Artificial selection has increased the weights at which cattle are marketed either at weaning or yearling ages (Figures
1, Additional file
1: Figures S4, and S13) while simultaneously decreasing the incidence of assisted births (Additional file
1: Figures S3 and S9), and the trends observed in our data set are very similar to those reported for the entire Angus breed
. Larger birth weights and yearling heights are both strongly associated with increased calving difficulty and genetic trend increased both traits until about the mid-1980s, after which both began to decrease (Figures
1 and Additional file
1: Figures S5). Breeders did not directly select to increase birth weight, but it increased as a correlated response to selection for increased weaning and yearling weights. Some breeders selected for increased yearling height to produce Angus cattle more comparable in frame size to the Continental European breeds, which were imported into the US during the 1970s
. However, once breeders appreciated the undesirable correlated response in calving ease, selection was practised to increase weaning and yearling weights while maintaining birth weight and yearling height constant.
Using EMMAX, only eleven loci were found to be significantly associated with birth date and thus under strong selection, but all loci simultaneously explained 53% of the variation in birth date. On the other hand, BayesCπ estimated that 2.11% of the SNPs were strongly associated with birth date, but all SNPs explained 72% of the variance in birth date. The difference in the heritability estimates between the GBLUP or EMMAX analyses compared to the BayesCπ analysis reflects the different model assumptions underlying these analyses. Whereas GBLUP and EMMAX assume the infinitesimal model under which all SNP ASEs are drawn from a distribution with constant variance
[33, 34], BayesCπ begins with a distribution with constant variance but shrinks the variance for small effect SNPs that are rarely fit in the model. As a consequence, GBLUP and EMMAX regress all SNP ASEs equally towards the mean of zero, while BayesCπ more aggressively regresses small effect SNPs and less aggressively regresses large effect SNPs, leading to a better model fit—as was found here—when there are loci under very strong selection. In the absence of selection, genotype frequency should be independent of time provided that the effects of drift are negligible and the heritability estimates should be close to zero. However, this was not the case for US registered Angus cattle and we conclude that a significant number of loci are rapidly responding to selection (our results suggest 2.11%) and that most of the genome (82.4% from the drift analysis) is responding more slowly to selection. Furthermore, in our nonlinear BayesCπ model, 72% of the variation in birth date could be explained by simultaneously using all SNP genotypes, suggesting that there are loci under very strong selection (large effect loci) that are not appropriately fit by the infinitesimal model.
The SNP ASEs for the 16 analysed traits indicate that, with the exception of the two large effect QTLs on BTA 7 and 20, the vast majority of QTLs underlying quantitative traits in beef cattle are of small effect. Of considerable interest, neither of these QTLs was found to be under very strong selection and this seems to be because of their large antagonistic pleiotropic effects on growth and calving difficulty. We postulate that when multiple traits are simultaneously selected, the genetic architecture of the population defined by pleiotropy and the chromosomal organization of QTL alleles (phase effects) constrains both the phenotypic and genotypic response to selection.
For selection to be effective, the selection intensity and effective population size must be sufficiently large to overcome the effects of genetic drift. We demonstrate that US registered Angus cattle have a sufficiently large effective population size to enable successful artificial selection, but more importantly, that large intergenerational changes in allele frequency are unlikely to occur due to drift alone. Furthermore, we found a considerable disparity between pedigree and genomic estimates of inbreeding coefficients. While others have argued that genomic relationship matrices should be adjusted to more closely resemble pedigree relationship matrices
, we assert that genomic relationship matrices provide a more accurate representation of the realized relationships among individuals that result from the Mendelian sampling of parental gametes and selection. The use of genomic relationship matrices in place of pedigree relationship matrices avoids the assumption of neutrality of loci both in the estimation of inbreeding coefficients and for the mean value of gametes inherited by progeny—both of which are assumed for the computation of the numerator relationship matrix
. The disagreement between genomic and pedigree estimates of F coefficients is likely to be due to the assumption that base animals are not inbred, errors in the pedigrees, and missing pedigree information likely due to the large-scale importation of Canadian Angus cattle in the 1940s and 1950s that were not carriers of dwarfism alleles, which had been driven to high frequency due to selection at the time
. This is supported by the closer agreement between pedigree and genomic F coefficients for the Wye herd that was largely derived from British stock with more complete pedigree records than the remaining US registered Angus cattle (Table
2 and Figure
We attempted to identify the relative selection intensities placed on each selected trait via the imprints that multi-trait selection had left on the Angus genome. Although this analysis assumed no change in relative selection intensities in time, an assumption that is clearly violated in view of the genetic trends in birth weight and yearling height, we were able to confirm that growth traits have historically been the most strongly selected in US registered Angus cattle. Because Angus is considered to be a maternal breed (i.e., motherly, used as dams in commercial beef production), it is logical that loci that influence calving ease, growth to weaning, and milking ability should have been found to be under selection. Angus breeders have successfully selected to increase calving ease and body weight by selecting for body shapes that allow a calf’s easy passage through the dam’s pelvis. This is supported by the finding of an enrichment of gene ontology terms related to limb morphogenesis and anatomical structure development within regions of the genome detected as responding to selection. It has previously been shown that calving ease is negatively correlated with several body measures, such as head circumference, head width, hip width, hip height, chest girth, and cannon bone circumference
[38–40]. Likely due to the Certified Angus Beef’s emphasis on quality grade
, Angus breeders have more recently selected to increase marbling. Conversely, traits such as fat thickness, docility, and heifer pregnancy rate have not been as intensely selected as growth traits, due to the differing breeding objectives of beef producers, genetic antagonisms constraining selection response, and the historic difficulty in collecting field data to allow the development of EPDs for these traits.
