Information on pedigree, heterotic group, and ecological adaptation of the 19 inbred lines (female) and five testers (male) used in this experiment is given in Table 1 . The 19 inbred lines were selected on the basis of previously determined GY and genetic diversity. Testers consisted of four RILs, developed by Li et al. (2018), and Q11, a widely used elite maize line in China. In the 2017 winter season, the five testers were crossed with the 19 inbred lines (female parents) in a line × tester design at Dehong, Yunnan, China, to generate 95 crosses.

Four recombinant inbred lines (RILs) were developed by crossing a Suwan1 inbred YML32 with a widely used in China Reid inbred line ‘Q11’ (Li et al., 2018). YML32 is a tropical inbred line selected from the Suwan1 population by the Institute of Food Crops, Yunnan Academy of Agricultural Sciences, China. It was used as the male parent for developing gray leaf spot (GLS)-resistant hybrid Yunrui 1 (Liu et al., 2016). The GLS-susceptible line Q11 was derived from HL1999, which had a high combining ability for GY and was widely used in maize breeding programs in southwest China. A line × tester experiment was conducted with five testers (i.e., the four RILs and their recurrent parent line Q11, used as ♂) and 19 elite maize inbreds (used as ♀).

The 95 crosses, along with their parents and one check (Yunrui62), were planted in the summer of 2018 (Experiment 2018) at three locations in Yunnan (China): Dehong (24°26 ‘N, 98°35 ‘E, elevation: 914 masl), Kunming (25°23 ‘N, 102°9 ‘E, elevation: 1970 masl) and Yanshan (23°60 ‘N, 104°4 ‘E, elevation: 1570 masl). A randomized complete block design with three replications was employed at each location. The experimental plot consisted of a single 3.5-m-long row, with a plot-to-plot spacing of 0.70 m and plant-to-plant spacing of 0.25 m within a row. Four seeds per hill were sown. Upon germination, thinning was done, and two seedlings per hill were maintained. The resulting plant density was approximately 60,000 plants ha^-1. Trials were managed according to standard agronomical practices. The GY was determined at each location from 10 plants taken from the middle of each row and was expressed as tons ha^-1.

The following general linear model was used for the analysis of variance (ANOVA) for GY

where Y_ijkl = observed value from each experimental unit; μ = population mean; α_l = location effect; b(a)_kl = replication effect within each location; υ_ij = F₁ hybrid effect = l _i + t _j + lt _ij (where l _i = ith line effect; t _j = jth tester effect; lt _ij = interaction effect between ith line and jth tester); (αυ) _ijl = interaction effect between ijth F₁ hybrid and lth location; and e_ijkl = residual effect.

The locations were treated as a fixed effect. Since the lines and testers were not selected at random, significance for lines, testers, and lines × testers and their interactions with the locations was determined against the overall experimental error term. Combining ability analysis was conducted according to the method proposed by Fan et al. (2009). Mid-parent heterosis (MPH) was determined as follows:

MPH = [(F₁ – MP)/MP] × 100

where F₁ = Mean of hybrid, MP = Mid-parent value = (P1 + P2)/2; where P1 = Parent 1 and P2 = Parent 2.

BPH = [(F₁ – BP)/BP] × 100

where F₁ = Mean of hybrid, BP = Better-parent value.

Fisher’s least significant difference (LSD) was used to separate MPH and GY means of testcrosses. Data analysis was conducted via the SAS 9.1.3 software package (SAS Institute, 2005).

Identity by state (IBS) distance matrix was applied to explain the genetic kinship between individuals within a group calculated by PLINK v.107. software (Purcell et al., 2007). The genetic distance (GD) between two parents of each hybrid was calculated as GD = 1- IBS (Geng et al., 2021).

Raw sequences with a 150-bp read length for the 24 parental lines were generated via the Illumina Novaseq platform (Illumina, San Diego, CA, USA). After quality checking and filtering, the high-quality paired-end reads were mapped to the B73 reference genome (Jiao et al., 2017) using Burrows-Wheeler Aligner software (Li and Durbin, 2009), with the parameters of ‘mem -t 4 -k 32 -M’. SNPs were then called using the Genome Analysis Toolkit software (McKenna et al., 2010) based on the B73 reference genome (Jiao et al., 2017). Finally, high-quality SNPs were filtered with the following parameters: depth for each individual ≥ 3, base quality ≥30, genotype quality for each individual ≥ 5, with a missing rate of ≤ 0.2. The identified SNPs were further annotated using the ANNOVAR software tool (v2013-05-20) (Wang et al., 2010).

