Theoretical and Applied Genetics

, Volume 126, Issue 10, pp 2587–2596

Genome-wide association analysis for nine agronomic traits in maize under well-watered and water-stressed conditions

Authors

  • Yadong Xue
    • National Key Laboratory of Crop Genetic ImprovementHuazhong Agricultural University
  • Marilyn L. Warburton
    • USDA-ARS Corn Host Plant Research Resistance UnitMississippi State University
  • Mark Sawkins
    • Generation Challenge Program
  • Xuehai Zhang
    • National Key Laboratory of Crop Genetic ImprovementHuazhong Agricultural University
  • Tim Setter
    • Department of Crop and Soil SciencesCornell University
  • Yunbi Xu
    • International Maize and Wheat Improvement Center
  • Pichet Grudloyma
    • Nakhon Sawan Field Crops Research Center (NSFCRC)
  • James Gethi
    • Kenya Agricultural Research Institute (KARI)
  • Jean-Marcel Ribaut
    • Generation Challenge Program
  • Wanchen Li
    • Maize Research InstituteSichuan Agricultural University
  • Xiaobo Zhang
    • National Key Laboratory of Crop Genetic ImprovementHuazhong Agricultural University
  • Yonglian Zheng
    • National Key Laboratory of Crop Genetic ImprovementHuazhong Agricultural University
    • National Key Laboratory of Crop Genetic ImprovementHuazhong Agricultural University
Original Paper

DOI: 10.1007/s00122-013-2158-x

Cite this article as:
Xue, Y., Warburton, M.L., Sawkins, M. et al. Theor Appl Genet (2013) 126: 2587. doi:10.1007/s00122-013-2158-x

Abstract

Drought can cause severe reduction in maize production, and strongly threatens crop yields. To dissect this complex trait and identify superior alleles, 350 tropical and subtropical maize inbred lines were genotyped using a 1536-SNP array developed from drought-related genes and an array of 56,110 random SNPs. The inbred lines were crossed with a common tester, CML312, and the testcrosses were phenotyped for nine traits under well-watered and water-stressed conditions in seven environments. Using genome-wide association mapping with correction for population structure, 42 associated SNPs (P ≤ 2.25 × 10−6 0.1/N) were identified, located in 33 genes for 126 trait × environment × treatment combinations. Of these genes, three were co-localized to drought-related QTL regions. Gene GRMZM2G125777 was strongly associated with ear relative position, hundred kernel weight and timing of male and female flowering, and encodes NAC domain-containing protein 2, a transcription factor expressed in different tissues. These results provide some good information for understanding the genetic basis for drought tolerance and further studies on identified candidate genes should illuminate mechanisms of drought tolerance and provide tools for designing drought-tolerant maize cultivars tailored to different environmental scenarios.

Introduction

Maize (Zea mays L.) is a staple for humans and other animals, and serves as raw material for production of starch, oil, protein, food sweeteners, and alcohol. Drought is the most serious environmental stress obstructing maize production and greatly reduces crop yields. It has been estimated that drought causes yield reductions ranging from 9.3 to 35.1 % in China (Wang and Li 2010), and even a transient drought may cause yield losses of 5–15.5 % annually in some area of the central US Corn Belt (Campos et al. 2004). Predicted long-term effects of global warming include increased drought conditions over much of the world (Cook et al. 2007). Therefore, drought-tolerant maize is urgently needed to maintain sufficient global production (Bruce et al. 2002). However, it is clear that the genetic mechanisms of drought tolerance are very complex, and successful genetic improvement programs of drought tolerance require information regarding genotype, environment, and genotype by environment (G × E) interactions (Yue et al. 2005; Bänziger and Araus 2007). Characterization of functional genes or markers closely linked to genes related to drought tolerance is a key step towards genomics-assisted plant breeding.

