Dissection of Leaf Angle Variation in Maize through Genetic Mapping and Meta‐Analysis

Abstract

Maize (Zea mays L.) hybrids have transitioned to upright leaf angles (LAs) over the last 50 yr as maize yields and planting densities increased concurrently. Genetic mapping and a meta‐analysis were conducted in the present study to dissect genetic factors controlling LA variation. We developed mapping populations using inbred lines B73 (Iowa Stiff Stalk Synthetic), PHW30 (Iodent, expired plant variety protection inbred), and Mo17 (Non‐Stiff Stalk) that have distinct LA architectures and represent three important heterotic groups in the United States. These populations were genotyped using genotyping‐by‐sequencing (GBS), and phenotyped for LA in the F₂ and F_2:3 generation. Inclusive composite interval mapping across the two generations of the mapping populations revealed 12 quantitative trait loci (QTL), and a consistent QTL on chromosome 1 explained 10 to 17% of the phenotypic variance. To gain a comprehensive understanding of natural variations underlying LA variation, these detected QTL were compared with results from 19 previous studies. In total, 495 QTL were compiled and mapped into 143 genomic bins. A meta‐analysis revealed that 58 genomic bins were associated with LA variation. Thirty‐three candidate genes were identified in these genomic bins. Together, these results provide evidence of QTL controlling LA variation from inbred lines representing three important heterotic groups in the United States and a useful resource for future research into the molecular variants underlying specific regions of the genome associated with LA variation.

Abbreviations
• BSSS
• Iowa Stiff Stalk Synthetic
• GBS
• genotyping-by-sequencing
• LA
• leaf angle
• LOD
• logarithm of odds
• NSS
• Non-Stiff Stalk
• PVP
• Plant Variety Protection
• QTL
• quantitative trait loci
• SNP
• single nucleotide polymorphism

Core Ideas

• Twelve QTL were discovered for maize leaf angle variation in the lower canopy.
• A consistently detected QTL explained 10 to 17% of the phenotypic variance on chromosome 1.
• Meta-analysis identified 58 genomic bins containing 33 candidate genes.

For the past 50 yr, maize yields and planting densities in the United States have increased concurrently. A comparative analysis of U.S. commercial maize hybrids released since the 1960s revealed that by selecting high yielding hybrids under high planting densities, breeders indirectly selected hybrids with upright LAs (Duvick et al., 2004). Similar results from other maize materials were obtained (Wang et al., 2011; Ma et al., 2014).

Simulation (Duncan et al., 1967; Hammer et al., 2009) and empirical evidence (Lee and Tollenaar, 2007; Zhao et al., 2015) suggest that upright LAs combined with higher planting densities improve light distribution within the canopy. For example, modern hybrids intercept 14% more light than older hybrids (Lee and Tollenaar, 2007), and the increased light intercepted from leaves near the ear is positively correlated with grain yield (Ma et al., 2014; Zhao et al., 2015). Furthermore, modeling has suggested that LAs should gradually transition from upright in the upper canopy to less upright in the lower as this configuration would distribute solar energy across multiple leaves while maximizing the total solar energy absorbed by the canopy (Duncan et al., 1967; Zhu et al., 2010). A recent study indicated that LA in the upper canopy was negatively correlated with grain yield, while LA in the lower canopy was positively correlated (Zhang et al., 2017), confirming the earlier suggestion about canopy configuration on grain yield.

Given LA's role in the adaptation of maize hybrids to high planting densities, it is beneficial to understand the genetic control of LA variation. Five maize genes that modify LA for multiple leaves in the canopy have been cloned: liguleless1 (lg1, Moreno et al., 1997), liguleless2 (lg2, Walsh et al., 1998), ZmTAC1 (Ku et al., 2011), CLA4 (Zhang et al., 2014), and nana plant2 (Best et al., 2016). These genes exhibit different types of dominance; for example, lg1 (Becraft et al., 1990) and lg2 (Harper and Freeling, 1996) are recessive mutations that result in upright LAs, whereas CLA4 exhibits incomplete dominance (Zhang et al., 2014).

Prior research has identified many QTL controlling LA variation in maize by phenotyping different leaves in the canopy, particularly leaves below the flag leaf (final leaf) and surrounding the ear. However, little research has investigated LA in the lower canopy, specifically below the ear. Previous studies used a meta-analysis to integrate a subset of reported QTL to identify multiple meta–QTL for LA and other canopy architecture traits (Ku et al., 2012; Wang et al., 2016; Zhao et al., 2018). Several known genes physically mapped near these meta–QTL, yet a large-scale meta-analysis for only LA to facilitate separating the genetic control of LA from other canopy architecture traits has not been conducted.

The US Plant Variety Protection (PVP) Act of 1970 allows private companies to protect the proprietary inbred lines they develop for a set period. As the protection expires for these inbred lines, they become publically available to explore their genetic diversity. These ex-PVP lines have attracted a lot of attention, with prior research suggesting that today's germplasm can be traced back to contributions from seven progenitor lines that include public inbred lines B73 and Mo17 and private inbred line PH207 (Mikel and Dudley, 2006; Nelson et al., 2008; Mikel, 2011). These inbred lines represent three important heterotic groups in maize: Iowa Stiff Stalk Synthetic (BSSS), Non-Stiff Stalk (NSS), and Iodent, respectively (Mikel and Dudley, 2006; Mikel, 2011). Investigating LA in lines representing these heterotic groups provides us an opportunity to identify significant regions of the genome that have contributed to modern hybrid's transition to upright LAs.

