The 2020 derecho revealed limited overlap between maize genes associated with root lodging and root system architecture

Abstract

Roots anchor plants in soil, and the failure of anchorage (i.e. root lodging) is a major cause of crop yield loss. Anchorage is often assumed to be driven by root system architecture (RSA). We made use of a natural experiment to measure the overlap between the genetic regulation of RSA and anchorage. After one of the most devastating derechos ever recorded in August 2020, we phenotyped root lodging in a maize (Zea mays) diversity panel consisting of 369 genotypes grown in 6 environments affected by the derecho. Genome-wide and transcriptome-wide association studies identified 118 candidate genes associated with root lodging. Thirty-four percent (40/118) of these were homologs of genes from Arabidopsis (Arabidopsis thaliana) that affect traits such as root morphology and lignin content, expected to affect root lodging. Finally, gene ontology enrichment analysis of the candidate genes and their predicted interaction partners at the transcriptional and translational levels revealed the complex regulatory networks of physiological and biochemical pathways underlying root lodging in maize. Limited overlap between genes associated with lodging resistance and RSA in this diversity panel suggests that anchorage depends in part on factors other than the gross characteristics of RSA.

Introduction

Lodging limits crops from fully realizing their yield potential and achieving sustainable yields in a world with a changing climate (Karthikeyan et al. 2020). Estimates of yield losses attributed to lodging in major crops are generally over 20% (Pinthus 1967; Noor and Caviness 1980; Carter and Hudelson 1988; Caierão 2006; Dorairaj et al. 2017). In maize (Zea mays), yield loss due to root lodging can be up to 30% (Lindsey et al. 2021). Lodging typically occurs during windstorms. Stalk lodging, or stalk breakage, typically occurs during late vegetative and reproductive stages (Carter and Hudelson 1988; Shah et al. 2019). Because the morphological changes associated with stalk lodging affect above-ground organs, they can be readily observed and characterized. In addition, stalk strength can be tested under laboratory conditions (Stubbs et al. 2020). Hence, it has been possible to identify the physiological and genetic mechanisms responsible for stalk lodging (Multani et al. 2003; Peiffer et al. 2013; Zhang et al. 2020; Guo et al. 2021).

In contrast, it has proved more challenging to determine the mechanisms responsible for root lodging, i.e. the failure of the root system to successfully anchor the plant in its original upright position. Although substantial progress has been made in biomechanical modeling of root lodging in maize (Ennos et al. 1993; Brune et al. 2018; Berry et al. 2021), the underlying biological mechanisms responsible for resistance to root lodging remain largely unknown. To date, only 1 major quantitative trait locus for root lodging, root-ABA1, has been identified and validated (Giuliani et al. 2005; Landi et al. 2007). In addition, the maize RTCS (rootless concerning crown and seminal roots) gene, which encodes a LOB domain protein (Taramino et al. 2007), is required for the formation of aerial roots and contributes to resistance to root lodging (Woll et al. 2005). Finally, an overexpression of a member of the ARGOS gene family, ARGOS8, has been shown to reduce root-lodging resistance in maize (Shi et al. 2019). This dearth of knowledge is due in part to the difficulties of studying at scale the below-ground root system architecture (RSA). Hence, some studies have focused on the impact of aerial (i.e. brace) roots (Hostetler et al. 2022) or surrogate traits (Tirado et al. 2021; Hostetler et al. 2022) on root lodging. Recently, however, an efficient procedure for excavating root systems has been reported (Zheng et al. 2020). The use of this procedure enabled the identification of genetic markers and genes that contribute to RSA, potentially enabling breeders to select for or generate genotypes with desirable RSAs (Zheng et al. 2020). Even so, relationships between structural features of the RSA and root anchorage, one of the multiple functions of root systems, have not been well characterized. Further, characteristics other than RSA, such as root strength conferred by chemical composition and the distribution within roots of structural compounds such as lignin and cell type, are likely to play important roles in root lodging (Shah et al. 2019). It would, therefore, be desirable to identify markers and genes that are directly associated with root lodging. It is, however, difficult to imagine how root lodging could be tested under laboratory conditions, and extant field-based tests for root lodging (Erndwein et al. 2020) have low throughput and are therefore not suitable for screening the large diversity panels required for association studies.

