Natural Variation at sympathy for the ligule Controls Penetrance of the Semidominant Liguleless narrow-R Mutation in Zea mays

Leaf architecture determines plant structural integrity, light harvesting, and economic considerations such as plant density. Ligules, junctions at the leaf sheath and blade in grasses, protect stalks from environmental stresses and, in conjunction with auricles, controls leaf angle. Previous studies in mutants have recessive liguleless mutants (lg1 and lg2) and dominant mutations in knotted1-like homeobox genes (Lg3-O, Lg4, and Kn1) involved in ligule development. Recently, a new semidominant liguleless mutant, Liguleless narrow (Lgn-R), has been characterized in maize that affects ligule and auricle development and results in a narrow leaf phenotype. We show that quantitative genetic variation affects penetrance of Lgn-R. To examine the genetic architecture underlying Lgn-R expressivity, crosses between Lgn-R/+ mutants in a B73 background and intermated B73 x Mo17 recombinant inbred lines were evaluated in multiple years and locations. A single main-effect quantitative trait locus (QTL) on chromosome 1 (sympathy for the ligule; sol) was discovered with a Mo17-contributed allele that suppressed Lgn-R mutant phenotypes. This QTL has a genetic-interaction with a locus on chromosome 7 (lucifer; lcf) for which the B73-contributed allele increases the ability of the solMo17 allele to suppress Lgn-R. Neither of the genetic intervals likely to contain sol or lcf overlap with any current liguleless genes nor with previously identified genome-wide association QTL connected to leaf architecture. Analysis of phenotypes across environments further identified a genotype by enviroment interaction determining the strength of the sol x lcf interaction.


genetic network development QTL epistasis
The ligule is an important architectural feature of grass leaves that protects the stalk from disease and environmental insults. In maize, the ligule and auricles allow the leaf blade to lean away from the sheath, which remains wrapped around the stalk (Foster and Timmermans 2009;Kiesselbach 1949). Study of liguleless mutants has identified genes with roles in meristem maintenance, leaf polarity, and leaf angle (Moreno et al. 1997;Walsh et al. 1998). The latter is a trait of critical importance to plant density and maize agriculture (Duvick 2005). In maize and other grass species, mutant studies, and quantitative trait loci (QTL) mapping have identifed the genomic locations of loci affecting leaf architecture (Mickelson et al. 2002;Tian et al. 2011;Yu et al. 2012;Zeng et al. 2009).
Classical and contemporary mutant studies have identified genes required for ligule development: lg1 is located on the short arm of chromosome 2; and lg2 is located on the long arm of chromosome 3 (Beckett 1975). Maize homozygous for recessive loss-of-function alleles at liguleless1 or liguleless2 lack a ligule and auricles (Becraft and Freeling 1991;Harper and Freeling 1996;Moreno et al. 1997), resulting in more erect leaves. lg1 encodes a protein similar to SQUA-MOSA PROMOTER-BINDING (Moreno et al. 1997) and lg2 encodes a basic leucine zipper protein transcription factor (Walsh et al. 1998). In addition to these, lg3, located on 3S (Kerstetter et al. 1994), and lg4, located on 8L (Fowler and Freeling 1991), were identified from dominant mutations that altered ligule formation. The semidominant Lig-uleless3 alleles cause the leaf blade, ligule, and auricle tissues to adopt a sheath-like phenotype (Fowler and Freeling 1996). The Liguleless4 dominant mutants, much like Liguleless3, show no auricle or ligule tissue, with blade tissue appearing to be sheath-like; however, the tissue disruption occurs near the leaf margins (Fowler and Freeling 1991). liguleless3 and liguleless4 encode class-I knotted1-like homeobox (KNOX) transcription factors that are duplicated within the grasses (Bauer et al. 2004;Kerstetter et al. 1994;Muehlbauer et al. 1999). The knotted1 transcription factor (Vollbrecht et al. 1991) is also involved in ligule development as well as affecting shoot meristem development (Kerstetter et al. 1994;Volbrecht et al. 2000). Kn1-dominant mutants displace the position of the ligule into the blade and cause knot structures composed of mislocalized sheath tissue to form along the lateral veins (Sinha and Hake 1990;Vollbrecht et al. 1991).
