Introduction

Maize is one of the important crops for its food, feed, and industry use. Maize has now become the most important grain crop in the world (FAO statistics data). In its growth periods, maize is almost variously affected by biotic and abiotic stress. Among these, drought stress is the major abiotic factor limiting grain yield and seed production in maize. Yield losses resulting from drought stress annually exceed the sum of other abiotic stresses. In developing countries, drought is especially severe because of a lack of irrigation facilities so that rainfall is the main source of water available to the crop growing (Banziger et al. 1999). Drought tolerance is a complex quantitative trait which is influenced by numerous genes (Wang et al. 2010) and regulated by several physiological and molecular pathways (Champoux et al. 1995; Lilley et al. 1996; Li et al. 2009). Selection of drought-tolerant varieties allows us to mitigate the yield loss caused by water stress and to reduce agricultural use of fresh water resources. The maize genome is rich in allelic variants and discovering these alleles from genes related to drought tolerance would be useful for maize breeding, as well as understanding the complex genetic and molecular mechanism underlying drought tolerance. Drought tolerance research focuses on morphological, physiological, and molecular features that influence the adaptive response of plants to drought stress (Bolanos and Edmeades 1993; Tuberosa and Salvi 2006; Messmer et al. 2009; Liu et al. 2009). Studies on maize drought tolerance through linkage mapping have made major progress for agriculture. Association analysis based on linkage disequilibrium (LD) is an effective method to identify functional allelic variation related to drought-tolerant traits.

As a well-known growth inhibitor, abscisic acid (ABA) modulates many key growth and physiological processes in plants, such as seed germination suppression, seed dormancy maintenance by inhibiting cell growth, stomatal closure induction to minimize transpiration for reducing water loss, and acceleration of abscission and senescence (Finkelstein et al. 2002; Fujita et al. 2005; Meng et al. 2009; Gao et al. 2010). Under water-limited conditions, there is evidence that ABA production is enhanced and it can effectively protect plants against drought stress (Shinozaki and Yamaguch-Shinozaki 2000; Xiong et al. 2002). ABA is mainly synthesized in an indirect way in which 9-cis-epoxycartenoid dioxygenase (NCED) encoded by the nced gene is the key regulatory enzyme of ABA biosynthesis in higher plants (Seo and Koshiba 2002). In maize, Vp14 (the same name as nced) was firstly cloned from the ABA-deficient mutant (Schwartz et al. 1997; Tan et al. 1997). The nced expression is induced in response to water deficit (Nambara and Marion-Poll 2005). Additionally, the maize rab28 gene (ABA-responsive gene), which encodes one of the late embryogenesis abundant proteins and is regulated by two bZIP transcription factors (Nieva et al. 2005) induced by ABA and water stress has been identified (Pla et al. 1991). The promoter region of the rab28 gene contains a cis-acting element CCACGTGG (ABRE, involved in ABA dependent pathway), which is induced by ABA induction (Pla et al. 1993). Therefore, the nced and rab28 genes are interesting candidate genes for further study of association analysis between nucleotide polymorphisms with phenotypic traits under water stress.

Based on the LD, association analysis can test the relationship between specific nucleotide sequence polymorphisms in candidate genes and phenotypic variation (Thornsberry et al. 2001). In maize, LD decays very rapidly (Remington et al. 2001; Tenaillon et al. 2001) so association mapping within gene resolution has a much greater precision than quantitative trait loci (QTL) mapping and provides an independent approach to test candidate genes identified in standard QTL experiments. At the same time, specific biological effects for significant polymorphisms can be hypothesized when association analysis links specific polymorphisms to trait variation. Although the spurious association of polymorphisms with traits is a potential obstacle due to relatedness rather than sequence variation function in association studies, there are now two ways to decrease the effect of false positive association. One is the structure population analysis which can be effectively estimated with the Structure 2.0 program using several models for linked and unlinked markers; another is the kinship analysis which includes the level of pairwise relatedness coefficients by SPAGeDi 1.2 software (Hardy and Vekemans 2002). Recently, Yu et al. (2005) developed a mixed linear model (MLM) which combines population structure information and kinship in the association analysis.

