Crop Modeling, QTL Mapping, and Their Complementary Role in Plant Breeding
Xinyou Yin,* Piet Stam, Martin J. Kropff, and Ad H. C. M. Schapendonk
ABSTRACT
massive amounts of information for plant breeding
(Stuber et al., 1999; Miflin, 2000), an option of improving
breeding efficiency is to develop and utilize a thorough
understanding of morphophysiological factors that determine yield (Bindraban, 1997). However, only in a
limited number of instances has plant physiology led to
crop improvement; rather, its role in breeding so far
has been to provide possible explanations for the improvements that have been achieved. Miflin (2000) suggested that this situation may change in the future if
the links between physiology and genetics are established.
This suggestion agrees with a main conclusion from an
extensive survey (Jackson et al., 1996) that there is a
general agreement among plant breeders and physiologists that physiological knowledge can be applied to
improve breeding efficiency in the future.
As early as in the 1960s, Donald (1968) proposed an
approach based much more explicitly on the design of
plants or ideotypes for target environment, using known
principles of physiology and agronomy. Rasmusson
(1987) suggested improvements to Donald’s approach,
considering correlations between traits, and designed a
barley (Hordeum vulgare L.) ideotype for the Midwest
USA with desired changes in culm, leaf, and head characteristics. Rasmusson (1991) reported that while some
characteristics appeared to afford little opportunities
for obtaining gains, others did show more promise. Based
on the concept of ideotype approach, International Rice
Research Institute (IRRI) launched a program in 1989
to develop a “new plant type” rice that combines multiple innovations (Peng et al., 1999). While so far these
new lines have not broken the yield barrier as hoped for,
progress is being made at IRRI with refined ideotype
designs (S. Peng, personal communication, 2002). A
convincing example of ideotype approach is the breeding for the superhybrid rice by Professor Yuan Longping’s group in China. Rather than count on heterosis
alone to raise yields, he also incorporated morphophysiological characters such as long, narrow and erect top
leaves and large panicles that hang close to ground, the
characteristics that physiologists have expected to enhance efficiency of crop light capture (Setter et al., 1995).
Field trials at four separate locations showed potential
yields of the superhybrid were 15 to 20% higher than
10.5 t ha⫺1 for existing hybrids (Normile, 1999).
Using crop physiology in ideotype breeding can be
more feasible than ever because of the development of
dynamic process-based crop growth simulation models
(crop models hereafter). These models quantify causal-
Crop modelers and geneticists have developed a vision of their roles
in plant breeding from their own perspective. However, to improve
breeding efficiency, interdisciplinary collaboration becomes increasingly important. The objective of this paper is to explore opportunities
for collaboration between modelers and geneticists in ideotype breeding for high crop yield. The advent of molecular markers enables variation of a complex trait to be dissected into the effects of quantitative
trait loci (QTL) and assists the transfer of these QTL into desired
cultivars or lines. A recent study in which QTL information was linked
to crop modeling has shown that QTL analysis removes part of random
errors of measured model input parameters and that QTL information
can successfully be coupled with crop models to replace measured
parameters. The QTL-based modeling overcomes the limitations in designing ideotypes by using models that ignore the inheritance of model
input traits. On the other hand, crop modeling can potentially be a
powerful tool to resolve genotype ⫻ environment interactions and to
dissect yield into characters that might be under simpler genetic control. Based on the complementary aspects of crop modeling and QTL
mapping, we propose an approach that integrates marker-assisted
selection into model-based ideotype framework to support breeding
for high crop yield. For this approach to be effective, there is a need
to develop crop models that are capable of predicting yield differences
among genotypes in a population under various environmental conditions.
B
reeding for high-yielding crop cultivars for specific environments is a major challenge to feed growing world populations. Through extensive selection, based
largely on empirical field observations, breeders have
been successful in creating high-yielding cultivars. In
many instances, progress has been attained from changes
in a relatively few genes, e.g., those involved in plant
height and photoperiodism (Miflin, 2000). However,
further improvement has been increasingly complex and
difficult (Bindraban, 1997), and in some crops such as
rice (Oryza sativa L.), no progress in increasing yield
potential has been achieved in the past decades (Peng
et al., 1999). Difficulties in manipulating yield are related to its genetic complexity: polygenic nature, interactions between genes (epistasis), and environmentdependent expression of genes (Ribaut and Hoisington,
1998). For more efficient crop improvement, joint interdisciplinary ventures to develop new knowledge and tools
are increasingly becoming important (Shorter et al., 1991).
