Computational protocol: The genetics of feed conversion efficiency traits in a commercial broiler line

Similar protocols

Protocol publication

[…] Extraction of genomic DNA was performed using Qiagen 96-well extraction kit (Qiagen, Hilden, Germany). A total of 864 DNA samples were genotyped employing chicken 60 K SNP Beadchips (Illumina, San Diego, CA, USA) and analyzed by the Illumina GenomeStudio Genotyping Module (v1.9.4). Three and two samples were excluded from analysis due to missing phenotypic data and sample call rates <0.99, respectively. In total, 44355 SNP markers were further examined after meeting the following selection criteria: (i) SNP call rate >95%, (ii) minor allele frequency ≥3%, and (iii) Chi2-test for deviation of Hardy-Weinberg equilibrium of p ≥ 0.0001. Missing genotypes in the filtered dataset were imputed using fastPHASE (v1.2) with 10 random starts and 50 iterations of the EM algorithm. All SNP with no assigned chromosome or linkage group based on the current chicken (Gallus gallus) genome assembly (Galgal4) were removed and led to 44035 mapped SNP being used for association and linkage analyses. Population structure and unequal genetic distances within an analyzed population could affect the results of a GWAS and are major sources for false positive associations. To evaluate whether the relationship between used individuals has to be considered for GWAS, the population stratification was tested employing the SNPRelate R package. Thereby, multidimensional scaling analysis was performed using identity-by-state distances between individuals. [...] Linkage disequilibrium (LD) (r2) between mapped SNP markers was calculated for each chromosome employing Haploview (v4.2). The output dataset was used to calculate average distances and r2 of adjacent markers. Additionally, the LD of markers included in non-overlapping sliding 1-kb windows were averaged leading to 30,000 windows covering marker distances from 0 to 30,000 kb with corresponding r2 values. LD values of the 1 kb windows were plotted against physical distances and nonlinear regression curves were generated by fitting a four-parameter Weibull function (type-1) using R software with ‘drc’ package (v2.5–12; https://cran.r-project.org/web/packages/drc/index.html). Based on the curves r2 values at distances of 10 kb, 100 kb, 500 kb, 1 Mb, and 10 Mb were estimated and compared between macro-, intermediate, and micro-chromosomes. The extent of LD was defined as ‘useful’ with r2 ≥ 0.25. Markers that exceed this threshold in the pairwise linkage analysis were assumed to share evidence in association analyses and were considered as useful for QTL mapping. [...] Genome-wide association analysis using single marker information was performed for BW36, BW49, FI, BWG, and FCR using generalized linear models (GLM) implemented in JMP Genomics 6 (SAS Institute, Cary, NC, USA). The following statistical model was used: where yj is the observation of the body weight and FCE traits; μ is the overall population mean for each trait; mj is the fixed effect of j-th numeric genotypes; βW is linear effect of BW39 as covariate, which is considered in the analyses of FI and BWG to account for differences in body weight, and ej is the random residual error. Bonferroni adjustment of the genome-wide significance threshold was based on the effective number of independent tests, to account for LD between SNP markers. The number of independent tests was estimated using simpleM. Therefore, the principal component parameter was set to account for 99.5% of the variance and resulted in the consideration of 19420 independent tests. Corresponding threshold p-values were set to -log10(p-value) = 4.3 (1/19420) and -log10(p-value) = 5.6 (0.05/19420) for suggestive and genome-wide significance, respectively. Manhattan plots were created using the qqman R script. […]

Pipeline specifications

Software tools GenomeStudio, fastPHASE, SNPRelate, Haploview, JMP Genomics, qqman
Organisms Gallus gallus
Chemicals Glycerides