The objectives of this study were to investigate the natural variation in a collection of maize inbred lines and drought-tolerant candidate genes in order to identify functional alleles and haplotypes in the coding sequence of the nced and rab28 genes (including 5′-UTR, 2 exons, one intron). Association analysis was performed with the MLM program to examine the correlation between sequence polymorphisms in candidate genes with phenotypic traits related to drought tolerance.

Materials and Methods

Plant Materials

A total of 196 maize inbreds (Zea mays L.) were used in this study mostly collected from five major maize-growing regions in the corn belt in China while others were from the USA. They represented most of the genetic diversity available to breeding and research programs in China. The seeds of the maize inbred lines were acquired from the Institute of Crop Science, Chinese Academy of Agricultural Science, China; detailed information on these materials is reported in the paper of Xie et al. (2008). These inbred lines were divided into six subpopulations based on 70 simple sequence repeat (SSR) marker data described by Xie et al. (2008): BSSS (American BSSS including Reid), PA (group A germplasm derived from a modern U.S. hybrid in China), PB (group B germplasm derived from a modern US hybrid in China), Lan (Lancaster Surecrop), LRC (derivative lines from Lvda Reb Cob, a Chinese landrace), and SPT (derivative lines from Sipingtou, a Chinese landrace).

The inbred lines were grown in one-row plots with an alpha (0,1) lattice design (Barreto et al. 1997) at two different field sites of Sanya, HaiNan province (winter, 2007, abbr 2007HN), and Urumqi, XinJiang province (summer, 2008, abbr. 2008XJ) in China, All materials in a random order were evaluated under two water regimes (well-watered, WW and water-stressed, WS), each with two replications. Plants were open-pollinated. The morphological and physiological traits were investigated in flowering periods of each line as a unit under water-stressed and well-watered situations, and then grain yield and yield component traits were acquired 20 drying days after harvesting. The following 22 traits were studied: grain weight per ear (GraW), single ear weight, grain yield (GY), ear weight, hundred kernal weight (HKW), total plant number (PlaN), total ear number (EarN), ears per plant (EPP), row number per ear (RowN), kernal number per row (KerN), ear length (EarL), ear diameter (EarD), bald tip length (TipL), plant height (PlaH), ear height (EarH), tassel length (TasL), days to anthesis (DtoA), days to silking (DtoS), anthesis-silking interval (ASI), stay green (StayG), leaf curling (LeafC), and water content of kernel (WatC). Of these, only GraW, SEarW, EPP, and ASI were indirectly calculated; all other traits were directly investigated from a minimum of five plants in each line.

With the average value of at least five plants to each line, phenotypic traits variance effects of genotype, location, replicate within location, and genotype–location interaction were evaluated through the general linear model (GLM) program in SAS (SAS 1989). Since many traits were highly correlated, we performed principal component analysis on the correlation matrix from adjusted means (FACTOR program in SAS). The mean value of two replicates for phenotypic trait in WW and WS regimes of each environment was used for association mapping.

Population Structure and Individual Kinship

Population structure is a major bias factor leading to false-positive associations (Flint-Garcia et al. 2003). To alleviate the effect of population structure, all inbred lines were genotyped with 70 genome-wide SSR molecular markers, and six groups were defined (Xie et al. 2008) using Structure version 2 software (Pritchard et al. 2000a). These independent group memberships were used as covariates in the genotype–phenotype association analyses. Based on recent studies (Yu et al. 2005), kinship coefficient matrices of 196 inbreds were calculated using the SPAGeDi version 1.2 software.

