The number of Al-regulated genes (differential expression between Al-treated and untreated plants) is over 1000 when cultivars are analyzed separately. However, only 263 genes are differentially expressed in all four cultivars. ANOVA analyses with stringent p values retained 83 of these genes as highly significantly regulated (Table 1; additional file 1). Among these genes, 20 were previously identified as up-regulated by Al (references in Table 1). We randomly selected 4 of the stress candidate genes for qRT-PCR validation (see bold probeset IDs in Table 1). Two of them (genes # 2 and 3) are strongly over-expressed in response to Al, one (gene # 5, oxalate oxidase) was previously shown to be up-regulated by Al and the fourth one (gene # 49) was represented by 5 different genes (genes # 49, 61, 73, 74, and 76).

As found for stress-associated genes, microarray analyses allowed the identification of more than 1000 genes potentially associated with Al tolerance when selection was performed using only one pair of wheat lines (differential expression between Atlas66/Bounty or Century-T/Century-S exposed to Al) but this number was reduced to 69 common genes when both pairs were considered. This result suggests that a large number of genes are associated with cultivar specific responses and only a limited number of them are associated with Al tolerance. This list was reduced to 25 genes using ANOVA analyses with stringent p values (Table 2; additional file 2). These genes were subdivided into constitutively or differentially expressed genes based on the difference in expression between Al-treated and non-treated control plants. Table 2A lists the 6 constitutively expressed genes classified in decreasing order of expression (average fold expression of the two tolerant compared to the two sensitive cultivars). The most strongly expressed gene is the previously identified Al malate transporter ALMT-1 33. We selected two of these genes for qRT-PCR validation (gene # 84 encodes ALMT-1 and is more expressed in tolerant cultivars while gene # 89 encodes a protein containing an F-box domain and is more expressed in sensitive cultivars; Table 2A). Table 2B lists the 19 genes that are regulated by Al and differentially expressed between the two tolerant and the two sensitive cultivars. Four of these genes were selected for validation by qRT-PCR (see bold probeset IDs in Table 2B).

We selected ten genes for validation of the microarray data by qRT-PCR (4 among the stress-associated candidates and 6 among the tolerance-associated candidates, as indicated above). Four different Al concentrations (0, 5, 50 and 250 μM Al) were used to determine the effect on gene expression. Four new replicates of each Al exposure conditions using all wheat lines were prepared. The values indicated in each histogram column represent the expression levels relative to the first column (cultivar Atlas66 not exposed to Al) (Fig. 1A–J). The level of expression of the tested genes under control conditions is generally similar in the tolerant and sensitive cultivars.

Genes were classified as associated with Al tolerance when there was a significant difference in expression in both tolerant cultivars (Atlas66 and Century-T exposed to 50 μM Al) compared to their respective sensitive counterparts (Bounty and Century-S exposed to 5 μM Al). These conditions allow the detection of either 1) Al-regulated genes that may be expressed differentially in tolerant and sensitive lines, or 2) genes not regulated by Al (constitutively expressed) but showing constitutive difference in expression between tolerant and sensitive lines. Once classified as associated with stress or tolerance, gene regulation by Al was confirmed by comparing the Al-treated and non-treated controls. qRT-PCR, analyses confirmed that three (Ta.192.1.S1_at, Ta.12921.1.S1_x_at and Ta.392.2.S1_at, Fig. 1A–C) of the four stress candidate genes are stress-associated while the 4^th gene (Ta.5557.1.S1_x_at, Fig. 1D) is more closely associated with Al tolerance. Among the six selected tolerance candidates genes, four were confirmed as associated with Al tolerance (Fig. 1E–H: Ta.30659.1.S1_at, TaAffx.16664.1.A1_at, Ta23271.1.S1_s_at, Ta.8545.1.S1_x_at) while the last two genes (Ta.21314.1.S1_x_at and Ta.24632.1.S1_at, Fig. 1I–J), were not clearly linked to either stress or tolerance.

