Browsing by Subject "Maiszüchtung"

Now showing 1 - 7 of 7

Biometrical Analyses of Epistasis and the Relationship between Line per se and Testcross Performance of Agronomic Traits in Elite Populations of European Maize (Zea mays L.)
(2005) Mihaljevic, Renata; Melchinger, Albrecht E.
Relations of yield and other important agronomic traits of inbred lines to the same traits in hybrids have been studied from the time of initiation of hybrid breeding to the present. Because crossing lines to a tester and conducting yield trials are expensive and time-consuming, reliable information on inbred lines that is indicative of their testcross performance is crucial for optimum testing schemes in hybrid breeding as well as simultaneous improvement of commercial hybrids and their inbred parents. It has therefore been of great importance to determine the magnitude of correlation between line per se performance (LP) and testcross performance (TP) and investigate if epistasis influences this correlation. The comprehensive study on hand was performed with five populations (F3 to F6 lines) differing in size (ranging from 71 to 344), level of inbreeding, and the number of common parents. The populations employed were derived from three biparental crosses within the heterotic pool of European elite flint maize (Zea mays L.). All five populations were evaluated for TP (using an unrelated dent tester inbred) of five agronomically important quantitative traits: grain yield, grain moisture, kernel weight, protein concentration, and plant height. Four of these populations were also evaluated for LP of the same five traits. The objectives were to (i) estimate phenotypic and genotypic correlations between LP and TP within four populations for all five traits, (ii) map quantitative trait loci (QTL) for LP and TP in four and five populations, respectively, for all five traits, (iii) validate estimated QTL effects and positions for TP by assessing QTL congruency among testcross populations differing in size and genetic background, (iv) determine the value of LP-QTL for the prediction of TP, (v) estimate the importance of epistatic effects for LP and TP of grain yield and grain moisture by generation means analysis as well as genome-wide testing for epistatic marker pairs, and (vi) draw conclusions regarding the prospects of marker-assisted selection (MAS). Genotypic correlations between LP and TP, rg(LP, TP), estimated herein were comparable with those obtained for European flint or U.S. dent material. The magnitude of rg(LP, TP) was trait-specific: for traits of high heritability, i.e. grain moisture, kernel weight, protein concentration, and plant height, estimates were generally larger than 0.7 across all four populations, whereas for grain yield, estimates were consistently lower and did not exceed the intermediate level of 0.5. For grain yield, lowest rg(LP, TP) were estimated with lowest precision (largest confidence intervals). This requires testing for both LP and TP and/or combining the data in a selection index to ensure sufficient inbred performance (seed production) and yield improvement. However, combined selection for LP and TP proved less efficient than sole selection for TP unless unadapted material was employed. For kernel weight, protein concentration, and plant height, we detected "large" congruent QTL across testcross populations derived from the same cross, which individually explained up to 46% of the validated genotypic variance p. However, as the p values estimated from validation were still below the corresponding heritability estimates, MAS will be superior to phenotypic selection only if it is more cost-efficient. For the above traits, similar numbers of QTL for LP and TP were detected across populations. More than half of the QTL regions detected for LP were in common for LP and TP in the largest population (N = 280). To assess the value of QTL identified for LP in predicting TP, we calculated the genotypic correlation rg(MLP, YTP). This parameter assesses QTL congruency for LP and TP quantitatively and is thus the key parameter for assessing the prospects of MAS. The number of common QTL for LP and TP (qualitative QTL congruency) was generally not indicative of the magnitude of rg(MLP, YTP) due to the differences in the effect size of the respective QTL detected for LP and used for the prediction of TP. For all traits, rg(MLP, YTP) were smaller than rg(LP, TP). This is because rg(MLP, YTP) is only predictive for the validated proportion of genotypic variance explained by the QTL for LP, which was generally below 50% because of the limited power of QTL detection, in particular with small sample sizes below 100. Only if QTL detected for LP explain a substantial proportion of the genotypic variance, MAS based on these QTL can be applied, provided it is more cost-efficient than an indirect phenotypic selection for TP based on LP. QTL detection power was drastically reduced for the complex trait grain yield with a presumably large number of small QTL underlying its genetic architecture. Thus, the number of common QTL for LP and TP as well as the QTL congruency across testcross populations was much lower for grain yield than the other four traits. Estimated gene action of QTL detected for LP was primarily additive for grain yield. Evidence for dominance and/or epistasis, which may be a reason for the low rg(LP, TP) and the low number of common QTL for LP and TP was generally weak. Both generation means analysis for LP and TP and genome-wide search for epistatic marker pairs yielded no evidence for epistasis. This is not only because the detected epistatic effects could not be validated, but also because there is low chance to find epistasis unless the generation examined displays the full epistatic variance such as expected from doubled haploids produced from an F1 cross. Thus, it is anticipated that the relative importance of epistatic effects in hybrid maize breeding may strongly increase with the currently happening shift in line development from recurrent selfing towards the production of doubled haploids.
