Published January 20, 2015
Genomic selection for the improvement of meat quality in beef
E. C. G. Pimentel1 and S. König
Department of Animal Breeding, University of Kassel, 37213 Witzenhausen, Germany
ABSTRACT: Selection index theory was used to
compare different selection strategies aiming at the
improvement of meat quality in beef cattle. Alternative
strategies were compared with a reference scenario with
three basic traits in the selection index: BW at 200 d
(W200) and 400 d (W400) and muscling score (MUSC).
These traits resemble the combination currently used
in the German national beef genetic evaluation system.
Traits in the breeding goal were defined as the 3 basic traits
plus marbling score (MARB), to depict a situation where
an established breeding program currently selecting for
growth and carcass yield intends to incorporate meat
quality in its selection program. Economic weights were
either the same for all 4 traits, or doubled or tripled for
MARB. Two additional selection criteria for improving
MARB were considered: Live animal intramuscular fat
content measured by ultrasound (UIMF) as an indicator
trait and a genomic breeding value (GEBV) for the target
trait directly (gMARB). Results were used to estimate
the required number of genotyped animals in an own
calibration set for implementing genomic selection
focusing on meat quality. Adding UIMF to the basic
index increased the overall genetic gain per generation by
15% when the economic weight on MARB was doubled
and by 44% when it was tripled. When a genomic
breeding value for marbling could be estimated with an
accuracy of 0.5, adding gMARB to the index provided
larger genetic gain than adding UIMF. Greatest genetic
gain per generation was obtained with the scenario
containing GEBV for 4 traits (gW200, gW400, gMUSC,
and gMARB) when the accuracies of these GEBV were
≥0.7. Adding UIMF to the index substantially improved
response to selection for MARB, which switched from
negative to positive when the economic weight on MARB
was doubled or tripled. For all scenarios that contained
gMARB in the selection index, the response to selection
in MARB was positive for all relative economic weights
on MARB, when the accuracy of GEBV was >0.7.
Results indicated that setting up a calibration set of ~500
genotyped animals with carcass phenotypes for MARB
could suffice to obtain a larger response to selection
than measuring UIMF. If the size of the calibration set is
~2,500, adding the ultrasound trait to an index containing
already the GEBV would bring little benefit, unless the
relative economic weight for marbling is much larger
than for the other traits.
Key words: genetic gain, genomic breeding value, selection index, single nucleotide polymorphisms
© 2012 American Society of Animal Science. All rights reserved.
INTRODUCTION
The value of genomic selection (GS) for practical
application mainly depends on accuracies of genomic
breeding values (GEBV) and on length of generation
intervals. Hence, increase in genetic gain was observed
in studies evaluating GS in dairy cattle (e.g., Schaeffer,
2006) or in horse breeding programs (Haberland
et al., 2012). The disadvantage of long generation
1Corresponding
author: [email protected]
Received December 7, 2011.
Accepted April 26, 2012.
J. Anim. Sci. 2012.90:3418–3426
doi:10.2527/jas2011-5005
intervals is also relevant for beef cattle. In the case of
beef breeding programs, especially if focusing on the
improvement of carcass or meat quality, some more
arguments encourage German breeding organizations
to set up suitable strategies for GS. Arguments are
based on difficulties and high costs for data recording
in the field, poor genetic connectedness for routine
genetic evaluation, heavy usage of natural service
sires having low accuracies for conventional EBV, and
the fact that meat quality cannot be measured in the
selection candidate itself. In Germany up to now, no
efforts have been made to include meat quality in an
overall breeding goal. Considering the potential use of
3418
Genomic selection for improving meat quality
genomic data facilitated by suitable structures of largescale beef cattle herds as prevalent in East Germany, new
opportunities exist. The idea is to set up a calibration
set of a number of steers to derive SNP effects using
phenotypes for meat quality. This approach may be
worthwhile to follow especially for high heritability
traits such as meat quality (e.g., Ríos-Utrera and Van
Vleck, 2004). A large number of meat quality phenotypes
can be collected in a standardized environment covering
a limited number of large-scale beef cattle farms located
in East Germany.
The objective of this study was to compare accuracies
of selection indices and genetic gain of a direct GS
strategy focusing on meat quality with scenarios using
correlated phenotypes or correlated GEBV of traits
which are currently used for routine genetic evaluation.
Results will be used to determine the required number
of genotyped animals with meat quality phenotypes for
setting up a database for SNP effect estimation.
MATERIALS AND METHODS
Animal Care and Use Committee approval was not
obtained for this study because no animals were used.
Breeding Program Scenarios
Selection index theory (Hazel, 1943) was applied to
assess accuracy of selection and expected genetic gain
per generation for different selection strategies aiming
at the improvement of meat quality in beef cattle. An
R script (SIG.R) covering a range of possibilities with
respect to information sources in the selection index
(e.g., phenotypic records or GEBV) was written and is
available as supplementary online material. A detailed
description of the equations and assumptions used in
the calculations is described in the next section. Six
scenarios differing in the traits and respective type
of information included in the selection index were
considered. The spectrum of traits contemplated in
these scenarios was defined to depict a situation where
an established breeding program currently selecting for
growth and carcass yield intends to incorporate meat
quality in its breeding goal. As it is one of the main traits
determining meat quality, marbling score (MARB) was
used as the target trait to be included in the selection
program. Traits in the breeding goal (additive genetic
value of a young sire) were BW at 200 d (W200), BW
at 400 d (W400), muscling score (MUSC) and MARB.
