Gene editing in breeding

Hyper-recombination as a strategy to extend genomic prediction accuracy across generations

---

Abstract

Linkage disequilibrium (LD) between markers and quantitative trait loci (QTL) enables genomic prediction (GP). However, recombination decays LD between markers and QTL, increasing genetic variation while lowering prediction accuracy in future generations. Increases in meiotic recombination rates offer new possibilities to exploit LD between tightly linked markers and QTL. Inducing hyper-recombination (HR) before model training could eliminate loosely linked marker-QTL without disrupting LD amongst tightly linked marker-QTL. If successful, GP models may be useful for more generations because tightly linked marker-QTL associations would be unlikely to decay under normal recombination (NR) and accuracy of selection would be maintained for more generations. Biotechnology tools aimed at increasing meiotic recombination are currently being researched and developed at The University of Queensland. Speed breeding, a plant breeding technique reducing generation intervals, may be applied to livestock embryos with these HR biotechnology tools in the future. Thus, prediction models would need to last many generations because new phenotypes would not be generated. We simulated whether inducing HR prior to training GP models could improve model longevity. We simulated 200 founders using AlphaSimR's "CATTLE" specification, followed by 10 generations of random mating to increase population size to 1853 individuals (heritability=0.5, 35k markers/chromosome, 30 chromosomes, 150 QTL/chromosome). For NR, we performed 10 to 15 generations of truncation selection (60% retention) on true breeding values (TBV) as burn-in to add historical inbreeding, LD, and selection signatures, using the 10th generation as the training population. The HR population underwent identical selection after multiplying the genetic map by 2 and 1000 to induce HR. After minor allele frequency filtering, we randomly sampled non-QTL marker sets (5k, 10k, 25k, 50k, 100k, 250k, and 500k) and trained BayesCπ models for both populations. We assessed true accuracy as the correlation between estimated breeding values (EBV) and TBV, heritability estimates against true heritability, and π estimates on the training population. The HR genetic maps were reset, and all populations underwent 10 generations of random mating to evaluate accuracy decay utilizing the training population predictions. The HR population showed lower accuracy than NR in the training population, heritability estimates closer to true heritability, lower π estimates, and greater marker variance. Prediction accuracy decayed at approximately the same rate in both populations, stabilizing 3-4 generations out. By generation 10, accuracy decreased 0.30 to 0.40 in both populations. Despite using many markers, limited marker-QTL LD may exist in the HR population due to extreme HR, or BayesCπ may be utilizing the wrong markers because of insufficient training population size. Plotting LD decay and marker-QTL effects along chromosomes will help resolve this. Further exploration of training population size, less aggressive HR, and marker density is needed to determine if HR is a viable strategy.

Keywords: 2026

How to Cite:

Dinglasan, E., Hayes, B., Hickey, L., Massel, K., Papin, V., Shaffer, W. & Villiers, K., (2026) “Hyper-recombination as a strategy to extend genomic prediction accuracy across generations”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286517. doi: https://doi.org/10.31274/wcgalp.24037

Rights: 1