Use of high-density SNP data to identify patterns of diversity and signatures of selection in broiler chickens.

The development of broiler chickens over the last 70 years has been accompanied by large phenotypic changes, so that the resulting genomic signatures of selection should be detectable by current statistical techniques with sufficiently dense genetic markers. Using two approaches, this study analysed high-density SNP data from a broiler chicken line to detect low-diversity genomic regions characteristic of past selection. Seven regions with zero diversity were identified across the genome. Most of these were very small and did not contain many genes. In addition, fifteen regions were identified with diversity increasing asymptotically from a low level. These regions were larger and thus generally included more genes. Several candidate genes for broiler traits were found within these 'regression regions', including IGF1, GPD2 and MTNR1AI. The results suggest that the identification of zero-diversity regions is too restrictive for characterizing regions under selection, but that regions showing patterns of diversity along the chromosome that are consistent with selective sweeps contain a number of genes that are functional candidates for involvement in broiler development. Many regions identified in this study overlap or are close to regions identified in layer chicken populations, possibly due to their shared precommercialization history or to shared selection pressures between broilers and layers.

Assessment of bagging GBLUP for whole-genome prediction of broiler chicken traits.

Bootstrap aggregation (bagging) is a resampling method known to produce more accurate predictions when predictors are unstable or when the number of markers is much larger than sample size, because of variance reduction capabilities. The purpose of this study was to compare genomic best linear unbiased prediction (GBLUP) with bootstrap aggregated sampling GBLUP (Bagged GBLUP, or BGBLUP) in terms of prediction accuracy. We used a 600 K Affymetrix platform with 1351 birds genotyped and phenotyped for three traits in broiler chickens; body weight, ultrasound measurement of breast muscle and hen house egg production. The predictive performance of GBLUP versus BGBLUP was evaluated in different scenarios consisting of including or excluding the TOP 20 markers from a standard genome-wide association study (GWAS) as fixed effects in the GBLUP model, and varying training sample sizes and allelic frequency bins. Predictive performance was assessed via five replications of a threefold cross-validation using the correlation between observed and predicted values, and prediction mean-squared error. GBLUP overfitted the training set data, and BGBLUP delivered a better predictive ability in testing sets. Treating the TOP 20 markers from the GWAS into the model as fixed effects improved prediction accuracy and added advantages to BGBLUP over GBLUP. The performance of GBLUP and BGBLUP at different allele frequency bins and training sample sizes was similar. In general, results of this study confirm that BGBLUP can be valuable for enhancing genome-enabled prediction of complex traits.

Dissection of additive genetic variability for quantitative traits in chickens using SNP markers.

The aim of this study was to separate marked additive genetic variability for three quantitative traits in chickens into components associated with classes of minor allele frequency (MAF), individual chromosomes and marker density using the genomewide complex trait analysis (GCTA) approach. Data were from 1351 chickens measured for body weight (BW), ultrasound of breast muscle (BM) and hen house egg production (HHP), each bird with 354 364 SNP genotypes. Estimates of variance components show that SNPs on commercially available genotyping chips marked a large amount of genetic variability for all three traits. The estimated proportion of total variation tagged by all autosomal SNPs was 0.30 (SE 0.04) for BW, 0.33 (SE 0.04) for BM, and 0.19 (SE 0.05) for HHP. We found that a substantial proportion of this variation was explained by low frequency variants (MAF <0.20) for BW and BM, and variants with MAF 0.10-0.30 for HHP. The marked genetic variance explained by each chromosome was linearly related to its length (R(2) = 0.60) for BW and BM. However, for HHP, there was no linear relationship between estimates of variance and length of the chromosome (R(2) = 0.01). Our results suggest that the contribution of SNPs to marked additive genetic variability is dependent on the allele frequency spectrum. For the sample of birds analysed, it was found that increasing marker density beyond 100K SNPs did not capture additional additive genetic variance.

Effect of allele frequencies, effect sizes and number of markers on prediction of quantitative traits in chickens.

