Using high-throughput multiple optical phenotyping to decipher the genetic architecture of maize drought tolerance

Abstract

Background: Drought threatens the food supply of the world population. Dissecting the dynamic responses of plants to drought will be beneficial for breeding drought-tolerant crops, as the genetic controls of these responses remain largely unknown.

Results: Here we develop a high-throughput multiple optical phenotyping system to noninvasively phenotype 368 maize genotypes with or without drought stress over a course of 98 days, and collected multiple optical images, including color camera scanning, hyperspectral imaging, and X-ray computed tomography images. We develop high-throughput analysis pipelines to extract image-based traits (i-traits). Of these i-traits, 10,080 were effective and heritable indicators of maize external and internal drought responses. An i-trait-based genome-wide association study reveals 4322 significant locus-trait associations, representing 1529 quantitative trait loci (QTLs) and 2318 candidate genes, many that co-localize with previously reported maize drought responsive QTLs. Expression QTL (eQTL) analysis uncovers many local and distant regulatory variants that control the expression of the candidate genes. We use genetic mutation analysis to validate two new genes, ZmcPGM2 and ZmFAB1A, which regulate i-traits and drought tolerance. Moreover, the value of the candidate genes as drought-tolerant genetic markers is revealed by genome selection analysis, and 15 i-traits are identified as potential markers for maize drought tolerance breeding.

Conclusion: Our study demonstrates that combining high-throughput multiple optical phenotyping and GWAS is a novel and effective approach to dissect the genetic architecture of complex traits and clone drought-tolerance associated genes.

Fig. 1

Combining high-throughput phenotyping and GWAS to study maize drought tolerance. a The phenotyping platform and experimental design. Left, the growth of the maize population under WW and DS conditions at D52 in greenhouse; middle and right, the capture of images with RGB, hyperspectral (HSI) and CT scanners under WW and DS conditions at different time point (D25-D98). b HSI, CT, and RGB image analyses and i-traits calculation with pipelines developed in this study. The details of these pipelines are shown in Additional file 9: Note S2 and Additional file 3: Video S2; Additional file 4: Video S3; Additional file 5: Video S4. All the images, phenotypic data, and genotype data are publicly available for reuse with the link: 10.6084/m9.figshare.14429003.v1. c A procedure showing the drought-related i-traits filtering and determining, GWAS, and candidate gene identification / validation

Fig. 2

General analyses of i-traits. a–c Examples showing the levels of RGB-derived (TPA, a), HSI-derived (dT233, b), and CT-derived (Hollow_area_700, c) i-traits, which effectively indicated the levels of drought stress at different time points. PCA of RGB-derived (d) and HSI-derived (e) i-traits collected at time points D34, D40, D46, and D52. f Broad heritability (H²) of all RGB-derived and top 60 HSI-derived i-traits. More detailed H² information is shown in Additional file 10: Figure S2. WW, well-watered; DS, drought-stressed; D, days after sowing

Fig. 3

Associations from I-trait-based GWAS and analysis of the candidate genes. a Gene-trait network showing the distribution of candidate genes and the clustering of genes enriched in the same pathways. I-traits and their related network were shown in the bottom layer. Genes and their enriched pathways are shown in the upper layer. ZmcPGM2 and ZmFAB1A were highlighted. b Genes enriched in the sugar metabolic pathway. ZmcPGM2 that catalyzes the invertible step of Gluc-6P to Gluc-1P was highlighted. c Genes enriched in inositol phosphate metabolic pathway. ZmFAB1A that catalyzes the step of PtdIns3P to PtdIns(3,5)P₂ was highlighted. d Density plot showing the P value distribution of most significant SNPs of the candidate genes and randomly selected genes. Ten thousand times of permutation tests with randomly selected genes were performed and compared to the candidate genes. e Number of i-traits associated with candidate genes

Fig. 4

eQTLs that were associated with the expression of candidate genes. a, b Density plots showing the explained expression variance by significant local or distant eQTLs under WW (a) or DS (b) conditions. c The amounts and their percentages of static and dynamic eQTLs from total, distant, or local eQTL groups. d The local TF eQTLs that constantly detected under both WW and DS conditions. e The local TF eQTLs that specifically detected or enhanced significance under DS conditions

