The genetic architecture of maize height

Abstract

Height is one of the most heritable and easily measured traits in maize (Zea mays L.). Given a pedigree or estimates of the genomic identity-by-state among related plants, height is also accurately predictable. But, mapping alleles explaining natural variation in maize height remains a formidable challenge. To address this challenge, we measured the plant height, ear height, flowering time, and node counts of plants grown in >64,500 plots across 13 environments. These plots contained >7300 inbreds representing most publically available maize inbreds in the United States and families of the maize Nested Association Mapping (NAM) panel. Joint-linkage mapping of quantitative trait loci (QTL), fine mapping in near isogenic lines (NILs), genome-wide association studies (GWAS), and genomic best linear unbiased prediction (GBLUP) were performed. The heritability of maize height was estimated to be >90%. Mapping NAM family-nested QTL revealed the largest explained 2.1 ± 0.9% of height variation. The effects of two tropical alleles at this QTL were independently validated by fine mapping in NIL families. Several significant associations found by GWAS colocalized with established height loci, including brassinosteroid-deficient dwarf1, dwarf plant1, and semi-dwarf2. GBLUP explained >80% of height variation in the panels and outperformed bootstrap aggregation of family-nested QTL models in evaluations of prediction accuracy. These results revealed maize height was under strong genetic control and had a highly polygenic genetic architecture. They also showed that multiple models of genetic architecture differing in polygenicity and effect sizes can plausibly explain a population's variation in maize height, but they may vary in predictive efficacy.

Keywords: GBLUP; height; maize; plant.

Figure 1

Distribution of plant height (PHT) line values within and between RIL families. Asymmetric transgressive segregation about mid-parent value was observed for PHT line values in many of the NAM and IBM families. Similar trends were noted for ear height (EHT), days to pollen shed (DTA), and node counts (NPH) (Figure S1).

Figure 2

Partitioning variation in PHT, EHT, DTA, and NPH. PHT, EHT, DTA, and NPH variation was attributed to genetic and environmental factors across the NAM families (A) and within the NCRPIS diversity panel (B).

Figure 3

Correlations of estimated PHT, EHT, DTA, and NPH line values. Pairwise trait correlations across all NAM families (A) and within the IBM family (B). Trait correlations within each of the NAM families were similar (Figure S2). The NCRPIS diversity panel differed, displaying higher correlations between PHT and DTA than PHT and EHT (C).

Figure 4

Correlations of estimated PHT, EHT, DTA, and NPH family-nested QTL effects. Allele effects for family-nested QTL from 100 models were estimated for PHT, EHT, DTA, and NPH. Correlations among estimated allele effects for each trait were then calculated. Family-nested QTL for EHT, DTA, and NPH were also estimated for all traits (Figure S4).

Figure 5

NIL families support of CML277 and CML333 alleles at a RIL family-nested QTL. Two families of recombinant lines with introgressions of CML277 (A) and CML333 (B) on chromosome 9 in a B73 background were queried for association with PHT to validate a NAM family-nested QTL (RefGenV1 Chr 9: 98,502,843) with significant allele effects when mapped independently in B73 × CML277 and B73 × CML333 RIL families. Within NIL families, t-tests for PHT associations using genotyping by sequencing (GBS) marker variants and Kaspar assays across a region of 10 Mb supported allele effect estimates. The smaller effect of the CML333 allele relative to CML277 also concurred within the families. Three associations (RefGenV1 Chr 9: 97,520,280; 100,367,415; and 100,371,640) identified to increase PHT relative to B73 by GWAS and segregating between B73 and CML277 or CML333 were also in the region.

Figure 6

Prediction of PHT across RIL families and within the NCRPIS diversity panel. A random sample of 20, 40, 60, and 80% of RILs in all NAM families calibrated GBLUP and family-nested QTL models to predict PHT variation of RILs not employed in calibration (A). All RILs were used to calibrate GBLUP and family-nested QTL models to explain their PHT variation (B). In the IBM family, random samples of RILs calibrated GBLUP and QTL models to predict PHT variation of the remaining RILs (C). All RILs in the IBM family were used to calibrate GBLUP and QTL models to explain PHT variation (D). Random samples of inbreds in the NCRPIS diversity panel calibrated GBLUP models to predict PHT variation of inbreds not employed in calibration (E). All inbreds in the NCRPIS diversity panel were used to calibrate a GBLUP model to explain their PHT variation (F). Similar levels of prediction accuracy were observed in GBLUP for EHT, DTA, and NPH (Figure S6).

Figure 7

Prediction of PHT within and between RIL families. The PHT explanatory ability of GBLUP within family (main diagonal) and prediction accuracy of models calibrated from one family and used to predict another were assessed (off diagonal). The nonparanthetical number of each off-diagonal element details the prediction accuracy of that row’s family when used to predict that column’s family. The number in parentheses details the prediction accuracy of that column’s family when used to predict that row’s family. Families are denoted by their unshared parent and ordered based upon clustering of their between-family prediction profiles across the families.