See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/263004318

Genetic evaluation of inbred plants based on

BLUP of breeding value and general combining
ability
ARTICLE in CROP AND PASTURE SCIENCE JULY 2011
Impact Factor: 1.48 DOI: 10.1071/CP11016

CITATIONS

READS

72

5 AUTHORS, INCLUDING:
Jos Marcelo Soriano Viana

Vincius Ribeiro Faria

Universidade Federal de Viosa (UFV)

Universidade Federal de Viosa (UFV)

82 PUBLICATIONS 279 CITATIONS

13 PUBLICATIONS 55 CITATIONS

SEE PROFILE

SEE PROFILE

Fabyano Fonseca Silva

Universidade Federal de Viosa (UFV)
226 PUBLICATIONS 1,122 CITATIONS
SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

Available from: Jos Marcelo Soriano Viana

Retrieved on: 28 January 2016

CSIRO PUBLISHING

www.publish.csiro.au/journals/cp

Crop & Pasture Science, 2011, 62, 515522

Genetic evaluation of inbred plants based on BLUP

of breeding value and general combining ability
Jos Marcelo Soriano Viana A,D, Ramon Vincius de Almeida A, Vincius Ribeiro Faria A,
Marcos Deon Vilela de Resende B, and Fabyano Fonseca e Silva C
A

Universidade Federal de Vic osa, Departamento de Biologia Geral, 36570-000, Vic osa, MG, Brazil.
Embrapa Florestas; Universidade Federal de Vic osa, Departamento de Engenharia Florestal, 36570-000,
Vic osa, MG, Brazil.
C
Universidade Federal de Vic osa, Departamento de Estatstica, 36570-000, Vic osa, MG, Brazil.
D
Corresponding author. Email: jmsviana@ufv.br
B

Abstract. The testcross method is considered efcient for identifying inbred families with superior general combining
ability. The objective of the present study was to assess the relative importance of the performance per se and in crossing in the
selection of inbred progenies using bi-trait best linear unbiased prediction. We analysed data for expansion volume (EV) and
grain yield from three tests of popcorn (Zea mays L. ssp. everta) S3 families and seven testcross trials, using the ASRemL
software. Four selection strategies were assessed based on: breeding value (strategy 1), general combining ability effect
(GCA) (strategy 2), additive value and GCA from strategies 1 and 2 (strategy 3), and breeding value and GCA predicted by
bi-trait analyses considering EV and yield of the families and testcrosses as different traits (strategy 4). The bi-trait analyses of
the same characteristic assessed in S3 families and topcrosses were generally more accurate and had greater heritabilites. The
greatest predicted gains in EV were obtained using strategy 4, which was inferior to the other strategies for the yield predicted
gains. Strategies 1 and 2 differed most for the families selected. Selection based on GCA maximised heterosis. All of the
strategies resulted in comparable realised gains, especially the strategies 3 and 4 based on breeding value and GCA. Selection
on S3 based on the additive value and GCA (strategies 3 and 4) resulted in inbred lines superior in number of favourable genes
and in general combining ability.
Additional keywords: inbred family, testcross, bi-trait analysis.

Introduction
The testcross method aims to assess the potential of inbred
families to produce superior hybrids from crosses with at least
one tester. It was suggested by Davis (1927) and became common
in maize hybrid breeding programs. In the 1980s, 78% of maize
breeders assessed general combining ability (GCA) by testcross,
using one of the three rst selng generations (Bauman 1981).
Many theoretical and applied results with maize showed that
testcross might be used efciently in inbred family selection.
Baktash et al. (1981) obtained positive signicant correlations
between the performances of maize testcrosses and singlecross hybrids for grain yield and other traits. In the studies of
Bernardo (1991), Lile and Hallauer (1994), and Bordes et al.
(2007) the genotypic correlation between the performance of the
testcrosses in early and late generations and that of their direct
homozygous descendants ranged from 0.71 for S1 lines, to 0.99
for S6 lines.
The choice of the tester is an important challenge to hybrid
crop breeders (Hallauer and Miranda Filho 1988). Hull (1945)
suggested the use of an inbred line with predominantly recessive
genes or a population with low favourable gene frequencies.
This kind of tester maximises the genotypic variance among
testcross families (Wright 1980; Cowen 1987). Comparisons

CSIRO 2011

among testers with narrow and broad genetic base

demonstrated discrimination relative to combining ability and
heterosis (Elias et al. 2000). For Hallauer and Lopez-Perez
(1979), the advantage of using a tester with a broad genetic
base is that it reduces the tester
environment and inbred
line
tester interactions. On the other hand, the testers with a
narrow genetic base present fewer gamete sampling problems.
Since Bernardo (1994) demonstrated the effectiveness of
the BLUP (best linear unbiased prediction) methodology in
the prediction of non-assessed maize single-cross hybrids, the
method has become, as had already occurred in animal and
forestry breeding, of regular use in annual crop breeding,
aiming to predict additive genetic or genotypic values, mainly
in the genetic evaluation of pure/inbred lines (Piepho et al.
2008). Considering that multi-trait analysis is theoretically
more advantageous than single-trait analysis when the absolute
value of the difference between the genetic and error correlations
is greater (Schaeffer 1984), if the genetic correlation between the
same characteristic assessed in the S3 families and the testcross
hybrids is high, bi-trait analysis can be an excellent alternative
for the prediction of additive values and the effects of general
combining ability because the residual correlation between
the same trait assessed in the inbred families and testcrosses is
10.1071/CP11016