On August 10, 2020, an intense and fast-moving thunderstorm complex known as a ‘derecho’ swept through the US corn belt, leaving a path of significant damage to crops from Nebraska to Indiana (https://www.washingtonpost.com/weather/2020/08/12/iowa-derecho-corn-damage/). In Iowa, the top corn-producing state in the United States, the derecho damaged 8.2 million acres of corn (maize) and 5.6 million acres of soybean (Glycine max), resulting in more than $7.5 billion of economic loss (https://iowaagriculture.gov/news/updated-derecho-impact-estimates-08142020; https://www.desmoinesregister.com/story/news/2020/10/17/iowas-august-derecho-most-costly-thunderstorm-us-history-7-5-billion-damages/3695053001/). Much of this crop damage was due to stalk and/or root lodging.

Fortuitously, that summer, we were growing a maize diversity panel (N = 369 inbreds in >2,100 plots) that was well characterized for both above-ground and root traits (Leiboff et al. 2015; Zheng et al. 2020). Because these plots were located at the epicenter of the derecho in Story and Boone Counties of Iowa (Hosseini et al. 2020), they suffered significant amounts of root lodging. By conducting genome-wide and transcriptome-wide association studies (TWASs) using phenotypic data from this natural experiment, we identified numerous candidate genes that expand our understanding of those aspects of root systems that contribute to resistance to root lodging.

Results

Phenotypic variation of root lodging

The field evaluation of root lodging (Fig. 1A; Materials and methods) of subsets of the 369 inbred lines from the SAM diversity panel (Leiboff et al. 2015) across 2 locations, each with 3 different planting schemes [in total of 6 planting density/plot type (2- vs. 4-row plot)/location combinations, each combination hereafter referred to as “environment,” see Materials and methods and Supplemental Table S1 for details] started on the second day after the derecho (i.e. August 11, 2020) to minimize the impact of “goose-necking” (Zhang et al. 2011) on the accuracy of phenotyping, and was completed within the same week. Substantial phenotypic variation for root lodging was observed among genotypes in the diversity panel for each of the 6 environments (Fig. 1B), with an overall narrow-sense heritability of 0.44 (Supplemental Tables S2 and S3). The phenotypic correlations of root lodging between environments ranged from low to moderate (Supplemental Fig. S1 and Table S3), consistent with the assumption that root lodging is heavily influenced by local environmental factors such as soil type, soil moisture, wind speed, wind direction, etc. (Berry et al. 2021). Further, we explored the phenotypic correlation of root lodging with a variety of published above-ground traits (Leiboff et al. 2015), RSA traits (Zheng et al. 2020), root pulling force (RPF) data (Woods et al. 2022), and previously unpublished brace root trait data we collected from prior years (Supplemental Text S1 and Table S4). The correlations of these traits with root lodging were generally low (i.e. Pearson correlation coefficients of between −0.22 and 0.26; Supplemental Table S5 and Fig. S1). Further, these above-ground and root-associated trait data collectively explain only ∼45% of the phenotypic variance for root lodging. By considering not only these trait data but also genetic marker data, it is possible to increase the fraction of phenotypic variance for root lodging that can be explained to at least 75% (Supplemental Text S2 and Table S6). This finding provides evidence that biochemical, physiological, and/or morphological pathways not closely associated with the traits described above contribute to resistance to root lodging. To identify these pathways and the genes that contribute to them, we conducted 2 types of association studies.

Figure 1.

Field evaluation of root lodging after the 2020 derecho. A) Quantification of root lodging with a protractor. The root-lodging (RL) scores ranges from 1 to 5, and the angle of deflection from vertical was converted to a 1-to-5 score, with 1 being unaffected and 5 being completely root lodged. B) Distribution of root-lodging scales (RL scale) across 6 environments after spatial correction. This correction results in some plots having RL scale values >5. See Materials and methods for details about the conversion from RL scores to RL scales. Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; points, outliers. Sample sizes for the 6 environments ranges from 326 to 341 (see Materials and methods for details).