Genetic variation in early maturing European maize germplasm for resistance to ear rots and mycotoxin contamination caused by Fusarium spp.
(2010) Bolduan, Christof; Melchinger, Albrecht E.
Ear rots of maize, caused by Fusarium spp., are of major concern because they lead to losses in grain yield and contamination with mycotoxins which harm animals and humans. In the absence of other strategies, breeding maize for genetic resistance is currently the most promising avenue to control these rots and mycotoxin accumulation. The predominant pathogens in Central Europe are F. graminearum, the causative agent of Gibberella ear rot (GER), and F. verticillioides, the causative agent of Fusarium ear rot (FER). GER causes contamination with deoxynivalenol (DON), nivalenol and zearalenone (ZEA), whereas FER causes contamination with fumonisins (FUM). Information on the resistance to GER and FER and mycotoxin contamination is lacking for maize adapted to the cooler climatic conditions of Central Europe. In this study we investigated (1) the resistance of early maturing European elite inbred lines against GER and FER and contamination of mycotoxins, (2) the genetic variances and heritabilities for ear rot ratings and mycotoxin concentrations, (3) the correlations of ear rot ratings with mycotoxin concentrations, (4) the correlations between line per se (LP) and testcross performance (TP) for GER rating and DON concentration, (5) the aggressiveness of and mycotoxins produced by different isolates of F. graminearum and F. verticillioides, and (6) the potential of near infrared spectroscopy (NIRS) to estimate concentrations of DON and FUM in maize grains under artificial inoculation. Significant genotypic variances and moderate to high heritabilities were found for GER, DON and ZEA among the inbred lines and for GER and DON among the testcrosses, as well as for FER and FUM among the inbred lines. Further, genotype x environment interaction variances were significant for all traits except FUM. Thus, the results underlined the presence of ample genotypic variation and the need to conduct multi-environment tests for reliable identification of resistant genotypes. Ear rot ratings and mycotoxin production of eight isolates each of F. graminearum and F. verticillioides differed significantly. Even though, isolate x inbred interactions were significant only in the case of F. graminearum, and no rank reversals occurred among the tested inbred lines. Most isolates differentiated the susceptible inbreds from the resistant ones for severity ratings. However, the differences between the two groups were smaller for the less aggressive isolates. Therefore, we recommend using a single, environmentally stable and sufficiently aggressive isolate for resistance screenings under artificial inoculation. Strong correlations between ear rot severity and mycotoxin concentrations indicated that selection for low ear rot severity under artificial inoculation will result in high correlated selection response for low mycotoxin concentration, particularly for GER and DON. Selection for ear rot severity is less resource-demanding and quicker than selection for mycotoxin concentration. Thus, it enables the breeder to maximize selection gain for a given budget. However, the selected elite material should be evaluated for mycotoxin concentrations in order to avoid ?false positives?. In this regard, NIRS showed high potential to predict DON concentrations in grain obtained from artificially inoculated maize. Compared to the commonly employed ELISA assay, NIRS assays are considerably cheaper, because no mycotoxin extractions and test kits are needed. We observed moderate positive correlations between GER and FER, and identified inbreds combining resistance to both ear rots. Therefore, selection for resistance to one pathogen is expected to result in indirect response to the other. Nevertheless, in advanced stages of each breeding cycle, lines preselected for other agronomically important traits should be evaluated for resistance to both pathogens. Genotypic variances for GER and DON were generally higher in LP than TP. Thus, assuming identical selection intensities for each scheme, the expected response to selection for LP should be higher than for TP. However, owing to moderate correlations between LP and TP for GER and DON, selection based on LP is not sufficient, because the ultimate goal is to develop resistant hybrids. Therefore, a multi-stage selection procedure is recommended with evaluation of agronomically promising lines for GER in only one environment in order to eliminate highly susceptible lines, followed by evaluation of TP of the selected lines for GER with one tester of moderate to high resistance level from the opposite heterotic pool in two to three environments.
