CHAPTER FOUR

A practical guide to genetic gain

Jessica E. Rutkoski*
International Rice Research Institute, Rice Breeding Platform, Metro Manila, Philippines
*Corresponding author: e-mail address: [email protected]

Contents
1. Introduction 218
2. The Basics of genetic gain and population improvement 219
2.1 Definition 219
2.2 How genetic gain is achieved 219
2.3 An example of genetic gain over cycles of breeding 222
2.4 Rationale for population improvement 223
3. Selection techniques useful for population improvement 225
3.1 Utilizing multiple sources of information for selection 225
3.2 Direct and indirect selection 227
4. Expected genetic gain 228
4.1 Components of phenotypic value 229
4.2 Expected breeding value of progeny and their parents 230
4.3 The Breeder’s equation 231
4.4 Equivalent versions of the Breeder’s equation 231
4.5 Stability of genetic variance 235
5. Realized genetic gain 236
5.1 Examples of genetic gain realized 236
5.2 Estimating realized genetic gain 238
6. How to improve rates of genetic gain 240
6.1 Foundational components 240
6.2 Advanced components 243
7. Conclusion 244
Acknowledgment 244
Appendix. R-function used for simulation 244
References 245

Abstract
An understanding of the inheritance of quantitative traits, those with a continuous phe-
notype, was first established in the early 1900s. This was instrumental for breeding
because quantitative genetic theory provides the basis for the development of methods
which can be used to increase the rate of genetic improvement, referred to as “genetic
gain,” within a breeding population over time. Today, the concept of genetic gain and
its basis in quantitative genetics is often not well understood among crop breeders and

Advances in Agronomy, Volume 157 # 2019 Elsevier Inc. 217

ISSN 0065-2113 All rights reserved.
https://doi.org/10.1016/bs.agron.2019.05.001
218 Jessica E. Rutkoski

scientists, often resulting in inefficient or ineffective crop improvement efforts. This

chapter aims to provide clarity on genetic gain to help those engaged in crop improve-
ment to take actions that will enable them to be more successful. To do so, a thorough
introduction to genetic gain and the population improvement cycle is provided along
with a review of selection techniques essential for breeding. Next, I demonstrate why
genetic improvement on a population basis is needed facilitate variety development. In
order to show that the genetic gain is tractable, the theory behind genetic gain and its
prediction is explained, followed by a discussion on realized genetic gain including a
review of methods that can be used for its estimation. Lastly, guidance is given on
how to improve rates of genetic gain in applied breeding programs.

1. Introduction
For the majority of agricultural history, selective breeding of animal
and plant species was done without a formal understanding of how selection
leads to genetic improvement. The basic principles of inheritance were first
described by Mendel (1866) and these principles could only fully explain the
inheritance of traits that fall into discrete categories. Important characters
like reproductive fitness in natural populations, as well as the yield of grain
crops and carcass weight of livestock were known to exhibit a continuous
range of variation. These characters are referred to as “quantitative traits.”
It was not until 1918 when the model for the inheritance of quantitative
traits based multiple Mendelian factors was comprehensively described
by the statistician and geneticist Fisher (1918). Also around that time, the
geneticist Sewall Wright described many quantitative genetic principles
that became instrumental for breeding (Wright, 1920, 1921). Animal
breeder, J.L. Lush was influenced by Wright’s work, and was one of the first
to apply quantitative genetics to breeding. The now widely recognized
breeder’s equation which describes how genetic improvement can be
predicted from one generation of selection to the next was just one of Lush’s
many important contributions (Lush, 1937).
Although ideas about the application of quantitative genetics for breed-
ing are over 80 years old, many plant breeding programs have not yet taken
full advantage of this knowledge to increase rates of genetic improvement.
This may be because there has been a great deal of emphasis on understand-
ing what are the genes or genomic regions which affect traits, in hopes
that this will enable a precise stacking of favorable genes. While such an
approach may be useful for monogenic or even oligogenic traits, it is not
a solution for the improvement for quantitative traits (Bernardo, 2008)
A practical guide to genetic gain 219

which are known to be influenced by a large number of genes. For such traits,
genetic improvement over cycles of selection is necessary so that favorable
alleles can be brought together gradually. Such a population improvement
approach is how natural selection works to drive adaptive evolution
(Fisher, 1930), and it is required for achieving genetic gain in breeding pro-
grams. This chapter aims to help improve the understanding and appreciation
of this concept for plant breeding.

2. The Basics of genetic gain and population

improvement
2.1 Definition
Genetic gain from selection, or simply “genetic gain,” is defined as the
improvement in average genetic value in a population or the improvement
in average phenotypic value due to selection within a population over cycles
of breeding (Hazel and Lush, 1942). Genetic gain may also be referred to as
response to selection which may be a better term for describing changes
that are not necessarily favorable. In this chapter, the symbol R is used to
represent genetic gain.
Discussions on genetic gain may be focused either on expected genetic
gain or realized genetic gain. Expected genetic gain is a prediction of the
actual change in phenotype that would occur due to the genetic changes
brought about by a proposed selection or a proposed breeding strategy.
Expected genetic gain can be estimated using parameters obtainable from
breeding experiments and given various assumptions. Realized genetic gain
is the observed gain due to selection over cycles. Occasionally, the term
genetic gain is used incorrectly to describe phenotypic differences between
individuals, such as when describing the yield advantage of a promising new
variety compared to a control. It is also incorrect to use the term genetic gain
to describe a trend in phenotype over time that was not the result of cycles of
selection within a population.

2.2 How genetic gain is achieved

Genetic gain can be achieved for virtually any trait as long as it is heritable.
It can also be achieved for total net merit, which is a combination of multiple
economically important traits. Selection for net merit is done using an
economic selection index (Hazel, 1943; Smith, 1936) which is a linear
combination of different traits of economic importance weighted optimally
so that selection based on the index maximizes expected genetic gain in net
220 Jessica E. Rutkoski

merit. Because the term genetic gain only applies to changes in population
mean over cycles of selection, in makes sense to talk about genetic gain in
traits that are heritable and conferred by multiple loci. This is because
improving such traits can be done most effectively over multiple cycles of
selection.
Explicitly stacking favorable alleles to achieve a desired genotype, back-
crossing for the introgression of favorable alleles, and mutation breeding are
breeding methods that do not produce genetic gain because these breeding
approaches do not lead to improvement the breeding population as a whole.
Even if these approaches were applied population-wide, once the targeted
set of desired loci are fixed, there can be no further improvement in
genetic value.
Strictly speaking, genetic gain can only be realized from executing at least
1 cycle of breeding. The breeding cycle required to realize genetic gain,
illustrated in Fig. 1, consists of four parts, generation, evaluation, selection,
and recombination. This process is also referred to as “population
improvement” or “recurrent selection.” Different breeding materials, such
as single non-inbred plants, or families such as inbred lines, may be used for
the processes of evaluation, selection, and recombination. For example, an
individual’s progeny from open-pollination may be used for evaluation and
its progeny from self-pollination for recombination (Goulas and Lonnquist,
1976). The breeding materials that are subject to selection are called

Fig. 1 The Breeding cycle. Recurrent cycles of breeding which consist of generation, eval-
uation, selection, and recombination, are required to achieve genetic gain over time.
A practical guide to genetic gain 221

selection units, those that are evaluated are called evaluation units, and those
that are recombined are called recombination units. In self-pollinated crops,
the same inbred lines are often utilized as the evaluation, selection, and
recombination units.
Generation is the development of the breeding materials that will be used
in the evaluation, selection, and recombination processes. The time required
to complete this step of the breeding cycle can be reduced by using off-
season nurseries or greenhouses (Collard et al., 2017; Watson et al.,
2018), or by using doubled haploid technology (Gallais and Bordes,
2007) when inbred lines are required.
Evaluation is the collection of phenotypic and/or genotypic data that
will be used in the selection process and it is the generally the most expensive
step in the breeding cycle. For phenotypic evaluation, conducting field trials
in multiple environments with appropriate field plot technique is a critical
component. High-throughput phenotyping and partially replicated exper-
imental designs can help to increase the efficiency of phenotypic evaluation
(Cullis et al., 2006; Haghighattalab et al., 2016; Williams et al., 2010),
thereby enabling evaluation of a larger number of breeding materials.
Selection is the identification of which selection units to recombine
based on the selection criteria which may be single phenotypic values, or
multiple phenotypic values combined using a selection index, Best Linear
Unbiased Prediction (BLUP) (Henderson, 1975) reviewed by Piepho
et al. (2008) or another method for predicting the selection units’ value
for breeding purposes such as those described by Gianola (2013). The selec-
tion units are then ranked based on the selection criteria and all those above
or below a certain threshold are selected. Selection done this manner is called
truncation selection. In truncation selection, there is no attempt made to
control how much each selection unit contributes to the next generation
nor to control the exact cross combinations made. Optimum contribution
selection (Meuwissen, 1997) is an alternative to truncation selection which
takes into account the selection criteria as well as the relationships between
the selection units to determine how much each one should contribute to
the next generation to control the rate of inbreeding. This allows one to
control short and long-term genetic gain.
Recombination is the reshuffling of allelic combinations found in the
selected breeding materials. This is done by open-pollination or controlled
crossing. If crosses are controlled it is possible to precisely plan which pairs to
cross, referred to as mate selection or mate allocation. Mate selection is useful
for increasing genetic gain for an index of multiple economically important
222 Jessica E. Rutkoski

