You are on page 1of 7

Efficiency of the Use of Doubled-Haploids in Recurrent Selection

for Combining Ability


A. Bouchez* and A. Gallais
ABSTRACT population improvement, it becomes a method of recur-
In population improvement for combining ability, the use of selfed rent varietal development.
progenies increases genetic advance per cycle, but it can unduly in- Griffing (1975) has already considered some aspects
crease the cycle length. Haplodiploidization (HD) can be very efficient of the use of HD in recurrent selection for improving
because it induces complete homozygosity in a short period. We com- combining ability; however, he considered neither the
pared, from a theoretical approach, the potential of recurrent selection use of S1 or S2 progeny testcrosses, nor the possible use
with a tester using doubled haploids (SDHT), with selection with
of off-season nurseries. Strahwald and Geiger (1988)
testcrosses of S0, S1, or S2 plants, with or without the use of off-season
nurseries, for an annual plant like maize (Zea mays L.). With the
have shown that with the use of off-season nurseries,
same selection intensity, and without off-season nurseries, SDHT with the advantage of HD, for the development of lines tends
a 4-yr cycle is the most efficient method. Efficiency increases with to disappear. In a theoretical study on optimizing sugar
lower heritability, with an advantage in comparison to the test of S0 beet breeding plans based on selection among testcross
plants (S0T) of 40 to 50% at low heritability (h2 ⬍ 0.15) and 12% at progenies, Borchardt and Geiger (1997) showed that
high heritability (h2 ⫽ 0.8). Use of off-season nurseries reduces the testing S3 lines maximizes genetic gain compared with
advantage of SDHT. With a 3-yr cycle, SDHT remains the best at S1 lines or doubled haploid lines.
low heritability with a gain of 27% (for h2 ⫽ 0.1) in comparison to In our paper, we consider only the use of recurrent
S0T with a 2-yr cycle. When using constant effective size, the advantage
of SDHT is further reduced or suppressed at the benefit of S0T in 2
selection with a tester (which can be a population), with
yr. The use of HD in recurrent selection for combining ability has its or without the use of an off-season nursery. Recurrent
biggest advantage when heritability is low. Consideration of variety selection with HD is compared with recurrent selection
development will give more advantage to HD. at the S0 plant level (i.e., without inbreeding) and with
selection at the S1 and S2 plant level (i.e., with one or
two generations of self-pollination before crossing to
F or the improvement by recurrent selection of the
value of lines that can be derived from a population,
haplodiploidization (HD) appears to be a very efficient
a tester).
The following theoretical development assumes no
epistasis and no genotype ⫻ environment (G ⫻ E) inter-
process (Griffing, 1975; Gallais, 1989, 1990a, 1993b; action. Gallais (1991) has shown that epistasis is ex-
Goldringer et al., 1996). This is due to the fact that it pected to have a low effect on relative efficiency of the
allows direct selection on line value, whereas by the use breeding methods for combining ability. Furthermore,
of partially inbred progenies, selection can be biased by in a large range of realistic situations, the presence of
heterozygosity. However, considering the improvement
genotype ⫻ location interaction has only a small effect
of combining ability of a population, the usefulness of
on the ratio of the two phenotypic standard deviations
HD may be questioned. Indeed, in the absence of epista-
associated to any pair of methods when selection is on
sis, the combining ability of lines derived from an S0
plant can be predicted from the combining ability of average performance across locations. Interaction with
the S0 plant (Gallais, 1990b). Nevertheless, HD can be an uncontrolled factor, such as years, decreases the ac-
efficient in increasing the variance among progenies tual heritability. Thus, here such an effect is considered
from testcrosses and thus the heritability. HD also has through heritability. Methods are compared on the basis
the advantage of giving more uniform progenies. If the of the genetic advance per unit of time, at the level of
cycle length is increased in comparison with combining the breeding population. The cost of the methods is not
ability tests at the S0 level, the genetic advance per unit considered. However, it is assumed that for a given year
of time will not necessarily be greater. As already shown of testing, the number of plots is fixed. Furthermore,
by Griffing (1975), the length of HD process is a critical time is considered through the cycle length. The consid-
parameter to consider. Thus, the advantage of HD in eration of the cost would need to assess the whole breed-
recurrent selection for combining ability with a tester is ing plan, including varietal development (Geiger and
not obvious. However, an advantage of HD in recurrent Tomerius, 1997). For such a development, to be realistic,
selection is to reduce the time required for varietal de- it would be necessary to solve the problem of response
velopment by pedigree selection. Indeed, with the use to recurrent selection on several cycles, and response
of HD, if the tester is used as a parent of the new to pedigree selection, which are not well resolved. Our
hybrids, recurrent selection is more than a method of study is valid for choosing, at a given cycle of the breed-
ing population, the recurrent selection scheme having
A. Bouchez, INRA-UPS-INA.PG, Station de Génétique Végétale, the greatest genetic advance per unit of time.
Ferme du Moulon, 91190 Gif Sur Yvette , France; A. Gallais, INA.PG,
16 rue Claude Bernard, 75321 Paris Cedex 05, France, and INRA-
UPS-INA.PG, Station de Génétique Végétale, Ferme du Moulon,
91190 Gif Sur Yvette, France. Received 13 Jan. 1999. *Corresponding Abbreviations: HD, haplodiploidization; SDHT, single-doubled-hap-
author (bouchez@mons.inra.fr). loid descent recurrent selection for combining ability with a tester;
S0T, S1T, S2T, recurrent selection for combining ability with a tester
Published in Crop Sci. 40:23–29 (2000). with respectively S0, S1, S2 plants.

