Genome-wide prediction for hybrids between parents withdistinguished difference on exotic introgressions in Brassica napus

Dndn Hu,Yusheng Zho,Jinxiong Shen,Xingxing He,Yiki Zhng,Yong Jing,Rod Snowdon,Jinling Meng,Johen C.Reif,*,Jun Zou,*

a National Key Laboratory of Crop Genetic Improvement,College of Plant Science &Technology,Huazhong Agricultural University,Wuhan 430070,Hubei,China

b Department of Breeding Research,Leibniz Institute of Plant Genetics and Crop Plant Research (IPK),Stadt Seeland 06466,Germany

c Department of Plant Breeding,IFZ Research Centre for Biosystems,Land Use and Nutrition,Justus Liebig University,Giessen 35392,Germany

Keywords:Hybrid Genome-wide prediction Exotic introgression Brassica napus Heterosis

ABSTRACT Extensive exotic introgression could significantly enlarge the genetic distance of hybrid parental populations to promote strong heterosis.The goal of this study was to investigate whether genome-wide prediction can support pre-breeding in populations with exotic introgressions.We evaluated seed yield,seed yield related traits and seed quality traits of 363 hybrids of Brassica napus (AACC) derived from two parental populations divergent on massive exotic introgression of related species in three environments.The hybrids presented strong heterosis on seed yield,which was much higher than other investigated traits.Five genomic best linear unbiased prediction models considering the exotic introgression and different marker effects(additive,dominance,and epistatic effects)were constructed to test the prediction ability for different traits of the hybrids.The analysis showed that the trait complexity,exotic introgression,genetic relationship between the training set and testing set,training set size,and environments affected the prediction ability.The models with best prediction ability for different traits varied.However,relatively high prediction ability (e.g.,0.728 for seed yield) was also observed when the simplest models were used,excluding the effects of the special exotic introgression and epistasis effect by 5-fold cross validation,which would simplify the prediction for the trait with complex architecture for hybrids with exotic introgression.The results provide novel insights and strategies for genome-wide prediction of hybrids between genetically distinct parent groups with exotic introgressions.

1.Introduction

Crop production should increase by 70% to 110% from the beginning to the middle of this century to meet the rising global demand [1,2].The systematic exploitation of heterosis in hybrid breeding has been one of the main drivers of yield increases in major crops in recent decades [3–9].Accumulated evidences showed that pronounced heterosis and increased hybrid yield could be achieved by crossing parents from genetically distinct populations[10–13].Significant genetic differences can be a result of intraspecies [14],inter-subspecific [15],or inter-subgenomic diversity [16–18].The latter approach is difficult to harness because exotic introgression contains not only favorable alleles but also unfavorable alleles,especially due to linkage drag [19–21].It is very challenging to identify the favorable and unfavorable alleles from exotic introgression one by one,especially for complex traits such as grain yield,which is influenced by thousands of genes and complex gene network [22,23].The method of genome-wide prediction [24] has the potential to provide a solution or assist to improve the efficiency of exploiting exotic introgression in breeding.

Rapeseed is a global oilseed crop that ranks 7th in the world with a 2.23% share of calorie production within 100 nations [2]and provides 13%–16% of the world’s vegetable oil production[25].Brassica napus(AnAnCnCn) is the major cultivated rapeseed species that has been developed during the past 7500 years [26].About 60% of the worldwide cultivatedB.napuswere hybrid varieties,but the extreme genetic bottlenecks caused by strict selection for certain seed quality traits (low erucic acid and low glucosinolate) and flowering and vernalization traits limited the diversity for breeding with genetically distinct heterotic groups.In order to broaden the genetic diversity ofB.napus,a considerable number of resynthesizedB.napuslines based onB.rapa(ArAr),B.oleracea(CoCo),B.carinata(BcBcCcCc) andB.juncea(AjAjBjBj) [27–31] were created.These resynthesizedB.napusshowed rich variations,large genetic difference to traditionalB.napus,and a strong heterosis potential in hybrid breeding [17,32–36].

