Aluminum tolerance mechanisms in Kenyan maize germplasm are independent from the citrate transporter ZmMATE1

Aluminum (Al) toxicity on acid soils adversely affects maize yields, which can be overcome by combining soil amendments with genetic tolerance. In maize, ZmMATE1 confers Al tolerance via Al-activated citrate release, whereby citrate forms non-toxic complexes with Al3+ in the rhizosphere. Here, we investigated Al tolerance mechanisms in maize germplasm originated from Kenya based on quantitative trait loci (QTL) mapping. Five QTLs and four epistatic interactions explained ~51% of the phenotypic variation for Al tolerance. The lack of Al tolerance QTL on chromosome 6 and the much lower expression of ZmMATE1 in both Kenyan lines than in Cateto Al237, which donates the superior allele of ZmMATE1, strongly indicate that this gene does not play a significant role in Al tolerance in neither parent. In turn, maize homologs to genes previously implicated in Al tolerance in other species, ZmNrat1, ZmMATE3, ZmWRKY and ZmART1, co-localized with Al tolerance QTL and were more highly expressed in the parent that donate favorable QTL alleles. However, these candidate genes will require further studies for functional validation on maize Al tolerance. The existence of Al tolerance mechanisms independent from ZmMATE1 suggests it is possible to develop highly Al tolerant cultivars by pyramiding complementary Al tolerance genes in maize.


Results
Variability of aluminum tolerance in the Kenyan maize population. Significant genotypic variation was detected for Al tolerance in F 2:3 progeny derived from 203B-14 (Al tolerant) crossed with SCH3 (Al sensitive) based on relative net root growth (RNRG) (Supplementary Table S1). The heritability for Al tolerance based on family means was high (0.97), and the coefficient of experimental variation was low (8.82%) (Supplementary  Table S1), indicating high quality of the phenotypic data.
The RNRG population mean was 35.2% (Supplementary Table S1) with minimum and maximum values of 17.6 and 68.3%, respectively (Fig. 1). The RNRG mean of the Al-tolerant parent 203B-14, 107.7%, was similar to the Brazilian Al-tolerant maize line, Cateto Al237 (95.2%), but higher than the most Al-tolerant F 2:3 progeny (Fig. 1). On the other hand, the Al-sensitive parent, SCH3, was less sensitive to Al than the most Al-sensitive F 2:3 progeny and the Brazilian Al-sensitive line L53 (Fig. 1).
Al tolerance QtLs. Out of the 198 markers genotyped in the population (183 SNPs, 14 SSRs and the ZmNrat1), 122 Mendelian markers (i.e. single locus segregation frequencies as expected for an F 2 population) were used in the linkage map, covering 1,192.2 cM of the ten maize chromosomes (Fig. 2). Chromosome 1 had the highest number of markers (22) spanning 227.6 cM, whereas chromosome 7 had only seven markers along 87.1 cM. Four gaps ranging from 35 to 42 cM were present on chromosomes 2, 4, 7 and 10. The order of the markers flanking these gaps was confirmed by their physical positions according to MaizeGDB.
Epistatic interactions between QTLs explained from 3.3 to 7.0% of the phenotypic variance for RNRG (Table 1), which was similar to the individual QTL main effects. In both epistatic interactions involving qALT8.05 (qALT8.05 × qALT1.09 and qALT8.05 × qALT9.01), F 2:3 progeny homozygous for the Al-sensitive (SCH3) allele at qALT8.05 showed the highest RNRG mean (blue dots, Fig. 3a,b). For the interaction qALT10.02 × qALT1.09, the most Al tolerant progeny were homozygous for alleles donated by the tolerant parent at both QTLs (red Figure 1. Distribution of aluminum tolerance based on relative net root growth (RNRG %) in Kenyan maize progeny. The population was composed by 180 F 2:3 progeny derived from SCH3 (Al-sensitive) × 203B-14 (Altolerant). RNRG means of Kenyan parents are shown by thick arrows, whereas that of the Brazilian standards for Al sensitivity and Al tolerance, L53 and Cateto Al237, respectively, are depicted by thin arrows.    Fig. 3c). The interaction between qALT9.01 × qALT1.09 showed that most of the Al sensitive progeny carried the unfavorable allele at qALT9.01 (Fig. 3d). expression pattern of ZmMATE1. The lack of an Al tolerance QTL at bin 6.00 suggests that ZmMATE1 does not play a role in Al tolerance in the 203B-14 x SCH3. Then, the time course expression of ZmMATE1 in these parents was compared to the Al-tolerant Cateto Al237 and the Al-sensitive L53. The expression of ZmMATE1 in the root tips of both Kenyan lines was much lower than in Cateto Al237, which donates the superior allele of ZmMATE1, and was even lower when compared to the Al-sensitive line from Brazil, L53 (Fig. 4). The time course expression of ZmMATE1 in the Brazilian maize lines was in agreement with Maron et al. 16 .