Genome-wide association studies for maize root lodging

A linear mixed model that accounts for individual relatedness and population structure (Yu et al. 2006) was used for genome-wide association studies (GWAS). For the SAM diversity panel, ∼800,000 single-nucleotide polymorphisms (SNPs) generated by combining reads from RNA-seq and GBS data were used as genetic markers (Leiboff et al. 2015). To address possible confounding factors for root lodging, such as the heights of a plot and its neighboring plots, field layout with regard to wind direction, etc., spatial correction was conducted for root-lodging trait values to isolate the genomic effect (see Materials and methods) that went into the association studies. In total, 38 significant root lodging–associated SNPs were identified when using a false discovery rate (FDR) cutoff of <0.05. To eliminate redundancy, only 1 SNP was retained from each 10 kb window, resulting in 21 unique SNPs (Supplemental Table S7). Twenty kilobase windows centered on each of these 23 SNPs were scanned for candidate genes. In total, 19 candidate genes (up to 1 per window) were identified (Supplemental Fig. S2 and Table S7), 47% (9/19) of which are homologs of Arabidopsis (Arabidopsis thaliana) genes known to affect lignin content and/or root traits, or which have been linked to physiological pathways and/or biochemical composition associated with root lodging in rice (Oryza sativa) and wheat (Triticum aestivum; (Supplemental Table S8; Shah et al. 2019). For example, Zm00001eb330790, a maize homolog of the Arabidopsis Mediator complex subunit MED5b (also known as RFR1), was associated with root lodging (Fig. 2A). MED5b along with MED5a can rescue the lignin deficiency and alter the gene expression profile of the lignin-related phenylpropanoid pathway mutant ref8 (Bonawitz et al. 2014). Another maize gene, Zm00001eb031980 was also associated with root lodging. Its Arabidopsis homolog is a large subunit GTPase 1 (AtLSG1-2), which is essential to ribosome biogenesis, affects auxin homeostasis, and is responsible for the dig6 (drought inhibited growth of lateral roots) mutant phenotype, which includes reduced lateral root numbers and shorter roots (Zhao et al. 2015).

Figure 2.

Manhattan plots of GWAS and TWAS for root lodging at Kelley 2-row low-density plots. A) Linear mixed model GWAS. B to D) TWAS using transcriptome data from (B) SAM-enriched tissue from 14 DAP seedlings, (C) SAM-enriched tissue from 5 DAP seedlings, and (D) root tip from 5 DAP seedlings.

eRD-GWAS for root lodging using transcriptome data from multiple tissues

TWASs directly associate variation in gene expression with phenotypic variation. These analyses are complementary to and less affected by linkage disequilibrium (LD) than conventional GWAS (Lin et al. 2017; Li et al. 2021). Furthermore, utilizing RNA-seq data from multiple tissues in TWAS, especially those relevant to the trait of interest, favors the identification of candidate genes with characteristics of true positives (Li et al. 2021).

Three different sources of transcriptome data from the SAM diversity panel (Materials and methods) were used to conduct eRD-GWAS, a Bayesian implementation of TWAS (Lin et al. 2017) on root lodging. From the 6 analyses (Supplemental Table S1), 99 unique candidate genes were identified using an arbitrary model frequency cutoff of >0.03 (Supplemental Table S9). Among these, 31% (31/99) have homologs in Arabidopsis that affect lignin content, root traits, and root traits that exhibit responses to abiotic stress (Supplemental Table S8 and Fig. S2). Three candidate genes, each of which was associated with root lodging via eRD-GWAS, are discussed below (Fig. 2, B and C). Zm00001eb008060 is a homolog of the Arabidopsis gene, Cop9 signalosome subunit 4 (AtCSN4, Fig. 2C). A mutant allele of AtCSN4 affects adventitious and lateral roots in other Arabidopsis root mutants, potentially by altering auxin signaling (Pacurar et al. 2017). Zm00001eb099150 (Supplemental Fig. S2) is a homolog of the Arabidopsis gene AtVIT1 (Gollhofer et al. 2014). A mutant of AtVIT1 has decreased root length compared with wild-type Arabidopsis plants. Finally, Zm00001eb096900 (Supplemental Fig. S2) is a homolog of the Arabidopsis gene AtCESA6, a mutant of which exhibits reduced root length, among other phenotypes (Hématy et al. 2007).