Grain yield (GY) under water stress is the primary trait used to assess the degree of drought tolerance in many crops, including maize. Correlated secondary traits, such as anthesis-silking interval (ASI), grain yield components, and plant height, are generally easier to measure than yield and show a higher heritability, and thus may be more suitable for improving maize selection response to water-stressed conditions (Bänziger et al. 2000; Setter 2012; Edmeades et al. 2000). ASI is commonly used as a selection criterion for drought-tolerant maize genotypes, as it has been shown to be highly correlated with grain yield under water-stressed conditions (Gutierrez-Rodriguez et al. 1998). At the International Maize and Wheat Improvement Center (CIMMYT), yield under water-stressed conditions was increased 3 % per cycle over eight cycles using this approach (Bolaños and Edmeades 1993). In addition, a limited water supply often postpones developmental stages in maize, reducing plant and ear height, and thus available photosynthate for grain production, resulting in drastic reduction of yield (Sari-Gorla et al. 1999; Lopes et al. 2011); thus, selection for taller plants and higher ears can enhance yield under drought conditions. Therefore, GY, along with these secondary traits, could be used to study the effects of drought stress on plants and to identify underlying functional genes (Messina et al. 2010).

Quantitative trait locus (QTL) mapping in biparental populations has been used to identify regions of the maize genome likely responsible for changes in morphological, metabolic, and enzymatic traits related to drought response (Ribaut et al. 2004; Welcker et al. 2007). Clustering of drought-related QTL including GY, OP (osmotic potential), LA (leaf surface area), and PH, consistent with pleiotropy or linkage, has been observed in several studies (Guo et al. 2008; Nikolic et al. 2011; Rahman et al. 2011). Mapping QTL associated with metabolic traits that are significantly correlated with drought tolerance has also uncovered QTL for carbohydrate and abscisic acid (ABA) content accumulation during stress (Tuberosa et al. 2002; Capelle et al. 2010). Biparental mapping cannot explore the full extent of allelic diversity that would be present in diverse germplasm. The limited resolution (10–20 cM; Holland 2007) associated with this technique necessitates further fine mapping to isolate the possible candidate gene(s).

Association analysis based on linkage disequilibrium has several advantages over biparental QTL mapping and is a useful tool to identify superior alleles for complex traits (Remington et al. 2001; Newton-Cheh and Hirschhorn 2005; Zhu et al. 2008; Myles et al. 2009; Rosenberg et al. 2010; Yan et al. 2011). Drought tolerance has also been subjected to association mapping to identify several SNPs associated with genes controlling the ABA pathway (Setter et al. 2010) and numerous SNPs directly associated with drought tolerance (Hao et al. 2011; Lu et al. 2010). The previous drought studies were run with a low marker density, which would be unable to capture the global genetic diversity for drought.

In this study, a commercial maize SNP50 array (Ganal et al. 2011) was used to perform genome-wide association analysis with a diverse panel of 350 maize genotypes phenotyped under well-watered and water-stressed conditions. The objective of this study was to identify candidate genes significantly associated with grain yield and related phenotypic traits under drought stress. In addition to the information gained on drought tolerance, this study may serve as a model for genetic dissection of other complex quantitative traits.

Materials and methods

Plant material

The association-mapping panel was comprised of 350 maize inbred lines selected to represent a wide range of diversity that could all be grown in a common environment, i.e., the tropical and subtropical field conditions typical of the developing world. The germplasm used in this experiment includes known drought-tolerant lines, known drought-susceptible lines (Setter et al. 2010), and others not previously tested (Online Resource 1). All inbred lines were testcrossed to a common tester, CML312, which has a good general combining ability and moderate drought susceptibility. CML312 was selected from CIMMYT Population 500, which is a subtropical, white dent population of intermediate maturity. It was generated by combining germplasm from DeKalb and Northrup King commercial hybrids, South Asia Pop.-3 × Suwan 1, Pools 32 and 20, and Populations 42 and 44. The backgrounds of the latter four pools and populations are very diverse, containing germplasm from South Asia, North, South, and Central America, the Caribbean, Europe, and the Middle East. Although it displays good general combining ability with most inbred lines tested to date, it displays specific combining ability with CML313, 314, 315, and 321. Of these, only CML321 is included in the current drought association-mapping panel. In a study of genetic relationships between CIMMYT derived inbred lines by Xia et al. (2005), CML312 was most closely related to CML311 (not included in the present study), and fairly equidistant to all other lines in the study. By crossing to CML312, the effects of the genes coming from the other inbred parent could be measured, and would not be masked by the effects of specific combining ability or narrow sense heterosis (except perhaps in the testcross with CML321). In addition, the weak drought tolerance coming from CML312 should not mask the tolerance coming from the other parent; nor should it contribute to poor performance in the hybrids, making tolerance impossible to measure. Genotypes were assigned to one of three precocity groups, and planting dates of each precocity group were staggered, so that all lines would experience the same drought stress during flowering time.

