A genome-wide association study (GWAS) identifies multiple loci linked with the natural variation for Al³⁺ resistance in Brassica napus

Keywords: acidity, aluminium, candidate genes, canola, genetic variation, genome-wide association analysis, pH, QTL, soil, tolerance, toxicity.

Acid soils limit crop production because they pose several stresses to plant growth. In addition to high concentrations of protons (H⁺), which are themselves toxic, acid soils can reduce growth, induce nutrient deficiencies (e.g. phosphorus and magnesium) and expose plants to several toxic minerals such as aluminium and manganese. Aluminium is the most damaging of the mineral toxicities because the prevalence of soluble trivalent Al³⁺ cations in acid soils can rapidly inhibit root growth at micromolar concentrations. Consequently, a common symptom of acid-soil stress is short, stubby roots which affect the uptake of water and nutrients. Additional symptoms that typically occur over longer periods due to nutrient imbalances include smaller canopies and chlorotic patches on the leaves (Foy 1984).

The most effective long-term strategy for ameliorating soil acidity is the application of lime (calcium carbonate) to raise soil pH, or other amendments that add basic cations to the soil or otherwise reduce the concentration of the harmful Al³⁺ cations. While important, these approaches are slow to take effect and it can take years for sub-soil pH to rise and for these changes to be reflected in better yields (Gazey and Davies 2009). Furthermore, liming may not be economically viable in some production systems or for subsistence farmers. In all these cases, the cultivation of acid-soil tolerant species or better-adapted germplasm allows production to continue while agronomic solutions gradually improve soil pH.

Substantial genotypic variation for acid soil tolerance has been identified in a range of important crop species including wheat (Triticum aestivum L.), barley (Hordeum vulgare L.), maize (Zea mays L.), rice (Oryza sativa L.), and sorghum (Sorghum bicolor L.) and this variation is correlated with greater resistance to Al³⁺ stress (Raman and Gustafson 2010). While several mechanisms appear to control this resistance to Al³⁺ (Ryan et al. 2011; Kochian et al. 2015), one mechanism that contributes at least part of the resistance to all these species is the release of organic anions from the root apices. The release of anions such as malate and citrate are believed to chelate the harmful Al³⁺ cations in the apoplast and reduce their damaging interactions with the meristematic and elongating zones of the root (Delhaize et al. 1993; Ryan and Delhaize 2010). The genes controlling the release of these organic anions for Al resistance have been identified in many species and they fall into two main families: the Aluminium-activated malate transporter (ALMT) family encodes anion channels that facilitate the release of malate and other compounds, and the Multi-drug and toxic compound exudation (MATE) family encodes a proton co-transporter that facilitates citrate release (Sasaki et al. 2004; Delhaize et al. 2012; Pereira and Ryan 2019). In some species, members of the Cys₂His₂-type zinc finger transcription factor family mediate signals between the perception of Al³⁺ in the soil and the induction of these transporters as well as genes involved in other mechanisms of resistance (Iuchi et al. 2007; Yamaji et al. 2009).

Canola (rapeseed, Brassica napus L.) is the third largest crop in Australia after wheat and barley and an important crop around the world. It is used for oil production, as source material for biofuels and, increasingly, as a winter forage crop. B. napus is generally considered to be sensitive to acid soils (Ryan 2018). Certainly, liming improves canola yields on acid soils and soil pH management is recommended by agronomists in Australia (GRDC 2015; Ryan 2018; French et al. 2019). Indeed, some recommend against growing canola in soils below pH 4.5, or even below pH 4.7 if the exchangeable aluminium levels are high.

B. napus likely originated from a spontaneous hybridisation between rape (Brassica rapa L.) and cabbage (Brassica oleracea L.). This event, together with recent selective breeding programmes to lower the erucic acid and glucosinolate contents in the seed, reduced the genetic diversity of modern varieties which make them susceptible to biotic and abiotic stresses (Kebede et al. 2010). A few studies have reported significant genotypic variation in Al³⁺ resistance in B. napus and among other members of the Brassicaceae. For instance, Huang et al. (2002) found that species containing either the R or C genomes such as radish (Raphanus sativus; 2n = 2× = 18, genome: RR), Brassica carinata (2n = 4× = 34, genome: BBCC), B. oleracea (2n = 2× = 18, genome: CC) and B. napus (2n = 4× = 38, genome: AACC) consistently showed greater Al³⁺ resistance than species without these genomes such as Brassica campestris (syn. B. rapa, 2n = 2× = 20, genome: AA), Brassica nigra (2n = 2× = 16, genome: BB), Brassica juncea (2n = 4× = 36, genome: AABB) and Arabidopsis thaliana (2n = 10, genome: ARAR). They proposed that a strategy for improving the acid soil tolerance of cultivated brassicas was to transfer the Al³⁺ resistance loci from R. sativus to the commercial species. Moroni et al. (2006) reported some genotypic variation in the acid soil tolerance of B. napus but concluded results from soil screens were different from the hydroponic screens. French et al. (2019) performed field trials and also found significant genotypic variation in acid soil tolerance among different B. napus cultivars. However, the variation between different trials made it difficult to rank cultivars with confidence and the authors suggested controlled pot trials would be more reliable. More recently, Gao et al. (2021) performed a genome-wide association study (GWAS) with 169 genotypes of B. napus to identify variation in Al³⁺ resistance during germination and early growth in Petri dishes. They identified eight single nucleotide polymorphisms (SNPs) associated with relative root length on chromosomes A03, A07, A09, A10, C05, C06, and C09 and five SNPs associated with root dry weight located on chromosomes A03, A04, A10, C05, and C07.

Although B. napus is generally considered to be sensitive to acid soils, it does show the same Al³⁺-dependent release of malate and citrate from roots usually associated with more resistant species (Ligaba et al. 2004). This may indicate that the magnitude of organic anion release is also important for determining the level of resistance. Malate release from B. napus is likely controlled by BnALMT1 and BnALMT2 (Ligaba et al. 2006) while citrate release from B. oleracea (one of the ancestral parents of B. napus) is likely controlled by a MATE-type transporter (Wu et al. 2014). What is not clear is whether differences in the magnitude of organic anion release contribute to the observed variation in Al³⁺ resistance in B. napus.