DNA Isolation, PCR Amplification, and DNA Sequencing

Genomic DNA samples were extracted from maize leaves using the CTAB method (Saghai-Maroof et al. 1984). Then, two PCR products were amplified using Taq DNA polymerase and primers based on the sequence of the maize vp14 mRNA (Genebank accession number, U95953) and the sequence of the maize rab28 gene (Genebank accession number, X59138), respectively. The combinations of forward- and reverse-primers were: (1) 5′-TCCAATTCCGTCAGGTTCTC-3′ and 5′-TGATGAAGGTGCCGTGGAA-3′ for nced gene, (2) 5′-ACCATCAAGTCTCAAGAGGC-3′ and 5′-ACATGGACGTAGGTAAAGCG-3′ for rab28 gene, respectively. Finally, the high-quality sequences of 162 inbred lines in the nced gene (with a length of 223–1,736 bp, the first nucleotide of U95953 marked +1 bp) and the high-quality sequences of 131 inbred lines in rab28 gene (with a length of 433–1,501 bp, the first nucleotide of X59138 marked +1 bp) were acquired and aligned.

PCR conditions were as follows: 200 ng of genomic DNA, 0.2 mM of each dNTP, 1 μM of each primer, 2× GC buffer (2 mM mg2+), and 2.5 U TransTaq high fidelity DNA polymerase (Transgen Biotechnology Corporation, Beijing, China), adjusted to a final volume of 25 μL. An initial denaturation of PCR reaction was at 94°C for 3 min, followed by 35 cycles at 94°C for 45 s, primer annealing at 58–60°C for 1 min (the annealing temperature were 58°C and 60°C for nced and rab28, respectively), then 72°C for 1 min, and the final extension at 72°C for 10 min.

Most PCR products were sequenced directly by ABI 3000. The others were cloned by ligating to pMD18-T vector using Escherichia coli strain DH5a as a host and then sequenced (Sambrook and Russell 2001). All sequences were sequenced by the public laboratory of National Key Facility of Crop Gene Resources and Genetic Improvement, Institute of Crop Science, Chinese Academy of Agriculture Science. The DNA was sequenced twice in both directions. Nucleic acid sequences were aligned using the DNAMAN software and adjusted by manual work.

The TASSEL software (http://www.maizegenetics.net/bioinformatics/tassel.index) was used to identify single nucleotide polymorphisms (SNPs) and insertions/deletions (indels). Nucleotide diversity was calculated as the average number of pairwise differences among sequences (Tajima 1989) and polymorphic sites. Haplotype diversities were calculated among different gene structure regions based on their position (favoring those in exons rather than introns) of polymorphic sites, potential functional role (favoring non-synonymous rather than synonymous changes), frequency (favoring balanced allele frequencies rather than rare alleles), and complementarity (avoiding redundancy among polymorphisms and favoring those that allow characterization of the highest number of haplotypes observed among the sequenced inbred lines).

Association Analyses

The mixed linear model pattern in TASSEL software was used to identify significant nucleotide variations associated with phenotypic traits under water-stressed regime. In addition, coefficient of drought resistance (CD) from GY under water-stressed regime divided by GY under well-watered regime was also used as a trait for association analysis. Both population structure and individual kinship relatedness were considered in association studies of all phenotypic traits with candidate genes. Consequently, association studies presented here are based on the Q+K model (mixed linear model) that is more complex but performs better for some phenotypic traits. Afterwards, the GLM pattern in the TASSEL software based on the Q model was used to do multiple testing of significant associations between the polymorphism and phenotype traits.

Results

Statistics of Phenotypic Traits Under WW and WS Conditions

ANOVA results showed that the maize inbred lines showed significant variance among all the investigated traits, which indicated that inbred lines hold diverse characteristics for genotyping under water deficit. Besides, there was no significant variation found between the WW and WS treatments for traits, including WatC, HKW, TipL, EarL, and EarD. For some traits including LeafC, WatC, PlaN, EarN, EarH, SearN, GY, and EarD, there were significant variations found between replicates, which might be caused by interaction between environment and genotype. Finally, StayG, PlaH, TasL, ASI, DtoA, DtoS, GraW, RowN, KerN, and EPP were significantly influenced by inbred lines, treatments, and interaction between environment and inbred lines (Table 1). Therefore, phenotypic data of the above 10 traits with two replicates were averaged for factor analysis using SAS software, whereas five factors represented more than 85% phenotypic variation. This indicated that several traits would be commonly regulated by the same variation underlying drought tolerance. To acquire accurate results, the above ten traits were used for the next association analysis between sequence variance of the nced and rab28 genes and phenotypic traits under water stress.