Leaf architecture features involving the ligule and the leaf angle have been examined using QTL studies across several species (Mickelson et al. 2002;Tian et al. 2011;Yu et al. 2012;Zeng et al. 2009). In rice, five flag-leaf ligule length QTL were mapped to chromosomes 2, 6, 10, and 12 as well as the rice ortholog of maize liguleless1 on the long arm of chromosome 4 (Zeng et al. 2009). Similarly, Tian et al. (2011) performed a large-scale QTL study in maize and found relatively largeeffect QTL controlling leaf angle mapped to the lg1 and lg2 genes.
Recently, a semidominant mutant was identified that defines a novel locus affecting ligule development. Liguleless-narrow (Lgn-R) was identified in an ethane methyl sulfonate2mutagenized B73 population (Moon et al. 2013). Lgn-R plants exhibit disrupted ligule development and reduced auricles toward the margins of the leaf. The leaves are both shorter and narrower. In addition, Lgn-R mutants produce fewer tassel branches and often fail to produce an ear. Expression of the Lgn-R mutant phenotype is clear in B73 as well as other inbred backgrounds (Moon et al. 2013).
In this work, we describe variation in the penetrance of Lgn-R affected by natural genetic variation. Lgn-R phenotypic expression is suppressed in crosses to Mo17. To map the Lgn-R modifiers, length and width measurements were made on liguleless-narrow mutants from Lgn-R/+ individuals in the B73 background crossed to Intermated B73 x Mo17 (IBM; Beavis et al. 1992;Lee et al. 2002) recombinant inbred lines (RILs). The IBM RIL x Lgn-R/+ F1 were grown in two locations in 2009, Indiana (Purdue Agronomy Center for Research and Education; West Lafayette, IN) and California (Gill Tract Farm; Albany, CA). A single main-effect QTL was identified on chromosome 1, called sympathy for the ligule (sol). The Mo17 allele at sol suppresses mutant phenotype expression in Lgn-R plants resulting in increased fertility, leaf length, and leaf width. The RIL genotypes of fertile, successfully suppressed IBM RIL x Lgn-R/+ F1 individuals were analyzed for shared regions of homozygosity, which, along with genome-wide pairwise analysis for epistasis, identified an epistatic interaction between sol and a region of chromosome 7, referred to as lucifer (lcf). Genotype by environment analysis revealed a dramatic requirement of environment for the expression of the sol x lcf interaction. We propose that this genetic network helps to integrate plant growth and architecture with environmental determinants of development.

Phenotypic measurements
The dominant Lgn-R mutant was isolated as a half-plant chimera from ethane methyl sulfonate mutagenesis of the inbred line B73 (Gerhold et al. 2005;Moon et al. 2013). Lgn-R/+ plants from introgression of the allele into the Mo17 and B73 backgrounds were used in this study. To determine the existence and location of Lgn-R suppressors, B73 Lgn-R heterozygotes were crossed as the pollen donor to 63 individuals from the core set of 94 IBM (Beavis et al. 1992) thus creating RIL x Lgn-R/+ F1s. Leaf length and width measurement of the 6th leaf from the tassel were recorded for between two and eight individual F1 (IBM RIL x Lgn-R) plants from these crosses grown at the Purdue Agronomy Center for Research and Education (West Lafayette, IN) and University of California Gill Tract farm (Albany, CA) in 2009. In both locations, Lgn-R/+ individuals were distinguishable from wild-type siblings due to dramatically narrower and shorter leaves. From these measurements, leaf area (LA, length x width) and length by width (LW, the ratio of length to width) were calculated. The location in Indiana consisted of two independent plantings, and measurements and was thus treated as replicates: IN_1 and IN_2. In addition to leaf phenotypes, stand counts of plants in the row (count) were also made. In 2010, a subset of the IBM RIL x Lgn-R/+ F1 individuals (IBM18, IBM30, IBM69, and IBM72), which previously displayed rescued or near wild-type sibling leaf phenotypes, were grown at the Purdue Agronomy Center for Research and Extension (West Lafayette, IN) to confirm action of the Lgn-R modifiers. An additional five RIL (IBM4, IBM25, IBM47, IBM55, and IBM65) were identified as substantially rescued to wildtype phenotypes among the RIL x Lgn-R F1s grown for the 2009 mapping experiment. These nine individuals are referred to as the "rescuing RILs." Data processing and QTL analysis Within and between years and locations, all leaf traits were positively correlated within individuals and between replicates of each IBM RIL x Lgn-R/+ F1 family. To better describe the genetic effects on phenotype expression and capture the genetic covariance responsible for trait correlation, principal components analysis (PCA) of covariances between traits was performed using the JMP software package (version 8.0.1, SAS Institute Inc., 2009) to derive additional quantitative traits for QTL mapping. All traits (e.g., length, width, count) were used for QTL mapping as well as the additional PCA derivatives presented in detail here: LW_all (length by width, all locations), LW_IN (length by width, IN locations), LW_GT (length by width, CA only), LA_all (leaf area, all locations), and LA_IN (leaf area, IN locations). Each IBM RIL had multiple length and width measurements and all individuals were used for PCA imputation. For comparisons between the IN and CA environments, leaf area and not principal components from CA was compared to leaf area from IN.
