S . Hermesch et al. Livestock Production Science 65 2000 239 –248
243
traits included breed and method of measurement for with estimates of 0.27 for model one and 0.10 for
intramuscular fat content. Weight of meat sample and model two while litter effect had a larger importance
2
hot carcass weight were fitted as linear covariables for this breed c : 0.20. In contrast, growth rate was
for drip loss percentage and intramuscular fat con- highly heritable in Landrace with estimates of 0.57
tent. and 0.48 for both models while litter effect was of
2
Estimation of variance components was based on less importance c : 0.08. The design of this project
an animal model applying a Restricted Maximum provided a similar data structure for both breeds in
Likelihood procedure using the DFREML program regard to piglets per litter and management system.
Meyer, 1993. A variance of the likelihood function This excludes systematic effects as a possible reason
2 8
values of less than 10 was chosen as a conver-
for these differences in heritabilities and litter ef- gence criterion. Approximate standard errors of
fects. Results from Klassen 1992, who analysed heritability estimates were obtained as described in
lifetime average daily gain for a number of Aus- Meyer 1989. Significance of random effects was
tralian herds, confirm higher heritabilities for Aus- tested through a likelihood ratio test. Based on this
tralian Landrace in comparison with Australian likelihood ratio test, litter effect was only significant
Large White. Klassen 1992 who did not fit litter for growth rate, weight of the back leg and lean meat
effect as an additional random effect estimated weight of the back leg as an additional random effect
heritabilities ranging from 0.23 to 0.40 for Large Table 2.
White and from 0.32 to 0.61 for Landrace pigs. It is also possible that the size or structure of data
sets in this study did not allow a reliable simulta-
3. Results and discussion neous estimation of additive genetic effects and litter
effects. To analyse the effect of size of the data set 3.1. Performance traits
on estimated parameters, data were subdivided in a separate analysis using only data until a certain
Heritability and litter effect estimates differed recording date with break off dates every six month.
substantially between Large White and Landrace for For both breeds, genetic parameters did not change
average daily gain from three to 18 weeks Table 3. significantly by adding more data once the data set
Heritability estimates were lower for Large White reached approximately 1000 animals. Therefore fur-
Table 3
2 2
Heritabilities h , litter effects c both with standard errors s.e. and variance components for growth rate from three to 18 weeks ADG1 full and reduced data set
2 2
2 2
2
Data set Model
h s.e. of h
c s.e. of c
s
p a
b
Full data LW
Model 1 0.27
0.08 3959
c
Model 2 0.10
0.05 0.20
0.04 3921
c
LR Model 1
0.57 0.08
4569 Model 2
0.48 0.09
0.08 0.04
4604 combined
Model 1 0.43
0.06 4307
c
Model 2 0.27
0.05 0.15
0.03 4210
Subset LW
Model 1 0.39
0.10 3906
c
Model 2 0.17
0.07 0.16
0.04 3843
c
LR Model 1
0.60 0.09
4544 Model 2
0.48 0.10
0.08 0.05
4545 combined
Model 1 0.47
0.07 4237
c
Model 2 0.27
0.06 0.14
0.03 3888
a
LW: Large White, LR: Landrace.
b
Model 1: random effects include animal effect and residual effect; Model 2: random effects include animal effect, litter effect and residual effect.
c
Significant model according to log likelihood ratio test.
244 S
. Hermesch et al. Livestock Production Science 65 2000 239 –248
ther data collection lead to a decrease in standard genetic parameters for the reduced data set in
errors for genetic parameters but did not greatly comparison to the full data set. For example, by not
influence the estimates themselves. fitting litter effect a larger part of the variation due to
The structure of the data set was based on a litter effect was picked up by the additive genetic
random choice of two male piglets per litter. How- variance. These results are in agreement with results
ever, it was not always possible to obtain two male from a simulation study by Meyer and Hill 1992
piglets and 14 of Large White pigs and 16 of who found that the sampling correlation does not
Landrace pigs had no littermates recorded. To better appear to be dependent on the number of sire
separate additive genetic effects and litter effects, families but is affected by their structure.
