Capture of large-scale i-traits in maize drought response
To gain insights into how maize plants respond to drought, we cultivated a maize association mapping population (AMP), which consists of 368 inbred lines and has 1 M SNPs among the population [25], in a greenhouse under well-watered (WW) and drought-stressed (DS) conditions (Additional file 1: Table S1; see “Methods”). By using an automatic platform for crop phenotyping developed based on our previous work (RAP [20, 26, 27]), the dynamic responses of each individual plant were captured in a noninvasive way via three types of scanners, RGB imaging, HSI, and CT, over a course of 98 days (Fig. 1a; Additional file 2: Video S1), which generated ~ 14 TB of images. To process the huge numbers of images, we further developed specific image analysis pipelines (Fig. 1b; Additional file 3: Video S2; Additional file 4: Video S3; Additional file 5: Video S4), with which a total of 26,910 i-traits (2010 RGB, 24,000 HSI, and 900 CT image-based traits) were extracted. After i-traits selection procedures (Fig. 1c), including filtering outliers, determination of drought-related i-traits using T-tests of WW/DS and multilayer perceptron (MLP), and heritability tests (Additional file 6: Video S5; Additional file 7: Video S6; Additional file 8: Video S7), 10,080 drought-related i-traits (37.46% of the rough i-traits, including 1503 RGB-derived, 7902 HSI-derived, and 675 CT-derived i-traits) were selected for further genetic study. The definitions of these i-traits are shown in Additional file 1: Table S2 and Additional file 9: Note S1. All the selected RGB, HSI, and CT i-traits are listed in Additional file 1: Table S3-5. All these images and related i-traits are open access to the public at https://doi.org/10.6084/m9.figshare.14429003.v1.
Additional file 6: Video S5. Operation procedure of filtering outlier.
Additional file 7: Video S6. Operation procedure of T-test.
Additional file 8: Video S7. Operation procedure of Multi-layer perception (MLP).
Effective and inheritable i-traits to reflect maize drought response
Many i-traits changed dynamically during the drought treatments and growth stages (Additional file 10: Figure S1a-c). For example, the RGB-derived i-trait TPA (total projected area), which has been reported as a good indicator of rice growth under drought stress [20], was indicative of different growth situations of maize plants under various drought stresses (Fig. 2a). The HSI-derived i-trait dT233, which is the first-order derivative of the total reflectance under 959 nm, has been reported to reflect internal water content [28]. We observed that dT233 increased under WW conditions and decreased under DS conditions, suggesting that it was an effective indicator of drought responses (Fig. 2b). The CT-derived i-trait hollow_area_700 reflected culm wall size also effectively indicated different levels of drought stress (Fig. 2c).
To further explore the potential of i-traits to reflect drought responses, principal components analysis (PCA) was performed to capture the phenotypic variations in the entire maize population. For the RGB- and HSI-derived i-traits, at D52 with more severe stress, PC1 alone explained more than 50% of the phenotypic variation, which clearly separated the WW plants from those undergoing DS (Fig. 2d, e). Interestingly, compared with RGB-derived and CT-derived traits, HSI-derived traits had better distinguishing ability even in early stress stages (Fig. 2d, e, Additional file 10: Figure S1d-f). Next, we calculated the broad-sense heritability (H2) of each individual i-trait over the growth period with or without drought stress, and the middle number of H2 of these i-traits was 0.4 (Fig. 2f, Additional file 10: Figure S2).
Genetic basis of i-traits in maize drought response
We performed GWAS of 10,080 i-traits with a mixed linear model (MLM) to detect significant SNP-trait associations. More than 2989 (29.6% of 10,080) i-traits had at least one significant associated SNP (P ≤ 1.8 × 10−6). We identified a total of 4322 distinct significant SNPs associated with 2989 i-traits (Additional file 1: Table S6 and 7). More significant SNPs (2378, ~ 55%) were detected with ratio i-traits as compared to those with i-traits from WW (972, ~ 22.5%) or DS (849, ~ 19.6%) conditions alone (Additional file 10: Figure S3a, Additional file 1: Table S7). Each SNP explained 5.3–22.6% of the observed phenotypic variance of the i-traits. The SNPs associated with CT-derived i-traits explained more phenotypic variance on average as compared to those with RGB and HSI-derived i-traits (Additional file 10: Figure S3b and c), suggesting either less complex genetic architecture or highly enriched diversity, both quantitatively and qualitatively, of CT-derived i-traits. We mapped the significant SNPs onto the maize chromosomes at 200-kb intervals (100 kb upstream and downstream of the significant SNP), and the mapped intervals were defined as QTLs controlling maize drought tolerance. In total, 1529 QTLs were identified (Additional file 1: Table S8). Of these, 71.4% (1092/1529) were co-localized with previously reported QTLs (Additional file 1: Table S9) [9, 29,30,31].