QTL analysis for all traits was conducted using R/qtl (Broman et al. 2003) using the 2009 data sets. Genotypes and map positions were based on the ISU IBM Map version 4 (http://www.maizeGDB.org) for the core 94 IBM population. Markers with more than 15% missing data among the 63 phenotyped IBM RIL were removed, leaving 3205 markers across the 10 maize chromosomes. Missing genotypes were imputed by the no double crossovers method, in which missing genotypes are filled in unless a recombination event is predicted to occur (R/qtl package manual; Broman et al. 2010). One-dimensional genome scans were performed using scanone by the Expectation Maximization method. Scanone, using method Expectation Maximization, assumes a single QTL model and follows standard interval mapping (R/qtl package manual; Broman et al. 2010). Experimentwide permutation thresholds (alpha , 0.05) were determined separately for each trait with 1000 permutations. Two-dimensional genome scan and tests for epistasis were performed by marker-pair regression analysis using the scantwo function (Broman et al. 2003). Scantwo tests a two-QTL model and result summaries provide multiple assessments of interaction. Scantwo results were summarized using the "int" flag for interaction summary and by the "best" method to look for additional QTL by two dimensional scanning. Experimentwide significance for the two dimensional scans (alpha , 0.05) were determined for each trait and procedure ("int" and "best") with 1000 permutations. The complete mapping procedure, scripts, primary data, and PCA outputs are all available as a supplement to this mansucript or from the authors by request (Supporting Information, File S1).
A Standard Least Squares model implemented in the JMP statistical software package (version 8.0.1; SAS Institute Inc. 2009) examined the genotype by environment interactions for data from 2009 from both locations. The genotypes of both sol, estimated with marker umc2145, and lcf, estimated with marker nbp1, were compared with the leaf area traits for each location and across both locations.
Validation of sol and lcf effect At the Gill Tract farm in 2011, a subset of IBM RIL x Lgn-R/+ F1 plants (IBM18, IBM30, IBM69, and IBM72) as well as BC1 IBM RIL x Lgn-R (IBM RIL x Lgn-R F1 backcrossed to B73) families segregating for the B73 and Mo17 alleles at sol were measured for leaf length and width of the 6th leaf from the tassel as well as plant height. A total of 187 individuals were measured. DNA was isolated from leaf samples for each individual and genotyped using the polymerase chain reaction markers umc2145 at chromosome one 94 cM for sol and bnlg1792 at chromosome seven 66 cM for lcf. Individuals were scored as either "B73," "Mo17," or heterozygous (B73/Mo17) for each marker. Because the RILs that effectively rescued Lgn-R plants were fixed for lcf B73 and the recurrent parent was B73, these BC1 plants were not segregating for lcf. To examine the effect of sol on mutant and wild-type leaf development, a comparison of leaf length and width data as well as plant height and the sol genotype was estimated by a Standard Least Squares model using the JMP software package (version 8.0.1, SAS Institute Inc. 2009).

