A replicated experiment on genomic selection in a maize landrace provides valuable insights on the design of rapid cycling recurrent pre-breeding schemes and the factors contributing to their success. The genetic diversity of landraces is currently underutilized for elite germplasm improvement. In this study, we investigated the potential of rapid cycling genomic selection for pre-breeding of a maize (Zea mays L.) landrace population in replicated experiments. We trained the prediction model on a dataset (N = 899) composed of three landrace-derived doubled-haploid (DH) populations characterized for agronomic traits in 11 environments across Europe. All DH lines were genotyped with a 600 k SNP array. In two replications, three cycles of genomic selection and recombination were performed for line per se performance of early plant development, a major sustainability factor in maize production. From each cycle and replication, 100 DH lines were extracted. To evaluate selection response, the DH lines of all cycles and both replications (N = 688) were evaluated for per se performance of selected and unselected traits in seven environments. Selection was highly successful with an increase of about two standard deviations for traits under directional selection. Realized selection response was highest in the first cycle and diminished in following cycles. Selection gains predicted from genomic breeding values were only partially corroborated by realized gains estimated from adjusted means. Prediction accuracies declined sharply across cycles, but only for traits under directional selection. Retraining the prediction model with data from previous cycles improved prediction accuracies in cycles 2 and 3. Replications differed in selection response and particularly in accuracies. The experiment gives valuable insights with respect to the design of rapid cycling genomic selection schemes and the factors contributing to their success.

Selection response in truncation selection across multiple sets of candidates hinges on their post-selection proportions, which can deviate grossly from their initial proportions. For BLUPs, using a uniform threshold for all candidates maximizes the selection response, irrespective of differences in population parameters. Plant breeding programs typically involve multiple families from either the same or different populations, varying in means, genetic variances and prediction accuracy of BLUPs or BLUEs for true genetic values (TGVs) of candidates. We extend the classical breeder's equation for truncation selection from single to multiple sets of genotypes, indicating that the expected overall selection response (Δ G_Tot) for TGVs depends on the selection response within individual sets and their post-selection proportions. For BLUEs, we show that maximizing Δ G_Tot requires thresholds optimally tailored for each set, contingent on their population parameters. For BLUPs, we prove that Δ G_Tot is maximized by applying a uniform threshold across all candidates from all sets. We provide explicit formulas for the origin of the selected candidates from different sets and show that their proportions before and after selection can differ substantially, especially for sets with inferior properties and low proportion. We discuss implications of these results for (a) optimum allocation of resources to training and prediction sets and (b) the need to counteract narrowing the genetic variation under genomic selection. For genomic selection of hybrids based on BLUPs of GCA of their parent lines, selecting distinct proportions in the two parent populations can be advantageous, if these differ substantially in the variance and/or prediction accuracy of GCA. Our study sheds light on the complex interplay of selection thresholds and population parameters for the selection response in plant breeding programs, offering insights into the effective resource management and prudent application of genomic selection for improved crop development.

Abstract Background Accuracy of genomic prediction depends on the heritability of the trait, the size of the training set, the relationship of the candidates to the training set, and the $$\text {Min}(N_{\text {QTL}},M_e)$$ Min ( N QTL , M e ) , where $$N_{\text {QTL}}$$ N QTL is the number of QTL and $$M_e$$ M e is the number of independently segregating chromosomal segments. Due to LD, the number $$Q_e$$ Q e of independently segregating QTL (effective QTL) can be lower than $$\text {Min}(N_{\text {QTL}},M_e)$$ Min ( N QTL , M e ) . In this paper, we show that $$Q_e$$ Q e is inversely associated with the trait-specific genomic relationship of a candidate to the training set. This provides an explanation for the inverse association between $$Q_e$$ Q e and the accuracy of prediction. Methods To quantify the genomic relationship of a candidate to all members of the training set, we considered the $$k^2$$ k 2 statistic that has been previously used for this purpose. It quantifies how well the marker covariate vector of a candidate can be represented as a linear combination of the rows of the marker covariate matrix of the training set. In this paper, we used Bayesian regression to make this statistic trait specific and argue that the trait-specific genomic relationship of a candidate to the training set is inversely associated with $$Q_e$$ Q e . Simulation was used to demonstrate the dependence of the trait-specific $$k^2$$ k 2 statistic on $$Q_e$$ Q e , which is related to $$N_{\text {QTL}}$$ N QTL . Conclusions The posterior distributions of the trait-specific $$k^2$$ k 2 statistic showed that the trait-specific genomic relationship between a candidate and the training set is inversely associated to $$Q_e$$ Q e and $$N_{\text {QTL}}$$ N QTL . Further, we show that trait-specific genomic relationship between a candidate and the training set is directly related to the size of the training set.