We extracted the candidate genes based on the significant SNPs, whose average LD decay in AMP used in this study have been reported to be 0.5 kb, reaching single-gene resolution [25]. In total, 2318 unique candidate genes related to 4322 significant associations were identified and annotated (Additional file 1: Table S7). Of which, only 95 genes (~ 4.1%) were consistently detected in two or more types of i-traits (Additional file 10: Figure S4a). Based on the genes and i-traits, we built a gene-trait network, in which the genes that involved in the same biological pathway were gathered in a group (Fig. 3a). This network would facilitate candidate gene identification and its function elucidation. We found that very few pathways were shared by genes associated with three types of i-traits, and many unique pathways were detected for genes associated with HSI-derived or RGB-derived i-traits (Fig. 3a, Additional file 1: Table S10 and Additional file 10: Figure S4a). For instance, although MAPK (mitogen-activated protein kinase) signaling and BR (Brassinolide) signaling pathways were shared by genes from HSI and RGB i-traits, several pathways, such as one carbon pool by folate, RNA degradation, and trypophan metabolism were unique to genes detected with RGB i-traits, and many other pathways, such as ABA signaling pathway, sugar metabolic pathway, and inositol phosphate metabolic pathway, were specific to genes associated with HSI-derived i-traits (Fig. 3b, c, Additional file 10: Figure S4b-d). These results indicated different genetic controls of these i-traits in drought responses. Further example of insights based on the data integration are shown below with case studies to ZmcPGM2 and ZmFAB1A in regulation of i-traits and drought tolerance.
The candidate genes were significantly enriched in GO terms response to different stimuli or stresses, suggesting the importance of these candidate genes in maize drought/stress responses (Additional file 10: Figure S5, Additional file 1: Table S11). GWAS showed that many genes were significantly associated with previously published drought-tolerant phenotype survival rates (SR) of AMP [12] (MLM, Additional file 1: Table S1, Additional file 1: Table S12). Permutation assay showed that there were enriched most significant P values of these candidate genes as compared to those from randomly selected genes (Pt.test = 9.11e-287, Ppermutation < 0.0001, Fig. 3d), suggesting that these associations are not false positive but real associations. Moreover, 25 previously identified drought-tolerant genes were detected in our candidate gene set (Additional file 1: Table S13). Taken together, these results indicated that the candidate genes were reliable and that the i-trait-based GWAS was powerful in mapping drought-responsive QTLs and causal genes.
Transcription factors (TFs) play key roles in plant drought tolerance [32]. In our GWAS results, there are 165 genes (7.1% or 165/2318) encoding TFs of 41 families, of which the NAC (14 genes) and AP2/EREB ERF (12 genes) TF families, which are well-known to control plant drought tolerance [32], were the families with the most members (Additional file 1: Table S14). The well-studied TF genes ZmNAC111 (GRMZM2G127379), Zmhdz10 (GRMZM2G041127), ZmDREB2A (GRMZM2G006745), and ZmDREB2.7 (GRMZM2G028386) [10, 11, 33, 34] were all detected by GWAS in this study. For example, ZmNAC111 positively regulates maize drought tolerance [11]. The most significant SNP chr10.S_2681198 of ZmNAC111 was significantly associated with the HSI-derived ratio i-trait ddT136_D46_R (the ratio of second-order derivative of the 725 nm total reflectance under drought stress to second-order derivative of the 725 nm total reflectance under well-water condition at 46 days after sowing) (P = 1.5 × 10−6, MLM) (Additional file 10: Figure S6a-d). There were two alleles of chr10.S_2681198. Plants with allele T had lower ddT136_D46_R levels (P = 1.39 × 10−6, t-test) but much higher (P = 2.65 × 10−4, t-test) ZmNAC111 expression under DS (Additional file 10: Figure S6e and f), implying that allele T of chr10.S_2681198 could be a favorable allele in AMP for regulating maize drought tolerance by enhancing ZmNAC111 expression. These analyses further suggested the reliability of the candidate drought-tolerant genes.
