Twenty-four yellow to orange endosperm maize inbred lines developed through biparental crosses and backcrosses involving tropical and temperate germplasm were used (Supplementary Table 1). These inbred lines were crossed in a factorial mating scheme (Hallauer et al., 2010) at IITA's main research station at Ibadan, Nigeria (7°29′11.99″N, 3°54′2.88″E, altitude 190 m) during two dry seasons (from December 2010 to April 2011 and December 2011 to April 2012). The parental inbred lines were divided into six different groups each containing four inbred lines to generate five sets of factorial crosses. The inbred lines were separated into different groups based on similarities in their heterotic affinities and carotenoid content determined in previous studies (Menkir et al., 2008; Azmach et al., 2013) to generate the five sets of crosses. Thus, each set of crosses involved inbred lines with different genetic backgrounds to maximize combining ability and heterosis. Generally, all the lines served as both male in one set and as a female parent in another set. The mating groups were crossed using the following arrangement: set 1: G-I × G-II, set 2: G-III × G-I, set 3: G-II × G-IV, set 4: G-V × G-III, and set 5: G-VI × G-V. This crossing scheme resulted in 80 F1 hybrids.

The 80 F1 hybrids along with a commercial F1 hybrid check were arranged in a 9 × 9 lattice design and planted at four testing sites in Nigeria in 2012 and at two testing sites in 2013, making six agro-environments. The testing sites were Ikenne 3o42′E, 6o54′N, altitude 30 m; Saminaka 8o39′E, 10o34′N, altitude 760 m; Bagauda 8o19′E, 12o01′N, altitude 520 m; and Zaria 7o45′E, 11o8′N, altitude 622 m. The parental inbred lines were also tested along with the F1 hybrid trial separately in a 4 × 7 alpha lattice design at the Zaria and Saminaka field sites in both 2012 and 2013. Both F1 hybrid and inbred line trials were planted in two replications, where each entry was planted in a 5-m-long single row plot with a plant spacing of 75 cm between rows and 25 cm within a row. Two seeds per hill were planted but later thinned out to maintain one plant. The trials received all the agronomic management practices recommended for the respective field-testing site.

The data were recorded on a plot basis. Agronomic data were collected from all the six environments (Ikenne, Saminaka, Bagauda and Zaria in 2012, plus Zaria and Saminaka in 2013), whereas carotenoid data were collected from the four environments (Zaria and Saminaka in both 2012 and 2013). Flowering dates were recorded based on the number of days at which 50% of the plants within a plot had set visible silks (as silking date) and started shedding sufficient pollen (as anthesis date). Ear and plant heights were measured in centimeters (cm) from the base of the plant to the attachment line of the lower ear and the first tassel branch, respectively. Ear aspect was scored on a 1 to 5 scale, where 1 represented the clean, well-filled, uniform, and larger ears, while 5 represented the diseased, poorly filled, variable, and smaller ears. Plant aspect was also scored on a 1–5 scale, where 1 represented the uniform, clean, vigorous, and good overall phenotypic appeal, while 5 represented the weak, diseased, and poor overall phenotypic appeal. Shelled grain weight of each F1 hybrid was measured in kilograms (kg) per plot, and moisture content of the representative sample grain was measured for each shelled plot using a portable moisture tester. The shelled grain weight per plot in kilograms was converted to grain yield in metric tons per hectare (t/ha) by first adjusting for 12.5% w/v moisture content. Grain yield was not recorded for the inbred line trial, since several plants in each plot were self-pollinated.

Five to six representative plants of each F1 hybrid and all typical plants of each inbred line were self-pollinated to produce seed samples for carotenoid analysis, from the field trials planted at Zaria and Saminaka during both 2012 and 2013 (i.e., four environments). Shelled seeds from self-pollinated ears were dried further at ambient temperature avoiding exposure to direct sunlight. Samples of about 100 seeds were drawn at random from each inbred line and F1 hybrid and subjected to carotenoid analysis.

