M . Serrano et al. Livestock Production Science 67 2001 253 –264
257
two or more lambs born for both breeds. Age at Latxa and smaller |100 ml for the 1st and 4th TD.
lambing was defined as a four class fixed effect. In TD protein percentage showed an opposite trend to
Latxa ewes, classes were defined in years 1, 2, 3 that observed for TD milk yield. There was an
and 4 or more years. In the Manchega breed, these increase of protein percentage from the 1st to the 4th
four classes were considered in months 15, 16– test day records approximately 1. Similar trends
17, 18–20 and 20 months. of milk yield and protein percentage along lactation
have been found in Churra ewes Fuertes et al., 1998.
3. Results and discussion Variance component estimates residual R and
additive genetic G associated with the first lactation Average days in milk DIM, milk yield and
TD milk yields under multivariate animal models are protein percentage only in Manchega breed for
shown in Table 3. In both data sets and breeds a each TD and breed in the first lactation data set are
decrease of residual variance from the 1st to the 4th presented in Table 2. The average DIM at each TD
TD was observed. This pattern is similar to that was larger in the Manchega breed. The standard
found in cattle by Swalve 1995. The first TD had deviation of DIM was high in all TD for Latxa ewes,
the highest residual variance estimate in both breeds due to the less strict criteria used to define the
and models. Differences were more evident in Latxa intervals between TD. Average TD milk yields in the
data. Uncontrolled environmental factors seem more Manchega breed were always higher than in the
likely to occur at the onset of lactation and might be Latxa breed, due to differences in lactation stage at
a possible explanation for this observed trend. the time of collection. These differences were higher
In the Latxa breed, residual variances estimated for the 2nd and 3rd TD 140 ml than for the 1st and
under the MHTD model were smaller than those 4th TD 75 and 40 ml, respectively. Standard
estimated with the MHYS model. This indicates, as deviations for the 2nd and 3rd TD milk yields in the
expected, that more environmental variation is re- Manchega breed were 160 ml larger than those for
moved by considering comparisons of animals within
Table 3 Test day milk yield ml residual R and genetic G variances diagonal and covariances above diagonal estimated under multivariate
animal models including HYS MHYS or HTD MHTD contemporary groups with first lactation data of Latxa and Manchega ewes Model
TD Latxa
Manchega 1
2 3
4 1
2 3
4 R
MHYS 1
134,648 47,732
26,841 14,188
118,286 48,325
37,205 28,219
2 90,889
35,809 20,914
101,217 47,096
36,299 3
57,602 17,997
73,621 38,828
4 49,396
60,737 MHTD
1 123,314
45,063 25,021
10,918 116,619
50,999 37,876
27,258 2
80,059 30,644
16,462 101,589
47,715 35,200
3 51,672
13,923 75,260
37,998 4
42,983 59,700
G MHYS
1 34,361
29,726 24,804
18,651 36,744
31,360 19,644
9884 2
30,414 23,716
17,366 35,038
26,304 15,472
3 24,477
21,893 26,146
16,338 4
21,644 12,470
MHTD 1
34,001 25,712
21,724 18,301
38,276 31,677
20,378 12,149
2 24,066
19,836 17,347
33,056 26,451
17,452 3
20,263 20,360
24,563 17,424
4 22,069
13,100
258 M
. Serrano et al. Livestock Production Science 67 2001 253 –264
the day on which the record is made than by In the Latxa breed, the TD milk yield genetic
comparing animals at the same period of parturition variance estimates were higher under the MHYS
Rekaya et al., 1999. However, in the Manchega model than under the MHTD model, these differ-
breed, the differences in residual variances between ences being more evident for the second and third
alternative models were not so clear. The first and TDs. Only the fourth TD showed a smaller genetic
fourth TDs showed smaller residual variances under variance. In the Manchega breed, the 2nd and 3rd
the HTD model, however these were higher for the TD records showed greater genetic variance esti-
second and third TDs. In the Latxa breed, HYS CG mates with the HYS model. On the other hand,
was widely spread over 5 months of the lambing higher genetic variance estimates were found for the
period two lambing periods in the year: September– 1st and 4th TD records under the HTD model. In
December and January–June. However, in Man- general, higher genetic variance estimates were
chega ewes, parturition season was defined in obtained in the Manchega than in the Latxa breed.