Functional interpretation of hotspot candidate genes
Next, we identified hotspot candidate genes that associated with no less than 10 i-traits. In total, 34 hotspot genes were detected (Fig. 3e, Additional file 1: Table S15), of which 29 were associated with HSI-derived i-traits (85% or 29/34). The gene GRMZM2G028386 (ZmDREB2.7) was associated with 13 HSI-derived i-traits and encoded AP2/EREBP ERF TFs. ZmDREB2.7 belongs to the AP2 DREB subfamily and positively regulates maize drought tolerance [10]. The most significant SNP chr1.s_201957847 in ZmDREB2.7 was significantly associated with HSI-derived i-trait lgA15_D34_WW (the logarithm of the 434 nm average reflectance under well-water condition at 34 days after sowing) (P = 7.5 × 10−7, MLM) and 12 other i-traits (Additional file 1: Table S15, Additional file 10: Figure S7a-d). This most significant SNP showed high linkage (R2 > 0.93) with two other SNPs in the coding region and the reported drought-tolerant causal allele, five polymorphic sites in the promoter region (R2 = 1) [10] (Additional file 10: Figure S7d). Based on the A/T alleles of the most significant SNP, plants with allele T had higher levels of i-trait lgA15_D34_WW (P = 4.27 × 10−8, t-test) and higher survival rates (P = 8.10 × 10−11, t-test) after drought stress (Additional file 10: Figure S7e and f), suggesting that allele T is a favorable allele in regulation of lgA15_D34_WW levels and maize drought tolerance.
Reactive oxygen species (ROS) are important signaling molecules in stress responses [35]. The membrane protein respiratory burst oxidase homolog D (RbohD) triggers ROS signaling at the very early stage of dehydration (e.g., in ~ 20 min) and plays positive roles in stomatal closure and ABA signaling [36, 37]. HSP proteins play key roles in maintaining ROS homeostasis and further in plant drought tolerance [38, 39]. GRMZM2G098167 (HSP20-like protein) was associated with 258 HSI-derived ratio i-traits and GRMZM2G300965 (ZmRbohD) was associated with 241 HSI-derived ratio i-traits, and both genes shared 34 associated i-traits (Fig. 3e, Additional file 1: Table S4, Additional file 1: Table S15). Intriguingly, all of these i-traits were calculated from the HSI images captured at D34 (the first time point for HSI imaging) with SM = ~ 20% (Fig. 1a), which was at the early drought stress stage. Based on these data, we deduced that ZmRbohD could play a key role in initiating ZmRbohD-dependent ROS signaling initiation, and HSP20-like could function to maintain ROS signaling homeostasis in maize drought tolerance.
Identification of the regulatory variants that control the candidate gene expression
The difference in gene expression could originate from changes in local and/or distant regulation [40]. We next investigated the expression QTLs (eQTLs) that associated with the expression of the 2318 candidate genes based on the transcriptome of 197 lines from 540 association mapping population treated with or without drought (M. Dai and L. Li unpublished RNA-seq data) [41]. Totally, 54.2% (1257/2318) of the candidate genes were controlled by 22,546 significant eQTLs (P ≤ 4.2 × 10−8, MLM, Additional file 1: Table S16-18). When the most significant SNP of an eQTL was located in a 20-kb region (upstream of downstream) of the expression trait (etrait) gene, this eQTL was defined as a local eQTL; otherwise, it was a distant eQTL. We found that distant eQTLs were identified for most (~ 63%) candidate genes under both WW and DS conditions (Additional file 1: Table S18); however, the local eQTLs had much larger effects on the expression of the etrait genes under both WW and DS conditions (Fig. 4a, b), indicating that local variations have great effect on gene expression regulation.