In our previous studies,we created a gene pool of new-typeB.napusby interspecific crosses introgressing the genomic components of the related oilseedBrassicaspecies of 122B.rapa(ArAr)and 74B.carinata(BcBcCcCc) cultivars followed by more than 10 years of artificial selection for important traits.This gene pool contained thousands of diverse lines with reorganized AC genome and rich genetic variations [35–38],generated new lines with favorable traits for selection[39–41],and showed a distinct genetic difference from the traditionalB.napuspopulation.Most importantly,this gene pool could be used as a novel heterotic group that benefits from inter-subgenomic heterosis in crosses with the traditionalB.napuslines [17,18].

In this study,we aimed to 1) evaluate the potential of intersubgenomic hybrids derived from new-typeB.napusand the traditionalB.napusfrom different ecotype and genetic groups,2)explore the feasibility of constructing genome-wide prediction models for inter-subgenomic hybrids,and 3) evaluate the impacts of exotic introgression and genetic relationship on prediction ability.

2.Materials and methods

2.1.Plant materials and field trials

Six traditionalB.napuslines,including the two semi-winter type recessive genic male sterile lines‘‘ZH3A”and‘‘MSL72A”(supplied by Zhejiang Academy of Agricultural Sciences and referred to as Tester1,Tester2),one semi-winter type cultivar ‘‘Ningyou 7”(bred by Jiangsu Academy of Agricultural Sciences,coded as Tester3)[42],two winter typeB.napuslines‘‘Express 617”(an inbred line of German oilseed rape cultivar,coded as Tester4 [43]) and‘‘Tapidor”(a French cultivar,coded as Tester5[44]),and one spring typeB.napusline (Tester6) were selected as female parents.61 elite new-typeB.napusinbred and double haploid male parental lines were selected from a gene pool whose genome was replaced with the subgenome from 122 accessions ofB.rapa(ArAr) and 74 accessions ofB.carinata(BcBcCcCc) [35,37].A total of 363 intersubgenomic hybrids were obtained by crossing the 6 traditionalB.napuslines with the 61 new-typeB.napuselite lines.

Parental lines and hybrids were evaluated with local commercial hybrids at two semi-winter environments (Wuhan ‘‘WH” and Xiangyang ‘‘XY”) and one spring environment (Jingtai ‘‘JT”) in China (Table S1).Considering that the hybrids with winter type female parent may not flower in the spring type environment with an average temperature of 21 °C during the growing season,these hybrids were only planted in semi-winter type environments.The experimental design was a randomized complete block design with 3 replications.Every plot comprised three or five rows with distance of 30 cm between rows and 15 cm between individuals.The agronomic traits assessed included seed yield per plot (SY,determined as Mg ha-1),flowering time (FT,the time when half of the plot were flowering),seed number per pod (SN),and thousand seeds weight(TSW).Six seed quality traits,including glucosinolate content (GLU),oil content (OC),erucic acid content (EAC),oleic acid content (OLE),linoleic acid content (LEI),and linolenic acid content (LEN) were measured using a Foss NIR Systems 5000 with near-infrared spectroscopy (NIR) as previously described [45].We also recorded the degree of susceptibility against biotic stresses and logging of every plot as a reference when dealing with outlier.

2.2.Phenotypic data analyses

We detected outliers using the method described by Anscombe and Tukey [46].The remaining data was used for estimating first and second-degree statistics.For single environments,we used following linear mixed model to obtain the best linear unbiased estimations (BLUEs) and variance component:

WhereYis the vector of phenotypic value,Gis the effect of genotype,Ris the effect of replication,andeis the residual.For estimating BLUEs,we set the effect of genotype as fixed,while for estimating variance components we treated it as random effect.The above model was solved by the restricted maximum likelihood method using the ASREML-R software[47].Repeatability was estimated asR=,where r refers number of replications,refers to genotypic variance,andrefers to residual variance.