Candidate genes co-localized with Al tolerance QTLs. The number of predicted genes within each Al tolerance QTL ranged from 69 to 447 (Table 2), which were described in the Supplementary Table S2. We selected  www.nature.com/scientificreports www.nature.com/scientificreports/ one to two candidates within each QTL interval based on similarity with other genes previously characterized for Al tolerance in other species. No previously characterized Al tolerance genes were found within the qALT1.09 and qALT9.01 intervals, suggesting that new candidate genes underlie these QTLs.
The qALT5.03 interval extended from 59.3 to 74.6 Mb on chromosome 5, and harbored two candidate genes, GRMZM2G065154 at 71.7 Mb and GRMZM2G168747 at 74.6 Mb ( Table 2). GRMZM2G065154 encodes a MATE member (ZmMATE3) that shared 21.9% of sequence identity with ZmMATE1 (Supplementary Table S3), and clustered together with other citrate transporters associated with Al tolerance in several plant species (Fig. 5a). GRMZM2G168747 (ZmNrat1) had 81.8% sequence identity with OsNrat1 ( Fig. 5b and   www.nature.com/scientificreports www.nature.com/scientificreports/ GRMZM2G068710 is located at 10.1 Mb on chromosome 10, which is coincident with the qALT10.02 interval (9.7-13.5 Mb). GRMZM2G068710 is 59.4% similar to OsART1 ( Fig. 5d and Supplementary Table S3), a transcriptional regulator of Al tolerance genes in rice 24 , and is predicted to contain C2H2 zinc finger motif. expression pattern of candidate genes in the parental lines. Temporal expression of candidate genes co-localized with Al tolerance QTLs was assessed in the root tips of the Al-tolerant and Al-sensitive parents, 203B-14 and SCH3, respectively. ZmMATE3 was up-regulated in the Al-tolerant line after 24 hours of Al treatment, and was expressed at higher levels in the Al-tolerant line compared to the Al-sensitive parent (Fig. 6a). ZmNrat1 expression was induced by Al in the root tip of the Al-tolerant line after 6 hours, reaching a plateau between 12 and 24 hours of Al exposure (Fig. 6b). ZmWRKY expression was higher in the Al-sensitive than in the Al-tolerant line, and was induced by Al in the Al-sensitive parent (Fig. 6c). ZmART1 was more highly expressed in the Al-tolerant than in the Al-sensitive parent but was neither induced by Al nor differentially expressed in any time-point (Fig. 6d).

Discussion
Several landraces collected at maize growing areas in Kenya showed high levels of Al tolerance in nutrient solution, including 203B, which was originated from an acid soil region in Muranga County 19 . The Al tolerant parent, 203B-14, was derived from selfing the landrace 203B. The existence of genetic variability for Al tolerance in landraces adapted to local agro-ecological regions can be used for molecular genetics studies and for breeding purposes. In this context, highly contrasting Kenyan maize lines for Al tolerance were selected and crossed to generate an F 2:3 population, which was used for QTL mapping.