Identification of essentially derived varieties in maize (Zea mays L.) using molecular markers, morphological traits, and heterosis
(2004) Heckenberger, Martin; Melchinger, Albrecht E.
The ‘breeder’s exemption’ as fixed in the UPOV convention on plant variety protec-tion allows the use of protected germplasm for the development of new plant varieties. The aim of this concept is the creation of new genetic variation to guarantee a continuous breeding progress. However, the use of molecular markers in backcrossing programs and genetic engineering has created the technical basis to develop new plant varieties without original breeding efforts. Therefore, the concept of ‘essential derivation’ was implemented into the 1991 Act of the UPOV convention to distinguish between varieties that resulted from intensive and creative selection programs and cultivars that were developed without major genetic changes from these former varieties. Accordingly, a variety is deemed to be essentially derived from an initial variety (IV), if it (i) was predominantly derived from the IV, (ii) is clearly distinguishable from the IV, and (iii) genetically conforms to the IV in the expression of it’s essential characteristics. The goal of this thesis was to evaluate and compare different approaches to assess conformity in the expression of the essential characteristics between IV and essentially derived varieties (EDVs) and to derive a theoretical and experimental basis for the devel-opment of thresholds to distinguish between independently derived varieties and EDVs in maize (Zea mays L.). The main focus was set on the evaluation of genetic distances based on ‘simple sequence repeats’ (SSRs) and ‘amplified fragment length poly¬morphisms’ (AFLPs) as well as the factors contributing to the GD between parental inbreds and their progeny lines. Furthermore, the ability of heterosis and morphological distances for identification of EDVs was examined. In detail, the objectives were to (1) analyze the factors influencing genetic distances (GD) based on SSRs and AFLPs between related maize inbred lines, (2) investigate the power of SSR- and AFLP-based GD estimates, morphological distances and heterosis for discriminating between progenies derived from F2, BC1, and BC2 populations, (3) exemplify theoretical and simulated results with experimental data, and (4) draw conclusions with regard to EDV thresholds suggested in the literature. A total of 220 flint, dent, and US maize inbred lines was genotyped with 100 SSRs equally distributed across the maize genome. The 220 lines comprised 163 triplets. A triplet consisted of one progeny and both parental lines, where the former was developed from an F2-, BC1-, or BC2 population. A subset of 58 lines (38 triplets) was genotyped addition-ally with 20 AFLP primer combinations. Furthermore, morphological traits and heterosis were observed for these 38 triplets in a field experiment over two years and three locations. The distributions of GD values for parental lines and their F2- and BC1-derived progeny overlapped for simulated as well as for experimen-tal data. Assuming that the derivation of a line from an F2 population was an accepted breed-ing procedure and the derivation from a BC1 population would not be accepted, we ob-served Type II errors (β) ranging from 0.23 to 0.37 depending on the germplasm pool for a given Type I error (α) of 0.05. For a threshold between BC1 and BC2, β ranged from 0.40 to 0.60 with an increasing tendency for higher BC levels. For fixed GD thresholds of T=0.25, 0.20, 0.15, and 0.10 suggested in the literature, substantial differences for α and β were found between different germplasm pools. Therefore, thresholds need to be gene pool specific and different thresholds for potential EDVs from intra-pool crosses than for progenies from inter-pool crosses must be applied. Discrimination of F2-, BC1-, and BC2-derived progeny lines on the basis of heterosis and morphological distances revealed β values ranging from 0.50 to 0.95 depending on the trait or combination of traits. Therefore, heterosis and morphological distances were fairly inappropriate tools for identification of EDVs due to the larger overlaps of F2-, BC1-, and BC2-distributions compared to GDs based on molecular markers. In general, SSRs and AFLPs were the most adequate tools to uncover close pedigree relationships between maize inbred lines and to discriminate among lines derived with ac-cepted or non-accepted breeding procedures. Therefore, the results presented in this study provide an example for identification of EDVs and can be transferred to other diploid crops by adjusting the corresponding thresholds.
Implementation and optimization of the doubled haploid technology for tropical maize (Zea mays L.) breeding programs
(2012) Prigge, Vanessa; Melchinger, Albrecht E.