1836-0947/11/060515

516

Crop & Pasture Science

nil. The objective of the present study was to assess the relative
importance of the performance per se and in crossing in the
selection of inbred progenies, using the BLUP methodology.
Materials and methods
Testcross and testers
The genotypic means of the testcrosses are:
GOA1 A1 xt MT qat MT GCAA1 A1
MT at =2aAA1 A1

q  p 

at MT GCAA1 A2
2
MT at =2aAA1 A2

GOA1 A2 xt MT

GOA2 A2 xt MT  pat MT GCAA2 A2

MT at =2aAA2 A2
where MT = m + (p + r 1)a + (p + r 2pr)d is the mean of the
testcrosses, at = [a + (s r)d] is the effect of a gene substitution in
the tester, a = [a + (q p)d] is the effect of a gene substitution in the
base population, m is the mean of the homozygotes, a is the
difference between the genotypic value of the homozygote of
greater expression and m, d is the deviation due to dominance, p
and q are the allelic frequencies in the base population, r and s are
the allelic frequencies in the tester (Hallauer and Miranda Filho
1988), and GCA and A stand for general combining ability effect
and additive genetic value, respectively (E(GCA) = E(A) = 0).
Thus, the variance of the genotypic means of the testcrosses is
the variance of the GCA effects of the Sn1 plants, given by:
X
s2GT s2GCA F
pqa2t
where F is the inbreeding coefcient of the inbred progenies.
If the tester is the base population, s2GT F2 s2A , where sA2 is
the additive variance in the base population.
The GCA effect, as the additive value, is an almost perfect
indicator of the superiority of the Sn1 plant for the number of
favourable genes, regardless of whether they increase or decrease
trait expression. The correlation between the additive value and
the GCA effect of an Sn1 plant is:
P
pqaat
rAS ; GCASn1 q
P
 P

n1
pqa2
pqa2t
For a gene, r = 1 with complete, partial, or no dominance.
In the case of overdominance, r = 1 or 1, depending on the
frequency of the dominant allele in the base population and in the
tester.
Experimental data
The analyses considered the data of expansion volume (EV) and
grain yield of three S3 family tests, derived from the Beija-Flor
and Vic osa popcorn (Zea mays L. ssp. everta) populations, and
seven testcross trials. Two sets of inbred families from the BeijaFlor population and one from the Vic osa population were
assessed in an experimental eld at the Federal University of
Vic osa (UFV), in Vic osa, MG, Brazil, in an incomplete block
design, with replication only of the common checks. In the trial

J. M. S. Viana et al.

conducted in the 19992000 growing season, 176 Beija-Flor

progenies (Program 1) and the ISLA 208 control, an open
pollination variety, were assessed. In the experiment
conducted in the 200001 growing season, 169 Beija-Flor
progenies (Program 2) and the IAC 112 control, a single
hybrid from the Agronomic Institute, Campinas, SP, were
assessed. In the 200103 growing season, 229 families from
the Vic osa population were assessed, with IAC 112 and the triple
hybrid Zlia, from Pioneer, as controls.
The testcrosses of the Beija-Flor families, obtained by
crossing with the Vic osa population, were assessed in four
experiments carried out in the 200001 growing season
(Program 1) and in two experiments carried out in the 2001
02 growing season (Program 2). Regarding the testcrosses of
Program 1, an experiment was carried out at the Agricultural
Research Company of the State of Rio de Janeiro, in Campos
dos Goytacazes, RJ, in a 10
10 lattice design with four
replications. A second test was carried out in the UFV
experimental station in Capinpolis, MG, in a 9
9 lattice
design with four replications. Two trials were conducted at
the experimental station of UFV in Coimbra, MG, in an 8
8
lattice design with three and two replications, respectively. The
testcross trials of Program 2 were carried out at the Maring
State University, PR, and in Capinpolis, in a 12
12 lattice
design with three replications. The controls in these experiments
were Beija-Flor, Vic osa, IAC 112, and Zlia. The testcrosses of
the Vic osa population, obtained by crossing with Beija-Flor,
were assessed in Capinpolis in the 200304 growing season,
in a 12
12 lattice design with three replications. In addition to
the mentioned controls, the ngela population, from Embrapa
Corn and Sorghum, and Beija-Flor Cycle 1, Beija-Flor Cycle 2,
Vic osa Cycle 1, and Vic osa Cycle 2, obtained by half-sib
selection, were included.
The realised gains were calculated using the data of the
corresponding tests of the S4 families, conducted at UFV in
the 200102 (Beija-Flor) and 200304 (Vic osa) growing
seasons, and also in an incomplete block design with
replication only of the common checks (IAC 112 and Zlia or
a inbred line).
In the family and testcrosses trials, each plot corresponded to
a 5-m row, with an ideal stand of 30 plants and 0.9-m spacing. The
following traits were assessed in each plot: nal stand (covariate),
grain moisture (covariate), grain yield, and EV. In the family tests,
EV was determined using a 1250-W hot-air popcorn machine and
30-g samples. The EV of the testcrosses was assessed in a Cretors
Metric Weight Volume Tester (C. Cretors & Co., Chicago, IL),
using 250-g samples.
Selection strategies and genetic gains
The following four selection strategies were considered, based on
the index proposed by Mulamba and Mock (1978) with weights of
3 for EV and 1 for yield: strategy 1, selection based on the additive
genetic values predicted by bi-trait analysis of S3 families;
strategy 2, selection based on the GCA effects predicted by bitrait analysis of the testcross experiments; strategy 3, selection
based on the additive values predicted by bi-trait analysis of the
family test and the GCA effects predicted by bi-trait analysis of
the testcross experiments; and strategy 4, selection based on the
additive values and GCA effects predicted by bi-trait analyses

