The authors have declared that no competing interests exist.
Conceived and designed the experiments: XL GS DS. Performed the experiments: TL HQ. Analyzed the data: TL HQ CL XL DS MSL GS. Wrote the paper: TL HQ CL XL DS MSL GS.
Newcastle disease (ND) and avian influenza (AI) are the most feared diseases in the poultry industry worldwide. They can cause flock mortality up to 100%, resulting in a catastrophic economic loss. This is the first study to investigate the feasibility of genomic selection for antibody response to Newcastle disease virus (Ab-NDV) and antibody response to Avian Influenza virus (Ab-AIV) in chickens. The data were collected from a crossbred population. Breeding values for Ab-NDV and Ab-AIV were estimated using a pedigree-based best linear unbiased prediction model (BLUP) and a genomic best linear unbiased prediction model (GBLUP). Single-trait and multiple-trait analyses were implemented. According to the analysis using the pedigree-based model, the heritability for Ab-NDV estimated from the single-trait and multiple-trait models was 0.478 and 0.487, respectively. The heritability for Ab-AIV estimated from the two models was 0.301 and 0.291, respectively. The estimated genetic correlation between the two traits was 0.438. A four-fold cross-validation was used to assess the accuracy of the estimated breeding values (EBV) in the two validation scenarios. In the family sample scenario each half-sib family is randomly allocated to one of four subsets and in the random sample scenario the individuals are randomly divided into four subsets. In the family sample scenario, compared with the pedigree-based model, the accuracy of the genomic prediction increased from 0.086 to 0.237 for Ab-NDV and from 0.080 to 0.347 for Ab-AIV. In the random sample scenario, the accuracy was improved from 0.389 to 0.427 for Ab-NDV and from 0.281 to 0.367 for Ab-AIV. The multiple-trait GBLUP model led to a slightly higher accuracy of genomic prediction for both traits. These results indicate that genomic selection for antibody response to ND and AI in chickens is promising.
Newcastle disease (ND) and avian influenza (AI) are regarded as two of the most important diseases of poultry worldwide and can lead to a flock mortality up to 100%
Previous genomic studies mainly focused on single-trait analysis; however, several traits of economic importance, which may be genetically related, were usually selected for animal breeding. Using the information from correlated traits, multiple-trait analysis could improve the accuracy of the estimated breeding value
This study was approved by the Animal Care Committee of the Institute of Animal Science, Guangdong Academy of Agricultural Sciences (Guangzhou, People's Republic of China), the Approval No. is GAAS-IAS-2009-73. Blood samples of birds were collected from the brachial vein by standard venipuncture procedure. The chickens were treated humanely, and none of them were sacrificed for this study.
The chicken population in this study originated from a cross between two divergently selected lines, i.e., the “High Quality chicken Line A” (HQLA) and the Huiyang Beard chicken (HB), as described in Sheng et al
Chickens were vaccinated with commercial vaccines in accordance with the instructions. At 25 days of age, the chickens were vaccinated against Newcastle disease with a commercially available LaSota strain (Intervet International B.V., Boxmeer, Netherlands) using the eye drop technique. After 25 days, a second vaccination against Newcastle disease was implemented. At 40 days of age, the chickens were vaccinated with a commercially Avian Influenza Inactivated H9 strain Vaccine using the eye drop technique. At 91 days of age, blood samples were collected. NDV and AIV antibody levels in the blood samples were measured using indirect ELISA and were expressed as the S/P value of the corresponding dilutions, according to the instructions for the commercial ELISA kit (BioCheck, Inc., Foster City, CA, USA).
Trait |
N | Mean | SD | Min value | Max value |
Ab-NDV | 511 | 3.63 | 1.57 | 0.67 | 9.10 |
Ab-AIV | 511 | 1.31 | 1.18 | 0.09 | 11.02 |
Ab-NDV = antibody response to Newcastle disease virus; Ab-AIV = antibody response to Avian Influenza virus.