Based on the ANNOVAR annotation, the SNPs in the promoters (1 kb upstream of the start codon), exons, untranslated regions, and stop codons were selected for analysis. These SNPs were designated as PEUS SNPs. If PEUS SNPs were the same at a point or locus, they were considered homozygous SNPs; if the PEUS SNPs were different, they were considered heterozygous SNPs. Then homozygous SNPs and heterozygous SNPs were counted in the 95 crosses and mean homozygous PEUS SNPs and mean heterozygous PEUS SNPs were calculated for the five testers and 19 lines.

Analysis of variance (ANOVA) results for GY is given in Table 2 . The mean squares for lines, testers, lines × testers, and locations for GY were statistically significant. These results implied that the differences for the GY should be large enough for the analysis of MPH and BPH.

To predict the MPH_GY and BPH_GY, we sequenced 24 maize parental lines from three different heterotic groups ( Figure 1A ) with a sequence depth of ~5×. A total of 58,986,791 high-confidence SNPs was retained after filtration (see Methods section). The numbers of the PEUS SNPs in the five testers and 19 lines are shown in Figure 1A . The results from Figure 1A showed that the number of the PEUS SNP were different in the five testers and among the 19 lines. This result was highly expected since the 24 inbreds belonged to three different heterotic groups ( Figure 1B ). The five testers and three lines from Reid heterotic group had more PEUS SNPs than those in the lines from both nonReid and Suwan1 heterotic groups ( Figure 1A ).

The number of reads for each of the 24 inbred lines is summarized in Supplemental Table 2 . The homozygous and heterozygous PEUS SNP are given in Figure 1C . Total numbers of reads from promoters, exons, untranslated regions (UTRs), and stop codons were accounted for (see Figure 1D ). The results showed that the promoter had the highest number of PEUS SNP. This result matched with the findings of Li et al. (2022), who found that the promoters and the regions of 10–50 kb distance from the gene bodies had been changed and accumulated more favorable SNPs for maize inbred lines during the history of maize breeding. We notice that the promoter has the highest numbers of PEUS SNP ( Figure 1D ) that may lead to high MPH_GY and BPH_GY ( Figures 2 , 3 ). Previous studies showed that 1) A multiple promoters may be involved to regulate the expression of one gene (Kühn et al., 2005); 2) A promoter may control the expression of several genes (Yang et al., 2008); 3) Individual exons in a same gene may be regulated by different promoters (Carol et al., 2000). The results from all these studies had suggested promoters are highly diversified. We can hypothesis that rich variations or polymorphisms are required by regulating different exon or different gene expressions in all creatures, especially in plant and human genomes.

Correlation between mean number of heterozygous PEUS SNPs or GD and mean MPH, BPH of GY from the 95 crosses averaged by the five testers and 19 lines at all locations

GY for all 95 crosses at each of the three locations, GD and mean GY across three locations are given in Supplemental Table 1 . The mean number of heterozygous PEUS SNPs, mean MPH_GY, and mean BPH_GY for the five testers are depicted in Figures 2A , B . The results showed that 1) the four RILs had more heterozygous PEUS SNPs and higher mean MPH_GY and higher mean BPH_GY than their recurrent parent (Q11). These results suggested that introgression of genes from the tropical donor parent YML32 (Zhang et al., 2012) increased the diversity of genetic background of the original recurrent parental line Q11, which likely increased the mean MPH_GY and BPH_GY of the crosses involving the four RILs; 2) MPH_GY and BPH_GY seem to be related to the mean number of heterozygous PEUS SNPs of the five testers in crosses with the 19 lines. A statistical analysis further showed that MPH_GY and the mean number of heterozygous PEUS SNPs of the five testers were significantly correlated (r=0.942, p=0.017). Similar result was found between BPH_GY and heterozygous PEUS SNPs (r=0.881, p=0.048). A regression plot ( Figure 2C ) demonstrated high determination coefficients for both MPH_GY (0.89) and BPH_GY (0.78) when plotted against the mean number of heterozygous PEUS SNPs. These results suggested that the number of heterozygous PEUS SNPs could be used for predicting the mean MPH_GY and BPH_GY of hybrids from their inbred parental lines. With the rapid development of DNA sequencing technology and due to its cost-effectiveness, breeders can select potential inbreds with high heterozygous PEUS SNPs to make the crosses. Then there would be a better chance of developing hybrids with high capacity than using methods without PEUS SNP information. Thus, in practice, breeders may select good inbreds with PEUS SNPs before making actual crosses in the field, and this should help improve breeding efficiency to a great extent.