Carotenoids were analyzed at the University of Wisconsin. Carotenoid extraction from maize grain samples was performed using the method of Howe and Tanumihardjo (2006). The equipment used for carotenoid separation and quantification was high-performance liquid chromatography (HPLC) system (Waters Corporation, Milford, MA). Details of the procedures for the analysis were described in the study by Azmach et al. (2018). The levels of α-carotene, β-carotene (cis and trans isomers), β-cryptoxanthin, lutein, and zeaxanthin were quantified in μg/g dry weight (DW). Total carotenoid was calculated as the sum of the concentrations of α-carotene, lutein, β-carotene, β-cryptoxanthin, and zeaxanthin. Provitamin A was calculated by adding the concentrations of β-carotene, and half of each of β-cryptoxanthin and α-carotene concentrations, since β-cryptoxanthin and α-carotene concentrations can provide only one molecule of retinol each as opposed to β-carotene, which can be converted to two molecules of retinol (US Institute of Medicine, 2001).

Mixed-model analysis using PROC MIXED procedure of SAS® software (SAS Institute, 2012) was employed for ANOVA among the F1 hybrids and the inbred line parents, and to obtain least-square means for each of the carotenoid traits and agronomic traits on a single and across environment basis after correcting for block effects. For traits with significant genotype-by-environment interaction (G×E), Spearman's rank correlation coefficients were calculated between environments in the PROC CORR procedure of SAS® software to determine whether the significant interactions of the F1 hybrids and parental inbred lines with the environment were of cross-over types.

The variability among the F1 hybrids within the mating sets was further analyzed using the data for the crosses formed through the factorial mating in our study (i.e., excluding the data for the check). This provided mean squares for females, males, and female-by-male interactions, including their interactions with the environments in the PROC GLM procedure of SAS® software. The linear mixed model used for the genetic analysis was:

where Y is the observed value from each experimental unit; μ is the grand mean; El is the effect of the lth environment; Bk(l) is the effect of the kth block nested within replication within each environment; Sm(l) is an mth set or mating group effect in lth environment; ESl(m) is the interaction effects between sets and the environment; Fi(m) is the main effect of the ith female within a set; EFi(m) is the interaction between females effect and the environment; Mj(m) is the main effect of jth male within the set; EMj(m) is the interaction between males effects and the environment; FMij(m) is the interaction effects of the cross between ith female and jth male; EFMijl(m) is the interaction between the SCA effects and the environment; and Error is the pooled residual effect.

In the mixed model, environments, replications within environment, and blocks within replication were considered as random effects, while mating sets, inbred lines (females and males), and hybrids were considered as fixed effects. GCA effects were inferred from variances of the main effects of males and females, while SCA effects were inferred from that of female-by-male interactions (Hallauer et al., 2010).

Family means-based heritability (H) estimates and the associated standard errors were calculated based on the method and code for SAS software provided by Holland et al. (2003), this time, assuming the genotypes were random samples representing a reference population.

where V(G) is the variance due to genetic effect, V(GE) is the variance due to genotype-by-environment interaction, V(E) is the variance due to random errors, e is the number of environments, and r is the number of replications per environment.

Simple correlation coefficients were also calculated among carotenoids and agronomic traits. Correlation analyses were performed using PROC CORR of SAS software (SAS Institute, 2012).

Mid- and high parent heterosis and their significance were calculated as follows, according to the formula indicated in Hallauer et al. (2010) and Khorzoght et al. (2010).

where mpH is the mid-parent heterosis, bpH is the best parent heterosis, F1 is the hybrid value, and mp is an average value of the corresponding parents for the hybrid.

The significance of each hybrid value with respect to the corresponding mp and bp values was evaluated using t-test (Khorzoght et al., 2010):

where Tmp is the calculated t-value for F1 to mp comparison and Tbp is for that of F1 to bp comparison, r is the number of repeats, and EMS is the mean square of error. The probability associated with the t-test was calculated using Microsoft Excel®.