Despite the expected yield benefits of better-adapted germplasm (Ryan 2018), breeding companies have not considered acid-soil tolerance to be a priority trait. The reason for this is partly because the genotypic variation for acid soil tolerance in B. napus is uncertain and robust genetic markers for the trait are not available. To address both these issues, we examined the responses of B. napus to Al³⁺ stress and conducted a genome-wide association study (GWAS) for Al³⁺ resistance using 352 diverse accessions of B. napus. GWAS can achieve a higher resolution than linkage mapping because the linkage disequilibrium decays much faster when association analysis is performed with a diverse set of germplasm compared with linkage analysis using a biparental population.

A series of preliminary experiments were performed to assess the responses of the Brassicaceae species to pH and Al³⁺ and to develop a robust screen. All preliminary experiments included the B. napus cultivars ATR-Bonito (still popular in Australia) and Oscar (older cultivar) but other B. napus hybrids (InvigorR5520P, PHR1712, GT42), an experimental line (3674-13) and close relatives of B. napus were included in different experiments. The related Brassicaceae included B. carinata (cv. Baltia), B. rapa (cv. 1B1337), B. juncea (cv. CSR18), and wild radish (Raphanus raphanistrum). The first experiment grew different genotypes for 10 days in different solutions from pH 3.8 to 6.2. A pair of near-isogenic lines of wheat, that differ in Al³⁺ resistance, named ET8 (Al³⁺ resistant) and ES8 (Al³⁺ sensitive) were directly compared with several B. napus genotypes. These wheat lines possess different alleles of the major Al³⁺ resistance gene TaALMT1 (Sasaki et al. 2006) and they were derived from germplasm described previously (Delhaize et al. 1993).

Germplasm used in the GWAS analysis consisted of 352 diverse accessions from the BnASSYST diversity panel which originated from different countries and includes winter, semi-winter and spring, fodder, kales, old European and Australasian oilseed rape, and rutabaga/swede morphotypes. A detailed description and their origins has been provided in earlier studies (Hasan et al. 2006; Bus et al. 2011; Körber et al. 2015).

The nutrient solution used for all hydroponic growth experiments contained (μM): Ca, 500; K, 1000; Mg, 200; NO₃⁻, 1000; NH₄⁺, 1000; PO₄³⁻, 50; SO₄²⁻, 200; Cl⁻, 2030; Na⁺, 100; Fe³⁺, 10; B(OH)₃, 5; Mn²⁺, 1; Zn²⁺, 0.15; Cu²⁺, 0.1; MoO₄⁻², 0.1. Iron was added as a FeCl₃ from a 0.1 M stock solution after the nutrient solution was adjusted to below pH 5.0. Aluminium was added from a stock of 20 mM AlCl₃. Solutions were transferred to plastic tanks containing 11 or 20 L of nutrient solution with added AlCl₃ concentrations and pH adjusted as mentioned for each experiment. Pre-germinated seedlings were wrapped in soft polyethylene foam (Ormonoid Abelflex™, Parexgroup) and positioned in one of the 70 holes (∼16 mm diameter) drilled into the plastic lids so their roots were suspended in the aerated nutrient solution (Supplementary Fig. S1b). For the pH treatments, germinated seedlings were transferred to aerated nutrient solutions adjusted to pH 3.9, 4.3, 5.0 and 6.0 for 10 days. The pH of these solutions was monitored daily. For comparing the Al³⁺ resistance of different species of wild radish, Oscar, wild radish, CSR18, 1B1337 and Baltia were planted in the same nutrient solution with or without 16 μM AlCl₃ (pH 4.3) for 12 days. To examine the effect of seed size on Al³⁺ resistance, seeds from each genotype were weighed individually and grouped into extreme size groups as presented in the Results. Seed of all genotypes was germinated and seedlings were grown hydroponically with or without 12 μM AlCl₃ (pH 4.3) for 12 days. To compare the Al³⁺ resistance between canola and wheat, seedlings of canola (Oscar, ATR-Bonito, GT42 and 45Y88 CL) and wheat (ET8 and ES8) were treated with 0, 10 and 20 μM AlCl₃ or 0, 8 and 16 μM AlCl₃ under pH 4.3 for 12 days. Relative root length was calculated as [(root length with Al³⁺)/(root length without Al³⁺)] × 100. The statistical comparison of these percentages was described previously (Zhou et al. 2013).

A total of 352 genotypes from the BnASSYST diversity panel were screened for Al³⁺ resistance in the nutrient solution described above. Seed were pre-germinated on damp tissue in Petri dishes placed at 4°C overnight and then 28°C for 3 days (Fig. S1a). The germinated seedlings were transferred to tanks (approximately 13 cm high × 37 cm long × 26 cm wide) containing 11 L nutrient solution with or without 12 μM AlCl₃ treatment (pH 4.3). The activity of free Al cations was estimated to be approximately 3.5 μM with Geochem EZ software (http://www.plantmineralnutrition.net). In order to screen the 351 genotypes (and one ‘check’) with three replicates in the control and Al³⁺ treatments (where each replicate had five plants), nine separate growth experiments were conducted over 7 months, each requiring 18 tanks. Therefore, there were 2268 experimental units. Each tank could hold 70 seedlings which meant that up to 14 different genotypes could be grown at one time (remembering there are five seedlings per genotype). One of the 14 genotypes was the ‘check’ (cv. Smart) which was selected at the start of the project to be included in all tanks of every run. Since a single experimental unit (replicate) was the average measurement of five seedlings, at least 20 seeds per experimental unit were germinated to ensure sufficient uniformity for planting. The tanks were placed in growth cabinets (six tanks in three growth cabinets) with the following conditions: 14 h/28°C and 10 h/22°C day–night cycle, 70% relative humidity and ∼300 μmol m⁻² s⁻¹ light intensity. Each tank was aerated with four pipes attached to aquarium air pumps (Fig. S1). The pH of the solution was checked every 1 or 2 days and adjusted as required using 1 M HCl or 1 M NaOH. All solutions were replaced after 7 days. After 12 days, the seedlings were harvested. Plants of each genotype were removed from the tank together, laid out on a dark cloth, and photographed next to a scale. The length of the longest root was recorded and then the shoots and roots were separated and dried at 80°C for 2 days to obtain dry weights (DW). The three main measurements of root length, root DW and shoot DW were averaged from the five plants (one replicate). Measurements of tissue dry weight were preferred over fresh weights because it is easy to introduce errors from the external water clinging to the plants or from the rapid wilting of small plants.