traits if at least one of the traits contributes to economic return in a non-

linear way (Allaire, 1980). An example of a trait that would have a non-linear
contribution to economic return would be days to flowering where inter-
mediate values are favored as opposed to maximum or minimum values.
Within the breeding cycle, selection may be done in stages. For example,
selected bulk and pedigree breeding methods involve evaluation and selec-
tion during multiple self-pollination generations (Weber, 1984). With these
breeding methods, the breeding cycle consists of additional generate-
evaluate-select steps prior to recombination.
Implementation of Genomic Selection (GS) (Meuwissen et al., 2001)
can reduce the length of the breeding cycle because the selection units
can be single plants that are evaluated by genome-wide genotyping and then
selected based on predicted breeding values, where breeding value is the
selection unit’s value as a parent. This can dramatically hasten the generation
and evaluation steps. For example, in self-pollinated crops, single non-
inbred plants can be genotyped, selected, and recombined without first
deriving inbred lines from these plants and evaluating them phenotypically.
However, with GS, generation of evaluation units suitable for phenotyping,
and then phenotypic evaluation of these materials is still eventually required
in order to generate data that will be used to predict breeding values in future
breeding cycles.

2.3 An example of genetic gain over cycles of breeding

As a specific example of genetic gain realized from a simple breeding
program of a cross-pollinated crop, consider a large randomly mating pop-
ulation of maize (Zea mays) that is variable in resistance to Northern corn leaf
blight (NLB) caused by Exserohilum turcicum. Resistance to NLB is a quan-
titative trait (Poland et al., 2011) and more resistant maize plants are desired.
The maize population has not yet been subject to selection and pedigree
relationships between individuals are not known. This population, referred
to as the base population, is then used as the initial population of a new
breeding program for NLB resistance. To start the first breeding cycle,
500 plants from the base population are grown (generation) and then phe-
notypically evaluated for resistance to NLB (evaluation). Based on these
measurements the breeder selects 50 plants with the highest levels of
NLB resistance after flowering (selection), and the selected plants open-
pollinate such that the male parent is randomly selected from the population
(recombination). The progeny of these crosses forms a new population
referred to as the “selected population” from cycle one of selection.
A practical guide to genetic gain 223

For the second cycle of selection, the generate-evaluate-select-

recombine process (Fig. 1) is repeated, but this time starting from the
selected population that was just created. This population is evaluated,
the best 50 individuals are identified and open-pollinated, and their progeny
become the selected population resulting from cycle two. The breeding
cycle is repeated many more times, each time using the selected population
that was just developed in the previous cycle in the generate-evaluate-select-
recombine process (Fig. 1).
The genetic gain in NLB resistance realized by the example maize breed-
ing program after each cycle of selection can be observed by comparing the
average level of NLB resistance within the selected populations with the
average level of NLB resistance of base population as shown in Fig. 2.
(Assume that resistance is scored such that larger values mean higher levels
of resistance and lower values mean lower levels of resistance.) For instance,
the mean level of NLB resistance of the selected population after 1 cycle
minus the mean level of NLB resistance of the base population is equal
to the genetic gain in disease resistance due to 1 cycle of selection. The
observed increase in NLB resistance in the selected population is solely
due to heritable genetic factors. The observed differences cannot be due
to environment effects because both the base population and the selected
population are observed alongside one another in the same environment.
As more cycles of selection are conducted, the difference between the
selected population mean and the base population mean increases (Fig. 2).

2.4 Rationale for population improvement

Improvement of a population over cycles of breeding within a mostly closed
gene pool is necessary in order to be able to develop a steady stream of vari-
eties that are superior to those released previously. Although it is possible to
identify superior new varieties by sampling from a population that is not
improving due to ongoing cycles of selection, such as a germplasm collec-
tion, achieving genetic gain over cycles of selection increases the probability
of identifying superior new varieties on an ongoing basis. Consider for
example a large and diverse population from which promising new varieties
can be identified. A sample from this population is evaluated and then the
best individual is identified for promotion as a variety. Another sample is
taken from this population for evaluation and then the best individual is
again identified for promotion as a variety. The probability that a superior
individual can be identified the second, third, fourth and fifth time in a row is
1/2, 1/6, 1/24 and 1/120, respectively. Thus, by selecting individuals from
224 Jessica E. Rutkoski

Fig. 2 An example of realized genetic gain over 5 cycles of selection. Histograms depict
distributions of phenotypic values for the base and selected populations. A vertical dot-
ted line is drawn at the mean of each distribution. With each advancing cycle the dif-
ference between the base and selected population means increases by a small
increment. By cycle five, the base and selected population means are significantly dif-
ferent in this example.

the same population there is some chance that superior individuals can be
identified each time, but the probability becomes increasingly smaller the
more times one attempts to identify an individual that is superior to all others
previously identified.
Now consider that instead of sampling from the same static population
each time, the population is subject to recurrent cycles of breeding and the
A practical guide to genetic gain 225

Table 1 Probability that an individual superior to all others can be identified every cycle
of selection during population improvement.
Number of attempts
h2 2 3 4 5
0 0.5 0.17 0.04 0
0.1 0.59 0.33 0.21 0.16
0.3 0.67 0.51 0.45 0.43
0.5 0.75 0.66 0.65 0.64
h2 ¼ 0 is equivalent to no population improvement.

best individual is sampled from each successive breeding cycle for promotion
as a variety. The probability that a superior individual can be identified each
time now partially depends on the genetic gain achieved from each selection
cycle. Using a simple simulation carried out in the R language and environ-
ment (R Development Core Team, 2016), described in Appendix, the
probabilities of successfully identifying a superior variety two to five times
in a row are estimated for different levels of narrow sense heritability, h2,
(Table 1) assuming a population size of 1000 and that 30 individuals are
selected for intermating at each cycle. The parameter h2 is the squared accu-
racy of selection and indicates how effective selection will be for achieving
genetic gain. The case of h2 ¼ 0 is equivalent to making repeated selections
from the same population that is not undergoing selection.
Assuming the heritability ranges from 0.1 to 0.5, the probability that a
superior variety is identified five times in a row from a population improving
over cycles of selection was shown to be 19–87 times greater than the prob-
ability of doing so without population improvement. Through this exercise,
it becomes clear that breeding is not merely a game of chance. Achieving
genetic gain through recurrent cycles of breeding can stack the deck in
ones favor by greatly increasing the probability of identifying superior
new varieties on an ongoing basis.

3. Selection techniques useful for population

improvement
3.1 Utilizing multiple sources of information for selection
Previously, an example of selection for increased NLB resistance in a maize
breeding population was given. In this example the selection units and the
evaluation units were the same single plants, and only single phenotypic
226 Jessica E. Rutkoski

measurements were used as the criteria for selection. This method is called
phenotypic mass selection or simply mass selection. Mass selection is the
oldest form selection used in breeding. It requires no record keeping and
it can be done by simply collecting seeds of desirable plants after pollination.
Mass selection done in this manner only selects on females, while the male
parent is chosen at random through open pollination. If mass selection is
done prior to pollination, selection can be done on both female and male
parents by only intermating the selected individuals. Because mass selection
is done on single observations only, the selection units are also used as the
evaluation units, and generating separate breeding materials for the purpose
of evaluation is not needed. The accuracy of mass selection, which is usually
low for quantitative traits, is the square root of the narrow sense heritability
of the trait of interest in the base population.
In contrast to mass selection where the selection is based on single phe-
notypic values per selection unit, the criteria for selection may be multiple
phenotypic observations combined in a linear fashion into a single value,
referred to as a selection index. There are three types of selection indices that
were developed so that selection based on the index maximizes genetic gain.
First, the economic selection index was developed by Smith (1936) and
Hazel (1943) in which indices are based on phenotypic observations on mul-
tiple traits weighted in an optimal way to maximize genetic gain in total net
merit. The economic value of each trait, as well as the phenotypic and
genetic covariances among traits are incorporated in the weighting proce-
dure. The second is the family selection index developed by Lush (1947)
where indices are based on multiple phenotypic observations of a single trait
on breeding materials that are related. The index weights in the family selec-
tion index take into account the additive genetic covariance among relatives
and the heritability of the trait. The third type of selection index combines
both information from relatives and information from multiple traits
(Henderson, 1963). These selection index weights take into account the
economic value of each trait, the phenotypic and genetic covariances among
traits, and the additive genetic covariance among relatives.
Multiple phenotypic observations can also be combined using BLUP.
BLUP is a procedure that improves upon selection index methods in that
it can control for non-genetic factors. In the absence non-genetic factors,
BLUP is equivalent to a selection index. With BLUP, breeding materials
may be considered unrelated, in which case the BLUP values, referred
to as BLUPs, for the breeding materials are based only on their own data.
A practical guide to genetic gain 227