The correlation between gene expression data obtained by microarray and qRT-PCR is presented in Fig. 2. Al-regulated genes are shown in Fig. 2A for the cultivars Atlas66 and Bounty. The fold change in gene expression (Al-treated VS control) obtained using microarray experiments compared to the fold change obtained using qRT-PCR gives a correlation coefficient of 0.61. A similar analysis was performed to compare the change in gene expression between the tolerant and sensitive cultivars (Atlas66 50 μM Al/Bounty 5 μM Al and Century-T 50 μM Al/Century-S 5 μM Al) calculated from the microarray or qRT-PCR data (Fig. 2B). In this case, a correlation coefficient of 0.77 was obtained between the two experimental methods. These correlation coefficients are very good considering that microarray data are semi-quantitative and subject to error for multigene families where different transcripts could hybridize to similar probes on the array. qRT-PCR data are more specific since the amplification of single transcripts are confirmed by melting curves and gel analyses. The Ta.5557.1.S1_x_at was initially selected as a stress-associated gene by microarray profiling (gene no 5 in Table 1) while it was reclassified as associated with Al tolerance using qRT-PCR (Fig. 1D). This may be related to the presence of multiple oxalate oxidase genes in wheat. Based on these results, we conclude that most of the genes identified in Table 1 and Table 2 are associated with Al stress and tolerance. However, real-time PCR analyses will be required to validate the other genes that were identified using microarrays.

We have previously shown that Al inhibits a redox reaction associated with root growth and Al tolerance 34. The perturbation of this redox reaction by Al could lead to the accumulation of different reactive oxygen species which can stimulate a redox signalling pathway and increase the expression of antioxidant enzymes 35. The accumulation of reactive oxygen species during Al exposure was observed in maize, soybean and wheat 363738. Al was also shown to affect mitochondrial functions leading to ROS production 39. Table 1 shows that the genes encoding WCI-5 (gene # 2), WAS-2 (genes # 10 and 23) and the EF-hand Ca²⁺-binding protein (genes # 40, 41, 43, 54 and 55) are up-regulated during Al exposure. In Arabidopsis, the WCI-5 and WAS-2 genes were shown to be induced via a MAP kinase pathway associated with the pathogen response 40. In wheat seedlings, the WCI-5 gene was shown to be induced by pathogen infection 41 and mediated by the EF-hand Ca²⁺-binding protein CCD1 42. These results suggest that a MAP kinase pathway is stimulated by Al in wheat roots as observed in cell suspension cultures of Coffea arabica L. 43. Three different protein kinase-like genes (genes # 37, 60, 63) are up-regulated during Al exposure suggesting that different signal transduction pathways may be activated. We identified two other potential signalling genes associated with Al tolerance. The first one encodes a putative C2H2 zinc finger transcription factor (gene # 107) that has a 33 aa region sharing 50% identity with the STOP1 C2H2 zinc finger protein required for Al³⁺ induction of ALMT1 in Arabidopsis 17. The other gene encodes an F-box containing protein (gene # 89), which is member of a family known for its involvement in the controlled degradation of target proteins. An Arabidopsis F-box protein homologous to a bacterial redox sensor was proposed to be involved in the cell wall receptor-like associated kinase (WAK1) signalling pathway 44. Since WAK1 was shown to be transiently overexpressed during Al exposure and associated with Al tolerance 27, the reduced expression of an F-Box protein may stimulate this pathway by allowing a stronger expression of WAK1 in tolerant cultivars. The TIR1 gene is another F-Box protein in Arabidopsis that interacts with different proteins in the SCF^TIR1 complex to mediate responses to auxin 4546. These two genes (C2H2 zinc finger and F-Box) require further characterization and functional analysis to understand their roles in the response to Al.