Three situations with respect to the relative economic
weights of the traits in the breeding goal were considered.
In the first situation, all 4 traits had an economic weight
of 1 monetary unit per genetic standard deviation of the
3419
trait. In the other 2 situations the weight on MARB was
doubled and tripled.
As the first and basic scenario a selection index
composed of W200, W400 and MUSC was chosen. This
combination of traits represents the composition of the
relative breeding value for production traits in beef cattle
currently being applied by the national beef breeding
evaluation system in Germany (Ruten et al., 1997).
Muscling score is a subjective visual score, which is meant
to measure the amount of muscle in key-points of the
body, such as forearm, loin, rump and round (Buchanan
et al., 1982). In Germany, scores are assigned within
a range from 1 to 9. This standard pattern of 3 traits is
recorded on a relatively large sample of steers and young
bulls in the field, and on a limited number of selection
candidates reared on station. The main information source
for estimating EBV of the 3 traits for a male selection
candidate is his own performance. Including MARB in the
overall breeding goal and improving MARB of the young
sire without changes in performance tests implies to use
correlated response of the 3 basic traits W200, W400, and
MUSC measured as the performance of each bull.
In scenario 2, intramuscular fat content measured
by ultrasound (UIMF) was incorporated into the basic
index as an indicator trait for marbling. Including live
animal ultrasound measures of carcass yield or quality
or both can significantly increase the accuracy of EBV
for carcass traits of young bulls that do not have progeny
with phenotypic carcass data (Crews and Kemp, 2002).
In scenario 3, instead of the phenotype of an indicator
trait, a GEBV for the target trait itself was added to the
basic index. Because the type of information in this case
was not a phenotype but a GEBV, the additional trait will
be denoted gMARB. The idea of scenario 3 is based on
the possibility of forming a calibration set for GS of meat
quality traits using large beef herds in the eastern part of
Germany. For such new traits, conventional EBV do not
exist. Consequently, phenotypes instead of conventional
EBV from genotyped steers have to be used for estimating
SNP effects. The success of such a strategy has been
shown by Buch et al. (2012) using simulated data for low
heritability health traits in dairy cattle, and is even more
promising for highly heritable meat quality traits.
Scenario 4 contained the 3 traits from Scenario 1
plus both UIMF and gMARB. Scenario 4 was created
to assess the value of a direct genomic selection strategy
on MARB for a situation where a valuable indicator
trait is already implemented. Hence, results of scenario
4 also will be relevant for international beef breeding
organizations that have used UIMF as an indicator for
improving meat quality over decades.
In scenario 5, instead of phenotypic records, GEBV
for the 3 basic traits were available (i.e., gW200, gW400
and gMUSC). As with scenario 1, genetic correlations
3420
Pimentel and König
Table 1. Simulated scenarios with respect to the traits
included in the selection index
Scenario
Traits in the selection index1
1
W200 + W400 + MUSC
2
W200 + W400 + MUSC + UIMF
3
W200 + W400 + MUSC + gMARB
4
W200 + W400 + MUSC + UIMF + gMARB
5
gW200 + gW400 + gMUSC
6
gW200 + gW400 + gMUSC + gMARB
1Weight at 200 (W200) and 400 (W400) days, muscling score (MUSC),
marbling score (MARB), and intramuscular fat content measured by
ultrasound (UIMF). A ‘g’ before the abbreviation indicates a genomic
breeding value for the given trait.
between these GEBV and the target trait MARB
determine the rate of genetic gain in MARB.
The last scenario (scenario 6) contained the same
traits as scenario 5 (i.e., GEBV for all 3 basic traits) plus
gMARB. A compact overview of the different scenarios
simulated is presented in Table 1.
For all the scenarios in which the type of information
from one of the traits was a GEBV (scenarios 3 to 6), 9
analyses were performed ranging the accuracy of the
GEBV from 0.1 to 0.9 in steps of 0.1. Here the accuracy
RI WKH *(%9 LV GH¿QHG DV WKH FRUUHODWLRQ EHWZHHQ WKH
GEBV and the true breeding value for the corresponding
trait. The information source for all traits in all cases
was assumed to be a single record (either a phenotypic
observation or a GEBV) from the own selection candidate.