The analysis of variance showed that both the inbred lines and F1 hybrids displayed significant variability in the carotenoid content of the seed endosperm (Table 1; Supplementary Table 2). The genetic variation among parents and F1 hybrids represented the largest proportion of the total variance for all the measured traits, followed by the variation due to environments, and F1 hybrid-by-environment interaction. Spearman's rank correlation analysis showed significant (p < 0.0001) positive pairwise correlations between pairs of environments for the concentration of each carotenoid, indicating that the environment-by-F1 hybrid interaction for carotenoids was not of the cross-over type, which means that genotypes did not differ in their relative carotenoid content with environmental changes. The strong impact of genetic effects on carotenoid variability was further confirmed by our estimates of heritability, which ranged from 57% for α-carotene to 95% for lutein in the parent trial, and 57% for α-carotene to 92% for β-cryptoxanthin in the F1 hybrid trial. The heritability estimates for provitamin A in both inbred line and F1 hybrid trials were 81%.

The mean carotenoid concentrations averaged over the environments are presented in Supplementary Tables 3, 4 for the inbred line and the F1 hybrid trials, respectively. The average proportions and distribution of the carotenoids in the inbred line and F1 hybrid trials were similar (Figures 1, 2). The distribution of all carotenoids measured in both the parents and F1 hybrids was broad, except for α-carotene (Figure 2). The average concentrations of the carotenoids in the inbred line trial were 5.53 μg/g for lutein, 10.29 μg/g for zeaxanthin, 2.01 μg/g for β-cryptoxanthin, and 2.49 μg/g for β-carotene, whereas, in the F1 hybrid trial, they were 5.79 μg/g for lutein 10.47 μg/g for zeaxanthin, 2.14 μg/g for β-cryptoxanthin, and 3.04 μg/g for β-carotene. The mean provitamin A content in the inbred line trial varied from 1.68 to 6.65 μg/g, whereas in the F1 hybrids, it varied from 1.44 to 6.53. The male and female parental lines of a commercial F1 hybrid check (Oba super 2) accumulated 3.08 μg/g and 2.72 μg/g provitamin A in their endosperm, respectively.

The set 2 hybrids (GIII × GI) displayed the highest mean concentration for zeaxanthin (8.58 μg/g), β-cryptoxanthin (2.76 μg/g), and total carotenoids (23.54 μg/g) (Table 2). The mean β-carotene (4.16 μg/g) and provitamin A (5.17 μg/g) concentrations of the F1 hybrids in set 1 (G-I × G-II) were found to be the highest, though they did not differ statistically from those of sets 2 and 3 for β-carotene, or those of sets 2, 3, and 4 for provitamin A. The F1 hybrids in set 5 (GVI × GV) displayed the lowest provitamin A carotenoid concentrations. This set displayed the highest average lutein concentration (6.92 μg/g).

The female and male main effects (GCA) and interaction of each with the environment were highly significant for most of the measured carotenoids (Table 1). Although the female-by-male interaction (SCA) had significant effects on lutein, β-cryptoxanthin and provitamin A, the contribution of SCA to the total variation in hybrids was small, ranging from 7% to 16%. The interaction of SCA with the environment was not significant for almost all of the carotenoids.

Both the F1 hybrids and the inbred line parents exhibited significant differences in all the agronomic traits evaluated (Table 4; Supplementary Table 6). Hybrid means averaged across environments ranged from 56 to 63 days for days to anthesis (AD), from 0.66 to 5.87 t/ha for grain yield (GY), from 2.3 to 4.36 for ear aspect (EA), and from 2.45 to 4.34 for plant aspect (PA) (Supplementary Table 7). The F1 hybrids 24 (12 × 3), 32 (12 × 1) and 52 (20 × 12) had the highest grain yield (5.7 to 5.9 t/ha), but were not significantly different from circa half of the F1 hybrids, including the commercial check. The ear aspect and plant aspect scores of these top-yielding F1 hybrids were acceptable (< score of 3), while their provitamin A contents were higher than 5 μg/g, which were not significantly different from the highest provitamin A concentration in the F1 hybrid trial (6.5 μg/g).