Genomic prediction of GCA effects based on model training with full-sib rather than half-sib families yields higher short- and long-term selection gain in reciprocal recurrent genomic selection for hybrid breeding, if SCA effects are important. Reciprocal recurrent genomic selection (RRGS) is a powerful tool for ensuring sustainable selection progress in hybrid breeding. For training the statistical model, one can use half-sib (HS) or full-sib (FS) families produced by inter-population crosses of candidates from the two parent populations. Our objective was to compare HS-RRGS and FS-RRGS for the cumulative selection gain ( ΣΔ G ), the genetic, GCA and SCA variances ( σ_G^2 , σ_gca^2 , σ_sca^2 ) of the hybrid population, and prediction accuracy ( r_gca ) for GCA effects across cycles. Using SNP data from maize and wheat, we simulated RRGS programs over 10 cycles, each consisting of four sub-cycles with genomic selection of N_e = 20 out of 950 candidates in each parent population. Scenarios differed for heritability ( h^2) and the proportion τ = 100 ×σ_sca^2 :σ_G^2 of traits, training set (TS) size ( N_TS ), and maize vs. wheat. Curves of ΣΔ G over selection cycles showed no crossing of both methods. If τ was high, ΣΔ G was generally higher for FS-RRGS than HS-RRGS due to higher r_gca . In contrast, HS-RRGS was superior or on par with FS-RRGS, if τ or h^2 and N_TS were low. ΣΔ G showed a steeper increase and higher selection limit for scenarios with low τ , high h^2 and large N_TS . σ_gca^2 and even more so σ_sca^2 decreased rapidly over cycles for both methods due to the high selection intensity and the role of the Bulmer effect for reducing σ_gca^2 . Since the TS for FS-RRGS can additionally be used for hybrid prediction, we recommend this method for achieving simultaneously the two major goals in hybrid breeding: population improvement and cultivar development.

Training sets produced by maximizing the number of parent lines, each involved in one cross, had the highest prediction accuracy for H0 hybrids, but lowest for H1 and H2 hybrids. Genomic prediction holds great promise for hybrid breeding but optimum composition of the training set (TS) as determined by the number of parents (nTS) and crosses per parent (c) has received little attention. Our objective was to examine prediction accuracy ( r_a ) of GCA for lines used as parents of the TS (I1 lines) or not (I0 lines), and H0, H1 and H2 hybrids, comprising crosses of type I0 × I0, I1 × I0 and I1 × I1, respectively, as function of nTS and c. In the theory, we developed estimates for r_a of GBLUPs for hybrids: (i) r̂_a based on the expected prediction accuracy, and (ii) r̃_a based on r_a of GBLUPs of GCA and SCA effects. In the simulation part, hybrid populations were generated using molecular data from two experimental maize data sets. Additive and dominance effects of QTL borrowed from literature were used to simulate six scenarios of traits differing in the proportion (τSCA = 1 r̃_a and r̂_a closely agreed with r_a for hybrids. For given size NTS = nTS × c of TS, r_a of H0 hybrids and GCA of I0 lines was highest for c = 1. Conversely, for GCA of I1 lines and H1 and H2 hybrids, c = 1 yielded lowest r_a with concordant results across all scenarios for both data sets. In view of these opposite trends, the optimum choice of c for maximizing selection response across all types of hybrids depends on the size and resources of the breeding program.

Mating designs determine the realized additive genetic variance in a population sample. Deflated or inflated variances can lead to reduced or overly optimistic assessment of future selection gains. The additive genetic variance V_A inherent to a breeding population is a major determinant of short- and long-term genetic gain. When estimated from experimental data, it is not only the additive variances at individual loci (QTL) but also covariances between QTL pairs that contribute to estimates of V_A . Thus, estimates of V_A depend on the genetic structure of the data source and vary between population samples. Here, we provide a theoretical framework for calculating the expectation and variance of V_A from genotypic data of a given population sample. In addition, we simulated breeding populations derived from different numbers of parents (P = 2, 4, 8, 16) and crossed according to three different mating designs (disjoint, factorial and half-diallel crosses). We calculated the variance of V_A and of the parameter b reflecting the covariance component in V_A, standardized by the genic variance. Our results show that mating designs resulting in large biparental families derived from few disjoint crosses carry a high risk of generating progenies exhibiting strong covariances between QTL pairs on different chromosomes. We discuss the consequences of the resulting deflated or inflated V_A estimates for phenotypic and genome-based selection as well as for applying the usefulness criterion in selection. We show that already one round of recombination can effectively break negative and positive covariances between QTL pairs induced by the mating design. We suggest to obtain reliable estimates of V_A and its components in a population sample by applying statistical methods differing in their treatment of QTL covariances.