Selection Index Calculations
Following standard selection index procedures, for
each scenario and assumed accuracy of GEBV (where
relevant), matrices P, G and C were set up. Matrix
P is the (co)variance matrix between all components
of the selection index in the given scenario, matrix C
is the genetic (co)variance matrix between all traits
in the breeding goal, and matrix G is the matrix of
covariances between the components of the selection
index and the additive genetic values for the traits in
the breeding goal. When the type of information in the
index was the phenotype, elements of P and G were
calculated as a function of phenotypic and genetic
constants relative to the traits, as described by Hazel
(1943). Assumed genetic and phenotypic parameters
XVHGLQWKHFDOFXODWLRQVZHUHWDNHQIURPWKHOLWHUDWXUH
.RRWVHWDODE*UHJRU\HWDO/HDÀHWHW
al., 1996; Archer et al., 2004; MacNeil and Northcutt,
2008) and are presented in Table 2. When the type
of information in the index was a GEBV, elements
of the P and G matrices were computed as described
E\ 'HNNHUV DFFRUGLQJ WR WKH GHULYDWLRQV LQ
Lande and Thompson (1990). Assuming a single own
record from the selection candidate and following
WKH HTXDWLRQV IURP 'HNNHUV WKH FRYDULDQFH
between a phenotype for trait i (Pi) and a genomic
breeding value for trait j (GEBVj) to be entered in
matrix P is equal to:
2
Cov ( Pi , GEBV j ) = rMG
ρ Gij σ Gi σ G j
j
Where rMG is the accuracy of the GEBV as a predictor of
WKHWUXHEUHHGLQJYDOXHDVGH¿QHGSUHYLRXVO\ȡG is the
genetic correlation between the traits i and jDQGıG is
the genetic standard deviation of the trait. Analogously,
the covariance between a phenotype and a GEBV for the
same trait i is equal to:
2
Cov ( Pi , GEBVi ) = rMG
σ G2i
i
$V SRLQWHG RXW E\ 'HNNHUV WKH *(%9 LV
incorporated into the index as a correlated trait with
a heritability of 1. Hence, the 2 equations used for
setting up matrix P as indicated above are also used
for computing the elements of matrix G giving the
covariance between the GEBV as a component of the
Table 2.$VVXPHGSKHQRW\SLFYDULDQFHVKHULWDELOLWLHVGLDJRQDOJHQHWLFȡGDERYHGLDJRQDODQGSKHQRW\SLFȡP,
below diagonal) correlations between the simulated traits
Trait
BW at 200 d (W200)
BW at 400 d (W400)
Muscling score (MUSC)
Intramuscular fat content (UIMF)
Marbling score (MARB)
0.241
2
0.103
4
2
Phenotypic variance ( σ 2P )
6252
1Koots
et al. (1994a).
et al. (1994b).
3Gregory et al. (1995).
4Archer et al. (2004).
5MacNeil and Northcutt (2008).
6/HDÀHWHWDO
2Koots
W200
2
0.331
0.123
4
0.142
1,4442
W400
0.313
0.143
0.643
3
2.023
MUSC
4
4
0.385
0.626
0.945
UIMF
2
2
3
0.665
0.455
0.615
MARB
3421
Genomic selection for improving meat quality
index and the additive genetic value of the trait in the
breeding goal. Finally, assuming that the proportion of
JHQHWLFYDULDQFHH[SODLQHGE\WKHPDUNHUVLQWKHSDQHO
is the same and equal to 1 for all traits, the element of
matrix P giving the covariance between a GEBV for
trait i and a GEBV for trait j is equal to:
2
Cov(GEBVi ,GEBV j ) = rMG
r2 ρ σ σ
i MG j Gij Gi G j
)ROORZLQJ'HNNHUVLIPDUNHUVDUHUDQGRPO\
distributed across the genome, the expected proportion
RIJHQHWLFYDULDQFHH[SODLQHGE\WKHPDUNHUVLVWKHVDPH
for both traits. Having set up all the elements of matrices
P, G and C VHOHFWLRQ LQGH[ FRHI¿FLHQWV E-values)
were calculated as b PíGw, where w is the vector of
relative economic weights expressed in monetary units
per measurement units of the traits. The variances of
the index (I) and of the aggregate genotype (H) were
calculated as σ 2I b’Pb and σ 2H w’Cw. The accuracy
of the index (i.e., the correlation between the index and
the aggregate genotype) was calculated as:
RIH =
σI
σH
The monetary overall genetic gain per generation
was calculated as ΔG = (i )RIH σH and the response to
selection per generation for each trait was calculated as:
S=
i
σI
b’G
where i is the selection intensity, which here was
assumed to be 1.
Size of Calibration Set
Daetwyler et al. (2010) proposed an equation for
calculating the expected accuracy of GEBV predicted
with a genomic linear model. Following their equation,
the expected number of genotyped animals in the
calibration set (Np) necessary to achieve a given level of
accuracy of GEBV was calculated as follows:
NP =
2
rMG
Mˆ e
2
2
)
h (1 − rMG
where h2 is the heritability of the trait and M̂ e is an
estimate of the number of independent chromosome
segments, calculated as:
Mˆ e =
2N e L
log(4 N e L)
where L is the genome length in Morgans and Ne is
the effective population size (Goddard, 2009). Here
WKHFDOLEUDWLRQVHWLVGH¿QHGDVWKHJURXSRIJHQRW\SHG
animals with phenotypic information used for estimating
the effects of single nucleotide polymorphism (SNP) to
be included in the prediction equation for GEBV.
RESULTS AND DISCUSSION
Accuracy of the Index
The accuracy of the index (RIH) for different
scenarios, accuracies of GEBV (rMG) and relative
economic weights on marbling are presented in Table 3.
As expected, RIH increased with increasing rMG for
all scenarios where GEBV were included in the index
(scenarios 3, 4, 5, and 6). Changes in RIH when altering
rMG were substantially large for scenarios 5 and 6, which
included 3 and 4 genomic traits in the index, respectively,
compared with scenarios 3 and 4 where only gMARB
was considered. In scenarios 1, 2 and 5, where gMARB
was not included as an information source in the index,
lower RIH was observed when the economic weight on
marbling was doubled or tripled.