Prediction of additive value and general combining ability

considering EV and yield of the families and testcrosses as

different traits. Regarding strategies 3 and 4, selection based
on the Mulamba and Mock index considered one index for EV and
one index for yield, both generically dened by the following
linear combination of additive value and GCA effect:
I b1 ASn1 b2 GCASn1
where b1 is the weight of the additive value and b2 is the weight
of the GCA effect.
Curiously, because of the proportionality between the additive
value and the GCA effect, the values of these weights that
minimise the variance of the difference between the index
and the mean of the additive value and the GCA effect
(Var(2I ASn1GCASn1)) are 1/2 and 1/2. As the predictions
of additive value and GCA effect do not have the same accuracy,
the index used was:
!
!
1  rGCA;GCA
1  rA;A~
~
~ Sn1
~ Sn1
A
GCA
I
rGCA;GCA
rA;A~
~
where rGCA;GCA
~ and r A;A
~ are the accuracies of the predictions of
breeding value and GCA effect.
Because the Mulamba and Mock index was used, the predicted
gains were computed based on the selection differential (SD)
and on the generalised measure of heritability (h2) proposed by
Cullis et al. (2006), by the function DM = SDh2 (parental control
equal to 1 was assumed). The proportion of selected families
was 30%.
The efciencies of the selection strategies, relative to the
identication of superior parents, were also assessed by means
of Pearsons correlation between breeding values and effects of
GCA predicted from distinct models, from the percentage of
common selected parents (coincidence) and from the predicted
and realised gains. The realised gains were calculated as the
difference between the average of the S4 families derived from
selected S3 progeny and the mean of the S4 generation, corrected
by the average of the common check (IAC 112).
BLUP analysis
Considering the assessment of Sn families and testcrosses, the
phenotypic values can be dened as:
PSn F MF ASn1 Do

fixed effectf

non-genetic random effectn error

PSn Fxt MT GCASn1

X
fixed effectf
f

non-genetic random effectn error

where MF is the mean of the Sn generation and Do is the average

dominance genetic value of the progeny (Viana et al. 2011).
The additive genetic values and the general combining
effects of the Sn1 plants, to identify the best Sn families, can
be predicted using the BLUP methodology, proposed by
Henderson (1974), with estimation of the variance components
by the residual/restricted maximum likelihood method (Patterson
and Thompson 1971).

Crop & Pasture Science

517

When assessing nf Sn families in an environment, in the

incomplete block design, and the testcrosses in more than
one environment, in the lattice design, in relation to v traits,
and admitting as random the effects of blocks, blocks/replication/
environment, and GCA
environment interaction, the models
for the multi-trait analyses can be represented in the following
matrix form:
y Xb Z1 u1 Z2 u2 Z3 u3 e
where y is the vector of phenotypic values; X is the incidence
matrix of the xed effects; b is the vector of the levels of each xed
factor (including the population means); Z1, Z2, and Z3 are the
incidence matrices of the random effects; and e is the residuals
vector. For inbred families, u1 is the vector of the additive genetic
values of the parents, u2 is the vector of the average dominance
genetic values of the progenies, and u3 is the vector of block
effects. For testcrosses, u1 is the vector of the GCA effects, u2 is
the vector of the GCA
environment interaction effects, and u3 is
the vector of block/replication/environment effects.
The criterion used to obtain the best linear unbiased predictor
of a random vector is the maximisation of the joint probability
density function of y and the random vector(s), or of the random
vector(s) and e, which obtains, under normality, the mixed model
equations (Henderson 1974).
For inbred families the variances of u1 and u2, and the
covariance between these random vectors, are (Viana et al. 2011):
G1 CovAv  Anf
where Cov(A)(v) is the additive genetic variance matrix, relative to
the v traits, A(nf) = {2rij} is the additive relationship matrix, and rij
is the coefcient of co-ancestry between the parents of the
progenies i and j;
G2 CovDo v  Dnf
where CovDo v is the dominance genetic variance matrix,
relative to the v traits, D(nf) = {uij} is the dominance
relationship matrix, and uij is the probability that the parents of
progenies i and j have genotypes identical by descent; and:
G4 CovASn1 ; Do v  Dnf
where CovASn1 ; Do v is the covariance matrix between the
additive value of the parent and the average dominance value of
the progeny, relative to the v traits.
For a trait, the parametric values of the variance of the average
dominance value of the inbred families and of the covariance
between ASn1 and Do are presented by Viana et al. (2011).
For testcrosses, G4 is a null matrix. Dening aT/2a = k, we
have:

X

X
kASn1i ;
kASn1j
CovGCASn1i ; GCASn1j Cov
 X

1
1
pqa2t 2rij s2GCA
2rij
2
2F
Then, as 2rii=1 + Fn1 = 2F, G1 = (1/2F)Cov(GCA)(v)  A(nf),
where Cov(GCA)(v) is the variance matrix of the general
combining ability effects, relative to the v traits.