In this study, we report the discovery of 12 QTL linked to LA in the lower canopy by genetic linkage mapping. We first developed biparental populations using three different parents (Fig. 1). Selected F₂ lines were genotyped using GBS, and both the selected F₂ lines and F_2:3 families were phenotyped for LA. Inclusive composite interval mapping was conducted to map QTL for LA in both generations. To gain a comprehensive understanding of natural variations underlying LA, we integrated our findings with other reported QTL into genomic bins and conducted a meta-analysis to identify regions of the genome associated with LA variation across a wide range of germplasm.

Fig. 1. Distinct leaf angle architecture for inbred lines B73, Mo17, and PHW30. An illustration depicting leaf angle phenotypes for three different leaves in the canopy for B73 (left), PHW30 (middle), and Mo17 (right). The phenotyped leaf was determined as the number of leaves above or below the ear leaf on each plant. The second leaf below the ear leaf was phenotyped as it contrasted the most among the three parents.

MATERIALS AND METHODS Genetic Materials

Genetic mapping populations were developed using three inbred lines with distinct LA architectures (Fig. 1) that represent three important maize heterotic groups. Inbred line B73 is a BSSS inbred released from Iowa State University in 1972 (Russell, 1972). It has broad leaves, upright LAs in the upper canopy, and slightly less upright LAs below the ear (Fig. 1). Inbred line PHW30 is an ex-PVP Iodent inbred (PVP number 9100102) developed by Pioneer Hi-Bred in the late 1980s. It has a distinct plant architecture with thin wavy leaves and upright LAs throughout the canopy (Fig. 1), characteristics similar to modern inbred lines. Inbred line Mo17 is a NSS inbred line the University of Missouri released in 1964 (Zuber, 1973). It has broad leaves and relatively flat LAs throughout the canopy (Fig. 1).

In the summer of 2013, we made reciprocal crosses between PHW30 and each of the two other inbreds. The F₁ plants were selfed in a 2014 winter nursery, and in the following summer, 500 F₂ seeds from each population were planted at 72,000 plants per hectare (∼29,000 plants per acre) in ∼5.5-m-long plots spaced ∼76 cm apart. We combined reciprocal populations, as there were no biologically significant LA differences (Table 1) and herein refer to them as B73 and Mo17 populations.

Table 1 Leaf angle phenotypic values for parents, hybrids, F₂, selected F₂, and F_2:3 plants.

Group	Leaf angle ± SD	No. of measurements	Range angle
	Degree		degree
Parents
B73	63.1 ± 5.3	46	53.5–73.8
Mo17	54.7 ± 8.1	35	32.4–67.3
PHW30	73.9 ± 3.4	90	66.6–82.9
F1
B73/PHW30	67.8 ± 5.4	48	53.2–76.4
PHW30/B73	67.9 ± 5.8	49	44.8–76.1
Mo17/PHW30	68.0 ± 4.4	48	58.4–76.7
PHW30/Mo17	66.0 ± 4.2	47	58.0–76.1
F2
B73/PHW30	68.1 ± 5.6	414	39.8–80.2
PHW30/B73	67.6 ± 6.0	440	46.0–80.9
Mo17/PHW30	68.9 ± 5.5	460	44.8–81.5
PHW30/Mo17	67.7 ± 5.4	346	50.0–82.3
Combined F2
B73 Population	67.8 ± 5.8	854	39.8–80.9
Mo17 Population	68.4 ± 5.5	806	44.8–82.3
Selected F2 Groups
B73-Flat	56.3 ± 3.4	45	47.1–62.0
B73-Average	67.8 ± 1.1	36	64.6–69.3
B73-Upright	77.0 ± 1.3	44	74.4–79.2
Mo17-Flat	58.3 ± 2.7	45	50.0–62.0
Mo17-Average	68.5 ± 1.2	34	66.0–70.3
Mo17-Upright	77.7 ± 1.9	46	74.0–82.0
F2:3
B73 Population	63.7 ± 5.0	123	48.1–74.0
Mo17 Population	55.0 ± 7.1	120	35.4–73.9

We applied a modified bidirectional selective genotyping strategy (Lander and Botstein, 1989) by including F₂ plants with phenotypic values around the population mean in addition to two trait extreme groups. From the B73 population, we selected 44 F₂ plants with the highest phenotypic values, 45 F₂ plants with the lowest, and 36 F₂ plants equal to the population mean, and from the Mo17 population, we selected 46, 45, and 34 plants, respectively. These three groups are referred to as the upright, flat, and average groups, respectively. In the summer of 2015, 25 F_2:3 seeds from selected F₂ plants were planted in a single replicate under the same planting conditions. Three representative plants from the middle of the plot were phenotyped for LA, and the calculated mean was recorded as the plot's phenotypic value. In total, 125 F_2:3 families were planted for each population, but only 123 F_2:3 families were phenotyped for the B73 population and 120 F_2:3 families for the Mo17 population.