23
24 CROP SCIENCE, VOL. 40, JANUARY–FEBRUARY 2000

MATERIALS AND METHODS lines can be derived from an S0 plant. Increasing the number
of lines per S0 decreases the number of S0 plants studied, and,
Description of the Methods on average, it is at the expense of the total variance and
From a practical point of view, we will consider the case effective population size (Gallais, 1989). The best approach
of an annual crop such as maize, for which it is possible to is to derive only one line per S0 plant, which is equivalent to
grow off-season nurseries. Basically, a cycle of recurrent selec- the classical single-seed descent method (Gallais, 1988). This
tion has four stages: (i) production of plants to be tested, i.e., leads to the single-doubled-haploid descent recurrent selection
the derivation of selection units: S1, S2 plants or doubled- method (SDHT). Its cycle length depends mostly on the length
haploid lines; (ii) crossing of these units to the tester, and of the HD process. We only consider cases in which the whole
simultaneous selfing in order to maintain tested plants; (iii) HD process is well controlled, and does not take more than
progeny testing; and (iv) intercrossing of selected units. Note 2 yr. With an annual plant without off-season generation, and
that with a plant like maize, with no prolificacy, it may be assuming that the whole HD process takes 2 yr, the complete
difficult to self and cross simultaneously (S⫹TC) with the SDHT cycle will be 2 yr longer than S0T, or 1 yr longer than
tester used as male parent. The tester can be used as a female. S1T. If doubled-haploid lines can be developed from S0 plants
However, for early types, it may be impossible to produce within 1 yr, then SDHT becomes comparable to S1T for cycle
enough seeds for trials. In this situation, many breeders prefer length. With maximum use of off-season resources, all four
to produce S1, S2, or S3 progenies in order to select at the S0, stages of S0T or S1T can be realized within 2 yr, while SDHT
S1, or S2 plant level, respectively, and thereafter to cross these as S2T will take at least one more year. Then, with DH lines, the
progenies in isolation with the tester used as male parent. We cycle length can vary from 3 to 5 yr, and cannot be shortened
assume that field testing has to take place in a normal growing because of field testing constraints.
season to get a representative agronomic evaluation.
The four selection stages of a cycle are scheduled differently
for each selection method, according to the use of off-season Expression of Genetic Advance
nurseries and S⫹TC (Table 1). As a result, each method can Consider the case of a breeding population and its tester.
be applied with varying cycle lengths. With selection among To simplify the notation, the combining ability of a S0 genotype
S0 plants (S0T) and use of S⫹TC, the length of the cycle will will be denoted AT , i.e., the additive effect of the S0 genotype
be three generations, i.e., 3 yr with an annual plant without crossed to a specific tester (Gallais, 1989; 1990b). The testcross
access to an off-season nursery and only 2 yr with an off- value with a given tester is equivalent to a quantitative charac-
season intercrossing. It cannot be shorter, because we assumed ter without dominance. Note that, with this notation, if the
agronomic tests will occur in a normal growing season. If S1 tester is the population itself, AT is one-half A, the classical
progenies of the S0 plants are used for crossing to the tester, additive effect for per se value. A general expression of the
the cycle length will be four generations long, and could last expected genetic advance per cycle has been given by Gallais
from 2 to 4 yr depending on the use of off-season nurseries; (1991)
a realistic length is 3 yr. With selection among S1 plants (S1T),
or S2 plants (S2T), the cycle needs one or two more generations ␪ covPTOT
⌬G ⫽ i , [1]
for selfing. Depending on the use of off-season nurseries, and √var PT
on the kind of progenies used to evaluate combining ability,
the cycle length can vary from 2 to 5 yr for S1T and from 3 where i is the selection intensity, ␪ the degree of selection
to 6 yr for S2T. control on both sexes, cov PTOT is the parent-offspring covari-
In the application of recurrent selection with HD, several ance for testcross value, and var PT the phenotypic variance