518

Crop & Pasture Science

J. M. S. Viana et al.

The accuracies of the predictions of the breeding values and

the GCA effects are (Mrode 2005):

rA;A~

s


PEVA~
1
~ 2A
2Fs

and rGCA;GCA
~

s


PEVGCA
~
1
~ 2GCA
s

where PEV is the prediction error variance.

Statistical analyses
ASRemL v3.00 software (Gilmour et al. 2009) was used for the
analyses (see code in Appendix 1). The additive and dominance
relationship matrices les were generated by a program
developed in REALbasic 5.5. To read the pair-wise prediction
error variance matrix in the le .pvs, another REALbasic 5.5
program was developed. In the analyses of inbred families, G4 is
assumed to be a null matrix. In the analyses, each common check
was inserted in the pedigree as an individual with unknown
parents. Since the checks were not S3 families or testcrosses,
we inserted a xed effect of population. To test the nullity of the
variance components, we used the likelihood ratio test (Gilmour
et al. 2009). The condence intervals for the variance components

were obtained from the standardised normal distribution (Littell

et al. 2006).
Results
Independent of the population and the BLUP analysis, the
likelihood ratio tests showed genetic variability for EV and
grain yield (Tables 1 and 2). The analysis of the Beija-Flor
population testcrosses, assessed in two and four environments,
showed interaction between the GCA effect of the S2 plant and the
environment for the two traits, with two exceptions. Regarding
the analyses of the S3 families, it is emphasised that, in general,
only the additive model could be tted. The additivedominant
model could only be tted for EV assessed in the S3 families of
Program 2 of the Beija-Flor population. This does not mean that in
the other cases there was no dominance. Analysis of the mean
heterosis (Table 3) for yield showed, as expected, gene action due
to dominance. In relation to EV, considering that the average
heterosis values were systematically close to zero and that
dominance was proved in one case, by the two BLUP
analyses, it can be considered that there was evidence of bidirectional dominance.

Table 1. Parameters estimated from the bi-trait BLUP/REML analyses of expansion volume (EV, mL/g) and grain yield (kg/ha) of S3 families and
testcross hybrids from the popcorn populations Vic osa and Beija-Flor
Parameter

Vic osa
EV

Beija-Flor
Yield

Program 1
EV

Additive variance
Cond. interval 95%
Dominance variance
Conf. interval 95%
Block
Error
Accuracy (A)
h2
Genetic correlation
Gain/yearA
Gain/yearB
Gain/yearC
S3 mean
GCA
Cond. interval 95%
GCA
Env.
Block/Rep./Env.
Error
Accuracy (GCA)
h2
Genetic correlation
Gain/yearB
Gain/yearA
Gain/yearC
Testcross mean
Hybrid mean
A

0.0166

2.5102
0.1359 4.8845

0.0000
14.4213

1E17

97 675
47 315 148 036

0
391 817

0.6181
0.6744
0.4374
0.4979
0.1224 0.3368
1.46
39.75
0.02
9.15
0.81
13.47
27.9
0.52940.0265
01.0800

0.0000
4.8461

1165.0
35 9960.0151
072 355

14 228
283 106

0.5739
0.5896
0.3656
0.3780
0.3811 0.3647
0.22
7.86
0.01
2.97
0.01
4.76
27.0
30.9

3243.2
3218.5

Selection based on inbred family (strategy 1).

Selection based on testcross (strategy 2).
C
Selection based on breeding value and GCA effect (strategy 3).
B

Program 2
Yield

0.0506

4.2106
09.1028

0.8651
28.3837

0.0773

23 876
055 195

6394
201 385

0.6115
0.5712
0.4314
0.3936
0.1882 0.4471
1.99
59.25
0.94
52.55
1.13
79.67
22.7
11.29451E18
6.449616.1394
1.99530.1310
0.2424
8.0713

1055.4
155 9151E18
84 386 227 444
15 9281E06
53 594
287 532

0.9228
0.9067
0.9703
0.8564
0.8452 0.0630
1.32
73.73
0.26
5.63
1.43
72.25
25.2
33.1

2545.5
2581.7

EV

Yield
0.0001

4.0082
014.4127
2.66100.8415
024.0649
0.0000
12.9116

21 8310.0003
636.329643 025

6059
129 363

0.6353
0.5839
0.4274
0.3806
0.2097 0.4541
1.69
53.81
0.43
5.30
0.77
18.31
23.8
1.94281E05
0.63983.2459
2.53591E11
0.2871
6.2324

783.2
13 2610.0110
026 903
30 0782E05
10 581
143 958

0.7579
0.6384
0.2317
0.2453
0.3194 0.2815
0.21
16.61
0.04
2.02
0.05
13.04
24.4
31.5