RESULTS
The dominant Lgn-R mutation results in narrower and shorter leaves, reduced ligule, fewer leaves, failure of ear development, and reduced tassel branches compared with recessive wild-type alleles ( Figure 1). Introgression of Lgn-R into the Mo17 background partially suppresses the mutant phenotype (Table 1). These observations suggest the presence of genetic modifiers of Lgn-R that are polymorphic between Mo17 and B73. To identify and localize modifying loci, we used the intercrossed RIL population developed from a B73 x Mo17 cross (referred  Length and width measurements are the mean value for four to eight individuals and leaf area (length · width) and LbW (length by width) measurements were calculated using length and width measurements. d All measurements are in cm units. to as IBM; Beavis et al. 1992;Lee et al. 2002). Crosses were made using Lgn-R pollen in the B73 background onto IBM lines to give rise to IBM RIL x Lgn-R/+ F1s. Stand counts, leaf number, leaf length, and leaf width measurements were made on the 6th leaf from the tassel of each F1 indivdual grown in 2 locations: Gill Tract, CA, and West Lafayette, IN, in the summer of 2009.
Leaf length and width measurements were highly correlated across environments within and across locations ( Figure 2). Correlation of length and width traits for location GT (CA) had an r = 0.7397 ( Figure  2A). Both length and width measurement correlations in location IN and location GT were statistically significant at P , 0.001 (data not shown). We calculated two other traits by using the leaf length and width measurements: leaf area (length · width) and a length width ratio (length:width). The calculated leaf area for each individual was also highly correlated across environments (r = 0.6839 and P , 0.001) as well as within location IN (r = 0.9346 and P , 0.001) ( Figure 2B). The correlation of leaf measurements allowed a PCA to generate single values that better represented the genetic effect on leaf development of each IBM RIL. Thus, use of the PCA should increase both the sensitivity and accuracy of QTL mapping.
To create values for the multiple traits measured, PCA of covariances defined a single trait for length and width measurements across both locations (LW_all) as well as locations separately (LW_IN and LW_GT); and finally, leaf area across all locations (LA_all) and leaf area in IN only (LA_IN). Principal component values for LW_all, LW_IN, LW_GT, LA_all, and LA_IN explained much of the variation in the measured traits (87.06%, 94.08%, 99.01%, 83.83%, and 96.74%, respectively).

QTL mapping
One-dimensional genome scans identified a single QTL, sympathy for the ligule (sol), on the short arm of chromosome 1 (Table 2). Multiple traits shared this QTL, and it was detected with: count, location IN2 (count_IN2); leaf area, location CA (LA_GT); leaf area, location IN (LA_IN); location IN (length and width PCA value; LW_IN); and length and width for all locations (LW_all) (Figure 3 and Table 2).
Localization of sol by a two LOD drop-off for the QTL affecting LW_all mapped sol to likely be between 92.9 and 121.6 cM on chromosome 1 in the IBM ISU v4 map. A 95% Bayesian credible interval was substantially less specific, estimating sol to be between 5.6 cM and 133.9 cM on chromosome 1.

Validation of extreme rescue phenotypes in IBM x
Lgn-R detects the lcf locus In 2010, a subset of the IBM RIL x Lgn-R F1 crosses (IBM18, IBM30, IBM69, and IBM72), which had previously exhibited suppression of the mutant phenotype, was grown at the Purdue Agronomy Center for Research and Education (Table 3 and Figure S1). This subset of IBM RILs rescued much of the Lgn-R phenotype, displaying leaf length and width measurements approaching wild-type siblings (Table 3 and Figure S1). In 2011, the same subset of IBM RILs was grown as IBM RIL x Lgn-R/+ F1 and (IBM RIL x Lgn-R) x B73 F1 (a B73 recurrent parent BC1 of the IBM RIL carrying Lgn-R) plants were grown at the Gill Tract farm in Albany, CA. Phenotypic rescue was again observed. Thus, over 3 years and in all locations, these RIL were able to rescue the Lgn-R mutant phenotype (data not shown). On close examination of the genotypes among these "rescuing" IBM RILs (Table 3 and Figure  S1) as well as those RILs identified in Table 1 (IBM4, IBM25, IBM47, IBM55, and IBM65), a region on chromosome 1 between 97.5 cM and 99.7 cM was invariantly Mo17 in these RILs ( Figure 4 and Table S2). This region is a subset of the previously identified sol interval, which mapped to 94 and 133.5cM. In addition, a region of chromosome 7 between 65.7 cM and 69 cM was invariantly B73 in the rescuing RILs ( Figure 4 and Table S2).