Cold reduces maize (Zea mays L.) production and delays sowings. Cold tolerance in maize is very limited, and breeding maize for cold tolerance is still a major challenge. Our objective was to detect QTL for cold tolerance at germination and seedling stages. We evaluated, under cold and control conditions, 919 Dent and 1009 Flint inbred lines from two nested association mapping designs consisting in 24 double-haploid populations, genotyped with 56,110 SNPs. We found a large diversity of maize cold tolerance within these NAM populations. We detected one QTL for plant weight and four for fluorescence under cold conditions, as well as one for plant weight and two for chlorophyll content under control conditions in the Dent-NAM. There were fewer significant QTL under control conditions than under cold conditions, and half of the QTL were for quantum efficiency of photosystem II. Our results supported the large genetic discrepancy between optimal and low temperatures, as the quantity and the position of the QTL were very variable between control and cold conditions. Furthermore, as we have not found alleles with significant effects on these NAM designs, further studies are needed with other experimental designs to find favorable alleles with important effects for improving cold tolerance in maize.

Discovery and enrichment of favorable alleles in landraces are key to making them accessible for crop improvement. Here, we present two fundamentally different concepts for genome-based selection in landrace-derived maize populations, one based on doubled-haploid (DH) lines derived directly from individual landrace plants and the other based on crossing landrace plants to a capture line. For both types of populations, we show theoretically how allele frequencies of the ancestral landrace and the capture line translate into expectations for molecular and genetic variances. We show that the DH approach has clear advantages over gamete capture with generally higher prediction accuracies and no risk of masking valuable variation of the landrace. Prediction accuracies as high as 0.58 for dry matter yield in the DH population indicate high potential of genome-based selection. Based on a comparison among traits, we show that the genetic makeup of the capture line has great influence on the success of genome-based selection and that confounding effects between the alleles of the landrace and the capture line are best controlled for traits for which the capture line does not outperform the ancestral population per se or in testcrosses. Our results will guide the optimization of genomeenabled prebreeding schemes.

The importance of phosphorus (P) in agriculture contrasts with the negative environmental impact and the limited resources worldwide. Reducing P fertilizer application by utilizing more efficient genotypes is a promising way to address these issues. To approach this, a large panel of maize (Zea mays L.) comprising each 100 Flint and Dent elite lines and 199 doubled haploid lines from six landraces was assessed in multi-environment field trials with and without the application of P starter fertilizer. The treatment comparison showed that omitting the starter fertilizer can significantly affect traits in early plant development but had no effect on grain yield. Young maize plants provided with additional P showed an increased biomass, faster growth and superior vigor, which, however, was only the case under environmental conditions considered stressful for maize cultivation. Importantly, though the genotype-by-treatment interaction variance was comparably small, there is genotypic variation for this response that can be utilized in breeding. The comparison of elite and doubled haploid landrace lines revealed a superior agronomic performance of elite material but also potentially valuable variation for early traits in the landrace doubled haploid lines. In conclusion, our results illustrate that breeding for P efficient maize cultivars is possible towards a reduction of P fertilizer in a more sustainable agriculture.