1769.9
1961.0

Prediction of additive value and general combining ability

Crop & Pasture Science

519

Table 2. Parameters estimated from the bi-trait BLUP/REML analyses of expansion volume (EV, mL/g) and grain yield (kg/ha), measured in S3
families and testcross hybrids from the popcorn populations Vic osa and Beija-Flor, considering the same trait measured in inbred families and
testcrosses as distinct variables
Parameter

Vic osa
EV

Beija-Flor
Yield

Program 1
EV

Additive variance
Cond. interval 95%
Dominance variance
Cond. interval 95%
Block
Error
Accuracy (A)
h2
Gain/yearA
GCA
Cond. interval 95%
GCA
Env.
Block/Rep./Env.
Error

2E06

7E23

5.5167
2.63858.3950

0.0000
10.0892

108370
56491 160 249

0
348 639.0

0.7925
0.7440
1.41

0.7377
0.5475
12.18

0.55250.0090
01.1339

0.0000
4.7322

6E21

1E17

0.9228
0.9041
2.44

44 0170.0062
316284 872

10 926
291 426

10.50187E21
6.008614.9950
1.97301E05
0.3084
6.3165

0.6289
0.3928
0.09

0.6884
0.4913
6.57

0.9018
0.9680
1.04

Correl. A, GCA

0.6160 0.2333

0.8069 0.2281

0.8182 0.0711

0.9651
0.7156

0.9022
0.8106

0.8826
0.9272

Correl. A, A
Correl. GCA, GCAB

EV

24.9701
315 975
15.025234.9150 169 016 462 934

0.0000
0
12.264
226 476

Accuracy (GCA)
h2
Gain/yearA

Program 2
Yield

4.6712
0.69588.6467
9.57237E06
3.837815.3068
0.0000
5.3639

0.8926
0.9542
79.06
106 9734E15
50 865163 081
10 3770.3537
58 138
369 680

Yield
0.0006

16 6180.0024
035 643

0
160 973

0.6977
0.5086
0.90
1.78608E08
0.55293.0190
2.52263E12
0.2923
6.9638

0.8235
0.7907
83.60

0.5067
0.1863
8.65
14 5090.0070
35528 664
27 3200.0002
9944
127 660

0.7517
0.1642
0.12

0.6419
0.2839
10.04

0.5136 0.1438

0.5500 0.2434

0.1967 0.4718

0.6090
0.9161

0.9648
0.9754

0.9649
0.9790

Selection based on breeding value and GCA effect (strategy 4).

Predictions from the two bi-trait analyses.

Table 3. Average heterosis, heterosis of selected S3 families, and realised gains for expansion volume (EV, mL/g) and grain yield (kg/ha), and
coincidence (%) of selected families, relative to four selection strategies
Selection strategies: 1, based on breeding value (A); 2, based on GCA effect; 3, based on A and GCA effect; 4, based on A and GCA effect from modied bi-trait
BLUP analysis
Pop.

Vic osa

B-Flor
Prog. 1

B-Flor
Prog. 2

Average heterosis
EV
Yield
0.4
(1.3%)

0.8
(3.2%)

0.1
(0.6%)

639.5
(24.6%)

299.7
(13.3%)

72.6
(4.3%)

Non-corrected for a common check.