Phenotyping for Leaf Angle

The second leaf below the ear leaf was phenotyped for LA on parents, hybrids, F₂ lines, and F_2:3 families starting after all plants completed anthesis. Leaf angle was phenotyped for this leaf because it contrasted the most among the three parents (Fig. 1). A digital image of each plant's LA was captured with the Field Book Phenotyping app (version 2.3.0; Rife and Poland, 2014) on a Google Nexus 7 Android Tablet. The angle between horizontal and middle of the midrib was measured on each image with ImageJ's angle tool (Schneider et al., 2012). The R open source statistical programming language and environment (R Core Development Team, 2017) and in-house R scripts were used to analyze the phenotypic data.

Genotyping, Filtering, and Imputation

To map QTL associated with LA variation, DNA was extracted from F_2:3 families. We collected one fresh tissue leaf punch from 10 F_2:3 plants within each F_2:3 family to reconstruct the original F₂ plant genotype (Schon et al., 1994). In total, 125 F_2:3 families from each population were tissue sampled.

DNA was extracted with a Qiagen DNA extraction kit and then digested with ApeKI and barcoded for 288-plex GBS (Elshire et al., 2011). Single nucleotide polymorphisms (SNPs) were called from the sequencing reads with the Tassel5-GBS Production Pipeline using the ZeaGBSv2.7 Production TOPM and reported as B73 RefGen_v2 physical positions (Glaubitz et al., 2014). FILLIN and maize donor haplotypes (V5) were used for SNP imputation (Swarts et al., 2014).

The SNPs were identified in the parents and then filtered and corrected for allelic dropouts in the progeny using in-house java scripts. The parental lines for each population were first compared to identify SNPs. These SNPs were selected in the progeny, and SNPs with >15% missing data and segregation distortion (by testing the Mendelian segregation ratio with a p-value <0.001) were removed. A sliding window algorithm with a 15-SNP window and an 1-SNP step (Su et al., 2017) was applied to address heterozygous genotypes called as homozygous because of low read coverage associated with the GBS method (Elshire et al., 2011).

Quantitative Trait Loci Mapping

IciMapping software (version 4.0.6.0; Meng et al., 2015) was used to map QTL. PHW30 was labeled as parent A for both populations and coded appropriately. Markers in linkage disequilibrium were binned using the following software options: missing rate percentage was set to 100, p-value distortion was set to 0 to omit this option, and markers in complete linkage disequilibrium were deleted randomly. The genetic maps were constructed with the binned markers and the Kosambi mapping function based on anchor (chromosome) and marker order (physical position) software options. Integrating the two genetic maps to construct a consensus genetic map was implemented with the R package LPmerge (version 1.6; Endelman and Plomion, 2014). Inclusive composite interval mapping of additive and dominant QTL was conducted with the consensus map and using the following software options: a 1-cM step, a 0.001 PIN size, and a 3.0 logarithm of odds (LOD) significance threshold. Physical positions for the detected QTL were converted to B73 RefGen_v3 using the assembly converter tool on Ensembl (Herrero et al., 2017). In addition, a genome-wide single-marker scan with physical distance and uncorrected genotypic data was conducted with 19,970 SNPs from the B73 population and 11,991 SNPs from the Mo17 population to corroborate the inclusive composite interval mapping results.

Meta-Quantitative Trait Loci Analysis

To anchor findings from this study, reported information from 19 previous studies was compiled and summarized. The position of the leaves used for phenotyping LA was summarized, and they were partitioned into six canopy levels to look for patterns connecting detected QTL and position within the canopy. The markers flanking the QTL were converted to physical positions with the locus search and BLAST function on MaizeGDB (Andorf et al., 2010; Cannon et al., 2011) and the bin location according to MaizeGDB or the original study. The assembly converter tool on Ensembl (Herrero et al., 2017) was used to convert all physical positions to B73 RefGen_v3.

We adapted a meta-analysis method (Wuschke et al., 2007) that uses physical positions of reported QTL to identify regions of the genome associated with LA variation. Reported QTL were declared significant with a LOD score, p-value, or likelihood ratio. The LOD scores were multiplied by 2 × ln(10) to obtain the likelihood ratio as it is asymptotically distributed as a χ² statistic. We obtained p-values for the χ² statistics using one degree of freedom. For QTL defined by a physical interval, the QTL was represented by the interval midpoint. Each chromosome was divided into 10-Mb genomic bins, and the reported QTL were mapped to a genomic bin within a chromosome. Physical map-based genomic bins were used to circumvent differences in marker systems and genetic map lengths reported from the different studies. By retaining QTL with the lowest p-value in genomic bins where multiple QTL were detected from the same study and canopy level, we obtained a set of nonredundant QTL. A Z-test weighted by sample size from the original study combined p-values within genomic bins (Zaykin, 2011) and was implemented with the combine.test function from R package survcomp (version 1.3; Schröeder et. al, 2011). An in-house R script administered the meta-analysis with a significance threshold for grouped p-values set at 7.11 × 10⁻¹¹ (LOD = 9.22) that accounts for multiple testing through a Bonferroni correction.

