Maize Cultivar Performance under Diverse Organic Production Systems

Maize (Zea mays L.) performance can vary widely between different production systems. The need for high-performing hybrids for organic systems with wide adaptation to various macroenvironments is becoming increasingly important. The goal of this study was to characterize inbred lines developed by distinct breeding programs for their combining ability and hybrid yield performance across diverse organic environments. Parent lines were selected from five different breeding programs to give a sample of publically available germplasm with potential for use in organic production systems with expired plant variety protection (Ex-PVP) and current commercial inbreds as benchmarks. A North Carolina Design II mating design was used to produce all possible cross combinations between seven lines designated as males and seven lines designated as females. A significantly positive general combining ability for the female inbred UHF134 suggests that it performs well in hybrid combination. Significant general combining ability was not observed for any male inbred line in this study. Several significantly positive specific combining abilities suggest that nonadditive genetic effects play an important role in determining yield in this germplasm. Further analysis revealed that hybrids containing either an Ex-PVP line or a commercial inbred line were on average superior to hybrids containing only inbreds developed by the cooperators of this study. This demonstrates the utility of testing inbreds from diverse sources when developing hybrids for organic production systems. R.D. Huffman, Dep. of Agronomy, Iowa State Univ., Ames, IA 50011; C.A. Abel, L.M. Pollak (retired), J.W. Edwards, and M.P. Scott, USDAARS Corn Insects and Crop Genetics Research Unit, Iowa State Univ., Ames, IA 50011; W. Goldstein, Mandaamin Institute, Elkhorn, WI 53121; R.C. Pratt, Dep. of Plant and Environmental Sciences, New Mexico State Univ., Las Cruces, NM 88003; M.E. Smith, Plant Breeding and Genetics, Cornell Univ., Ithaca, NY 14853; K. Montgomery, Montgomery Consulting, Maroa, IL 61756; L. Grant, Agricultural Experiment Station, New Mexico State Univ., Las Cruces, NM 88003. Received 14 June 2017. Accepted 26 Sept. 2017. *Corresponding author (paul.scott@ars.usda.gov). Abbreviations: AEA, average environment axis; Ex-PVP, expired plant variety protection; GCA, general combining ability; G  ́ E, genotype ́ environment; GGE, genotype and genotype  ́ environment; SCA, specific combining ability. Published in Crop Sci. 58:253–263 (2018). doi: 10.2135/cropsci2017.06.0364 © Crop Science Society of America | 5585 Guilford Rd., Madison, WI 53711 USA This is an open access article distributed under the CC BY license (https:// creativecommons.org/licenses/by/4.0/). Published January 15, 2018