Selection strategy:
1

EV

Yield

EV

Yield

EV

Yield

EV

Yield

Het. of sel. families

%
Realised gain
Coincidence
1
2
3

0.3
1.2
2.7

623.7
24.0
630.0

0.8
3.0
2.3

597.9
23.0
653.5
18.8

0.2
0.6
3.0

664.7
25.5
653.8
91.3
21.7

0.1
0.3
2.8

666.2
25.6
615.3
82.6
26.1
88.4

Het. of sel. families

%
Realised gainA
Coincidence
1
2
3

0.3
1.1
6.2

312.9
13.9
156.6

1.9
7.2
6.7

127.5
5.7
259.3
45.3

2.1
8.2
7.7

131.0
5.8
307.0
56.6
69.8

1.3
5.1
6.9

88.3
3.9
267.2
69.8
62.3
58.5

Het. of sel. families

%
Realised gain
Coincidence
1
2
3

0.2
0.8
1.5

55.7
3.3
114.7

1.6
6.7
2.9

208.0
12.3
158.8

1.4
5.6
2.5

133.0
7.8
163.9

1.3
5.4
2.4

143.3
8.4
163.1

41.2

64.7
72.5

66.7
74.5
96.1

520

Crop & Pasture Science

The bi-trait analyses of EV and yield (Table 1) should have

been advantageous relative to the single-trait analyses only for the
topcrosses because the absolute values of the difference between
the genetic and error correlations were 0.43, 0.97, and 0.15. In the
S3 family tests, the differences were 0.04, 0.09, and 0.03.
Compared with the bi-trait analyses of EV and yield, the bitrait analyses of the same characteristic assessed in the S3 families
and topcrosses generally resulted in increased prediction
accuracy of additive value and the GCA effect and increase in
heritability (Tables 1 and 2), especially in the analyses of the S3
family tests. The increases in accuracy of the additive values
ranged from 9.4 to 56.3%. The increases in accuracy of the GCA
effects ranged from 0.5 to 16.8%. The increase in heritability of
the S3 families and topcrosses ranged from 10 to 142.4% and from
7.4 to 30.0%, respectively. These results can be attributed to
the magnitude of the absolute value of the genetic correlation
between additive value and the GCA effect (because the
correlation of the error is zero), which ranged from 0.55 to
0.82 for EV and from 0.20 to 0.81 for yield.
The predictions of additive value and GCA effect by the two
BLUP analyses presented positive, high-magnitude correlations
for both EV and yield (Table 2). The correlation between the
additive values ranged from 0.88 to 0.96 for EV and from 0.61 to
0.96 for yield. The intervals for the predictions of the GCA effect
were 0.72 to 0.97 for EV and 0.81 to 0.98 for yield. The
correlations between breeding value and the GCA effect for
the same trait ranged from 0.55 to 0.82 for EV and from 0.20
to 0.81 for yield. Because the heritability values were generally
greater in the analysis of the S3 family tests compared with the
topcross experiments, the gains predicted per year were superior
for both EV and yield when calculated based on the phenotypic
values of S3 progenies (Tables 1 and 2).
Discussion
It is noteworthy that ignoring the covariance between additive
and average dominance values and tting the additive model
when there is dominance does not signicantly affect the
inferences. Based on simulation, de Boer and van Arendonk
(1992) demonstrated that ignoring both the covariance between
additive and dominance effects with inbreeding and the change
in dominance variance due to inbreeding did not signicantly
bias the prediction of additive and dominance effects in selected
or unselected populations with inbreeding. Even concluding
that the additivedominant model should be tted wherever
possible, Viana et al. (2010a, 2010b) reported a high positive
correlation between the breeding values predicted by the additive
and additivedominant models, for the EV and yield.
Conrming theoretical results of Schaeffer (1984), our study
also showed that the accuracy of the bi-trait analysis, relative to
the single-trait approach, is greater when there is large difference
between the genetic and error correlations. Viana et al. (2010b)
reported greater accuracy and efciency of bi-trait analysis in
selection among half-sib families, but equivalence for withinfamily selection, because the absolute values of the difference
between the genetic and error correlations were 0.38 and 0.09.
Based on the estimates of the correlation between the breeding
values and GCA effects, we can state that selection based on the
two predictions is a priori a most adequate strategy. In the studies

J. M. S. Viana et al.

of Bekavac et al. (2008) and Mihaljevic et al. (2005), for grain

yield the genetic correlations between line per se and testcross
performances ranged from 0.40 to 0.59. For other traits the values
varied from 0.52 to 0.94.
To assess the relative importance of the performances per se
and in crossing, we considered the predicted and realised gains,
heterosis, coincidence among the selected families, and the results
of the single-crosses tests. Considering only the S3 family tests,
the predicted gains per year were greater in EV and better in yield,
because of the negative predictions of lower magnitude, with
selection based on additive value (strategy 1), emphasising that
for the gain per cycle in EV, selection based on additive value and
GCA effect (strategy 3) resulted in equivalent predicted gain. S3
progeny selection based solely on the GCA effect (strategy 2) was
associated with lower predicted gains in EV and yield. Using only
the topcross experiments, the gains predicted per year in EV were
greater with selection based on the GCA effect. The greatest
predicted gains in EV were obtained using strategy 4, calculated
based on the phenotypic values of the S3 families. However, this
strategy was inferior to the other strategies for predicted gains in
yield, regardless of the phenotypic values used in the calculation
of the selection differential. Concerning the realised gains, all of
the strategies gave comparable estimates for EV and alteration in
the yield, especially strategies based on additive value and the
GCA effect (Table 3). Regarding the coincidence of selected
families, in the Vic osa population sharp differences were
observed in strategy 2 compared with the other strategies, with
coincidences ranging from 19 to 26%, and there were similarities
among strategies 1, 3, and 4, with coincidences ranging from 83
to 91% (Table 3). In relation to the Beija-Flor population, the
strategies with less coincidence in the selected families were
strategies 1 and 2. Heterosis analysis of the topcrosses of the
selected families showed that selection based on the GCA effect
maximised heterosis in EV and tended to maximise heterosis in
yield (Table 3). The index-based strategies tended to produce
similar results. Improvement in GCA and heterosis with selection
based on testcrosses has also been proved in previous studies
(Obaidi et al. 1998; Menkir and Kling 1999).
Some of the inbred lines derived from the S3 families
selected by at least one of the four strategies under analysis
were assessed by J. M. S. Viana, M. S. F. Valente,
C. A. Scapim, M. D. V. Resende, F. F. Silva (unpubl. data).
The diallels included 28 inbred lines of Vic osa and 18 of
Beija-Flor, all also assessed per se. Of the 28 Vic osa inbred
lines, 1016 were derived from selected S3 families, depending
on the strategy considered. Regarding the 18 Beija-Flor inbred
lines, three to 10 were derived from selected S3 progenies.
Regardless of the population, the analysis of the additive
values and the GCA effects of the parents of inbred lines,
predicted by BLUP single-trait analyses, did not show clear
superiority or inferiority of one selection strategy in S3. With
some exceptions, >65% of the inbred lines derived from selected
S3 presented a positive additive value and GCA effect for both the
EV and yield. Thus, selection in S3 based on additive value and/or
GCA effect resulted in inbred lines superior in the number of
favourable genes for quality and yield, with superiority also in
GCA. Also, with few exceptions, the inbred lines derived from
selected S3 families presented a mean additive value and mean
GCA effect greater than those for the inbred lines derived from

