Recurrent selection (RS) has been used in plant breeding to successively improve synthetic and other multiparental populations. Synthetics are generated from a limited number of parents (Np), but little is known about how Np affects genomic selection (GS) in RS, especially the persistency of prediction accuracy (rg,g^) and genetic gain. Synthetics were simulated by intermating Np= 2–32 parent lines from an ancestral population with short- or long-range linkage disequilibrium (LDA) and subjected to multiple cycles of GS. We determined rg,g^ and genetic gain across 30 cycles for different training set (TS) sizes, marker densities, and generations of recombination before model training. Contributions to rg,g^ and genetic gain from pedigree relationships, as well as from cosegregation and LDA between QTL and markers, were analyzed via four scenarios differing in (i) the relatedness between TS and selection candidates and (ii) whether selection was based on markers or pedigree records. Persistency of rg,g^ was high for small Np, where predominantly cosegregation contributed to rg,g^, but also for large Np, where LDA replaced cosegregation as the dominant information source. Together with increasing genetic variance, this compensation resulted in relatively constant long- and short-term genetic gain for increasing Np > 4, given long-range LDA in the ancestral population. Although our scenarios suggest that information from pedigree relationships contributed to rg,g^ for only very few generations in GS, we expect a longer contribution than in pedigree BLUP, because capturing Mendelian sampling by markers reduces selective pressure on pedigree relationships. Larger TS size (NTS) and higher marker density improved persistency of rg,g^ and hence genetic gain, but additional recombinations could not increase genetic gain.

Genomic selection (GS) offers the possibility to estimate the effects of genome-wide molecular markers, which can be used to calculate genomic estimated breeding values (GEBVs) for individuals without phenotypes. GEBVs can serve as a selection criterion in recurrent GS, maximizing single-cycle but not necessarily long-term genetic gain. As simple genome-wide sums, GEBVs do not take into account other genomic information, such as the map positions of loci and linkage phases of alleles. Therefore, we herein propose a novel selection criterion called expected maximum haploid breeding value (EMBV). EMBV predicts the expected performance of the best among a limited number of gametes that a candidate contributes to the next generation, if selected. We used simulations to examine the performance of EMBV in comparison with GEBV as well as the recently proposed criterion optimal haploid value (OHV) and weighted GS. We considered different population sizes, numbers of selected candidates, chromosome numbers and levels of dominant gene action. Criterion EMBV outperformed GEBV after about 5 selection cycles, achieved higher long-term genetic gain and maintained higher diversity in the population. The other selection criteria showed the potential to surpass both GEBV and EMBV in advanced cycles of the breeding program, but yielded substantially lower genetic gain in early to intermediate cycles, which makes them unattractive for practical breeding. Moreover, they were largely inferior in scenarios with dominant gene action. Overall, EMBV shows high potential to be a promising alternative selection criterion to GEBV for recurrent genomic selection.

Synthetics play an important role in quantitative genetic research and plant breeding, but few studies have investigated the application of genomic prediction (GP) to these populations. Synthetics are generated by intermating a small number of parents ([Formula: see text] and thereby possess unique genetic properties, which make them especially suited for systematic investigations of factors contributing to the accuracy of GP. We generated synthetics in silico from [Formula: see text]2 to 32 maize (Zea mays L.) lines taken from an ancestral population with either short- or long-range linkage disequilibrium (LD). In eight scenarios differing in relatedness of the training and prediction sets and in the types of data used to calculate the relationship matrix (QTL, SNPs, tag markers, and pedigree), we investigated the prediction accuracy (PA) of Genomic best linear unbiased prediction (GBLUP) and analyzed contributions from pedigree relationships captured by SNP markers, as well as from cosegregation and ancestral LD between QTL and SNPs. The effects of training set size [Formula: see text] and marker density were also studied. Sampling few parents ([Formula: see text]) generates substantial sample LD that carries over into synthetics through cosegregation of alleles at linked loci. For fixed [Formula: see text], [Formula: see text] influences PA most strongly. If the training and prediction set are related, using [Formula: see text] parents yields high PA regardless of ancestral LD because SNPs capture pedigree relationships and Mendelian sampling through cosegregation. As [Formula: see text] increases, ancestral LD contributes more information, while other factors contribute less due to lower frequencies of closely related individuals. For unrelated prediction sets, only ancestral LD contributes information and accuracies were poor and highly variable for [Formula: see text] due to large sample LD. For large [Formula: see text], achieving moderate accuracy requires large [Formula: see text], long-range ancestral LD, and high marker density. Our approach for analyzing PA in synthetics provides new insights into the prospects of GP for many types of source populations encountered in plant breeding.

Thousands of landraces are stored in seed banks as "gold reserves" for future use in plant breeding. In many crops, their utilization is hampered because they represent heterogeneous populations of heterozygous genotypes, which harbor a high genetic load. We show, with high-density genotyping in five landraces of maize, that libraries of doubled-haploid (DH) lines capture the allelic diversity of genetic resources in an unbiased way. By comparing allelic differentiation between heterozygous plants from the original landraces and 266 derived DH lines, we find conclusive evidence that, in the DH production process, sampling of alleles is random across the entire allele frequency spectrum, and purging of landraces from their genetic load does not act on specific genomic regions. Based on overall process efficiency, we show that generating DH lines is feasible for genetic material that has never been selected for inbreeding tolerance. We conclude that libraries of DH lines will make genetic resources accessible to crop improvement by linking molecular inventories of seed banks with meaningful phenotypes.

A major application of genomic prediction (GP) in plant breeding is the identification of superior inbred lines within families derived from biparental crosses. When models for various traits were trained within related or unrelated biparental families (BPFs), experimental studies found substantial variation in prediction accuracy (PA), but little is known about the underlying factors. We used SNP marker genotypes of inbred lines from either elite germplasm or landraces of maize (Zea mays L.) as parents to generate in silico 300 BPFs of doubled-haploid lines. We analyzed PA within each BPF for 50 simulated polygenic traits, using genomic best linear unbiased prediction (GBLUP) models trained with individuals from either full-sib (FSF), half-sib (HSF), or unrelated families (URF) for various sizes (Ntrain) of the training set and different heritabilities (h2). In addition, we modified two deterministic equations for forecasting PA to account for inbreeding and genetic variance unexplained by the training set. Averaged across traits, PA was high within FSF (0.41–0.97) with large variation only for Ntrain<50 and h2 <0.6. For HSF and URF, PA was on average ∼40–60% lower and varied substantially among different combinations of BPFs used for model training and prediction as well as different traits. As exemplified by HSF results, PA of across-family GP can be very low if causal variants not segregating in the training set account for a sizeable proportion of the genetic variance among predicted individuals. Deterministic equations accurately forecast the PA expected over many traits, yet cannot capture trait-specific deviations. We conclude that model training within BPFs generally yields stable PA, whereas a high level of uncertainty is encountered in across-family GP. Our study shows the extent of variation in PA that must be at least reckoned with in practice and offers a starting point for the design of training sets composed of multiple BPFs.