2006
). Producing new organic maize varieties with high performance and yield stability is of great interest due to the continued increase in consumer demand for products derived from organic maize.
Approaches for meeting the challenge of producing broadly adapted, high-yielding hybrids for organic production systems have not been explored extensively.Due to differences in cultivar performance between different production systems, it has been suggested that separate breeding programs may be needed to optimize yields for organically produced grain (Murphy et al., 2007); however, a study of testcrosses derived from a set of recombinant inbred lines evaluated in organic and conventional systems found high genetic correlations for grain yield across the two production systems (Lorenzana and Bernardo, 2008).One explanation for that finding may be associated with the prior selection of parental inbreds from a circumscribed germplasm base developed under conventional conditions.
Production practices in organic systems are very distinct from conventional systems and vary widely from farm to farm, factors that may contribute to increased cultivar ´ environment interaction in organic systems and in comparisons with conventional systems (Murphy et al., 2007;Lammerts van Bueren et al., 2011).This has prompted several breeding programs, including those contributing inbreds for this study, to develop hybrids specifically for organic production systems where integrated farm management practices are followed.These breeding programs use diverse germplasm that do not contain transgenes, and they emphasize selection under organic production conditions in both the inbreeding and testcross evaluation programs whenever possible.Additionally, they focus on traits of interest to organic producers such as native insect and disease resistance, tolerance to the presence of weed competition and mechanical weed control, and productivity in zero-or low-synthetic-chemical management systems (Goldstein et al., 2012).
The need for developing organic maize hybrids with high yield performance in differing environments presents both a challenge and an opportunity for breeding programs in both the public and private sector.The success of a hybrid maize breeding program targeted for organic production requires the testing of all germplasm under optimal conditions to determine the yield potential, along with multienvironment testing in diverse locations to screen for biotic and abiotic stress tolerance (Braun et al., 1996;Troyer, 1996).Generating hybrids from inbred parents adapted to different environments may increase the adaptability of those hybrids across diverse organic environments.Our programs have attempted to produce widely adapted organic hybrids by crossing genetically diverse inbred lines with varying maturities and adaptation to diverse environments and geographical locations.
The objective of this study was to determine under certified organic conditions the combining ability of maize inbred parents developed by different breeding programs.We assessed the performance of a series of hybrids produced by intercrossing inbred parents selected for their yield potential and adaptation to organic systems.By using a North Carolina Design II, or factorial mating design (Comstock and Robinson, 1952;Beil and Atkins, 1967;Cukadar-Olmedo et al., 1997), we were able to determine the general combining ability (GCA, defined as the average performance of a given parent in hybrid combination) and specific combining ability (SCA, defined as the average performance of a hybrid relative to the average performance of each parent) (Sprague and Tatum, 1942).The suitability of the hybrids for organic production systems was evaluated across macroenvironments in a multiyear study to identify those hybrids with high yield performance and stability.We used the genotype and genotype ´ environment (GGE)biplot method to explore multienvironment trial data and evaluate hybrid performance.(Yan, 1999;Yan et al., 2000;Yan and Tinker, 2005).Results from this study demonstrate the utility of combining inbred lines from diverse origins in hybrid combination and testing in a wide variety of environments.

Field Procedures
Forty-nine hybrids were produced by crossing seven inbred lines designated as females to seven different inbred lines designated as males using a North Carolina Design II (Comstock and Robinson, 1952).Each of five cooperating institutions submitted one male and one female line developed in their breeding programs.These lines were selected by each institution for their high overall agronomic performance in organic production environments.Some were developed in organic environments, and some were developed in conventional environments but were selected based on traits desired by organic producers (Table 1).In addition, one male and one female inbred were expired plant variety protection (Ex-PVP) lines, and one male and one female inbred were current proprietary commercial inbreds.Both Ex-PVP inbred lines, LH82 (PVP certificate no.8500037, became available in 2003) and LH132 (PVP certificate no.8300148, became available in 2003), were originally developed by Holden's Foundation Seeds (Williamsburg, IA).A summary of the lines used in this study is presented in Table 1.Hybrid seed was produced in 2013 at the Iowa State University Agronomy Farm near Ames, IA, and at the Montgomery Consulting nursery near Maroa, IL.The resulting 49 hybrids were then evaluated at 10 location-year combinations during 2014 and 2015.In addition to the 49 hybrids in the study, one organic commercial hybrid was used as a check at all locations.All yield trial test sites were certified organic and complied with USDA certification.The name and code of each test location and hybrid combination can be found in Tables 2 and 3, respectively.Each location contained two replications planted in a randomized complete block design.All entries were grown in which Y ijkl is the observed yield value for each experimental unit, m is the grand mean, f i is the female effect, m j is the male effect, (fm) ij is the interaction effect between the ith female and jth male, v k is the environment effect, r l(k) is the replication effect nested within each environment, (vf ) ik is the interaction effect between the ith female and environment, (vm) jk is the interaction effect between the jth male and environment, (vfm) ijk is the interaction effect between the female, male, and environment, and e ijkl is the random residual error.The female, male, and female ´ male interaction model terms were fit as fixed effects, whereas all the remaining terms were fit as random effects.Normal quantile-quantile plots of studentized residuals were examined to check for normality and potential outliers.No observations appeared as outliers on residual plots, and all observations were within 3.5 SD of predicted value.The significance of the male, female, and male ´ female variance components were tested against the male ´ environment, female ´ environment, and male ´ female ´ environment random effects, respectively, and declared significant at P £ in two-row plots averaging 5.5 m long with 0.76 m between rows except at the New Mexico locations, which used one-row plots on 1.02-m centers in an irrigated bed and furrow system.Management practices were as uniform as possible across locations, but variation in equipment, fertilizer, and timing of management activities was unavoidable.Each plot was machine harvested with grain weight and moisture recorded.Plots in New Mexico were hand harvested and shelled using an Almaco stationary sheller.Plot yield was determined using the grain weight of each plot adjusted to 15.5% grain moisture.