Phenotyping of hybrids

All hybrids were grown over 2 years (2006–2007 and 2007–2008 dry seasons) in four locations in Tlaltizapan, Mexico, Sichuan, China (2006–2007 only), Nairobi, Kenya, and Nakhon Sawan, Thailand, typical of arid climates in developing countries. Fields were planted with two replications in an alpha lattice design, using 5-m rows per plot and two seeds per hole, and plants were thinned after emergence. Water stress was applied at flowering of each maturity group, and continued through grain filling. The following traits under well-watered (WW) and water-stressed (WS) conditions were measured on five plants and averaged over the plot: grain yield (GY); hundred kernel weight (HKW); kernel number (KNO); ear height (EH); plant height (PH); relative ear position (EPO); female flowering, (days to silk DTS); male flowering, (days to anthesis DTA); and anthesis-silking interval (ASI). The correlation between the two replications within each location and treatment was calculated to estimate repeatability and confidence in the data. The difference in yield between WW and WS materials was calculated to ensure that plants subjected to drought treatment were under more stressful conditions.

Phenotypic data analysis

Repeatability (w2) for each trait was calculated across environments, where an environment was defined as each field × year combination. Repeatabilities were computed as follows:
$$w^{2} = \sigma_{\text{G}}^{2} /\left[ {\sigma_{\text{G}}^{2} + ({{\sigma_{\text{GE}}^{2} } / n}) + \sigma_{\text{e}}^{2} /(nr)} \right]$$
where \(\sigma_{\text{G}}^{2}\) is the genotypic variance, \(\sigma_{\text{GE}}^{2}\) is the genotype × environment variance, \(\sigma_{\text{e}}^{2}\) is the error variance, n is the number of environments, and r is the number of replications. The estimates of \(\sigma_{\text{G}}^{2}\), \(\sigma_{\text{GE}}^{2}\), and \(\sigma_{\text{e}}^{2}\) were analyzed by analysis of variance (ANOVA) using the aov function in R (version 2.14.1, R Foundation for Statistical Computing, http://www.r-project.org/). Correlation coefficients were obtained using Pearson’s statistic as implemented in the cor procedure in R. Best linear unbiased prediction (BLUP) was done by fitting the following mixed linear model in R package “lme4” for estimation of breeding values of all testers:
$$Y = \left( {1|{\text{LINE}}} \right) + \left( {1|{\text{ENV}}} \right) + (1|{\text{REP}}\% {\rm in} \% {\text{LINE}}:{\text{ENV}}) + (1|{\text{LINE}}:{\text{ENV}})$$
where Y is trait data, the parentheses indicate random effects, “1|” means groups, and “:” means interactions. LINE refers to all testcrosses used, ENV to environments, each of which is a combination of years and locations, and REP to the replications in one ENV.