pigs with no littermates available were removed from For further illustration, the log likelihood was
the data. Applying model one, the heritability esti- plotted for a range of heritability and litter effect
mate was 0.39 in Large White Table 4. When estimates for the full Large White data in Fig. 1. The
model two was used in the reduced data set top area of the log likelihood represents combina-
heritability increased to 0.17 while litter effect tions of heritability and litter effects estimates which
decreased to 0.16. In contrast to Large White, are not significant different from the maximum log
estimates of heritabilities and litter effects did not likelihood at a heritability of 0.10 and a litter effect
change significantly between the full and the reduced of 0.20. This area covers a range of heritability
data set in Landrace. estimates of 0.03 to 0.22 and a range of estimates of
Meyer and Hill 1992 stated that parameters are litter effects of 0.04 to 0.22 illustrating that a
correlated in models with multiple parameters imply- substantial cross-substitution can occur between
ing that change in one parameter will lead to a these random effects until the log likelihood is
corresponding change in the other parameter. Sam- changed significantly.
pling correlations between parameters are therefore Heritability estimates did not differ significantly
often largely negative and in the situation of variance between breeds for further performance traits. There-
components estimation, a considerable cross-substi- fore, heritability estimates are only presented from
tution between parameters is possible before a the combined analysis. The second growth rate trait
significant change in the log likelihood occurs. recorded in the test station ADG2 was lowly
2
The analysis showed that sampling correlations heritable h : 0.13 Table 4. This trait was char-
did not change significantly after the data set reached acterised by a high variation. This drastic increase in
a size of approximately 1000 animals in both breeds. variation might be caused by random differences in
However, for Large White the sampling correlation gut fill at test begin and test end. Given the short
between heritability and litter effect estimates was testing period of four weeks these random differ-
2 0.43 for the total data set while it increased in
ences have a large influence on recorded weight gain magnitude 20.54 for the reduced data set. The
and therefore growth rate. It was not possible to higher sampling correlation implies that cross-substi-
account for these effects in the model and the tution between the two parameters was larger in the
environmental variation was therefore increased. reduced data set. This helps to explain the change in
Heritability estimates were moderate for feed intake
2 2
h : 0.23 but low for feed conversion ratio h : 0.13. These estimates of heritabilities were similar
to estimates presented by Cameron and Curran
Table 4
2 2
Heritabilities h , litter effects c both with standard errors s.e.
1994 but generally lower than heritabilities pre-
a
and phenotypic variance for production traits full data set
¨ sented by Hofer and Schworer 1995, de Vries et al.
2 2
2 2
2
Trait h
s.e. of h c
s.e. of c s
1994b and Mrode and Kennedy 1993 indicating
p
that the testing period was too short to measure
ADG1 0.27
0.05 0.15
0.03 4210
growth rate and feed conversion ratio reliably. In
ADG2 0.13
0.04 0.08
0.03 29024
FDINT 0.23
0.04 0.140
addition, Archer et al. 1997 showed that a 70-day
FCR 0.15
0.04 0.265
test was required to measure growth rate and feed
a
For abbreviation of traits see Table 1.
conversion ratio without compromising the accuracy
S . Hermesch et al. Livestock Production Science 65 2000 239 –248
245
Fig. 1. Log likelihood values for different estimates of heritability and litter effects for growth rate until 18 weeks in Large White full data set.
of the measurement in cattle whereas a 35-day test were slightly lower FDP2: 0.46, FD3 4: 0.45.
was sufficient to record feed intake reliably. Similar high heritability estimates were obtained by
Li and Kennedy 1994 analysing four breeds using 3.2. Carcass traits
large data sets and by de Vries et al. 1994a who analysed Large White and Landrace data in an
Backfat characteristics measured on the live ani- Australian herd. Overall these estimates are in agree-
mal with real time ultrasound LFDP2, LFD3 4 ment with average heritability estimates presented in
2
were highly heritable h : 0.60 and 0.62 Table 5. the reviews of Stewart and Schinckel 1989 and
Heritability estimates for these two backfat measure- Ducos 1994 cited in Sellier 1998.