Female and male main effects were highly significant for all agronomic traits (Table 6). The effects of female-by-male interaction (SCA effects) were significant for all the traits. Rank correlation coefficients for the agronomic characters were non-significant, suggesting the presence of changes in ranking of the hybrids across environments (data not presented). The proportion of female GCA effects relative to the total variation for the F1 hybrids ranged from 26% for plant height to 59% for SD, and the proportion for male GCA effect ranged from 27% for AD to 43% for EA. The SCA component of variance ranged from 14% for SD to 35% for PA, which were higher than the SCA component of variance observed for carotenoids.

The hybrids displayed either significant positive or negative correlations between combinations of carotenoids, except for β-carotene with lutein, zeaxanthin, and β-cryptoxanthin (Table 6). In the inbred trial, α-carotene with lutein (r = 0.42, p < 0.05), α-carotene with β-carotene (r = 0.49, p < 0.05), and β-cryptoxanthin with zeaxanthin (r = 0.66, p < 0.01) were found to be significantly correlated. The strength and direction of correlation reflected the positions of the carotenoid biomolecules along the linear representation of the carotenoid biosynthesis pathway and the underlying enzymes involved in the bioconversions and synthesis of the different carotenoid molecules. The correlation between β-carotene and β-cryptoxanthin was not significant, whereas the correlation between α-carotene and β-carotene were strong and positive.

Correlations between the carotenoids and agronomic traits were in general not significant (Table 7). However, the correlation of α-carotene with each of the agronomic traits was significant and negative, except with ASI and PA, whereas the correlations of total carotenoid with SD, AD, PH, and GY were significant and positive. Beta-carotene and β-cryptoxanthin were not associated with GY.

The regression of the carotenoid content of the F1 hybrids on mid-parent values detected significant (p < 0.01) but weak associations for lutein (r = 0.29), zeaxanthin (r = 0.39), and β-cryptoxanthin (r = 0.39) (Table 7). The regression coefficients for β-carotene, provitamin A, and total carotenoids were non-significant.

Fifteen of the F1 hybrids displayed significant high parent heterosis (hpH%) (Table 8), while 42 of the F1 hybrids displayed positive mid-parent heterosis (mpH%) for at least one of the carotenoids (Supplementary Table 9). The largest number of hybrids with positive mpH% heterosis was observed for β-carotene (22 hybrids) followed by β-cryptoxanthin (21 hybrids), zeaxanthin (17 hybrids), and lutein (14 hybrids). Significant mid-parent heterosis effects were detected for β-carotene, provitamin A, and total carotenoids, varying from 15 to 55%. Hybrid 30 (10 × 1) exhibited mpH for all carotenoids and had provitamin A content of 5.3 μg/g, but was poor in its agronomic traits. Mid-parent heterosis reached up to 77% for zeaxanthin and 114% for lutein. Hybrids 22 (10 × 3) and 23 (11 × 3) displayed a significant positive high parent heterosis for both β-cryptoxanthin and β-carotene, ranging from 30 to 56% with mean provitamin A concentrations of 5.27 and 6.0 μg/g, respectively. Hybrid 12 (4 × 6), which showed favorable SCA effects for the most important agronomic traits GY, EA, and PA, expressed high parent heterosis of 32% for provitamin A (6.06 μg/g) and 37% for total carotenoid (25.58 μg/g). This hybrid was comparable to the best-performing hybrid in terms of its agronomic traits, having 4.9 t ha-1 for GY, and 2.6 for EA and PA. Hybrid 38 (6 × 15) also showed significant positive high parent heterosis for provitamin A (21%), high provitamin A concentration (5.32 μg/g), and good agronomic characters, 5.2 t/ha for GY, 2.3 for EA, and 2.5 for PA.

Some of the F1 hybrids displayed heterosis for multiple carotenoids. For example, F1 hybrid 30 showed a significant mid-parent heterosis for all carotenoids, while F1 hybrids 15, 57 and 58 showed significantly positive mid-parent heterosis for all β-branch carotenoids. Most F1 hybrids that displayed significantly positive heterosis for carotenoids in the β-branch did not show significantly positive heterosis for lutein. Similarly, most of the F1 hybrids that displayed significantly positive heterosis for at least one provitamin A carotenoid did not show significant heterosis for the non-provitamin A carotenoids zeaxanthin and lutein.