Following linear mixed model was applied to estimate the BLUEs across environments:

whereYis the vector of phenotypic value in each environment,Gis the effect of genotype,Eis the effect of environment,G×Eis the genotype-by-environment interaction effect,R×Eis the effect of replication in each environment,andeis the residual.In this model only genotype effect” was treated as fixed,and the remaining effects were treated as random effects.The total variance of hybrids was further decomposed into the variance due to general combining ability effects (GCA) of males and females,and the variance due to specific combining ability (SCA) effects of crosses.

The model used to estimate the variance components was:

whereYis the vector of phenotypic value in each environment,His the fixed effect of hybrids and parents,Pis effect of parent lines,GCAFemale,GCAMaleand SCA are the general and specific combining ability effects of hybrids,respectively.GCAFemale×Edenote the interaction between GCAFemaleand environment,GCAMale×Edenote the interaction between GCAMaleand environment,and the rest interaction effect are present in the same way.

Broad-sense heritability of hybrids was calculated as the ratio of genotypic to phenotypic variance:

Commercial heterosis was calculated for each hybrid as CH=F1-Check,where F1denotes the seed yield performance of the hybrid and check refers to the average seed yield of the two commercial checks closed to the hybrid.

2.3.Genotypic data analysis

67 parental lines were genotyped by sequencing using the double-digested restriction-site associated DNA(ddRAD)sequencing method [48] but with paired-end reads of 150 bp.Genotype calling of the 67 parental lines was performed based on the reference genomes ofB.napus(AnAnCnCn) ‘Darmor-bzh’ (version 4.1)[26],B.rapav2.5(ArAr)[49],B.oleraceav2.1(CoCo)[50],andB.nigra(BnBn) [51] to detect more species-specific markers.The details of single nucleotide polymorphisms (SNP) calling have been presented previously [37].Clean reads were mapped to the reference genome with Bowtie2 v2.3.4 (https://sourceforge.net/projects/bowtie-bio/files/bowtie2/2.3.4/) using the default parameters.SAMtools v1.8 (https://sourceforge.net/projects/samtools/files/samtools/1.8/) was used to convert and sort mapping results into the BAM format.Duplicate reads were filtered using the Picard v2.12.1 package (https://github.com/broadinstitute/picard/tree/2.12.1),and the indexing for the BAM file was constructed using SAMtools v1.8.SNP detection was performed using GATK v4.0.3.0(https://github.com/broadinstitute/gatk/releases/tag/4.0.3.0),and the details were as follows:1) Using the HaplotypeCaller package runs per-sample to generate an intermediate GVCF file;2) Using the CombineGVCFs package to combine per-sample GVCF files into a multi-sample GVCF file;3)Using the GenotypeGVCFs package to perform joint genotyping on more samples pre-called using HaplotypeCaller,and 4)Using the VariantFiltration package to filter variant calls based on INFO and/or FORMAT annotations.The markers detected based on the traditionalB.napusgenome were coded asB.napusmarkers (SNPT),the markers detected based onB.rapa,B.oleraceaandB.nigrawere coded as ABC markers,then we aligned the ABC markers to theB.napusmarkers using a BLAST alignment,and ABC markers with no BLAST alignments were coded as speciesspecific exotic introgression markers (SNPS).A total of 48,602 SNP markers were obtained(Fig.S1),43,106 of them were denoted asB.napusmarkers (SNPT),and 5496 of them were denoted as species specific markers (SNPS) that were detected only in the mapping against the exotic introgressions of the A,B,and C genomes.All 48,602 markers were used to calculate the Nei’s genetic distance[52] between the parental lines,and the genetic distance matrix was used to obtain an unrooted phylogeny reconstruction in the Neighbor-Joining method using PowerMarker V3.25 with the obtained tree visualized via MEGA 6.0 [53].