This is referred to as BLUP with genotypes independent and identically dis-

tributed (i.i.d.). Alternatively, breeding materials may be considered as
related based on pedigree or genome-wide marker based relationship
(Hayes et al., 2009), referred to as pedigree BLUP and genomic BLUP,
respectively. Pedigree and genomic BLUP enables the sharing of informa-
tion among different breeding materials that are related thereby improving
selection accuracy.
If multiple traits are phenotyped, multi-trait BLUP, which takes into
account the genetic and phenotypic covariances between traits, can be used
to utilize all sources of available information. Use of multi-trait BLUP
(Henderson, 1973) facilitates the use of economic selection indices because
the economic values are directly used to weight the multi-trait BLUPs when
estimating net merit for each individual.
Selection indices and BLUPs are estimates of genetic value or they are
estimates of breeding value if individuals are assumed related based on addi-
tive genetic relationships estimated using pedigree or genome-wide
markers. The benefit of using a selection index or BLUP for selection is that
that all phenotypic information can be utilized in an optimal way so that
selection accuracy is maximized. The accuracy of selection based on a selec-
tion index or BLUP is the square root of the reliability. The reliability
depends on the amount and strength of the information utilized as well as
heritability of the trait(s).

3.2 Direct and indirect selection

The example maize breeding program described previously illustrated a
form of direct selection. With direct selection, the selection criteria (either
a phenotypic value, selection index, or BLUP) is based on the trait(s) of
interest. For example, if the trait of interest is grain yield in the target envi-
ronment, selections are made using grain yield data from trials conducted in
the target environment. Selection based on traits that may be associated with
grain yield such as biomass yield, grain size, number of inflorescences, etc.,
would be considered indirect selection, and such traits are called secondary
traits. Likewise, selection based on the trait of interest measured in an envi-
ronment that does not represent the target environment can be considered
indirect selection (Cooper et al., 1993; Cooper and DeLacy, 1994). For
instance, if the target environment is tropical, grain yield measured in a tem-
perate environment would be considered a secondary trait and selection
228 Jessica E. Rutkoski

based on this trait would be considered indirect selection. Genomic selection

is a form of indirect selection because selection is based on a genotype and
not directly on phenotypic values.
Indirect selection is often used to eliminate a fraction of the selection
units prior to selection on the trait of interest especially when the trait of
interest is expensive to measure. For example, in rice breeding, the cost
of conducting sensory evaluations to assess grain quality is so high, quality
can only be measured on relatively few evaluation units. Therefore, to breed
for good or acceptable quality, rice breeders measure and select based on
traits which are associated quality such as amylose content.
While indirect selection can be a useful. It has been often used ineffec-
tively. For example, U.S. maize breeding in the early 1900s relied on selec-
tion based on appearance of the ear judged at “corn shows” (Bowman and
Crossley, 1908) in attempt to develop better yielding maize varieties. In
small grain crops, visual selection for yield component traits was often done
to try to improve grain yield. This strategy has since been shown to be an
ineffective (McGinnis and Shebeski, 1968; Rasmusson and Cannell, 1970).
To determine if indirect selection would be more effective than indirect
selection, a metric called the relative efficiency can be computed. The rel-
ative efficiency of indirect vs. direct selection is simply the ratio of the
expected genetic gain due to indirect selection to the expected gain due
to direct selection (Lerner and Cruden, 1947). Assuming the population
sizes and number of individuals selected is constant, the relative efficiency
of indirect and direct selection depends the selection accuracy of indirect
selection as well as the genetic correlation between the indirect and direct
selection criteria (Fig. 3). For the case of indirect selection based on markers,
indirect selection accuracy equals one, because a marker genotype is
completely heritable. In most cases direct selection is more effective than
indirect selection. When much larger population sizes are possible with indi-
rect selection compared to direct selection, there is a greater chance that
indirect selection would be a more favorable strategy.

4. Expected genetic gain

Now that genetic gain and the process that generates it has been
described, the theory behind why the breeding cycle leads to genetic gain
and how rates of genetic gain can be increased will be shown. To do so,
A practical guide to genetic gain 229

Fig. 3 Relative efficiency of indirect vs. direct selection phenotypic selection for differ-
ent levels of heritability (h2) of the trait of interest, h2 of the trait under selection, and
additive genetic correlation (rg) between the two traits. Equal selection intensities are
assumed for indirect and direct selection. Indirect selection is more effective than direct
selection wherever the relative efficiency values are greater than one.

two basic genetic principles are explained and then used to show how to
predict genetic gain per cycle and per unit time from simple phenotypic mass
selection and selection based on selection indices or BLUPs.

4.1 Components of phenotypic value

Characteristics of animals and plants are due to genetic and environmental
causes. For example, its well understood that animals or plants that receive
better than average nutrition will have better than average growth. This
environmental cause of better growth will not be passed down to progeny
when these individuals (single animals or plants) are mated. The heritable
portion of an individual’s observed phenotypic value is its breeding value.
Breeding value is the value of an individual as a parent and it is the sum total
of the additive effects that an individual’s alleles exert on the trait. Only
the additive allele effects contribute to breeding value because alleles, and
not allelic combinations, are transmitted to progeny due to independent
assortment during meiosis.
230 Jessica E. Rutkoski

Although observed phenotypic values are not completely heritable, they

can give some indication of breeding value. Using simple linear regression,
an individual’s breeding value can be predicted using its own phenotypic
value for a single trait or from a linear combination of multiple phenotypic
values in a selection index or BLUP. The regression formula that allows us to
predict breeding value based on phenotypic values, selection indices, or
BLUPs is g^i ¼ g + bxg ðxi  xÞ where g^i is the predicted breeding value of indi-
vidual i, g is the mean breeding value of the population from which i was
sampled, bxg is the regression coefficient between breeding value and phe-
notypic or selection index values, xi is the phenotypic or selection index
value of individual i and x is the mean phenotypic or selection index value
of the population from which i was sampled.
To express this in a different way, g can be subtracted from each side of
the equation to get g^i  g ¼ bxg ðxi  xÞ. The expression g^i  g is the genetic
superiority of individual i relative to the population mean, and xi  x is the
superiority of individual i in terms of its phenotype, selection index, or
BLUP, relative to the population mean. In other words, genetic superiority
can be predicted from superiority in phenotypic or selection index values,
and the degree of correspondence between the two superiorities is the
regression coefficient bxg. In the case of simple phenotypic selection, bxg is
equal to the narrow sense heritability, h2, which is the proportion of the phe-
notypic variation that can be transmitted to progeny and indicates the effec-
tiveness of selection for realizing genetic gain.
In the same way that genetic superiority of an individual can be
predicted, the average genetic superiority for a group of selected individuals
can be predicted using g^∗  g ¼ bxg ðx∗  xÞ where g^∗ is the average breeding
value of the selected individuals and x∗ is the average phenotypic value of
the selected individuals. Later it will be shown how the expression g^∗  g ¼
bxg ðx∗  xÞ is critical for predicting gain from selection.

4.2 Expected breeding value of progeny and their parents

When two individuals are mated, the average breeding value of their prog-
eny will be equal to the average of their own breeding values. Or in other
words, most progeny will be as good as their parents. This is expressed as:
g1 ¼ 12 gp1 + 12 gp2 , where g1 is the average breeding value of a progeny gener-
ated from a cross between parent one and two, gp1 is the true breeding value
of parent one and gp2 is the true breeding value of parent two. Assuming the
parents were not completely inbred, individual progenies will have breeding
A practical guide to genetic gain 231

values that deviate from the average in positive and negative directions
(Franklin, 1977; Hill, 1993) because progeny do not receive the same set
of alleles from each parent.