Prediction of additive value and general combining ability

non-selected families, or at least a positive value, although smaller

than the mean values of the non-selected progenies. Several
reported experimental results have also proved the efciency
of identication of superior inbred lines based on the evaluation of
testcrosses (Bohn et al. 2003; Gustafson et al. 2010).
Considering that, in general, heritabilities were greater for
inbred progenies and genetic correlations between line per se and
testcrosses were intermediate to high, it can be concluded that
selection based on breeding value and GCA effect tends to be
more efcient than selection based solely on the additive value or
GCA effect, as demonstrated by Gallais (1997).
Acknowledgments
We thank the Foundation for Research Support of Minas Gerais State
(Fapemig), the Brazilian Federal Agency for Support and Evaluation of
Graduate Education (Capes), and the National Council for Scientic and
Technological Development (CNPq) for nancial support.

References
Baktash FY, Younis MA, Al-Younis AH (1981) Correlation coefcients
between topcrosses and diallel crosses in corn (Zea mays L.).
Mesopotamia Journal of Agriculture 15, 5158.
Bauman LF (1981) Review of methods used by breeders to develop superior
corn inbreds. In Proceedings of the 36th Annual Corn and Sorghum
Industry Research Conference. pp. 199208. (American Seed Trade
Association: Washington, DC)
Bekavac G, Purar B, Jockovic D (2008) Relationships between line per se and
testcross performance for agronomic traits in two broad-based populations
of maize. Euphytica 162, 363369. doi:10.1007/s10681-007-9527-y
Bernardo R (1991) Correlation between testcross performance of lines at early
and late selng generations. Theoretical and Applied Genetics 82, 1721.
doi:10.1007/BF00231272
Bernardo R (1994) Prediction of maize single-cross performance using
RFLPs and information from related hybrids. Crop Science 34, 2025.
doi:10.2135/cropsci1994.0011183X003400010003x
Bohn M, Magg T, Klein D, Melchinger AE (2003) Breeding early maturing
European dent maize (Zea mays L.) for improved agronomic performance
and resistance against the European corn borer (Ostrinia nubilalis HB.).
Maydica 48, 239247.
Bordes J, Charmet G, de Vaulx RD, Lapierre A, Pollacsek M, Beckert M,
Gallais A (2007) Doubled-haploid versus single-seed descent and S-1family variation. Euphytica 154, 4151. doi:10.1007/s10681-006-9266-5
Cowen NM (1987) Testcross selection theory. Theoretical and Applied
Genetics 73, 371378. doi:10.1007/BF00262503
Cullis BR, Smith AB, Coombes NE (2006) On the design of early generation
variety trials with correlated data. Journal of Agricultural, Biological &
Environmental Statistics 11, 381393. doi:10.1198/108571106X154443
Davis RL (1927) Report of the plant breeder. In Annual reports. pp. 1415.
(Agricultural Experiment Station: Puerto Rico)
de Boer IJM, van Arendonk JAM (1992) Prediction of additive and dominance
effects in selected or unselected populations with inbreeding. Theoretical
and Applied Genetics 84, 451459.
Elias HT, Carvalho SP, Andr CGM (2000) Comparac o de testadores
na avaliac o de famlias S2 de milho. Pesquisa Agropecuaria
Brasileira 35, 11351142. doi:10.1590/S0100-204X2000000600009
Gallais A (1997) Combined testcross and S-1 selection for the improvement
of testcross and inbred performances. Crop Science 37, 11261133.
doi:10.2135/cropsci1997.0011183X003700040016x