2.4.Five genome-wide prediction models

The SNP data was coded as following:the homozygous ‘‘AA”class as-1,the homozygous‘‘BB”class as 1,and the heterozygous‘‘AB” class as 0.The genomic data of 363 hybrids were deduced from their parents.We considered different genome-wide prediction scenarios including SNPS,SNPTand a combination of SNPSand SNPT.The first two genome-wide best linear unbiased prediction(GBLUP)models were based on additive and dominance effect for markers SNPT(aTanddT) and SNPS(aSanddS):

whereYrefers to the BLUEs of the genotypes.μ refers to the mean of the population.The genetic effects and residualefollow normal distributionsaT～dS～N(0,DS),ande～N(0,).ATand ASdenote the additive relationship matrices for SNPTand SNPSmarkers,DTand DSdenote the dominance relationship matrices for SNPTand SNPSmarkers,respectively.These numerator relationship matrices were calculated according to VanRaden[54]and Zhao[55].

In order to test whether the marker effects of different marker groups have different contribution to the prediction ability,we also tested two GBLUP models assuming common or special distributions from different groups of markers:

Y=μ +a+d+e(Model3,M3)

Y=μ +aT+aS+dT+dS+e(Model 4,M4)

For model M3,aanddare assumed to follow normal distributionsa～with and A,D referring to the numerator relationship matrices calculated from the combined marker data of SNPTand SNPS.For model M4,aanddeffects of SNPTand SNPSare assumed to follow normal distributionsas defined above.In addition,we implemented the following model considering additive,dominance and also epistatic effects:

whereaa,adandddare additive×additive,additive×dominance,and dominance × dominance genetic effects,respectively.These epistatic effects are assumed to follow normal distributions:aa～andaa～,A and D,are the numerator relationship matrices used in model 3,and ‘‘*” denotes the Hadamard product of matrices.

All of the above GBLUP models were implemented using the R package BGLR (https://cran.r-project.org/web/packages/BGLR/index.html).

2.5.Chess-board-like and 5-fold cross validations

First,we used the chess-board-like cross validation to test the prediction ability of all five above outlined prediction models.The training set consists of 4 out of 6 female and 40 out of 61 male parental lines as well as 100 hybrids derived from them.The remaining hybrids were divided into 3 different test set TS2 (Test set 2),TS1 (Test set 1),and TS0 (Test set 0),which share two,one,or no parental lines with the parental lines of the training set,respectively (Fig.S2).

Previous studies have shown that the prediction ability increases substantially when the size of training set increases from 100 to 300 [56,57].Therefore,we evaluated the prediction ability of the five models using 5-fold cross validation.The training set for 5-fold cross validation has 357 genotypes,including all 67 parental lines and 290 random sampled hybrids(80%of hybrids),and the test set is the remained 73 hybrids(20%of hybrids).The above procedure was repeated 100 times,all five models were tested in each round.The prediction ability for each test set was estimated as the Pearson correlation coefficient between the predicted and the observed hybrid performance was standardized with the square root of the heritability.