Five contrasting accessions were selected for estimating organic acid efflux from roots. The selection was based on the predicted means of the three measures of Al³⁺ resistance (see below), seed availability and germination rate. Seeds were surface sterilised by washing with 70% ethanol for 1 min followed by 20 min in 1% sodium hypochlorite on a shaker. All procedures were performed using sterile procedures in a laminar flow cabinet. The seed were then thoroughly rinsed using seven to eight flushes with sterile water and placed on 2% water agar in Petri dishes. These were wrapped in foil and kept at 4°C overnight before being moved to room temperature for three more days. The germinated seedlings (15 per pot) were gently transferred onto wire grids suspended over 45 mL of nutrient solution (pH 6.0) in sterile culture pots. Each pot was a single replicate for that genotype. The pots were placed on a shaker in a growth cabinet. The nutrient solution was renewed every 3 days and signs of contamination were checked by plating some of the old solution on Nutrient Agar (Difco) plates. Exudate collection began after 9 days. The mesh was lifted from the pots with sterile forceps and transferred into a new sterile pot containing 15 mL sterile 0.2 mM CaCl₂ solution (pH 4.3) and placed on a shaker for 20 min to wash the seedlings. Note that the wire grid could be lowered so the seedling roots were again fully submerged in the reduced volume of solution. After 20 min this wash solution was discarded and another 15 mL 0.2 mM CaCl₂ (pH 4.3) was added to collect the exudates. After 3 h of gentle shaking in a cabinet the solution was collected (control treatment) and replaced with 15 mL of 0.2 mM CaCl₂ + 50 μM AlCl₃ (pH 4.3). After 2 min of shaking this rinse, the solution was replaced with another 15 mL 0.2 mM CaCl₂ + 50 μM AlCl₃ and placed on the shaker again to collect exudates. The solutions were sampled for exudates after 3 h and 20 h and results expressed as the average efflux over 3 h without Al³⁺ (control) and the average efflux over 3 h and 20 h with Al³⁺. Malate and citrate concentrations in the solutions were assayed as described previously (Kawasaki et al. 2021).

The preliminary experiments were analysed using analysis of variance (ANOVA). If the sums of squares associated with genotypes was statistically significant at the 5% level then Duncan’s multiple range test was used to identify significant mean differences between pairs of genotypes. Data generated from the 352 genotypes associated with the BnASYST diversity panel was analysed using linear mixed models. An incremental Al³⁺ resistance index (ARI) was developed within the linear mixed model framework for all traits considered including root length, root DW, shoot DW. An ARI is defined as the deviation from the regression of the Al treatment against the control treatment. Similar indices have been developed and described in weeds research (Lemerle et al. 2006) and in research on the drought tolerance of B. napus genotypes (Raman et al. 2020). For all traits considered non-genetic sources of variation were fitted as random effects and included terms associated with runs, tanks within runs and the layout of tanks within a glasshouse. Variance heterogeneity between runs was considered for all analysed traits. The main effect of Al treatment was fitted as a fixed effect and effects associated with the interaction of Al treatment and genotype as random effects. A broad-sense heritability (H²) for an ARI was estimated using the approach described in Cullis et al. (2006). All linear mixed model analyses were conducted using the asreml statistical software package (Butler et al. 2017) within the R computing environment (https://www.R-project.org/).

Genomic DNA was isolated from leaf tissue collected from 2-week old seedlings of each accession using a modified cetrimonium bromide (CTAB) method (Raman et al. 2009). The BnASSYST panel was genotyped with genotyping-by-sequencing (GBS)-based DArTseq markers (Raman et al. 2014). The DArTseq markers, comprising both in-silico DArT and DArTseq SNP with call frequency ≥ 0.8 and minor allele frequency ≥ 0.05 were selected. In addition, SNPs with 0.25 heterozygous genotypes were discarded. A total of 35 729 high-quality markers were selected and pruned based on the linkage disequilibrium (LD) using the composite haplotype method (r² = 0.5). This has enabled us to exclude redundant SNPs that were in high LD. A subset of 8841 LD pruned markers that are distributed across all 19 chromosomes of B. napus among 344 genotypes were used for genome-wide association analyses. To reduce the false-positive associations, we accounted for both population stratification and kinship matrices, as described previously (Yu et al. 2006; Kang et al. 2008). Population structure was estimated with principal components analysis (PCA) of the marker genotypes (Price et al. 2006) and the first 10 principal components (PC) were used for GWAS. The first four principal components (PC1 to PC4) explained 39% (PC1), 24.6% (PC2), 7.3% (PC3) and 5.8% (PC4) of the genetic variation, respectively. GWAS was conducted using efficient mixed−model association expedited (EMMAX) single-locus (SLMM) and multi-locus mixed model (MLMM) using SNP & Variation Suite ver. 8.6.0 software (Golden Helix, Inc., Bozeman, MT, www.goldenhelix.com) to identify the association between phenotypic trait data (in the Al³⁺ and control treatments) and SNP markers (Kang et al. 2010; Segura et al. 2012). MLMM uses both forward and backward stepwise approaches to select markers as fixed effect covariates as described previously (Segura et al. 2012). Kinship/relatedness estimates were obtained using the identity-by-descent (IBD) method implemented in the same SNP & Variation Suite software. The MLMM association procedure uses pairwise relatedness between samples as a random effect and principal components as a covariate and it was implemented to correct for confounding effects and to control for Type I and Type II errors. The DArTseq markers that showed significant associations with the incremental Al³⁺ resistance indices for root length (ARI-RL), root DW (ARI-RDW) and shoot DW (ARI-SDW) were identified with a probability (P) of ≤0.0001 (−log₁₀P ≥ 4) threshold and corrected with the Bonferroni multiple testing procedure. The SNP associations with Bonferroni correction (P ≤ 0.05), false discovery rate (FDR) ≤ 0.05 and −log₁₀P ≥ 4 were treated as significant markers associated with Al³⁺ resistance traits. To identify loci for genetic interaction, we performed GWAS analysis by assessing the relationship between the dependent variable (e.g. root weight) and the covariate (e.g. root length) using SNP & Variation Suite (SVS) ver. 8.6.0. The Manhattan plots were generated using the SVS software.