months over the year. In the first case, the MHTD Genetic covariance estimates, as for the residual
model captured environmental variation that was not ones, were higher between adjacent TD records. In
accounted for by the MHYS model. In Manchega, the Latxa breed, these values were higher for the
the MHTD model did not show the same advantage, MHYS model than for the MHTD model. In the
because of the greater similarity between the CG, Manchega breed, the opposite trend was observed.
HYS and HTD definitions. Thus HTD models seem Higher genetic covariances were estimated under the
to decrease residual variances in relation to HYS MHTD model than under the MHYS model.
models, depending on the definition of HYS CG. Although at the level of variance component
Residual covariance estimates showed higher val- estimation some advantages of HTD over HYS were
ues between adjacent TD records under both models observed in this study, it is important to take into
and in both breeds. These covariances had higher account the reduction in CG size produced by fitting
values in the MHYS model than in the MHTD model HTD rather than HYS effects. Using HTD resulted
for the Latxa breed but not in Manchega, where both in considerably smaller CG subclasses which would
models showed similar residual covariance values. affect sire evaluation. Meyer et al. 1989 found in
Genetic variances had a similar trend along lacta- their study a great increase in terms of percent
tion as observed for the residual variances. In both information lost of single sire subclasses when fitting
breeds and models, genetic variance estimates de- HTD rather than HYS effects. In our case, the CG
creased with time, from the 1st to the 4th TD record. was similarly affected in the multivariate TD ap-
However, these changes among TDs were smaller proach. In Latxa, changes in CG size ranged from 24
than in the previous case. Again, the 1st TD record to 58 CG with less than six records when fitting
had the highest variance value. In this case, results HYS or HTD effects, respectively. In Manchega, this
did not agree with those obtained in cattle Swalve, reduction was smaller due to more stringent data
1995; Kettunen et al., 1997; Rekaya et al., 1999, selection, which included only HYS CG with at least
where additive genetic variance estimates increased six records. In this case, 8 of CG presented less
from the 1st to the 4th TD milk yield. This could be than six records in the HTD model. This last case is
due to the fact that, in cattle, there is no suckling not representative of practice, because, in the whole
period because calves are weaned 2 or 3 days from milk recording program of this breed, there are
parturition. Thus, the first TDs were collected in the numerous herds of small size, which also implies
ascendant phase and peak of the lactation curve. small CG sizes. This is avoided in the Manchega
However, generally, TD collection in sheep begins selection scheme by fitting HYS CG with mobile
after the suckling period average 30 days from periods of parturition Jurado et al., 1995. However,
parturition and, therefore, the first TDs are collected this kind of grouping is not possible for HTD effects.
after the peak of the lactation curve. Rekaya et al. Therefore, one of the most important disadvantages
1999 reported a decrease in genetic variance from in considering a multivariate approach of TD models
the 4th to the 10th TD. Our results might be in in sheep is the great reduction in CG size when the
accordance with this last estimate. HTD effect is fitted. Also, variance component
M . Serrano et al. Livestock Production Science 67 2001 253 –264
259 Table 4
estimates was observed. Residual variances increased
Residual and genetic variances diagonal and covariances above
along lactation, except for the 2nd TD, which
diagonal test day protein percentage, estimated under multivariate
showed the lowest residual variance for both models.
animal models including HYS MHYS and HTD MHTD
Additive genetic variance presented the same pattern,
contemporary groups with first lactation data of Manchega ewes
increasing from the 1st to the 4th TD.