3.Results

3.1.The genetic relationships between the parental lines

Our study is based on 61 elite semi-winterB.napusinbred lines(N0–N61) that been selected from a new-typeB.napusgene pool with massive exotic introgression from 122 accessions ofB.rapa(ArAr)and 74 accessions ofB.carinata(BcBcCcCc)[35,37].Moreover,we used six eliteB.napuslines without massive exotic introgressions as female parents.These represent three sub-populations ofB.napus:the group of semi-winter(Tester1–Tester3),winter (Tester4,Tester5),and spring type(Tester6).Genotyping by sequencing of the 67 parental lines yielded 48,602 SNP markers,including 43,106 SNPTmarkers and 5496 SNPSmarkers that covered the entire A and C genomes.The density of SNPTmarkers was almost 9 times higher than that of the SNPSmarkers (Fig.S1).The traditionalB.napusfemale parental lines were genetically different from the new-typeB.napusmale parental lines(Fig.1a).The average genetic distance between male and female lines was 0.368 when using the SNPTmarkers.The distance was higher (0.395)when the marker set SNPSwas also taken into account.This means that ignoring specific markers can lead to an underestimation of the genetic distance between the new-type and traditionalB.napuslines.The average genetic distance between the three ecotypes of female parent lines and new-typeB.napusmale parent lines showed the following trend:winter type lines (Tester4,Tester5)(genetic distance of 0.401)>spring type line(Tester6)(genetic distance of 0.361)>semi-winter type lines(Tester3,Tester2,Tester1)(genetic distance of 0.349) (Fig.1b).This trend is in accordance with the adaptation of theB.napusmale parental lines,which were bred for semi-winter environments.

Fig.1.The genetic relationship of the parents.(a)Neighbour-Joining tree of the 67 parental lines.Blue line means the new-type B.napus male parental lines.(b)The genetic distance between female parental line and male parental lines.

3.2.High quality of the phenotypic data,heritability and distribution of the investigated traits

The 363 inter-subgenomic hybrids and their 67 parents were examined for 10 important agronomic and quality traits in three environments(Table S1).Within the three environments,repeatability was high for the 10 traits and ranged from 0.60 to 0.99(Fig.S3).For the analyses across the three environments,we observed moderate to high estimates of heritability for the 10 traits,ranging from 0.38 to 0.99 (Table 1).BLUEs for the 10 traits showed continuous variation in most cases (Fig.2;Table S2) with exceptions observed for the traits FT,EAC,GLU,and OLE.These traits showed a discontinuous distribution,most likely caused by the use of testers covering three rapeseed ecotypes with marked differences in FT,EAC,GLU,and OLE.This is further reflected by pronounced variance due to general combining ability effects of the female pool for these traits (Table 1).

3.3.Inter-subgenomic hybrid advantages for different traits

For the 363 inter-subgenomic hybrids,the midparent heterosis,i.e.the difference between the hybrids and the mean performance of their parents,varied among the different traits (Table 2;Fig.2).For instance,a pronounced average midparent heterosis of 70.7%was observed for seed yield.For the other nine traits,the average midparent parent heterosis ranged from -4.4% for LEI to 7.7% for SN.The better-parent heterosis,i.e.the difference between the hybrids and the better parent,also varied among the differenttraits(Fig.3;Table S3).The average better-parent heterosis ranged from-20.49%for EAC to 37.65%for SY.For seed yield,the commercial heterosis,i.e.hybrid performance in contrast to the commercial hybrid variety,averaged -12.9%,while the highest commercial heterosis reached 71%,and 15 hybrids showed a significant (P<0.05) positive commercial heterosis (Table 3).In the individual environments the number of hybrids that showed a significant(P<0.05)positive commercial heterosis for seed yield with a range from 1 hybrid for the JT environment to 10 hybrids for the WH environment.This indicates that the new-typeB.napushas potential in rapeseed hybrid breeding across a broad range of environments.

Table 1 Variances and heritability of 10 traits evaluated across three environments.

Table 2 Summary statistics for midparent heterosis with best linear unbiased estimations cross environments.

Fig.2.Distribution of best linear unbiased estimations across environments of the 363 hybrids and their 67 parental lines evaluated for 10 traits.SY,seed yield;FT,flowering time;SN,seed number;TSW,thousand seeds weight;GLU,glucosinolate content;EAC,erucic acid content;OC,oil content;OLE,oleic acid content;LEI,linoleic acid content;LEN,linolenic acid content.