Genomic selection is a well-investigated approach that facilitates and supports selection decisions for complex traits and has meanwhile become a standard tool in modern plant breeding. Phenomic selection has only recently been suggested and uses the same statistical procedures to predict the targeted traits but replaces marker data with near-infrared spectroscopy data. It may represent an attractive low-cost, high-throughput alternative but has not been sufficiently studied until now. Here, we used 400 genotypes of maize (Zea mays L.) comprising elite lines of the Flint and Dent heterotic pools as well as 6 Flint landraces, which were phenotyped in multienvironment trials for anthesis-silking-interval, early vigor, final plant height, grain dry matter content, grain yield, and phosphorus concentration in the maize kernels, to compare the predictive abilities of genomic as well as phenomic prediction under different scenarios. We found that both approaches generally achieved comparable predictive abilities within material groups. However, phenomic prediction was less affected by population structure and performed better than its genomic counterpart for predictions among diverse groups of breeding material. We therefore conclude that phenomic prediction is a promising tool for practical breeding, for instance when working with unknown and rather diverse germplasm. Moreover, it may make the highly monopolized sector of plant breeding more accessible also for low-tech institutions by combining well established, widely available, and cost-efficient spectral phenotyping with the statistical procedures elaborated for genomic prediction - while achieving similar or even better results than with marker data.

Genetic variation is the basis of selection, evolution and breeding. Maize landraces represent a rich source of allelic diversity, but their efficient utilization in breeding and research has been hampered by their heterogeneous and heterozygous nature and insufficient information about most accessions. While molecular inventories of germplasm repositories are growing steadily, linking these data to meaningful phenotypes for quantitative traits is challenging.Here, we present comprehensive molecular and phenotypic data for ∼1,000 doubled-haploid (DH) lines derived from three pre-selected European maize landraces. Due to their full homozygosity, the DH lines can be multiplied ad libitum and represent a powerful biological resource available to the community. The DH lines allow high-precision phenotyping in repeated experiments and reveal the full additive genetic variance of the population. The DH lines were evaluated for nine agronomically important, quantitative traits in multi-environment field trials comprising seven locations and two years. The DH populations revealed high genetic variance and high heritability for the analysed traits. The DH lines were genotyped with 600k SNP markers. After stringent quality filtering 500k markers remained for further analyses.This is the largest resource of landrace derived DH material in maize, unprecedented in its structure and dimension. The presented data are ideal for linking molecular variation to meaningful phenotypes. They can be used for genome-wide association studies, genomic prediction, and population genetic analyses as well as for developing and testing statistical methods. All plant material is available to the community for conducting additional experiments, extending the panel of traits and environments, and for testing the landrace-derived lines in combination with other genetic material.

Model training on data from all selection cycles yielded the highest prediction accuracy by attenuating specific effects of individual cycles. Expected reliability was a robust predictor of accuracies obtained with different calibration sets. The transition from phenotypic to genome-based selection requires a profound understanding of factors that determine genomic prediction accuracy. We analysed experimental data from a commercial maize breeding programme to investigate if genomic measures can assist in identifying optimal calibration sets for model training. The data set consisted of six contiguous selection cycles comprising testcrosses of 5968 doubled haploid lines genotyped with a minimum of 12,000 SNP markers. We evaluated genomic prediction accuracies in two independent prediction sets in combination with calibration sets differing in sample size and genomic measures (effective sample size, average maximum kinship, expected reliability, number of common polymorphic SNPs and linkage phase similarity). Our results indicate that across selection cycles prediction accuracies were as high as 0.57 for grain dry matter yield and 0.76 for grain dry matter content. Including data from all selection cycles in model training yielded the best results because interactions between calibration and prediction sets as well as the effects of different testers and specific years were attenuated. Among genomic measures, the expected reliability of genomic breeding values was the best predictor of empirical accuracies obtained with different calibration sets. For grain yield, a large difference between expected and empirical reliability was observed in one prediction set. We propose to use this difference as guidance for determining the weight phenotypic data of a given selection cycle should receive in model retraining and for selection when both genomic breeding values and phenotypes are available.

Association mapping with immortalized lines of landraces offers several advantages including a high mapping resolution, as demonstrated here in maize by identifying the causal variants underlying QTL for oil content and the metabolite allantoin. Landraces are traditional varieties of crops that present a valuable yet largely untapped reservoir of genetic variation to meet future challenges of agriculture. Here, we performed association mapping in a panel comprising 358 immortalized maize lines from six European Flint landraces. Linkage disequilibrium decayed much faster in the landraces than in the elite lines included for comparison, permitting a high mapping resolution. We demonstrate this by fine-mapping a quantitative trait locus (QTL) for oil content down to the phenylalanine insertion F469 in DGAT1-2 as the causal variant. For the metabolite allantoin, related to abiotic stress response, we identified promoter polymorphisms and differential expression of an allantoinase as putative cause of variation. Our results demonstrate the power of this approach to dissect QTL potentially down to the causal variants, toward the utilization of natural or engineered alleles in breeding. Moreover, we provide guidelines for studies using ancestral landraces for crop genetic research and breeding.