Finally, a literature search identified 75 candidate genes associated with LA from maize, rice (Oryza sativa L.), and sorghum [Sorghum bicolor (L.) Moench] that were mapped into the genomic bins (Supplemental Table S1). Candidate genes from rice and sorghum were only included if they had an ortholog in maize. We recorded physical positions for the maize genes as B73 RefGen_v3.

RESULTS Phenotypic Analysis

Reciprocal crosses between inbred lines PHW30 and both B73 and Mo17 were developed. Compared with PHW30, inbreds B73 and Mo17 have less upright LAs, most notably for the second leaf below the ear leaf (Fig. 1, Table 1). The LA for both B73 hybrids was about equal to the midparent value, whereas LA for both Mo17 hybrids was greater than the midparent value (Table 1). We tested the maternal effect for LA in the B73 and Mo17 hybrids and observed a small difference only between Mo17 hybrids (2.0°, P = 0.026). Since reciprocal differences were not biologically significant for both of the hybrids (Table 1), we combined reciprocal populations.

After phenotyping the combined F₂ populations, F₂ plants were selected into three groups corresponding to their LA: upright, average, and flat. The mean of the flat group from the Mo17 population was larger than the mean of the Mo17 parent, while the mean of the flat group from the B73 population was smaller than the mean of the B73 parent (Table 1). Both of the upright group's means were larger than the mean of the PHW30 parent (Table 1). Phenotypic distributions across populations in the F₂ and F_2:3 generations were normally distributed (Supplemental Fig. S1), varied widely (Table 1) and were slightly lower in both of the F_2:3 generations (Supplemental Fig. S1).

Construction of Genetic Map and Quantitative Trait Loci Detection

A modified selective genotyping method was used to select 125 F₂ plants from each population, and DNA was collected from 10 F_2:3 plants to reconstruct the F₂ genotype. Genotyping-by-sequencing was used to generate the genotypic data for both populations. Using the filtered progeny genotypic data, two genetic maps were developed and combined into a consensus map. We identified 94,078 SNP sites segregating in the B73 population and 46,724 segregating in the Mo17 population. After filtering, imputation, and binning, 2710 bins developed the B73 population linkage map, and 2129 bins developed the Mo17 population linkage map. Genetic map lengths were 1726.5 cM for the B73 population and 1628.88 cM for the Mo17 population. Combining the two genetic maps resulted in a consensus map of 1808.9 cM.

Inclusive composite interval mapping with the consensus map was conducted separately for each population and generation. In total, 12 QTL were detected for LA (Fig. 2). Six QTL were detected across both generations of the B73 population (Fig. 2A,B), and all six were nonoverlapping according to the QTL interval's physical position (Table 2). We detected three QTL in each of the two generations (Fig. 2A,B). Six QTL were detected across both generations of the Mo17 population (Fig. 2C,D), and five were nonoverlapping according to the QTL interval's physical positions (Table 2). Five QTL were detected in the F₂ generation, and one was detected in the F_2:3 generation (Fig. 2C,D). A single-marker scan with physical mapping of unprocessed genotypic data provided supporting evidence for these results (Supplemental Fig. S2).

Fig. 2. Linkage mapping of leaf angle. Logarithm of odds (LOD) plots for selected chromosomes from the inclusive interval mapping. Results for LA in the B73 population using selected F2 plants (A), F2:3 families (B), Mo17 population using selected F2 plants (C), and F2:3 families (D). Chromosomes were selected for plotting if it contained a significant QTL among all results. The horizontal red line indicates the significance threshold, and vertical dotted lines separate the selected chromosomes.

Table 2 Quantitative trait loci detected across two combined populations and two generations.

Population	Generation	Chromosome	Position	Physical interval (RefGen_v3)	LOD	Phenotypic variance explained	Additive effect	Dominance effect
			cM	bp		%
B73	F2	1	191.00	275,062,657–275,835,496	6.25	16.60	5.19	0.03
Mo17	F2	1	192.91	275,843,509–276,767,207	4.74	10.44	3.32	−0.02
Mo17	F2:3	1	192.91	275,843,509–276,767,207	3.91	14.43	3.66	1.06
B73	F2:3	2	159.00	180,806,386–182,824,212	4.82	13.01	−2.19	1.28
Mo17	F2	2	240.84	227,218,624–227,302,619	6.03	13.29	3.60	1.16
Mo17	F2	3	91.74	29,175,464–29,212,960	7.43	17.17	4.12	−0.66
B73	F2:3	3	92.00	36,470,328–37,674,938	4.88	13.78	2.40	0.48
B73	F2	3	131.00	175,591,546–176,377,186	3.41	8.77	3.72	0.26
Mo17	F2	3	216.74	215,170,248–215,298,141	4.95	10.99	−3.69	−0.31
B73	F2:3	3	246.00	225,010,933–225,429,141	4.46	11.54	−2.34	2.02
Mo17	F2	4	89.00	192,416,147–193,325,151	4.66	9.89	3.68	1.40
B73	F2	8	39.00	102,435,103–102,836,776	3.04	7.54	1.41	4.77

The QTL near ∼275 Mb on chromosome 1 was consistently detected in both generations of the Mo17 population and in the F₂ generation of the B73 population (Table 2). A peak in this region was also detected in the F_2:3 generation of the B73 population, but its LOD score (2.41) was below the significance threshold (Fig. 2B). Additionally, the QTL on chromosome 3 that was detected in the F_2:3 generation of the B73 population and the F₂ generation of the Mo17 population maps to the same genetic position (∼92 cM) but physically map ∼7 Mb apart (Table 2). The detected QTL were located on chromosomes 1, 2, 3, 4, and 8 (Table 2).