The QTLs and QTL interactions detected in this Kenyan population explained approximately 51% of the total variance for Al tolerance, which was lower but comparable to a QTL mapping study in a Brazilian population derived from Cateto Al237 × L53 (63%) 15 . Al tolerance in Cateto Al237 and 203B-14, parents of both the Brazilian and the Kenyan populations, respectively, was similar (~100% RNRG), which makes these lines highly Al tolerant 15,[19][20][21]25 . The slight bias of derived F 2:3 progeny towards Al sensitivity could be explained, at least in part, by the presence of three epistatic interactions between QTLs. High Al tolerance was observed in progeny that are double homozygous for favorable alleles at each pair of interacting QTLs (Fig. 3a-c). Taking two unlinked QTL loci whose alleles are assorted randomly, only 1/16 of the progeny are expected to be double homozygous for the favorable alleles. Hence, very few progeny should be highly Al tolerant, whereas the majority would express from medium to low Al tolerance.
No Al tolerance QTL was mapped on chromosome 6 (bin 6.00), which harbored the Al tolerance gene, ZmMATE1 16 , whose expression was much lower in 203B-14 and SCH3 than in Cateto Al237. Consistent with these results, both Kenyan parental lines showed lower ZmMATE1 expression 6 hours after Al stress compared to Cateto Al237 21 . As Al-induced citrate transporter expression in the root tips leads to ZmMATE1-mediated Al tolerance in Cateto Al237 16,18 , low ZmMATE1 expression in 203B-14 strongly supports the lack of this functional www.nature.com/scientificreports www.nature.com/scientificreports/ allele in 203B-14. Thus, the high Al tolerance in this Kenyan line is likely controlled to a large extent by mechanisms other than Al exclusion mediated by ZmMATE1. As other Al-tolerant maize genotypes from Kenya also showed low ZmMATE1 expression 21,25 , it is possible that functional ZmMATE1 alleles are not present in Kenyan maize germplasm.
We selected candidate genes co-localized with Al tolerance QTLs, which will be discussed in the light of QTL effects and gene expression profiles. The Al tolerance qALT5.03 explained 5.4% of the Al tolerance phenotype and two candidate genes ZmMATE3 (GRMZM2G065154) and ZmNrat1 (GRMZM2G168747) are located within this genomic region. A positive correlation between citrate exudation and Al tolerance (r = 0.51, P < 0.05) in 12 Kenyan maize accessions has been previously reported 25 , suggesting that Al exclusion mediated by citrate release takes place in Kenyan maize germplasm. Phylogenetic studies of maize MATE-like proteins grouped ZmMATE3 with other MATE members previously characterized as citrate transporters in different plant species (Fig. 4b and Guimaraes et al. 15 ). Different to the early induction reported for ZmMATE1 in Cateto Al237 16 , ZmMATE3 was induced by Al after 24 hours in 203B-14, which donated the qALT5.03 allele improving Al tolerance. As ZmMATE1 was not functional in this population, ZmMATE3 could provide such Al tolerance mechanism to 203B-14.
ZmNrat1 is 83% similar to OsNrat1, a plasma membrane Al 3+ transporter required for Al detoxification in rice 22 . A reduction of OsNrat1 expression in rice resulted in shorter roots with higher Al concentration in the cell wall of root tips of the Al-sensitive parent compared to the Al-tolerant parent under Al stress 26 . In addition, OsNrat1 was found to underlie the Al tolerance QTL on rice chromosome 2 22,26 . ZmNrat1 (GRMZM2G168747), which was also co-localized with qALT5.03, was up-regulated after 6 hour of Al exposure in the root tip of 203B-14, which donates the favorable qALT5.03 allele. It is thus plausible to hypothesize that the induction of ZmNrat1 in the root tip of the 203B-14 could lead to Al tolerance based on a mechanism similar to that controlled by rice Nrat1, via Al transport reducing Al 3+ concentration in the cell wall of apical cells 22 . Therefore, ZmNrat1 and ZmMATE3 could coordinate internal detoxification and exclusion mechanisms of Al tolerance, contributing to the high level of Al tolerance in 203B-14, conferred by qALT5.03.
