| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
ANIMAL GENETICS |
Department of Animal and Aquacultural Sciences, Norwegian University of Life Sciences, Ås, Norway
 |
Abstract |
|---|
 Top Abstract INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Key Words: breed • genotype by environment interaction • phenotypic plasticity • sheep • weaning weight
|
INTRODUCTION |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
). Genetic differences in plasticity will show up in a statistical analysis as genotype x environment (G x E) interactions. In Norway’s rangeland-based sheep industry, the G x E interaction is potentially important: the pasture environments where lambs spend most of their life before weaning are greatly variable, both spatially and temporally (Mysterud et al., 2001; Steinheim et al., 2004b). Increasing use of rams across the population through AI (Eikje et al., In press) also calls for studies of G x E interactions in Norwegian sheep breeding. Previous studies have demonstrated G x E effects on lamb growth traits (e.g., Carter et al., 1971; Rajab et al., 1992; Osoro et al., 1999; Lewis et al., 2002; MacFarlane et al., 2004).
In Norway, the long-tailed Norwegian White sheep (NWS) and the Nordic short-tailed Spel sheep (Spel) are the most common breeds, constituting 80 and 15% of the sheep population, respectively. Both are spread across greatly varying environments. Using Hofmann’s (1989) ruminant classification, we expected the BW of Spel lambs to be less sensitive to environmental variation compared with NWS, because the Spel sheep may be classified as more selective grazers than the NWS (Steinheim et al., 2003, 2005). Thus, with the generally low stocking levels on Norwegian rangelands (Mysterud, 2000) allowing for selectivity, Spel sheep should be superior in compensating for forage quality variation. Steinheim et al. (2004a) indicated that Spel sheep were less sensitive than NWS to an environmental gradient. A G x E interaction effect on lamb weaning weights, using breed as the genotype, was thus hypothesized and tested by examining whether breed-specific variance components were different on the micro- or macroenvironmental level, and whether the correlation between the environmental rankings of the 2 breeds was imperfect.
|
MATERIALS AND METHODS |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Animal Care and Use Committee approval was not obtained for this study because the data were obtained from an existing database.
Farms that had kept both NWS and Spel sheep during 1989 through 1999 were identified by using the database of the national Sheep Recording System (containing information on approximately 30% of all Norwegian sheep flocks). Litters with more than 4 lambs and litters by ewes above 7 yr of age (few ewes above that age are kept in the breeding stocks) were excluded, as were litters not born during April or May and litters in which lambs had been removed or added by the farmer. Individual lambs were removed if they weighed <20 kg at weaning and if their weaning age was <111 d or >179 d. Altogether, the restrictions resulted in removal of 
2.5% of lambs (from farms stocking both breeds).
Further, the data set was restricted to flocks with a yearly minimum of 18 lambs of each breed. A total of 40 flocks, with a wide geographical distribution (Figure 1), met the requirements, and the final data set included 37,338 NWS and 30,075 Spel lambs, from 20,887 NWS and 16,629 Spel litters. For details on variables and the number of observations in the data set, see Table 1.
|
|
). Adult Spel ewes (double-fleeced), a breed originating from old Nordic short-tailed sheep breeds (Drabløs, 1997), weigh 60 to 70 kg (Trodahl, 1989; Drabløs, 1997). The NWS have a greater digestive capacity, relative to body size (Steinheim et al., 2003), and tend to select a more grass-dominated diet compared with the Spel breed (Steinheim et al., 2005). The breeds also differ in flock behavior, with Spel sheep being more gregarious (Drabløs, 1997) and more reactive in their responses to predator-like stimuli (Hansen et al., 2001). For images and descriptions of the breeds, see Oklahoma State University’s Breeds of Livestock database (Oklahoma State University, 2007); for NWS refer to the similar and overlapping Dala breed. The 2 breeds have different histories as domesticated sheep. The (mainly) British origins of the NWS imply grazing on grassland and heather pastures, being kept outdoors in winter, mainly on Calluna heathers. The northern European history of the Spel means adaptation to more diverse and heterogeneous forest and mountain pastures during summer and, in general, low-level indoor feeding during winter. Historically (and today), Scandinavian sheep have also been exposed to greater risks of predator attacks (by the wolverine, lynx, red fox, brown bear, wolf, and golden eagle) compared with sheep on the British Isles (mainly red fox). All of these factors may have contributed to breed differences with respect to utilization of unfenced rangeland pastures.