4.3 The Breeder’s equation

The breeder’s equation allows one to predict gain from selection and it can
be derived using the two basic principles just described. Assume that unbi-
ased estimates of breeding value are available, then estimated breeding values
of parents can be substituted for true breeding values in g1 ¼ 12 gp1 + 12 gp2
giving g1 ¼ 12 g^p1 + 12 g^p2 . Thus, its possible to predict the population mean
breeding value of a population of progeny simply based on estimated breed-
ing values of the parents. In other words, g^∗, which was defined earlier as the
average estimated breeding value for a group of selected individuals, would
be equal to the average breeding value of the progeny of these selected indi-
viduals, denoted as g1 , if they were mated. Substituting g for g^∗ in the pre-
diction equation mentioned above we can arrive at g1  g ¼ bxg ðx∗  xÞ.
Notice that g1  g is simply the difference in the mean breeding value
that occurred due to 1 cycle of selection. This is also referred to as the
response, R, to 1 cycle of selection. Substituting R for g1  g, response to
1 cycle of selection can be predicted using R ¼ bxg ðx∗  xÞ. In the case of
phenotypic mass selection, recall that bxg ¼ h2, and the formula for one
response to 1 cycle of selection can be expressed as R ¼ h2 ðx∗  xÞ. The
expression x∗  x is commonly referred to as the selection differential, S.
Substituting S for ðx∗  xÞ we arrive at the R ¼ h2S, which is a commonly
used formula for the prediction of response to phenotypic mass selection.
When selection is imposed differently on males and females. Response is
predicted by R ¼ 12 Rm + 12 Rf . Where Rm is the expected response in males
and Rf is the expected response in females. Phenotypic mass selection after
flowering imposes no selection in males, thus the expected response in this
case is R ¼ 12 h2 S:

4.4 Equivalent versions of the Breeder’s equation

Other algebraic rearrangements of g^i  g ¼ bxg ðxi  xÞ and g1 ¼ 12 gp1 + 12 gp2
allow us to express the breeders equation in different ways (Table 2) which
are useful for predicting genetic gain per cycle in different scenarios. To
demonstrate, let us assume that selection is based on estimated breeding
values obtained using BLUP or a selection index. Also assume that the
232 Jessica E. Rutkoski

Table 2 Useful formulas for predicting genetic gain.

Equation Descriptiona
R ¼ h2S Expected gain from phenotypic mass selection when the selection
differential, S, is known
R ¼ kh2σ p Expected gain from phenotypic selection
R ¼ krxgσ g Expected gain per cycle based on multiple observations combined
using a selection index or BLUP
R ¼ khihjrgσ pi Expected gain per cycle in trait i based selection based phenotypic
values of trait j
R ¼ khjrgσ gi Expected gain per cycle in trait i based on multiple observations of
trait j combined using a selection index or BLUP j
R ¼ 12 Rm + 12 Rf Overall expected gain when selection is imposed differently on
males and females
Rt ¼ R/L Expected gain per unit time based on expected gain per cycle and
the breeding cycle duration L
h is the heritability, σ p is the phenotypic standard deviation, k is the selection intensity, rxg is correlation
a 2

between true and estimated breeding values referred to as the selection accuracy, σ g is the additive genetic
standard deviation, h is the square root of the heritability, rg is the additive genetic correlation between
traits i and j, Rm and Rf is the genetic gain per cycle in males and females, respectively.

selection, evaluation, and recombination units are the same individuals and
selection is imposed equally on both males and females.
Recall that the average breeding value for a group of selected individuals
can be predicted using g^∗ ¼ g + bxg ðx∗  xÞ, where g^∗ is the average breed-
ing value, g is the average breeding value of the population, bxg is the regres-
sion coefficient between breeding values and estimated breeding values, x∗
is the average estimated breeding value for the selected individuals, and x is
the average estimated breeding value of the population.
Also recall that x∗  x is also known as the selection differential, S. Thus
the prediction equation can be written as g^∗ ¼ g + bxg S. The selection differ-
ential, S, can be expressed in units of standard deviation, referred to as k, using
k ¼ S/σ x where σ x is the additive genetic standard deviation of the estimated
breeding values. Through rearrangement, S ¼ kσ x. Substituting kσ x for S in
g^∗ ¼ g + bxg S we arrive at g^∗ ¼ g + bxg kσ x . Subtraction of g from both sides
gives g^∗  g ¼ bxg kσ x . Recall that g^∗ is equal to g1 , therefore g^∗  g becomes
g1  g which is simply the change in population means due to selection, R.
Through substitution of R for g^∗  g in g^∗  g ¼ bxg kσ x , we arrive at
R ¼ bxgkσ x.
A practical guide to genetic gain 233

It is more useful to express the breeder’s equation in terms of the selec-

tion accuracy, rxg, which is the correlation between the selection indices or
BLUPs and the breeding values. To do so, we start with the definitions of
regression and correlation coefficients in terms of variance and covariance:
rxg ¼ σ xg/σ gσ x and bxg ¼ σ xg/σ 2x. The parameter σ 2x can be expanded to σ xσ x,
thus bxg ¼ σ xg/σ xσ x. Then based on the definition of rxg, we can substitute
σ xg/σ x with rxgσ g giving bxg ¼ rxgσ g/σ x. Now, rxgσ g/σ x can be substitued
for bxg in R ¼ bxgkσ x giving R ¼ kσ xrxgσ g/σ x. Then σ x cancels out giving:
R ¼ rxgσ gk. If the parameters in R ¼ rxgσ gk can be assumed to remain constant
over cycles of selection then R is the expected response, also called expected
genetic gain from selection per cycle. The expected gain per unit time, Rt,
referred to as the rate of genetic gain, is simply R/L, where L is the length of
time required to complete one breeding cycle.
Note that several of these equations refer to the selection intensity k
which is the selection differential is expressed in units of standard deviation.
This is useful because assuming the trait phenotypes are normally distributed,
k can be estimated based on percentage of individuals selected as parents
from the population of selection units (referred to as percent selected)
using properties of the normal distribution. The relationship between the
percent selected and the selection intensity is shown in Fig. 4. In breeding
programs, the percent selected, and consequently the selection intensity, are
often increased by increasing the number of selection units that are
evaluated.
The breeder’s equations just described can only accurately predict
genetic gain in quantitative traits per cycle or per unit time assuming that
the selection units, evaluation units, and recombination units are the same
breeding materials, selection is done in one stage, generations do not over-
lap, and the breeding population is reasonably large (Burrows, 1972). Accu-
rate prediction of genetic gain when these conditions are violated, which is
often the case in breeding programs, requires additional formulas. However,
the breeder’s equations are still useful for understanding the factors that affect
genetic gain and their relative importance. For example, it is clear that the
accuracy of selection is an important factor for achieving genetic gain. As
shown in Fig. 5, there is a direct linear increase between the selection accu-
racy and the rate of genetic gain per cycle for all levels of percent selected.
Decreasing levels of the percent selected are also associated with greater rates
of genetic gain per cycle, as shown in Fig. 6, but the relationship is not linear
and it is weaker when the selection accuracy is low. Another critical variable
234 Jessica E. Rutkoski

Fig. 4 Selection intensity vs. percent selected for percent values ranging from 1 to 50.

Fig. 5 The rate of genetic gain per cycle with increasing levels of selection accuracy.
A clear linear relationship between the selection accuracy and the rate of genetic gain
in units of genetic standard deviation is demonstrated.

is the duration of the breeding cycle which has an indirect relationship with
the rate of genetic gain per unit time. For example, reducing the breeding
cycle from 6 to 3 years would double the rate of genetic gain per unit time,
assuming the accuracy, additive genetic variance, and percent selected,
A practical guide to genetic gain 235

Fig. 6 The rate of genetic gain per cycle with decreasing levels of percent selected.
Lower values of percent selected are associated with greater rates of genetic gain
per cycle, especially when the selection accuracy is moderate to high.

remain unchanged. Of all the variables in the breeder’s equation, the length
of the breeding cycle duration has probably the greatest potential to be
manipulated to increase rates of genetic gain in plant breeding programs.

4.5 Stability of genetic variance

Predictions of per-cycle gain from selection also assume that parameters in
the breeder’s equation remain constant over cycles. Genetic variance
remains constant over cycles assuming that the trait under selection con-
forms follows the infinitesimal model of inheritance (Crow and Kimura,
1970), which states that there are an infinite number of loci, each with a
small, additive effect on the trait (Fisher, 1918). Selection does in fact affect
the genetic variance, a phenomenon referred to as the Bulmer effect
(Bulmer, 1971). Immediately after selection and prior to recombination,
genetic variance is reduced depending on the selection accuracy and the per-
cent of the population selected. Fig. 7 shows the proportion of the additive
genetic standard deviation remaining after selection for different values of
selection accuracy and percent selected. However, much of the genetic var-
iance is restored during recombination. After a few cycles of selection a bal-
ance between loss of variance induced by selection and generation of
variance due to recombination is obtained (Bulmer, 1971; Gomez-Raya
and Burnside, 1990), thus it is reasonable to assume that the genetic variance
236 Jessica E. Rutkoski

Fig. 7 The proportion of the additive genetic standard deviation remaining immedi-
ately after truncation selection and prior to recombination for different values of selec-
tion accuracy and percent selected. As selection accuracy increases the additive genetic
standard deviation remaining after selection decreases.

will remain constant for a reasonable length of time if the trait under
selection is quantitative and the population has already been under selection
for a few cycles.