Genotyping and filtering

In this study, the MaizeSNP50 BeadChip maize array (Illumina) was used to genotype all lines. SNP content and selection criteria are described in Ganal et al. (2011). Briefly, 56,110 markers that were evenly spaced along the B73 reference sequence (AGPv2) were used to cover the whole maize genome. These markers were selected from several published and unpublished sources. Of 350 maize lines, 333 were successfully genotyped. After removal of unsuccessful calls, SNPs that were missing in more than 20 % of the tested lines, and SNPs exhibiting >20 % heterozygosity (unexpected in inbred lines), usable information was reported for 50,989 SNPs. Because SNPs with low MAF (minor allelic frequency) in the preliminary data analysis can often result in false positive associations, markers with MAFs <5 % were excluded from the association analysis, leaving 43,990 SNPs. The association panel had been previously genotyped with the SNP1536 chip (Setter et al. 2010), and these two data sets were compared to confirm the quality of the current genotyping assays. Five SNPs and 15 lines for which the differences in SNPs were >10 % between the two data sets were removed from the analysis. A final total of 44,314 markers from the two chips scored in 318 lines were considered sufficiently robust and consistent to be employed in the present analysis.

Association analysis

The linkage disequilibrium measurement parameter r2 was used to estimate LD between SNPs on each chromosome via the software package TASSEL3.0 (Bradbury et al. 2007). A principal components analysis (PCA) was used to correct for population stratification (Price et al. 2006; Zhang et al. 2009b, 2010), and a kinship matrix was calculated using the Loiselle algorithm (Loiselle et al. 1995) to determine relatedness among individuals. Zhao et al. (2007) demonstrated that the PCA method provides results comparable with alternative approaches such as the STRUCTURE algorithm. The SNP data set from 318 inbred lines was analyzed with the GWAS tool GAPIT (Genome Association and Prediction Integrated Tool-R package) (Lipka et al. 2012). The PC matrix was generated automatically by setting GAPIT parameters pca.total to 6. The general linear model (GLM) with PCs for population structure control (QQ plots showed over fitting of PC + K model, data not shown) was used for genome-wide association mapping with 44,314 SNPs (MAF > 0.05) on data of each environment. A Bonferroni-corrected threshold probability based on individual tests was calculated to correct for multiple comparisons, using 1/N, where N is the number of individual trait-SNP combinations tested.

Results

Analysis of phenotypes under WW and WS

The repeatability values (w2) of the nine target traits and their average phenotypic performance based on BLUP values across the seven environments are shown in Table 1. Non-significant differences were observed between the means for DTA under WW and WS conditions using F test, most likely because drought stress was imposed just prior to flowering. Average ASI was three times higher under WS than under WW conditions, which was expected because ASI reflects susceptibility to drought (Welcker et al. 2007). The repeatability values for DTA and DTS were high under both WW and WS, while w2 for ASI was lower (58 % under WW and 47 % under WS). WS decreased GY by 54 % on average, which indicated that severe drought conditions had been experienced. Average repeatability of GY was 61 % in the WW environment and 54 % under WS conditions. The results for KNO were analogous to those of GY except that KNO repeatability was very low under WS. HKW repeatability was also significantly decreased, in WS compared with WW conditions, and repeatability was higher for this trait than for other yield-related traits. PH and EH were lower under WS conditions because plants were unable to fully develop under drought. The upper part of the plant (above the ear) was more affected by drought, and this was reflected in much lower WS/WW ratio for PH than for EH. Although the difference in EPO between WW and WS was significant, the difference itself (WS/WW = 1.1) was not as great as for PH or EH.
Table 1