Statistical Analysis
Data from experimental hybrids, excluding the common check, were analyzed with an ANOVA corresponding to the following linear model using the software JMP Pro 11.0 (SAS Institute, 2013):  0.05.Least significant differences, reported in Tables 4 and 5, are calculated using Fisher's LSD at a significance threshold of 0.05.The LSD value for differences between males is not reported because the male effect was not significant.
Estimates and significance for the GCA and SCA were determined using terms in the linear model outlined previously.Male and female model effects are estimators of the GCA for each parent, respectively, whereas the female ´ male interaction model effect estimates SCA (Hallauer et al., 2010).Specific contrasts were used to further partition each GCA or SCA effect to determine if any individual parent lines or testcrosses had a significant difference in their average performance based on definitions outlined by Sprague and Tatum (1942).
A genotype ´ environment (G ´ E) matrix was constructed using the mean yield of each genotype in each environment.
Phenotypic trait measurements (such as grain yield) include the combined effects for genotype, environment, and the G ´ E interaction.Not all of these effects are relevant in cultivar evaluation, as individual environments cannot be replicated.Thus, only the GGE are considered when evaluating the performance of individual hybrids (Yan, 1999;Yan et al., 2000).Each biplot is constructed using the primary and secondary effects, or Principle Components 1 and 2 respectively, as main axes.The GGE biplots were generated from this matrix using R software version 3.1.2and the 'GGEBiplotGUI' package outlined by Frutos et al. (2014).Biplot parameters included: no data scaling, testercentered G + GE, and the column metric preserving singular value partitioning method.Environmental yield stability was determined by the GGE biplot as the distance from each hybrid to the average environment axis (AEA).effects were the largest variance component.The impact of environmental effects on yield is shown in Table 5.
The G ´ E variance components were significant, but the magnitudes of these effects were small.The female and male model effects are estimates of GCA for each parent, whereas the female ´ male interaction estimates the SCA.
With the GCA and SCA reflecting additive and nonadditive genetic effects, respectively, the significance of each indicates the types of gene action responsible for determining yield (Hallauer et al., 2010).Further partitioning of the female parent GCA model term revealed that two inbred lines, AR2 and UHF134, had significant GCAs (Table 8).With UHF134 having the only positive GCA, this inbred line had the best GCA of all the lines included in this study.A lack of significance for any male parent suggests that none of the male lines outperformed the rest of the group in this study.Similar to the female effect, further partitioning of the female ´ male interaction (SCA) effect revealed that 10 hybrids had significant SCA estimates (Table 8).Four hybrids had positive SCA estimates, and the other six hybrids had negative SCA estimates.Significance of genotype effects supports the use of a GGE-biplot for yield data analysis.High performing genotypes can be easily identified using these plots which utilize a principle component analysis to generate major axes.This analysis showed that principle component one (PC1) accounted for 33.5% of the total variability alone while principle component two (PC2) accounted for 16.6%.Thus, principle components 1 and 2 captured ~50% of the total variability (Fig. 1-4).It should be noted