In the studies by König and Swalve (2009) and by
Haberland et al. (2012), values of RIH incorporating
genomic and phenotypic information of selection
FDQGLGDWHV LQ WKH LQGH[ ZHUH XVHG WR PDNH GHFLVLRQV
regarding the necessity of a central station test for
potential bull dams and young stallions. Following
results from both studies for moderate to high rMG, it
Table 3. Accuracy of the index (RIH) for different sce
narios, accuracies of genomic breeding values (GEBV)
and relative economic weights on marbling (w1 VDPH
w2 GRXEOHGw3 WULSOHG
Scenario
w1
1
2
3
4
5
6
w2
1
2
3
4
5
6
w3
1
2
3
4
5
6
Accuracy of GEBV
0.4
0.5
0.6
0.1
0.2
0.3
0.8
0.9
0.60
0.61
0.60
0.61
0.12
0.12
0.60
0.61
0.60
0.61
0.24
0.25
0.60
0.61
0.60
0.61
0.36
0.36
0.60
0.61
0.61
0.62
0.46
0.60
0.61
0.61
0.62
0.56
0.60
0.61
0.62
0.62
0.65
0.60
0.61
0.62
0.63
0.60
0.61
0.63
0.63
0.80
0.84
0.60
0.61
0.64
0.64
0.86
0.92
0.45
0.52
0.46
0.53
0.09
0.11
0.45
0.52
0.53
0.18
0.21
0.45
0.52
0.49
0.54
0.31
0.45
0.52
0.51
0.56
0.35
0.41
0.45
0.52
0.54
0.58
0.42
0.51
0.45
0.52
0.60
0.49
0.61
0.45
0.52
0.61
0.63
0.55
0.45
0.52
0.65
0.66
0.61
0.80
0.45
0.52
0.66
0.90
0.32
0.46
0.33
0.06
0.10
0.32
0.46
0.36
0.48
0.12
0.19
0.32
0.46
0.39
0.50
0.18
0.29
0.32
0.46
0.44
0.53
0.24
0.38
0.32
0.46
0.50
0.56
0.29
0.48
0.32
0.46
0.56
0.61
0.34
0.58
0.32
0.46
0.63
0.65
0.38
0.68
0.32
0.46
0.43
0.32
0.46
0.89
3422
Pimentel and König
was shown that testing genotyped selection candidates
on station to improve RIH would not be needed. Also,
when referring to scenario 6 and equal economic
weights in the present study, which extends the
problem to several (partly antagonistic) genomic traits
in the index, use of GEBV with rMG ≥ 0.7 resulted in
acceptable values for RIH ranging from 0.76 to 0.92.
Figure 1. Expected overall genetic gain per generation for different scenarios with respect to the traits included in the index: weight at 200 d (W200)
and 400 d (W400), muscling score (MUSC), marbling score (MARB) and
intramuscular fat content measured by ultrasound (UIMF). A ‘g’ before the
abbreviation indicates a genomic breeding value for the given trait.
Following these results, further performance tests in
central stations would not be required.
Overall Genetic Gain
The expected overall genetic gain per generation
from the different scenarios with respect to the traits in
the index and the relative economic weights is presented
in Figure 1. The results in terms of genetic gain depicted
in Figure 1 reflect the same trends as observed for
accuracies of indices (Table 3). Therefore most of the
discussions regarding the comparison between scenarios
will be focused on genetic gain, making references to
Figure 1. Similar patterns could be observed in the three
situations considered: when the economic weights were
the same (i.e., one monetary unit per genetic standard
deviation of the trait), or the economic weight on MARB
was doubled or tripled. Because W200, W400 and
MUSC are all positively correlated with each other and
negatively correlated with MARB, when the economic
weights were all the same most of the emphasis was
placed on growth and muscling. Therefore only minimal
differences were observed among the different strategies
employed in scenarios 1 to 4 for equal economic weights
of the traits used in the overall breeding goal. These
differences were magnified and a clearer distinction
between all the simulated scenarios became evident
when the economic weight on MARB was doubled and
even more pronounced when it was tripled.
Adding UIMF to the basic index increased the overall
genetic gain per generation by 15% when the economic
weight on MARB was doubled and by 44% when it was
tripled. This increase was determined by the improved
response to selection on MARB when UIMF was used as its
indicator. Sapp et al. (2002) conducted an experiment with
Angus bulls to investigate the relationship between these two
traits and found highly significant regression coefficients of
MARB from steer progeny on EBV for UIMF of the sires.
Strong linear associations between changes in the EBV of
the sire for UIMF and progeny phenotype for MARB were
also found by Crews et al. (2004) with Simmental field data.