Statistics
Mean lamb weaning weight per breed and year was estimated by using the Summary procedure in the SAS System for Windows, release 8.2 (SAS Institute Inc., Cary, NC), and variables relevant to the analysis were described (Table 1
). The final statistical analyses were performed with mixed linear models by using ASREML software (Gilmour et al., 2002). Likelihood-ratio tests were used to determine the significance of the random effects by extending the model through a 4-step process. Phenotypic sensitivity and G x E interaction were inferred from the random effects and from the correlation between breeds in how they responded to the environment classes. Such a partitioning of the G x E interaction into 1) effects caused by heterogeneous variances and 2) imperfect correlations of environmental rankings among genotypes was suggested by Muir et al. (1992) through a sums-of-squares approach. Yang (2002) added the method to a REML framework, allowing sound statistical testing through comparing log-likelihoods of reduced and full mixed models.
The main effect of breed was defined as fixed, representing only these 2 breeds, whereas the flock x year environmental effect was assumed to be random (sensu Muir et al., 1992). Breed-specific estimates (REML) of between- and within-environment variance components were used for estimation of phenotypic plasticity, following an approach resembling the one by Bytyqi et al. (2007). The breed-specific residual variances are measures of the breeds’ relative sensitivity to microenvironmental changes within macroenvironmental settings. The breed-specific flock x year variances indicate the breeds’ relative sensitivities to macroenvironmental (between flock and year) changes (Lynch and Walsh, 1998).
The differences between breed-specific variance components and the correlation between flock x year environmental effects or breeds determine whether a G x E interaction effect is present. In addition, the correlation gives a measure of the strength of association between the 2 breeds in how they "perceive" the environment by using an intuitively understandable and well-known measure.
The basic model (model 1) was
 |
In this model, y is the vector containing all observations of weaning weight (for both breeds), and β is a vector containing the fixed classification effects of breed (NWS or Spel), age of ewe (1 to 7 yr), sex of lamb (male or female), litter size (1 to 4 lambs), and year (11 classes, 1989 to 1999). Vector β also contains the continuous regression coefficient for age of lamb at weighing (111 to 179 d). Vector u [N(0, Iu
)] contains the random effect solutions for flock x year effects (40 flocks in 11 yr), with variance component , and e [ N(0, In
)] is the residual error term associated with each of the n observations (variance component: ).). Matrices X and Z are appropriate incidence matrices for the fixed and random effects, respectively, whereas I is the identity matrix of appropriate dimensions. In this model, identical variance components of flock x year effects were assumed for the 2 breeds, as it was for the residual effect.
Model 2 was identical to model 1, except that breed-specific residual variances were assumed to be
 |
where 
e12 and e22 are the residual variances for the NWS and Spel breed, respectively, with n1 and n2 observations.
Model 3 (an extension of model 2), was made by nesting the random flock x year effect within breed. More specifically,
 |
where Xb is a diagonalization of the row incidence vector for breed NWS (part of X), and u1 and u2 are the vectors of random flock x year effect solutions for the NWS and Spel breed, respectively. Further,
 |
where
 |
In this model, the correlation among breed-flock-year effects was fixed close to unity,
 |
to obtain a positive definite (co)variance matrix for breed-flock-year effects (U). The final model, model 4, was identical to model 3 except that the correlation among breed-flock-year effects was estimated from the data.
The different models were compared by using likelihood-ratio tests comparing nested models with the same fixed structure. The likelihood-ratio test statistic for 2 models i and j, where i is nested within j, is given by
 |
where Li and Lj are the restricted likelihoods of the 2 models, and 
i and j are the corresponding number of (co)variance components of those models. Testing model 2 against model 1 is equivalent to testing a null hypothesis that assumes homogeneous residual variance across breeds against a hypothesis that assumes breed-specific residual variances. Testing model 3 against model 2 means testing a null hypothesis that assumes homogeneous flock-year effects across breeds against a hypothesis that assumes proportional flock-year effects (ru1,u2 = 1) with different variances. Finally, testing model 4 against model 3 corresponds to testing the null hypothesis that assumes that flock-year effects of the 2 breeds are proportional (ru1,u2 = 1) with different variances against a hypothesis that assumes the flock-year effects of the 2 breeds have different variances and a correlation lower than unity.