Gene ontology term enrichment

To further profile the functionality of the candidate genes identified for root lodging via the association studies described above, gene ontology (GO) enrichment analysis was conducted. We expanded the number of genes of interest by including the direct interaction partners of the root-lodging candidate genes on the transcriptional and translational levels: predicted interaction partners of the candidate genes from a maize gene regulatory network (GRN; Walley et al. 2016), referred as the transcriptional candidate gene set (N = 714 unique maize genes) and protein–protein interaction network (PPIN; Zhu et al. 2016), referred to as the translational candidate gene set (N = 1,069) were extracted (Supplemental Table S10) and separately subjected to GO enrichment analysis using Cytoscape's (Shannon et al. 2003) plugin BiNGO (Maere et al. 2005; Materials and methods). Using the hypergeometric test at FDR of 0.05, 497 GO categories and 210 GO categories were identified for the transcriptional candidate gene set and the translational gene set, respectively (Supplemental Table S11). Among the significantly enriched GO terms, some biological processes and pathways involved in root lodging from studies in crops such as rice and wheat (Shah et al. 2019) were identified (Fig. 3). These GO terms include: (i) root development and subcategories; (ii) lignin and hemicellulose biosynthetic and metabolic processes; (iii) hormones involved in root lodging such as auxin, gibberellin, and ethylene signaling pathways (Shah et al. 2019); and (iv) responses to stress and subcategories. These examples suggest the existence of intertwined physiological and biochemical regulatory networks working at both the transcriptional and translational levels that confer lodging resistance.

Figure 3.

Representative GO pathways that were significantly (FDR < 0.05) enriched among GWAS/TWAS candidate genes and their direct GRN and PPIN interacting patterners. Biological processes and pathways involved in stress response, root morphology, lignin and hemicellulose biosynthesis, and key hormones involved in root lodging including auxin, ethylene, and gibberellin are presented.

Discussion

Although root lodging causes substantial annual yield losses across the globe, for reasons discussed earlier, our understanding of the genetic basis of this trait lags behind that of other abiotic stresses. The devastating 2020 derecho provided a natural experiment that enabled us to directly associate markers and genes with root lodging. Although Hostetler et al. (2022) similarly measured root lodging after a storm, their experiment included such a limited number of genotypes that it was not possible to conduct association studies. In addition, brace roots and other published root-related traits explain only a limited percentage of the phenotypic variance for root lodging. In contrast, because our natural experiment involved 369 genotypes grown in >2,100 plots, we could begin to describe the complex genetic architecture of root lodging by associating DNA sequence variation and variation in gene expression levels with phenotypic variation. Despite the relatively moderate heritabilities of root lodging, we identified high-confidence candidate genes with characteristics of true positives. For example, 34% of candidate genes had homologs in other plant species that are known to have functions related to RSA and hence presumably to root lodging. Similarly, the candidate genes identified via GWAS and TWAS that are associated with lignin biosynthesis highlight the importance of composition for root lodging. Further, the candidate genes and their predicted GRN and PPIN interaction partners are enriched in morphological, physiological, and biochemical pathways, consistent with pathways associated with root lodging–related traits in other crop species (Shah et al. 2019), providing opportunities for further basic research on root biology. Consequently, our candidate genes have the potential to serve as breeding targets to confer resistance to root lodging in a world in which extreme wind events such as derechos are expected to occur at higher frequencies (Mudd et al. 2014).

Among the 118 candidate genes associated with lodging resistance via GWAS and TWAS in this study, only 1 (gene ID: Zm00001eb335120 in RefGenV5 or GRMZM2G444567 in RefGenV2) was found to have been previously associated with RSA in the same diversity panel, but grown in a different environment (Supplemental Tables S6 and S8 of Zheng et al. 2020). Consistently, when comparing the 118 root-lodging candidate genes from the current study with candidate genes identified in another GWAS on different sets of RSA traits using a different diversity panel grown in a different environment (Supplemental Table S8 of Ren et al. 2022), no overlapping candidate genes were identified. This suggests that anchorage (the focus of the current study) depends at least in part on factors other than gross characteristics of RSA analyzed by the 2 earlier studies of RSA. These factors could include, for example root characteristics such as root strength (which could be a function of the chemical composition of roots and/or the spatial distribution of structural compounds within roots) and fine features of RSA not measured in either of the 2 earlier studies, as well as above-ground plant features such as canopy architecture, stem flexibility, etc. (Shah et al. 2019). For root strength, we compared the overlap between the 118 candidate genes identified in the current study with those from a GWAS for RPF, a proxy trait that reflects root strength under different irrigation treatments (Supplemental Table S4 of Woods et al. 2022); only a single candidate gene Zm00001eb191650 was identified in both studies. This gene is a maize homolog of an Arabidopsis gene, AtPHO1 involved in both phosphate and cytokinin signaling pathways; a mutant of this gene exhibits reduced root length when compared with wild type (Lan et al. 2006).