We tested for the effect of the region on chromosome 7 on suppression of Lgn-R expression while controlling for the effect of segregation at sol. Phenotypic data from the full 63 RIL mapping population ( Figure 4 and Table S2) were analyzed in pairwise marker regression tests using the marker umc2145 as a proxy for sol and each marker in the chromosome 7 segment (Figure 4 and Table S2). Each regression included the umc2145 genotype, the genotype of a marker from chromosome 7, and the pairwise interaction between these two markers. The inclusion of a marker from the chromosome 7 region increased the fit of the model to the data and returned significant sol x lcf interaction terms indicating an epistatic relationship between the sol and lcf variants. Thus, this region contains a second QTL interacting with sol, which we name lucifer (lcf). Of all the markers tested within the invariantly B73 region of the rescuing RILs, the marker npb1 had the strongest interaction with sol. Figure 5 shows the effect of genotype at sol (marker umc2145) and lcf (marker nbp1) on the first principal component of leaf measurements. The sol Mo17 allele partially suppresses the effect of Lgn-R ( Figure 5) and occurs as an invariant region within the rescuing RIL ( Figure 4 and Table S2). Consistent with the occurrence of a B73 segment from chromosome 7 in the rescuing RIL (Figure 4), the lcf B73 allele enhances rescue of Lgn-R by sol Mo17 (Figure 5). For all phenotypic traits examined, except for stand count data, lcf (marker nbp1) modifies sol (marker umc2145) such that when sol Mo17 and lcf B73 are found together, the Lgn-R phenotype is substantially suppressed (Figure 5 and Figure S2).

Two-dimensional genome-wide scan for genetic interactions
The detection of a single pairwise interaction and positive transgressive epistasis demonstrated that we missed QTL due to epistasis in our single-dimension QTL scan. The scantwo function (R/qtl, method marker regression) was used to examine all pairwise combinations of markers by regression (Broman et al. 2003). Experiment-wide permutation thresholds (1000 permutations, threshold of alpha , 0.05) were calculated from the scantwo results for each phenotype. To simplify the output, summaries display only the best QTL-pair for each chromosome pair. Putative interactions were detected two ways. First the data were summarized to display differences between the full model (two QTL model with interactions) and reduced models (e.g., a two QTL model with no interaction; Broman et al. 2006). The second method maximized the interaction LOD for a pair of positions on each chromosome, thus returning the pair with the greatest synergistic effect (Broman et al. 2006). LODs were calculated for each summary type for each trait. Significant interactions are summarized in Table 4. Almost all interactions above their respective experiment-wide permutation thresholds included the sol region on chromosome 1. This finding is not surprising, given the dramatic impact of sol and the observation that it was the only main effect QTL identified, thereby returning greater LOD scores for models that include this QTL. Genome-wide multiple testing correction via permutation was used and significant interaction was detected between chromosome 1 at the sol QTL and chromosome 7 at lcf (Table 4), confirming our observation within the suppressing RIL. lcf was detected using the PCA value for length and width using either CA or the PCA from all locations (IN and CA). Additional two-QTL models were significant at unlinked locations on chromosome 1 (  (Table 4). A paralog of lgn, sister of liguleless narrow (sln; GRMZM2G009506), is also located on chromosome 5 but at an unlinked position at the opposite end of the chromosome, excluding the possibility that sln encodes this QTL.