Table 1. Cycle description for four methods of recurrent selection for combining ability with a tester, in the case of an annual crop such
as maize. Plants are grown in summer (SUM) or winter in off-season nurseries (WIN).
Year 1 Year 2 Year 3 Year 4 Year 5 Year 6
Selection Cycle length
method† SUM WIN SUM WIN SUM WIN SUM WIN SUM WIN SUM
years
S0 T 4 S‡ – TC – T – IC
3 S⫹TC – T – IC
3 S TC T – IC
2 S⫹TC – T IC
2 S TC T IC
S1T 5 S – S – TC – T – IC
4 S – S⫹TC – T – IC
4 S S TC – T – IC
3 S S TC – T IC
2 S S⫹TC T IC
S2T 6 S – S – S – TC – T – IC
5 S – S – S⫹TC – T – IC
5 S – S S TC – T – IC
4 S – S – S⫹TC – T IC
4 S – S S TC – T IC
3 S S S⫹TC – T IC
3 S S S TC T IC
SDHT 5 ⇐ HD ⇒ – TC – T – IC
4 ⇐ HD ⇒ – TC – T IC
4 HD ⇒ TC – T – IC
3 HD ⇒ TC – T IC
3 ⇐ HD ⇒ TC T IC
† Methods of recurrent selection for combining ability with testcrosses of S0 plants: S0T, of S1 plants: S1T, of S2 plants: S2T, or of doubled haploids: SDHT.
‡ S: selfing, TC: crossing with a tester, T: agronomical test, IC: intercrossing, S⫹TC: selfing and testcrossing on the same plant, HD: complete haplodiploidiza-
tion process.
BOUCHEZ & GALLAIS: DOUBLED-HAPLOIDS IN RECURRENT SELECTION FOR COMBINING ABILITY 25