In a simulation study, Kahi and Hirooka (2005) compared
a number of selection strategies using or not ultrasound
information. They reported increases in genetic gain ranging
from 17 to 43% when carcass traits measured by ultrasound
scanning on live animals were used as an additional source
of information in the selection index. When a genomic
breeding value for marbling could be estimated with an
accuracy of 0.5, then adding gMARB to the index provided
larger genetic gain than adding UIMF. Using the equation of
Daetwyler et al. (2010) and assuming a genome length of 30
Morgans and an effective population size of 100, as reported
by de Roos et al. (2008) for Australian Angus, the numbers
of animals in the calibration set needed to achieve the
3423
Genomic selection for improving meat quality
different levels of accuracy of GEBV considered here were
calculated (Table 4). Under these assumptions, an rMG = 0.5
for gMARB is expected to be achieved with a calibration set
of 473 genotyped animals (Table 4). MacNeil et al. (2010)
used a calibration set of 444 Angus sires to predict gMARB
and reported a genetic correlation (± SE) of 0.38 ± 0.10
between MARB and gMARB. Following Dekkers (2007)
the genetic correlation between MARB and gMARB is
equal to the rMG of gMARB. However, the equation used by
MacNeil et al. (2010) for predicting gMARB was derived
from conventional EBV and genotypes for 40 pre-selected
markers. Prediction equations derived from genotypes on
a denser panel of markers are expected to improve rMG.
Brito et al. (2011) showed with stochastic simulation that
when the marker density increased from 40k to 800k, rMG
moved from 0.39 to 0.48, for a heritability of 0.4 and using
EBV from 480 animals in the calibration set. Veerkamp et
al. (2011) presented results from real data on a number of
traits in dairy cattle and reported approximated rMG which
were in close agreement with (or even greater than) values
predicted with the equation of Daetwyler et al. (2010).
Adding UIMF to an index that already contained gMARB
LQFUHDVHGWKHJHQHWLFJDLQSHUJHQHUDWLRQEXWWKHEHQH¿WRI
including UIMF became only marginal when gMARB was
predicted with a high accuracy. For an accuracy of gMARB
of 0.8, which is expected to be obtained with a calibration set
of 2,524 animals (Table 4), the additional gain of including
UIMF was only 1.6% when the economic weight on MARB
was doubled, and 2% when it was tripled. Crews et al.
(2004) compared a model with only phenotypic measures
of carcass traits with a model including both carcass and
live animal ultrasound measurements. They reported a
correlation of 0.95 between sire EBV for MARB estimated
from the 2 models and suggested that genetic evaluations
for carcass traits should be based on both carcass phenotype
and live ultrasound data. This recommendation was also
made by MacNeil and Northcutt (2008). Our results are in
agreement with these 2 previous studies, as a slight increase
in RIHDQGǻG was observed when both UIMF and gMARB
were included in the index. Nevertheless, as pointed out
by Kahi and Hirooka (2005) a decision on which traits to
include in the index should take into account whether the
additional gain in RIHDQGǻG more than compensate the
costs of adopting ultrasound technology.
Greatest genetic gain per generation was obtained
with scenario 6, where gW200, gW400, gMUSC and
gMARB were used in the selection index and the
DFFXUDFLHVRIWKHVH*(%9ZHUH•)RUWKHWUDLWZLWK
the lowest heritability (W200) such an accuracy is
expected to be achieved with a calibration set of around
2,557 genotyped animals (Table 4).
Response to Selection in Single Traits
The response to selection per generation (S) in
MARB for the 6 scenarios and 3 different relative
economic weights on MARB are depicted in Figure 2.
As expected from the genetic correlations between the
traits (Table 2), when only the BW and muscling traits
are included in the selection index (either as phenotypes
or as GEBV; i.e., scenario 1 or 5) the response to
selection in MARB was negative for all levels of rMG
and all relative economic weights on MARB. For
selection based on GEBV, the more accurate they were,
the greater was the loss in genetic merit for MARB. As
suggested by the study of Sapp et al. (2002), selection
decisions using UIMF as one of the selection criteria
FDQ VLJQL¿FDQWO\ LPSURYH 0$5% ,QFUHDVHG JHQHWLF
gain for carcass traits by selecting on ultrasound
scanning of corresponding traits in live animals were
also reported by Kahi and Hirooka (2005). Moving from
scenario 1 to scenario 2 (i.e., adding UIMF to the index)
substantially improved S for MARB, which switched
from negative to positive when the economic weight
on MARB was doubled or tripled (Figure 2). This
improvement was responsible for the superior ΔG from
scenario 2 compared with scenario 1 (Figure 1), because
adding UIMF caused a decrease in S for the other 3
traits in the breeding goal (results not shown). For all
scenarios that contained gMARB in the selection index
(i.e., scenarios 3, 4, and 6), the response to selection in
MARB was positive for all relative economic weights
on MARB, when rMG was greater than 0.7. For an rMG
of 0.8, the differences between having just gMARB or
gMARB plus UIMF in the index (i.e., scenario 3 vs. 4)
were 4.8% when the economic weight on MARB was
doubled and 3% when it was tripled. For the other three
traits in the breeding goal, S was always positive across
all scenarios and relative economic weights on MARB,
with the exception of W400, for which S was negative
for scenarios 3 and 4 when rMG •
Table 4. Number of animals in the calibration set (NP)
needed to achieve a given level of accuracy of GEBV
(rMG) for BW at 200 d (W200) and 400 d (W400),
muscling score (MUSC) and marbling score (MARB)
Trait
rMG
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
W200
27
111
263
507
887
1,497
2,557
4,732
11,347
W400
20
81
191
369
645
1,089
1,860
3,441
8,252
MUSC
10
42
99
190
333
561
959
1,774
4,255
MARB
14
59
140
270
473
798
1,364
2,524
6,052
3424
Pimentel and König
König et al. (2009) investigated the change of
selection response per generation in a production trait
with moderate heritability and a functional trait with low
heritability for different values of rMG. In their study,
both traits (i.e., production and functionality) were
included in the breeding goal, and information sources in
the index were GEBV for both traits. For an rMG = 0.70
in both traits, the overall annual genetic gain was mostly
due to response to selection in production, whereas for
extremely high accuracies of 0.90 or 0.99, response to
selection in both traits was almost identical.