Table 1 Combined variance analysis (GLM) of 22 traits and their significance levels
Full size table

Sequence Polymorphisms and LD Analysis of nced and rab28 Genes

For the coding region of the nced gene, we identified 20 SNPs with a frequency of more than 10% of the rare alleles among the 162 lines and no indels. Among SNP, four variants were non-synonymous resulting in changes in amino acids, while the remaining ones were synonymous. Nucleotide diversity showed that the highest allelic variation frequency occurred around 600–750 bp among sequenced region of the nced gene with an average Pi of 0.00741 (Table 2). LD was estimated between all pairs of polymorphic sites in the sequenced region of the nced gene. A plot of r 2 against physical distance for polymorphic pairs indicated that the LD level remained high (r 2 > 0.2) along the entire sequenced region of the nced gene (Fig. 1). However, LD was not evenly distributed among these loci. The highest level of LD was identified by the detection of two recombination events: one hot spot between sites 234 and 310, and the other between sites 609 and 831 of all the alignments (r 2 > 0.8). Moreover, sites 420 and 672, 498 and 672, 672 and 1,554, and sites 720 and 831 showed higher LD (r 2 > 0.5). A relatively high degree of LD in the polymorphic sites of nced gene indicated its conserved evolution process of maize inbred lines.

Table 2 Nucleotide polymorphisms of the sequenced regions of nced and rab28 genes
Full size table
Fig. 1
figure 1

LD level among pairwise polymorphisms in the alignment of nced gene. Lower left triangle, P values derived from Fisher’s exact test. Upper right triangle, r 2 values. The left and lower number indicated the polymorphic sits of nced gene

Full size image

For the rab28 gene, we identified 26 SNPs and six indels in the sequenced region of the rab28 gene with a frequency above 10% for the rare alleles among 131 inbred lines. Of these, one indel (five nucleotides) was found in the 5′UTR region, two SNPs, and two indels (four nucleotide deletion) were found in the intron region of the rab28 gene, 11 SNPs were in exon-2, while 13 SNPs and three indels (two include 18 nucleotides and one includes three nucleotides) were in exon-1. In the coding sequence of the rab28 gene, there were 12 synonymous variances (SNP) and 12 SNPs were non-synonymous. At the same time, three indels in the coding sequence of the rab28 gene did not result in frame shift mutation. Nucleotide diversity showed that the highest allelic variation frequency occurred around 600–950 bp among the rab28 gene sequenced region with an average Pi of 0.03054. LD was estimated between all pairs of polymorphic sites in the sequenced region rab28 gene. The same result with the nced gene, a plot of r 2 against physical distance for polymorphism pairs indicated that LD level remained high (r 2 > 0.2) for the entire length of the rab28 gene-sequenced region. As is indicated in Fig. 2, LD is not evenly distributed along the locus. The highest level of LD was supported by the detection of two regions; one from sites 877 to 946, and the other from sites 1,399 and 1,424 of the alignment (r 2 > 0.5). Between sites 532 and 541, there was higher LD (r 2 = 1), these two sites were thought to be significant polymorphisms. Other regions showed different levels of LD.