Crop & Pasture Science

521

Gilmour AR, Gogel BJ, Cullis BR, Thompson R (2009) ASReml User Guide
Release 3.0. (VSN International Ltd: Hemel Hempstead, UK)
Gustafson TJ, Coors JC, de Leon N (2010) Selection for forage yield and
composition on the Wisconsin Quality Synthetic maize population. Crop
Science 50, 17951804. doi:10.2135/cropsci2009.12.0725
Hallauer AR, Lopez-Perez E (1979) Comparisons among testers for
evaluating lines of corn. In Proceedings of the 34th Annual Hybrid
Corn Industry Research Conference. pp. 5775. (Pioneer Hi-Bred:
Washington, DC)
Hallauer AR, Miranda Filho JB (1988) Quantitative genetics in maize
breeding. 2nd edn (Iowa State University Press: Ames, IA)
Henderson CR (1974) General exibility of linear model techniques for sire
evaluation. Journal of Dairy Science 57, 963972. doi:10.3168/jds.
S0022-0302(74)84993-3
Hull FH (1945) Recurrent selection for specic combining ability in corn.
Journal - American Society of Agronomy 37, 134145.
Lile SM, Hallauer AR (1994) Relation between S2 and later generation
testcrosses of two corn populations. The Journal of the Iowa Academy
of Science 101, 1923.
Littell RC, Milliken GA, Stroup WW, Wolnger RD, Schabenberger O (2006)
SAS for mixed models. 2nd edn (SAS Institute Inc.: Cary, NC)
Menkir A, Kling JG (1999) Effect of reciprocal recurrent selection on grain
yield and other traits in two early-maturing maize populations. Maydica
44, 159165.
Mihaljevic R, Schon CC, Utz HF, Melchinger AE (2005) Correlations and
QTL correspondence between line per se and testcross performance for
agronomic traits in four populations of European maize. Crop Science
45, 114122.
Mrode RA (2005) Linear models for the prediction of animal breeding
values. 2nd edn (CABI Publishing: Oxfordshire, UK)
Mulamba NN, Mock JJ (1978) Improvement of yield potential of Eto Blanco
maize (Zea mays L.) population by breeding for plant traits. Egyptian
Journal of Genetics and Cytology 1, 4057.
Obaidi M, Johnson BE, Van Vleck LD, Kachman SD, Smith OS (1998)
Family per se response to selling and selection in maize based on testcross
performance: a simulation study. Crop Science 38, 367371. doi:10.2135/
cropsci1998.0011183X003800020016x
Patterson HD, Thompson R (1971) Recovery of inter-block information when
block sizes are unequal. Biometrika 58, 545554. doi:10.1093/biomet/
58.3.545
Piepho HP, Mhring J, Melchinger AE, Bchse A (2008) BLUP for
phenotypic selection in plant breeding and variety testing. Euphytica
161, 209228. doi:10.1007/s10681-007-9449-8
Schaeffer LR (1984) Sire and cow evaluation under multiple trait models.
Journal of Dairy Science 67, 15671580. doi:10.3168/jds.S0022-0302
(84)81479-4
Viana JMS, Almeida IF, de Resende MDV, Faria VR, Silva FF (2010a) BLUP
for genetic evaluation of plants in non-inbred families of annual crops.
Euphytica 174, 3139. doi:10.1007/s10681-009-0094-2
Viana JMS, Sobreira FM, Resende MDV, Faria VR (2010b) Multi-trait BLUP
in half-sib selection of annual crops. Plant Breeding 129, 599604.
doi:10.1111/j.1439-0523.2009.01745.x
Viana JMS, Faria VR, Silva FF, Resende MDV (2011) Best linear unbiased
prediction and family selection in crop species. Crop Science, (In press).
Wright AJ (1980) The expected efciencies of half-sib, testcross and S1
progeny testing methods in single population improvement. Heredity 45,
361376. doi:10.1038/hdy.1980.78

Manuscript received 28 January 2011, accepted 25 May 2011

522

Crop & Pasture Science

J. M. S. Viana et al.

Appendix 1.
ASRemL code for bi-trait analysis considering the same trait in S3 family and in testcross
BLUP analysis, S3 families and testcrosses, Beija-Flor population, program 2 (title line)
a 306 !p # additive value/gca effect
avd 306 # average dominance value
env 3 # environment
pop 7 # population
blof 21 # block relative to family test
evf !m0 # ev of S3 family; !m0 recodes 0 as missing values
yf !m0 # yield of S3 family
s # nal stand
m # grain moisture
rep 3 # replication
blot 36 # block relative to testcross test
evt !m0 # ev of testcross
yt !m0 # yield of testcross
ped.txt # pedigree le
A.grm # additive relationship matrix
D.grm # dominance relationship matrix
data.asd !asuv # data le; !asuv allows an error variance other than IS
!ddf # correction of d.f. by the Kenward-Roger method
!continue # more than 10 iterations (default), if necessary
!maxit 100 # maximum number of iterations
evf evt ~Trait Tr.pop at(Trait,2).env at(Trait,2).env.rep !r at(Trait,1).blof at(Trait,2).env.rep.blot Tr.a at(Trait,2).a.env at(Trait,1).avd !f mv # mv estimate
missing values
#yf yt ~Trait at(Trait,2).pop at(Trait,2).env at(Trait,2).env.rep Tr.s Tr.m !r at(Trait,2).env.rep.blot Tr.a at(Trait,2).a.env !f mv
predict a !vpv # !vpv save the prediction error variance matrix
!pin !dene
R gc 7 : 9 # to compute the standard error of the genetic correlation
2 1 5 # one R structure (diagonal) and ve G structures
0 0 id !S2 = 5 # initial value for evf error variance
0 0 id !S2 = 7 # initial value for evt error variance
at(Trait,1).blof 1
at(Trait,1).blof 0 id 1 # 1 is the initial value of var.(blof) for evf
at(Trait,2).env.rep.blot 1
at(Trait,2).env.rep.blot 0 id 1
Tr.a 2
Trait 0 us 5 1 1 # var.(evf) cov(evf, evt) var.(evt); initial values
a 0 ainv # ainv or giv1
at(Trait,2).a.env 1
at(Trait,2).a.env 0 id 2
at(Trait,1).avd 1
at(Trait,1).avd 0 giv2 10
The qualier !m0 is optional, since missing values can be specied as dot (.). In this case the qualier !mvinclude must be included

http://www.publish.csiro.au/journals/cp