Results from a Calibration Set of 2,500 Animals
As mentioned earlier, the greatest overall genetic
gain per generation occurred when the traits in the
selection index were gW200, gW400, gMUSC, and
gMARB and rMG was around 0.7. In the following we
depict a situation in which a calibration set of 2,500
genotyped animals is formed. For a set of this size, the
expected accuracies of gW200, gW400, gMUSC, and
gMARB are 0.70, 0.75, 0.85, and 0.80, respectively.
Selection index calculations were performed and results
in terms of RIH, ΔG, and S for each trait according to
each scenario and relative economic weight on MARB
are presented in Table 5. The greatest RIH was obtained
with scenario 6 and equal economic weights and the
greatest ΔG was obtained with scenario 6 and a tripled
weight on MARB. The latter also yielded the greatest S
on MARB, while keeping S positive for the other three
traits, providing a more harmonic trend in response to
selection in all four traits in the breeding goal.
The considered number of 2,500 genotyped animals
with phenotypes is below the number of genotyped bulls
Table 5. Accuracy of the index ( RIH), overall genetic gain
per generation (ΔG) and response to selection for each trait
(S), according to different scenarios and relative economic
weights on marbling (w1 = same, w2 = doubled, w3 =
tripled), with a calibration set of 2,500 animals
Figure 2. Response to selection in MARB for different scenarios with
respect to the traits included in the index: BW at 200 d (W200) and 400 d
(W400), muscling score (MUSC), marbling score (MARB) and intramuscular
fat content measured by ultrasound (UIMF). A ‘g’ before the abbreviation
indicates a genomic breeding value for the given trait.
Scenario
w1
1
2
3
4
5
6
w2
1
2
3
4
5
6
w3
1
2
3
4
5
6
S for each trait
W400
MUSC
RIH
ΔG
W200
MARB
0.60
0.61
0.63
0.63
0.77
0.81
1.34
1.37
1.41
1.42
1.72
1.82
6.17
6.00
5.63
5.60
9.05
8.60
8.65
8.16
6.90
6.85
14.34
12.63
0.72
0.70
0.66
0.66
0.66
0.61
-0.10
-0.06
0.03
0.04
-0.13
0.00
0.45
0.52
0.65
0.66
0.58
0.78
1.17
1.34
1.67
1.70
1.48
1.99
6.11
5.20
3.71
3.66
9.03
6.77
7.49
5.71
2.17
2.10
13.59
8.14
0.73
0.62
0.46
0.45
0.66
0.46
-0.08
0.03
0.23
0.24
-0.12
0.17
0.32
0.46
0.70
0.71
0.40
0.77
1.03
1.48
2.21
2.26
1.26
2.44
5.81
3.90
2.02
1.99
8.81
4.66
5.69
2.83
-1.12
-1.14
12.26
3.88
0.72
0.48
0.27
0.26
0.65
0.30
-0.06
0.11
0.33
0.34
-0.10
0.28
Genomic selection for improving meat quality
with highly accurate conventional EBV currently being
used for deriving SNP effects in dairy cattle GS programs.
Examples of such programs include the calibration sets
of 5,025 genotyped Holstein bulls formed in Germany
(Liu et al., 2010) and the 15,966 bulls put together in
the EuroGenomics project (Lund et al., 2010). The latter
strategy (i.e., mixing sires from different countries in a
calibration set), implies a harmonization of traits and
EBV across countries as successfully implemented for
international genetic evaluations for production traits,
conformation, fertility, milkability, and longevity in
dairy cattle. A further strategy could be to build up a
calibration set for new meat quality traits (e.g., MARB)
based on phenotypes collected on a number of steers,
and using correlated response to selection from bulls
with GEBV for correlated traits (e.g., BW gain or
feed intake) available in large scale. This strategy was
evaluated by Calus et al. (2011) for GS of new health
traits in dairy cattle. For example, correlated response
to selection on GEBV for somatic cell score can be
used for increasing genetic gain for the new health trait
‘mastitis’. However, success of this attempt strongly
depends on the correlation between the trait of interest
and the available indicator traits.
For some beef breeds in Germany, herd structures
favor the implementation of a systematic collection
of phenotypes for carcass traits to be used in GS.