Doubled haploid (DH) technology is currently the fastest way to achieve homozygosity in maize and it offers numerous quantitative genetic, operational, and economic advantages. Hybrid maize breeding with DH lines is common in temperate areas, yet adoption of this technology is still to be realized in tropical areas. Therefore, the main goal of my thesis project was to establish and validate the DH technology for tropical maize breeding programs at the International Maize and Wheat Improvement Center (CIMMYT) in Mexico. In vivo production of maternal haploids and DH lines involves four steps: (i) inducing haploidy by pollinating source germplasm with pollen of a haploid inducer; (ii) identifying seeds with haploid embryos based on a visually scorable marker; (iii) duplicating chromosomes of putative haploids by treating the seedlings with a mitotic inhibitor; and (iv) self-pollinating DH plants to multiply their seed. To impart knowledge on each of the above steps, we compiled a detailed protocol and produced a publicly available video which will be very useful for capacity building. Lack of reliable information on the performance of temperate inducers under nontemperate conditions is one reason for the slow adoption of DH technology in tropical maize breeding programs. Therefore, we assessed haploid induction rates (HIR) and agronomic performance of three temperate inducers in tropical lowland environments in Mexico. HIR obtained under tropical conditions were similar to those previously reported from evaluations under temperate conditions, indicating that temperate inducers can be used for initiation of DH breeding programs in the tropics. However, the inducers showed poor pollen production, poor seed set, and strong susceptibility to tropical leaf diseases. Hence, better adapted inducers would be advantageous for large-scale induction of haploidy in tropical DH programs. To develop better adapted haploid inducers, segregating populations were generated from crosses between temperate inducers and eight tropical CIMMYT maize lines (CML) from Mexico and Zimbabwe. Mass selection of individual F2 plants was conducted for visually scorable and highly heritable traits, followed by family-based selection for HIR and agronomic traits. Several tropical inducer candidates (TIC) were identified with HIR of up to 10% and notably improved agronomic performance under tropical lowland conditions. Compared to backcrosses to the inducers, backcrosses to the CML showed similar HIR combined with a significantly later anthesis date and improved plant vigor. Hence, backcrossing to the adapted parent may be a suitable approach to improve adaptation of new inducers while maintaining high HIR levels. Furthermore, we screened randomly chosen South American maize accessions and observed HIR of up to 3%, suggesting that novel sources of haploid induction ability may be present in CIMMYT?s vast germplasm collection. Although extensively exploited in DH line production, the genetic mechanisms underlying in vivo induction of maternal haploids in maize are still largely unknown. We conducted comparative quantitative trait locus (QTL) mapping for HIR to explore the genetic architecture of this phenomenon. Segregating populations were generated from four crosses composed of two temperate haploid inducer lines and three non-inducer lines. One major QTL on chromosome 1 (qhir1; bin 1.04) explaining up to 66% of the genotypic variance was detected in the three populations involving non-inducer lines. Hence, bin 1.04 represents an interesting region for map-based cloning. Further, qhir1 was affected by strong segregation distortion against the inducer allele, indicating that natural selection disfavors haploid induction ability. Seven QTL with smaller effects were detected in the CAUHOI×UH400 population. Further, we proposed a conceptual genetic framework for inheritance of in vivo haploid induction ability in maize. Common methods for artificial duplication of haploid chromosome sets mostly involve toxic and costly reagents and are extremely labor-intensive. This leads to serious bottlenecks during DH line development. When screening haploid populations derived from 260 diverse temperate and tropical source germplasm, we observed significant genetic variation for fertility-related traits, suggesting that haploid fertility can be effectively improved by recurrent selection. This may facilitate abolishment of artificial chromosome doubling during DH production, which seems particularly relevant for enabling small national maize breeding programs and seed companies in developing countries to adopt the DH technology. To study the suitability of different population types for DH line extraction, we developed 131 DH lines from five tropical elite single crosses (SC) and five tropical open-pollinated populations (OP) and evaluated them for testcross performance in Mexico. While testcross grain yield means of the two population types did not differ significantly, significant genetic variance was only revealed for OP-derived DH lines. Several DH lines from OP excelled in testcross performance and may be useful for tropical hybrid breeding programs. In addition, tropical OP may harbor valuable untapped genetic variation that can effectively be exploited with DH technology. This thesis work demonstrated that established protocols for in vivo DH line development can be readily applied to tropical maize breeding programs. Adoption of the DH technology promises to greatly increase the efficiency of breeding programs and DH lines are also an exciting tool to (i) immortalize genetic resources, (ii) conduct high-resolution genetic analyses of important traits, and (iii) accelerate the arrival of improved varieties to farmers? fields.