The objective was to assess goodness of fit and predictive ability of subsets of single nucleotide polymorphism (SNP) markers constructed based on minor allele frequency (MAF), effect sizes and varying marker density. Target traits were body weight (BW), ultrasound measurement of breast muscle (BM) and hen house egg production (HHP) in broiler chickens. We used a 600 K Affymetrix platform with 1352 birds genotyped. The prediction method was genomic best linear unbiased prediction (GBLUP) with 354 564 single nucleotide polymorphisms (SNPs) used to derive a genomic relationship matrix (G). Predictive ability was assessed as the correlation between predicted genomic values and corrected phenotypes from a threefold cross-validation. Predictive ability was 0.27 ± 0.002 for BW, 0.33 ± 0.001 for BM and 0.20 ± 0.002 for HHP. For the three traits studied, predictive ability decreased when SNPs with a higher MAF were used to construct G. Selection of the 20% SNPs with the largest absolute effect sizes induced a predictive ability equal to that from fitting all markers together. When density of markers increased from 5 K to 20 K, predictive ability enhanced slightly. These results provide evidence that designing a low-density chip using low-frequency markers with large effect sizes may be useful for commercial usage.

Accuracy of genomic prediction using an evenly spaced, low-density single nucleotide polymorphism panel in broiler chickens.

One approach for cost-effective implementation of genomic selection is to genotype training individuals with a high-density (HD) panel and selection candidates with an evenly spaced, low-density (ELD) panel. The purpose of this study was to evaluate the extent to which the ELD approach reduces the accuracy of genomic estimated breeding values (GEBV) in a broiler line, in which 1,091 breeders from 3 generations were used for training and 160 progeny of the third generation for validation. All birds were genotyped with an Illumina Infinium platform HD panel that included 20,541 segregating markers. Two subsets of HD markers, with 377 (ELD-1) or 766 (ELD-2) markers, were selected as ELD panels. The ELD-1 panel was genotyped using KBiosciences KASPar SNP genotyping chemistry, whereas the ELD-2 panel was simulated by adding markers from the HD panel to the ELD-1 panel. The training data set was used for 2 traits: BW at 35 d on both sexes and hen house production (HHP) between wk 28 and 54. Methods Bayes-A, -B, -C and genomic best linear unbiased prediction were used to estimate HD-marker effects. Two scenarios were used: (1) the 160 progeny were ELD-genotyped, and (2) the 160 progeny and their dams (117 birds) were ELD-genotyped. The missing HD genotypes in ELD-genotyped birds were imputed by a Gibbs sampler, capitalizing on linkage within families. In scenario (1), the correlation of GEBV for BW (HHP) of the 160 progeny based on observed HD versus imputed genotypes was greater than 0.94 (0.98) with the ELD-1 panel and greater than 0.97 (0.99) with the ELD-2 panel. In scenario (2), the correlation of GEBV for BW (HHP) was greater than 0.92 (0.96) with the ELD-1 panel and greater than 0.95 (0.98) with the ELD-2 panel. Hence, in a pedigreed population, genomic selection can be implemented by genotyping selection candidates with about 400 ELD markers with less than 6% loss in accuracy. This leads to substantial savings in genotyping costs, with little sacrifice in accuracy.

Genetic analysis of sexual dimorphism of body weight in broilers.

Variation in sexual dimorphism (SD) is particularly marked in meat-type chickens. This paper investigates the genetic basis of SD in an important economic trait, i.e. body weight (BW) at 35 days of age, in broilers by applying quantitative genetic analysis. A large dataset comprising 203,323 BW records of a commercial line of broiler chicken was used. First, a bivariate approach was employed treating BW as a sex-specific trait. During this approach, seven bivariate models were applied and variances due to direct additive genetic, maternal genetic and maternal environmental effects were estimated via the restricted maximum likelihood method. The best-fitting model included direct additive genetic, maternal genetic and maternal environmental effects with a direct-maternal genetic covariance. Differences between male and female direct heritabilities were non-significant (0.28 vs. 0.29 for males and females, respectively), implying no need for sex-specific selection strategies. The direct-maternal genetic correlation was more strongly negative in males than in females (-0.72 vs. -0.56), implying a more profound antagonism between direct additive and maternal genetic effects in this particular gender. The direct genetic correlation of BW between the two sexes was as high as 0.91, i.e. only slightly lower than unity. Second, variance components and genetic parameters of two measures of SD, i.e. the weight difference (Δ) and the weight ratio (R), between the genders were estimated. Direct heritabilities for both measures were significantly different to 0 but of low magnitude (0.04). Apart from the additive-maternal covariance, no other random effects were found to be of importance for Δ and R. The results of the present study suggest that only minimal selection responses due to the selection of Δ and/or R and a small capacity for amplifying or reducing the BW differences between the sexes are to be expected in this specific population. Furthermore, selection pressure on BW is expected to amplify SD.