Trait performance based on the BLUP value across seven environments

Category

Trait

Treatment

Average ± SD

Range

WS/WW

Repeatability

Plant architecture

PH

WW

219.9 ± 8.48

184.7–240.60

0.79**

0.840

WS

173.5 ± 4.41

156.2–185.22

0.522

EH

WW

107.6 ± 8.57

85.2–133.91

0.85**

0.897

WS

92.0 ± 5.36

76.4–112.04

0.744

EPO

WW

0.5 ± 0.03

0.4–0.55

1.1**

0.899

WS

0.5 ± 0.03

0.5–0.68

0.823

Flowering time

DTA

WW

70.4 ± 1.96

65.4–75.02

0.99

0.906

WS

70.3 ± 2.08

64.6–74.90

0.897

DTS

WW

72.5 ± 2.15

67.2–77.99

1.05**

0.915

WS

76.4 ± 2.05

70.7–82.32

0.781

ASI

WW

2.0 ± 0.50

0.6–3.43

3.11**

0.577

WS

6.2 ± 0.68

4.5–8.04

0.466

Yield components

GY

WW

2.7 ± 0.15

2.2–3.04

0.46**

0.606

WS

1.2 ± 0.12

0.9–1.59

0.544

KNO

WW

1,022.2 ± 39.72

884.0–1,116.15

0.45**

0.49

WS

457.7 ± 17.44

416.0–523.91

0.161

HKW

WW

33.4 ± 1.95

28.5–39.44

0.85**

0.844

WS

28.4 ± 1.63

23.5–34.55

0.699

PH plant height, EH ear height, EPO relative ear position, DTA days to anthesis, male flowering, DTS days to silk, female flowering, ASI anthesis-silking interval, GY grain yield, KNO kernel number, HKW hundred kernels weight

*, ** Significant at P = 0.05, 0.01, respectively

Correlations

Under WW conditions, no phenotypic correlations were observed between GY and flowering traits. A negative correlation (P ≤ 0.01) between ASI and GY (−0.425) was observed under WS conditions, as well as between DTS and KNO (−0.389) (Table 2). There was a correlation between GY and HKW, and it was higher than the correlation between GY and KNO under WS, but lower under WW while correlations between PH and EH with GY were higher under both conditions.
Table 2

Correlation matrix for all traits based on BLUP value across seven environments under water stress (WS, under diagonal) and well-watered (WW, above diagonal)

Trait

PH

EH

EPO

GY

KNO

HKW

DTA

DTS

ASI

PH

1

0.77**

0.36**

0.40**

0.33**

0.08

0.48**

0.46**

0.03

EH

0.66**

1

0.87**

0.30**

0.35**

−0.04

0.60**

0.55**

−0.07

EPO

0.002

0.74**

1

0.14*

0.25**

−0.11*

0.51**

0.45**

−0.12*

GY

0.26**

0.21**

0.04

1

0.45**

0.30**

0.08

0.06

−0.03

KNO

0.14**

0.02

−0.12*

0.37**

1

−0.44**

0.27**

0.25**

−0.02

HKW

0.17**

0.10

−0.01

0.43**

−0.23**

1

−0.10

−0.10

0.04

DTA

0.25**

0.47**

0.42**

−0.24**

−0.19**

−0.01

1

0.94**

−0.04

DTS

0.11*

0.29**

0.31**

−0.44**

−0.39**

−0.02

0.83**

1

0.28**

ASI

−0.18**

−0.22**

−0.11*

−0.43**

−0.42**

−0.01

−0.06

0.50**

1

PH plant height, EH ear height, EPO relative ear position, DTA days to anthesis, male flowering, DTS days to silk, female flowering, ASI anthesis-silking interval, GY grain yield, KNO kernel number, HKW hundred kernels weight

*, ** Significant at P ≤ 0.05, 0.01, respectively

Linkage disequilibrium

All 44,314 SNPs with a MAF > 0.05 were loaded into Plink 1.07 (Purcell et al. 2007) to calculate genome-wide LD in this panel of tropical/subtropical maize germplasm. A rapid decline in LD was observed with increasing physical distance on all chromosomes (Online Resource 2), but the decay rate varied over different chromosomes: equilibrium was reached within 30–45 kb on chromosome 1, 50–60 kb on chromosomes 4 and 5, and 80–150 kb on the rest of the chromosomes. The mean LD decay across all chromosomes was 80–100 kb (r2 = 0.1). If we relax the cut-off to r2 = 0.2, the mean LD decay was very rapid reach to gene level as 5 kb.

