J. Anim. Sci. 2002. 80:2263-2266
© 2002 American Society of Animal Science
Joint evaluation of purebreds and crossbreds in swine
E. Lutaaya*,1,
I. Misztal*,2,
J. W. Mabry*,3,
T. Short
,
H. H. Timm
and
R. Holzbauer
* Department of Animal and Dairy Science, University of Georgia, Athens 30602;
and
Pig Improvement Company, Franklin, KY; and
and
Deutsch Pig Improvement Company, Germany
2 Correspondence:
phone: 706-542-0951; E-mail:
ignacy{at}uga.edu.
 |
Abstract
|
|---|
Data from two purebred swine lines A (n = 6,022) and B (n = 24,170), and their reciprocal, cross C (n = 6,135), were used to examine gains in reliability of combined purebred and crossbred evaluation over conventional within-line evaluations using crossbred and pureline models. Random effects in the pureline model included additive, parental dominance, and litter. In the crossbred model, effects were as in the pureline model except traits of each line were treated as separate traits and two additive effects were present. The approximate model was the same as the pureline except it was used for all lines disregarding breed differences. The traits in the evaluation were lifetime daily gain (LDG) and backfat. When separate line evaluations were replaced by evaluations with crossbreds, mean reliabilities of predicted breeding values increased by 2 to 9% for purebreds and by 21 to 72% for crossbreds. Rank correlations between these breeding values were > 0.99 for purebreds but 0.85 to 0.87 for crossbreds. Rank correlations between predicted breeding values obtained from crossbred and approximate models were 0.98 to 0.99 for purebreds and 0.96 to 0.98 for crossbreds. When the number of crossbreds was small in comparison to purebreds, the increase in reliability by using the crossbred data and the crossbred model as opposed to purebred models was small for purebreds but large for crossbreds. The approximate model provided very similar rankings to the crossbred model for purebreds but rankings were less consistent for crossbreds.
Key Words: Crossbreds • Pigs • Purebreds • Reliability • Selection
|
Introduction
|
|---|
In swine, commercial animals are crossbreds but genetic selection is carried out based on purebred performance. There is interest in combined evaluation of purebreds and crossbreds to evaluate purebreds more precisely and for crossbred performance (Comstock et al., 1949; Wei and Van Der Werf, 1994).
Accurate theories for modeling variances in crossbred populations are very complex. Models that account for all additive and dominance (co)variances among all crosses of two pure lines (Lo et al., 1995) require a large number of parameters and are not practical. A simplified model in which the only cross allowed is the terminal cross (F1) by Lo et al. (1997) results in more realistic computations. This model, which contains two additive effects, allows for different variances in each pureline and in crossbred lines, for less than unity correlation between genotypes expressed in crossbreds and purebreds, and for different covariances among half-sib groups dependent on the breed of the common parent. Spilke et al. (1998) compared that model with a simpler multiple-trait model that contained only one additive effect. Estimates of genetic correlations from the simpler model were biased, although the loss of efficiency compared with the full model was small.
Lutaaya et al. (2001) applied the model by Lo et al. (1997) to lifetime daily gain (LDG) and backfat (BF) from a terminal cross. Some estimates of the genetic correlation were below 0.4 and the estimate of dominance variance approached 0.39 for LDG. Low genetic correlations and large dominance variance would suggest that the use of the crossbred model has merits.
The objectives of this study were 1) to examine the gains in reliability of joint purebred and crossbred evaluations over conventional within-breed analyses and 2) to compare animal rankings from within-line, crossbred, and simplified crossbred model analyses.
|
Materials and Methods
|
|---|
Data for this study were obtained from Deutsch Pig Improvement Company (DPI) and spanned a period of 8 yr (1991 to 1998). The data consisted of two purebred lines A (6,022 animals) and B (24,170 animals) developed from Landrace and Large White breeds respectively, and one crossbred line C (6,135 animals), consisting of reciprocal crosses of lines A and B. Line A was used predominantly as a paternal line and line B as a maternal line; parents of line C were 173 animals from line A and 1,208 animals from line B. Traits were lifetime daily gain (LDG) and backfat (BF). Distribution of records by line and sex is shown in Table 1
and mean and standard deviation of LDG and BF for each line are shown in Table 2. More details about the data are available in Lutaaya et al. (2001).