RESULTS
The mean yield for all hybrid combinations and locations included in the study was 7.62 Mg ha −1 .Mean yields for all 49 entries evaluated in this study are presented in Table 4. Row and column means were calculated to determine the average performance of a given inbred line across all seven hybrids.Mean yield data with statistical separations are shown in Supplemental Table S1, and all data used in the study are presented in Supplemental Table S2.
Specific contrasts of hybrid groupings were performed to test hypotheses of interest (Table 6).First, we tested the hypothesis that hybrids made with inbreds contributed by different cooperators performed differently than hybrids of inbreds contributed by the same program.No significant difference was found between the within-group and between-group hybrids, leading us to reject the hypothesis.Second, we tested the hypothesis that hybrids made from inbreds contributed by the cooperators performed differently than hybrids containing an Ex-PVP line or (in a separate contrast) a commercial line.Hybrids containing an Ex-PVP line or a commercial line were significantly different than hybrids of inbreds contributed by the cooperators.
All sources of variation investigated were significant except the male parent GCA (Table 7).Environmental   that while this method is widely used to evaluate yield trial data, it does not provide a statistical test of mean comparisons.Conclusions drawn from this method should be regarded with an appropriate level of caution.
A "which-won-where" biplot (Fig. 1) allows for the best genotypes in each environment to be easily recognized.Identification of superior cultivars requires a polygon to be drawn by connecting the genotypes farthest from the biplot origin.Lines perpendicular to each polygon side and originating from the biplot origin are added to separate the plot into several sectors (Frutos et al., 2014).Hybrids occupying a polygon vertex showed superior performance in the environments contained in the same sector.In this study, the which-won-where biplot revealed that all 10 environments were contained in four sectors (Fig. 1).The clustering of environments within relatively few of the sectors is in part a consequence of the low level of G ´ E observed.
The "discriminativeness vs. representativeness" biplot (Fig. 2) determines an average environment based on genotype performance at each location and draws an AEA to allow the representativeness and discriminating power of each environment to be visualized.The average environment is estimated by averaging the coordinates of all environments with the AEA determined by transecting these coordinates with a line originating from the biplot origin.The farther an environment vector is from the biplot origin, the more discriminating (or informative) power that environment has, whereas a shorter vector indicates less discriminating ability.Additionally, those environments having a small angle with the AEA are more representative of other test environments , GCA effect of male lines.
Fig. 1.Which-won-where view of the genotype and genotype ´ environment (GGE)-biplot showing all 49 maize hybrids and 10 environments.Polygon vertices represent the highestor lowest-yielding hybrid in each environment for a defined sector.The identity of each hybrid and environment is shown in Tables 2 and 3.
Of the environments included in this study, Jefferson, IA-2015 (E6b) was found to be the most discriminating and representative of all environments based on its vector length and angle with the AEA.
The "mean vs. stability" biplot (Fig. 3) also defines the average environment and draws an AEA through these coordinates.The single arrow on the AEA line points towards genotypes with the highest mean yield across all environments, whereas the length of the vector from the AEA line represents yield stability.A longer vector correlates with lower yield stability, whereas a short vector correlates with high yield stability (Yan, 2001).
Further analysis of hybrid yield stability revealed that crosses between inbreds from the same breeding program Environment vector length represents discriminating ability, whereas the vector's angle with the AEA represents representativeness.
Longer vectors are most discriminating, and smaller angles with the AEA are most representative.

Fig. 3. Mean vs. stability view of the genotype and genotype
´ environment (GGE)-biplot comparing all 49 hybrids.Mean yields for each hybrid increase moving left to right on the average environment axis (AEA, solid green line), whereas those genotypes with the highest yield stability have markers closest to the AEA.contained similar yield stabilities to crosses between inbreds from different breeding programs.As shown in Fig. 5, within-program hybrids expressed high to very low yield stability, whereas between-program crosses showed very high to very low yield stabilities.Three of the withinprogram crosses had high stability, with another three having low stability and one hybrid having very low stability.Over 50% of the between program crosses had high to very high yield stability, with 85% of the remaining hybrids having low stability.Of the nine hybrids found to have very high yield stability, all were between-program crosses, which may support the notion that high yield stability can be achieved by crossing inbred lines developed from geographically diverse breeding programs.
The "ranking environments" biplot (Fig. 4) defines a hypothetical ideal environment based on the test environments and allows for comparisons to be made between actual environments and a hypothetical ideal environment.The ideal environment is defined as the most discriminating and representative, with the biplot generating a ranking of the test environments based on both criteria.The hypothetical ideal environment is located at the center of the concentric circles (Fig. 4), with the test environment closest to the center of the concentric circles being the most discriminating and representative (Yan, 2001).As previously suggested in Fig. 2, the ranking environments biplot also identified Jefferson, IA-2015 (E6b) as the environment closest to an ideal testing location for the hybrids evaluated in this study (Fig. 4  ´ environment (GGE)-biplot comparing all 10 environments based on both discriminating ability and representativeness.
A hypothetical ideal environment for testing is located at the center of the concentric circles, with those environments closest to the ideal environment the most representative of this environment.

