Predicting the performance of untested single crosses is an important objective in hybrid breeding programs. The effectiveness of best linear unbiased prediction (BLUP) of grain yield, moisture, stalk lodging, and root lodging of untested maize (Zea mays L.) single crosses was investigated. Multilocation data, from 1990 to 1995, for 4775 single crosses were obtained from the hybrid testing program of Limagrain Genetics. For each of 16 heterotic patterns, the performance of m untested single crosses was predicted from the trait phenotypes (T-BLUP) of n tested single crosses as YM = CMP Cpp−1 YP, where YM = m × 1 vector of predicted performance of untested single crosses; CMP = m × n matrix of genetic covariances between untested single crosses and tested single crosses; Cpp = n × n phenotypic covariance matrix among tested single crosses; and yp = n × 1 vector of average performance of tested single crosses, corrected for yield trial effects. For one heterotic pattern with available parental RFLP data, performance also was predicted from (i) marker genotypes (M-BLUP) at loci with significant general combining ability effects and (ii) both trait phenotypes and marker genotypes (TM-BLUP). Correlations between predicted and observed performance were obtained by a delete-one cross-validation procedure. Across heterotic patterns, the correlations for T-BLUP ranged from 0.463 to 0.770 for yield, 0.868 to 0.936 for moisture, 0.466 to 0.685 for stalk lodging, and 0.164 to 0.518 for root lodging. Out of 74 RFLP loci, 9 had significant (P ≤ 0.1) effects on yield, 14 on moisture, and 22 on stalk lodging. Correlations between predicted and observed yield were 0.764 for T-BLUP, 0.765 for TM-BLUP, and 0.341 for M-BLUP. Simplifying assumptions regarding marker effects and variances were used in TM-BLUP and M-BLUP, and whether a more complex model would lead to better TM-BLUP and M-BLUP predictions is yet to be determined. The results indicated that BLUP based on trait data alone is useful for routine identification of superior single crosses prior to field testing.
Commercial maize breeders routinely evaluate several hundred single crosses in their respective breeding programs each year. Typically, <1% of the tested single crosses eventually become commercial hybrids (Hallauer, 1990). Heterosis is the basis for the development of single-cross cultivars, but prediction of heterosis itself is not an important objective in commercial maize breeding programs.
Rather, breeders are interested in predicting single-cross performance prior to making the actual crosses and evaluating them in the field.
Best linear unbiased prediction based on trait phenotypes (T-BLUP) has been found useful for selection of untested maize single crosses. Correlations between predicted and observed performance, when the number of tested single crosses was > 100, have ranged from 0.426 to 0.762 for yield, 0.754 to 0.933 for moisture, 0.300 to 0.739 for stalk lodging, and 0.l64 to 0.532 for root lodging (Bernardo, 1996). An attractive feature of T-BLUP is that the predictions are made from data routinely generated in yield trials. Restriction fragment length polymorphism (RFLP) marker data on the parental inbreds of single crosses have become increasingly available in commercial maize breeding programs. If single cross trait data and parental marker data are available, BLUP may be useful for detecting markers with significant general combining ability (GCA) effects for different traits. Single-cross performance then may be predicted from (i) marker genotypes (M-BLUP) at loci with significant GCA effects or (ii) both trait phenotypes and marker genotypes (TM-BLUP). But the usefulness ofT-BLUP, TMBLUP, and M-BLUP has not been compared.
My objectives are to: (i) present results on the use of T -BLUP for predicting yield, moisture, stalk lodging, and root lodging with data sets typically encountered in a commercial maize hybrid breeding program; (ii) describe how BLUP may be used for routine detection of marker-trait associations; and (iii) compare the effectiveness of T-BLUP, TM-BLUP, and M-BLUP in predicting single-cross performance.