5. Realized genetic gain

5.1 Examples of genetic gain realized
Hundreds of selection experiments have been conducted since the early
1900s. More than 60 selection experiments conducted using animal species
have been summarized in a 1988 review by Sheridan (1988), and more than
50 selection experiments conducted in self and cross-pollinated crop species
have been reviewed by Hallauer and Darrah (1985). Outcomes of selection
experiments have confirmed that the rate of genetic gain per cycle depends
on the selection accuracy, selection intensity, and additive genetic variance
as expected based on theory. Random fluctuations in selection responses are
also observed, confirming theoretical expectations of variability in selection
response due to genetic drift (Wright, 1930).
In addition to confirming theoretical expectations, selection experiments
have also demonstrated the impressive ability of populations to respond to
A practical guide to genetic gain 237

selection over the long term. For example, in the Illinois Long-Term Selec-
tion experiment in maize (Zea mays L.), recurrent selection for high percent
oil and high percent protein in maize kernels has produced consistent
changes in phenotype for more than 100 cycles of selection. For example,
at the start of the Illinois Long-Term Selection experiment in 1896 average
percent oil was 4.69, and by 2004 the population being selected for higher
percent oil reached an average of 22.1% oil (University of Illinois Urbana-
Champaign Department of Crop Science, 2007a, 2007b). In addition to the
changes in means, the ranges of the populations observed in 1896 and 2004
are also drastically different (Fig. 8). This demonstrates the effectiveness of
selection within a single population for bringing about major changes in
the phenotypic means and ranges.
The long-term genetic gain from selection observed in the Illinois Long-
Term Selection experiment is consistent with a hypothesis that there are
many loci affecting oil and protein content, and alleles that have a positive
effect on these traits were relatively rare in the initial population (Dudley,
2007). Similar observations have been observed in animal species. For

Fig. 8 Realized genetic gain from selection for increased percent oil of kernels in maize
in the Illinois Long-Term Selection experiment.
238 Jessica E. Rutkoski

example, in the Virginia Chicken lines selection experiment, continued

response to selection for body weight was observed for 38 cycles, and
response to selection in this experiment was found to be due the effect of
many small effect loci ( Jacobsson et al., 2005).
Genetic gain has not only been observed in controlled experiments. Sev-
eral applied animal and crop breeding programs have also documented gain
from selection across many years of selective breeding (Burnside and Legates,
1967; Chen et al., 2003; Cloete et al., 2004; Duvick, 2005; Laidig et al.,
2014; Piepho et al., 2014). In summary, nearly 100 years of experimentation
and breeding has confirmed that if a trait is heritable it will respond to selec-
tion and it will continue to respond to selection for many cycles of selection
if conferred by many loci.

5.2 Estimating realized genetic gain

The success of a plant breeding program has traditionally been assessed based
on the adoption of the varieties it has released (Brennan and Byerlee, 1991)
because its ultimate goal is to develop varieties grown by farmers; however,
variety adoption is also heavily affected by the efficiency of the seed system
and can only be assessed years after variety release. The genetic trend, or real-
ized rate of genetic gain from a breeding program could be a useful indicator
of a breeding program’s success independent of the seed system. On the
other hand, realized rates of genetic gain are affected by factors that are
not under one’s control including the genetic nature of the target traits,
and genetic drift in addition to those under the breeder or breeding organi-
zation’s control including the design of the breeding program and its oper-
ational efficiency. Estimates of realized genetic gain can also only be assessed
after many years of breeding. Thus, although the rate of genetic gain realized
may be a good indicator of a breeding program’s success, it may be a poor
indicator of the performance of the breeder or breeding organization.
Measuring the rate of genetic gain realized by a breeding program for
individual traits or for net merit is conceptually straightforward. All that is
required are estimates of the mean breeding value for the trait(s) of interest
for the breeding population per year or per breeding cycle. The estimate of
realized gain per year or per cycle, is simply the slope of the regression line of
mean breeding value on year or cycle number (Eberhart, 1964;
Garrick, 2010).
In practice, accurately estimating the rate of genetic gain realized by a
breeding program can be problematic because non-genetic trends due to
A practical guide to genetic gain 239

changes in agronomic practices or climate change are confounded with

genetic trends. For example, increases in minimum night temperature are
known to negatively impact yield in both rice (Oryza sativa) and wheat
(Triticum aestivum L.). For these crops, change in average yield observed over
time may be the result of a reduction in average yield due to climate change
and an increase in average yield due to breeding.
To overcome the problem of confounding, breeding materials from dif-
ferent cycles or years can be phenotypically evaluated in a common set of
environments. Then the trend in the average phenotypic value over breed-
ing cycles or years must be due change in average breeding value over time,
assuming the breeding materials were developed through the population
improvement process. Ideally, the phenotypic evaluation is done using
recently multiplied seed, to avoid possible confounding of genetic trend
with seed age, and using large samples of breeding materials from each year
or cycle of breeding to minimize sampling error (Hallauer et al., 2010). This
is often done when estimating realized genetic gain from selection experi-
ments. Unfortunately, most applied breeding programs keep only the most
elite breeding materials for a short period of time.
In a series of experiments conducted between 1977 and 2004 to assess
genetic contributions to improvement in U.S. corn yields, Duvick (2005)
evaluated varieties released over many years in the same environments
and regressed variety mean yield on the year of release. This method,
referred to as the “ERA trial” method, has become popular for estimating
rates of genetic gain realized by plant breeding programs because seed of
varieties released over time is more readily available than large samples of
breeding materials representing multiple cycles or years of breeding. Many
examples of this approach are available (Brancourt-Hulmel et al., 2003;
Donmez et al., 2001; Peng et al., 2000). However, the ERA trial method
does not necessarily provide an accurate estimate of the rate of genetic gain
because released varieties are not representative of the breeding population.
Historical breeding trial data could be useful for estimating realized
genetic gain for the trait(s) of interest, like what is done in livestock
(Garrick, 2010). In fact, some studies have used crop variety trial data to esti-
mate realized gain (Laidig et al., 2014; Mackay et al., 2011; Piepho et al.,
2014). However, historical plant breeding trial datasets often do not meet
the requirements necessary to obtain a reliable estimates of mean genetic
value per year or cycle. The most frequent problem is confounding of
genetic and year effects due to a lack of genetic connectivity between years
caused by testing only a small proportion of breeding materials for more than
240 Jessica E. Rutkoski

1 or 2 years. Estimates of mean breeding value per year estimated using ped-
igree or genomic relationship may be less confounded with year because
genetic connectivity between years can be increased through the pedigree
or genomic relationship. If a control population or multiple check varieties
were grown in all years, then data on the control population can be used to
help separate the genetic and non-genetic effects over time. However, it is
relatively rare to find an applied breeding program that has evaluated a con-
trol population or the same set of checks for several consecutive years.

6. How to improve rates of genetic gain

Although theory indicates that realizing genetic gain is possible as long
as a population improvement strategy is followed and the traits of interest are
heritable, there are also a series of components that should be in place in a
breeding program to ensure that genetic gain can be realized. Once these
components are well established, a breeding program can then incorporate
additional components to help it achieve a higher rate of genetic gain. Fol-
lowing the example of Maslow’s hierarchy of needs (Maslow, 1943), what is
needed in a breeding program to realize genetic gain and then improve it
further can be described as having a hierarchical structure (Fig. 9) with dif-
ferent components building on each other to achieve a more effective breed-
ing program.