DISCUSSION
One novel aspect of this study is that we combined inbred lines developed by five different breeding programs to make hybrids.For each breeding program, one hybrid was made with both parents originating from that program.The other hybrids in the study were made from inbreds contributed by different breeding programs.A male and a female Ex-PVP and a male and a female commercial line were included as benchmarks.This design gave us the opportunity to compare the performance of hybrids developed within cooperators' breeding programs with the performance of hybrids made of inbreds from different breeding programs.We reasoned that when combining inbreds from different cooperators' breeding programs, the added diversity may result in superior hybrids.This hypothesis was not supported by the results, however, because the means of the within-group and between-group hybrids were not significantly different from each other (Table 6).Combining inbreds between breeding programs still has the benefit of greatly expanding the germplasm available to breeders.In addition, we compared the performance of hybrids made exclusively with cooperators' inbreds with the performance hybrids made with either Ex-PVP or commercial lines.Hybrids made with cooperators' inbreds alone did not perform as well as hybrids made between cooperators' inbreds and either Ex-PVP or commercial lines, suggesting that even in organic systems, current or past commercial germplasm that was not selected in organic production systems can still contribute to high-yielding hybrids.This is consistent with conclusions drawn by Lorenzana and Bernardo (2008).It is important to note that we only evaluated yield in this study and organic producers are often interested in other traits that were not evaluated, such as nutritional quality or impact on soil health.Comparison of inbreds developed for organic production systems with commercial inbreds for traits of interest to the organic community would be particularly interesting.
An ANOVA revealed that many of the linear model terms were significant (Table 7).The significance of the environment effect and interactions between environment and female, male, and female ´ male indicate that there was a significant difference in the mean yields and hybrid rankings in the different environments evaluated.Table 5 illustrates the variation in yields across the geographically diverse yield trial locations and provides data with a commercial check hybrid for comparison.Although each location was certified organic, one major difference in management practices was that the New Mexico locations were irrigated.These data support previous reports of differing hybrid performance when evaluated under different management practices (Carr et al., 2006;Lammerts van Bueren and Verhoog, 2006).We conclude that while most inbred parents used to produce each hybrid were identified for their potential use by organic producers, the resulting hybrids varied in performance when grown across diverse organic locations, and significant though small G ´ E effects were observed.Testing each hybrid in diverse growing conditions not only allowed us to identify which hybrids performed well in specific organic locations, but also those with high yield performance across several organic locations evaluated.Although local target environment testing is clearly needed to identify the best hybrids for a location, our data suggest that there is value in testing hybrids outside of their local area of adaptation as well to identify additional hybrids that perform well in a given location.Although not a direct objective of this study, identifying high-yielding hybrids with locally specific adaptation is of interest to organic producers, as private seed companies invest a majority of their resources into developing hybrids for the larger global seed market.
The lack of a significant male genetic effect is due to the combination of a low male effect variance component estimate and a high male ´ environment effect variance component estimate.The low male effect may indicate a lack of diversity in the male parents.Examination of the pedigrees of these lines (Table 1) does not support this conclusion.The male lines used in study represent the Iodent, Lancaster Sure-Crop, and Iowa Stiff Stalk (BSSS) families, as well as other germplasm.We conclude that the large male ´ environment interactions are an important contributor to the lack of male genetic effects.
Estimates for the genetic effects regulating grain yield revealed that both the GCAs and SCAs were significant.Since the GCA is associated with additive genetic effects, selection and use of lines with positive values would be beneficial, as favorable alleles could be passed on to progeny.Conversely, the SCA is associated with nonadditive genetic effects such as dominance and epistasis.With UHF134 having the only positive GCA for either female or male lines, using this inbred line as a female parent would be advantageous for hybrid development and breeding.Although none of the male parents had significant GCAs, several SCAs were found to be significant, resulting in some individual hybrids such as UHF134/ LMPNG28, which had an average yield >8.8 Mg ha −1 .
The which-won-where biplot was used to identify which hybrid performed best at each location.One of four hybrids won in all 10 locations, with eight locations won by either UHF134/NuMex-01 or UHF134/LMPNG28.UHF134/NuMex-01 (G42) won in Iowa and Wisconsin locations, whereas UHF134/LMPNG28 (G41) won in Iowa and New York.This biplot view also allowed us to determine if hybrids produced by crossing two inbreds developed in the same breeding program had an advantage in their "home environments."None of the hybrids evaluated in this study won in its home location, with all of the highest-yielding hybrids having been generated