Phenotype, Population, and Quantitative Trait Loci Summary

Different leaves were phenotyped in dissecting LA variation in previous genetic mapping studies (Fig. 3, Table 3). Phenotyped leaves varied from one leaf below the flag leaf to the second leaf below the ear, with most of the studies phenotyping the leaves near the main ear (Fig. 3, Table 3). Eight studies phenotyped multiple leaves but used the mean phenotypic value of all leaves measured as the final trait value (Table 3). Nine studies, including this one, phenotyped a single leaf, while three studies phenotyped consecutive leaves and mapped them as separate traits (Table 3).

Fig. 3. Phenotyped leaves from 20 maize leaf angle studies. A representative maize plant is next to blue bars representing the leaves previous studies in maize phenotyped for leaf angle. Blue bars spanning more than one leaf indicates those leaves were phenotyped and the mean value was used for genetic mapping. Letters above blue bars relate the phenotyped leaf to the study referenced in Table 3.

Table 3 Leaf angle quantitative trait loci (QTL) studies in maize.

Source	No. of populations	Population size	No. of QTL	Leaves measured	Canopy level	Fig. 3 reference
Mickelson et al., 2002	1	180	10	Average of all leaves above ear leaf	Mid-upper	B
Yu et al., 2006	2	120; 114	9	Third leaf below flag leaf	Upper	H
Lu et al., 2007	1	397	6	Average of first through third leaf above ear leaf	Mid-upper	D
Ku et al., 2010	1	229	3	Average of first leaf above ear leaf, ear leaf, and first leaf below ear leaf	Middle	E
Tian et al., 2011	25	∼195 each	30	One leaf below flag leaf	Upper	G
Ku et al., 2012	1	256	5	Average of first leaf above ear leaf, ear leaf, and first leaf below ear leaf	Middle	E
Wassom, 2013	1	93	3	Ear leaf	Middle	K
Liu et al., 2014	1	144	5	Third leaf above ear leaf	Mid-upper	H
			5	Second leaf above ear leaf	Mid-upper	I
Potts, 2014	9	∼35 each	12	Third leaf below flag leaf	Middle	H
Chen et al., 2015	1	144	5	Second leaf above ear leaf	Mid-upper	I
			6	Third leaf above ear leaf	Mid-upper	H
Ding et al., 2015	1	305	14	Average of first through fourth leaf above ear leaf	Upper	C
Hou et al., 2015	1	266	14	First leaf above ear leaf	Middle	J
Li et al., 2015	3	183; 172; 183	17	Average of first leaf above ear leaf, ear leaf, and first leaf below ear leaf	Middle	E
Chang et al.,2016	1	150	17	First leaf above ear leaf	Middle	J
			19	Ear leaf	Middle	K
			16	First leaf below ear leaf	Middle	L
Ku et al., 2016	4	215; 208; 212; 223	21	Average of first through fourth leaf above ear leaf	Upper	C
Pan et al., 2017	10	∼188 each	149	First leaf above ear leaf	Middle	J
Wang et al., 2017	1	220	17	Ear leaf	Middle	K
Zhang et al., 2017	1	196	24	Average of all leaves in canopy	All	A
			19	Average of all leaves above ear	Mid-upper	B
			24	Average of all leaves below ear	Mid-lower	F
Zhao et al., 2018	2	202;218	33	Ear leaf	Middle	K
This study	2	125;125	12	Second leaf below ear leaf	Lower	L

In addition to the phenotyped leaves, the parents and genetic mapping populations used by the different studies were summarized (Supplemental Table S2, S3). The populations were developed from genetic crosses using 133 inbred lines, 71 of which were unique (Supplemental Table S2). Inbred line B73 was used 32 times in a cross (∼24% of all inbred lines used), while 50 parents were only used once (Supplemental Table S2). Additionally, the number of genetic mapping populations per study ranged from 1 to 25, and the size of those populations ranged from 94 to 397 (Table 3).

Overall, 495 QTL associated with LA were reported through linkage mapping within individual populations or joint linkage mapping across populations (Supplemental Table S3). Reported QTL per study ranged from 3 to 149 (Table 3), and the phenotypic variance explained by each QTL ranged from 0.41 and 85.05% (Supplemental Table S3). Furthermore, the estimated additive effect for the reported QTL ranged from −8.51 to 6.67, and the estimated dominance effect ranged from −2.88 to 10.77 (Supplemental Table S3).