It is noteworthy that although there are significant correlations between RPF and root-lodging phenotypes in some environments (Supplemental Table S5), there is limited overlap between the candidate genes associated with these 2 traits, even though the 2 studies were conducted on the same diversity panel. This may be due to the substantial differences in growing conditions (including soil characteristics) and growth stages at which these traits were phenotyped. Such differences could reduce statistical power and/or raise genotype × environment (G × E) issues that would complicate the coidentification of genes from the 2 studies, even if those genes affect both traits. Alternatively, it is possible that some finer features of RSA that were not captured by Zheng et al. (2020) and Ren et al. (2022) could contribute to root-lodging resistance. The phenotyping methods for both studies were predominantly focused on the crown root system in the top 30 cm of soil. However, the complexity and dynamics of the maize root system, to which many genetic and biochemical pathways contribute (as summarized in Hochholdinger et al. 2018) cannot be fully captured by current phenotyping methods. For example, the maize mutant, rum1, only affects seminal and lateral root initiation without affecting shoot-borne crown roots (von Behrens et al. 2011). Genotypes may have substantial phenotypic variation in seminal and lateral root number and density, which could lead to variation in root-lodging resistance, but they are not distinguishable in terms of the overall RSA structure as measured using current high-throughput phenotyping technologies. In addition, other fine features of RSA, such as numbers, distributions, and lengths of root hairs, which dramatically increase the contact surface area of the root system with the surrounding soil were not phenotyped during the cited association studies (Wen and Schnable 1994), despite the fact that they could reasonably affect root-lodging resistance under extreme conditions such as those experienced during a derecho. Lastly, given the difficulties associated with high-throughput empirical studies of root lodging, modeling studies involving above- and below-ground plant architectures and root strength may provide an efficient path toward hypothesis generation.

A genomic prediction model developed for root lodging across all 6 environments had a prediction accuracy of 0.25, similar to traits such as maize grain yield (Supplemental Text S3). Of course, the optimal RSA for resistance to root lodging may not be optimal for other agronomic traits such as nutrient and water uptake, yield, and soil-based carbon sequestration. As such, breeders may need to rely on multi-objective optimization methods, such as those proposed by Moeinizade et al. (2020).

Materials and methods

Germplasm and field experimental design

Subsets of the 369 maize (Z. mays) genotypes from the SAM diversity panel (Leiboff et al. 2015) along with checks were planted at the Bennett Farm (41.99 N, 93.68 W) and the Kelley Farm (42.05 N, 93.72 W) of Iowa State University, Ames, IA, USA. In each of the 2 locations, these inbred genotypes were planted under different planting densities and plot types (2- vs. 4-row plot). Each planting density/plot type/location combination is referred to as an “environment.” Within each environment, only a single fully randomized replication was planted. These entries were blocked by plant height with checks included (for details, see Supplemental Table S1 and Text S4).

Phenotyping for root lodging

The phenotyping of root lodging was conducted within the week of August 11, 2020. The severity of root lodging was assessed using a customized protractor scale from 1 to 5 (Fig. 1A), with 1 being fully vertical plants (i.e. with no root lodging) and 5 being fully horizontal plants (i.e. completely root lodged). The plants were scored on a per-plot basis, i.e. recording the most and second-most prevalent root-lodging scales and the corresponding percentages of the plants in each plot having those scores. When designing the scale, the plant's angle with the ground was targeted to be measured to minimize the impact of “goose-necking” (Zhang et al. 2011) on the measurement of root lodging, especially in the later stages of phenotyping after the first few days.