Fig
Fig. 2. Discrimitiveness vs. representativeness view of the genotype and genotype ´ environment (GGE)-biplot comparing all 10 environments.The solid blue line represents the average environment axis (AEA).Environment vector length represents discriminating ability, whereas the vector's angle with the AEA represents representativeness.Longer vectors are most discriminating, and smaller angles with the AEA are most representative.

Fig. 4 .
Fig. 4. Ranking environments view of the genotype and genotype´ environment (GGE)-biplot comparing all 10 environments based on both discriminating ability and representativeness.A hypothetical ideal environment for testing is located at the center of the concentric circles, with those environments closest to the ideal environment the most representative of this environment.

Fig. 5 .
Fig. 5. Heat map of lines used in this study and their resulting hybrid yield stability across all test environments.The environmental yield stabilities for within-program crosses are included in the main diagonal of the table (boldface type), whereas betweenprogram crosses are included above and below the main diagonal.Dark green (very high stability, 0-0.61 Mg ha −1 location −1 ), light green (high stability, 0.62-1.56Mg ha −1 location −1 ), pink (low stability, 1.57-3.14Mg ha −1 location −1 ), red (very low stability, >3.15 Mg ha −1 location −1 ).

Table 1 .
Inbred lines used in the North Carolina Design II mating design.Org indicates that this inbred was tested in certified organic conditions in development.Con indicates that the inbred was evaluated in noncertified production systems during development but was selected for this study based characteristics desirable for organic production systems.

Table 2 .
Environment codes for biplot analysis.
† Data collected by ncdc.noaa.govusing nearest weather station.‡ Received six to eight irrigations equivalent to ~10 cm of rainfall per irrigation.crop science, vol.58, january-february 2018

Table 3 .
Hybrid codes for biplot analysis.

Table 4 .
Mean yield for the 49 hybrid combinations.

Table 5 .
Mean yield for all 10 location-year combinations.A single commercial hybrid check was grown at all locations for yield comparison.

Table 6 .
Contrasts between least squares means of three hybrid groupings.
*, *** Significant at a = 0.05 and 0.001, respectively.† Within-group hybrids are made with two inbreds contributed by the same breeding program, whereas between-group hybrids are made with both inbreds contributed by different breeding programs.This analysis excludes all hybrids made with Ex-PVP and commercial inbreds.‡ Contributed hybrids are those with both parents contributed from a cooperators breeding program.§ The Ex-PVP group consists of all hybrids with at least one parent containing an Ex-PVP line.Hybrids containing commercial inbreds are excluded from this group.¶ The commercial group consists of all hybrids with at least one parent containing a commercial inbred line.Hybrids containing Ex-PVP lines are excluded from this group.

Table 7 .
Variance components for the yield linear model.

Table 8 .
General (GCA) and specific (SCA) combining ability estimates for inbred lines and hybrids for yield.