Among the total eQTLs, the majority (69%, or 15,668/22,546) were dynamic (detected under WW or DS condition), only 31% of the eQTLs were static (detected under both WW and DS conditions). Similar ratios of dynamic (74%) and static (26%) eQTLs were observed in distant (17,088) eQTLs (Fig. 4c), indicating vast and dynamic gene regulatory networks in maize i-trait formation. We took closer look at the local eQTLs as they averagely have greater effects on gene expression regulation than distant eQTLs (Fig. 4a, b). Totally, 2383 static and 3075 dynamic local eQTLs were detected based on the gene expression under WW and DS conditions (Fig. 4c, Additional file 1: Table S18). For instance, very specific and significant eQTL peaks were constantly detected under WW and DS conditions for genes involved in IAA biosynthesis: GRMZM2G048295 (myb15), GRMZM2G163848 (iap3), GRMZM2G045404 (ibr5), sugar metabolism: GRMZM2G111324 (ogh17), GRMZM2G318780 (scs3), GRMZM2G171373 (hk1) and peroxide metabolism: GRMZM2G162688 (sip), GRMZM5G872256 (gs1). In addition, many significant peaks were repeatedly detected for genes encoding TFs that regulate multiple biological processes or stress responses (Fig. 4d). These data strongly indicated that the local regulatory variations have significant effects on the expression of its own. Dynamic significant peaks were detected under DS conditions for genes regulate BR biosynthesis: GRMZM2G472625 (pk), GRMZM2G012391 (p450), protein phosphorylation: GRMZM2G002100 (mapk6), GRMZM2G146553 (cipk3), heat stress response: GRMZM2G428391 (hsp70). Some significant peaks or enhanced significance of the peaks were detected for TF genes under DS conditions (Fig. 4e). Therefore, the local regulatory variants of these genes could be more specific for stress-responsive gene expression regulation in AMP.
ZmcPGM2 contributed to the diversity of HSI i-trait ddT200_R and maize drought tolerance via regulating the changes of sugar contents
To further interpret the findings from GWAS, we tested two genes ZmcPGM2 (cytosolic phosphoglucomutase) and ZmFAB1A (1-phosphatidylinositol-4-phosphate 5-kinase or forms aploid and binucleate cells 1) which are annotated in sugar metabolic pathway and inositol phosphate metabolic pathway, respectively (Fig. 3b,c, Additional file 10: Figure S8a and b). In Arabidopsis, cPGM proteins regulate starch-dependent protein synthesis balance and are required for male and female gametophyte function [42, 43], but they have not been reported in regulation of plant drought tolerance.
The ZmcPGM2 locus (GRMZM2G109383) showed significant (P = 2.57 × 10−7, MLM) association with i-trait ddT200_D40_R (the ratio of second-order derivative of the 880 nm total reflectance under drought stress to second-order derivative of the 880 nm total reflectance under well-water condition at 40 days after sowing) (Fig. 5a). The most significant SNP chr5.S_10856121, which explained 8.4% of the phenotypic variance (Additional file 1: Table S7), was located in the coding region of ZmcPGM2 and had strong LD (R2 > 0.76) with four other less significant SNPs (P < 10−4) (Fig. 5b–d). There are two alleles of SNP chr5.S_10856121 and plants in the maize population with T allele had higher levels of ddT200_D40_R than plants with the G allele (Fig. 5e). A mutant Zmcpgm2, which had a stop mutation at Trp(504) of ZmcPGM2 (Fig. 5f), was obtained from a maize EMS mutant bank [8]. Zmcpgm2 plants were grown under WW and DS conditions and the HSI i-traits ddT200 were captured and calculated (Fig. 5g). We observed that the levels of ratio i-traits ddT200_R were lower in Zmcpgm2 than those in B73 wild type (WT) plants when there was no stress, but the levels of this i-trait were higher in Zmcpgm2 than those in WT plants when the stress was more severe (SM ≤ 15%) (Fig. 5 h), demonstrating a role of ZmcPGM2 in regulation of i-trait ddT200_R.
cPGM reversibly converts glucose-1P to glucose-6P and plays important roles in regulation of sugar biosynthesis [42] (Fig. 5i). Previous studies showed that ddT200 reflects the cellular sugar contents [44]. We investigated the sugar contents of Zmcpgm2 and WT plants treated with or without drought (Additional file 1: Table S19). Under WW conditions, the main sugars showed lower levels in Zmcpgm2 than in WT plants, and drought promoted the levels of all these sugars in both Zmcpgm2 and WT plants, but the changes in all these sugars (ratios of sugar contents under DS/WW conditions) were much higher in Zmcpgm2 than in WT plants (Fig. 5j, k). These results demonstrated important roles of ZmcPGM2 in regulation of maize sugar contents and suggested the consistence of ddT200_R with the changes in sugar contents during maize drought responses.