The physical position of qALT8.05 (111.1-133.8 Mb) coincides with that of marker umc1202 at 109.2 Mb on chromosome 8, and umc1202 was linked with an Al tolerance QTL detected based on root re-growth and hematoxylin staining 14 . This suggests that qALT8.05 is conserved in different maize populations. In the 203B-14 × SCH3 population, qALT8.05 harbors GRMZM2G034421 that encodes a group III WRKY protein. ZmWRKY has 80% sequence identity with SbWRKY1, and 20.6% and 17.1% with OsWRKY22 and AtWRKY46, respectively. SbWRKY1 activates SbMATE expression in sorghum 23 , OsWRKY22 activates the expression of OsFRDL4 in rice 27 , whereas AtWRKY46 is a transcriptional repressor of AtALMT in Arabidopsis 28 , with all these genes being shown to control Al tolerance in the target species. Interestingly, ZmWRKY was induced by Al and showed higher expression in the Al-sensitive (SCH3) compared to the Al-tolerant parent. SCH3 donated the favorable allele of qALT8.05 that interacted with both qALT1.09 and qALT9.01 alleles, improving the RNRG in the population (Fig. 3a,b). These data suggest that ZmWRKY would be a candidate gene underlying qALT1.09, encoding an activator of Al tolerance genes located within qALT1.09 and qALT9.01 regions.
The Al tolerance QTL, qALT10.02 (8.7-13.5 Mb), was coincident with an Al tolerance QTL flanked by umc130 at 13.6 Mb, which was associated with an Al tolerance QTL in an F 2 population derived from C100-6, another highly Al-tolerant Cateto line 12 . The predicted gene GRMZM2G068710, located within the qALT10.02 interval, encodes a C2H2 zinc finger protein with 59.4% of sequence identity to OsART1, a transcription factor that regulates several Al tolerance genes in rice 24 . ZmART1 was phylogenetically closer to OsART1 and clustered with ART1/STOP1 transcription factors from Arabidopsis (AtSTOP1 29 ), sorghum (SbSTOP1 30 ) and wheat (TaSTOP1A 31 ), which have been shown to control Al tolerance in these species. Although controlling the expression of several genes under Al stress, OsART1 and AtSTOP1 were not responsive to Al in rice 24 and Arabidopsis 28 , similarly to ZmART1 that was not differentially expressed by Al in both Kenyan maize lines. However, ZmART1 presented higher expression in 203B-14, which donated the favorable allele of qALT10.02. Additionally, qALT10.02 interacted with qALT1.09, with progeny double homozygous for the 203B-14 alleles at both QTL showing high Al tolerance. Thus, qALT10.02 allele from 203B-14 could enhance Al tolerance by harboring ZmART1, which would activate transcriptionally other genes.
A remarkable difference of this Kenyan population was the relative high effect of epistatic interactions between QTLs, which has never been detected in other mapping study of maize Al tolerance [12][13][14][15] . The existence of different Al tolerance mechanisms in Kenyan maize germplasm, independent from ZmMATE1, brings the opportunity to develop superior maize cultivars by introducing exotic lines harboring functional ZmMATE1 allele, such as Cateto Al237. Al-tolerant cultivars should benefit maize production on acidic soil regions worldwide.

Material and Methods
plant material. The plant material consisted of 180 F 2:3 progeny derived from a cross between Kenyan maize inbred lines previously characterized as extremely tolerant (203B-14) and sensitive (SCH3) to Al 20 . Additionally, the Brazilian standard lines for Al tolerance (Cateto Al237) and sensitivity (L53) were used as checks in hydroponics.
DnA extraction and markers genotyping. DNA was isolated from young leaves of the parental lines and F 2 plants using a modified CTAB method as described by Saghai-Maroof et al. 32 . Genotyping of single nucleotide polymorphisms (SNPs) was performed using the Kompetitive Allele-Specific PCR (KASP TM ) assays by LGC Genomics (www.lgcgenomics.com). The parental lines were screened with 1,250 random SNPs for polymorphism detection and F 2 individuals were genotyped with 183 SNP markers.