Rapid phenotyping of different maize varieties under drought stress by using thermal images
(2011) Müller, Joachim; Romano, Guiseppe; Zia, Shamaila; Spreer, Wolfram; Cairns, Jill; Araus, Jose Luis
The development of maize genotypes with high yields under drought is of pivotal relevance for the International Maize and Wheat Improvement Centre (CIMMYT). Thermal images of the canopy of different 92 maize genotypes were acquired in the time interval between anthesis and blister stage with each picture containing five plots of different genotypes. Mean temperature differences of more than 2°C between different genotypes under water stress were then detected using thermal images. Genotypes better adapted to drought exhibiting lower temperatures. A canopy thermal image is a potential promising method to accelerate the screening process and thereby enhance phenotyping for drought adaptation in maize.
Response to modified recurrent full-sib selection in two European F2 maize populations analyzed with quantitative genetic methods
(2006) Flachenecker, Christian; Melchinger, Albrecht E.
Many plant breeding strategies lead to a reduction in the genetic variance of the source population. However, a sufficient genetic variance is essential for the long-term selection response. Hence, the aim of recurrent selection (RS) is a continuous increase in the frequencies of favorable alleles while maintaining genetic variability in the population. Several intrapopulation RS methods have been proposed in maize: e.g., mass selection, half-sib selection, full-sib (FS) selection, S1 selection. Among them, recurrent FS selection is characterized by a short cycle length, complete parental control, and a high selection response. The goal of this thesis was to investigate the changes in the population structure over several cycles of a modified recurrent FS selection program in two European F2 maize populations. In detail, the objectives were to (i) monitor trends across selection cycles in the estimates of population mean, inbreeding coefficients, and variance components, (ii) determine selection response for per se and testcross performance, (iii) compare predicted with realized selection response, (iv) extend the population diallel analysis under full consideration of inbreeding depression due to random genetic drift, (v) separate genetic effects due to selection from those due to random genetic drift, and (vi) investigate the usefulness of best linear unbiased prediction (BLUP) estimates of parents for predicting progeny performance under the recurrent FS selection scheme applied. Four early maturing European flint inbreds were used as parents to produce two F2 populations (A×B and C×D). Both populations were three times intermated by chain crossing to reduce the gametic phase disequilibrium. Starting from the F2Syn3 population obtained in this manner, a modified recurrent FS selection program was conducted over four cycles in population A×B and over seven cycles in population C×D. In each cycle, 144 FS families were tested in field trials and, in parallel, six plants from each FS family were selfed. The selfed ears of the 36 families with the highest selection index (SI = 2 × dry matter content + grain yield) were selected and intermated according to a pseudo-factorial mating scheme. In this mating scheme, the gametic contribution of the best selected FS families is doubled compared with the gametic contribution of the remaining selected FS families. Afterwards, all cycles of both populations were tested in two population diallel analyses in six environments. Based on the known pedigree records, the inbreeding coefficient of each FS family and the coancestry coefficients among them were calculated. Variance components and BLUP values were obtained using phenotypic means and coancestry coefficients. For grain yield, the selection response per cycle, which could be expected after correcting for the effects of random genetic drift, was higher than reported in the literature (14.1% and 8.3% in populations A×B and C×D, respectively). We ascribe the comparatively high selection response mainly to the pseudo-factorial mating scheme. This mating scheme is expected to increase the selection response compared with commonly applied random mating schemes, without a major reduction in the effective population size (Ne). In this study, the expected Ne was 32, suggesting a minor influence of random genetic drift compared with that of selection. This assumption was verified by an extended population diallel analysis, showing that random genetic drift reduced the selection response only by about 1-2% in both populations. In contrast to an estimation of variance components with moment estimators, the REML procedure has no special requirements on the mating scheme and accounts for any relationship among families in a breeding population. As expected from the high Ne applied in our study, we observed only a moderate decrease in additive variance for grain yield and grain moisture in both populations. Nevertheless, the variance components were still associated with high standard errors, which prevented the revealing of trends across cycles. A larger number of test locations and larger population size would reduce the standard errors of variance components at the cost of oversized and expensive field trials. Methods for predicting the performance of progenies are important to optimize RS programs. Due to simplifying assumptions, a prediction with phenotypic means is often inaccurate. An alternative method is BLUP, which was suggested for predicting the performance of untested single-cross hybrids but has not been applied in RS programs. In our study, the prediction of progeny performance based on BLUP was only marginally better than prediction based on the phenotypic mean. However, higher degree of relationship between the entries and lower heritabilities would increase the advantage of BLUP compared with phenotypic means.