among selection units. cov PTOT is equal to (1 ⫹ F)/2 ␴A2 T, ␴A2 T of replications and of the plot-heritability h2S01 at the S0 level:
being the additive variance for testcross value, F the coefficient
of inbreeding of selection units: F ⫽ 0 for S0 plants or S1 (1 ⫹ F ) hS01 ␴AT 1
⌬G ⫽ i . [5]
families, 1/2 for S1 plants or S2 families, 3/4 for S2 plants or S3 √(1 ⫹ F ) h2S01 ⫹ (1 ⫺ h2S01)/b t
families and 1 for doubled haploid lines. As in our situation
␪ ⫽ 2, it results It can be noted that when heritability is low, i.e., there
is a great environmental variance in comparison to genetic
(1 ⫹ F ) ␴2AT variance, phenotypic variances under different methods can
⌬G ⫽ i . [2] be considered as approximately equal. Then, differences in
√var PT genetic advance only depend on the genetic variance among
With an experimental design with n individuals per plot progenies multiplied by (1 ⫹ F ) and divided by cycle length.
and b replications, the phenotypic variance among selection At the other extreme, when heritability is high, differences in
units is genetic advance are due to the square root of genetic variance
among progenies multiplied by (1 ⫹ F )1/2, and divided by the
var PT ⫽ ␴2GBT ⫹ ␴2p/b ⫹ (␴2WGT ⫹ ␴2e)/bn, cycle length. These two extremes for the genetic advance in
where ␴2GBT is the genetic variance between progenies, ␴2p the standard units of the genetic variance (␴AT ⫽ 1) are easy to
environmental variance between plots, ␴2e the environmental compute (Table 2).
variance at the plant level within plot, and ␴2WGT the within
progeny genetic variance. It can be shown (see appendix) that Comparison of the Methods
␴2GBT ⫽ (1 ⫹ F ) ␴2AT The different methods will be compared according to their
genetic advance per unit of time with or without the use of
and off-season nurseries (Table 1). Methods can be compared at
the same selection intensity or on the basis of the same effec-
␴2WGT ⫽ (1 ⫺ F ) ␴2AT ⫹ ␴2WT, tive size of the population of intercrossed plants. At the same
␴2WT being the genetic variance due to heterogeneity of the selection intensity, their relative efficiencies depend only on
tester. When selection units are doubled-haploid lines, the three parameters: heritability, cycle length, and inbreeding
within progeny genetic variance depends only on the heteroge- coefficient. Such a comparison favors the schemes with the
neity of the tester, ␴2WT being then zero when the tester is a most inbred selection units. Ignoring the other parameters,
the selection of N completely homozygous individuals leads
homozygous line.
to an effective size that is half that with the selection of N
The general expression [1] of genetic advance per cycle can
non-inbred individuals. This is equivalent to having a greater
be transformed to get explicit heritabilities of the associated
selection intensity. To ensure comparable effective population
testing systems. Then
size for the various methods, if pS0 is the selection rate for
(1 ⫹ F ) ␴2AT S0T, pS1, pS2 and pDH for S1T, S2T and SDHT, respectively, it
⌬G ⫽ i√1 ⫹ F hF ␴AT with h2F ⫽ . [3] is necessary to set pS1 ⫽ 1.5 pS0, pS2 ⫽ 1.75 pS0 and pDH ⫽
var PT
2 pS0. More generally, with an inbreeding coefficient F among
Assuming that the number of plants per plot is sufficiently selection units, the selection rate pF applied to the method
great to neglect the contribution of the within-plot genetic must equal (1 ⫹ F ) pS0. For high values of F, the intensity
variance in comparison to the between-plot environmental may become unrealistic. For example, to compare SDHT and
variance and to the genetic variance among progenies, i.e., S0T with the same effective genetic size, if pS0 is 0.15 then
n ⬎ 30 (Gallais, 1993a), the phenotypic variance becomes pDH becomes 0.30. Considering the cost of SDHT, such a low
var PT ⫽ (1 ⫹ F ) ␴2AT ⫹ ␴2p/b. selection intensity is quite unrealistic. Thus, both types of