Similar analyses for the mean number of heterozygous PEUS SNPs, mean MPH_GY, and mean BPH_GY were conducted for the 95 crosses grouped by lines and three heterotic groups ( Figures 3A, B ). The Figure revealed that the high mean number of heterozygous PEUS SNPs in a line usually resulted in high mean MPH_GY and BPH_GY for a line in crosses with the testers, with a few exceptions. The trend was more obvious for the lines in the nonReid heterotic group than those in the other two heterotic groups. The difference might have happened since there were only three lines each in the Suwan1 and Reid heterotic groups. The correlations between the mean number of heterozygous PEUS_SNPs and MPH_GY and BPH_GY were calculated for the remaining 13 lines from the nonReid heterotic group. The results showed that the correlation between the mean number of heterozygous PEUS SNPs and MPH_GY was statistically significant (r=0.719, p=0.006). Similarly, the correlation between the mean number of heterozygous PEUS SNPs and BPH_GY was also found to be statistically significant (r=0.611, p=0.027). The regression plot for the mean MPH_GY and BPH_GY by the mean number of PEUS SNP for the 13 lines was constructed ( Figure 3C ). The determination coefficients for both mean MPH_GY (0.51) and BPH_GY (0.37) were smaller than those of the five testers. The possible reason for this is that the five testers were crossed with the 19 lines that belonged to three heterotic groups and were genetically highly diversified. On the contrary, the 13 lines had been crossed only with the lines from the five testers, and they all belonged to the Reid heterotic group and were less diverse. Thus, to effectively explore its potential of improving the GY of maize, the maize inbred with a higher number of heterozygous PEUS SNPs should be crossed with the lines with diverse genetic backgrounds or possibly from different heterotic groups.

Since GD is widely used for the prediction of heterosis, we calculated the correlation between GD and MPH_GY/BPH_GY in 95 crosses. Then we compared the correlation coefficient between the number of heterozygous PEUS SNP and MPH_GY/BPH_GY from the same 95 crosses at the three locations, Dehong (DH), Kunming (KM) and Yanshan (YS) ( Figure 4 ). It showed that all the correlation coefficients were highly significant except GD at the KM location. The results revealed that the correlation coefficients were higher between the number of heterozygous PEUS_SNP and MPH_GY/BPH_GY than those between GD and MPH_GY/BPH_GY in all three individual locations. A t-test showed that the correlation coefficient of the number of heterozygous PEUS SNP was significantly higher than that of GD (P=0.03). The likely reason underlining the difference is that GD was calculated based on all genetic differences between two parents. In comparison, the number of heterozygous PEUS SNP was calculated from selected genomic regions between two parents (see Methods). Recently, Li et al. (2022) assembled 1,604 historically utilized maize inbred lines belonging to various female heterotic groups and male heterotic groups. They found that the frequency of favorable alleles at the associated SNPs exhibited convergent increases and accumulated in both the female and male heterotic groups for the convergently improved traits such as grain yield per plant (GYPP) and yield-related traits during modern maize breeding. Since the selected heterozygous PEUS SNPs are directly related to gene expressions (i.e., exons, promoters, stop codons), the chances of these SNPs likely covering most of the accumulated favorable alleles/genes are higher. As described by Li et al. (2022), this could have determined the GY of maize, whereas some other markers present in different genomic regions covered by GD were not included in the heterozygous PEUS SNPs in an inbred line. Therefore, our results strongly suggested that the number of heterozygous PEUS SNPs would be better for heterosis prediction than GD.

We selected the top 10 crosses with high GY from our experiment and found that 7 out of the 10 crosses involved RIL R2_2 as the male parent, and 4 out of 10 crosses involved CML395 or CML 373 as the female parent. As expected, the RIL R2_2 had the highest mean number of heterozygous PEUS SNPs among the five testers ( Figure 2 ), whereas CML395 and CML373 had the highest mean number of heterozygous PEUS SNPs among the 19 lines. These results showed that the maize breeder should be able to confidently select inbreds with an increased number of heterozygous PEUS SNPs to develop hybrids with the possibility of getting higher GY. With SNP technology being available worldwide, breeders can use high-throughput DNA sequencing technologies to obtain SNP information from each line and calculate the number of heterozygous PEUS SNPs for each inbred line. Then, the mean number of heterozygous PEUS SNP for all individual inbred lines could be computed and finally, the lines with a higher mean number of heterozygous PEUS SNP could be selected to make the crosses. By using PEUS SNP information, breeders would need to select fewer inbreds to make the required crosses to reach their breeding goals aimed at developing hybrids with higher GY. Thus, the breeding efficiency can be substantially improved. So, the results of the present investigation should have possible applications in maize breeding programs.