A set of λ was investigated. Finally, parameter λ was set as 0.05 and 0.5 for Ab-NDV and Ab-AIV, respectively. After transformation, the two traits met normal distribution and the analysis was based on the transformed data.
The birds were genotyped using the Illumina Chicken 60K SNP Beadchip
In this study, breeding values were estimated using two models, i.e., a pedigree-based best linear unbiased prediction model (BLUP) and a genomic best linear unbiased prediction model (GBLUP). Both single-trait and multiple-trait analysis were implemented. Hereafter single-trait analysis is denoted as BLUPST or GBLUPST, and multiple-trait analysis as BLUPMT or GBLUPMT, respectively.
The pedigree-based BLUP model
The GBLUP model based on genomic information is:
The variance and covariance components were estimated using Average Information Restricted Maximum Likelihood (AIREML)
A four-fold cross validation was used in this study. Two scenarios were considered with regard to training and validation sets. The first cross validation scenario is random family sampling (CVF), in which all of the data for 8 half-sib families were randomly divided into four subsets, i.e., each subset has 2 half-sib families. The second cross validation scenario is random individual sampling (CVR), in which all of the data (511 birds) were randomly split into four subsets. In each fold of validation, one data set was used as the test data set and the other three data sets as training datasets. In the family sample scenario, the test birds did not have any sibs as training birds, and thus had a distant relationship to the training birds. Conversely, in the random sample scenario, the test birds had many sibs as training birds, and thus had a close relationship with the training birds. To account for population structure and sampling variation, splitting was repeated 10 times in CVF and 50 times in CVR. The numbers of birds in the test and training data sets for each fold of validation are shown in
Scenario |
Training data | test data | Scenario | Training data | test data |
CVF_fold1 | 381 | 130 | CVR_fold1 | 383 | 128 |
CVF_fold2 | 383 | 128 | CVR_fold2 | 383 | 128 |
CVF_fold3 | 385 | 126 | CVR_fold3 | 383 | 128 |
CVF_fold4 | 384 | 127 | CVR_fold4 | 384 | 127 |
CVF = the cross validation scenario by random family sampling (an example from 10 repeats); CVR = the cross validation scenario by random individual sampling.
In this study the accuracy of prediction was defined as the correlation between prediction and corrected phenotypic value (yc), where yc was calculated as the phenotypic value corrected for fixed sex and batch effects. A paired t test was implemented to test the differences among the correlations obtained from these prediction models. The paired t test treated a fold of validation as a subject and took a pair of correlation coefficients for the fold from two models as a matched pair of observations.
As shown in
Trait |
Method | |||
Ab-NDV | BLUPST | 0.306 | 0.335 | 0.478±0.142 |
BLUPMT | 0.314 | 0.330 | 0.487±0.144 |
|
GBLUPST | 0.206 | 0.367 | 0.360±0.075 |
|
GBLUPMT | 0.210 | 0.364 | 0.366±0.073 |
|
Ab-AIV | BLUPST | 0.142 | 0.329 | 0.301±0.123 |
BLUPMT | 0.136 | 0.332 | 0.291±0.120 |
|
GBLUPST | 0.158 | 0.291 | 0.351±0.077 |
|
GBLUPMT | 0.155 | 0.292 | 0.347±0.077 |
Ab-NDV = antibody response to Newcastle disease virus; Ab-AIV = antibody response to Avian Influenza virus.
* Significantly different from 0 at P<0.05.