Strategies for selecting high-yielding and broadly adapted maize hybrids for the target environment in Eastern and Southern Africa
(2012) Windhausen, Sandra Vanessa; Melchinger, Albrecht E.
Maize is a major food crop in Africa and primarily grown by small-holder farmers under rain-fed conditions with low fertilizer input. Projections of decreasing precipitation and increasing fertilizer prices accentuate the need to provide farmers with maize varieties tolerant to random abiotic stress, especially drought and N deficiency. Genetic improvement for the target environment in Eastern and Southern Africa can be achieved by: (i) direct selection of grain yield in random abiotic stress environments, (ii) indirect selection for a secondary trait or grain yield in optimal, low-N and/or managed stress environments, or (iii) index selection using information from all test environments. At present, the maize hybrid testing programs of the International Maize and Wheat Improvement Center (CIMMYT) select primarily for grain yield under managed stress and optimal environments and subdivide the target environment according to geographic and climatic differences. It is not known to what extend the current strategy contributes to selection gains. The same holds true for genomic prediction, a strategy that is not yet implemented into the CIMMYT maize breeding program but that may accelerate breeding progress and reduce cycle length by predicting genotype performance based on molecular markers. Regarding the different strategies mentioned for selecting high-yielding and broadly adapted maize hybrids, the breeder needs to decide which of them are most promising to increase genetic gains. Consequently, the objectives of my thesis were to (1) evaluate the potential of leaf and canopy spectral reflectance as novel secondary traits to predict grain yield across different environments, (2) estimate to what extent indirect selection in managed drought and low-N stress environments is predictive of grain yield in random abiotic stress environments, (3) investigate whether subdividing the target environment into climate, altitude, geographic, yield level or country subregions is likely to increase rates of genetic gain, and (4) evaluate the prospects of genomic prediction in the presence of population structure. The measurement of spectral reflectance (495 ? 1853 nm) of both leaves and canopy at anthesis and milk grain stage explained less than 40% of the genetic variation in grain yield after validation. Consequently, selection based on predicted grain yield is only suitable for pre-screening, while final yield evaluation will still be necessary. Nevertheless, the prospect of developing inexpensive and easy to handle devices that can provide, at anthesis, precise estimates of final grain yield warrants further research. Based on a retrospective analysis across 9 years, more than 600 trials and 448 maize hybrids, it was shown that maize hybrids were broadly adapted to climate, altitude, geographic and country subregions in Eastern and Southern Africa. Consequently, I recommend that the maize breeding programs of CIMMYT in the region should be consolidated. Within the consolidated breeding programs, genotypes should be selected for performance in low- and high yielding environments as the genotype-by-yield level interaction variance was high relative to the genetic variance and genetic correlations between low- and high-yielding environments were moderate. Genetic gains were maximized by index selection, considering the yield-level effect as fixed and appropriately weighting information from all trials. To allow better allocation of resources, locations with high occurrence of random abiotic stress need to be identified. Heritability in trials conducted at these locations may be increased by the use of row- and column designs and/or spatial adjustment. Furthermore, resources invested into managed drought trials should be maintained during early breeding stages but shifted to the conduct of low-N trials at later breeding stages. Investments in a larger number of low-N trials may increase selection gain, because performance under low-N and random abiotic stress was highly correlated and genotypes can be easily selected under different levels of soil N. Prospects are promising to accelerate breeding cycles by the use of genomic prediction. Based on two large data sets on the performance of eight breeding populations, it was shown that prediction accuracy resulted primarily from differences in mean performance of these populations. Genomic prediction may be implemented into the CIMMYT maize breeding program to predict the performance of lines from a diversity panel, segregating lines from the same or related crosses, and progenies from closed populations within a recurrent selection program. The breeding scenarios in which genomic prediction is most promising still need to be defined. Generally, the construction of larger training sets with strong relationship to the validation set and a detailed analysis of the population structure within the training and validation sets are required. In conclusion, combining index and genomic selection is the most promising strategy for providing high-yielding and broadly adapted maize genotypes for the target environments in Eastern and Southern Africa.