The identification of stress-regulated genes provide new tools to reduce Al stress, as shown by the ectopic overexpression of some Al stress-regulated genes 23242526. Among the 83 candidate genes regulated by Al stress (Table 1), several could play a role in alleviating phosphate deficiency and provide energy to fight oxidative stress. Nutrient deficiency, and especially phosphate, occurs in the presence of Al due to the precipitation of Al-phosphate 47. Under phosphate deficiency, several pathways are hindered due to the reduced availability of ATP and related nucleoside phosphates. These imbalances lead to the induction of alternative pathways of glycolysis to maintain energy and carbon skeletons for key metabolic processes 48. The fructose-1,6-bisphosphatase (gene # 101) associated with Al tolerance (Table 2) is normally inhibited by phosphate but is stimulated under low phosphate availability (Fig. 3, Box 3). This activity produces fructose-6-P and liberates Pi. The glycolytic alternative enzyme pyrophosphate phosphofructokinase is stimulated by low phosphate and can resynthesise fructose-1,6-bisphosphate and liberate more Pi. This apparently futile cycle is useful to conserve ATP and recycle Pi that is readily available in the pyrophosphate molecule 48. Furthermore, since sucrose is the most abundant photosynthate available in the phloem sap, fructose (from sucrose) is readily available and will enter the glycolytic pathway at the level of fructose-6-P thereby maintaining a stable supply of this metabolite. The high availability of fructose-6-P (and other hexose-P) can provide sufficient metabolites to support both the pentose phosphate pathway and organic acid synthesis (Fig. 3, Box 3). Al-tolerant rye was found to maintain a higher content of glucose-6-phosphate than Al-sensitive wheat in response to Al exposure 49. This metabolite is an important precursor of the pentose phosphate pathway (with glucose-6-phosphate dehydrogenase (G6PDH) as the first enzyme) and is an important source of NADPH production. The importance of energy metabolism in modulating Al tolerance is exemplified in Pseudomonas fluorescens where G6PDH was shown to play a pivotal role in enhancing NADPH production and protecting against ROS toxicity 50. The up-regulation of pyruvate dehydrogenase (gene # 65) and alternative oxidase (gene # 72) supports the involvement of an alternative pathway to maintain energy production (Fig. 3, Box 3) 48. Members of the alternative oxidase gene family were found to be highly responsive to oxidative stress and to reduce mitochondrial ROS production while maintaining NADH supply 5152. In Al tolerant plants, a higher NADH supply (and NADPH through the pentose phosphate pathway) would provide reducing power to regenerate ascorbate and glutathione to fight oxidative stress (Fig. 3, Box 3).

Exudation of organic acids has received much attention as an Al tolerance (resistance) mechanism. The secretion of malate or other organic acids (citrate, oxalate) can chelate Al in the rhizosphere and has been associated with Al tolerance in several species 7891033. Enzymes involved in malate synthesis (PEP carboxylase and malate dehydrogenase, Fig. 3 Box 3) were shown to be stimulated during Al stress 5354. These enzymes are also stimulated during Pi deficiency where malate is known to mobilize external Pi 48. These two actions are complementary since the chelation of Al will also free Pi from precipitated Al-phosphate in the rhizosphere. Some of the genes identified in this study may complement the ALMT1 malate transporter to improve Al exclusion. An interesting possibility is that organic acids, such as malate or oxalate, could form intracellular Al-bound complexes that are transported outside the cell (Fig. 3, Box 3). It is possible that the ALMT1 malate transporter (gene # 84) associated with Al tolerance can transport free malate but that it would require a co-factor to transport the larger Al-bound malate complex. Oxalate is another organic acid known to be a strong Al chelator associated with Al tolerance in Taro 55. This organic acid is another potential molecule that could form an Al complex transported out of the cell by an anion transporter such as ALMT1 or the ABC transporter (gene # 22) that is up-regulated during Al exposure.

Several proteins encoded by genes that are up-regulated during Al exposure have a high content of cysteine (Fig. 3, Box 2). The high amount of thiol groups in these proteins may help protect enzymes containing thiol sensitive sites by providing new targets of oxidation. The presence of 11 cysteine residues in the small protein (128 aa) encoded by gene # 90 may explain why this gene is associated with Al tolerance (Fig. 3, Box 2). The up-regulation of four different GSTs (genes # 1, 70, 91, 96), containing cysteine residues in their active sites, indicate that these enzyme may be inactivated by oxidative stress. The synthesis of new enzyme molecules during Al exposure may partially compensate for GST inactivation. On the other hand, two of the four GSTs are associated with Al tolerance (genes # 91, 96). These GSTs may help to reduce oxidative stress as suggested for parB, a tobacco GST isoform 25.