Additionally, 14 fluorescently labeled SSR (Simple Sequence Repeat) markers were genotyped in the population. PCR reactions were performed using 50 ng of DNA, 1X PCR Buffer, 2.5 mM MgCl2, 166 µM of each evaluation of aluminum tolerance in nutrient solution. Al tolerance was assessed in a growth chamber under nutrient solution according to Guimaraes et al. 15 . Briefly, four-day old seedlings were transferred to polyethylene cups organized into containers filled with nutrient solution 33 at pH 4.0 under continuous aeration. After 24 h of acclimatization, the initial root length (IRL) was measured and the seedlings were cultivated with and without {39} µM of Al 3+ activity supplied as AlK(SO 4 ) 2 .12H 2 O (brackets denote free Al 3+ activity estimated with GEOCHEM-EZ software 34 that corresponds to 222 µM of Al concentration). The final root length (FRL) of each seedling was measured five days after the treatments and net root growth (NRG) was calculated as FRL -IRL under Al treatment (NRG +Al ) and control conditions, without Al (NRG −Al ). The phenotypic index used to evaluate Al tolerance was Relative Net Root Growth (RNRG) calculated as NRG +Al /NRG −Al (x100).
The F 2:3 progeny and the parents were evaluated in six experiments carried out in a completely randomized design with three replicates and two common checks (Cateto Al237 and L53). Analysis of variance was performed with RNRG data using PROC GLM of SAS software version 6.1.7601. Broad sense heritability (H 2 ) was estimated based on family means. Linkage analysis and QtL mapping. Marker loci were tested for goodness-of-fit to the expected single locus segregation ratio in an F 2 population (1:2:1) using the chi-square test (P < 0.05). The linkage map was constructed using the MapMaker/EXP 3.0 35 , with a minimum LOD of 3.0 and a maximum recombination frequency of 0.4. The Kosambi mapping function 36 was used to convert recombination frequencies into map distances in centiMorgans (cM).
QTL mapping was performed using multiple interval mapping (MIM) 37 implemented in QTL Cartographer version 2.5 for Windows 38 . The final model was selected using forward selection based on the Bayesian Information Criterion (BIC) with the penalty function c(n) = log(n), in which n = 180. The QTL position was defined based on the closest marker to the QTL maximum LOD value, and those markers were used to calculate the RNRG mean of genotypic classes for each QTL and for combinations of epistatic QTLs. Confidence intervals were established using the LOD-1 criterion 39 . The QTLs (q) were named using the acronym of Al tolerance (ALT) followed by their genetic position in chromosomal bins.
Searching for candidate genes within the Al tolerance QtLs. Genes previously associated with Al tolerance in other species were searched within the confidence intervals of Al tolerance QTLs based on sequence similarity using Phytozome (phytozome.jgi.doe.gov) considering the B73 genome sequence version 4.0. phylogenetic analysis of candidate genes. Protein sequences encoded by the candidate genes co-localized with Al tolerance QTLs and other similar sequences from maize were aligned with their respective homologs controlling Al tolerance in other plants using the M-COFFEE package available at T-COFFEE (tcoffee. crg.cat). Percent of identity was based on Clustal Omega (www.ebi.ac.uk/Tools/msa/clustalo/) and phylogenetic trees were constructed based on maximum likelihood using the software Mega v 10.0.5 40 . expression analysis of candidate genes. The expression profiles of the candidate genes co-localized with Al tolerance QTLs were evaluated by quantitative real-time PCR (RT-qPCR) using the ABI Prism 7500 Fast System (Applied Biosystems, Thermo Fisher Scientific, Inc.). Maize seedlings were grown in hydroponics as described in the section Evaluation of aluminum tolerance in nutrient solution with seven seedlings representing each sample. The first centimeter of the root tips were collected after 0, 1, 6, 12 and 24 hours of treatment with and without {39} μM of Al 3+ activity in the contrasting parents 203B-14 and SCH3. Total RNA was extracted using the RNeasy Plant Mini Kit (Qiagen, Germantown, MD) and the first-strand cDNA was synthesized using the High Capacity cDNA Reverse Transcription kit (Applied Biosystems, Thermo Fisher Scientific, Inc.) according to the manufacturer's instructions. Transcripts were quantified using cDNA (5 ηg for target genes and 0.005 ηg for the endogenous control 18S rRNA), 2.5 μM of each primer and Fast SYBR Green Master Mix 1×(Applied Biosystems, Thermo Fisher Scientific, Inc.) in a final volume of 10 μL. Primers for each target gene were designed using the Primer-Blast tool (www.ncbi.nlm.nih.gov/tools/primer-blast/) (Supplementary Table S4). Calculation of relative gene expressions were performed using 2 −ΔΔCt method 41 , with three biological and three technical replicates for each biological sample.