3.4.The genetic relationship between training and test set affected the prediction ability

Five GBLUP models,which differed with respect to the modelling of marker effects and the genetic effects considered,were examined for their prediction ability of the hybrid performance.The chess-board-like cross validation (Fig.S2) showed that the genetic relationship between training and test set dramatically affected the prediction abilities (Table 4).The test set TS2,which includes hybrids sharing both parental lines with the training set,showed the highest prediction ability of at least 0.561 for OC and at most 0.768 for FT.Test set TS1,which includes hybrids that share one parental line with the training set,showed a slightly lower prediction ability ranging from 0.377 for SY to 0.770 for FT.The most distant test set TS0,which does not share parental lines with the training set,showed a very low prediction ability ranging from -0.250 for EAC to 0.583 for FT.

Fig.3.Better-parent heterosis of 363 hybrids for 10 traits with best linear unbiased estimations cross environments.SY,seed yield;FT,flowering time;SN,seed number;TSW,thousand seeds weight;GLU,glucosinolate content;EAC,erucic acid content;OC,oil content;OLE,oleic acid content;LEI,linoleic acid content;LEN,linolenic acid content.

Table 3 Estimates of commercial heterosis on seed yield in three environments.

Table 4 Prediction abilities of 10 traits by five genome-wide prediction models with Chess-board-like cross validation.

3.5.Specific SNPS markers and epistatic effects influence prediction ability

The model M2,which considered specific SNPSmarkers,showed a higher prediction ability for the five seed quality traits GLU,EAC,OLE,LEI,and LEN but a lower prediction ability for the other five traits SY,FT,SN,TSW,and SN compared to model M1,which considered theB.napusSNPTmarkers only(Table 4).For the traits SN,TSW,GLU,and EAC,the addition of specific SNPSmarkers (model M3 and M4) slightly increased the prediction ability compared to the model M1 that only consideredB.napusSNPTmarkers with test set TS1 and TS2.We also compared model M4,which assumed that the marker effects for SNPSand SNPTwere drawn from one distribution,with model M3,which assumed that the marker effects for SNPSand SNPTwere drawn from two distributions.We found that for most traits (such as FT,SN,GLU,E AC,and OC),the prediction ability of the M4 model is slightly higher than that of the M3 model,while for SY and LEI is lower.This suggests that effects ofSNPSand SNPTshould be treated individually.We extended the M3 model for epistatic effects (model M5),and found an increase in the prediction ability for most traits except for SY (Table 4).

Table 5 Prediction abilities of 10 traits within each experiment with 5-fold cross validation.

3.6.Genome-wide prediction ability of the hybrid performance for inter-subgenomic hybrids

Chess-board-like cross validation showed only moderate prediction abilities for the closely related data sets for complex traits such as seed yield.This is most likely due to the moderate size of the training set with 144 lines.Consequently,we also used 5-fold cross validation that allowed to test the prediction ability of a larger training set of 357 lines.We found that as the size of training set increases,the prediction ability increased,and the difference in the prediction abilities between different models and marker data became smaller (Table 4 (TS2),Table 5 (cross environment)).In addition,for most traits,the prediction ability in a single environment is similar or lower than the ability for BLUEs across the environments,except for the prediction ability of OC in experiment JT and the TSW in experiment XY (Table 4).Comparing the genome-wide prediction models under the two cross validation methods (Tables 3 and 4),revealed that the simplest model M1,based on 5-fold cross validation,B.napusmarkers‘‘SNPT”,additive effect and dominance effect,is comparable to that of other models for genome-wide prediction of inter-subgenomic hybrids.