6.1 Foundational components

If a breeding program follows a generate-evaluate-select-recombine process
Fig. 1, it can expect to realize genetic gain as long as a set of essential com-
ponents are in place. Three of these essential components that are the foun-
dation for all others are (1) clear objectives, (2) meticulous data management,
and (3) effective operations. Once these are in place, it is possible to achieve
the next level of required components, (4) effective phenotyping and (5)
selection based on BLUP.
Having clear objectives enables the direction of the breeding program to
set towards a specific goal. A breeding program without clear objectives is
like a ship traveling without a compass. No matter how fast or efficiently the
ship moves, it will not reach any particular destination. When setting clear
objectives, the target traits and their relative importance should be well
defined. Defining the set of environments experienced by the farmers that
the breeding program aims to serve, referred to as the target population of
A practical guide to genetic gain 241

Fig. 9 A Breeding program’s hierarchy of needs. Each section of the pyramid represents
a different component that is important for realizing genetic gain. The first two levels,
shaded in a darker color, consists of the foundational components that must be in place
to ensure that genetic gain can be realized. The upper part of the pyramid consists of
advanced components that can help increase the rate of genetic gain realized by the
breeding program.

environments (TPE) (Cooper et al., 1993; Cooper and DeLacy, 1994), is

also required. Because genetic gain is realized in small increments over time,
it is important that the objectives of the breeding program not change fre-
quently, otherwise a meaningful level of improvement in the target traits in
the TPE will not be achieved. Determining the economic weights for each
target trait helps to remove ambiguity in the breeding objectives and reduces
human error in the process of selection towards these objectives.
Meticulous data management is needed to ensure that the breeding pro-
gram’s phenotypic, genotypic, and pedigree data can be used for selection
and to avoid losses in selection accuracy due to errors in data management.
Without this component, resources used to generate and utilize data will be
wasted. For example, if phenotypic data is kept in simple electronic spread-
sheets, a small mistake in sorting data or combining different sources of
information could lead to zero selection accuracy. As another example, if
242 Jessica E. Rutkoski

spreadsheets containing plot yield data do not also include harvested area of
the plots, the yield data may no longer be usable. Proper use of a breeding
data management system can help in avoiding data such errors. Several such
data management systems are now available.
Effective operations means that routine activities such as planting,
harvesting, crossing, and managing germplasm are done correctly so that
the identity of the breeding materials is preserved during the breeding process.
If a breeding program cannot execute routine activities correctly, it will lose
control over which breeding materials ultimately get selected and recom-
bined. Mistakes during the execution of routine activities can be minimized
by ensuring that population sizes are manageable given the resources available,
by ensuring staff are adequately trained and managed, and by consistently fol-
lowing a set of well thought out standard operating procedures. Detecting
mistakes when they occur is also important so that they can be mitigated
and prevented in the future. Morphological checks, visually distinct plants
or families that can be easily identified, can help detect errors in planting
and harvesting. If these checks are not planted in the correct plots or if they
are mislabeled after harvest, this can indicate that a mistake occurred during
the field operations and the data and germplasm resulting from these opera-
tions may be unusable. Genotyping can also be useful for detecting seed mix-
tures or failed crosses so that they may be discarded (Chen et al., 2016).
Effective phenotypic evaluations are needed to ensure that the breeding
program will realize genetic gain in the target traits according to the objec-
tives of the breeding program. For phenotypic evaluations to be effective
they should be done in a way so that the phenotypic data collected is pre-
dictive of that which would be observed in farmers’ fields. This requires
proper phenotyping methods, and appropriate statistical design which
includes conducting trials in multiple environments that adequately sample
the TPE. For example, if the target trait is grain yield and farmers grow the
crop in their fields as a uniform stand (as opposed to a mixture), then grain
yield should be measured in field conditions, using field sites that that rep-
resent the farmer’s environments, and using homogenous breeding materials
grown in reasonably large plots (Rebetzke et al., 2014). Research to assess
patterns of genotype by environment interaction (GxE) and different
phenotyping methods can help in determining what field trial sites to use
and how to conduct phenotypic evaluations for improving the traits of inter-
est in the TPE. In many crops, research studies on phenotyping methods for
various traits have already been conducted and can be readily adopted.
A practical guide to genetic gain 243

While it is possible to use single phenotypic observations for selection,

selection based on BLUP is critical for improving low heritability traits
and traits that are affected by GxE such as yield. Assuming that the statistical
design used for phenotypic evaluation is adequate, phenotypic data across
multiple years and environments can be combined using BLUP to improve
selection accuracy. Selection accuracy can also be improved by using ped-
igree and/or marker information in BLUP. If the breeding program targets
multiple traits, selection based on multi-trait BLUPs combined using eco-
nomic weights is optimal (Hazel, 1943). Finally, it is critical that BLUPs
or an index of multi-trait BLUPs be the criteria for selection such that
the average BLUP or index of the selected breeding materials is superior
to that of the population average.

6.2 Advanced components

Once a population improvement-based breeding program has all the
required components in place to ensure that it can realize genetic gain, it
can begin to conduct research aimed at improving the breeding process.
For example, empirical and/or simulation experiments can be used to eval-
uate techniques for reducing the time needed to generate breeding materials,
such as off-season nurseries, greenhouses or doubled haploids; and to eval-
uate different ways of leveraging techniques such as genomic selection and
high-throughput phenotyping in the breeding cycle. Before making
changes to the breeding process, this kind of “breeding methods research”
is critical to ensure that the proposed changes will actually improve the
breeding program’s efficiency or rate of realized genetic gain. Examples
of recent breeding methods research studies include Heffner et al. (2010),
Endelman et al. (2014), Lorenz (2013), Rutkoski et al. (2015), Massman
et al. (2013), and Longin et al. (2015).
Simulation is useful tool which allows testing different hypotheses about
breeding methods quickly and at low cost. Simulations many be determin-
istic, relying on formulas for predicting genetic gain such as the breeders
equation, or they may be stochastic, relying computers to model each pro-
cess in the breeding cycle. Stochastic simulations are useful when the breed-
ing scenarios to be tested are complex. Several stochastic simulation
platforms useful for plant breeding are now available such as Breeding
Scheme Language (Yabe et al., 2017), AphaSim (Faux et al., 2016) and
QuGene (Podlich and Cooper, 1998).
244 Jessica E. Rutkoski

7. Conclusion
The objective of this chapter was to increase awareness and knowledge
about genetic gain so that breeding programs can begin to take actions that
will enable them to be more successful. Genetic gain was defined as the
improvement in average genetic value in a population or the improvement
in average phenotypic value due to selection within a population over cycles
of breeding assuming that the effect of environment remains constant. The
importance achieving genetic gain over cycles of selection was demonstrated
using a simple simulation, where achieving genetic gain over multiple cycles
of selection was shown to dramatically increase the probability of identifying
promising new varieties on an ongoing basis compared to repeated selecting
from a population that is not improving such as a germplasm collection.
Next, key selection techniques such as combining multiple sources of infor-
mation for selection using selection indices and BLUP and direct and indi-
rect selection were explained and discussed to help point out when these
techniques are useful. Then, to show that genetic gain is readily achievable,
quantitative genetic principles were used to explain how genetic gain can be
predicted from one generation to the next. This was followed by a discussion
of realized genetic gain including impressive examples from selection exper-
iments, and a review of methods used to estimate realized rates of genetic
gain. Lastly, a hierarchy of different components needed in a breeding pro-
gram was provided, first to ensure that the breeding program can realize
genetic gain and second to enable the breeding program to realize genetic
gain at a faster rate.

Acknowledgment
This work was supported by the Bill and Melinda Gates Foundation project: Transforming
Rice Breeding and Excellence in Breeding Platform (EiB) of the Consultative Group on
International Agricultural Research (CGIAR).

Appendix. R-function used for simulation

The argument nsim indicates the number of simulations, ngen is the
number of times identification of superior varieties is attempted, n is the
number of selection units, h2 is the heritability, Vg is the additive genetic
variance, and nsel is the number of parents selected breeding cycle.
A practical guide to genetic gain 245

SimuSel<- function(nsim=10, ngen=3, n=100, h2=0.5, Vg=1, nsel=30){

r<- sqrt(h2)
p<- nsel/n
i<- dnorm(qnorm(p))/p
Ve<- Vg/h2- Vg
stdG<- sqrt(Vg)
stdE<- sqrt(Ve)
succall<-c()
for(j in 1:nsim){
mn<- 0
g<- rnorm(n,sd=stdG)
e<- rnorm(n,sd=stdE)
p<- g+e
val0<- g[which.max(p)]
succ<-c()
for(i in 1:c(ngen-1)){
mn<- mn+c(stdG*i*r)
g<- rnorm(n,mean=mn, sd=stdG)
e<- rnorm(n,sd=stdE)
p<- g+e
val<- g[which.max(p)]
suc<- val>val0
if(suc){
val0<- val
}
succ<- append(succ, suc)
}
succall<- append(succall, !FALSE %in% succ)
}
probsuccess<- sum(succall)/nsim
return(probsuccess)
}

References
Allaire, F.R., 1980. Mate selection by selection index theory. Theor. Appl. Genet. 57 (6),
267–272. https://doi.org/10.1007/BF00264953.
Bernardo, R., 2008. Molecular markers and selection for complex traits in plants: learning
from the last 20 years. Crop Sci. 48 (5), 1649–1664.
Bowman, M.L., Crossley, B.W., 1908. Corn. Kenyon, Des Moines, IA.
246 Jessica E. Rutkoski

