Genotype x environment interactions in pig breeding programmes

A pig breeding programme generally consists of different levels in a pyramidal structure, indicated as nucleus, multiplication and commercial level. Selection takes place at all levels but improvements generated in the nucleus determine eventually the rate of annual genetic change. Selection at nucleus level for growth and carcass traits is generally based upon performance testing, sometimes supplemented with sib information. These tests usually take place under standardized environmental conditions to allow a fair comparison of the tested pigs. However, these sophisticated conditions may deviate from the conditions at the multiplication level and certainly also from the conditions at commercial fattening where the breeding goal is defined. As a consequence changes in rank order for genotypes between these environments may occur and lower the efficiency of pig breeding programmes. These changes in rank order of genotypes between environments are indicated as genotype x environment interaction (G x E). The size of G x E may be represented by the genetic correlation between the genotypic values of the trait in different environments.
At the end of the 1970's several non-unit estimates of genetic correlations between the different levels of pig breeding programmes were reported in the literature. These results were considered as serious indications of G x E in pig breeding programmes, that might have serious drawbacks for the Dutch pig industry, e.g. for the Dutch herdbook breeding programme in which three levels can be distinguished; nucleus herds with testing at central stations, multiplication herds with on-farm testing and commercial herds with fattening pigs. This encouraged further research into the Dutch herdbook breeding programme on cause and effect of G x E.

The first main object of the project was the investigation of environmental effects in central test, on-farm test and commercial fattening results and the estimation of up-to-date genetic parameters for the traits measured at these levels of the breeding programme. The analyses of G x E may give biased results in case the appropriate definition of environmental effects and up-to-date genetic parameters are not used. Routinely collected central test and on-farm test data from Dutch Landrace (NL) and Dutch Yorkshire (GY) pigs tested between 1979 and 1983 were used. The fattening data of crossbred pigs were obtained in a progeny test on commercial fattening herds of 65 central and 42 on-farm tested GY- AI-boars.
To investigate the environmental effects within test stations (Chapter 2), different definitions of environmental effects were included separately in models for analysis of variance. Batch effects were significant (P < 0.001) for daily gain on test and feed conversion ratio, month effects were significant (P < 0.05) for backfat measurements and ham + loin %. Indications for an optimal classification of the environmental effects were shown only for daily gain and feed conversion ratio. For the carcass characteristics no balance could be found between chance and environmental fluctuations. The estimated heritabilities in central test for daily gain on test, feed conversion ratio and ultrasonic backfat thickness were 0.22, 0.23, 0.26 for NL and 0.14, 0. 9, 0.29 for GY respectively. Differences between the two breeds in heritabilities were reported, especially for ham + loin % (NL, h ²= 0.34; GY, h ²= 0.75), which may be the result of the selection against halothane-positive animals in NL.
Herd effects were an important source of environmental variation in onfarm test results (Chapter 3) and explained 9 to 20 % of the variance within herdbook regions. A part of these herd effects was due to differences in use of AI-boars between herds. Within herdbook regions these differences were small owing to intensive use of AI. However, across regions indications were found for moderate genetic herd differences. The estimated heritabilities for weight corrected for age, backfat thickness corrected for weight and the performance index were 0.13, 0.39, 0.26 for NL and 0.19, 0.27, 0.22 for GY respectively.
Also in the commercial fattening data (Chapter 5), herd effects were an important source of environmental variation next to seasonal effects. The heritability estimates for daily gain during the fattening period, daily gain during life, the score for backfat thickness and the score for type were 0.05, 0.08, 0.10 and 0.10 respectively. Also carcass weight was analysed and had next to a heritability of 0.05, also a high genetic correlation with the two growth traits.

From these results it was concluded that the evaluation of central and onfarm test results may be improved by an appropriate correction for batch or month effects in central test and for herd effects in on-farm test. Moreover, the genetic parameters used for these evaluation procedures should be replaced by the estimates reported, especially for the evaluation of NL in central test. In commercial fattening data genetic variance was present but the heritabilities were low if compared to the heritabilities for similar traits in central or on-farm test.