ments, recorded with the Hennesy Chong machine Heritability estimates were also higher for muscle
depth recorded with real time ultrasound than for muscle depth recorded with Hennesy Chong equip-
Table 5
2 2
2 2
Heritabilities h and litter effects c ; second row with standard
ment h : 0.21 for LMD3 4; h : 0.02 for MD3 4.
a
errors s.e. and phenotypic variance for carcass traits
The low heritability for muscle depth recorded with
2 2
2 2
2
Trait h c
s.e. of h c s
Hennesy Chong was mainly due to a five times
p
larger residual variance and as a consequence use of
LFDP2 0.60
0.05 4.51
this trait in genetic evaluations is limited. No esti-
LFD3 4 0.62
0.05 4.49
LMD3 4 0.21
0.04 15.10
mates of heritabilities were found in the literature for
FDP2 0.46
0.06 7.20
muscle depth.
FD3 4 0.45
0.07 6.34
Heritabilities differed substantially between both
MD3 4 0.02
0.05 60.87
breeds for weight of the whole back leg with
BLW 0.22
0.05 1.340
estimates of 0.08 for Large White and 0.46 for
0.14 0.03
LMW 0.38
0.07 0.450
Landrace. This weight measurement includes skin,
0.13 0.03
bones and fat of the back leg and is an indicator of
a
For abbreviations of traits see Table 1.
the growth rate of an animal. Differences in
246 S
. Hermesch et al. Livestock Production Science 65 2000 239 –248
heritabilities for this trait between breeds therefore comparison to Large White. Luxford 1995 ex-
correspond to differences in heritability estimates for amined the effect of the halothane gene on meat
growth rate. Heritabilities for lean meat of the whole quality characteristics in these two breeds. The
back leg were 0.27 for Large White and 0.59 for halothane gene did not influence pH24 but influenced
Landrace. Pooled estimates were 0.22 for back leg colour and drip loss percentage. This finding sup-
´ weight and 0.38 for lean meat weight Table 5.
ports results presented by Gueblez et al. 1995, These estimates not only correspond to results from
McPhee and Trout 1995, Peschke et al. 1993 and ¨
Ollivier 1983 and Sonnichsen and Kalm 1984 Wittmann et al. 1993. Furthermore, McPhee and
who analysed ham weight but also correspond to Trout 1995 found that the unfavourable effect of
estimates from recent studies by Scholz and Triebler the halothane gene is of higher magnitude in a line
1992 and Hofer et al. 1992 who analysed the that had been selected for lean meat growth. Since
weight of primary cuts. Landrace is a slightly leaner breed and has a higher
incidence of the halothane gene, extreme cases of 3.3. Meat quality traits
PSE meat characterised by a high drip loss per- centage are anticipated.
Both pH measurements had low heritability esti- Heritability estimate for intramuscular fat content
mates 0.15 for pH45 and 0.14 for pH24 Table 6. of 0.35 was lower than estimates presented by
However, in addition to these low heritabilities, Cameron 1990, Lo et al. 1992, de Vries et al.
additive genetic variances were low for both pH 1994b and Goodwin 1995. This low heritability
measurements especially for pH24, which limits might be due to differences in the two measurement
possible genetic progress in these traits. Heritability techniques applied in this study. However, no differ-
estimates were 0.29 and 0.30 for the two colour ences were found in coefficient of determination for
measurements and 0.23 for drip loss percentage. intramuscular fat content within measurement tech-
Overall, these estimates are in agreement with esti- niques and variance components were not signifi-
mates presented in recent studies by Lo et al. 1992, cantly different for the two techniques.
Bidanel et al. 1994a,b, de Vries et al. 1994b, ¨
Goodwin 1995, and Hofer and Schworer 1995 as well as mean heritability estimates presented in the
4. Conclusions