To predict candidate genes related to the incremental Al³⁺ resistance traits for root length (ARI-RL), root DW (ARI-RDW) and shoot DW (ARI-SDW) the public reference genome for B. napus cv. Darmor-bzh assembly ver. 4.1 (Chalhoub et al. 2014) was used. The DArTseq marker sequences that showed significant trait-marker associations were searched for against the reference sequence using a local basic local alignment search tool (BLAST) function (Altschul et al. 1990) in the Geneious package (https://www.geneious.com/prime). Putative candidate genes were identified based on their physical distance from the significant SNPs on the reference genome assembly (only hits with high e-values were considered). A previous study determined the LD (r²) in the BnASSYST panel and it was estimated to drop to less than 0.2 within 494 kb for the A subgenome and within 3389 kb for the C subgenome (Schiessl et al. 2015). In this study, the closest B. napus genes from the significant SNPs, as well as those which were localised within 100 kb for the A and C subgenomes, were treated as candidate genes associated with one of the Al³⁺ resistance traits.

Root and shoot growth was reduced as pH decreased as shown in a typical response for ATR-Bonito (Fig. S2a). For all genotypes, including wild radish, the net root growth fell from between 10 and 13 cm at pH 6.2 to almost zero at pH 3.8 (Fig. S2b) indicating they all had a similar sensitivity to low pH.

The relative Al³⁺ resistance of B. napus (cv. Oscar) was then compared with a selection of close relatives including B. rapa, B. juncea, B. carinata, and Raphanus raphanistrum. The Al³⁺ treatment significantly inhibited root growth in all species except for wild radish (Fig. S3a). Relative root length (i.e. Al³⁺ treatment/control) for the Brassica species was reduced to approximately 30% whereas relative root length for wild radish was 81% (Fig. S3b). Al³⁺ treatment also decreased total DW for all genotypes but relative DW for wild radish was significantly different (P < 0.05) and almost two-fold greater than the other species (Fig. S3c, d). These results indicate that wild radish is significantly more resistant to Al³⁺ toxicity than the other species tested.

The effect of seed size on Al³⁺ resistance was investigated by individually weighing the seed of three B. napus genotypes (Oscar, ATR-Bonito and 3674-13) and preparing two groups containing the largest seed and smallest seed of each genotype. For Oscar, the 100 seed weight of the small seed and large seed were 221 mg and 355 mg, for ATR-Bonito it was 404 mg and 625 mg and for 3674-13 it was 200 mg and 343 mg, respectively. Therefore, the large seeds were between 55 and 70% heavier than the small seeds in each case. These seeds were grown with and without 16 μM Al³⁺ (pH 4.4) and root lengths and total DW were measured after 12 days. In the control solution, the plants from the large seed of each genotype were approximately two-fold larger than those from the small seed. Al³⁺ treatment inhibited growth by a similar amount in all cases (Fig. S4a) and relative total DW (Al³⁺/control) was not significantly different between the small and large seed of Oscar and ATR-Bonito. For 3674-13, the relative total DW of plants from the large seed was lower than those from the small seed (Fig. S4b). The large seed generated slightly longer roots than those from the smaller seed in control but Al³⁺ reduced growth by about 50% in all cases and no differences in relative root length were detected between the small and large seed (Fig. S4c, d). These results indicate that seed size influences early seedling vigour but not Al³⁺ resistance.

The Al³⁺ resistance of B. napus and wild radish was directly compared with wheat in two experiments using different Al³⁺ concentrations. The B. napus genotypes included were Oscar, ATR-Bonito and the hybrid GT42 while the wheat genotypes were a pair of near-isogenic lines that vary in Al³⁺ resistance: ET8 (Al³⁺-resistant) and ES8 (Al³⁺-sensitive). Since root growth is a more sensitive indicator of Al³⁺ toxicity in cereals than root dry weight or shoot biomass, the length of the longest root was recorded after 12 days of growth with and without Al³⁺ so that relative root growth could be calculated. Results from both the two experiments were generally consistent with one another. Al³⁺ treatment inhibited root growth in the B. napus genotypes and ES8 wheat whereas root growth was not affected in ET8 wheat and even slightly increased in wild radish (Fig. 1). These results indicate that the response of B. napus to Al³⁺ toxicity is more similar to Al³⁺-sensitive wheat than to Al³⁺-resistant wheat.

Fig. 1.  Direct comparison of the Al³⁺ resistance in B. napus and wheat. The Al³⁺-resistance of a group of B. napus cultivars was compared with the near-isogenic wheat lines ET8 (Al³⁺-resistant) and ES8 (Al³⁺-sensitive) in two experiments. The first experiment (a, b) used 0 and 10 μM AlCl₃ and the second experiment (c, d) used 0, 8 and 16 μM AlCl₃ (pH 4.3). Data show means and s.e. (n = 8). In (b) and (d) relative root length is calculated as [(root length with Al)/(root length in control)] × 100. The hatched columns indicate wheat data. The statistical comparison of percentages used the approach described by Zhou et al. (2013) and data with different letters are significantly different (P < 0.05).