The ability to maintain root elongation under Al stress may require complementary mechanisms such as Al sequestering (outside the cell, or Al exclusion) and increased expression of genes involved in detoxification or root growth. We identified several genes that could play a role in maintaining root growth. Different enzymes such as sugar hydrolases and transferases may participate in cell wall remodelling/loosening in the elongation zone (genes # 32, 36, 46, 77). During the cell wall loosening process, ascorbate provides an electron to reduce Cu²⁺ to Cu⁺ (Fig. 3, Box 1) 5657, which participates in the Fenton reaction with H₂O₂ to ultimately produce •OH radicals. The non-enzymatic cleavage of cell wall polysaccharides thus requires the presence of three basic molecules: ascorbate, Cu⁺² and H₂O₂ (grey boxes in Fig. 3) 5657. The enzymatic reactions required to maintain/regenerate ascorbate and provide H₂O₂ are essential components to maintain root growth and are likely associated with Al tolerance. The redox activity previously shown to be associated with Al tolerance 34 may represent the transfer of electrons required to regenerate the apoplasmic ascorbate. The great avidity of Al³⁺ for electrons (Fig. 3, Box 1) could prevent the regeneration of apoplasmic ascorbate and inhibit root growth. Oxalate oxidase may help maintain the production of H₂O₂ for the Fenton reaction (Fig. 3). However, H₂O₂ could become a substrate for cell wall peroxidases which are not inhibited when ascorbate levels are reduced 58. The inhibition of peroxidase activity was associated with Al tolerance 5960 and may help maintain H₂O₂ levels needed for the non-enzymatic wall loosening process. In our study, three peroxidases (genes # 13, 21, 58) are up-regulated and could be related to the reduced growth rate under our experimental conditions. Transcripts encoding blue copper binding proteins (BCBP) (gene # 25) is possibly up-regulated to stimulate the Cu²⁺-dependent Fenton reaction needed for cell wall loosening. When apoplasmic ascorbate loses an electron, it is transformed into monodehydroascorbate/dehydroascorbate (MDHA/DHA) and new electrons are needed to regenerate ascorbate and avoid degradation into oxalate (Fig. 3, Box 1). The up-regulation of an ascorbate dependent oxidoreductase (gene # 24) could be associated with the regeneration of apoplasmic ascorbate (Fig. 3, Box 3). Impaired ascorbate metabolism was proposed to be involved in the reduction of root growth in squash roots exposed to Al 38 while a higher level of ascorbate and gluthathione was shown to be associated with Al tolerance in tobacco 61. Furthermore, feeding with D-glucose or L-galactono-γ-lactone to enhance ascorbate levels was able to improve Al tolerance in rice 62. These results indicate that maintaining a high ascorbate level is an essential aspect of Al tolerance. Gene # 30 annotated as an arabinonolactone oxidase is homologous to galactonolactone oxidase involved in ascorbate synthesis (Fig. 3, Box 3). This last gene participates in the regeneration of cytoplasmic ascorbate while providing additional NADH (from glucose, Fig. 3, Box 3). The arabinonolactone oxidase is known to synthesize erythroascorbate in yeast. Interestingly, the overexpression of the yeast arabinonolactone oxidase enzyme in rice improves Al tolerance 62.

Two wheat cultivars with a high tolerance to Al (Triticum aestivum L. cv. Atlas66 and the near isogenic line OK91G106, named Century-T in this work; 32) and two wheat cultivars with low tolerance to Al (T. aestivum L. cv. Bounty and the near isogenic line OK91G108, named Century-S in this work; 32) were grown as previously described, and treated under conditions where Al remains mostly in the Al³⁺ form 19. To reduce pH variations and ensure that Al speciation was stable throughout the experiment, at least 100 ml of solution was used for each plant. The root growth inhibition (RGI) is expressed as 100 × [1-(root growth of Al-treated seedling divided by the root growth of control seedlings)]. Replicate experiments were performed on different days with one series of Al concentration per day per cultivar in order to rapidly collect root tips.