Brancourt-Hulmel, M., Doussinault, G., Lecomte, C., Berard, P., Le Buanec, B.,
Trottet, M., 2003. Genetic improvement of agronomic traits of winter wheat cultivars
released in France from 1946 to 1992. Crop Sci. 43, 37–45. Crop Science Society of
America, Madison, WI, https://doi.org/10.2135/cropsci2003.3700.
Brennan, J.P., Byerlee, D., 1991. The rate of crop varietal replacement on farms: measures
and empirical results for wheat. Plant Var. Seeds 4, 99–106.
Bulmer, M.G., 1971. The effect of selection on genetic variability. Am. Nat. 105 (943),
201–211. The University of Chicago Press for the American Society of Naturalists,
https://doi.org/10.2307/2459550.
Burnside, E.B., Legates, J.E., 1967. Estimation of genetic trends in dairy cattle. J. Dairy Sci.
50 (9), 1448–1457.
Burrows, P.M., 1972. Expected election differentials for directional selection. Biometrics 28
(4), 1091–1100. Wiley, International Biometric Society, https://doi.org/10.2307/2528642.
Chen, P., Baas, T.J., Mabry, J.W., Koehler, K.J., Dekkers, J.C., 2003. Genetic parameters
and trends for litter traits in US Yorkshire, Duroc, Hampshire, and Landrace pigs.
J. Anim. Sci. 81 (1), 46–53.
Chen, J., Zavala, C., Ortega, N., Petroli, C., Franco, J., Burgueño, J., Costich, D.E.,
Hearne, S.J., 2016. The development of quality control genotyping approaches: a case
study using elite maize lines. PLoS One 11 (6), e0157236. Public Library of Science,
https://doi.org/10.1371/journal.pone.0157236.
Cloete, S.W.P., Gilmour, A.R., Olivier, J.J., Van Wyk, J.B., 2004. Genetic and phenotypic
trends and parameters in reproduction, greasy fleece weight and liveweight in merino
lines divergently selected for multiple rearing ability. Aust. J. Exp. Agric. 44 (8),
745–754.
Collard, B.C.Y., Beredo, J.C., Lenaerts, B., Mendoza, R., Santelices, R., Lopena, V.,
Verdeprado, H., et al., 2017. Revisiting rice breeding methods—evaluating the use of
rapid generation advance (RGA) for routine rice breeding. Plant Prot. Sci.
1008, 1–16. Taylor & Francis, https://doi.org/10.1080/1343943X.2017.1391705.
Cooper, M., DeLacy, I.H., 1994. Relationships among analytical methods used to study
genotypic variation and genotype-by-environment interaction in plant breeding
multi-environment experiments. Theor. Appl. Genet. 88 (5), 561–572. https://doi.
org/10.1007/BF01240919.
Cooper, M., Byth, D.E., DeLacy, I.H., Woodruff, D.R., 1993. Predicting grain yield in
Australian environments using data from CIMMYT international wheat performance
trials. 1. Potential for exploiting correlated response to selection. Field Crop Res.
32 (3–4), 305–322. https://doi.org/10.1016/0378-4290(93)90039-P.
Crow, J.F., Kimura, M., 1970. An Introduction to Population Genetics Theory. Harper &
Row, New York, NY.
Cullis, B.R., Smith, A.B., Coombes, N.E., 2006. On the design of early generation cultivar
trials with correlated data. J. Agric. Biol. Environ. Stat. 11, 381–393.
Donmez, E., Sears, R.G., Shroyer, J.P., Paulsen, G.M., 2001. Genetic gain in yield attributes
of winter wheat in the great plains. Crop Sci. 41, 1412–1419. Crop Science Society of
America, Madison, WI, https://doi.org/10.2135/cropsci2001.4151412x.
Dudley, J.W., 2007. From means to QTL: the Illinois long-term selection experiment as a
case study in quantitative genetics. Crop Sci. 47, S20–S31. Suppl. https://doi.org/
10.2135/cropsci2007.04.0003IPBS.
Duvick, D.N., 2005. The contribution of breeding to yield advancements in maize (Zea
mays L.). Adv. Agron. 86, 83–145.
Eberhart, S.A., 1964. Least squares method for comparing progress among recurrent selection
methods. Crop Sci. 4, 230–231.
Endelman, J.B., Atlin, G.N., Beyene, Y., Semagn, K., Zhang, X., Sorrells, M.E.,
Jannink, J.L., 2014. Optimal design of preliminary yield trials with genome-wide
markers. Crop Sci. 54 (1), 48–59. https://doi.org/10.2135/cropsci2013.03.0154.
A practical guide to genetic gain 247

Faux, A.M., Gorjanc, G., Gaynor, R.C., Battagin, M., Edwards, S.M., Wilson, D.L.,
Hearne, S.J., Gonen, S., Hickey, J.M., 2016. AlphaSim: software for breeding program
simulation. Plant Genome 9 (3).
Fisher, R.A., 1918. The correlation between relatives on the supposition of Mendelian
inheritance. Philos. Trans. R. Soc. Edinb. 52, 339–433.
Fisher, R.A., 1930. The fundamental theorem of natural selection. In: The Genetical Theory
of Natural Selection. The Clarendon Press, Oxford, pp. 22–47.
Franklin, I.R., 1977. The distribution of the proportion of the genome which is homozygous
by descent in inbred individuals. Theor. Popul. Biol. 11 (1), 60–80. https://doi.org/10.
1016/0040-5809(77)90007-7.
Gallais, A., Bordes, J., 2007. The use of doubled haploids in recurrent selection and hybrid
development in maize. Crop Sci. 47, S190–S201. Crop Science Society of America,
Madison, WI, https://doi.org/10.2135/cropsci2007.04.0019IPBS.
Garrick, D.J., 2010. An animal breeding approach to the estimation of genetic and environ-
mental trends from field populations. J. Anim. Sci. 88 (13 Suppl), E3–E10. https://doi.
org/10.2527/jas.2009-2329.
Gianola, D., 2013. Priors in whole-genome regression: the Bayesian alphabet returns.
Genetics 194 (3), 573–596. https://doi.org/10.1534/genetics.113.151753.
Gomez-Raya, L., Burnside, E.B., 1990. The effect of repeated cycles of selection on genetic
variance, heritability, and response. Theor. Appl. Genet. 79 (4), 568–574. https://doi.
org/10.1007/BF00226169.
Goulas, C.K., Lonnquist, J.H., 1976. Combined half-sib and S1 family selection in a
maize composite population. Crop Sci. 16, 461–464. Crop Science Society of America,
Madison, WI, https://doi.org/10.2135/cropsci1976.0011183X001600040005x.
Haghighattalab, A., Perez, L.G., Mondal, S., Singh, D., Schinstock, D., Rutkoski, J., Ortiz-
Monasterio, I., Singh, R.P., Goodin, D., Poland, J., 2016. Application of unmanned
aerial systems for high throughput phenotyping of large wheat breeding nurseries. Plant
Methods 12 (1), 35. BioMed Central.
Hallauer, A.R., Darrah, L.L., 1985. Critical reviews in plant sciences compendium of recur-
rent selection methods and their application. Crit. Rev. Plant Sci. 3 (1), 1–33.
Hallauer, A.R., Carena, M.J., Miranda Filho, J.B., 2010. Quantitative Genetics in Maize
Breeding, third ed. Springer-Verlag, New York, LLC.
Hayes, B.J., Visscher, P.M., Goddard, M.E., 2009. Increased accuracy of selection by using
the realized relationship matrix. Genet. Res. 91 (1), 47–60. https://doi.org/10.1017/
S0016672308009981.
Hazel, L.N., 1943. The genetic basis for constructing selection indices. Genetics 28 (6),
476–490. http://www.genetics.org/content/28/6/476.abstract.
Hazel, L.N., Lush, J.L., 1942. The efficiency of three methods of selection. J. Hered.
33, 393–399.
Heffner, E.L., Aaron, J., Lorenz, J.-L.J., Sorrells, M.E., 2010. Plant breeding with genomic
selection: gain per unit time and cost. Crop Sci. 50 (5), 1681–1690. Crop Science Society
of America: http://cat.inist.fr/?aModele ¼ afficheN{\&}cpsidt ¼ 23164725.
Henderson, C.R., 1963. Selection index and expected genetic advance. In: Statistical Genet-
ics and Plant Breeding 982, pp. 141–163. Washington, DC.
Henderson, C.R., 1973. Sire evaluation and genetic trends. In: Animal Breeding and Genet-
ics Symposium in Honor of Dr. Jay L. Lush. American Society of Animal Science;
American Dairy Science Association, Champaign, IL, pp. 10–41.
Henderson, C.R., 1975. Best linear unbiased estimation and prediction under a selection
model. Biometrics 31 (2), 423–447. Wiley, International Biometric Society, https://
doi.org/10.2307/2529430.
Hill, W.G., 1993. Variation in genetic identity within kinships. Heredity 71, 652. The
Genetical Society of Great Britain, https://doi.org/10.1038/hdy.1993.190. http://10.
0.4.14/hdy.1993.190.
248 Jessica E. Rutkoski