Genome-wide association studies

A total of 126 trait × environment × treatment combinations were analyzed, resulting from measurements of nine traits in seven environments and two watering treatments. Population structure was controlled with the PC matrix. Previous work with this same panel indicated that population structure was relatively modest, with five PC axes explaining approximately 7.2 % of the variation in population structure (Setter et al. 2010). In the current study, population structure was re-estimated using a larger number of SNPs, and a similar result was obtained, with six PC axes explaining about 6.9 % of the variation. Quantile–quantile plots (Fig. 1 and Online Resource 3) showed that the GLM model with six PC axes effectively accounted for population substructure of all traits. A total of 51 associations involving 42 SNPs, located in 33 genes, were identified at the P < 2.25 × 10−6 (0.1/N) level (Online Resource 4). Seven of these genes were associated with more than one trait. These results are consistent with the quantitative nature of drought tolerance, which is known to be controlled by a large number of genes with small effects. Detailed association results are presented in Table 3 and Online Resource 5. More associations were identified under WW than WS for most traits. With respect to plant architecture traits, 4, 10, and 11 significant SNPs/treatments were detected for PH, EH, and EPO, respectively, across each environment; for flowering time traits, 4 associations were detected for DTA, 2 for DTS, and 5 for ASI; and for yield-related traits, 6, 2, and 3 associations were detected for GY, KNO, and HKW, respectively.
https://static-content.springer.com/image/art%3A10.1007%2Fs00122-013-2158-x/MediaObjects/122_2013_2158_Fig1_HTML.gif
Fig. 1

Quantile–quantile plot for associations with ear height in water-stressed condition in Thailand in 2008. P values are shown on a −log10 scale and the dashed horizontal line indicate Bonferroni-corrected threshold 0.1/N (color figure online)

Table 3

Significant SNP-trait associations for nine agronomic traits

Trait

C07

K07

K08

M07

M08

T07

T08

Total

SNPs

Genes

PH

 WW

0

0

0

0

0

1

1

4

4

4

 WS

0

0

0

1

0

0

1

EH

 WW

1

2

0

0

0

2

3

12

10

9

 WS

1

0

0

0

0

0

3

EPO

 WW

2

3

0

1

0

2

0

12

11

7

 WS

1

0

0

0

2

1

0

DTA

 WW

0

1

0

0

0

0

0

4

4

4

 WS

1

2

0

0

0

0

0

DTS

 WW

0

0

0

0

1

0

2

3

2

2

 WS

0

0

0

0

0

0

0

ASI

 WW

0

2

1

0

0

2

0

5

5

5

 WS

0

0

0

0

0

0

0

GY

 WW

0

3

0

0

0

1

1

6

6

5

 WS

0

1

0

0

0

0

0

KNO

 WW

0

0

0

0

1

0

1

2

2

2

 WS

0

0

0

0

0

0

0

HKW

 WW

0

0

0

0

0

1

1

3

3

3

 WS

1

0

0

0

0

0

0

Total

51

42

33

PH plant height, EH ear height, EPO relative ear position, DTA days to anthesis, male flowering, DTS days to silk, female flowering, ASI anthesis-silking interval, GY grain yield, KNO kernel number, HKW hundred kernel weight, C China, K Kenya, M Mexico, T Thailand, SNPs numbers of different, SNPs detected, Genes numbers of distinct genes detected

Numbers of significant SNP-trait associations of measured traits at p = 2.25e−06 from column C07 to column total

Discussion

Maize yields are most damaged by water shortage during flowering and early kernel development (Setter et al. 2001). In our study, withdrawal of irrigation before these important stages led to the expected severe water-stressed condition, and decreased PH and EH substantially, delayed DTS and increased ASI, and reduced KNO, and thus GY (Table 2). HKW was less affected by drought because plants, as part of their drought response, were able to distribute resources to pollinated kernels to ensure the optimal development of at least some of them, even though fewer kernels would be pollinated under drought. Higher correlations between PH and EH with GY under WW than under WS conditions show that tall plants have a larger photosynthetically active leaf area and more stem reserves (Sari-Gorla et al. 1999; Lopes et al. 2011) and, thus, better capacity for grain filling than short plants.