Model TD
Manchega
In this case, as expected, MHYS residual vari-
1 2
3 4
ances were slightly higher than those of MHTD for
Residual MHYS
1 0.212
0.064 0.039
0.025
all TD records. Genetic variances showed an oppo-
2 0.180
0.054 0.041
site trend. Slightly higher values were obtained for
3 0.266
0.114
the MHTD model than for the MHYS model. Using
4 0.356
the HTD effect seems to enhance the detection of
MHTD 1
0.205 0.058
0.036 0.026
differences among animals at the environmental
2 0.175
0.054 0.043
level, but not at the genetic level for this trait.
3 0.256
0.113
Residual covariances had similar trends as those
4 0.350
observed for TD milk yield records. Higher values
Genetic MHYS
1 0.058
0.040 0.044
0.041
were found for adjacent TD records. However,
2 0.072
0.080 0.067
genetic covariance estimates showed different be-
3 0.089
0.075
havior. Genetic covariances among the 1st and other
4 0.118
TDs tended to be smaller than those estimated
MHTD 1
0.058 0.045
0.047 0.043
between the 2nd and 3rd, 2nd and 4th and 3rd and
2 0.074
0.081 0.065
4th TDs.
3 0.092
0.076
Heritabilities and genetic correlations among TD
4 0.113
milk yield and TD protein percentage only in Manchega breed records for both breeds using
estimation seems to be affected when a HYS or a alternative MHYS and MHTD models are shown in
HTD CG is fitted. Models with HTD CG showed Table 5. Standard deviations ranged from 0.01 to
smaller residual variances in most cases. 0.09.
Table 4 shows TD protein percentage variance In the Latxa breed, no substantial differences in
component estimates in the Manchega breed under genetic parameter estimates were observed between
the two models considered, MHYS and MHTD. An alternative models. Similar heritability and genetic
opposite trend to that of TD milk yield variance correlations were found with the MHYS and MHTD
Table 5
2
Test day milk yield ml and protein percentage genetic parameters: heritability h at diagonal and genetic correlations above diagonal r
g
estimated under multivariate animal models including HYS MHYS and HTD MHTD contemporary groups with first lactation data of
a
Latxa and Manchega ewes Model
TD Latxa
Manchega Milk yield ml
Milk yield ml Protein percentage
1 2
3 4
1 2
3 4
1 2
3 4
MHYS 1
0.20 0.92
0.85 0.68
0.23 0.87
0.63 0.46
0.21 0.61
0.60 0.50
2 0.25
0.87 0.67
0.25 0.87
0.74 0.28
0.99 0.72
3 0.29
0.95 0.26
0.90 0.25
0.73 4
0.30 0.17
0.25 MHTD
1 0.21
0.90 0.82
0.66 0.24
0.89 0.66
0.54 0.22
0.67 0.64
0.53 2
0.23 0.89
0.75 0.24
0.92 0.84
0.29 0.98
0.70 3
0.28 0.96
0.24 0.97
0.26 0.74
4 0.33
0.18 0.24
a
Standard deviations for the heritability and genetic correlations ranged from 0.01 to 0.09.
260 M
. Serrano et al. Livestock Production Science 67 2001 253 –264
models. The first TD milk yield record showed the decreasing to 0.66 when records were three TD
lowest heritability value in both cases, due to the apart. In Manchega, genetic correlations between
higher residual variance, which agrees with several adjacent TD took values of 0.87, decreasing to 0.46
results in cattle Meyer et al., 1989; Wiggans and when records were three TD apart. Higher genetic
Goddard, 1997; Vargas et al., 1998; Rekaya et al., correlations in Latxa could be due to the less strict