Statistical modeling

A Bayesian hierarchical multinomial probit model for ordered categorical data (Munkin and Trivedi 2008; Gelman et al. 2014) was used to compute the spatially and covariate adjusted genomic mean log-lodging angles (genomic effects). Specifically, the first prevalent lodging scores were linked to latent angles of lodging (angle to the vertical line) through a probit link function, and the log-transformed lodging angles were modeled through a mixed linear model with genotypes, percentage of lodging of first the prevalent score, and the heights of the plants in the 8 neighboring plots as effects. The genotypic and covariates’ effects were assumed to be Gaussian with inverse gamma priors on the variance components (see Supplemental Text S4 for details). The model was fitted for each of the 6 environments (Kelley and Bennett; 2-row plots with low density and 4-row plots with low or high densities, Supplemental Table S1) through Gibbs sampler (Wang and Roy 2018) and posterior means of the genotypic effects were used for subsequent analyses.

Tissue collection, library construction, and RNA sequencing

For the 369 genotypes from the SAM diversity panel (Leiboff et al. 2015), 15 seeds for each genotype were grown in paper rolls in Percival A100 growth chambers (Percival Scientific, Perry, IA, USA) with a 16 h light (28 °C) and 8 h dark (21 °C) cycles. After 5 d of germination, SAM-enriched tissue and root tips of 1 cm from 3 to 10 seedlings were pooled for each genotype for RNA extraction. All RNA extractions were conducted using Qiagen RNeasy Plant Mini Kits (Qiagen, Hilden, Germany) following the manufacturer's protocol. RNA for all the samples was diluted with 30 μL of RNase free water and was subjected to initial quality control using Qubit RNA Integrity and Quality Assay Kits (Thermo Fisher Scientific, Waltham, MA, USA). RNA samples with concentrations >100 ng μL⁻¹ were used to construct RNA-sequencing libraries. Indexed RNA-seq libraries were prepared from RNA samples from the SAM-enriched tissue using the TruSeq Stranded mRNA Kit (Illumina, San Diego, CA, USA) and sequenced using NovaSeq6000 S4 (2 × 150 bp) instrument(s). Indexed RNA-seq libraries were prepared from RNA samples from root tips using QuantSeq 3′ mRNA-seq Library Prep FWD for Illumina (Lexogen, Vienna, Austria), and sequenced using NovaSeq6000 S2 (100-cycle single-end) instrument(s). RNA-seq reads were processed and aligned to maize RefGen V5 (Hufford et al. 2021) as described in previous studies with the same diversity panel (Leiboff et al. 2015; Lin et al. 2017). Conversions of gene IDs among different RefGen versions to enable comparisons among studies were conducted using the “Translate Gene Model IDs” functional module of MaizeGDB (https://www.maizegdb.org/gene_center/gene#xref).

Linear mixed model GWAS for root lodging

For the GWAS, ∼800,000 SNPs for up to 366 genotypes of the SAM panel combined from RNA-seq and GBS (Leiboff et al. 2015) filtered for minor allele frequency >0.01 were used as genetic markers. GWASs were conducted using linear mixed models implemented in GEMMA (Zhou and Stephens 2012). The first 3 principal components and kinship matrix calculated using TASSEL 5.0 (Bradbury et al. 2007) were included in the GWAS model to control for population structure and individual relatedness. GWASs of root lodging were conducted separately for 2- and 4-row plots under each density for each environment (Supplemental Table S1). A FDR cutoff (Benjamini and Hochberg 1995) of 0.05 was used to select significant trait-associated SNPs (TAS). LD values of the SAM panel were calculated using PLINK v1.90 (Purcell et al. 2007). Based on the average LD of the panel, a 20-kb window centered at each TAS was used to search for candidate genes. Only a single gene among all genes within each 20-kb window was selected as a candidate gene. This selection was based on the relevancy of gene's functional annotation: i.e. if the predicted functional annotation of a gene was reported in root-lodging studies in maize and other crop species (Shah et al. 2019), that gene was assigned a higher priority than other genes in the same window. If none of the genes within a window had a relevant functional annotation, the gene closest to the TAS was selected as a candidate gene.

eRD-GWAS for root lodging

In addition to the SNP-based GWAS, eRD-GWAS (Lin et al. 2017), a Bayesian-based implementation of TWAS, was used to identify genes associated with root lodging in the 6 environments described above (Supplemental Table S1). RNA-seq data from 3 different tissue/developmental stages, namely, 14-d-after-planting (DAP) SAM-enriched tissue (Leiboff et al. 2015) for up to 366 genotypes of the SAM panel; 5 DAP SAM-enriched tissue for up to 324 genotypes; and 5 DAP root tips for up to 334 genotypes, were used as explanatory variables to gain more comprehensive insights into cross-tissue/stage gene expression's impact on phenotypic variation (Li et al. 2021). An arbitrary cutoff of model frequency >0.03 was used for candidate gene discovery.