Confirmation of sol effects on leaf development in segregating material
In the 2011 field season at Gill Tract Farms (Albany, CA), an evaluation of sol and lcf effects on leaf and plant morphology was performed in segregating families. Four IBM lines that provided rescue of Lgn-R in the aforementioned experiments (IBM18, IBM30, IBM69, and IBM72; Table 3) were backcrossed to B73. Leaf measurements as well as plant height were recorded in the IBM RIL x Lgn-R/+ BC1 F1s. Individuals were classified as either severe Lgn-R mutants, mild expressing mutants, or nonmutants. The genotype of marker umc2145 was used as a proxy for sol. The markers bnlg1792 and npb1 are both located at 66 cM on chromosome 7 and were used to confirm the lcf genotype, which was not segregating in these families because of the homozygosity for the lcf B73 allele in the RIL and B73 being the recurrent parent. A genotype of either B73, Mo17, or B73/Mo17 heterozygote was noted for each locus in each individual (File S1). The effects of the IBM RIL genetic background, sol genotype, Lgn-R mutation, and the interaction of Lgn-R and sol were estimated using a Standard Least Squares model for each of the three phenotypes: leaf length, leaf width, and plant height ( Table 5). The genetic background effect is defined by the filial generation and genetic background (e.g., B73, Mo17, or the 63 RILs) separate from the genotype at sol. The sol effect estimates the significance of the genotype at umc2145. Finally, the n a RIL x Lgn-R/+ F1 genotypes were identified as rescued phenotypes in the 2010 West Lafayette, IN, growing season. b Length, width, leaf area, and length by width (LbW) were phenotypic measurements made on the 6th leaf from the tassel. Leaf area (length · width) and LbW (length by width) measurements were calculated using length and width measurements.
effect the genotype at lgn was also included, as these familes are segregating 1:1 for Lgn-R/+ and wild-type lgn homozygotes. All three primary parameter estimates as well as the interaction of sol with Lgn-R were statistically significant ( Table 5). The interaction between the sol genotype and mutant phenotype indicate that sol impacts phenotype within the IBM x Lgn-R/+ individuals to a greater degree than in lgn wild-type siblings.

Genotype x environment interaction
Genotype by environment (GxE) interactions were estimated in the 63 IBM RIL mapping crosses using a Standard Least Squares model. We determined the effects of sol and lcf and their interaction (using markers umc2145 and nbp1, respectively) while also considering the growth environments as factors. Across all environments (CA and IN), we determined GxE and (GxG) x E interaction ( Table 6). The interactions between sol and lcf (P , 0.0001) as well as sol x lcf x environment (P , 0.0003) were both statistically significant (Table 6). If we examine the leaf area phenotype within each location, sol was significant in both environments (P = 0.0223, CA and IN both needed) but the interaction between sol and lcf was only detected in IN (P , 0.001; Table 7). It appears that some aspect of the environment or cultural practice that differs between the IN and CA field sites influences the expression of Lgn-R phenotypes and with it lcf and the interaction effect of lcf and sol.

Leaf morphology QTL comparison
Tian and Co-Workers (2011) previously carried out a genome-wide QTL mapping experiment for leaf morphology using the nestedassociation panel, which includes a Mo17-B73 contrast. We compared our sol QTL region (94 cM to 133.5 cM on the IBM ISU v4 map; Table S1) Figure 4, and Table S2) from 97.5 cM and 99.7 cM was converted to 62,995,272 bp to 66,027,760 bp on chromosome 1 in the AGPv1 build. Not all SNPs could be mapped to AGPv2; those SNPs that were unambiguously mapped to locations in the AGPv2 maize assembly (http://www. maizesequence.org) also are presented in Table 8. In total, four jointlinkage analysis QTL, 21 significant GWA SNPs, and 1 indel fell within the Bayesian credible interval for sol for leaf angle, leaf length, or leaf width. The narrowed QTL region did not include any QTL or SNP/indels from the work of Tian and et al. (2011). Also, no traitassociated SNPs (Yu et al. 2012) for leaf morphology traits overlapped with sol. Likewise, of the leaf morphology QTL identified in the IBM population by Mickelson et al. (2002), none were linked to sol. Thus, by using natural variation to search for mutant suppressors and genetic interaction, we have detected the gene, sol, for leaf morphology. In addition to detecting sol, we have also detected an interacting locus, lcf, opening up the possiblity of constructing a genetic pathway including n Table 4 A summary of the maximum interaction LOD scores above an experiment-wide permutation threshold for a two-QTL model (scantwo, R/qtl)   the products of lgn, sol, and lcf all heretofore unknown determinants of leaf growth and architecture in maize.