A selection of 352 lines from the BnASSYT diversity panel was evaluated for Al³⁺ resistance in a hydroponic nutrient solution with and without 12 μM AlCl₃ (pH 4.4). Over the nine runs performed for the screen, the Al³⁺ treatment reduced root DW by approximately 35%, shoot DW by 49% and root length by 25% compared with the controls (data not shown, see Fig. S5). The Al³⁺ resistance for each accession was first estimated by dividing the root length, root DW and shoot DW in the Al³⁺ treatment with the control treatment using the raw data (relative to the check genotype included in each tank). The range of these values was approximately 5.0-fold for shoot DW, 3.5-fold for root DW and 1.6-fold for root length (Fig. 2). The results indicate that Al³⁺ stress not only affected root-related traits (root length and root DW), but significantly inhibited shoot growth as well. Indeed, the variation in shoot DW was three-fold greater than for root length, which contrasts with many similar studies in cereals where root growth is typically affected more than shoot growth (Kerridge 1968).

Fig. 2.  Range of Al³⁺ resistance using the raw data. Data show the variation in Al³⁺ resistance among the BnASSYST accessions estimated from the raw data for (a) shoot DW, (b) root DW and (c) root length in the Al³⁺ treatment compared with the control treatment. The data are the average of the three replicates from each treatment and relative to the check accession included in each tank. The variation for shoot DW was ∼5.0-fold, the variation for root DW was 3.5-fold and the variation for root length was 1.6-fold.

To compare the different BnASSYST accessions in a more robust statistical manner for the GWAS analyses, an incremental Al³⁺ resistance index (ARI) was calculated for each accession based on their root length (ARI-RL), root DW (ARI-RDW) and shoot DW (ARI-SDW) in the control and Al³⁺ treatments. The ARI used the ‘check’ accession (cv. Smart) that was included in every tank of each experimental run and accounted for the genetic differences in plant vigour and variation between experimental runs (Lemerle et al. 2006). Pair-wise correlations between three measures of Al³⁺ resistance revealed a strong positive correlation between ARI-SDW and ARI-RDW (r = 0.82), and moderate correlations between ARI-RL and ARI-RDW (r = 0.52) and between ARI-RL and ARI-SDW (r = 0.5) (Fig. S6). All correlations were statistically significant (P < 0.001).

Table 1 lists the accessions that were among the most resistant and most sensitive to Al³⁺ based on ARI-SDW, ARI-RDW and ARI-RL. Many accessions appeared in more than one category of resistance. For instance, of the 40 most resistant accessions listed in each category, 28 appeared in at least two of these categories and 11 accessions appeared in all three. Similarly, of the 30 least resistant accessions in each category, 24 appeared in at least two of these categories and nine accessions appeared in all three. The H² ranged from 51% (for mean ARI-RL and mean ARI-SDW) to 59% (mean ARI-RDW), suggesting that the genotypic variation is sufficient for the identification of genomic loci associated with Al³⁺ resistance.

Table 1.  The most Al³⁺-resistant and least Al³⁺-resistant accessions screened from the diversity panel.

Results of the phenotypic screen were confirmed by comparing some of the most Al³⁺-resistant accessions and least resistant accessions in a separate growth experiment with and without AlCl₃ treatments. Results were consistent with the GWAS screen with the more resistant accessions accumulating almost two-fold more total biomass in the Al³⁺ treatments than the Al³⁺-sensitive accessions (data not shown).

Since the 100-seed weight varied four-fold among the BnASSYST accessions we investigated how this affects early seedling vigour and resistance to Al³⁺ stress. Seed weight showed significant positive correlations with plant growth in the control treatment (minus Al³⁺) where it explained 24% of the variation in shoot DW (P = 0.004), 27% of the variation in root DW (P < 0.001) and 13% of the variation in root length (P < 0.001) (Fig. 3). By contrast, no significant correlations were detected between seed weight and ARI-RL (P = 0.22) or ARI-RDW (P = 0.20). Seed weight was significantly correlated with ARI-SDW (P < 0.001) but it only explained 3.7% of the variation. These results are generally consistent with the results in Fig. S4 which examined the effect of seed weight on plant vigour and Al³⁺ resistance within three B. napus genotypes.

Fig. 3.  Seed size affects early seedling vigour but not Al³⁺ resistance. The 100-grain weight for each line in the diversity panel was plotted against (a) shoot dry weight, (b) root dry weight and (c) root length in the control treatment. Note that these values are presented relative to the check line (cv. Smart) included in every tank. The 100-seed weight was also plotted against the Al³⁺ resistance indices (ARI) for (d) shoot dry weight (SDW), (e) root dry weight (RDW) and (f) root length (RL). Also included are the regression lines and r² values for the correlation.

We utilised BLUPs (best linear unbiased predictions) to identify quantitative trait loci (QTL) associated with ARI-RL, ARI-RDW, and ARI-SDW using two GWAS analysis approaches that account for both population structure and relatedness: MLMM and the SLMM. Using the MLMM approach, we identified a total of five significant DArTseq SNP associations for ARI-RL, ARI-RDW and ARI-SDW with threshold values of −log₁₀(P) ≥ 4 and Bonferroni corrections and FDR of ≤ 0.05 (Tables 2, S1). Among the significant SNPs, two were identified for ARI-RL on chromosomes A04 and A05, one for ARI-RDW on A06, and two for ARI-SDW on chromosomes A02 and C05 (Table 2, Fig. 4). These five significant SNPs each explained between 5.96 and 7.02% of the variation (Table 2). The SLMM based GWAS approach did not identify any significant SNP associations for Al tolerance measures based on ARI. However, nine SNP associations underlying ‘suggestive’ QTL for ARI-RL, ARI-SDW and ARI-RDW were detected on A01, A02, A06, A08, C04, C05, C09, and Cnn-random chromosomes. These ‘suggestive’ QTL were identified with both SLMM and MLMM approaches using slightly less stringent parameters where −log₁₀(P) remained ≥ 4, but the Bonferroni corrections and/or FDR scores had higher thresholds (> 0.05) (Table S1). Interestingly, two significant SNP marker-trait associations for ARI-SDW on A02 and ARI-RDW on A06 (Table 2) co-located with the markers linked with ‘suggestive’ QTL identified using the SLMM approach (Table S1). Finally, two of the suggestive SNP associations identified with MLMM for ARI-RDW on chromosome C05 (delimited by 3183982|F|0-18:T>G-18:T>G) and for ARI-RL on C09 (delimited by marker 3097126|F|0-6:C>A-6:C>A), mapped to ‘suggestive’ QTL using the SLMM approach but with a less stringent threshold (−log₁₀P of ≥ 3.74; Table S1).