1999. These results confirm that there is more criteria used to consider TD intervals in this breed.
variation not accounted for by the models in the In this case, the period between consecutive TD
initial stage of lactation. As reported in cattle studies, records are more spread over time 20 to 40 days
heritabilities were generally lower in early lactation required between successive TD than those of
Danell, 1982; Wilmink, 1987. Heritability values Manchega records 25 to 35 days. Therefore, in the
increased along lactation, the last 4th TD being the Latxa breed, TD traits were mixed because the time
one with the highest value 0.30 and 0.33 in MHYS interval for a TD trait was very long and thus genetic
and MHTD models, respectively. The first and correlations were higher. Higher genetic correlations
second TD records showed heritability values near among TD records were obtained by Barrillet and
that estimated for standardized 120 day milk yield, Boichard 1994 with a sire model, concluding that
0.23 Ugarte et al., 1996. However, the 3rd and 4th considering TD data as repeated measures of the
TD heritability values were clearly higher than those same trait seems to be a reasonable hypothesis. In
estimated for the 120 day milk yield. the present case, the results did not lead to the same
Heritability values in the Manchega breed pre- conclusion. Also, our estimates were lower than
sented some differences to those found in Latxa. In those obtained by Georgoudis et al. 1997 in a study
this breed, heritability of the 1st, 2nd and 3rd TD with a single herd of Chios and Kymi sheep breeds,
records showed similar values within and across which found genetic correlation values among TD
models. In this case, the 4th TD presented the milk yield records ranging from 0.73 to 0.99.
smallest heritability values, 0.17 and 0.18 in MHYS Heritability for protein percentage TD records in
and HTD models, respectively, due the smaller the Manchega breed showed slightly higher values
residual and genetic variances. Also, insubstantial with the MHTD model. For this trait, the first TD
differences were observed among genetic parameter record presented the smallest heritability value in
estimates under alternative models. These heritability both alternative models. All heritability values were
estimates were smaller than those of the standardized clearly lower than estimated in this breed for the
120 day milk yield, 0.33 with REML Serrano et al., standardized 120 day protein percentage, 0.35 with
1996b and 0.28 with a Bayesian approach Jurado et REML Serrano et al., 1996b, and similar to the
al., 1997. estimate with a Bayesian approach, 0.29 Jurado et
These results are in accordance with those ob- al., 1997. Also, Barrillet and Boichard 1994 found
tained by Barrillet and Boichard 1994. These higher heritability for complete lactation 0.45 than
authors found lower heritability estimates for TD for TD traits 0.35. Genetic correlations showed the
milk yield average 0.25 than for complete lactation same trend as for milk yield. Higher genetic correla-
0.30. Our TD heritability estimates were similar to tions were obtained between adjacent TDs than for
those obtained by Georgoudis et al. 1997, which two or three TDs apart.
ranged from 0.19 to 0.26. However, these authors Table 6 shows variance components and genetic
found lower heritability for commercial milk yield parameters heritability and repeatability estimated
0.16 than for TD traits. These estimates had large under alternative UHYS and UHTD univariate ani-
standard errors, which ranged from 0.12 to 0.15. mal models considering TD milk yield and protein
In both breeds and with the two models, genetic percentage only in Manchega breed records as
correlations were always large between adjacent TD repeated measures of the same trait within first
for milk yield. Generally, genetic correlations were lactation. Standard deviations of genetic parameters
higher in the Latxa than in the Manchega breed ranged from 0.005 to 0.02.
under both models. In the Latxa breed, genetic Small changes in genetic and permanent environ-
correlations for adjacent TDs were at least 0.87, mental variances across alternative UHYS and
M . Serrano et al. Livestock Production Science 67 2001 253 –264
261 Table 6
2 2
2
Test day milk yield ml and protein percentage residual s , genetic s and permanent environmental s variances and genetic
e a
c 2
parameters, heritability h and repeatability R estimated under univariate animal models including HYS UHYS and HTD UHTD
a
contemporary groups with first lactation data of Latxa and Manchega ewes
2 2
2 2
Breed Trait
Model s
s s
h R
e a
c
Latxa Milk yield
UHYS 75,230
24,847 27,000
0.19 0.40
UHTD 58,014
23,973 27,883
0.21 0.47
Manchega Milk yield
UHYS 86,093
22,103 29,786
0.16 0.37
UHTD 58,806
24,723 40,342
0.20 0.52
Protein UHYS
0.322 0.065
0.024 0.16
0.21 percentage
UHTD 0.226
0.071 0.050
0.20 0.35
a
Standard deviations for heritability and C squared ranged from 0.005 to 0.02.