There is also evidence that natural selection has occurred in this population. The gene ontology enrichment results indicate that genes affecting immune response, such as the MHC, NOD2, C3, and DBH, have strongly responded to selection (Additional file
2, Additional file
3, and Additional file
4), presumably due to the exposure of Angus cattle to novel pathogens following their introduction into the US in 1873
 and the continuous co-evolutionary “arms race” between bovine and pathogen genomes
[43, 44]. The Bovine HapMap Consortium
 found that the MHC had some of the lowest Fst values in the entire genome when compared between breeds. Our analyses have identified the MHC as being under strong selection. Taken together, these results suggest that the MHC or certain of the numerous olfactory receptors which occupy the same region on chromosome 23 are under strong convergent selection.
Furthermore, natural selection may also be attempting to buffer the cellular environment against the deleterious effects of inbreeding. We found that spermatogenesis was an enriched ontology term describing the function of genes within the strongly selected regions of the genome (Additional file
3). Seminal plasma proteins have been associated with bull fertility
, and selection may be increasing fertility to counter act the inbreeding depression of reproduction (alternatively, the use of AI may be selecting for increased fertility). Genes involved in response to oxidative stress were also identified; response to oxidative stress has been tied to mitigating inbreeding depression
. We also inferred that at least 6 heat shock proteins are under selection in Angus and that protein folding was an enriched biological process (Additional file
4). It has been hypothesized that heat shock proteins assist the organism to cope with protein instability and misfolding caused by homozygous nonsynonymous mutations that are elevated in frequency by inbreeding
One of the greatest difficulties encountered in identifying genomic signatures of selection is in distinguishing changes that have occurred due to demographic as opposed to selective forces
. Our Birth Date Selection Mapping approach utilizing mixed models specifically accounts for pedigree relationships and explicitly deconvolutes their confounding effects on time-dependent allele frequency changes, which are due to the fact that not all pedigrees are sampled equally deeply in terms of the numbers of genotyped individuals. Rather than fit generations as the dependent variable
, which are poorly estimated when pedigrees are incomplete, fitting birth date allows unknown or complex pedigrees with overlapping generations to be analysed. Furthermore, the genomic relationship matrix accounts for pedigree relationships between samples, allowing closely related samples to be analysed. However, one of the limitations of Birth Date Selection Mapping is the requirement of a temporally stratified sample of genotyped individuals. The results for the analyses of the reduced data subsets suggest that sampling over extended time periods or large sample sizes—but not necessarily both—will be necessary to identify strongly selected loci. This will currently limit the utility of the approach in human populations due to a lack of preserved DNA samples across multiple generations. However, this limitation may be alleviated as it becomes more practical to extract quality DNA from formalin-fixed, paraffin-embedded tissue section samples and ancient remains. Nevertheless, Birth Date Selection Mapping is clearly most easily applied to organisms with temporally stratified DNA samples and genome-wide genotypes.
Using the estimated birth date ASEs as informative priors in the development of genomic selection programs
 is another interesting application of our method. Loci with small birth date ASEs are either of small effect on the selection objective or represent genes of large effect that have undesirable pleiotropic effects (or closely linked loci with antagonistic phase relationships). Loci that have large birth date ASEs must have large effects on the selection objective that are less constrained by antagonistic pleiotropic effects allowing them to more rapidly respond to selection.