Meta-Quantitative Trait Loci Analysis

After dividing the genome into 210 genomic bins ∼10 Mb in size, the 495 reported LA QTL were mapped into 143 genomic bins across all 10 chromosomes. Within a genomic bin, multiple QTL from the same study and canopy level were removed, leaving 325 nonredundant QTL (∼66%). After calculating the grouped p-value for each genomic bin using a weighted Z-test, there was evidence of linkage (P < 7.1 × 10⁻¹¹) of LA to 58 genomic bins distributed across all 10 chromosomes (Fig. 4, Supplemental Table S4). These genomic bins contained 203 nonredundant LA QTL (∼41%), representing a moderate enrichment (∼2.3-fold) of LA QTL in these regions. To identify any pattern that connected canopy position with the linked genomic bins, each study's phenotyped leaves were partitioned into six canopy levels (Table 3). When the 58 linked genomic bins were compared with the canopy level used to detect the QTL, no clear pattern was observed (Fig. 4).

Fig. 4. Meta-analysis of quantitative trait loci (QTL) for leaf angle in maize. The maize genome was divided into 210 genomic bins of approximately equal size (∼10 Mb), and 325 nonredundant QTL were mapped to a genomic bin based on the physical position of the QTL interval midpoint. Grouped p-values were calculated for 141 genomic bins containing 325 nonredundant QTL, and the horizontal red line indicates the significance threshold (upper panel). Within each genomic bin, colored bars indicate the number of nonredundant QTL detected from phenotyping leaves in different levels of the canopy, while the total height of all the bars indicates the total number of non-redundant QTL detected (lower panel).

Meta-Analysis Candidate Genes

From the 58 genomic bins that were associated with LA, we focused on summarizing the 15 most prominent genomic bins (Fig. 4, Table 4). These 15 genomic bins contained 82 of the 325 nonredundant QTL (∼25%) and represented a ∼3.5-fold enrichment of LA QTL in these regions. All chromosomes except chromosome 6 contained a prominent genomic bin (Table 4) and chromosomes 1, 2, 3, 7, and 10 contained multiple (Table 4).

Table 4 Candidate genes identified in the 15 prominent genomic bins from the meta-analysis using the nonredundant set of reported quantitative trait loci (QTL).

Rank	Chromosome	Genomic bin	No. of QTL	Grouped p-value	Candidate genes	Species	Maize gene model
1	2	1	12	5.40 × 10−56	lg1	Maize	GRMZM2G036297
2	1	3	10	2.14 × 10−44	drl1	Maize	GRMZM2G088309
3	2	2	6	1.20 × 10−25	RAVL1	Rice	GRMZM2G102059
4	7	14	6	2.83 × 10−25	–	–	–
5	3	18	3	1.33 × 10−23	lg2	Maize	GRMZM2G060216
6	10	15	6	3.09 × 10−23	OsLG1	Rice	GRMZM2G058588
7	4	13	1	1.30 × 10−22	–	–	–
8	1	2	6	1.02 × 10−21	OsMDP1	Rice	GRMZM2G059102
9	7	13	5	1.09 × 10−21	–	–	–
10	8	17	5	1.57 × 10−21	lrs1	Maize	AC232238.2_FG004
11	9	15	4	4.47 × 10−21	LPA1	Rice	GRMZM2G074032
					drl2	Maize	GRMZM2G102218
12	5	18	6	4.95 × 10−21	–	–	–
13	1	4	5	1.40 × 10−19	LPA1	Rice	GRMZM2G465595
					pmk1	Maize	GRMZM2G019260
					DWF4	Maize	GRMZM2G065635
14	3	3	4	8.58 × 10−19	–	–	–
15	10	2	3	8.26 × 10−18	–	–	–

We identified 12 candidate genes in 10 of the 15 prominent genomic bins: seven genes from maize and five maize orthologs from rice genes (Fig. 4, Table 4). Prominent genomic bins contained two maize genes known to alter LA: lg1 and lg2. Twelve studies reported 28 QTL in the same genomic bin as lg1 (Supplemental Table S4). Two other maize genes also known to alter LA, droopy leaf1 (Strable et al., 2017) and droopy leaf2 (Strable et al., 2017) were identified. Other prominent genomic bins contained liguleless related sequence1 (Langham et al., 2004), which is a paralog of lg2, CLA1 (MaizeGDB), and DWF4 (Liu et al., 2007). Maize orthologs of rice genes known to alter cell sizes in the lamina joint that modifies leaf bending were also identified in these prominent genomic bins. These genes include RAVL1 (Je et al., 2010), OsMDP1 (Duan et al., 2006), and LPA1 (Wu et al., 2013). Two maize orthologs were identified for both OsLG1 (Lee et al., 2007) and LPA1 (Supplemental Table S3), and all four orthologs were identified in prominent genomic bins (Table 4), as one of the orthologs for OsLG1 is lg1.

The meta-analysis integrated the 12 newly identified QTL into 10 genomic bins, and we explored those that contain candidate genes (Table 2). Four of these genomic bins contain candidate genes, and lg2 was included in the 15 prominent genomic bins (Table 4). The three QTL located on chromosome 1 physically map near one of the six maize phytochrome genes, phytochrome C1 (PhyC1; Sheehan, 2004). Additionally, the cloned maize gene ZmTAC1 (Ku et al., 2011) was identified as a candidate gene. Finally, another maize ortholog of a rice gene that modifies the bending of the leaf's lamina joint, TLD1 (Zhang et al., 2009), was identified as a candidate gene.