comparisons were developed, with pS0 ⫽ 0.05 and 0.10 for the
Then, the design heritability at the S0 level is studies with the same effective size.
Methods are compared first on the basis of the same number
␴2AT of replications, and second at their optimum number. To deter-
h2S0 ⫽ .
␴2AT ⫹ ␴2p/b mine such an optimum, we consider a fixed total number of
plots for a given year of testing. Then, for a given number of
Heritabilities for designs with the same number of replications intercrossed plants, the rate of selection with b replications is
but different F values can be expressed in terms of h2S0 pb ⫽ b p1. Increasing the number of replications increases the
design-heritability but decreases the selection intensity. The
(1 ⫹ F ) h2S0
h2F ⫽ . first effect is favorable, the second unfavorable. Then an opti-
(1 ⫹ F ) h2S0 ⫹ (1 ⫺ h2S0) mum number may result. At this optimum, comparisons have
been developed on the basis of the same number of inter-
When expressed per unit of Time t, genetic advance becomes crossed plants and on the basis of the same effective size.
⌬G ⫽ (i√1 ⫹ F hF ␴AT)/t
or RESULTS
(1 ⫹ F ) hS0 ␴AT 1 Selection with Same Selection Intensity
⌬G ⫽ i . [4]
√(1 ⫹ F ) h ⫹ (1 ⫺ h ) t
2
S0
2
S0
Without Off-Season Nurseries
The breeding methods can be easily compared on the basis Without off-season nurseries, considering the shortest
of the same number of replications. However, the optimum possible schemes with same selection intensity (Fig. 1),
number of replications can vary according to the level of SDHT in 4 yr is the best method whatever the heritabil-
inbreeding. To determine this optimum, Formula [4] was trans- ity. Its superiority is all the greater when the heritability
formed to express the genetic advance in terms of the number is low. With low S0-heritability (h2S0 ⫽ 0.10), the gain in
26 CROP SCIENCE, VOL. 40, JANUARY–FEBRUARY 2000

Fig. 2. Genetic advance per year (in genetic standard deviations) as a


Fig. 1. Genetic advance per year (in genetic standard deviations) as a function of S0 heritability under different recurrent selection methods,
function of S0 heritability under different recurrent selection methods, with use of off-season (with same selection rate pS0 ⫽ 0.05).
when no off-season is used (with same selection rate pS0 ⫽ 0.05).

length (Table 1). A 3-yr cycle has also been considered


efficiency in comparison to S0T in 3 yr is 43%, and tends
for S1T, because the 2-yr cycle is quite complicated.
towards 50% as heritability approaches zero (Table 2).
Whatever the heritability, S1T in 2 yr obviously becomes
With h2S0 ⫽ 0.40, the gain is still 27%, and for h2S0 ⫽ 0.80,
more competitive than all other methods. It is followed
the gain is only 12%. Gains for SDHT in 5 yr are 25% by SDHT in 3 yr, S0T in 2 yr and S1T and S2T in 3 yr.
less. S2T in 5 yr is clearly the worst whatever the herita- In 4 yr, SDHT is the least efficient, especially for high
bility and at most equal to S0T at low heritability. Other heritabilities. In 3 yr, it is still a competitive method for
methods are relatively close to S2T when S0-heritability medium to low heritability. The relative efficiencies with
is low. When S0-heritability increases, S0T in 3 yr and same selection intensity (SI) can also be expressed in
S1T in 4 yr are significantly better than S2T and SDHT reference to S0T in 2 yr (Table 3). These results show
in 5 yr. the strong effect of the cycle length on the genetic ad-
vance per unit of time. The rank of the methods is
With the Use of Off-Season Nurseries approximately the rank of the cycle length. This illus-
To restrict the number of curves on Fig. 2, each selec- trates the strong effect of off-season nurseries. Notations
tion method was studied at its shortest realistic cycle for selection method will be followed by the cycle length