Table 2.  Significant DArTseq marker associations using incremental aluminium resistance indices (ARI) for root length (ARI-RL), shoot dry weight (ARI-SDW), and root dry weight (ARI-RDW) in the BnASSYST diversity panel.

Fig. 4.  Manhattan plots showing significant markers for Al³⁺ resistance. Plots show −log₁₀P scores for associations between DArTseq markers and Al³⁺ resistance, evaluated as incremental Al³⁺ resistance indices for (a) the root length (ARI-RL), (b) root DW (ARI-RDW) and (c) shoot DW (ARI-SDW). The x-axes represent most of the B. napus chromosomes and the y-axes are the −log₁₀P scores where the red lines indicate −log₁₀P 4.0. The five markers indicated are significantly associated with their traits because they all have −log₁₀P scores above 4 and Bonferroni corrections and FDR scores of P < 0.05. The red solid line indicates the threshold value for significant SNPs at −log₁₀P ≥ 4.

Fig. 5.  Organic anion efflux from selected accessions of the B. napus diversity panel. Malate and citrate release from B. napus accessions were measured in the absence (3 h) and presence (3 h and 20 h) of 50 μM AlCl₃ in 0.2 mM CaCl₂ (pH 4.3). The Al³⁺-resistant accessions are Daichousen-Nakano (V297), Odin (V328) and Bronze Top (V202) and the other accessions among the least resistant to Al³⁺ are Jantar (V130) and Giant Xr707 (V302). Invigor R5520P is a commercial hybrid cultivar. Data show the mean and s.e. (n = 4–6).

Since Al³⁺ stress affects both the root and shoot growth, we investigated genomic regions potentially involved in root–shoot interactions. Firstly, using SDW (ARI-SDW) as a covariate, GWAS analysis identified one significant QTL for ARI-RDW on chromosome A02, which accounted for 13.15% of the phenotypic variance (Table 2). Another four ‘suggestive’ QTL for interactions were identified on chromosomes A03_random, C02, C05 and C09 (Table S1). When ARI-RL was used as a covariate, no significant interactions with RDW were detected. However, six ‘suggestive’ QTL for interactions were found on A03, A07, C07_random, C09 and Cnn_random chromosomes each accounting for 10.8–12.31% of the genotypic variance (Table S1). We then used ARI-RDW as a covariate to identify QTL for interactions with ARI-SDW and identified one significant QTL on C02, delimited with marker 3156059|F|0-17:T>G-17:T>G to 39.83 Mb sequence which explained 12.95% of the genotypic variance (Table S1). One of these suggestive SNP for ARI-RDW interactions on chromosome C09 (3081487|F|0-26:C>T-26:C>T) was detected when ARI-RL or ARI-SDW were used as covariates (genetic variance, r² = 11.25–12.31%), suggesting that RDW is dependent on both RL and SDW.

Analysis of the control (minus Al³⁺) treatment provided an opportunity to identify QTL for early seedling vigour at low pH. Five significant QTL for RL and RDW in the control treatment were identified on chromosomes A03, A10, Ann-random, C01, and Cnn_random, each accounting for between 6.18 and 8.72% of the genotypic variance (Tables 1, 3). All five of these were identified using the MLMM, while SLMM identified the same SNP on Ann-random (Table 3). An additional 13 ‘suggestive’ QTL for seedling vigour in the control solution were identified on chromosomes A01, A03, A03_random, A04, A06, A09, Ann_random, C01, C02, C04, C06, C07 and Cnn_random (Table S1). Six of those ‘suggestive’ QTL on A01, A03, A04, Ann_random, C06 and a Cnn_random chromosomal region were identified as being ‘suggestive’ with the same markers using the SLMM approach. Our data show that none of the significant QTL identified for traits measured in the control treatment was the same as identified using ARI values (Table S1) which indicates that this study reliably identified QTL for Al³⁺ resistance rather than for other traits related to seedling vigour.

Table 3.  Significant DArTseq marker associations for root length (RL), and root dry weight (RDW) evaluated in the control treatment at low pH but without Al³⁺.