ZmcPGM2 was also significantly associated with CT i-trait Culm_diameter_700_D98_R (the ratio of stem thickness under DS/WW conditions), and the most significant SNPs were chr5.S_10857363 and chr5.S_10858751 (P = 3.46 × 10−7, MLM), which were completely linked to each other (R2 = 1) and highly linked with Chr5.S_10856121 (R2 = 0.81) (Additional file 10: Figure S8c-f). Plants with allele C had higher levels of Culm_diameter_700_D98_R than those of plants with allele A (from chr5.S_10857363) (Additional file 10: Figure S8g). Under WW conditions, the levels of i-trait Culm_diameter_700_D98_R in WT plants were higher than those in Zmcpgm2 mutants, but after severe stress (SM = 15% or 10%), the levels of this i-trait in Zmcpgm2 mutants were higher than those in WT plants (Additional file 10: Figure S8h-j). The ratios of these i-traits were larger in Zmcpgm2 mutants than in WT plants under both WW and DS conditions (Additional file 10: Figure S8k). These results suggested a role ZmcPGM2 in regulation of relatively higher (~ 10%) maize stem thickness.
ZmcPGM2 expression was inhibited by severe drought stress [45] (Fig. 6a). The SNP chr5.S_10857363 of ZmcPGM2 was significant associated with maize SR (P = 5.6 × 10−3, GLM plus 3PCs), and plants with the A allele showed higher survival rates than those with the C allele [13] (Fig. 6b). These results indicated a role of ZmcPGM2 in regulation of maize drought tolerance. SNP Chr5.S_10856121 had strong LD (R2 > 0.8) with four other less significant SNPs (chr5.S_10855874, chr5.S_10855875, chr5.S_10857363, and chr5.S_10858751). Re-sequencing to the genomic DNA of ZmcPGM2 in the maize populations did not detect more significant genomic variations. Further analyses showed that SNPs Chr5.S_10856121, chr5.S_10857363 and chr5.S_10858751 are synonymous variations, while chr5.S_10855874 and chr5.S_10855875 are located in ZmcPGM2 3′-untranslated region, and showed significant associations with i-trait ddT200_D40_R and SR (Additional file 10: Figure S9a-e), indicating that SNPs chr5.S_10855874 and chr5.S_10855875 could be the potential causal variants that regulate i-traits and drought tolerance. We next used Zmcpgm2 mutants to test a possible role of ZmcPGM2 in maize drought tolerance. Detached leaves from Zmcpgm2 mutants lost water more slowly than WT leaves under dehydration conditions (Fig. 6c). More Zmcpgm2 mutants than WT survived after drought stress (Fig. 6d, e), indicating that Zmcpgm2 mutants were more tolerant to drought and that ZmcPGM2 had a negative role in maize drought tolerance. Although the photosynthetic rate, stomatal conductance, transpiration rate, and water use efficiency (WUE) showed slightly higher levels in WT plants under WW conditions, these indices were significantly higher in Zmcpgm2 mutants after severe drought stress (SM < 15%) (Fig. 6f–i). We deduced that the weaker role of ZmcPGM2 promoted higher WUE and photosynthetic rates under DS conditions, which benefitted maize drought tolerance. The anthesis-silking interval (ASI) is an important maize flowering trait, the shorter the ASI, the better for pollen and silk to meet with each other to produce seeds. We observed that the ASIs of Zmcpgm2 mutants were significantly shorter than those in WT plants under both WW and DS conditions in the field, indicating that ZmcPGM2 could also play important roles in flowering regulation.
ZmFAB1A was a key regulator of i-trait dT233_R and maize drought tolerance
The Arabidopsis FAB1A/B regulates the endomembrane homeostasis of pleiotropic developmental processes and is required for pollen development [46, 47], but their roles in crop stress responses remain elusive. There were 11 SNPs in the ZmFAB1A locus (GRMZM2G132373) that showed significant association with i-trait dT233_D40_R (the ratio of first-order derivative of the 959 nm total reflectance under drought stress to first-order derivative of the 959 nm total reflectance under well-water condition at 40 days after sowing under DS/WW conditions) (Additional file 10: Figure S10a-c, Additional file 1: Table S7). The most significant SNP chr6.S_117795068 (P = 1.51 × 10−6, MLM) explained 7.2% of the phenotypic variance and had high linkage with 10 other significant SNPs (R2 = 0.9) (Additional file 10: Figure S10d). Plants with the allele G of the most significant SNP had higher levels of dT233_D40_R than those with the allele C (Additional file 10: Figure S10e). A premature stop mutant Zmfab1a, which had a stop mutation at Gln (409) (Additional file 10: Figure S10f), was obtained to further verify the function of ZmFAB1A. We grew B73 WT and ZmFAB1A mutant plants under WW and DS conditions and investigated the i-traits dT233 and dT233_R at different growth/stress stages (Additional file 10: Figure S10g). The results showed that the levels of dT233_R were higher in Zmfab1a than in WT plants after slight or severe drought stress (Additional file 10: Figure S10h-j) and demonstrated that ZmFAB1A had a role in regulation of i-trait dT233_R.