UHTD models were observed in the Latxa breed. In the case of protein percentage, TD records
However, the estimated residual variance with the residual variances were lower in the UHTD model.
UHTD model was smaller, as reported by Meyer et However, additive genetic and permanent environ-
al. 1989. Heritability values were similar to those mental variances are greater in the UHTD model
of the 1st TD in the multivariate analysis, 0.19 and than in the UHYS model. Heritability estimates were
0.21 for UHYS and UHTD, respectively. The re- similar to those obtained for TD milk yields, 0.16
peatability value was higher in the UHTD model and 0.20 for UHYS and UHTD respectively, and
0.47 than in the UHYS model 0.40, and superior smaller than values obtained in the multivariate
to the estimate from a repeatability animal model approach.
along lactation 0.42 Ugarte et al., 1996. In Churra ewes with a similar repeatability animal
In Manchega data, changes in variance component model, Baro et al. 1994 found a heritability value
estimates with UHYS and UHTD models were of 0.35 for test day milk yield. The inclusion of
observed. Also in this case, the residual variance for natural mating and AI sires in the pedigree file
the UHTD model was the lowest. However, perma- contributes to a better heritability estimation due to
nent environmental variance showed a substantial more
relationships being
considered. However,
increase from the UHYS to the UHTD model. In this heritability of protein percentage 0.13 was smaller
case, there were large differences between different than values found in Manchega ewes. In Chios sheep
models in heritability estimates. The UHTD model Georgoudis et al., 1997, the heritability estimate
showed a higher value of this parameter 0.20 than for the first four TD records in a repeatability animal
that observed with the UHYS model 0.16. Both model was 0.28 and for the seventh TD record 0.17.
heritability estimates were clearly smaller than those Repeatability estimates were 0.39 and 0.40 in each
of the multivariate approach. Repeatability estimates case.
presented more differences across models than in the Under the univariate model, the CG size reduction
case of Latxa. Repeatability was much higher in the had less impact. When a univariate approach is
UHTD model than in the UHYS model in the considered by taking TD records as repeated mea-
Manchega breed, 0.52 and 0.37, respectively. sures of the same trait within lactation, HTD CG
Variance estimates were higher than those found includes all animals recorded on the same day
by El-Saied et al. 1998 under a univariate TD independent of TD number. Thus HTD classes have
model with HTD CG. Heritability and repeatability a larger size than in the multivariate approach. In
values for TD milk yield were 0.14 and 0.44, Latxa, in the univariate TD approach, there were 8
respectively. In this case, HTD CG was not nested to and 33 of CG with less than six records when HYS
TD number and multiple lactations were considered and HTD effects were fitted respectively. In Man-
in the analysis. A single permanent environmental chega, 2 of CG groups had less than six records
effect was fitted to account for TD repeated measures under the UHTD model.
within and among lactations. The higher genetic variances and heritabilities
262 M
. Serrano et al. Livestock Production Science 67 2001 253 –264
found under the multivariate than under the uni- using a multiple trait approach and a method to
variate approach, in both breeds, traits and models, combine all information in a single animal breeding
could be due to a better adjustment of environmental value may be a reason to use a repeatability model
effects in the first case. CG and regression coefficient instead. However, the pattern in genetic correlations
in DIM were different effects for each TD in among different TD traits suggests that a multiple
multivariate models, however in the univariate ap- trait approach is more accurate than the repeatability
proach, these are the same effects for all TD records. model. The use of random regression models RRM
Therefore, multivariate models might fit the data or covariance functions CF would also allow
better and subsequently higher heritability values are acknowledgement of changes in the trajectory of
obtained. random effects genetic and or environmental and
their associated covariances with time. However, the advantage of RRM or CF over the multi-trait
4. Conclusions approach in terms of goodness of fit and computing