Genetic Dissection Of The General Combining Ability Of Yield-Related Traits And Genomic Selection In Maize

Maize(Zea mays ssp.mays L.)is not only an important food crop in the world,but also an important source of feed and bioenergy raw materials.The planting area of maize in China has exceeded 40,000,000 hectares,which has became the most important cereal crop of China.With the increasing demand for maize,how to further improve maize yield is a major challenge for breeders.Maize is also one of the first crop to use heterosis in the world,and the use of heterosis has made great contributions to world food security.The key to the successful utilization of heterosis lies in the breeding of parents with high yield potential,but the actual breeding program requires a large number of field trials by breeders.The phenotype of hybrid cannot be predicted by their parental performance,but largely depends on the potential of their parents to generate excellent offspring.The concept of combining ability(CA)provides a quantitative index,which also greatly accelerates the research and utilization of heterosis.In the present,CA has been widely used to guide maize breeding practice,but how to accurately estimate the CA effects based on breeding practice is still a key issue that needs to be solved in maize and other crop breeding.In addition,little information is currently known about the genetic basis of CA.Currently,there is also little information on the optimal selection gain of different statistical methods in maize genomic selection breeding.This study mainly includes the following four aspects:1 The present study used 266 inbred lines with abundant genetic variation,which were also commonly used in maize breeding practice.Genotyping-by-Sequencing(GBS)was applied for genotyping,and 199,000 SNP markers were obtained for analysis.The present study used 61,468 SNP high-quality SNP markers to simulate the genotype and phenotype values of all possible hybrids(266*265*0.5=35,245),and investigated the effects of different statistical methods,sample sizes of different hybrids,and the distribution of parental inbred lines involved in hybrids on the accuracy of GCA prediction.The results showed that GP-Ⅰ and GP-Ⅱ methods were significantly superior to OLS method,indicating that the genome information on SPDC design can help improve the accuracy of GCA prediction for inbred lines.In addition,the GP-Ⅱ method performed better than GP-I and OLS for predicting some traits with low heritability and small sample population.2 In the present study,using 108,541 SNPs,266 inbred lines were divided into five heterosis groups based on population structure analysis,NJ cluster analysis and pedigree information.A total of 945 hybrids were crossed with the 266 inbred lines by SPDC design.Each inbred line participated in at least one cross and a maximum of 24 crosses,with an average of 7.1 crosses per inbred line.Phenotypic values of yield related traits,including ear weight(EW),ear grain weight(EGW),ear row number(ERN),kernel number per row(KNR),ear diameter(ED),ear length(EL),cob diameter(CD),plant height(PH),and ear height(EH)were collected from 945 hybrids and 266 inbred lines in two environments.For EW and EGW,Lvdahonggu×P,P×Reid showed the highest mean and median,and these two patterns also showed higher PH,KNR,ERN,EL,ED,and relatively lower EH and CD,indicating that these two patterns are the optimal heterotic patterns and can be further utilized in maize breeding practice.The GCA values of nine yield related traits for the 266 inbred lines were accurately predicted using genomic information.The GCA values of each trait showed normal distribution,and showed a large range of variation.Variance analysis showed that there were significant differences in GCA among different heterotic groups.For the P group,EW,EGW,ERN,KNR,and PH showed the highest GCA values,while ED and CD had the lowest GCA values,indicating that the P group has great breeding potential in terms of yield and ear type.However,there is also a risk of increasing plant height and reducing ear diameter in this group.In order to investigate the predictability of GCA,GBLUP,LASSO and BayesB were used for predicting GCA of the nine traits.The results showed that GBLUP had a prediction accuracy(r2)of 0.433 to 0.694 for different traits of GCA,with GCACD,GCAED,GCAERN,GCAEGW,and GCAEW having higher prediction accuracy(>0.5);The prediction accuracy of LASSO were 0.273～0.606,with GCACD,GCAED,and GCAEGW having higher prediction accuracy(>0.5);The prediction accuracy of BayesB were 0.415～0.684,with higher prediction accuracy(>0.5)for GCACD,GCAED,GCAERN,GCAEGW,and GCAEW.GBLUP and BayesB have relatively high prediction accuracy for various traits,but the difference between the two methods was not significant,while LASSO had significantly lower prediction accuracy.3 Correlation analysis showed that except for KNR and ERN,ED and KNR,EL and ERN,CD and EL,EH and EL,all other traits had