Optimal use of regression models in genome-wide association studies.

The performance of linear regression models in genome-wide association studies is influenced by how marker information is parameterized in the model. Considering the impact of parameterization is especially important when using information from multiple markers to test for association. Properties of the population, such as linkage disequilibrium (LD) and allele frequencies, will also affect the ability of a model to provide statistical support for an underlying quantitative trait locus (QTL). Thus, for a given location in the genome, the relationship between population properties and model parameterization is expected to influence the performance of the model in providing evidence for the position of a QTL. As LD and allele frequencies vary throughout the genome and between populations, understanding the relationship between these properties and model parameterization is of considerable importance in order to make optimal use of available genomic data. Here, we evaluate the performance of regression-based association models using genotype and haplotype information across the full spectrum of allele frequency and LD scenarios. Genetic marker data from 200 broiler chickens were used to simulate genomic conditions by selecting individual markers to act as surrogate QTL (sQTL) and then investigating the ability of surrounding markers to estimate sQTL genotypes and provide statistical support for their location. The LD and allele frequencies of markers and sQTL are shown to have a strong effect on the performance of models relative to one another. Our results provide an indication of the best choice of model parameterization given certain scenarios of marker and QTL LD and allele frequencies. We demonstrate a clear advantage of haplotype-based models, which account for phase uncertainty over other models tested, particularly for QTL with low minor allele frequencies. We show that the greatest advantage of haplotype models over single-marker models occurs when LD between markers and the causal locus is low. Under these situations, haplotype models have a greater accuracy of predicting the location of the QTL than other models tested.

The application of random regression models in the genetic analysis of monthly egg production in turkeys and a comparison with alternative longitudinal models.

Random regression models (RR) have become a popular methodology for the genetic study of longitudinal data since the last decade. The first objective of the current study was to investigate the application of RR models for the genetic analysis of egg production in turkeys. Data collected from a heavy dam line were used to estimate genetic parameters with 2 RR models, one having second-order Legendre polynomials as regression over time (RR2) and another with third-order polynomials (RR3). The second objective was to benchmark the performance of RR models with more conventional methods, so genetic parameters were reestimated using a multitrait (MT) and a repeatability model. To assess the model efficiency of predicting missing values, a reduced data set was used, and for each model, the predicted values of the deleted records were compared with the true values. The RR models were further compared against each other by eliminating the last period and estimating the MS error of the predictions for both models. The repeatability model had the poorest performance in predicting missing values. Heritability estimates from RR2 and MT models were close, whereas the RR3 model estimates were different. Both RR models demonstrated better prediction ability than the MT model. However, when RR models were compared solely, the RR2 model resulted in the smallest MS error. The results indicated that the RR3 model overfitted the data, suggesting that the choice of the appropriate polynomial order requires careful consideration. The present study illustrated that the application of RR models for the genetic analysis of egg production in turkeys is not only feasible but also offers a high accuracy of prediction.

Genetic parameters for a heavy female turkey line: impact of simultaneous selection for body weight and total egg number.

1. The objective of this study was to investigate the strength of the genetic association between growth and reproduction traits in turkeys selected for body weight, conformation and egg production. 2. Two distinct populations but derived from the same heavy turkey female line and situated in different locations (UK and USA), were used to estimate genetic parameters using multivariate REML for the following traits: body weight at 14 (BW14), 19 (BW19) and 24 (BW24) weeks of age and total egg number (EGG). 3. A Box-Cox transformation was applied to egg production data to reduce the impact of non-normality. 4. The heritability estimates for each trait for the UK and USA populations, respectively, were: BW14 0.37 and 0.48; BW19 0.34 and 0.43; BW24 0.28 and 0.43; EGG 0.22 and 0.34. 5. The genetic correlation between the body weight at all ages and the total egg production was strongly negative, reaching a value of -0.75 for the UK and -0.55 for the USA population. 6. The comparison of our results with published estimates in turkeys suggests that the genetic correlation may get stronger in magnitude following selection for increased body weight. 7. This could arise from fixation during selection of genes favouring larger weights but with minimal effect on egg production, leaving the segregating genetic variation dominated by pleiotropic loci with antagonistic effects on the traits. 8. Thus, in order to avoid continued selection for body weight reducing egg production to a point where natural selection offsets selection gains, alternative selection strategies should be considered.