One such example is the Angus breed, with herds of
considerable large size located in East Germany. Those
herds could be used as a general base for accurate meat
quality phenotypes to be used in a calibration set for
GS within the Angus breed. Extension to other beef
breeds may also be possible. For some QTL or genes
associated with beef meat quality, the same favorable
alleles have been reported across breeds. A list of
studies reporting frequencies of favorable alleles of SNP
related to meat quality in different breeds was reported
by Van Eenennaam et al. (2007). Following results
from de Roos et al. (2008), the persistence of linkage
disequilibrium phase with panels of ~300k markers (i.e.,
an intermarker space of ~10 kb) should be enough to
obtain consistent marker effects across breeds such as
Angus, Jersey, or Holstein-Friesian. Rolf et al. (2010)
used a set of markers with mean spacing of 4.8 kb
harboring the μ-calpain gene and found that for 1 SNP
the same allele increased tenderness in Angus, Charolais,
Hereford, Limousin, and Simmental. They further
argue that with the currently available marker panels
of ~800k SNP (e.g., Illumina or Affymetrix), the intermarker distance is ~3.8kb, which should be enough to
implement across-breed GS in beef. Costs of genotyping
with high density panels of SNP keep decreasing over
time, which continues to make the genomic selection
approach more affordable. Recent advances in methods
3425
for imputing genotypes (e.g., VanRaden et al., 2011) will
further decrease costs as a proportion of the animals can
be genotyped for a lower density panel.
Implications
In the case where an established breeding program
is already routinely collecting ultrasound measures of
intramuscular fat, keeping this trait in the selection index
and continuing to measure it whilst incorporating a GEBV
for marbling into the index may be still advantageous.
This would be especially worthwhile for markets in
which MARB is of outstanding economic importance, as
it does for example for the Japanese beef industry (Kahi
and Hirooka, 2005). In the case of a breeding program
that does not collect ultrasound information, if a decision
has to be made whether to start measuring it or not, than
it may be better to invest in setting up a calibration set for
estimating GEBV for marbling directly.
Genomic selection might also help solve other
problems often observed in beef cattle genetic evaluations.
Routine conventional genetic evaluation may be biased
due to potential mistakes in pedigrees and poor genetic
connectedness across herds. These problems are not
relevant in genomic evaluations using information from
high density SNP arrays. Instead of the usual additive
genetic relationships built up from pedigree, genomic
relationships based on SNP data are used. Shifting from
probabilistic to realized relationships also allows better
estimation of mendelian sampling effects. Additionally,
successful implementation of GS would benefit natural
service sires. These bulls would be fully competitive
with sires used for artificial insemination.
Mainly due to the wish of consumers, German beef
breeding organizations are encouraged to include meat
quality in the overall breeding goal. From the current
perspective, 2 possibilities exist: incorporating ultrasound
technology to measure meat quality traits in live animals
(e.g., intramuscular fat content on selection candidates in
the field), or forming a calibration set of genotyped steers
with meat quality information measured in the carcass
for implementing GS. The presented results indicate
that forming a calibration set of ~500 genotyped animals
for estimating genomic breeding values for marbling
could suffice to obtain a larger response to selection
than collecting phenotypic ultrasound measures of
intramuscular fat content. If the size of the calibration set
is ~2,500, adding the ultrasound trait to an index already
containing the GEBV would hardly bring a benefit, unless
the relative economic weight for marbling is many-fold
larger than for the other traits.
3426
Pimentel and König
LITERATURE CITED
Archer, J. A., S. A. Barwick, and H. U. Graser. 2004. Economic evaluation
of beef cattle breeding schemes incorporating performance testing
of young bulls for feed intake. Aust. J. Exp. Agric. 44:393–404.
Brito, F. V., J. Braccini Neto, M. Sargolzaei, J. A. Cobuci, and F. S.
Schenkel. 2011. Accuracy of genomic selection in simulated
populations mimicking the extent of linkage disequilibrium in beef
cattle. BMC Genetics 12:80.
Buch, L. H., M. Kargo, P. Berg, J. Lassen, and A. C. Sørensen. 2012. The
value of cows in reference populations for genomic selection of new
functional traits. Animal 6:880–886.
Buchanan, D. S., M. K. Nielsen, R. M. Koch, and L. V. Cundiff. 1982.
Selection for growth and muscling score in beef cattle. I. Selection
applied. J. Anim. Sci. 55:516–525.
Calus, M. P. L., Y. de Haas, M. Pszczola, and R. F. Veerkamp. 2011.
Predicted response of genomic selection for new traits using
combined cow and bull reference populations. Interbull Meeting,
Stavanger, Norway.
Crews, D. H., Jr., and R. A. Kemp. 2002. Genetic evaluation of carcass
yield using ultrasound measures on young replacement beef cattle. J.
Anim. Sci. 80:1809–1818.
Crews, D. H., Jr., E. J. Pollak, and R. L. Quaas. 2004. Evaluation of
Simmental carcass EPD estimated using live and carcass data. J.
Anim. Sci. 82:661–667.
Daetwyler, H. D., R. Pong-Wong, B. Villanueva, and J. A. Wooliams.
2010. The impact of genetic architecture on genome-wide
evaluation methods. Genetics 185:1021–1031.
Dekkers, J. C. 2007. Prediction of response to marker-assisted and
genomic selection using selection index theory. J. Anim. Breed.
Genet. 124:331–341.
de Roos, A. P. W., B. J. Hayes, R. J. Spelman, and M. E. Goddard. 2008.
Linkage disequilibrium and persistence of phase in HolsteinFriesian, Jersey and Angus cattle. Genetics 179:1503–1512.
Goddard, M. E. 2009. Genomic selection: Prediction of accuracy and
maximisation of long term response. Genetica 136:245–252.