Root tips (5–10 mm) were isolated after 24 hours of exposure to Al and frozen on dry ice. Total RNA was isolated from the root tips (5 mm long) of 50 plants collected from the various genotypes using the RNeasy Plant Mini Kit (Qiagen). The RNA quality was assessed on agarose gels and with the Bioanalyzer 2100 (Agilent). Microarray profiling was performed according to Affymetrix protocols at the Functional Genomics Platform of McGill University and Génome Québec Innovation Centre using the Affymetrix GeneChip^® Wheat Genome Array. The microarray results were deposited in the database of ArrayExpress under the accession number E-TABM-454. Three biological replicates of the different lines were treated with Al concentrations resulting in 50% RGI: 50 μM for the tolerant cultivars Atlas66 and Century-T and 5 μM for the sensitive cultivars Bounty and Century-S. The wheat NILs Century-T and Century-S were used only to prepare the Al-treated plants due to the limited number of seeds available. In this case, the average value obtained for the untreated (not exposed to Al) Atlas66 and Bounty was used as the reference.

The level of expression of selected genes was validated by Real-Time PCR. Probes were designed using the primer3 software and primer specificity was verified on agarose gels and by T_m measurements at the end of the Real-Time PCR reactions. Total RNA was prepared (RNeasy Plant Mini Kit; Qiagen) from the root tips (5 mm long) of 50 plants collected from the various genotypes and reverse transcribed using SuperScript II reverse transcriptase and random hexamers (SuperSript™ First-Strand; Invitrogen). Quantitative Real-Time was performed on an ABI 7000 Real Time Cycler using the specific primers described in additional file 3. One-tenth dilutions of the cDNAs were used as template for the qRT-PCR in a total volume of 25 μL as follows: 12.5 μL SYBR Green (Platinum^® SYBR^® Green qPCR SuperMix-UDG, Invitrogen) 1.0 μL primer mix (50:50 mix of forward and reverse primers at 10 pmol/μL each) and 2 μL template. The reaction conditions were: 10 min at 95°C followed by 40 cycles of 1 min at 95°C and 30 s at 60°C.

The microarray data were analyzed using the robust multi-array average (RMA) software (RMA version 0.2) of background adjusted, normalized, and log transformed perfect match values 63. A two-fold cut-off value (log₂ ≥ 1) was arbitrarily set to indicate differential gene expression between two samples. Genes showing a differential expression greater than two-fold (RMA differential expression of log₂ ≥ 1) between the Atlas66 and Bounty controls were excluded from the reference set as these may represent inherited cultivar basal expression levels that could bias the analysis. Overall, this excluded 1.26% of the genes on the microarray (773 of 55,052) demonstrating that most genes are expressed at a similar level between the two cultivars. Genes that are differentially expressed to the same extent in all four cultivars exposed to Al (giving 50% RGI) compared to the non-treated cultivars were selected and classified as candidates for stress-associated genes. The differentially expressed genes between the two tolerant cultivars exposed to Al (Atlas66 and Century-T) and their respective sensitive counterparts (Bounty and Century-S) exposed to Al concentrations resulting in 50% RGI were classified as candidate genes associated with Al tolerance. These genes were subdivided in two groups (constitutively expressed or Al-regulated) based on the average level of regulation in the two tolerant cultivars. Genes were considered constitutively expressed when the differential signal between Al-treated and non-treated samples was less than two-fold (log₂ < 1). An analysis of variance was performed using GraphPad InStat 3 to select genes that are differentially expressed under the conditions specified for each analysis.

For Quantitative Real-Time PCR analyses, the amplification efficiency (90% to 100%) for the different primer sets was determined by amplification of cDNA dilution series using 80, 20, 10, 5, 2.5, and 1.25 ng per reaction (data not shown). The variance (standard error) was very small between the PCR replicates for a same biological sample (Real-Time experimental replicate) compared to the variance between the different biological replicates. Calculations and statistical analyses were performed by ANOVA on the mean ΔC_T (C_T of each gene – C_T of 18S RNA used as load control) of different biological replicates, as described in the Results section.