Improved management and breeding increased maize ( Zea mays L.) yields over the last century, but nutritional efficiency was usually not the focus. This study investigates whether old and recently released flint and dent maize seedlings vary in the phosphorus (P) acquisition and utilization. P use efficiency (PUE) and related traits were measured and compared at two P levels in a calcareous soil. PUE and P acquisition efficiency (PAE) from founder flints to elite flints declined over the last decades. This was associated with smaller root systems, reduced ability to exploit external P, decreased rhizosphere pH and shorter root hairs in low P. Comparing flints with doubled haploid landraces (DH_LR), old and elite dents and hybrids revealed that dents started to acquire exogenous P earlier and had improved PUE. Most DH_LRs had similar PUE as elite flints. When evaluating root traits associated with P efficiency, seed P was also critical, and it is important to stack different root traits to optimize PUE, P utilization efficiency (PUtE) and PAE in breeding programmes. The root hair length, the ability to acidify the rhizosphere and the root diameter in flint and dent pools may be utilized to improve P use in maize germplasm.

Aims: High maize yields in modern well-managed agroecosystems depend on the use of elite varieties and hybrids. Unfortunately, because of repeated selection at high fertilizer rates, some beneficial traits, such as the interaction with arbuscular mycorrhizal fungi or the release of organic acid anions for phosphate mobilization and for attracting beneficial microorganisms, might be gradually declining in modern elite genotypes. However, old founder lines and landraces possibly carry genetic relicts that originate from pre-green revolution times that are useful for breeding elite material for low input farming systems. Methods: Seedling colonization with arbuscular mycorrhizal fungi (AMF) and organic acid anion release were measured in flint lines that were released over more than five decades ago and in six preselected doubled haploid (DH) lines from landraces. P-uptake-related root traits were compared under P-sufficient and P-deficient conditions. Results: Weak trends for the loss of AMF colonization or changes in organic acid anion release at low P supply were detected in modern varieties. One DH line from a landrace was found with increased mycorrhization, whereas others were similar to modern elite lines. Overall, substantial genetic variance was encountered for these traits. Conclusions: The concern that modern elite maize varieties have lost beneficial traits for nutrient acquisition is not substantiated for the flint pool of maize, although weak trends exist. Lines associated with better P-acquisition efficiency under limited P availability should be utilized for breeding more sustainable varieties.

A breeding strategy combining genomic with one-stage phenotypic selection maximizes annual selection gain for net merit. Choice of the selection index strongly affects the selection gain expected in individual traits. Selection indices using genomic information have been proposed in crop-specific scenarios. Routine use of genomic selection (GS) for simultaneous improvement of multiple traits requires information about the impact of the available economic and logistic resources and genetic properties (variances, trait correlations, and prediction accuracies) of the breeding population on the expected selection gain. We extended the R package “selectiongain” from single trait to index selection to optimize and compare breeding strategies for simultaneous improvement of two traits. We focused on the expected annual selection gain (ΔGa) for traits differing in their genetic correlation, economic weights, variance components, and prediction accuracies of GS. For all scenarios considered, breeding strategy GSrapid (one-stage GS followed by one-stage phenotypic selection) achieved higher ΔGa than classical two-stage phenotypic selection, regardless of the index chosen to combine the two traits and the prediction accuracy of GS. The Smith–Hazel or base index delivered higher ΔGa for net merit and individual traits compared to selection by independent culling levels, whereas the restricted index led to lower ΔGa in net merit and divergent results for selection gain of individual traits. The differences among the indices depended strongly on the correlation of traits, their variance components, and economic weights, underpinning the importance of choosing the selection indices according to the goal of the breeding program. We demonstrate our theoretical derivations and extensions of the R package “selectiongain” with an example from hybrid wheat by designing indices to simultaneously improve grain yield and grain protein content or sedimentation volume.