Fig. 5

Roles of candidate gene ZmcPGM2 in regulation of i-traits and sugar biosynthesis. a Zoom in on the view of Manhattan plot of chromosomal 5 region 9.4~12.4 Mb, where there were significant associations of SNPs with i-trait ddT200_D40_R. b–d Distribution of SNPs (b) in gene model ZmPGM2 (c) and their LD to each other (d). The most significant SNP chr5.S_10856121 is highlighted with red dots in b. In panel c, filled black boxes indicate exons and black lines indicate introns of ZmcPGM2. e Plants with the T allele of chr5.S_10856121 showed significantly higher levels of i-trait ddT200_D40_R than those with the G allele in the AMP. f ZmcPGM2 gene structure and position of the EMS mutation. g Growth of B73 wild type and Zmcpgm2 mutant plants under WW and DS conditions. The soil moisture (SM) is shown on the top of the panels. Bar = 20 cm for all plants shown in this panel. h The levels of i-trait ddT200 in B73 wild type and Zmcpgm2 mutant plants under WW, DS and ratio (DS/WW) conditions. The arrows and numbers show the fold decrease or increase in this trait in Zmcpgm2 mutants as compared to those in B73 wild type plants. i cPGM2 is responsible for reversibly converting glucose-1p to glucose-6p in sugar biosynthesis. Adapted and edited based on the KEGG database. Enzymes and their abbreviations: phosphoglucomutase (PGM), UTP-glucose-1-phosphate uridylyltransferase (UGP), UDP-glucose 4-epimerase (UGE), inositol 3-α-galactosyltransferase (IGT), galactinol-sucrose galactosyltransferase (GSGT), α-galactosidase (GTD), sucrose synthase (SUS), sucrose phosphorylase (SPP), glucose-1-phosphate phosphodismutase (GPPD), Hexokinase (HXK), glucose-6-phosphatase (GPP), glucose-6-phosphate isomerase (GPI), invertase (IVT). Arrows indicate the direction of the reaction. Sugars identified with GC-MS in this study are highlighted in red. j Sugar contents of B73 wild type and Zmcpgm2 mutant plants grown under WW and DS conditions. k Fold increase in sugar contents (DS/WW) in B73 wild type and Zmcpgm2 mutant plants. Statistical significance was determined by Student’s t-test: *P < 0.05; **P < 0.01; ***P < 0.001

Fig. 6

Roles of ZmcPGM2 in regulation of maize drought tolerance. a Expression of ZmcPGM2 in maize plants grown under WW or DS conditions. DS2-4 indicates different stress levels. b Plants with different alleles (A/C) of chr5.S_10857363, which showed high LD with chr5.S_10856121 (R² = 0.81), showed significantly different survival rates in the maize population. c Comparison of water loss rate between detached leaves of B73 wild type and Zmcpgm2 mutants. d Growth of B73 wild type and Zmcpgm2 mutant plants under well-watered (WW) and drought-stressed (DS) conditions followed by re-watering. Bar = 20 cm for all plants shown in this panel. e Comparison of the survival rates of B73 wild type and Zmcpgm2 mutant plants after drought stress. f–i Comparison of the photosynthetic rates (f), stomatal conductances (g), transpiration rates (h), and water use efficiencies (WUE, i) of B73 wild type and Zmcpgm2 mutant plants after ceasing watering at different time points. Days indicate the time after irrigation ceased. The embedded graph in (f) indicates the soil moistures (SM) at each time point without irrigation. Statistical significance was determined by Student’s t-test: *P < 0.05; **P < 0.01; ***P < 0.001. j, k Anthesis-silking intervals (ASI) of B73 and Zmcpgm2 mutant plants grown under WW (j) and DS (k) conditions. Means with letters a and b show significantly different by t test (P < 0.05)

Fig. 7

Prediction of maize drought tolerance by candidate genes and i-traits. a Drought tolerance selection accuracies by different amounts of candidate and random genes with RR-BLUP and Bayes A models (see “Materials and methods”). The significances were determined by t test: ***, P < 0.001. b Survival rate predicted by combining 15 i-traits across the 4 time points. c–f Prediction of four known spectral indexes: red valley reflectance (c), green peak reflectance (d), green peak area (e), and red edge area (f), respectively