The RIL population consisting of 204 F8 lines was derived from a cross between two inbred lines, DH1M (female parent) and T877 (male parent) using single-seed descent. Plants were grown in a greenhouse located on the campus of Yangzhou University. The experiment used a completely randomized design with three replicates per line. Maize seeds of 204 lines were sterilized for 20 min in a 10% solution of H₂O₂, washed with distilled water, soaked in saturated CaSO₄ for 6 h, and then germinated in the dark on moist filter paper at 28°C in a germination chamber for 2 days. Three seedlings of uniform size from each line were transplanted to round black plastic pots (18 cm in depth, 20 cm in diameter). Each pot was filled with 3 L light nutritional soil, and 2 days before planting, all pots were watered to saturation with deionized water. In the first 7 days, all pots received 100 mL of deionized water every 2 days. Then, 200 mL of deionized water used to irrigate plants in the WW treatment group every 4 days, and the WS treatment plants received no further irrigation until harvest. Seven days after sowing, seedlings were thinned to one per pot.

The seedlings were harvested at 35 days after sowing. Plant height (PH, cm) was measured from the coleoptilar node to the tip of the longest leaf. The third leaf (unfolded leaf) was selected for measuring leaf length (LL, cm), leaf width (LW, cm), and chlorophyll content (SPAD). SPAD was measured using a SPAD-502 PLUS chlorophyll meter (Minolta, Japan). Total shoot fresh weight (SFW, g) and dry weight (SDW, g) were measured. Shoot dry weights were determined after oven-drying at 70°C to a constant weight, and the shoot dry weight was measured using an electronic balance. Roots were separated from the soil by vigorous rinsing through a 2-mm sieve. Roots were stored at −20°C before the measurements. The length of primary root (PRL, cm), seminal root (SRL, cm), and crown root (CRL, cm) was measured with a ruler. The number of seminal roots (SRN) and crown roots (CRN) was counted. Three roots on third whorl were selected to measure crown root angle (CRA) and diameter (CRD, mm). Root angles were measured with a protractor as degrees from horizontal: Horizontal roots were classified as 0°, vertical roots as 90° (Trachsel et al., 2013). Selected crown roots were cut off, and a Vernier caliper was used to measure root diameter.

All RILs and parental inbred lines were genotyped using an Affymetrix microarray CGMB50K SNP Array containing 56,000 maize SNPs at China Golden Marker (Beijing) Biotech, China. The cosegregating SNP markers were considered as belonging to the same recombination bin using a home-made Perl script. All called bins were used to construct the genetic linkage map using JoinMap version 4.0 software (Van Ooijen, 2006), with the Kosambi mapping function used to calculate the genetic distance between markers. The genotype and phenotype data can be downloaded from the website https://pan.baidu.com/s/1geDpnJt.

The R software package (R Development Core Team, 2013) was used to perform statistical analyses on the raw data from each experiment. Analysis of variance (ANOVA) was used to test for significant differences between treatments, lines, and interactions (genotype–environment interaction, GEI). Mean values of each line across the two experiments were used for subsequent phenotypic summarization, correlation analysis, principal component analysis (PCA), and QTL mapping. PCA was performed using the “prcomp” function in R, with further visualization performed using the “ggfortify” package. The “lme4” package was used to estimate genotypic variance (σG2), G–E interaction variance (σG×E2), and error variance (σe2). The broad-sense heritability (h²) of each measured trait was calculated as previously described by Hallauer and Miranda (1981).

QTL mapping based on data for root and shoot traits under WW and WS conditions was conducted using ME analysis and a MT approach to detect main QTLs, GEI and pleiotropic QTLs with the QTL library in GenStat for Windows v18 (VSN International, Hemel Hempstead, UK). A step size of 10 cM, a minimum cofactor proximity of 30 cM, a minimum separation between selected QTLs of 20 cM, and a genome-wide significance level of P < 0.05 were used for the QTL analysis. The whole genome was first scanned using simple interval mapping (SIM) with selected cofactors for two rounds of composite interval mapping (CIM; Malosetti et al., 2004; Boer et al., 2007). A final QTL model was selected by backward selection of the selected cofactors, with the allelic effect of each QTL estimated in each environment (ME) or each trait (MT).

Maize plants were grown under WW and water-stressed (WS) conditions in a greenhouse, and seven root traits, including crown root (CRA, CRD, CRL, and CRN), PRL, and seminal root number (SRL and SRN), and six shoot traits, including plant height (PH), leaf traits (LL and LW), shoot biomass (SDW and SFW), and SPAD, were evaluated three times. Water stress decreased CRD, CRL, CRN, PH, LL, LW, SDW, SFW, and SPAD and increased CRA, PRL, and SRL (Figure 1, Table 1). Water stress had the greatest influence on SFW and SPAD; the values for these parameters were reduced by 16.62 and 17.96%, respectively. Frequency distributions of the different traits in the RIL population showed that all variables exhibited normal distributions under WW and WS conditions (Figure 1). Coefficients of variation ranged from 14.0 to 56.7% under WW conditions and 15.43–57.42% under WS conditions, indicating the existence of considerable phenotypic variation (Table 1). Genotypic variation was significant for all investigated traits at P < 0.001, and 12 traits showed significant variation under different water regimes, except SRN. CRA, CRD, CRL, PRL, SRL, LL, SDW, SFW, and SPAD showed significant G × E interaction effects (Table 1). The heritability (h²) of root-related traits were moderate, ranging from 30.4% (CRA) to 86.6% (CRL) under WW and from 37.9 to 74.0% under WS conditions. The heritability (h²) of shoot-related traits was rather high, varying from 69.5 to 87.4% under WW and from 78.7 to 83.3% under WS condition (Table 1).