Fig. 2
figure 2

LD level among pairwise polymorphisms in the alignment of rab28 gene. Lower left triangle, P values derived from Fisher’s exact test. Upper right triangle, r 2 values. The left and lower number indicated the polymorphic sits of rab28 gene

Full size image

The Tajima’D test is a widely used test of neutrality in population genetics by measuring the allele frequency distribution of nucleotide sequence data. Tajima’D test result of both gene in this study showed no significant natural selection existed in their entire sequenced regions (P < 0.05). The non-coding regions of both genes gave negative values for selection, while the two exon regions of the rab28 gene gave positive values. Fu and Li’s tests showed similar results with Tajima’s D test in sequenced regions of both genes at the significant level of P < 0.05 (Table 2). Low genetic diversity and weak evidence of selection could result in low nucleotide polymorphisms and high r 2 in LD analysis.

Phenotype–Genotype Associations of the nced and rab28 Genes

MLM of association analysis that controls the effects of population structure and individual relative relationship were used to identify the association of phenotypic traits under water stress and genotypic variation data in the nced and rab28 genes. All nucleotide polymorphisms, including SNPs and indels, with a frequency of more than 10% of the rare alleles were considered in the association analysis of phenotype–genotype in both genes.

By MLM analysis and multiple testing, seven polymorphisms in the nced gene-sequenced region (positions 498; 672; 876; 918; 1,009; and 1,554) were significantly associated (P < 0.05) with traits KerN, RowN, EPP, DtoS, and CD. There were no associations identified for GraW, TasL, PlaH, DtoA, ASI, and StayG in the 2007HN experiment. In the 2008XJ experiment, eight polymorphisms in the nced gene sequenced region (positions 310; 420; 597; 720; 918; 1,128; 1,164; and 1,554) were significantly associated (P < 0.05) with GraW, KerN, RowN, EPP, DtoS, DtoA, ASI, TasL, and PlaH. No associations were identified for StayG and CD. It indicated that grain yield components such as KerN, RowN, and morphological traits such as PlaH, DtoA, and ASI were highly associated with the nced gene under water stress. Based on the associated polymorphisms in 2007HN and 2008XJ, best allele at each site was identified for nced gene in all sequenced maize inbred lines, which was thought to be the superior haplotype of nced gene related to drought tolerance (Table 3).

Table 3 Candidate-gene association analysis of phenotypic traits related to drought tolerance in both environments based on polymorphic sites in the nced sequenced coding region of 162 maize inbred lines
Full size table

By MLM analysis and multiple testing, four polymorphisms in the rab28 gene sequenced region (positions 532; 541; 889; and 1,269) were significantly associated (P < 0.05) with RowN and PlaH. There were no associations identified for GraW, EPP, KerN, TasL, DtoA, DtoS, ASI, StayG, and CD in the 2007HN experiment. In the 2008XJ experiment, seven polymorphisms in the rab28 gene-sequenced region (positions 553; 1,108; 1,172; 1,404; 1,407; 1,416; and 1,424) were significantly associated (P < 0.05) with EPP, RowN, and StayG. There were no associations identified for GraW, KerN, TasL, DtoA, DtoS, ASI, PlaH, and CD. It indicated that grain yield component RowN was highly associated with the rab28 gene under water-stressed conditions. Based on the associated polymorphisms, best allele at each site were identified for the rab28 gene in all sequenced maize inbred lines, which was thought to be the superior haplotype of the rab28 gene related to drought tolerance (Table 4).

Table 4 Candidate-gene association analysis of phenotypic traits related to drought tolerance in both environments based on polymorphic sites in the rab28 sequenced coding and non-coding regions of 131 maize inbred lines
Full size table

Discussion

Nucleotide Diversities of the nced and rab28 Genes Among Chinese Germplasms

In the study, there were 20 SNPs identified in the sequenced region of nced gene, four of which were non-synonymous and the others were synonymous. However, unlike the nced gene, there were 15 synonymous polymorphisms in the sequenced region of the rab28 gene, and the other 12 nucleotide variances were non-synonymous (including SNPs and indels). Compared to other functional genes in maize previously reported, nucleotide diversity of the sequenced regions of the nced and rab28 genes were slightly higher. In maize, SNP has been detected even more frequently because of its characteristic of cross pollinator. It has been reported that there is one SNP approximately every 48 bp (Tenaillon et al. 2001) and every 130 bp (Rafalski 2002) in un-translated and coding regions, respectively. In this study, we detected on average one SNP every 80 bp in the nced gene in the coding region and one SNP every 41 bp and one indel every 178 bp in the rab28 gene. This result was quite similar or a little higher than the average level previously reported for other maize genes.