The expression of ZmFAB1A was increased under severe drought stress (Additional file 10: Figure S10k). Plants with the G allele had higher survival rates after drought stress than those with the C allele (Additional file 10: Figure S10l), which suggested a role of ZmFAB1A in maize drought tolerance. Re-sequencing the genomic DNA of ZmFAB1A in the maize populations did not detect new significant genomic variations. Analyses to the 11 significant SNPs (tightly linked to each other, R2 = 0.9) showed that 4 were synonymous variations and 7 were missense variations, including chr6.S_117795068 (46Asp/Glu), chr6.S_117795706 (231Asp/Asn), chr6.S_117795706 (592Glu/Val), chr6.S_117795706 (665Ala/Val), chr6.S_117795706 (1020Pro/Arg), chr6.S_117795706 (1072Met/Thr), chr6.S_117795706 (1112Gln/Pro), which could be potential causative variations. We further verified the function of ZmFAB1A in drought tolerance and the results showed that Zmfab1a mutants had higher survival rates than those of WT plants after drought stress (Additional file 10: Figure S10m and n). Moreover, as compared to WT plants, Zmfab1a mutants had higher photosynthetic rates, stomatal conductance, and transpiration rates after drought stress with SM < 20% (Additional file 10: Figure S10o-q), and higher WUE after severe drought stress (SM = 12%) (Additional file 10: Figure S10r). Together, these data demonstrated an important role of ZmFAB1A in regulation of maize photosynthesis, WUE and drought tolerance.
Potential utilization of the candidate genes and i-traits
Genomic selection (GS) is helpful in rapid selection of the superior genetic components that associated with given phenotypes. Because GS utilizes all genetic makers to predict the performance of certain candidates in selection, it is therefore a very useful and effective approach to predict the values of certain genetic makers in breeding [48]. Based on the i-traits collected in this study, we identified more than two thousands of candidate drought-tolerant genes. We performed GS with ridge regression best linear unbiased predictor (RR-BLUP) [49] and Bayes A (Method) to the candidate genes to see the accuracy of their certain combinations in selection of AMP drought-tolerant phenotype survival rates. The randomly selected same amount genes from maize genome (excluded candidate genes) were used in the control analysis. The results showed that the selection accuracies of maize drought tolerance by the candidate genes were significantly higher than those by random genes (Fig. 7a), indicating that these candidate genes could be potential genetic markers in drought-tolerant maize selection and breeding.
To know if the i-traits could be potential biomarkers, we evaluated 1311 ratio i-traits (DS/WW, with significant trait loci associations) in explaining the phenotypic variance of survival rates using a linear stepwise regression model. The results showed that up to 60% of the phenotypic variance in survival rates could be explained by combining 15 i-traits across the 4 time points (Fig. 7b; Additional file 1: Table S20), indicating that these i-traits could be used as markers to select drought-tolerant maize germplasm. Interestingly, 53% of these marker i-traits, including the i-trait ddT200 that was associated with new drought-tolerant gene ZmcPGM2, had wavelengths of 780–1000 nm (Fig. 5; Additional file 1: Table S20). These 15 marker i-traits were further compared with four known spectral indexes including red valley reflectance, green peak reflectance, green peak area, and red edge area, which are widely used in agricultural remote sensing to indicate the chlorophyll or water content, and crop health [50, 51]. The results showed good correlation of these marker i-traits with the four indexes (Fig. 7c–f). For example, 58% of the phenotypic variance of red edge area was explained using two markers A248, ddT200 (Fig. 7f), indicating that these markers reflected the change in chlorophyll or water content and could be used to dynamically monitor drought responses and screen for maize accessions with higher drought resistance. The cross-validation of the observation for survival rates and four spectral indexes are shown in Additional file 1: Table S20.