To predict candidate genes related to significant QTL for ARI-RL, ARI-RSW, and ARI-RDW, we identified 40 orthologues of annotated A. thaliana genes within 100 kb of the peak significant SNPs on the reference assembly of the Darmor-bzh genome (Table 1). Five B. napus genes; BnaA02g19610D, BnaA04g10260D, BnaA05g20870D, BnaA06g12540D and BnaC05g33650D were located within 2.6 kb from the significant SNP markers associated with different measures of ARI (Table 2). We further localised the physical map positions of some candidate genes previously found to influence Al³⁺ resistance in Arabidopsis (also in the Brassicaceae) (see Kochian et al. 2015; Nakano et al. 2020). These include members of the Aluminum-activated malate transporter family (AtALMT1, AT1G18420.1; ALMT12, AT4G17970), Regulation of AtALMT1 expression (RAE1, AT5G01720), Multi-drug and toxic compound extrusion (AtMATE; At1g51340), the zinc finger Cys₂His₂ Transcription Factor Sensitive to proton rhizotoxicity 1 (STOP1, At1g34370), NEP-INTERACTING PROTEIN 1 (NIP1, AT2G17750), TON1-RECRUITING MOTIF 28 (AtTRM28) and THIOREDOXIN H-TYPE 1 (AtTRX1), WRKY46 (repressor of Al³⁺ resistance; At2g46400), the activator of ALMT1 expression CALMODULIN-BINDING TRANSCRIPTION ACTIVATOR2 (AT5G64220), GLUCAN SYNTHASE-LIKE 6, (AT1G05570.1), GALACTAN SYNTHASE 1 (GALS1, AT2G33570.1) and the ATP binding cassette-like genes (AtSTAR1 AT1G67940.1, At2g37330, AT5G39040.1). Several genes of interest were identified (Tables 2, 3, S1). For example, the B. napus orthologue to AtALMT1 was found to reside on chromosome A06 within 6 kb of significant SNPs for ARI-RDW (Table S1) and orthologues of the AtMATE on A01 (41.4 kb) and A03 (93.7 kb), the Major transport facilitator (At1g75220.1) on Cnn_random and a STOP1-like gene on C09 were located near significant/suggestive SNP marker associations (Table 1). Furthermore, we located the genes located within 22 kb from the proximal and distal ends of significant SNPs associated with seedling vigour in the control treatment and the putative functions of these genes include cation/H⁺ exchanger 27 (A03), pectin lyase/glycoside hydrolase (A10), d-tagatose-1,6-bisphosphate aldolase and Inositol-tetrakisphosphate 1-kinase (Ann_random); GA20 oxidase (C01) and cytochrome P450 (Cnn_random) (Tables 3, S1).

We performed a preliminary experiment to determine whether organic anion efflux from roots might explain part of the variation in resistance between the BnASSYST accessions. Malate and citrate efflux was measured in a selection of the most resistant accessions (Daichousen-Nakano, Odin, Bronze Top) and least resistant accessions (Jantar, Giant Xr707, Table 1). A current commercial B. napus hybrid of Australian origin (variety Invigor 5520R) was also included as a comparison. The results indicated that Al³⁺ significantly increased the release of malate and citrate from all accessions, including the hybrid variety, with malate efflux being at least 10-fold greater than the citrate efflux (Fig. 5). Two resistant accessions Daichousen-Nakano and Odin showed the greatest malate efflux and Odin also showed the greatest citrate efflux. However, the other resistant accession, Bronze Top, released much less malate and a similar citrate efflux as the sensitive accessions. While no consistent differences were detected between these Al³⁺-resistant and sensitive BnASSYST accessions, the results indicate that organic anion efflux still could account for part of the difference in Al³⁺ resistance.

The present study confirmed that B. napus is sensitive to both low pH and Al³⁺ toxicity, the major factors limiting growth on acidic soils. We demonstrated that Al³⁺ toxicity inhibited the growth of B. napus plants by a similar degree to Al³⁺-sensitive wheat, which provides evidence for the need to increase the Al³⁺ resistance of commercial germplasm if sufficient variation is available. We compared B. napus with several of its close relatives in the Brassicaceae and found they were all similarly sensitive to Al³⁺ except for wild radish. Wild radish was substantially more resistant to Al³⁺ toxicity but similarly sensitive to low pH. Indeed, wild radish showed a slight increase in root growth with Al³⁺ treatment. This response has been reported previously (Kinraide and Parker 1990) and interpreted as being caused by low concentrations of Al³⁺ ameliorating the stress associated with the highly toxic H⁺ by displacing it from the root membrane surface (Kinraide et al. 1992).

The genotypic variation of Al³⁺ resistance in B. napus was assessed by measuring root length (RL), root dry weight (RDW) and shoot dry weight (SDW) in 352 diverse BnASSYST accessions grown in hydroponics with and without Al³⁺. The incremental Al³⁺ resistance indices (ARI) for each accession accounted for the inherent variation in plant vigour. The same approach was utilised earlier to describe variation in weed competitiveness and drought tolerance (Lemerle et al. 2006). Significant variation in the ARI-RL, ARI-RDW and ARI-SDW was detected among the accessions and some individuals consistently ranked among the most resistant in all three ARI measures. The ARI measures displayed a moderate broad-sense heritability indicating that these traits could be targeted by breeders to improve the acid soil tolerance of canola and related species. Interestingly, ARI-RDW showed a stronger correlation with ARI-SDW (r = 0.82) than with ARI-RL (r = 0.52), which could be related to the common finding that Al³⁺ toxicity not only inhibits root growth but makes the roots thicker and stubbier (Foy 1984).

GWAS analysis with a Multi-locus Mixed Model-based method identified at least five significant QTL for the three ARI measures on chromosomes A02, A04 and A05 A06, and C05 and identified significant interactions on A02 and C02 (Table 2). Our findings overlap little with those of Gao et al. (2021) who screened for Al³⁺ resistance during germination. They reported eight QTL associated with relative root length on chromosomes A03, A07, A09, A10, C05, C06, and C09, and five QTL for root dry weight on chromosomes A03, A04, A10, C05, and C07. Only two genomic regions on A04 (∼9.1 Mb) and A10 (15.9 Mb) are potentially similar to the ones identified in our study. It is unsurprising the results differ because Gao et al. (2021) screened younger seedlings in Petri dishes and so the growth conditions were very different. They also screened half the number of accessions and used less stringent parameters in their GWAS analysis. A diverse and unstructured population (used herein) is essential for a GWAS panel to reveal the genetic control/architecture of trait variation.

We found that some of the significant SNP detected with one of the mixed model methods were also ‘suggestive’ with the second method. For example, a significant marker for ARI-RDW identified using a MLMM (3131293|F|0-23:A>G-23:A>G; Cnn_random, 14 827 173 bp) was a ‘suggestive’ QTL using SLMM (Table S1). Similar observations were made for ARI-SDW on A02 (12 181 001 bp), ARI-RDW on A06 (6 515 485 bp) and RL (minus Al) on C01 (11 877 612 bp) (Table S1). Our results indicate that these ‘suggestive’ SNP associations should not be ignored because they could well be contributing significantly to trait variation.