Significantly different residual variances for the 2 breeds imply that the breeds differ with respect to their microenvironmental sensitivity (Lynch and Walsh, 1998), whereas differences in flock-year variances represent breed differences on a macroenvironmental level. Additionally, a correlation (from model 4) between flock-year effects of the different breeds significantly lower than unity implies that the breeds not only vary with respect to environmental sensitivity, but also that there are differences in how the environments affect them (Muir et al., 1992).
|
RESULTS |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
. Mean lamb weaning weight of NWS was 42.99 (± 0.04) kg and that of Spel was 41.23 (± 0.04) kg. Mean lamb weaning weights (Figure 2) demonstrate a relatively similar pattern for the 2 breeds over the years. Log-likelihood tests demonstrated significant improvements (P < 0.001) from model 1 through 4 (Table 2). Hence, substantial breed differences were found for both residual variance and flock x year variance. Further, the correlation between flock-year effects of the 2 breeds of 0.82 (± 0.02) was significantly (P < 0.001) different from unity (Table 3).
|
|
|
), residual SD of the NWS and Spel breeds were 5.62 and 5.24 kg, whereas SD for effects of flock x year were 3.37 kg (NWS) and 3.16 kg (Spel; Table 3). The larger variances obtained for NWS compared with the Spel breed indicate systematic breed differences with respect to phenotypic plasticity on both the micro-and macroenvironmental levels. These differences could not be attributed to scale because analyses using transformed data (weaning weight to the power of 0.75, i.e., 
metabolic BW) gave similar results (G. Steinheim, unpublished data). Further, because the model estimated a correlation of 0.82 between flock-year effects for the 2 breeds, there was evidence of the 2 breeds ranking the flock-year environments differently. |
DISCUSSION |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
). The present method of analysis has been applied within plant breeding (Yang, 2002), but has not, to the authors’ knowledge, previously been used in animal breeding. The large, high-resolution sample data compiled for the Norwegian sheep industry and the existence of multibreed flocks allowed for such methods to be used. The analyses used a REML approach and a partitioning of the G x E interaction (Muir et al., 1992) with exact likelihood tests (Yang, 2002).
The weaning weight of NWS lambs was expected to be more sensitive to environmental variation than that of Spel lambs, and weaning weight was tested both for the micro- and macroenvironmental levels (Lynch and Walsh, 1998). The expectation was based on knowledge of anatomy, domestication history, and behavior, and was confirmed by the results of the analysis: the Spel was the more stable breed, significantly if not substantially.
The consistency of differences in plasticity on the micro- and macroenvironmental scale is interesting, and may represent a general pattern in the way genotypes respond to their environments (Lynch and Walsh, 1998). It should be noted that the measure of microenvironmental sensitivity (variance between sheep within year, flock, and breed) may be biased if it also contains genetic variance because of segregation (Lynch and Walsh, 1998). In the current study, both breeds were subject to a national breeding scheme with systematic exchange of rams within breeds across the population. In addition, all farms in the study were members of the Norwegian Sheep Recording System, and thus were provided with breeding indices calculated through the national sheep breeding scheme.
Breed-specific variances describe phenotypic plasticity, whereas the correlation of flock-year effects of the breeds determines the part of the G x E interaction caused by deviation from a perfect correlation of how the same environments affect the 2 breeds (Muir et al., 1992). The correlation (0.82, which in this case was different from unity) is a useful and easily understandable measure of the strength of association between environmental effects on the 2 breeds. This linear correlation of environmental rankings of genotypes is appropriate in situations in which few genotypes are applied to a greater number of environments (the latter was then considered a random effect; Muir et al., 1992). It should be noted that even if the correlation in this study were different from unity, the 2 sheep breeds would generally vary together across environments.
If studying many genotypes (defined as random) in a few (fixed) environments, one would similarly estimate the correlation between genotypes across environments. This correlation of genotypes across environments is the genetic correlation commonly used in G x E studies within breeds (Falconer, 1952; Stearns, 1992; Snowder and Knight, 1995; Lynch and Walsh, 1998; de Jong and Bijma, 2002). Whether these genetic correlations are different from unity can then easily be tested by using likelihood ratios.