4.Discussion

4.1.New-type B.napus lines are adapted to the target environments and represent a rich reservoir of novel diversity

A significant differentiation between the gene pool of new-typeB.napuslines and the traditionalB.napuspopulation has been described in previous studies [35,37,38].The focus was on exotic introgressions of the parent-related species,novel allelic genetic variation,and genomic structural variations.In line with these findings,the new-typeB.napuslines used in our study also showed pronounced genetic differences from the six elite tester lines(Fig.1).The use of SNPScalled on the basis of the specific genome fragments of the parent species that do not map to the reference genome ofB.napusmade the differences between the two gene pools clearer,compared to the use of markers called on the basis of the reference genome of traditionalB.napus(SNPT).Thus,the lines of the new-typeB.napusrepresent a rich reservoir of novel molecular diversity.It is important to note that the genetic differences have most likely still been underestimated as not all exotic introgresssion and newly induced variations have been detected using the currently available reference genome sequence representingB.rapaandB.carinata.Additionally,there was possibility that a few SNPSmarkers from the exotic introgression might be mixed in the panel of SNPTmarkers because of high homology of the progenitor reference genomes withB.napus.On the other hand,a few SNPTmarkers might also be mixed in the panel of SNPSmarkers because of the missing assembly of the reference genome Darmor4.1 but with high homology with the A and C progenitor diploid reference genomes.The ongoing activities to develop pan-genomes forB.napusand related species [26,34,42,49,50,58–60] will complement our picture on inter-subgenomic diversity in the future.

The lines of the new-typeB.napuswere selected in the past on the basis of important agronomic characteristics such as grain yield,other agronomic and seed quality characteristics.The selection took place in the WH semi-winter environment,and most of the lines could be adapted to the spring environment without requirement of vernalization when grown in the northwest of China.As a result,the lines were adapted,which is reflected in the flowering of the 61 new-typeB.napuslines,which were in the range of the adapted elite lines.Furthermore,the lines per se performance in seed yield was on average only 19%below the average performance of the semi-winter tester lines and 27%above that of the spring tester lines (Table S2).The success in breeding for quality is documented by 83.6% of the lines of the new-typeB.napusthat achieved a ‘‘double low” or ‘‘single low” quality level,i.e.a glucosinolate content<30 μmol g-1and an erucic acid content<3%.In summary,the new-typeB.napuslines not only represent a rich reservoir of novel variety,but are also adapted to the needs of rapeseed cultivation in the target environments.

4.2.Promising heterotic pattern between new-type and elite spring/winter B.napus lines

InBrassicacrops,introgression with related species is frequently achieved to broaden the genetic basis [37,61–67].For example,the AC subgenome ofB.napuswere reorganized by massive introgression or even resynthesized with its precursor species[27–31,35].This novel diversity is particularly interesting forB.napushybrids to boost heterosis and,with an acceptable lineper seperformance,also the hybrid performance.The potential of hybrids between new-type and eliteB.napusis documented by a total of 15 single-crosses,which showed significant commercial heterosis (Table 3).Interestingly,the new-typeB.napuslines showed on average the highest hybrid performance when crossed with the elite spring-type tester6 in spring environment.Besides,hybrids derived from winter type tester showed a similar average hybrid performance with hybrids derived from semi-winter type tester in XY semi-winter environment,which is not the adapted environment for winter typeB.napus.This suggests that newtype and elite spring/winterB.napuslines represent an interesting heterotic pattern that can be exploited in the future.To further improve the heterosis by crossing with winter-type testers grown in winter environment,the new-typeB.napuswould be selected or edited on the genes involving vernalization,flowering time and cold tolerance.

4.3.Genome-wide prediction in pre-breeding

The phenotypic results of our study showed the success of the comprehensive pre-breeding activities.However,it has to be taken into account that pre-breeding is time consuming and resource intensive:Based on the crosses between the founder lines,it took 13 years to breed this adapted set of 61 new-typeB.napuslines[35,37,38].The long period of time was needed to eliminate undesirable quality traits,increase adaptation to the target environment,and at the same time enrich promising introgressions for seed yield.The latter is particularly challenging because the ultimate goal is hybrid but not lineper seperformance and both were only weakly correlated in our study for seed yield with an average correlation of 0.24 for the 6 test crosses.To solve this challenge,we explored the possibility to select for hybrid performance using genome-wide prediction.