Table 2. Relative efficiency of four recurrent selection methods for combining ability with a tester, for various heritability values (h2S0),
and for realistic cycle length using off-season generation or not (italics). Relative efficiency is expressed in reference to the genetic
advance for S0T in three years, for 3 situations: same Selection Intensity (SI), same Effective Size (ES) with selection rate ps0 ⫽ 0.05,
and same Effective Size (ES) with ps0 ⫽ 0.10.
Cycle length
2 years 3 years 4 years 5 years
Heritability Selection
h2S0 method SI ES (0.05) ES (0.10) SI ES (0.05) ES (0.10) SI ES (0.05) ES (0.10) SI ES (0.05) ES (0.10)
S0T 150 150 150 100 100 100 75 75 75 – –
0.0† S 1T 225 206 199 150 137 133 113 103 100 – – –
S2T –‡ – – 175 154 147 131 116 110 105 93 88
SDHT – – – 200 170 160 150 128 120 120 102 96
0.1 S1T 220 201 194 146 134 130 110 100 97 – – –
S2T – – – 169 149 142 127 112 106 101 89 85
SDHT – – – 191 162 152 143 122 114 114 97 91
0.4 S1T 205 188 182 137 125 121 103 94 91 – – –
S2T – – – 153 135 129 115 101 97 92 81 77
SDHT – – – 169 144 135 127 108 101 101 86 81
0.8 S1T 190 174 168 127 116 112 95 87 84 – – –
S2T – – – 138 122 116 104 91 87 83 73 70
SDHT – – – 149 127 119 112 95 89 89 76 71
1.0 S1T 184 168 163 122 112 108 92 84 81 – – –
S2T – – – 132 117 111 99 87 83 79 70 67
SDHT – – – 141 120 113 106 90 85 85 72 68
† Limit when heritability decreases towards zero.
‡ Unrealistic situation for maize.
BOUCHEZ & GALLAIS: DOUBLED-HAPLOIDS IN RECURRENT SELECTION FOR COMBINING ABILITY 27

Table 3. Relative efficiency of five recurrent selection methods


for combining ability with a tester, for various heritability val-
ues (h2S0) with same selection intensity, and according to their
cycle length (in parentheses). Relative efficiency is expressed
in reference to the genetic advance for S0T in 2 yr.
Selection method
Heritability
h2S0 S1T(2) S1T(3) S2T(3) SDHT(3) SDHT(4)
0.1 146 97 112 127 95
0.4 136 91 102 112 84
0.8 126 84 92 100 75

in parentheses; S1T(2) means S1T in 2 yr. To get the


advantage of S1T(2) (46% for h2S0 ⫽ 0.10 and still 26%
for h2S0 ⫽ 0.80), it is necessary to conduct simultaneously,
in off-season, selfing and crossing to the tester. If we
look (in Table 3) at the four most comparable schemes,
with the maximum realistic use of off-season nurseries
for a species like maize, i.e., S0T(2), S1T(3), S2T(3), and Fig. 3. Domains of efficiency for SDHT compared with S0T, S1T and
SDHT(3), then SDHT(3) gives a gain of 27% for h2S0 ⫽ S2T, according to relative cycle length (ratio of the SDHT length
to the length of the compared method), and to S0 plot-heritability
0.10, and is equal to S0T(2) for h2S0 ⫽ 0.80. S2T(3) is only with (same selection intensity).
about 10% less than SDHT(3) whatever the heritability.
SDHT in 4 yr is at most equal to S0T in 2 yr at low are nearly vertical. This indicates a low effect of herita-
heritability, and is 25% less at high heritability. bilities on these boundaries, mainly for the boundary
S2T-SDHT. The boundary S0T-SDHT clearly depends
Selection with Same Effective Size on S0-heritability. This means that the relative cycle
As expected, the relative efficiencies for SDHT, S2T, length of SDHT can be longer at low heritabilities than
and S1T are reduced relatively to selection at the same at high heritabilities. For example, with SDHT in 4 yr
selection intensity, and all the more as the tested plants and S0T in 2 yr, the ratio is 2, and then whatever the
are more inbred (Table 2). With a 5% rate of selection heritabilities, S0T will be better than SDHT (approxi-
in S0T, the reduction in relative efficiency is about 10% mately equal at very low heritabilities). With a shorter
for S1T, 12% for S2T, and 15% for SDHT. With this SDHT cycle (3 yr) and S0T in 2 yr, with S0-heritability
selection intensity, it is clear that the results are not less than 0.80, SDHT will be the best, whereas if h2S0 ⬎
drastically changed. When a method has a strong advan- 0.80, it will be S0T.
tage on the basis of the same selection intensity for all Because S1T can be a competitive method, domains
methods, it also has a great advantage on the basis of of efficiency were calculated for this method (Fig. 4).
the same effective size. Without off-season nurseries, This method appears more efficient than SDHT only
this increases the inferiority of S2T and SDHT in 5 yr. when its cycle is approximately 20% shorter. The
S1T in 4 yr also becomes significantly inferior to S0T. boundary between S1T and S2T is nearly vertical at a
The advantage of SDHT in 4 yr is still 22% at low relative cycle length of 0.9. This means that for the same
heritability but disappears at high heritability. With the cycle length, as expected, S2T is more efficient than S1T.
use of off-season nurseries, the advantage of S1T in 2 yr If the cycle length for S1T is shorter than that for S2T,
in comparison to S0T in 2 yr is 34% for h2S0 ⫽ 0.10 and
16% for h2S0 ⫽ 0.80. All other methods are inferior to
S0T in 2 yr, except SDHT in 3 yr at low heritabilities.
In particular, SDHT in 3 yr remains better than S0T in
3 yr, even at high heritabilities. With a lower selection
intensity (pS0⫽ 0.10), the reduction in relative efficiency
of SDHT in 3 or 4 yr is about 6% greater (Table 2).