Regardless of different water regimes, CRA showed no significant correlation with other traits, except a very small correlation with CRD (WW, r = 0.189), CRL (WS, r = −0.184), and PRL (WW, r = −0.176). CRD was significantly correlated with shoot traits under WW (r = 0.262–0.436) and WS (r = 0.387–0.506), but SPAD was not (Figure 2). As expected, all shoot-related traits measured under WW and WS conditions showed a positive correlation (r = 0.260–0.875), except for SPAD. In general, the water treatment had little effect on the correlation coefficient between traits, indicating that the root traits and shoot traits synergistically respond to WS. Principal component analyses (PCA) were conducted among the RILs under WW and WS conditions (Figures 3A,B). The first two major principal components (PC1 and PC2) are projected, respectively, onto x- and y-axes in the figure. The first component (PC1) represented more than 30% of the variability under both WW and WS conditions and accounted primarily for most traits except for SPAD and CRA. PC2 explained 13.4% variation and accounted primarily for SPAD and CRA under WW, while 17% variation mainly accounted for SPAD under WS (Figure 3).

The RIL populations were genotyped using the CGMB50K SNP Array, and a total of 56,000 SNPs were detected. After quality control, 9,780 high-quality polymorphic SNPs were used to construct linkage maps. The co-separation markers were considered as recombination bin, resulting in a map with 1,868 marker bins (Figure 4A, Table S1). The total length of the linkage map was 3,081.8 cM with an average interval of 1.65 cM, the largest interval on the linkage map was 12.85 cM. To examine the quality of the genetic map, QTL mapping was conducted for leaf sheath color, which has high heritability. Two QTLs were identified, the largest one located on chromosome 10 with a peak at 150.7 Mb (LOD = 112.9; Figure 2C). The cloned r1 (colored1) gene was located in the confidence interval of this QTL (Figure 4B).

To identify the genetic basis of maize root plasticity to water availability, we conducted QTL mapping in the RIL population using a ME approach. In total, 48 QTLs were mapped for the thirteen analyzed traits under WW and WS environments (Table 2). A total of 15 QTLs, associated with 5 root traits (CRA, CRL, CRN, PRL, and SRL) and 4 shoot traits including LL, LW, SFW, and SPAD, showed significant Q × E interactions, of which two QTLs (CRA2 and CRL1) had antagonistic pleiotropic effects in the two different environments, other QTLs with QEI were only identified in one of the conditions. The phenotypic effects of four QTLs: CRN1, PRL3, SFW3, and SPAD2 were 26-, 9.72-, 13.01-, and 23.38-times higher, respectively, in one environment than the other (Table 2). Of the identified QTLs, 28 and 11 possessed a favorable allele from T877 and DH1M under both conditions (Table 2).

A total of 26 identified QTLs were associated with root traits (Table 2). Four putative QTLs for CRA were detected. Only CRA2 at SNP251 had a significant QEI, and it was detected on chromosomes 1. CRA2 had antagonistic pleiotropic effects in the two water conditions. The T877 allele had a positive effect under WW conditions, whereas DH1M contributed a favorable allele under WS. Only one QTL associated with CRD was detected under WW and WS conditions. Sixteen QTLs were mapped for root length (three QTLs for PRL, nine QTLs for SRL, and four QTLs for CRL). Of these, six QTLs with significant QEI (CRL1, PRL2, PRL3, SRL2, SRL7, and SRL9) were detected. CRL1 had antagonistic pleiotropic effects in the two water conditions, PRL2 and PRL3 were detected only under WW conditions, whereas SRL7 and SRL9 were detected only under WS conditions. SRL2 had the largest contribution to SRL, 14.4% under WW conditions, whereas it only explained 1.2% of the phenotypic variation under WS conditions. Root number was mapped to five loci (three QTLs for SRN and two QTLs for CRN), all CRN-QTL were detected with significant QEI which were identified only under WW conditions. No Q × E interactions were detected for SRN-QTL. A total of 22 QTLs were identified for shoot traits, including 3, 6, 2, 4, 3, and 4 QTLs for PH, LL, LW, SDW, SFW, and SPAD, respectively. Six QTLs with significant QEI were detected (LL1, LL6, LW1, SFW3, SPAD2, and SPAD3) including five QTLs (except SFW3) that had antagonistic pleiotropic effects in the two watering conditions. LL1 and LW1 were detected only under WS conditions, and the other four were detected only under WW condition.

Phenotypic correlation analysis revealed that most traits were correlated with one another (Figure 2). Traits are genetically correlated due to pleiotropic QTLs or closely linked QTLs. To identify which QTLs showed pleiotropic effects, we conducted MT analysis for 13 traits under WW and WS conditions. In total, 14 regions were identified as harboring 71 putative QTLs for the 13 analyzed traits in WW and WS environments (Figure 5, Table S2). Each region consisted of 5.1 QTLs on average with a range of 2–8. SNP1347 (7@121.9 cM) consisted of eight QTLs, most of these were identified under WW conditions, and the T877 allele increased the trait value. Three loci (SNP253 at 1@407.8 cM, SNP526 at 3@288.8 cM, and SNP1579 at 9@9.2 cM) were only identified under WW conditions. SNP253 and SNP526 showed an increased effect with the DH1M allele. One locus (SNP418 at 2@307.7), which was only identified under WS conditions, showed an increased effect with the T877 allele. SNP20 (1@79.2 cM) was associated with PRL, SRL and CRL under WS conditions, and the higher value allele was from T877. SNP1057 (5@234.5 cM) was mainly associated with shoot-related traits (PH, LL, LW and SFW). Under both WW and WS, all QTLs had a positive contribution of DH1M (Figure 5).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.