While this study did not specifically examine the variation in sensitivity to low pH, we could compare the growth of the accessions in the control treatment (low pH without Al³⁺), to identify genetic markers linked with early seedling vigour in these conditions. At least five significant SNP associations with the shoot and root traits were identified, none of which overlapped with the QTL associated with Al³⁺ resistance. These results suggest that the genetic architecture for Al³⁺ resistance and low pH tolerance in B. napus is distinct and complex. Similar conclusions that tolerance to low pH and resistance to Al³⁺ stress are controlled by distinct genomic regions have been made for rice bean (Vigna umbellate), Arabidopsis and barley (Fan et al. 2015; Nakano et al. 2020; Szurman-Zubrzycka et al. 2021). Nakano et al. (2020) extended their investigation in Arabidopsis by using reverse genetics to confirm that some genes from the genetic regions they identified did influence aluminium resistance. Among these were novel resistance genes such as TON1-RECRUITING MOTIF 28 (AtTRM28) and THIOREDOXIN H-TYPE 1 (AtTRX1), as well as known resistance genes including the aluminum-activated malate transporter, AtALMT1.

GWAS analysis enabled us to identify candidate resistance genes by examining those localised near significant SNP associations for Al³⁺ resistance (Tables 2, 3). We searched for candidate genes within 100 kb from the significant SNPs (50 kb upstream and 50 kb downstream region). However, the strong linkage disequilibrium (LD) in the BnASSYST accessions reported on the A subgenome of ∼400 kb and the C subgenome of ∼1000 kb (Schiessl et al. 2015) means the genes responsible for the trait variation may be located farther away from the significant SNPs listed here. Nevertheless, several of the significant markers were located in the vicinity of genes known to be linked with Al³⁺ toxicity and resistance in other species, especially those that affect the cell wall and plasma membrane. Some of these genes include multicopper oxidase, aquaporin like genes, callose synthase 1, cation/H⁺ exchanger 27, cell wall receptor kinase, methyltransferase, pectin lyase, lactate/malate/succinate/fumarate dehydrogenase, inositol tetrakisphosphate 1 kinase and glutathione S transferase (Kochian et al. 2005, 2015).

Notably, the orthologue of AtALMT1 (BnaA06g12560D) was located 5906 bp from a significant SNP marker on chromosome A06 and AtMATE genes were located within 100 kb of a marker on A03 (Table S1). These genes control malate and citrate exudation from Arabidopsis (Li et al. 2009) and similar mechanisms operate in several other crop species (Ma et al. 2001; Delhaize et al. 2007; Furukawa et al. 2007; Magalhaes et al. 2007; Wang et al. 2007; Ryan et al. 2009). A preliminary experiment compared some of the most resistant and least resistant accessions to determine whether differences in organic anion release could explain the variation but the results were equivocal. We can confirm that Al³⁺ treatment does increase malate and citrate release from B. napus as shown previously (Ligaba et al. 2006) and that two of the resistant accessions released more malate than the sensitive accessions. However, the malate efflux and citrate efflux from a third resistant accession tested were similar to the sensitive accessions. It remains possible that organic anion efflux is one of several resistance mechanisms operating in B. napus. Further work is required to compare the organic anion release from the resistant and sensitive accessions in more detail.

GWAS analysis using the control data identified significant QTL for early seedling vigour in these low pH growth conditions. The genes that mapped close to the associated markers include a few involved in plant development and abiotic stress including BnaC01g17320D (ARABIDOPSIS THALIANA GIBBERELLIN 20-OXIDASE 1), a gene involved in the gibberellin biosynthesis pathway.

Interestingly, we found that the shoots of B. napus were quite sensitive to Al³⁺ toxicity over the relatively short screening period of 12 days. This contrasts with many other species, especially cereals, where root growth is far more sensitive to Al³⁺ stress than shoot growth. Certainly, root elongation in wheat and maize can begin to decrease within minutes of exposure to Al³⁺ (Ryan et al. 1992) whereas for phenotypes to develop in the shoots typically takes days and these are likely associated with nutrient imbalances (Kerridge 1968). This is the reason why most screens for Al³⁺ resistance examine root growth (length or biomass) or use stains to detect Al³⁺ accumulation (hematoxylin and erichrome cyanine), callose production or hypersensitivity responses in those tissues (Polle et al. 1978; Massot et al. 1999; Wang et al. 2006a, 2006b). The sensitivity of B. napus shoots to Al³⁺ might be related to their smaller seed. Since Al³⁺ is known, directly or indirectly, to inhibit the uptake of many nutrients including calcium, magnesium, potassium and phosphorus (Kochian et al. 2005), B. napus seedlings might deplete the nutrient reserves in their seed more rapidly and require functioning roots to take up nutrients sooner than species with larger seeds. The present study found that seed size was positively correlated with seedling vigour in control solution. Indeed, seed size explained 27% of the variation in RDW among the accessions in the controls. However, no association was detected between seed size and Al³⁺ resistance using any of the ARI values demonstrating that seed size does not affect Al³+ resistance.

In conclusion, the genetic variation for Al³⁺ resistance among a diverse set of B. napus germplasm is substantial and likely sufficient for breeding purposes. We identified five genomic regions for Al³⁺ resistance explaining up to 13% of the variation and shortlisted candidate genes near the significant SNPs. Further studies wBus accessions and DArTseq SNP markers) would enable canola breeders to improve Al³⁺ resistance of canola cultivars via genomic and marker-assisted selection. Improving the acid soil tolerance of canola would expand its area of cultivation and increase its usefulness as a rotation crop with cereals and pulses.

Supplementary material is available online.

The authors will provide data as requested.

The authors declare no conflicts of interest.

The authors acknowledge the China Scholarship Council for supporting the visit of Hanmei Du to CSIRO in Canberra. Other activities were supported by internal CSIRO Agriculture and Food funding.