When accurate grain yield prediction under water deficit during flowering time can be carried out using ASI, selection gain is increased and phenotyping cost is decreased in breeding programs (Bolaños and Edmeades 1993; Edmeades et al. 1999; Monneveux et al. 2006; Ribaut et al. 2009; Barker et al. 2010). In the current study, ASI was only correlated with GY under water stress, confirming the predictive power of ASI on drought tolerance and also indicated that late silking is unfavorable for kernel set and grain yield under water deficit. Strong selection on elite germplasm, however, has narrowed the genotypic variation of this trait (Monneveux et al. 2008). In the present study, we have presented drought-related information collected for a diverse range of germplasm in replicated field experiments across four locations over 2 years. The availability of this data set not only advances our understanding of drought tolerance in maize, but also may be used by the maize community as a resource to deepen the drought resistance gene pool.

Genome-wide association mapping of drought tolerance

The association panel used in this study was comprised mainly of tropical and subtropical lines, and is part of a larger association resource genetically characterized by Yan et al. (2009), who estimated average LD decay to be 5–10 kb. In the current study, LD calculated based on 44,314 SNPs and the subset panel was higher, averaging 80–100 kb across the whole genome. The current panel lacks the diversity from the temperate lines included in the larger panel, which would have uncovered more recombinations and thus smaller linkage blocks. The higher LD in the current study suggests the need for fewer SNP markers to adequately cover the genome.

Large-scale genome-wide association analyses have been employed to dissect complex traits (Buckler et al. 2009; Brown et al. 2011; Weng et al. 2011; Riedelsheimer et al. 2012; Li et al. 2013) and have proven to be an effective way to find candidate genes for these traits. In this study, 42 SNPs were associated with one or more phenotyped traits, and some of these SNPs were located within consensus QTL found by a previous meta-analysis of drought tolerance in maize (Online Resource 5) (Li et al. 2010). Using a candidate gene association strategy, several SNPs in previous studies were found to be associated with metabolite (Setter et al. 2010) or agronomic (Lu et al. 2010; Hao et al. 2011) traits under water stress; however, none of those SNPs were detected in our study. This may be due to the low correlations between metabolite and agronomic traits (Schauer et al. 2006; Fu et al. 2009; Sulpice et al. 2010; Kooke and Keurentjes 2011) or the different populations under study. In this study, the testcrosses used and the different environments in which they were measured are further reasons that no common SNPs were detected compared with the study of Setter et al. (2010). A more recent genome-wide association study on metabolite traits in maize leaves under WS conditions by Riedelsheimer et al. (2012) found several significant associations; these included one SNP, PZB00083.3, associated with gama-aminobutyric acid (P = 2.05 × 10−5 <0.1/N) and also with an agronomic trait (GY_TS8) under WS in our study. The gene identified (a phytochrome receptor gene) is worthy of further investigation in drought-stressed maize ears.

Many of the 33 candidate genes identified in this study were associated with plant architecture or flowering time traits that had low correlation with GY under either WW or WS. Other studies have shown that these traits are related to drought tolerance and can be used to improve maize yield under drought (Zhang et al. 2009a; Saeed et al. 2011; Tardieu 2011; Lopes et al. 2011). Finding direct strong correlations between grain yield and secondary traits such as PH, EH, root growth, and root system architecture are often difficult; even though such traits may be associated with drought tolerance, they may have opposite influences on drought tolerance at different levels of drought severity. SNPs and genes found to be associated with specific drought environments may be used to design lines with enhanced drought tolerance under the same conditions (Tardieu 2011). Although impressive outcomes obtained in one drought scenario may have a limited application to other drought prone environments, the results of this study are still a valuable resource for improving drought tolerance of maize in the tested areas. In addition, SNPs and genes may highlight the important underlying drought-tolerance mechanisms, and may provide information for dissecting the genetic basis of drought tolerance and further molecular breeding.

Detailed analysis of candidate genes

A meta-analysis of published QTL related to drought tolerance identified over 400 QTL and 79 discrete consensus genomic regions (or mQTL) related to drought tolerance in maize (Li et al. 2010). Physical coordinates of the consensus genomic regions based on the B73 maize reference sequence version 2 overlapped with 3 of 33 associated genes identified in our study, which were located in three mQTL regions. For further examination, we chose annotated genes from within the consensus regions, and which were associated with flowering time or yield-associated traits in WS or in both WS and WW conditions. Of the 33 most significant genes, seven were associated with multiple traits. The highest number of simultaneous trait associations was seen with the gene GRMZM2G125777, which harbored the SNP PZE-104036909. This gene was associated with four different traits (EPO, DTA, DTS, and HKW), and encodes NAC domain-containing protein 2, which is expressed in all associated tissues (Sekhon et al. 2011).