DISCUSSION
We have used natural variation affecting the penetrance of the leaf morphology traits conferred by the semidominant Lgn-R mutant to identify a locus on the short arm of chromosome 1, sol. In addition to the sol locus, two-dimensional QTL scans identified a second locus, lcf, on chromosome 7 that interacts with sol to determine the overall degree of phenotype rescue in Lgn-R mutants. Lgn-R was shown previously to affect ligule and auricle development (Moon et al. 2013) but can also cause termination of the development of the maize plant before ear and tassel production. In the 2009 field season, IBM RIL x Lgn-R/+ F1 individuals grown in two different locations, IN and CA, were used to detect and localize sol and lcf by regression-based QTL analysis methods. The sol QTL is a novel gene as recent maize QTL mapping experiments (Mickelson et al. 2002;Tian et al. 2011;Yu et al. 2012) identified QTL affecting leaf morphology not contained within the narrow interval for sol defined by the IBM RILs that rescued the Lgn-R phenotype to near wild-type leaf measures (Table 3, Figure 4, Figure S1, and Table S2). Four IBM RIL x Lgn-R/+ F1s displayed a near-rescued phenotype in 3 years at four locations (IBM18, IBM30, IBM69, and IBM72; Table  3 and Figure S1). These RILs, as well as five others with near wild-type leaf measurements from the 2009 RIL mapping experiments (IBM4,  IBM25, IBM47, IBM55, and IBM65; Table 1), are invariantly Mo17 at the sol locus. We referred to these at the "rescuing RILs." Likewise, these nine IBM RIL had B73 alleles at the location of lcf (Figure 4 and Table S2). These near wild-type phenotypes were recorded at four locations in 3 years both as IBM RIL x Lgn-R/+ crosses and as B73 x (RIL x Lgn-R/+) BC1 F1 plants for IBM18, IBM30, IBM69, and IBM72, indicating the additivity or dominance of sol Mo17 .
The marker effect plots for umc2145 and npb1 linked to sol and lcf demonstrate transgressive positive-epistasis such that the optimum combination of sol Mo17 /lcf B73 is significantly rescued and all other combinations are indistiguishable ( Figure 5 and Figure S3). The fact that these interactions occur in IBM RIL x Lgn-R/+ F1 material and that Lgn-R was isolated in a B73 background further support the additive or dominant nature of sol Mo17 but do not clarify whether lcf B73 is additive or recessive. Lgn-R x Mo17 BC1 and BC2 generations exhibited partially suppressed Lgn-R phenotypes (data not shown). This finding is consistent with lcf enhancing the sol Mo17 suppression of Lgn-R but not being required for sol Mo17 expression.
The interaction between sol and lcf was demonstrated in the marker effect tests as well as the genotypes of a set of fully suppressed RIL x Lgn-R/+ crosses. All phenotypes except stand counts in CA (count_GT) are closer to wild type when the sol allele (umc2145) is Mo17 (blue line; Figure 5 and Figure S2). However, sol B73 IBM RIL x Lgn-R/+ F1s have narrower leaves much like the original B73 Lgn-R isolate described by Moon et al. (2013) (Figure 1). Similarly, the lcf alleles impact all phenotypes coordinately and the lcf B73 allele always promoted wild-type phenotype expression ( Figure 5).
Because the loss-of-function allele at lgn had no visible phenotype in Albany, CA, the Lgn-R allele is predicted to be an antimorph (Moon et al. 2013). Thus, trans-regulatory polymorphisms increasing transcription from the lgn locus would be expected to enhance the Lgn-R phenotype. Lgn-R/+ plants exhibit increased accumulation of the mRNA encoded by the paralog, sln, in shoot apices whereas lossof-function alleles did not alter sln mRNA accumulation (Moon et al. 2013). The expression levels of lgn and sln in RNAseq data from B73 and Mo17 were made available by the Pat Schnable lab at Iowa State University, ahead of publication, via James Schnable's qTeller tool n (http://qteller.com/). Consistent with the penetrance of the Lgn-R phentypes in B73 and Mo17, lgn mRNA is accumulated at a greater level in the apices of B73 as compared with Mo17 ( Figure S3). The opposite expression difference was observed for sln, which exhibited a greater accumulation of mRNA in RNA isolated from Mo17 shoots than was observed for B73 ( Figure S4).