DISCUSSION

Research has shown that commercial hybrids released over the last 50 yr have increasingly more upright LAs (Duvick et al., 2004; Wang et al., 2011; Ma et al., 2014). This study investigated three maize inbred lines with distinct LA architectures from important US heterotic groups. Inbred line B73 has slightly less upright LAs in the lower canopy than PHW30, while Mo17 has the least upright. The LA for F₁ plants from the B73 crosses was about equal to the average of the parents, while LA for F₁ plants from the Mo17 crosses was greater than the average of the parents. Other studies reporting LA phenotypes on F₁ plants and the corresponding parents reported F₁ plants about equal to the midparent value (Ku et al., 2010) and F₁ plants greater than both parents (Hou et al., 2015). Transgressive segregation in both the F₂ and F_2:3 generations was observed in all populations, which is consistent with previous studies (Mickelson et al., 2002; Ku et al., 2010; Chen et al., 2015; Hou et al., 2015). Slight differences in the distributions of the F₂ and F_2:3 generations could be explained by the decrease in heterozygosity between the F₂ and F_2:3 generation that can result in less vigorous plants therefore causing the plants to become less upright.

All 12 QTL detected across both populations and generations in this study were supported by at least one prior study. These 12 QTL were detected on chromosomes 1, 2, 3, 4, and 8 by phenotyping a leaf in the lower canopy. The QTL at ∼227 Mb on chromosome 2, ∼29 Mb on chromosome 3, and ∼176 Mb on chromosome 3 were near or within QTL intervals detected from other studies that phenotyped leaves in different parts of the canopy (Yu et al., 2006; Ku et al., 2011; Tian et al., 2011; Potts, 2014). The QTL in these three regions all have positive additive effects indicating PHW30 as the source of the allele and explain substantial amounts of phenotypic variance (∼9–17%). Furthermore, two of the three QTL are near cloned maize genes, lg2 (chromosome 3, ∼176 Mb) and ZmTAC1 (Chromosome 2, ∼221 Mb) that alter LA in multiple leaves throughout the canopy. In a separate experiment, we developed a backcross family with PHW30 and Mo17 as the recurrent parent that appears to be segregating for a single gene causing upright LAs for all leaves in the canopy. We are currently genotyping and phenotyping this family to help determine the underlying molecular variant that explains the phenotype observed in this family.

Light sensitivity reduction contributed to modern maize's transition to upright LAs (Fellner et al., 2003, 2006). Maize gene PhyC1 (Chromosome 1, ∼227 Mb; Sheehan, 2004) physically maps within the three QTL intervals detected on chromosome 1. A previous study phenotyped a leaf near the ear and used ex–PVP lines and B73 to detect a QTL ∼4 Mb upstream of this gene (Potts, 2014). Of the 20 studies summarized, these are the only two targeted genetic mapping studies to use modern inbred lines to identify QTL for LA. The PhyC1 gene has been cloned in maize (Sheehan, 2004), and previous results in maize and rice suggested that phyC can perceive far-red light signals (Takano et al., 2005; Dubois et al., 2010). High planting densities reduce red light and increase far-red light, resulting in a decrease in the red to far-red ratio. This decrease can trigger shade-avoidance responses in plants that can include altering LA and other architectural modifications to increase the amount of unfiltered light reaching their leaves (Franklin and Whitelam, 2005). Positive additive effects for all three QTL from this study indicated that the more modern and upright LA inbred, PHW30, is the source of the allele, thus suggesting PHW30 might have a different PhyC1 allele with decreased sensitivity to red to far-red light ratios.

Under high planting densities, the ideal LA architecture for the maize canopy should gradually move from upright LAs in the upper canopy to less upright LAs in the lower (Duncan, 1971; Long et al., 2006; Zhang et al., 2017). Previous research dissecting the genetic control of LA in maize has focused on phenotyping a single leaf or averaging the phenotypes from multiple leaves. Without phenotyping and genetically mapping multiple individual leaves in the canopy, it is difficult to differentiate QTL controlling LA specifically in one part of the canopy from QTL controlling LA in multiple parts. For example, leaves from the upper canopy to the mid-lower canopy were phenotyped in the studies that detected QTL within the genomic bin on chromosome 2 (0–10 Mb) that contains lg1 (Moreno et al., 1997). Even though this particular mutant is known to affect LA in all parts of the canopy (Becraft et al., 1990), for other QTL, without phenotyping and genetically mapping for LA in multiple leaves in the canopy, it is difficult to make this differentiation. Previous research has genetically mapped LA for individual leaves and identified leaf specific QTL, but consecutive leaves were phenotyped rather than leaves in different parts of the canopy (Liu et al., 2014; Chen et al., 2015; Chang et al., 2016). To differentiate leaf-specific from canopy-wide QTL that control LA, we are currently genotyping doubled haploid lines developed from the selected F₂ plants from this study and phenotyping four leaves throughout the canopy for LA.

In addition to differentiating leaf-specific from canopy-wide QTL, parents with contrasting LAs in the canopy should be used for developing genetic mapping populations. Past LA studies in maize, including this one, have developed genetic mapping populations by crossing inbreds with upright LAs throughout the canopy by inbreds with flat LAs (Ku et al., 2010, 2012; Zhang et al., 2014; Chen et al., 2015; Ding et al., 2015; Li et al., 2015). These types of crosses resulted in 12 studies reporting QTL in the same genomic bin as lg1.