Ching et al. (2002) and Flint-Garcia et al. (2003) found nucleotide diversity decreased while LD increased at a given locus because of population bottlenecks and selection. In the study, there was no distinct positive evidence of selection for nced and rab28 genes through Tajima’s (1989) and Fu and Li’s (1993) neutrality tests (Table 2). Both genes had similar results, which were in agreement with results previously reported for maize functional genes. For example, Tenaillon et al. (2001) found that there was a rare departure from neutrality in random maize loci. Our research proved that negative or weak selection occurred for both genes.

Remington et al. (2001) and Tenaillon et al. (2001) reported a rapid breakdown of LD (r 2 < 0.1 within few hundred base pairs) for several loci in diverse sets of maize germplasm. Our research obtained the same result where there was a rapid decay of LD for more than 750 bp in the nced gene and for more than 700 bp in the rab28 gene. It has been shown that LD decay level varies in different materials in maize. Ching et al. (2002) and Jung et al. (2004) reported extended LD might be up to hundreds of kilobase in sets of inbred lines. In contrast, the rab28 gene had a more rapid decay than the nced gene.

Association Analysis of the nced and rab28 Genes with Drought Tolerance

Association analysis based on linkage disequilibrium (LD) has recently emerged as an alternative approach to mapping QTL and genes associated with some comprehensive traits. Compared with the QTL mapping, association analysis identified the allele variations of gene sequence with no need for construction of mapping populations. In this study, 13 significant polymorphisms in the sequenced region of nced gene and 11 significant polymorphisms in the sequenced region of rab28 gene were identified associated with drought tolerance through the MLM procedure and multiple testing in TASSEL software. Although the association analysis method is an effective way to find true polymorphisms with phenotypic traits in candidate gene, population structure is one of the major reasons that lead to a false positive association (Knowler et al. 1988; Pritchard et al. 2000a, b). For example, population structure among inbred lines may cause candidate genes to lose their significance because of non-functional polymorphism and the related functional polymorphism coinciding with population structure. The latter can lead to a false negative result. In this study, the MLM procedure based on Q+K model decreased the chance of a positive association and then the GLM procedure based on Q model tested the significance of detected associations. The highly significant associations with lower P value were always existed. However, there were two associations with traits StayG (P = 0.044) and GraW (P = 0.049) in 2008XJ for nced gene, two associations with trait RowN (P = 0.048 and 0.036) in 2008XJ for rab28 gene, which were not detected through multiple test and were deleted in the final result. It indicated that the multiple testing could adjust the accuracy of association analysis or the P threshold value we had chosen could be lower. In this study, two environments of 2007HN and 2008XJ located different latitudes of 18°14′N and 43°54′N, which suitable planting maize cultivars were greatly different in both locations. It caused the field phenotyping result highly varied, especially for GraW. So, the complex interaction of genotype by environment should be considered and studied in the future program.

Functional markers were frequently used for molecular-assisted breeding in recent years, in particular that association analysis accelerated their deployment for breeding purposes. Generally, they are derived from causative polymorphisms. For this research, the non-synonymous SNPs in the nced gene-sequenced region and the rab28 gene significant polymorphisms could be firstly considered candidate variation sites related to drought tolerance. Allelic effects of these polymorphisms could be further elucidated in a gene expression and enzyme activity study. Consequently, further research should focus on the large number of genes involved in the drought tolerance network combining several genes acting on metabolism pathways related to drought tolerance. Additionally, by using TILLING technique, a series of drought tolerance gene mutants could be produced for comparisons of single polymorphisms, especially in complete LD with isogenic backgrounds.