Norwegian sheep share their rangeland pastures with wild ungulates, substantially overlapping in diets and habitat use patterns (Mysterud, 2000). Year-to-year variation in body growth between the different ungulate species seem to be positively correlated, at least for domestic sheep, semidomestic reindeer (Rangifer tarandus), and moose (Alces alces; Sæther, 1985; Weladji et al., 2003). Weladji et al. (2003) reported a correlation of 0.6 between the yearly juvenile BW of reindeer (carcass weights) and sheep (weaning weights) using the same grazing area in mid-Norway. Sæther (1985) found an even greater correlation (0.7) between 1.5-yr-old moose carcass weights and lamb carcass weights in a grazing area in northern Norway. Thus, when it comes to utilizing rangeland resources for meat production, responses to environmental variation are not always much more different between species than between genotypes within a species. The method used in this study can be used to examine how systems with sympatric ungulates respond to environmental variation.
A potentially important use of the ranking of environments for the 2 breeds (flock x year solutions) will be in contributing to the establishment of objective environmental characterizations relevant to production by using parameters such as length of the plant growing season and pasture altitude, longitude, and latitude. Such environmental variables or, alternatively, environmental clusters will be important to advance long-range G x E studies, including those within breeds.
The important difference between the present approach and reaction norm (RN) and multitrait models is that the present method does not depend on a predefined environmental gradient. If one has not defined an appropriate environmental descriptor, the multitrait and RN approaches may reject the presence of a true G x E interaction. Another assumption in the commonly used models is a joint environmental gradient for all genotypes, whereas in our model 4, the breed-specific flock-year solutions acting as environmental gradients are correlated to a degree estimated from the data. The stronger the G x E interaction, especially if the part attributable to imperfect correlations is large, the less accurate and relevant the estimation of a common environmental gradient may be in biological terms; it may, however, still be economically important and, as such, the "correct" gradient to use in analysis. The appeal of the present approach is that it is able to take complex biological realities into account; the downside is that results are (currently) difficult to apply to breeding schemes and management. The method should be developed so that results may be presented in an intuitively informative manner, such as the graphic presentations from RN models.
Our results demonstrated that the smaller breed, the short-tailed Spel, was less sensitive to environmental variation than the heavier NWS breed. The same pattern of environmental sensitivity between the light and heavier sheep breeds was demonstrated by MacFarlane et al. (2004) for the carcass weight of Suffolk and Scottish Blackface lambs on varying pasture qualities, and by Rajab et al. (1992) for body growth of 3 tropical hair sheep breeds. The same is also indicated by Osoro et al. (1999) for the performance of one light and one heavier breed when grazing pastures with varying Calluna and grass proportions.
The current analysis assumes that both breeds were offered the same available environment within flock and year. In production systems based on free-ranging generalist herbivores on heterogeneous pastures with a wide array of available food resources, a G x E interaction may be a result of genotype differences in diet choice, habitat use, or digestive anatomy, especially if low stocking rates make selective foraging possible.
The NWS lambs were heavier than the Spel lambs in an average environment, but their BW had a greater phenotypic plasticity. The Spel will thus compete better, relative to the NWS, in harsh environments; this result is in accordance with Steinheim et al. (2004a). However, to determine what breed would be the best in what environments, one needs to complete economic studies of farm systems, not only including income generated from the sale of lambs, but also including the costs (e.g., indoor feeding during winter) that are related to each sheep breed.
The different phenotypic plasticities and the imperfect environmental correlation of the 2 sheep breeds call for investigations into the potential benefits of incorporating G x E effects into sheep breeding programs, especially because the trait studied, the lamb weaning weight, through its strong correlation with slaughter weight, is economically important in sheep production. Before such extensions of breeding schemes are undertaken, objective environment characterization based on obtainable variables should be established. In addition, the differences in how the 2 breeds studied respond to environmental changes is another argument in favor of maintaining breed diversity (Hall and Bradley, 1995): the breeds may have different optimal production environments. These differences may be especially important in sheep management systems such as those in Norway, where the environmental variation is large today and where future environmental conditions for sheep production may be affected by large-scale climatic change (Mysterud et al., 2001).
|
Footnotes |
|---|
2 Corresponding author: geir.steinheim{at}umb.no
Received for publication January 15, 2007. Accepted for publication September 26, 2007.
|
LITERATURE CITED |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
This article has been cited by other articles:
|
|
 |
|
  N. Long, D. Gianola, G. J. M. Rosa, K. A. Weigel, and S. Avendano Marker-assisted assessment of genotype by environment interaction: A case study of single nucleotide polymorphism-mortality association in broilers in two hygiene environments J Anim Sci, December 1, 2008; 86(12): 3358 - 3366. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
