Large-scale genotyping and phenotyping of a worldwide winter wheat genebank for its use in pre-breeding

Abstract

Plant genetic resources (PGR) stored at genebanks are humanity's crop diversity savings for the future. Information on PGR contrasted with modern cultivars is key to select PGR parents for pre-breeding. Genotyping-by-sequencing was performed for 7,745 winter wheat PGR samples from the German Federal ex situ genebank at IPK Gatersleben and for 325 modern cultivars. Whole-genome shotgun sequencing was carried out for 446 diverse PGR samples and 322 modern cultivars and lines. In 19 field trials, 7,683 PGR and 232 elite cultivars were characterized for resistance to yellow rust - one of the major threats to wheat worldwide. Yield breeding values of 707 PGR were estimated using hybrid crosses with 36 cultivars - an approach that reduces the lack of agronomic adaptation of PGR and provides better estimates of their contribution to yield breeding. Cross-validations support the interoperability between genomic and phenotypic data. The here presented data are a stepping stone to unlock the functional variation of PGR for European pre-breeding and are the basis for future breeding and research activities.

Fig. 1

Number of counts according to the geographical origins of the 9,616 genotypes considered in the current study. (a) World map of 7,639 out of 9,145 winter wheat plant genetic resources (PGR) from the IPK genebank with 55 known geographical origins. (b) Map of Europe (excluding Russia), portraying releasing/obtention countries for 471 elite genotypes (340 European cultivars plus 131 German breeding lines). In (a) and (b), territories in gray lack entries. All maps were generated with Datawrapper.

Fig. 2

Distribution of the number of PGR (orange columns), elite cultivars (green) and breeding lines (blue) considered in this study according to their years of acquisition, release or obtention. When present, the exact counts number of genotypes per year are included within brackets [].

Fig. 3

Molecular neutral diversity and linkage disequilibrium decay in genebank and elite plant material. Molecular diversity portrayed by the first two principal coordinates (PCos) from Rogers’ distance matrices calculated using genotyping-by-sequencing (GBS, (a)) and whole-genome sequencing (WGS, (b)). Intra-chromosomal linkage disequilibrium (r²) as a function of the genomic physical distance (Mb) in GBS (c) and WGS (d). GBS was conducted for 7,745 plant genetic resources (PGR) samples from the IPK genebank and 325 European elite cultivars. WGS was performed for 191 European elite cultivars, 131 German elite breeding lines and 446 PGR samples from the IPK genebank. Percentage of variation explained by PCos are included in brackets (). For r² decay, distances between SNP pairs correspond to RefSeq v1.0 of Chinese Spring while cubic splines were fitted to whole genomes but only the first 20 Mb are portrayed.

Fig. 4

Distribution of the best linear unbiased estimations (BLUEs) across experiments for outlier-corrected yellow rust (YR, Puccinia striiformis f. sp. tritici) infections of plant genetic resources (PGR or SSD-PGR) and elite cultivars (Elite) tested in precision (boxplot, upper left corner), large-scale screening (boxplot, lower right) or both types of field experiments (scatter plot, upper right). YR infections were scored using an ordinal rating scale between 1 and 9, where 1 means complete absence of YR leaf symptoms and 9 denotes fully infected leaves. BLUEs that lie outside of the 1–9 parametric space are due to the unorthogonal structure of unbalanced experiments. In total, 19 field experiments were conducted between harvest years 2015 and 2020 considering five German locations. Large-scale screenings fully relied on natural YR infections, while five out of seven precision experiments were artificially inoculated. The exact numbers of genotypes according to each category are included within brackets []. In boxplots, boxes enclose 50% of the central data, including median (black bold line) and mean (black diamond), while whiskers are ± 1.5 × interquartile range and dots represent extreme values. In the scatter plot, ** denotes the significance [-log₁₀(p-value) = 128.4] of the correlation between YR scores from precision and large-scale screening experiments.

Fig. 5

Using yield breeding value estimates of plant genetic resources (PGR) to initiate pre-breeding programs in wheat. (a) Kernel density distribution of yield breeding values (Mg/ha) for 707 PGR. Breeding values were estimated using yield data of ‘Elite × PGR’ F₁ hybrids from 22 field experiments conducted between harvest years 2016 and 2020. Based on preliminary data from 2016, 13 PGR with superior breeding values were used as male parents in two- (Elite₁ × PGR) and three-way (Elite₂ × [Elite₁ × PGR]) crosses involving 11 adapted elite cultivars. Vertical dashed lines indicate the breeding values of selected PGR estimated across the full set of 22 experiments. (b) After two-stage selection for high leaf health and reduced plant height, 173 advanced F_3:4 PGR-derived progenies tracing back to 32 initial crosses were tested together with 15 elite cultivar checks (black dots) and 16 additional IPK pre-breeding lines (gray dots) in yield validation experiments conducted in two locations during harvest years 2020 and 2021. The best linear unbiased estimations of yield (Mg/ha) computed across validation experiments for the tested material are portrayed and grouped according to each initial cross. The color legend of PGR-derived populations matches that of the selected PGR parents used in initial crosses. Horizontal dotted and dashed lines indicate the yield performances of the best newest cultivar (‘LGCharacter’) and the mostly grown cultivar during the last decade (‘RGTReform’) in Germany, respectively.

Fig. 6

Distributions of cross-validated interoperability between genomic and phenotypic data. Genotyping platforms were genotyping-by-sequencing (GBS) and whole-genome sequencing (WGS, 3-fold coverage), while phenotypes corresponded to the best linear unbiased estimates for yellow rust (YR) scorings computed across large-scale screening or precision experiments, as well as yield breeding values (BV) computed across estimation experiments. Interoperability was estimated as the genomic prediction accuracy using 80% of the integrated data as training and 20% as validation set in 100 cross-validation runs. Total number (N) of samples with phenotypes and polymorphic SNP markers used for cross-validations according to each genotyping platform are portrayed as table on the left side. In case of GBS*, the same training and validation phenotypes used for WGS were considered. For more details on cross-validations, please see Methods. In distributions, diamonds, horizontal and vertical lines correspond to the average, standard deviation and median, respectively. Violin plots were obtained using the vioplot R package (v0.3.7).