Initial evaluation of genomic selection to improve wood property in Eucalyptus nitens breeding population
By Jaroslav Klápšte, Mari Suontama, Emily Telfer, Natalie Graham, Charlie Low, Toby Stovold, Russell McKinley, Heidi Dungey , June 2016.
Download SWP-T006 (pdf)
The E. nitens genetic improvement programs are predominantly based on open-pollinated progeny tests. This approach produces high levels of hidden relatedness whose ignorance causes upward bias in genetic parameters. Development of high-throughput genotyping technologies enables generation of large amounts of genomic markers that can provide information to construct matrices of realized genetic relationship. The advantage of marker based relationship matrices is to fill gaps in pairwise relatedness produced by shallow and simple pedigrees commonly present in forest tree genetic evaluations and thus reduce the standard errors of genetic parameters. This in turn results in more precise selection of valuable genotypes.
Our analysis found the marker based approach has improved the accuracy of genetic parameter estimates and also resulted in higher predictive accuracy in cross-validation evaluation. The likely source of improvement is the utilization of all the available information in the populations through complete pairwise relationship matrix compared to very sparse pedigree-based relationship matrix. This besides the faster progress in genetic improvement and delivery are a major benefits to the implementation of genomics in forest tree breeding when generally only shallow and simple pedigrees are available. The marker based approach found generally lower heritability estimates in Tinkers compared to Waiouru which is probably a consequence of a higher selection intensity applied in the Tinkers population compared to Waiouru which resulted in a fixation of part of the genetic variance. Surprisingly, in the pedigree based approach we found the opposite results in several traits such as a15, a16, a17, a39, a40, a41 and ht1 which is probably caused by the smaller sample size used to obtain reliable heritability estimates based on pedigree information. In addition, breeding values were less accurate in Tinkers compared to the Waiouru population. The across seed orchard heritability and breeding values accuracy estimates converted to intermediate values between both population estimates. Surprisingly, a larger sample size did not result in higher accuracy of genetic parameters. This could be a consequence of merging two populations with different selection histories.
We performed cross-validation at both an individual and family. The individual based cross- validation found that Tinkers population produced higher predictive ability compared to Waiouru population which is contrary to the results from heritability and theoretical accuracy estimates. The higher predictive accuracy in the Tinkers population can be explained by larger haploblocks which are built in populations created under higher selection intensity and thus the whole genetic complex can by efficiently captured even by a sparse marker array. The across population cross- validation produced again intermediate predictive accuracies between both populations (Waiouru and Tinkers) and an increase in training population sample size did not help to improve the estimates above the Tinkers population. Therefore, the decrease in effective number of genomic segments through building of larger haploblocks is more efficient than an increase in training population sample size in our population. The family based cross-validation relies purely on linkage disequilibrium between markers and QTLs which is the most stable part of genomic prediction. Generally, we can find higher predictive ability in Tinkers population which is related to the larger haploblocks from more intensive selection.
Generally, it is highly recommended to capture a large proportion of the genetic variability in training populations in order to build robust prediction models, making it important to keep a broad spectra of genetic material in training populations. Therefore, in genomics based breeding programs, the breeding arboretum should be established independently of the production population due to different requirements on genetic diversity vs. genetic gain trade-offs to utilize genomics at maximum efficiency.
No posts yet