Effect of the Cycle Length


As long as the cycle for SDHT is not longer than for
other methods, SDHT is obviously the most efficient
whatever the heritability (Fig. 3). When its cycle is about
25% longer than that for S1T or about 10% in compari-
son to S2T, it becomes less efficient. For example, with
SDHT in 4 yr and S1T in 3 yr, the ratio of cycle length
is 1.33, and then S1T will be better than SDHT. If SDHT
Fig. 4. Domains of efficiency for S1T compared with S0T, S2T and
is in 5 yr and S2T in 4 yr, the cycle length ratio is 1.25, SDHT, according to relative cycle length (ratio of the S1T length
and then S2T will be better than SDHT. Note that the to the length of the compared method), and to S0 plot-heritability
separations of the domains S2T-SDHT and S1T-SDHT (same selection intensity). (a) S1T ⬎ S2T and S1T ⬍ SDHT.
28 CROP SCIENCE, VOL. 40, JANUARY–FEBRUARY 2000

DISCUSSION
When off-season nurseries are not used, it appears
that SDHT can increase genetic advance per unit of
time in comparison to the other methods when a cycle
can be achieved in 4 yr. With the use of off-season
nurseries, SDHT in a 3-yr cycle has an advantage only
with low heritability values. This advantage is reduced,
and even suppressed if comparisons are made on the
basis of the same effective genetic size. In this case,
excluding S1T in 2 yr, which can be difficult to apply in
a species like maize, S0T in 2 yr appears to be the most
efficient except at very low heritability (Table 2). This
is due to the shortness of its cycle. As previously men-
tioned, the cycle length has a greater effect than the
coefficient of inbreeding and the rank of the methods
corresponds approximately to their cycle length, the
worst having the longest cycle. Indeed, reducing the
cycle length from three to 2 yr increases genetic advance
per year by 50%. One self from S0 to S1 can increase
the expected genetic advance per cycle by 50% at low
Fig. 5. Genetic advance per cycle (in genetic standard deviations) as heritability, and the gain per year is only 22%. With
a function of the number of replications of selection units, for high heritability, the genetic advance per year is reduced
the same number of intercrossed plants, with the same selection to 8% (Table 2). This illustrates that the additional gain
intensity (pS0 ⫽ 0.05 for design without replication) and with S0 plot-
heritability ⫽ 0.2.
expected by the use of inbreeding can be suppressed by
the associated longer cycle. The use of inbreeding is
for example, S1T in 3 yr and S2T in 4 yr, the relative efficient, mainly at low heritability, if it can be accom-
cycle length is 0.75, and then, S1T is better than S2T. plished with the use of off-season nurseries. Even with
The boundary between S1T and S0T depends more on SDHT, it is necessary to have a well controlled HD
S0-heritability. For high heritabilities, h2S0 ⫽ 0.80, with a process to have SDHT among the best methods for the
cycle length ratio greater than 1.30 (for example, S1T genetic advance per unit of time. It now appears possible
in 4 yr and S0T in 3 yr) S0T is more efficient than S1T. in maize to complete the HD process in 12 or 18 mo
When heritability is lower S1T remains the best method. by the in situ gynogenesis process (Bordes et al., 1997).
For low heritability, with a ratio of 1.50 (S1T in 3 yr and The development in off-season nurseries for simultane-
S0T in 2 yr), S0T tends to be the best. ous pollination and crossing to the tester (S⫹TC) can
It is obvious from Fig. 3 and 4 that the cycle length decrease the cycle length for several schemes (Table 1)
ratio has a stronger impact than the heritability value. and then increase genetic advance per unit of time. In
The relative cycle length determines most of the relative this case, S1T in 2 yr will be one of the best schemes.
efficiency, while heritability modulates this relative effi- The difficulty with such a scheme, for a plant like maize,
ciency to a lesser extent. Consideration of the same will be the low quantity of seeds produced, even by
effective size still decreases the effect of heritability, crossing the candidate plant to several plants of the
even for the comparison between S0T and SDHT. tester, a solution which implies the use of a pure line
or a single-cross as tester.
Comparison at the Optimum As far as the influence of heritability is concerned,
The optimum number of replications depends on the we must note that in maize breeding for grain yield,
selection method, i.e. on the inbreeding coefficient (F) plot heritability at the S0 level is often about 0.40 to
of selection units. Obviously, the cycle length has no 0.50, leading to design–heritability greater than 0.50 with
influence on such an optimum. With SDHT, two replica- two or three replicates (Moreno-Gonzalez and Hal-
tions are optimum only when h2S0 is lower or equal to lauer, 1982; Gallais, 1996; Sampoux and Gallais, 1996).
0.2 (Fig. 5). As the selection units have a smaller in- With design heritabilities of 0.60 to 0.70 and the same
breeding coefficient, the optimum number of replica- selection intensity, the relative genetic gain obtained by
tions increases up to three for S0T at h2S0 ⫽ 0.2. The the use of SDHT, in comparison to S0T, is between 0
optimum is relatively flat, and thus the differences in and 25%. With the same effective size, there will be no
genetic advance with 1, 2, or 3 replications are small. significant gain, and worse, SDHT can be lower than
When heritability is greater or equal to 0.5 , an evalua- S0T. Note that a gain of 10% is low in absolute value.
tion based on single replicate trials ensures the best Thus, considering HD as a costly process, its interest
genetic advance whatever the method. For a given heri- for the improvement of combining ability in recurrent
tability, the optimum is approximately the same for each selection is not as clear as for the improvement of line
method. With a constant effective size, conclusions value. However, genotype ⫻ year interaction will fur-
about the optimum number of replications are not ther reduce heritability. If genotype ⫻ year interaction
changed. represents 50% of the total genetic variance, operational
BOUCHEZ & GALLAIS: DOUBLED-HAPLOIDS IN RECURRENT SELECTION FOR COMBINING ABILITY 29