significant or extremely significant correlations;Except for GCAED and GCAEL,GCAEL and GCACD,GCACD and GCAPH,and GCACD and GCAEH,the GCA of all other traits had significant or extremely significant correlations;Each trait had a highly significant positive correlation with its corresponding GCA,with correlation coefficients ranging from 0.41 to 0.78.Among them,the correlation coefficient between ERN and its corresponding GCA was the highest(0.78),and the correlation coefficient between KNR and its corresponding GCA was the lowest(0.41).Genome-wide association study was conducted on the yield related traits and their corresponding GCA using the five methods of Blink,FarmCPU,FASTmrMLM,FASTmrEMMA,ISIS EM-BLASSO.The results showed that there were 7,16,13,14,14,10,25,13,and 15 loci significantly associated with GCAEW,GCAEGW,GCAERN,GCAKNR,GCAED,GCAEL,GCACD,GCAPH,and GCAEH,respectively;there were 6,5,9,1,4,5,16,3,and 6 loci significantly associated with EW,EGW,ERN,KNR,ED,EL,CD,PH,and EH,respectively.With the maizeGDB and NCBI database,a total of 277 candidate genes were screened and functionally annotated based on the significant SNPs detected by at least three statistical methods.In order to further analyze the role of these loci in hybrid prediction,this study selected 187 inbred lines from 266 inbred lines to construct a new hybrid population.A total of 257 hybrids were successfully crossed to verify the prediction accuracy of marker sets on hybrids,which was composed of different types of significant loci.The results showed that the prediction accuracy of 9 yield related traits on hybrids by 9 marker sets showed similar tendency.The prediction accuracy of GCA significant marker sets(GCA SNP2 and GCA SNP1)for the 9 traits was significantly higher than other marker sets,while the prediction accuracy of Random Inbred SNP2 was the lowest.In addition,the analysis of the favorable genotypes of GCA significant loci for different traits in inbred lines and hybrids showed that for ED,EGW,EL,ERN,EW and KNR,the corresponding GCA value in inbred lines and the corresponding hybrid phenotypes showed an upward tendency as the accumulation of favorable genotypes.For CD,EH,and PH,with the accumulation of favorable genotypes,the GCA of corresponding traits in inbred lines and the corresponding phenotypes in hybrids show a downward tendency,indicating that the accumulation of favorable genotypes can effectively improve the GCA of ED,EGW,EL,ERN,EW and KNR and the phenotypes of corresponding hybrids.Simultaneously,the accumulation of favorable genotypes can also effectively reduce the GCA of CD,PH,EH,and the phenotype of corresponding hybrids.4 Based on the SPDC population composed of the 945 hybrids,we firstly evaluated the prediction accuracy of three genome-wide selection(GS)methods,BayesB,GBLUP and LASSO on the 9 traits using cross validation.The results showed that the difference in predictive ability(r2)between the different GS methods was small,while there was significant differences on predictive ability between different traits.For example,the predictive ability of PH and ERN reached above 0.6,while that of KNR was less than 0.3.Furthermore,this study used three GS methods,BayesB,GBLUP,and LASSO,to predict the ear weight(EW)of all possible 35245 potential hybrids from 266 parents,and selected 100 hybrids with the highest EW(top 100)and 100 hybrids with the lowest EW(bottom 100),respectively.Then,the selection gain brought by GS was evaluated based on the average increase ratio of top100 to bottom100.The top hybrids selected by BayesB,GBLUP,and LASSO in the field validation were 108.3%,105.4%,and 108.6%higher than the bottom hybrid,respectively,and 35.1%,34.5%,and 35.2%higher than the estimated mean of all potential hybrids,respectively.Furthermore,of the top hybrids verified in the field,21 hybrids had EW higher than the check variety,and 16 hybrids had EW higher than the check variety by more than 5%.In addition,the GCA of the inbred lines related to the top 100 hybrid selected in this study was mostly at a high level,while the GCA of the inbred lines related to the bottom 100 hybrids always ranks behind,which further confirms the correlation between the phenotype of the hybrid and their corresponding GCA.The top 100 hybrids involved about 50 inbred lines,including some parents of widely promoted varieties such as DH382,H2671,A489,and S181.Finally,analysis of the accumulation of favorable genotypes at significant loci associated with GCAEW in the top 100 and bottom 100 hybrids showed that the bottom hybrids selected by GS had little or no accumulation of favorable genotypes,while the top hybrids were able to extensively aggregate excellent alleles.For example,the four hybrids,A345/A351,A017/A037,A037/A169,and A345/A438 in top100 have all aggregated 6 favorable genotypes,of which A017/A037 and A037/A169,which had been registered by the Crop Cultivar Registration Committees,and named Suyu 161 and Tongyu 1701,respectively.