Jacobsson, L., Park, H.-B.B., Wahlberg, P., Fredriksson, R., Perez-Enciso, M., Siegel, P.B.,
Andersson, L., 2005. Many QTLs with minor additive effects are associated with a large
difference in growth between two selection lines in chickens. Genet. Res. 86 (2),
115–125. https://doi.org/10.1017/S0016672305007767.
Laidig, F., Piepho, H.-P., Drobek, T., Meyer, U., 2014. Genetic and non-genetic long-term
trends of 12 different crops in German official variety performance trials and on-farm
yield trends. Theor. Appl. Genet. 127, 2599–2617. https://doi.org/10.1007/s00122-
014-2402-z.
Lerner, M.I., Cruden, D.M., 1947. The heritability of accumulative monthly and annual egg
production. Poult. Sci. 27, 67–78. no. 1918.
Longin, C.F.H., Mi, X., W€ urschum, T., 2015. Genomic selection in wheat: optimum
allocation of test resources and comparison of breeding strategies for line and hybrid
breeding. Theor. Appl. Genet. 128 (7), 1297–1306. https://doi.org/10.1007/s00122-
015-2505-1.
Lorenz, A.J., 2013. Resource allocation for maximizing prediction accuracy and genetic gain
of genomic selection in plant breeding: a simulation experiment. G3 (Bethesda) 3 (3),
481–491. https://doi.org/10.1534/g3.112.004911.
Lush, J.L., 1937. Animal Breeding Plans, third ed. Iowa State College Press, Ames, IA.
Lush, J.L., 1947. Family merit and individual merit as bases for selection. Part I. Am. Nat.
81 (799), 241–261. University of Chicago Press, American Society of Naturalists,
http://www.jstor.org/stable/2457881.
Mackay, I., Horwell, A., Garner, J., White, J., McKee, J., Philpott, H., 2011. Reanalyses of
the historical series of UK variety trials to quantify the contributions of genetic and envi-
ronmental factors to trends and variability in yield over time. Theor. Appl. Genet.
122 (1), 225–238. https://doi.org/10.1007/s00122-010-1438-y.
Maslow, A.H., 1943. A theory of human motivation. Psychol. Rev. 50 (4), 370, American
Psychological Association.
Massman, J.M., Jung, H.J.G., Bernardo, R., 2013. Genomewide selection versus marker-
assisted recurrent selection to improve grain yield and Stover-quality traits for cellu-
losic ethanol in maize. Crop Sci. 53 (1), 58–66. https://doi.org/10.2135/cropsci2012.
02.0112.
McGinnis, R.C., Shebeski, L.H., 1968. The reliability of single plant selection for yield in F2.
In: Proceedings of the Third International Wheat Genetics Symposium, Canberra, Aus-
tralia, pp. 109–114.
Mendel, G., 1866. Versuche u €ber Pflanzen-Hybriden. In: Verhandlungen Des
Naturforschenden Vereins Br€ unn. vol. 4, pp. 3–47.
Meuwissen, T.H.E., 1997. Maximizing the response of selection with a predefined rate of
inbreeding. J. Anim. Sci. 75 (4), 934–940. https://doi.org/10.2527/1997.754934x.
Meuwissen, T.H., Hayes, B.J., Goddard, M.E., 2001. Prediction of total genetic value using
genome-wide dense marker maps. Genetics 157 (4), 1819–1829. Genetics Society of
America.
Peng, S., Laza, R.C., Visperas, R.M., Sanico, A.L., Cassman, K.G., Khush, G.S., 2000.
Grain yield of rice cultivars and lines developed in the Philippines since 1966. Crop
Sci. 40, 307–314. Madison, WI. https://doi.org/10.2135/cropsci2000.402307x.
Piepho, H.P., M€ ohring, J., Melchinger, A.E., B€ uchse, A., 2008. BLUP for phenotypic selec-
tion in plant breeding and variety testing. Euphytica 161, 209–228. https://doi.org/10.
1007/s10681-007-9449-8.
Piepho, H.-P., Laidig, F., Drobek, T., Meyer, U., 2014. Dissecting genetic and non-genetic
sources of long-term yield trend in German official variety trials. Theor. Appl. Genet.
127, 1009–1018. https://doi.org/10.1007/s00122-014-2275-1.
Podlich, D.W., Cooper, M., 1998. QU-GENE: a simulation platform for quantitative anal-
ysis of genetic models. Bioinformatics 14 (7), 632–653.
A practical guide to genetic gain 249

Poland, J.A., Bradbury, P.J., Buckler, E.S., Nelson, R.J., 2011. Genome-wide nested
association mapping of quantitative resistance to northern leaf blight in maize. Proc.
Natl. Acad. Sci. 108, 6893–6898. http://www.pnas.org/content/early/2011/04/07/
1010894108.abstract.
R Development Core Team, 2016. R: A Language and Environment for Statistical Com-
puting. R Foundation for Statistical Computing, Vienna, Austria. https://www.
r-project.org/.
Rasmusson, D.C., Cannell, R.Q., 1970. Selection for grain yield and components of yield in
barley. Crop Sci. 10, 51–54. Madison, WI: Crop Science Society of America, https://doi.
org/10.2135/cropsci1970.0011183X001000010020x.
Rebetzke, G.J., Fischer, R.A., Van Herwaarden, A.F., Bonnett, D.G., Chenu, K.,
Rattey, A.R., Fettell, N.A., 2014. Plot size matters: interference from intergenotypic
competition in plant phenotyping studies. Funct. Plant Biol. 41 (2), 107–118. https://
doi.org/10.1071/FP13177.
Rutkoski, J., Singh, R.P., Huerta-Espino, J., Bhavani, S., Poland, J., Jannink, J.L.,
Sorrells, M.E., 2015. Genetic gain from phenotypic and genomic selection for quanti-
tative resistance to stem rust of wheat. Plant Genome 8 (2). https://doi.org/10.3835/
plantgenome2014.10.0074.
Sheridan, A.K., 1988. Agreement between estimated and realized genetic parameters. Anim.
Breed. Abstr. 56, 877–889.
Smith, H.F., 1936. A discriminant function for plant selection. Ann. Hum. Genet. 7 (3),
240–250.
University of Illinois Urbana-Champaign Department of Crop Science, 2007a. Data File:
Means by Year and Generation of the Oil and Protein Concentration Measured Each
Year (1896–2004). University of Illinois Urbana-Champaign. http://hdl.handle.net/
2142/3526.
University of Illinois Urbana-Champaign Department of Crop Science, 2007b. Data File:
Raw Data From Each Ear Analyzed Each Year of the Illinois Long-Term Selection
Experiment for Oil and Protein in Corn (1896–2004). University of Illinois Urbana-
Champaign. http://hdl.handle.net/2142/3528.
Watson, A., Ghosh, S., Williams, M.J., Cuddy, W.S., Simmonds, J., Rey, M.-D., Asyraf Md
Hatta, M., et al., 2018. Speed breeding is a powerful tool to accelerate crop research and
breeding. Nat. Plants 4 (1), 23–29. https://doi.org/10.1038/s41477-017-0083-8.
Weber, W.E., 1984. Selection in early generations. In: Lange, W., Zeven, A.C.,
Hogenboom, N.G. (Eds.), Selection in Early Generations. Proceedings of the 10th Con-
gress of Eucarpia. Pudoc, Wageningen, pp. 72–81.
Williams, E., Piepho, H.-P., Whitaker, D., 2010. Augmented p-rep designs. Biom. J. 53 (1),
19–27. Wiley-Blackwell, https://doi.org/10.1002/bimj.201000102.
Wright, S., 1920. The relative importance of heredity and environment in determining the
piebald pattern of Guinea-pigs. Proc. Natl. Acad. Sci. 6, 320–332.
Wright, S., 1921. Systems of mating. Genetics 6, 111–178.
Wright, S., 1930. Evolution in Mendelian populations. Genetics 16, 97–159. https://doi.org/
10.1007/BF02459575.
Yabe, S., Iwata, H., Jannink, J.-L., 2017. A simple package to script and simulate breeding
schemes: the breeding scheme language. Crop Sci. 57 (3), 1347–1354. https://doi.org/
10.2135/cropsci2016.06.0538.