Improving genetic predictions by accounting for Mendelian sampling or inbreeding depression

Abstract

Current genetic evaluation procedures (Best Linear Unbiased Prediction (BLUP) animal model) employed in the livestock industries throughout the world provide accurate predictions of breeding values. However, accuracy of predictions can still be enhanced. For example, BLUP methodology uses genetic relationships computed from pedigree information to allow for optimal utilization of all sources of information (performance of the individual and its relatives). Each source of information is weighted differently depending on the proportion of genes "shared" in common between relatives. However, these relationships among collateral relatives assume an average covariance. This averaging can sometimes limit the accuracy of prediction. Joint use of pedigree and currently available genetic (DNA) marker information can provide relationships that account for the variation in the covariance among collateral relatives due to Mendelian sampling. Furthermore, most current evaluation procedures do not account for decline in performance associated with increasing levels of inbreeding (inbreeding depression). The effective population sizes in many livestock populations are small and often in decline, causing increased inbreeding. In view of these opportunities for enhancement of accuracy of prediction, two independent investigations were conducted in this study. For convenience, the two investigations were categorized as the inbreeding study and the genetic relationship study. The primary objectives of the inbreeding study were to 1) assess the level of inbreeding, and 2) quantify inbreeding depression on early female reproduction, long-term cow productivity and carcass traits in cattle registered with the Red Angus Association of America. Inbreeding coefficients were computed using 829,882 pedigree records of animals born between 1930 and 2001. Most animals (89%) had both parents known. The inbreeding trend was evaluated over the period 1960 to 2001. Effects on performance of inbreeding of the individual (Fd) and that of its dam (Fm) were estimated. Reproductive traits included: heifer calving success (HCS, n = 1,197); heifer calving ease (HCE, n = 636); and cow stayability (CS, n = 14,268). Carcass traits were: hot carcass weight (HCW, n = 951); rib-eye area (REA, n = 947); backfat thickness (BFT, n = 767); and marbling score (MRB, n = 947). Individual inbreeding coefficients ranged from 0 to 51% with an average inbreeding level of 3.1% for the entire population. Considering the period 1992 to 2001, where pedigrees were more complete, the rate of inbreeding was 0.08%±0.002 per year. This rate of inbreeding indicates that inbreeding is accumulating at a minimal rate in the registered Red Angus population. However, the fact that individuals exist with inbreeding coefficients as high as 51% is suggestive of occasional deliberate or inadvertent intense inbreeding. The partial regressions of female performance on Fd expressed on a probability scale were 0.75, -1.44, and -0.56 %/% for HCR, HCE and CS, respectively. Corresponding partial regressions of female performance on Fm were -0.40, 0.20, and 0.12 %/%. All partial regression coefficients were nonsignificant (P > 0.10) except for CS on Fd (P = 0.0007) indicating that a percentage increase in inbreeding coefficient of the cow was associated with a 0.56% decline in the probability of the cow to maintain production until or beyond six years of age. Partial regressions of carcass traits on Fd were -0.71 kg/%, -0.04 cm2/%, 0.0005 cm/%, and -0.011 per % for HCW, REA, BFT, and MRB, respectively. Corresponding values for Fm were -0.45 kg/%, -0.12 cm2/%, -0.0012 cm/%, and -0.003 per %. The only significant partial regression coefficients were those of HCW on Fd (P = 0.0238) and REA on Fm (P = 0.0111). Results from this study provided little evidence of unfavorable relationships between inbreeding and performance in the registered Red Angus population at least for the current levels of inbreeding. The objectives of the relationship study were to 1) derive weighting factors for different sources of information available for predicting genetic merit of an individual, and 2) evaluate the change in accuracy of genetic prediction when the inverse of the marker-based numerator relationship matrix (AM) was substituted for the inverse of the standard (pedigree) numerator relationship matrix (AP) in the mixed model equations. The data used to achieve the first objective comprised of daughter yield deviation (DYD) records on one grandsire and half-sib sons. The grandsire and his sons had several hundred daughters with yield deviation records. The DYD is an average of the daughters' yields adjusted for fixed and non-genetic random effects of the daughters and genetic effects of the dam. Predicted transmitting abilities (PTA) were obtained using a sire model. The PTA for each son was expressed as a linear function of his DYD if available and PTA of his relatives. The weights for a half-sib bull using Apare always zero except for his DYD and the grandsire's PTA indicating that only its daughters' records (adjusted for the merit of the mate) and parent contribute directly to the evaluation of the individual. When AP was replaced with AM, the evaluation of a half-sib bull had non-zero weights for all half-sibs with the weight on each half-sib varying with the proportion of alleles shared in common indicating that half- and full-sibs can also contribute directly to the evaluation of the individual. This result demonstrate that the genetic merit of a bull with no daughters could be more reliably predicted using AM rather than AP if he had half-sibs with daughters because the inferiority or superiority of his Mendelian sampling could be assessed to some extent. The data set used to achieve the second objective comprised of records on 1,811 progeny-tested Holstein bulls. Each record comprised DYD for milk, fat and protein yield and genotypic information on 52 microsatellite markers. The markers were located in interesting quantitative trait loci (QTL) regions on six chromosomes. Three sets of analyses were conducted to obtain breeding values. The first set of breeding values (EBV-ALL) was obtained using all sources of information (e.g. own DYD and those of all relatives) incorporating the inverse of AP in the mixed model equations. The second set of breeding values (EBV-PED) was computed as in the first analysis except that the sire's own DYD (but not those of its relatives) was excluded when predicting its breeding value. The third set of breeding values (EBV-MRK) was computed as in the second analysis except that the inverse of AM was substituted for the inverse of AP. Correlations of EBVs and of their ranks were computed between EBV-ALL and EBV-PED or EBVMRK to evaluate the change in accuracy and ranking of sires when AP was replaced by AM. Considering all sires without sons in the data set (n = 849), the accuracy of prediction increased by 4.3% for milk yield but did not change for fat and protein yields when AP was replaced by AM (computed across chromosomes). Considering AM computed within chromosomes, the use of marker information resulted in at least an improvement in accuracy of prediction for milk and protein yield and sometimes a decline in accuracy for fat yield. These results suggest that different AM may be required for different traits. The rank order correlations were consistently higher across traits when AM was used, suggesting that use of markers provide better ranking of sires compared to use of pedigree information only. Results from the current study demonstrated that marker information could be used successfully to enhance the accuracy of genetic prediction in routine genetic evaluation particularly for young animals without their own performance or progeny information. Genetic evaluation has evolved over time as a result of research focusing on different aspects of the prediction process. Results from this study suggest that little will be gained in accuracy of genetic prediction by accounting for inbreeding depression in genetic evaluation at least for the ranges of inbreeding investigated except for CS, HCW, and REA. However, the relationship study demonstrated that using genetic marker information could enhance accuracy of genetic prediction.