Two genes, GRMZM2G140082 and GRMZM2G313643, were associated with grain yield-related traits measured under WS environment (HKW_CS7 and GY_KS7, respectively) and are plausible candidate genes for drought tolerance. GRMZM2G140082 and GRMZM2G313643 encode proteins with the same function, tyrosine-protein kinase. This kind of protein is involved in growth, proliferation, dead, survival, etc. (Li and Hristova 2006; Sharma et al. 2009). The two genes, which functioned in ear, may have the potential to improve GY under stress. Further knowledge about how these two genes work together, and with other genes associated in this and other studies, will aid targeted efforts to increase maize grain yield under drought stress.

Given that ASI is strongly correlated with GY under water stress (Bänziger et al. 2000), genes related to ASI are of particular interest. Two genes, GRMZM2G137961 and GRMZM2G119079, associated with ASI_KW7, were identified in this study (Online Resource 5). GRMZM2G119079 encodes a protein tyrosine phosphatase (PTPase), which functions in cell cycle control and cell growth, proliferation, differentiation, and transformation (Fordham-Skelton et al. 1999; Corellou et al. 2000; Ghelis 2011). GRMZM2G137961 encodes a member of Acyl-CoA N-acyltransferase superfamily that is involved in Krebs cycle, fatty acid metabolism, etc. It was highly expressed in mature tassel (Sekhon et al. 2011). There are three more genes associated with ASI. These genes are good candidates for future study, because regulation of cell proliferation and growth under drought conditions are factors that might determine ASI and drought tolerance (Ribaut et al. 2009; Setter et al. 2010; Lorković 2009).

Most of the remaining genes reported in Online Resource 5 were associated with traits pertaining to plant architecture. Although plant architecture traits have a low correlation with GY and drought tolerance, a full understanding the mechanism of these traits would assist breeders in the development of ideotypic maize phenotypes for high yield under WW as well as WS conditions.

The GWAS has been successfully used at a high resolution to uncover associations involving complex traits from field-grown maize inbred lines with genetic variants (Buckler et al. 2009; Tian et al. 2011; Riedelsheimer et al. 2012; Li et al. 2013). With the development of next-generation sequencing technology, the cost of sequencing has dropped dramatically, and the large number of markers needed to adequately cover the maize genome for GWAS is now available at a reasonable cost. Although gene level resolution can now be achieved in some diverse lines using GWAS (Yan et al. 2009), the technique still only provides a statistical (i.e., indirect) clue for connecting traits with their associated genomic sequences. It is therefore necessary to validate the uncovered associations using biological evidence obtained from approaches such as transgenic or mutation techniques (RNAi, antisense methods, or production of knock-out mutants for inducing loss-of-function point mutations in candidate genes) or the slower creation of near isogenic lines for comparative analysis of pairs of maize lines with and without the genes of interest.

Acknowledgments

This research was supported by the Generation Challenge Program and the National Hi-Tech Research and Development Program of China (2012AA10A307) and the National Natural Science Foundation of China (31101156).

Conflict of interest

The authors declare that they have no conflict of interest.

Supplementary material

122_2013_2158_MOESM1_ESM.pdf (74 kb)
Supplementary material 1 (PDF 73 kb)
122_2013_2158_MOESM2_ESM.pdf (358 kb)
Supplementary material 2 (PDF 357 kb)
122_2013_2158_MOESM3_ESM.pdf (2.6 mb)
Supplementary material 3 (PDF 2707 kb)
122_2013_2158_MOESM4_ESM.pdf (3.4 mb)
Supplementary material 4 (PDF 3471 kb)
122_2013_2158_MOESM5_ESM.pdf (63 kb)
Supplementary material 5 (PDF 62 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg 2013