Genetic variation is of crucial importance for selection and genetic improvement of crops. Landraces are valuable sources of diversity for germplasm improvement, but for quantitative traits efficient strategies for their targeted utilization are lacking. Here, we propose a genome-based strategy for making native diversity accessible for traits with limited genetic variation in elite germplasm. We generated ~ 1,000 doubled-haploid (DH) lines from three European maize landraces, pre-selected based on molecular and phenotypic information. Using GWAS, we mapped haplotype-trait associations for early development traits at high resolution in eleven environments. Molecular haplotype inventories of landrace derived DH libraries and a broad panel of 65 breeding lines based on 501,124 SNPs revealed novel variation for target traits in the landraces. DH lines carrying these novel haplotypes outperformed breeding lines not carrying the respective haplotypes. Most haplotypes associated with target traits showed stable effects across populations and environments and only limited correlated effects with undesired traits making them ideal for introgression into elite germplasm. Our strategy was successful in linking molecular variation to meaningful phenotypes and identifying novel variation for quantitative traits in plant genetic resources.

High genetic variation in two European maize landraces can be harnessed to improve Gibberella ear rot resistance by integrated genomic tools. Fusarium graminearum (Fg) causes Gibberella ear rot (GER) in maize leading to yield reduction and contamination of grains with several mycotoxins. This study aimed to elucidate the molecular basis of GER resistance among 500 doubled haploid lines derived from two European maize landraces, “Kemater Landmais Gelb” (KE) and “Petkuser Ferdinand Rot” (PE). The two landraces were analyzed individually using genome-wide association studies and genomic selection (GS). The lines were genotyped with a 600-k maize array and phenotyped for GER severity, days to silking, plant height, and seed-set in four environments using artificial infection with a highly aggressive Fg isolate. High genotypic variances and broad-sense heritabilities were found for all traits. Genotype-environment interaction was important throughout. The phenotypic (r) and genotypic ( $${r}_{g}$$ ) correlations between GER severity and three agronomic traits were low (r =  − 0.27 to 0.20; $${r}_{g}\hspace{0.17em}$$ =  − 0.32 to 0.22). For GER severity, eight QTLs were detected in KE jointly explaining 34% of the genetic variance. In PE, no significant QTLs for GER severity were detected. No common QTLs were found between GER severity and the three agronomic traits. The mean prediction accuracies ( $$\rho $$ ) of weighted GS (wRR-BLUP) were higher than $$\rho $$ of marker-assisted selection (MAS) and unweighted GS (RR-BLUP) for GER severity. Using KE as the training set and PE as the validation set resulted in very low $$\rho $$ that could be improved by using fixed marker effects in the GS model.

In vivo haploid induction based on maternal haploid inducers is the first step in deriving completely homozygous maize doubled haploid (DH) lines. Haploid induction rate (HIR) is influenced by both pollen parent inducing haploidy and the maternal source germplasm used in induction crosses. This study was aimed at analyzing the influence of source germplasm on HIR using 671 tropical inbred lines organized in two association mapping panels. These two association mapping panels (AMP1 and AMP2) were crossed to two different Tropically Adapted Inducer Lines (TAILs). For HIR assessment, seeds from induction crosses were planted in the field and ploidy status of each surviving plant was assessed using a gold standard classification based on visual differences between the haploid and diploid plants. The analysis revealed significant variation for HIR and led to identification of several tropical inbred lines that respond very positively to haploid induction. Use of HIR data in a genome wide association study (GWAS) led to identification of twenty-seven and two SNPs that were significantly associated with HIR in AMP1 and AMP2, respectively. Meta-analysis of AMP1 and AMP2 GWAS led to identification of 52 SNPs with significant effect on HIR across both studies. Genome-wide prediction revealed moderate to high prediction accuracy within AMPs using random SNPs. Inclusion of the SNPs detected in GWAS into the prediction model led to improvement in prediction accuracy. Overall, the study revealed that the maternal influence on HIR is controlled by a few moderate and many small effect QTL.

Genetic variation is of crucial importance for crop improvement. Landraces are valuable sources of diversity, but for quantitative traits efficient strategies for their targeted utilization are lacking. Here, we map haplotype-trait associations at high resolution in ~1000 doubled-haploid lines derived from three maize landraces to make their native diversity for early development traits accessible for elite germplasm improvement. A comparative genomic analysis of the discovered haplotypes in the landrace-derived lines and a panel of 65 breeding lines, both genotyped with 600k SNPs, points to untapped beneficial variation for target traits in the landraces. The superior phenotypic performance of lines carrying favorable landrace haplotypes as compared to breeding lines with alternative haplotypes confirms these findings. Stability of haplotype effects across populations and environments as well as their limited effects on undesired traits indicate that our strategy has high potential for harnessing beneficial haplotype variation for quantitative traits from genetic resources.