Models
Three models were used in the evaluation. The purebred model was:
where y is a vector of observations; ß is a vector of fixed effects of contemporary group, sex, and for, BF only, a covariable for end weight at end of the test period; u is a vector of additive effects, f is a vector of parental dominance effects; t is a vector of litter effects; e is a vector of residuals; and X, W, Z, and S are appropriate design matrices. Variances are: var(u) = A
a2; var(f) = Ff2; var(t = I x 2t; var(e = I x 2e, where A is additive relationship matrix, F is parental dominance relationship matrix, and a2, f2, 2t and 2e are appropriate variance components with values as estimated by Lutaaya et al. (2001).
The crossbred model by Lo et al. (1997) can be written as an extension of the previous model as follows:
where subscripts A, B, and C denote vectors/matrices for appropriate lines, uAC (uBC) is vector of additive effects in line C as passed by line A (B). Only two groups of effects are correlated: (uA, uAC) and (uB, uBC); the other effects are uncorrelated. Variance components were as estimated by Lutaaya et al. (2001). For the purebred model, they were as follows:
The parameters for the crossbred model for LDG were:
where Ax was the additive numerator relationship matrix, Fx was the parental dominance matrix, Iwx was an identity matrix, x denoted a particular line, and w denoted a particular effect.
The parameters for the crossbred model for BF were:
The approximate crossbred model was the same one as used for purebreds but with the genetic parameters as in the largest line B; the choice of parameters in single-trait models is not very important because these models are robust with respect to small changes in parameters. An implicit but very important assumption in the approximate model is that genetic correlations between purebreds and crossbreds equal unity.
Computations
Additive values were calculated for pigs in all lines using all three models by program BLUPF90 (Misztal, 1999). Improvement in additive evaluations resulting from the change from purebred to crossbred model was determined by comparison of reliabilities. Reliabilities were obtained by inversion using the formula: rij2 = 1 - pevij/j2, where rij2 is reliability for animal i and breed/trait j, pevij is the corresponding prediction error variance, and j2 is the additive variance for trait/breed j. Differences among additive values for various lines were determined by rank correlation.
|
Results and Discussion
|
|---|
Table 3 presents the mean reliability of predicted purebred breeding values (PBV) obtained from purebred and crossbred models. Mean reliability of line A was improved by 0.02 to 0.03 and mean reliability of line B was improved by 0.01. The improvement was larger for breed A because 1) it was mainly a paternal line and 2) line C was as large as line A. Sires in line A had approximately as many progenies in line A as in line C. Increases in reliability of these sires due to the crossbred information also benefited females in line A. Smaller improvement was obtained for line B because 1) it was a predominantly maternal line and 2) line C was only one-quarter the size of line B. In swine, the number of crossbreds is usually much higher than the number of purebreds. With a larger number of crossbreds in the analyses, the improvement in reliabilities of purebreds would be higher.
Table 4 presents the mean reliability of PBV for crossbreds obtained from purebred and crossbred models. The improvement was 0.05 to 0.11 for the predicted breeding values originating from line A (uAC) and 0.19 to 0.21 for line B (uBC). The proportion of improvement was reversed compared with purebred PBV because males from line A could be relatively well evaluated using the crossbred information only but females from line B could not unless records from line B were available.
Table 5 presents rank correlations between PBV from purebred and crossbred analyses for each of the lines. The correlations were greater than 0.99 for the purebreds. This indicates that improvements in reliabilities of purebreds after adding crossbred data are small when the amount of crossbred data is small. The correlations were lower, 0.85 to 0.87, for crossbreds because a substantial part of information in crossbreds was derived from the purebreds.
Correlations between PBV obtained from the crossbred and the approximate models, where all lines were evaluated by a single-trait model, are in Table 6. Correlations for purebreds are high at 0.98 to 0.99; correlations for crossbreds are 0.96 and 0.98. It appears that the approximate model produces PBV almost identical to the more complex crossbred model, especially for the purebreds. In crossbreds, a higher correlation was found for LDG, where variances were very similar in all three lines.
In view of high correlations, it seems that the approximate method is sufficiently accurate to evaluate purebreds and crossbreds. However, this holds only if the number of crossbreds is small. If the number of crossbreds is much higher than the number of purebreds, then differences among the methods will be larger.
The usefulness of the crossbred evaluation will depend on the purpose of the evaluation (Wei and Van der Werf, 1994; Bijma and Van Arendonk, 1998). If the only purpose is to improve performance of pure lines and the number of crossbreds is low, the purebred evaluation would be the best strategy. If data on crossbreds are used to increase the reliability of purebred evaluation, the genetic correlations between lines are high, and variances are similar within the lines, the approximate model would be appropriate but could be associated with several problems. First, records from lines with lower variances would receive higher weight. Then, evaluations of crossbreds based on purebred data, and vice versa, would be inflated due to assuming unity genetic correlations. Although these problems may not result in large differences in ranking, genetic gains as predicted through inflated evaluations would not be realized. Also, reliabilities based on correlated lines would be inflated.
