Pedigree Effective Population Size Affects Genomic Prediction Accuracy in Katahdin Sheep

Abstract Katahdin sheep are a composite hair breed of medium-size, relatively prolific, with potential to exhibit resistance to gastrointestinal parasites. Recently, genomic predictions of breeding values were introduced in Katahdins. Prediction accuracy depends on the genetic variation present, and selection using genomically-enhanced breeding values may reduce diversity. We aimed to characterize the current genetic variability in Katahdins utilizing their extensive pedigree. The National Sheep Improvement Program provided the pedigree of 92,030 sheep born from 1984-2019. Two reference populations were defined for animals born between 2017-2019: all Katahdins (n = 23,494), and Katahdins with at least three generations of known Katahdin ancestry (n = 9,327). Using ENDOG and PopReport, effective population sizes (Ne) were estimated. We then considered the effect of Ne on the size of the genomic reference population needed to achieve certain levels of prediction accuracy. Weaning weight, a lowly heritable trait for this breed (h2 = 0.13), was evaluated. Based on individual increase in inbreeding, inbreeding coefficient of animal adjusted for equivalent complete generations, the estimated Ne for the full pedigree, reference 1, and reference 2 were 92, 104, and 87, respectively. When inbreeding coefficients were regressed or log-regressed on birth date and adjusted for generation interval, the Ne estimates were 53 and 49 for reference 1, and 76 and 67 for reference 2, respectively. For our average Ne = 75 with 3,960 effective chromosomal segments, the accuracy of prediction for weaning weight was estimated for genomic reference population sizes (N) of 10k, 15k, 20k, and 25k. Starting at N = 10k, the prediction accuracy was 0.58, and as N increased by 5k increments the accuracy increased to 0.65, 0.70, and 0.74. Conversely, with our smallest Ne = 49 with 2,548 effective chromosomal segments, the prediction accuracy was 0.64, 0.71, 0.75, and 0.78 for N = 10k, 15k, 20k, and 25k, respectively. Estimated Ne varied depending on the method utilized, and tended to be larger for reference 1 than 2. However, this was not the case for Ne based on regression. Inbreeding coefficients were greater in reference 2 than 1, and the generation interval was shorter in reference 2 (2.84 years) than in 1 (2.97 years). These combined differences led to larger Ne estimates for reference 2 than 1, reflecting the impact of different criteria used to obtain this statistic. The generally small, estimated Ne benefitted genomic prediction due to a greater shared number of effective chromosomal segments. Investing in larger genomic reference populations will only marginally increase prediction accuracy even for lowly heritable traits. Selection using genomically-enhanced breeding values will accelerate genetic gains, potentially reducing the Ne of Katahdins and thereby increasing prediction accuracy. Yet, long-term gains may be at risk due to the loss of genetic diversity.