heritability will not be 0.60 to 0.70 but 0.30 to 0.35. Then, From [1] :
the advantage of SDHT will be increased. Whatever ␴2Gtot ⫽ E (␣ip)2 ⫹ E (␣ti)2 ⫹ E (␤ijpt )2.
the situation, there still remains an interest in SDHT
because it allows rapid development of new hybrids, According to [2], the first term is equal to 2␴2AT and the last
particularly when off-season nurseries are not used. two represent variance due to the heterogeneity of the tester
(␴2WT). Then,
Moreover, HD produces completely homozygous lines,
while some residual heterozygosity is possible when ␴2Gtot ⫽ 2␴2AT ⫹ ␴2WT
lines are derived by single-seed descent or pedigree and from [4]
selection. Then, if the HD process is expensive, it could
be used every second cycle of recurrent selection to ␴2GWT ⫽ (1 ⫺ F )␴2AT ⫹ ␴2WT
quickly develop hybrids, in alternation with the use of For a set of loci, with a population in linkage equilibrium and
S0T. S0T in 2 yr is the shortest selection scheme available without epistasis, the formulae for variances remain valid,
that allows separation of self-pollination and crossing except that variance components represent summation of one-
to the tester. Note also that, from the point of view of locus components over the set of loci.
the existence of genotype ⫻ year interaction, with the
same expected genetic advance, a short cycle will be ACKNOWLEDGMENTS
better than a long cycle. In 6 yr, three complete S0T The authors are very grateful to the reviewers and to Prof.
cycles can be developed, whereas only two cycles will Kendall R. Lamkey, Technical Editor, for their helpful sugges-
be developed for SDHT in 3 yr. Obviously, the breeder tions and revision of the English.
has to consider the cost of genetic advance. However,
it is sometimes efficient to invest more, even if this REFERENCES
increases the cost of genetic advance, in order to achieve Borchardt, D.C., and H.H. Geiger. 1997. Theoretical studies on opti-
quicker varietal development, allowing success on the mizing sugar beet breeding plans. p. 61–66. In P. Krajewski and
Z. Kaczmarek (ed.) Biometrics in Plant Breeding, Proc. 10th Meet-
market, which will make the investment profitable. Use ing of the EUCARPIA section, Poznan. EUCARPIA, Wagen-
of doubled-haploids, if well controlled, is a tool to satisfy ingen, the Netherlands.
this goal. Bordes, R., R. Dumas de Vaux, A. Lapierre, and M. Pollacseck.
1997. Haplodiploidization of maize (Zea mays L.) through induced
APPENDIX gynogenesis assisted by glossy markers and its use in breeding.
Agronomie 17:291–297.
Derivation of Between and Within Gallais, A. 1988. A method of line development using haplodiploidiza-
Progenies Variance tion: The single doubled-haploid descent recurrent selection. Theor.
Appl. Genet. 75:330–332.
Consider, for one locus, a random mating population with Gallais, A. 1989. Optimization of recurrent selection on the phenotypic
alleles A1p, A2p, . . . Aip, Ajp and a tester population with alleles value of doubled haploid lines. Theor. Appl. Genet. 77:501–504.
A1t , A2t , . . . Ait, Ajt. Gallais, A. 1990a. Quantitative genetics of doubled haploid popula-
The value of a genotype Aip, Ajt from the population x tester tions and application to the theory of line development. Genet-
ics 124:199–206.
cross can be written Gallais, A. 1990b. Application of the concepts of the test value and of
varietal value to the study of genetic advance in recurrent selection.
Yijpt ⫽ ␮T ⫹ ␣ip ⫹ ␣tj ⫹ ␤ijpt [1] Euphytica 48:197–209.
where ␮T is the mean of all testcross progenies, ␣ip the additive Gallais, A. 1991. A general approach for the study of a population of
test-cross progenies and consequences for the recurrent selection.
effect of allele Aip in combination with the tester, ␣jt the additive Theor. Appl. Genet. 81:493–503.
effect of allele Ajt in combination with the population, and Gallais, A. 1993a. Field of efficiency of breeding methods for per se
␤ijpt the dominance effect, i.e. the interaction between two al- value or combining ability in plant breeding. Agronomie 13:467–
leles, one from the population and the other from the tester. 480.
Now consider the combining ability of a genotype Aip Ajp Gallais, A. 1993b. Efficiency of recurrent selection methods to im-
from the S0 population. It is prove the line value of a population. Plant Breeding 111:31–41.
Gallais, A. 1996. Combined testcross and S1 selection for the improve-
T(ij) ⫽ ␮T ⫹ 1/2(␣ip ⫹ ␣jp ) ment of testcross and S1 performances. Crop Sci. 37:1126–1133
Geiger, H.H., and A. Tomerius. 1997. Quantitative genetics and opti-
or mum breeding plans. p. 15–26. In P. Krajewski and Z. Kaczmarek
T(ij) ⫽ ␮T ⫹ T ␣i ⫹ T ␣j (ed.) Biometrics in Plant Breeding, Proc. 10th Meeting of the EUC-
ARPIA section, Poznan, Poland. EUCARPIA, Wageningen, the
with Netherlands.
Goldringer, I., P. Brabant, and A. Gallais. 1996. Theoretical compari-
T ␣i ⫽ 1/2 ␣ip. [2] son of recurrent selection methods for the improvement of self-
The variance among progenies will be pollinated crops. Crop Sci. 36:1171–1180
Griffing, B. 1975. Efficiency changes due to use of doubled-haploids
␴2GBT ⫽ ␴2AT in recurrent selection methods. Theor. Appl. Genet. 46:367–386.
Moreno-Gonzalez, J., and A.R. Hallauer. 1982. Combined S2 and
with crossbred family selection in full-sib reciprocal recurrent selection.
Theor. Appl. Genet. 61:353–358.
␴2AT ⫽ 2E (T ␣2i ) Sampoux, J.P., and A. Gallais. 1996. Expected genetic gains in testcross
and more generally with inbred plants and S1 performances for several populations of silage maize with
combined S1-testcross selection. Maydica 41:167–179.
␴2GBT ⫽ (1 ⫹ F )␴2AT. [3] Strahwald, J.F., and H.H. Geiger. 1988. Theoretical studies on the
usefulness of doubled haploids for improving the efficiency of re-
The within progeny variance will be current selection in spring barley. p. 1–13. In Biometrics in Plant
Breeding, Proc. 5th Meeting of the EUCARPIA section, Aas,
␴2GWT ⫽ ␴2Gtot ⫺ ␴2GBT. [4] Norway. EUCARPIA, Wageningen, the Netherlands.

You might also like