Integration of QTL, genetic interaction, and mRNA expression data provide a potential molecular mechanism for the modulation of leaf morphology and shoot development by sol and lcf. The sol QTL on chromosome one could be explained by an allele at a regulatory gene in Mo17 that suppresses the expression of lgn, resulting in suppression of the lgn phenotype in the presence of the dominant mutant allele. The presence of multiple splice forms of lgn (http://www. maizesequence.org) make this a possible candidate mechanism for the sol QTL, as well. Alternatively, expression difference need not underlie the sol or lcf QTL. Coding sequence changes to known genes or the presence of coding sequence in the Mo17 chromosome not present in the B73 reference, and therefore unannotated for expression, could be responsible for the suppression of Lgn-R. The expression of sln is greater in shoots of Mo17 and suppression of sln expression by the n Mo17 are indicated as "yes." f QTL were previously identified by Tian et al. (2011) and are partitioned in the three different phenotypes: leaf angle, length, and width. g Indicates that SNPs have yet to be mapped to B73v2 map. h Indicates the SNP or indel listed falls outside the QTL supporting interval position (cm). n Table 7 Examination of 2009 data set for main effects of sol, lcf, and location as well as genotype x environment (GxE) interactions calculated for leaf area, treating locations separately a DF is the degrees of freedom for each parameter in the model. The full model and error DFs are included. b x indicates interaction between two or more parameters within the model. Genotype of sol and lcf were used for genotype x environment (GxE); IN and CA define the location parameter. Main effects and interactions were calculated using a Standard Least Squares model (JMP, version 8.0.1, SAS Institute Inc., 2009). Ã Denotes statistical significance at a = 0.05. ÃÃ Denotes statistical significance at a = 0.01. ÃÃÃ Denotes statistical significance at a = 0.001. B73 allele at lcf could provide the mechanism for sol Mo17 -dependent mutant phenotype suppression by that QTL. The hypothesis of a lgn-affecting regulatory QTL at sol is consistent with the requirement of sol Mo17 for the expression of lcf B73 and the previous proposal that lgn phenotype manifestation is due to up regulation of sln in the shoot apex by the dominant Lgn-R allele (Moon et al. 2013). Testing of candidates genes for sol and lcf, and the roles they may play in the expression regulation and alternative splicing of lgn and the unusual effect of the 8bp UTR insertion allele lgn-dAc (Moon et al. 2013), await future experiments.
Neither lcf nor the interaction between the sol and lcf QTL were detected in CA (Table 1 and Figure 3). Multiple leaf morphology measures differed between the growing areas used throughout the 4 years of experimentation. According to the National Oceanic and Atmospheric Administration, the mean maximum temperatures during the summer of 2009 indicate that of the two sites used, IN was 6213°F warmer ( Figure S5) than Gill Tract, CA. The expressivity of the lgn phenotype was strongest in West Lafayette, IN, than any of the other locations. The penetrance of the Lgn-R allele was so strong that gross morphological and quantitative measures of phenotype expression in IN look similar to the Lgn-R homozygotes in other areas. For example, heterozygous Lgn-R/+ plants exhibit a complete failure of all reproductive structures in the B73 background in West Lafayette, IN. Although any differences in these two locations including soil characteristics and management techniques could also explain the lack of lcf detection in this CA location. As compared with the Gill Tract, CA site, stronger expression of Lgn-R suppression was observed in the 2010 growing season in IN (West Lafayette), during the 2012 season in Davis, CA, and in Valle de Banderas, Nayarit, MX during the winter of 2012 (data not shown). All three of these sites were substantially warmer than the Gill Tract Farm in Albany, CA. Regardless of the cause, the elucidation of a novel genetic pathway affecting leaf morphology and plant architecture including Lgn-R, lcf, and sol, and multiple epistatic and environmental interactions was identified in this study of natural variation affecting expression of a dominant mutant phenotype.