A goal in understanding quantitative trait variation is to determine the underlying molecular variation (Mackay, 2001). Integrating detected QTL from many independent studies through a meta-analysis increases the resolution of those QTL and brings us closer to determining the causal molecular variations. From our meta-analysis results, we identified cloned genes, lg1 and lg2, as candidate genes that were corroborated by other LA meta-analyses (Ku et al., 2012; Wang et al., 2016; Zhao et al., 2018) providing evidence to the power these methods have in detecting candidate genes. The recently proposed omnigenic model postulates that there are a moderate number of genes that directly affect a particular trait, or core genes, and any variant in the numerous genes that compose the highly connected cell regulatory networks can affect these core genes (Boyle et al., 2017). For example, genes like lg1 and lg2 may be core genes since they directly affect LA, and any variants in the numerous genes involved in leaf development, such as the other genomic bins detected in the meta-analysis, may affect the regulation or function of lg1, lg2, or other core genes associated with LA.

Many candidate genes detected from the meta-analysis were maize orthologs of rice genes that modify LA through changes in cell size and bending in the lamina joint (Supplemental Table S1). While both are members of the grass family, maize and rice have slightly different leaf structures. In maize, the ligule and auricle are structures that separate the blade and sheath. The auricle can act as a hinge, allowing the leaf to bend away from the stem and help support the weight of maize's wider and heavier leaf. Loss of ligule and auricle (Becraft et al., 1990; Harper and Freeling, 1996), variations in the auricle size (Kong et al., 2017; Strable et al., 2017), and proteomic differences and changes in midrib cells (Wang et al., 2015; Strable et al., 2017) are known biological mechanisms that can alter maize LA. On the other hand, rice's ligules and auricles are outside of the leaf and act more as appendages, and the lamina joint separates the blade and sheath. There are cloned rice genes that alter LA by modifying leaf bending through changing cell size in the lamina joint (Duan et al., 2006; Je et al., 2010; Jiang et al., 2012; Wu et al., 2013; Gao et al., 2014; Feng et al., 2016). Changes in cell size near the leaf collar in maize have been reported in near isogenic lines differing only by LA related gene CLA4 (Zhang et al., 2014). Combined with the results of our meta-analysis results, cell size near the abaxial boundary between the blade and sheath near the midrib may also contribute to LA variation in maize.

The results from our study provide additional knowledge on QTL associated with LA in the lower canopy and a stronger framework for determining the molecular variations underlying LA variation in maize. Our genetic mapping data provide evidence of QTL associated with maize LA variation in the second leaf below the ear. Questions remain on differentiating QTL controlling LA in one part of the canopy from QTL controlling LA in multiple parts. Current research is underway to address these remaining questions by phenotyping and genetically mapping LA for multiple leaves in the canopy using material developed from this study.

CONCLUSION

Understanding the genetic control of complex traits like LA involves linking observed phenotypic variation to regions of the genome. Our study investigated LA variation in three inbred lines with distinct LA architectures that represent important US heterotic groups. We have identified unique regions of the genome linked to LA variation, with a major effect QTL consistently detected on chromosome 1. Integration of these results with other reported QTL through a meta-analysis identified 58 regions of the genome associated with LA variation. While some of these regions contain genes known to affect LA in maize, other regions contain maize orthologs from rice genes that are associated with modifying cell size and bending in the leaf's lamina joint. These could be useful regions of the genome to target for future comparative genomics research into identifying additional molecular variants associated with LA variation. Together, these results advance our knowledge about LA variation in maize to assist breeders in developing hybrids with a canopy-wide LA architecture that are adapted to future increases in planting density.

Supplemental Information Available

Supplemental Fig. S1. Histograms of phenotypic values for LA across the F₂ and F_2:3 generation for the B73 and Mo17 populations.

Supplemental Fig. S2. Linkage mapping compared with single marker scan results for LA. Scans were done with the uncorrected genotypic data for LA across B73 population using selected F₂ plants (A), F_2:3 families (B), Mo17 population using selected F₂ plants (C), and F_2:3 families (D). Top sections represent the linkage mapping results, while the bottom sections represent the single marker scan results.

Supplemental Table S1. List of candidate genes known to affect LA from maize, rice, and sorghum, and the corresponding maize gene model.

Supplemental Table S2. Summary of the parents the 20 studies used to develop genetic mapping populations.

Supplemental Table S3. Summary of the 495 reported QTL reported from 20 studies including the present study.

Supplemental Table S4. Details about the 58 genomic bins, candidate genes, and reported QTL linked to LA variation detected from the meta-analysis.

Data and code used in this study was uploaded in an online public repository: https://github.com/mdzievit/Meta_QTL-LA-Mapping-Paper.

Conflict of Interest Disclosure

The authors declare that there is no conflict of interest.

Acknowledgments

This work was supported by the Iowa State University Raymond F. Baker Center for Plant Breeding, the DuPont Pioneer Graduate Assistantship, and the Iowa State University Plant Sciences Institute.