Gregory, K. E., L. V. Cundiff, and R. M. Koch. 1995. Genetic and
phenotypic (co)variances for growth and carcass traits of
purebred and composite populations of beef cattle. J. Anim. Sci.
73:1920–1926.
Haberland, A. M., U. U. von Borstel, H. Simianer, and S. König. 2012.
Integration of genomic information into Sport Horse Breeding
Programs for optimization of accuracy of selection. Animal 6:1369–
1376.
Hazel, L. N. 1943. The genetic basis for constructing selection indexes.
Genetics 28:476–490.
Kahi, A. K., and H. Hirooka. 2005. Genetic and economic evaluation
of Japanese Black (Wagyu) cattle breeding schemes. J. Anim. Sci.
83:2021–2032.
König, S., H. Simianer, and A. Willam. 2009. Economic evaluation of
genomic breeding programs. J. Dairy Sci. 92:382–391.
König, S., and H. H. Swalve. 2009. Application of selection index
calculations to determine selection strategies in genomic breeding
programs. J. Dairy Sci. 92:5292–5303.
Koots, K. R., J. P. Gibson, C. Smith, and J. W. Wilton. 1994a. Analyses of
published genetic parameter estimates for beef production traits. 1.
Heritability. Anim. Breed. Abstr. 62:309–338.
Koots, K. R., J. P. Gibson, and J. W. Wilton. 1994b. Analyses of published
genetic parameter estimates for beef production traits. 2. Phenotypic
and genetic correlations. Anim. Breed. Abstr. 62:825–853.
Lande, R., and R. Thompson. 1990. Efficiency of marker-assisted
selection in the improvement of quantitative traits. Genetics
124:743–756.
Leaflet, A. S., M. Izquierdo, D. E. Wilson, G. P. Rouse, and V. Amin. 1996.
Relationship Between Carcass End Points and USDA Marbling
Quality Grades: A Progress Report. ISU Beef Research Report, ASL
R1336. Iowa State University, Ames, IA.
Liu, Z., F. Seefried, F. Reinhardt, G. Thaller, and R. Reents. 2010. Dairy
cattle genetic evaluation enhanced with genomic information. 9th
World Congr. Genet. Appl. Livest. Prod., Leipzig, Germany.
Lund, M. S., A. P. W. de Roos, A. G. de Vries, T. Druet, V. Ducrocq, S. Fritz,
F. Guillaume, B. Guldbrandtsen, Z. Liu, R. Reents, C. Schrooten,
F. Seefried, and G. Su. 2010. Improving genomic prediction by
EuroGenomics collaboration. 9th World Congr. Genet. Appl. Livest.
Prod., Leipzig, Germany.
MacNeil, M. D., J. D. Nkrumah, B. W. Woodward, and S. L. Northcutt.
2010. Genetic evaluation of Angus cattle for carcass marbling using
ultrasound and genomic indicators. J. Anim. Sci. 88:517–522.
MacNeil, M. D., and S. L. Northcutt. 2008. National cattle evaluation
system for combined analysis of carcass characteristics and
indicator traits recorded by using ultrasound in Angus cattle. J.
Anim. Sci. 86:2518–2524.
Ríos-Utrera, A., and L. D. Van Vleck. 2004. Heritability estimates for
carcass traits of cattle: A review. Genet. Mol. Res. 3:380–394.
Rolf, M. M., S. D. McKay, M. C. McClure, J. E. Decker, T. M. Taxis, R.
H. Chapple, D. A. Vasco, S. J. Gregg, J. W. Kim, R. D. Schnabel,
and J. F. Taylor. 2010. How the next generation of genetic
technologies will impact beef cattle selection. 42nd BIF Annual
Meeting, Columbia, MO.
Ruten, W., C.-S. Schäfer, and E. Tholen. 1997. Zuchtwertschätzung für
Fleischrinder - Rasse Limousin. Accessed Nov. 28, 2011. http://www.
vit.de/fileadmin/user_upload/vit-fuers-rind/zuchtwertschaetzung/
publikationen/zws-limousin-in-deutschland.pdf.
Sapp, R. L., J. K. Bertrand, T. D. Pringle, and D. E. Wilson. 2002. Effects
of selection for ultrasound intramuscular fat percentage in Angus
bulls on carcass traits of progeny. J. Anim. Sci. 80:2017–2022.
Schaeffer, L. R. 2006. Strategy for applying genome-wide selection in
dairy cattle. J. Anim. Breed. Genet. 123:218–223.
Van Eenennaam, A. L., J. Li, R. M. Thallman, R. L. Quaas, M. E. Dikeman,
C. A. Gill, D. E. Franke, and M. G. Thomas. 2007. Validation of
commercial DNA tests for quantitative beef quality traits. J. Anim.
Sci. 85:891–900.
VanRaden, P. M., J. R. O’Connell, G. R. Wiggans, and K. A. Weigel. 2011.
Genomic evaluations with many more genotypes. Genet. Sel. Evol.
43:10.
Veerkamp, R. F., D. P. Berry, E. Wall, Y. de Haas, S. McParland, M.
Coffey, and M. P. L. Calus. 2011. Use of phenotypes from research
herds to develop genomic selection for scarcely recorded traits like
feed efficiency. Interbull Meeting, Stavanger, Norway.