gene identification. Since ACA8 and MGT7 knockout mutants had previously been characterized in A. thaliana, they were not included in further experiments in this study.

The preliminary A. thaliana phenotyping experiment identified four mutant lines with lower shoot Ca concentrations and two with lower shoot Mg concentrations than wild type plants. The most notable of these was At5g48650.2, the only line in which both shoot Ca and Mg was affected. The gene mutated in this line encodes NUCLEAR TRANSPORT FACTOR 2 (NTF2). This protein is proposed to function in the import of RAN, a multifunctional GTPase involved in nucleocytoplasmic transport (Zhao et al., 2006). It is the first time that it has been characterized with a shoot Ca and Mg phenotype in A. thaliana. Further investigation of this line showed that it had lower Ca and Mg concentration than wild type plants in all tissues except rosette leaves, suggesting it could be a promising candidate for manipulating the translocation of Ca and Mg to specific tissues in crop plants. At5g07320 encodes the ATP-Mg/Pi transporter APC3. Despite being annotated as an Mg/Pi transporter, mutants in this gene were only found to have reduced shoot Ca concentration. Tissue specific characterisation of this line showed silique and stem Ca concentrations were lower than wild type plants. This suggests that, at least in these conditions, the effects of the mutation are limited. However, effects at different external Ca or Mg concentrations might be different. A further line mutated in a gene encoding an ABC transporter was found to have both lower Ca and Mg concentrations in siliques compared to wild type plants. Identifying candidate genes controlling silique nutrient traits is particularly important in B. napus, which is mostly grown for the harvest of seeds which have a secondary use in animal feed. All A. thaliana experiments in this study took place using a high-nutrient compost. This could have masked the phenotypes of mutations in a number of candidate genes which may have been able to maintain normal Ca and Mg concentrations due to sufficient soil concentrations. In addition, it is possible that the mutations characterized here would show greater defects in shoot Ca and Mg concentrations when grown in nutrient limiting conditions. As well as this, all plants in both A. thaliana experiments were harvested at a single growth stage and other phenotypes might be seen at other growth stages. Despite this, four candidate genes analyzed here have proven to be potential targets for altering Ca and Mg concentrations in B. napus. These are orthologs of the Arabidopsis thaliana genes At2g13610, At5g07320, At5g08670, and At5g48650.