The second main object of the project was the analysis of G x E. The problem of G x E was analysed as (1) the genetic correlations (r _G ) between identical traits measured in the nucleus, multiplication and commercial fattening level and (2) the genetic correlations (r _g ) among identical traits measured in the various environments within each of the three levels. The data used were the same as in the first part of the project.
Because the traits used in the different levels of the breeding programmes are not identical, genetic correlations between the various definitions of both growth rate and carcass quality were estimated on the basis of central test data (Chapter 1). The genetic correlations between different definitions of growth rate were all close to one (r _g = 0.81 - 1.0). However, the genetic correlations between different definitions of carcass quality (e.g. carcass backfat thickness, ultrasonic backfat thickness and the score for carcass backfat thickness) clearly showed differences in genetic background which should be taken into account in the comparison of these traits across levels of the breeding programme.
In central test results (Chapter 1) sire x batch and sire x month interactions were not significant (P>0.05) for the traits included in the selection; the genetic correlations within the nucleus level (r _gI ) were equal to one. In on-farm test data (Chapter 4), the sire x herd interaction was significant (P < 0.001) for all test characteristics and explained a large part of the total variance. The genetic correlations between sires' progeny performance in different multiplication herds (r _gII ) varied between 0.3 and 0.7 for weight corrected for age (SC W) and between 0.6 and 0.9 for backfat thickness corrected for weight (SC UB). Non-random mating, preferential treatment of pigs and environment-specific genes are discussed as possible causes of these sire x herd interactions.
At the level of commercial fattening (Chapter 5) the sire x herd interaction was significant (P < 0.001) for the growth traits but not for the carcass characteristics. Genetic correlations between sires' progeny performance in different fattening herds (r _gIII ) were 0.29 for daily gain during the fattening period and 0.52 for daily gain during life. As there are so many environmental differences between fattening herds, environment- specific genes are expected to be responsible for the low genetic correlations among herds.

The genetic correlations between the different levels of the breeding programme (Chapter 6) were derived from the correlations between best linear unbiased predictions of breeding values at the different levels. Moderate genetic correlations were calculated between central and on- farm test; for backfat thickness r _G1 - 0.3 -.7, for daily gain r _G1 - 0.3 - 0.65. Differences in definition of the traits and differences in sex of the progeny were only partly responsible for the moderate relationships. For identical traits measured in central and on-farm test on progeny of the same sex r _G1 - 0.41 for daily gain and r _G1 - 0.70 for backfat thickness. Sire x herd interaction in on-farm test data was found to be the responsible factor for the moderate correlations between central test and on-farm test.
Between progeny results in commercial fattening and performances of the sires in central test no clear relationship was found for daily gain, r _G2 = -0.48 - 0.17, but high correlations for identical carcass characteristics, r _G2 = 0.57 - 0.64. These results agreed closely with the presence of sire x herd interactions in commercial fattening for only daily gain. The genetic correlations between on-farm test and commercial fattening were high for daily gain, r G3= 1.0, but low for carcass characteristics, r _G3 = 0. The presence of sire x herd interaction in both levels of the breeding programme may be responsible for these inconsistent relationships.

From the analyses of G x E within and between levels of the breeding programme it was concluded that there exist moderate genetic relationships between the different levels of the Dutch herdbook breeding programme. The sire x herd interactions within multiplication and commercial fattening levels are responsible for this. Since the differences between herds, multiplication as well as commercial fattening herds, are numerous and sometimes undefinable, selection of genotypes for suitability under commercial fattening conditions is desirable.

Finally, the consequences of the moderate genetic correlations for the design and efficiency of pig breeding programmes were investigated (Chapter 7).
In general, the accuracy of selection across levels of the breeding programme is directly proportional to (r _G /[r _gP * r _gH ] ^½) where r _gP and r _gH are the genetic correlations within respectively the level where the index information is collected and the level where the breeding goal is defined. However, for a fixed r _G , the highest genetic progress may be reached if r _gP - r _G . A limited number of test places are best used by distributing the representatives of the genotype over as many herds as possible. The size of r _G in comparison with r _g is discussed further as this has a large impact on the efficiency of the breeding programmes.
Furthermore, some testing strategies were compared for their expected genetic progress with values for r _g and r _G as reported in the different chapters. It was concluded that in general testing of boars and their paternal half sibs in on-farm test or commercial fattening simultaneously is depending on the genetic correlations almost three times more efficient than central testing only. Also two-stage selection with progeny testing in commercial fattening appeared an efficient alternative (1.5 - 2.25 times more efficient than central testing only) under the circumstances of G x E.