GRN and PPIN

Over a million predicted transcription factors and potential target gene pairs were downloaded from a maize GRN quantified at the protein level (Walley et al. 2016). Over 2.7 million predicted and experimentally validated protein–protein interactions were downloaded from a maize PPIN (Zhu et al. 2016). Candidate genes detected from linear mixed model GWAS and eRD-GWAS, along with their high-confidence, direct (first-degree) interaction partners from the GRN and PPIN were extracted from the full networks and tested for GO term enrichment. Arabidopsis (A. thaliana) putative orthologues of the maize genes mentioned above were retrieved from Gramene Mart (http://www.gramene.org; release 65, accessed June 9, 2022). GO annotations of A. thaliana were downloaded from ftp://ftp.arabidopsis.org/home/tair/Ontologies/Gene_Ontology/(accessed June 9, 2022). GO enrichment tests were conducted by using a hypergeometric test and a <0.05 FDR cutoff using the BiNGO (Maere et al. 2005) plugin of Cytoscape (Shannon et al. 2003).

Accession numbers

RNA-seq data are available through the NCBI sequence read archive accession PRJNA935698.

Acknowledgments

The authors thank Ms Lisa Coffey (Schnable Lab) for coordinating field studies and Ms Lauren Docherty (former member of the Schnable Lab) for assistance with collecting phenotypic data; Dr Lakshmi Attigala and Ms Sasha Fetty (both former members of the Schnable Lab) for generating some of the RNA-seq data; Drs Aaron Kusmec (Schnable Lab) and Dan Nettleton (Department of Statistics, Iowa State University) for useful discussions.

Author contributions

Z.Z. and P.S.S. conceived and designed the experiments; Z.Z. and H.L. collected the data; Z.Z., B.G., S.D., V.R., H.L., and P.S.S. analyzed and interpreted the data; Z.Z. and P.S.S. wrote and edited the manuscript, with inputs from B.G., S.D., and V.R.

Supplemental data

The following materials are available in the online version of this article.

Supplemental Figure S1. Correlations of root-lodging scales with above-ground and root traits.

Supplemental Figure S2. Manhattan plots for GWAS and eRD-GWAS.

Supplemental Text S1. Collection of brace root trait data.

Supplemental Text S2. Phenotypic correlation between root lodging and root traits and above-ground traits.

Supplemental Text S3. Genomic prediction for root lodging.

Supplemental Text S4. Statistical model for estimating genetic effects of root lodging.

Supplemental Table S1. Experimental design for each environment.

Supplemental Table S2. Narrow-sense heritability of root lodging for each environment.

Supplemental Table S3. Root-lodging scales of the SAM panel across 6 environments and correlations.

Supplemental Table S4. Brace root traits of the SAM panel.

Supplemental Table S5. Correlation between root-lodging scales with above-ground and root traits.

Supplemental Table S6. Phenotypic variance of root-lodging scales explained by root and above-ground traits collectively and PCs calculated from SNPs.

Supplemental Table S7. Significant trait-associated SNPs and candidate genes.

Supplemental Table S8. Annotations of candidate genes.

Supplemental Table S9. Candidate genes from eRD-GWAS.

Supplemental Table S10. Maize root-lodging candidate genes and predictive partners in GRN and PPIN.

Supplemental Table S11. Significant GO categories for the transcriptional (GRN) and translational (PPIN) gene sets.

Funding

The information, data, or work presented herein was funded in part by the Advanced Research Projects Agency-Energy (ARPA-E), US Department of Energy, under Award Number DE-AR0000826. This work was partially supported by Plant Breeding for Agricultural Production program grant no. 2020-68013-30934/project accession no. 1022368 from the USDA National Institute of Food and Agriculture. This work was partially supported by the USDA National Institute of Food and Agriculture, Hatch project IOW03717, and also partially supported by the Plant Sciences Institute at Iowa State University.

Author notes

Conflict of interest statement. The authors declare no conflict of interest.