Most studies on genomic prediction of hybrids employ genetic markers as the main carrier of information. Very few use transcriptomic or metabolomic data despite the fact that the end product of gene expression, i.e., the protein, might carry more information than genetic markers. The main goal of the present study was therefore to investigate whether gene expression profiles can be employed successfully for hybrid prediction in maize.With RR-BLUP, similar accuracies were found for ALFP markers and mRNA transcription profiles for prediction of hybrid maize grain yield and grain dry matter content within a set of 98 factorial crosses and within an unbalanced set of 230 maize hybrids.No investigations on the number of required mRNA transcripts for reliable predictions had been conducted so far. Comparable prediction accuracies were found for 10k, 2k and 1k mRNA transcripts, and 1k AFLP markers. This means that also in terms of the number of mRNA transcripts required, the mRNA transcription profiles are comparable to genetic markers.A major challenge is the successful prediction of hybrids whose parents are not in the training set. Transcriptome-based binary distances based on 10k mRNA transcripts had been shown to be advantageous in this situation. However, prediction with binary distances based on 2k mRNA transcripts was found to be inferior to prediction with RR-BLUP in most cases, especially in grain dry matter content, a trait with low heterosis. Apparently, a large number of mRNAs must be available to select from for meaningful transcriptome-based distances. Pre-selection of a core set of genes for hybrid prediction with transcriptome-based distances in order to save resources is therefore not a promising approach.Even more than the generation of marker data, the calibration of models based on appropriate training sets is very resource-intensive. It would therefore be beneficial for breeders if material from previous breeding cycles could be employed for that purpose. Prediction accuracies of RR-BLUP with ALFP markers and mRNA transcription profiles were evaluated when one or two of the four sets of factorial crosses formed the training set and the remaining factorials formed the validation set. Mean prediction accuracies in grain yield and grain dry matter content were higher than 0.55 in type 1 hybrids and 0.16 to 0.38 in type 0 hybrids. Thus, prediction with models calibrated with material from previous breeding cycles seems to be possible if sufficient relatedness is ensured.In conclusion, mRNA transcription profiles can be regarded as promising predictors of hybrid performance in maize.
Verknüpfung zu Publikationen oder weiteren Datensätzen