In agreement with earlier results with rapeseed [56],but also other crops such as wheat [68],maize [69],sunflower [70],or rice[71],we observed a strong dependence of the prediction ability on the relationship between the training and test population:Predictions of single crosses can be made with medium to high accuracy if both parental lines were also used as parents in the training population (Table 4).Our results therefore clearly underlined the potential of genome-wide hybrid prediction also for intersubgenomicB.napussingle crosses in the case of related training and test populations.Interestingly,the prediction ability of the simplest M1 based on 5-fold cross validation was nearly equal compared to the other four models (Table 5) suggesting that the choice of the biometrical model is of less importance.Furthermore,our results suggest that in the case of inter-subgenomicB.napushybrids,including markers that specifically target introgression(SNPS) (M3 or M4),the prediction ability of certain traits (SN,TSW,GLU,EAC,OC)for small training populations(about 144 individuals in TS2)can be increased(Table 4).A part of the SNPTmarkers may be closely linked with SNPSmarkers because of the homologous of A,B and C genome inBrassicaspecies,as a result,all traits can also obtain a relatively good prediction ability based onlyB.napusmarkers (M1) (Tables 4 and 5).Besides of exotic introgression,rich genomic rearrangement events such as translocation,inversion,large deletion and duplication also happened in the gene pool [35–38].Future studies are needed to evaluate if and how these genomic rearrangement events would influence prediction ability.

In addition,the challenge and at the same time another interesting situation for pre-breeding is to predict the performance of unrelated hybrids.With the training population composed in our study,this is only feasible with low prediction ability (Table 4).The low prediction ability of the hybrid performance can be interpreted as an indicator that the effects of individual exotic introgression can only be estimated with low accuracy and cannot be used for pre-breeding.Thus,in order to exploit fully the potential of genome-wide prediction in pre-breeding,larger training populations are required.However,compiling these populations is too costly for individual pre-breeding programs.An alternative might be to combine data from different pre-breeding programs with the process of exotic introgression in different plant materials,to facilitate more precise and universal genome-wide selection.

5.Conclusions

Using phenotypic values and genotypic data of 363 hybrids ofB.napusderived from two parental populations divergent on massive exotic introgression of related species,we observed strong heterosis on seed yield and high level of genome-wide prediction ability.The comparison between genome-wide prediction models under two cross validation methods revealed that the choice of the genome-wide prediction model was not impacting substantially the prediction abilities.In contrast,trait complexity,exotic introgression,genetic relationship between the training set and testing set,training set size,and environments affected the prediction ability.Overall,our findings suggest that genome-wide prediction are promising and can support breeding of hybrids between genetically distinct parent groups with exotic introgressions.

CRediT authorship contribution statement

Dandan Huperformed the research and wrote the manuscript;Yusheng Zhaoperformed the analysis on genome prediction;Xiangxiang Heperformed parts of the field experiment;Yikai Zhangperformed the SNP calling;Jinxiong Shen,Rod Snowdon,and Jinling Menghelped the research materials and field experiments;Yong Jianganalyzed the data;Jun Zoudesigned the research,analyzed the data,and wrote the manuscript;Jochen C.Reifanalyzed the data and wrote the manuscript.The authors declare that they have no competing interests.All authors revised,read,and approved the final manuscript.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (NSFC-DFG,31861 133016;NSFC,31970564).We acknowledged the assistance of Prof.Xuhui Mao from Gansu Academy of Agricultural Sciences,Prof.Wancang Sun from Gansu Agricultural University,and Dr.Guiping Bai from Xiangyang Academy of Agricultural Sciences for the field experiment.And we also acknowledged Prof.Guangsheng Yang from Huazhong Agricultural University and Prof.Wanchang Sun to provide the commercial check hybrids grown in semi-winter and spring environment,respectively.

Appendix A.Supplementary data

Supplementary data for this article can be found online at https://doi.org/10.1016/j.cj.2020.11.002.