The crossbred model by Lo et al. (1997) may be useful in a few cases. The first case is when the genetic correlations are low, when both purebred and crossbred evaluations are of interest, and when substantial crossbred information is available. The second case would be when some traits are recorded for the purebreds but others only in crossbreds (H. Van Der Steen, 1999, personal communication). In either case, the crossbred methodology would provide appropriate scaling and weighting of observations from all the lines, and without overestimation of reliabilities.
The terminal cross model is of only limited interest in swine because commercial animals are up to five-way crosses. However, the theory involving variances of multi-way crosses is very complicated (Lo et al., 1995), and recording on crosses beyond two-way, which are at the multiplier level, may not be economically feasible. Still, the evaluations with the terminal cross models may be more accurate for the purpose of selecting parents of commercial animals than the evaluations with simpler models, especially when the amount of data on terminal crosses is relatively large.
|
Implications
|
|---|
The crossbred model in swine is not justified when the main interest is in evaluations of purebreds and when the amount of crossbred data relative to purebred data is small. Such a model is more justified when crossbred evaluations are of interest. The purebred model can also be successfully used for the joint evaluation but its usefulness will depend on values of genetic correlations and variances within lines. When the interest is in both purebred and crossbred evaluations, and when some traits are recorded predominantly on crossbreds, the use of the crossbred model will result in properly scaled and unbiased evaluations.
|
Footnotes
|
|---|
1 Currently at Univ. of Namibia, Windhoek, Namibia. This study was partly funded by a grant from the Pig Improvement Company, Franklin, KY. Assistance by Kath Donoghue and Shogo Tsuruta is gratefully acknowledged. 
3 Presently at Iowa State Univ., Ames.
Received for publication June 25, 2001.
Accepted for publication February 15, 2002.
|
Literature Cited
|
|---|
Bijma, P., and J. A. M. Van Arendonk. 1998. Maximizing genetic gain for the sire line of a crossbreeding scheme utilizing both purebred and crossbred information. Anim. Sci. 66:529–542.
Comstock, R. E., H. F. Robinson, and P. H. Harvey. 1949. A breeding procedure designed to make maximum use of both general and specific combining ability. Agron. J. 41:360–367.[Free Full Text]
Lo, L. L., R. L. Fernando, R. J. C. Cantet, and M. Grossman. 1995. Theory for modelling means and covariances in a two-breed population with dominance inheritance. Theor. Appl. Genet. 90:49–62.
Lo, L. L., R. L. Fernando, and M. Grossman. 1997. Genetic evaluation by BLUP in two-breed terminal crossbreeding systems under dominance. J. Anim. Sci. 75:2877–2884.[Abstract/Free Full Text]
Lutaaya, E., I. Misztal, J. W. Mabry, T. Short, H. H. Timm, and R. Holzbauer. 2001. Genetic parameter estimates from joint evaluation of purebreds and crossbreds in swine. J. Anim. Sci. 79:3002–3007.[Abstract/Free Full Text]
Misztal, I. 2000. BLUPF90 program. Available at: fttp://nce.ads.uga.edu/pub/ignacy/blupf90/. Accessed Dec. 1, 2000.
Spilke, J., E. Groeneveld, and N. Mielenz. 1998. Joint purebred and crossbred (co)variance component estimation with a pseudo multiple trait model: loss in efficiency. J. Anim. Breed. Genet. 115:341–350.
Wei, M., and J. H. J. Van der Werf. 1994. Maximizing genetic response in crossbreds using both purebred and crossbred information. Anim. Prod. 58:401–413.
This article has been cited by other articles:
|
|
|
|
 
J. M. Hickey, M. G. Keane, D. A. Kenny, A. R. Cromie, H. A. Mulder, and R. F. Veerkamp
Estimation of accuracy and bias in genetic evaluations with genetic groups using sampling
J Anim Sci,
May 1, 2008;
86(5):
1047 - 1056.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
P. M. VanRaden, M. E. Tooker, J. B. Cole, G. R. Wiggans, and J. H. Megonigal Jr.
Genetic Evaluations for Mixed-Breed Populations
J Dairy Sci,
May 1, 2007;
90(5):
2434 - 2441.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
J. B. Cole, R. C. Goodling Jr., G. R. Wiggans, and P. M. VanRaden
Genetic Evaluation of Calving Ease for Brown Swiss and Jersey Bulls from Purebred and Crossbred Calvings
J Dairy Sci,
April 1, 2005;
88(4):
1529 - 1539.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Introduction