** Significantly different from 0 at P<0.01.
Trait |
Scenario |
BLUPST | BLUPMT | GBLUPST | GBLUPMT | Scenario | BLUPST | BLUPMT | GBLUPST | GBLUPMT |
Ab-NDV | CVF_mean | 0.086a | 0.087a | 0.223b | 0.237b | CVR_mean | 0.391a | 0.389a | 0.424b | 0.427b |
CVF_fold1 | 0.093 | 0.099 | 0.249 | 0.270 | CVR_fold1 | 0.402 | 0.399 | 0.433 | 0.439 | |
CVF_fold2 | 0.064 | 0.072 | 0.182 | 0.206 | CVR_fold2 | 0.402 | 0.401 | 0.435 | 0.440 | |
CVF_fold3 | 0.138 | 0.128 | 0.229 | 0.239 | CVR_fold3 | 0.379 | 0.379 | 0.410 | 0.412 | |
CVF_fold4 | 0.050 | 0.049 | 0.233 | 0.234 | CVR_fold4 | 0.379 | 0.377 | 0.417 | 0.418 | |
Ab-AIV | CVF_mean | 0.094a | 0.080a | 0.332b | 0.347c | CVR_mean | 0.284b | 0.281a | 0.364c | 0.367d |
CVF_fold1 | 0.071 | 0.066 | 0.360 | 0.383 | CVR_fold1 | 0.298 | 0.295 | 0.375 | 0.378 | |
CVF_fold2 | 0.073 | 0.034 | 0.306 | 0.313 | CVR_fold2 | 0.292 | 0.289 | 0.367 | 0.368 | |
CVF_fold3 | 0.061 | 0.064 | 0.345 | 0.364 | CVR_fold3 | 0.277 | 0.275 | 0.359 | 0.364 | |
CVF_fold4 | 0.170 | 0.155 | 0.316 | 0.328 | CVR_fold4 | 0.268 | 0.266 | 0.354 | 0.359 |
Ab-NDV = antibody response to Newcastle disease virus; Ab-AIV = antibody response to Avian Influenza virus.
CVF = the cross validation scenario by random family sampling; CVR = the cross validation scenario by random individual sampling.
within a row, estimates without a common superscript differ significantly (P<0.05), according to the paired t test.
In single-trait analysis, compared with conventional EBV, the accuracies of the genomic predictions from GBLUPST for Ab-NDV and Ab-AIV increased by 0.137 and 0.238 in CVF, and 0.033 and 0.080 in CVR, respectively. In multiple-trait analysis, in contrast to conventional EBV, the accuracies of the genomic predictions from GBLUPMT for Ab-NDV and Ab-AIV increased by 0.150 and 0.267 in CVF, and 0.038 and 0.086 in CVR, respectively. For both traits, the multiple-trait model (GBLUPMT) led to an increase in accuracy of genomic prediction.
Cross-breeding is widely applied in beef, pigs and chickens
In addition, the present study showed a favorable moderate genetic correlation between Ab-NDV and Ab-AIV, which indicates that selection for one trait will led to a correlated favorable response for the other trait.
Genomic prediction based on genome-wide dense markers captures not only the information about linkage disequilibrium between markers and QTL but also the additive genetic relationship between individuals
The Bayesian model is another type of model that was widely implemented in genomic selection. Several previous studies have reported differences in the accuracy of genomic prediction between GBLUP and Bayesian models. Su et al.
In this study, the data were split into training and test data sets in two different ways, and cross-validation was carried out based on these data sets to evaluate the accuracy of genomic predictions. The accuracies in CVR were higher than those in CVF. This is because, in contrast with CVF, the test birds had sibs in the training data in CVR. Therefore, the genetic ties between test and training birds are much stronger in CVR, which confirms that the accuracies were higher with more close genetic relationship between test and training animals, as reported by previous studies
Comparisons between single-trait and multiple-trait models for genomic prediction have been reported in previous studies. Christensen et al.
Genetic correlations among traits impacts the accuracy of the multiple-trait model. Using simulation data, Calus et al.
This is the first study on genomic prediction for Ab-NDV and Ab-AIV. It was found that Ab-NDV and Ab-AIV were moderately heritable. Genomic prediction can greatly improve the accuracy of estimated breeding values. The genomic prediction using the multiple-trait model was more accurate than prediction using the single-trait model. The results indicate that genomic selection for Ab-NDV and Ab-AIV is promising.
The authors acknowledge Ning Li and Xiaoxiang Hu (College of Biological Science, China Agricultural University) for genotyping assistance.