Besides recent developments in genomics (such as
genome sequencing) that will provide useful tools and
X. Yin and M.J. Kropff, Crop and Weed Ecol. Group, Wageningen
Univ., P.O. Box 430, 6700 AK Wageningen, the Netherlands; P. Stam,
Lab. of Plant Breeding, Wageningen Univ., P.O. Box 386, 6700 AJ
Wageningen, the Netherlands; and A.H.C.M. Schapendonk, Plant
Dynamics, Englaan 8, 6703 EW Wageningen, the Netherlands. Joint
contrib. from Plant Res. Int. and the C.T. de Wit Graduate School for
Prod. Ecol. and Resour. Conserv. Received 1 May 2001. *Corresponding author ([email protected]).
Abbreviations: AFLP, amplification fragment length polymorphism;
G ⫻ E, genotype ⫻ environment interaction; h2, heritability; MAB,
marker-assisted breeding; QTL, quantitative trait locus or loci; QTL ⫻
E, quantitative trait loci ⫻ environment interaction; RFLP, restriction
fragment length polymorphism; RIL, recombinant inbred line; SLA,
specific leaf area.
Published in Agron. J. 95:90–98 (2003).
90
YIN ET AL.: THE COMPLEMENTARITY OF CROP MODELING AND QTL MAPPING
ity between relevant physiological processes and responses of these processes to environmental variables
and therefore allow yield predictions not restricted to
the environments where the model parameters are derived. As model parameters can represent certain genetic characteristics, crop modeling has been considered
a useful tool to assist breeding (Loomis et al., 1979; Whisler et al., 1986; Boote et al., 1996). Shorter et al. (1991)
proposed collaborative efforts among breeders, physiologists, and modelers, using models as a framework to
integrate physiology with breeding. For such, understanding the inheritance of the model parameters is required (Stam, 1998).
An important development during the last decade in
quantitative genetics was the ability to identify genome
regions responsible for variation of a trait due to the
advent of molecular markers (Paterson et al., 1988). The
term QTL has come to refer to polygenes underlying a
quantitative trait. Numerous studies have been reported
on identifying QTL for various traits in humans, animals, and plants. Similar to other quantitative traits,
individual input parameters of a crop model are amenable to QTL analysis (Yin et al., 1999b).
In this paper, we discuss potentials and limitations
of crop modeling and QTL analysis in assisting plant
breeding. The complementary aspects of crop modeling
and QTL analysis are explored to develop an integrated
approach for ideotype breeding.
CROP MODELING AS A TOOL
TO ASSIST PLANT BREEDING
Generally, crop models require two types of inputs:
environmental inputs (i.e., weather variables and management options) and physiological inputs. The latter
inputs are used as model parameters for characterizing
genotypic differences. These parameters are also referred
to as genetic coefficients (Hunt et al., 1993; White and
Hoogenboom, 1996; Mavromatis et al., 2001) or model
input traits (Yin et al., 2000a), reflecting the awareness
that model input parameters may be under genetic control.
Given the expectation that crop models based on
physiologically sound mechanisms have the potential to
quantify and integrate crop yield responses to genetic
and environmental factors, physiologists and modelers
have explored potential uses of crop models in various
aspects of breeding:
• To identify main yield-determining traits (Bindraban, 1997; Yin et al., 2000b)
• To define optimum selection environments (Aggarwal et al., 1997a)
• To optimize single-trait values (Boote and Tollenaar, 1994; Setter et al., 1995; Yin et al., 1997)
• To design ideotypes consisting of multiple traits
(Penning de Vries, 1991; Dingkuhn et al., 1993;
Kropff et al., 1995; Haverkort and Kooman, 1997)
• To assist multilocation testing (Dua et al., 1990)
and explain genotype ⫻ environment interactions
(G ⫻ E) (Mavromatis et al., 2001)
A common endpoint of these studies, based on model
91
simulations, is suggestions that breeders may use. Given
that direct experimental confirmation and objective
comparisons of modeled suggestions with those already
used in breeding programs are rare, Stam (1998), from
a geneticist’s and breeder’s point of view, expressed
his concerns about this model-based approach. First, a
practical problem of breeding is that the majority of
model input traits to be assessed are difficult to accurately measure. Second, the inheritance of the model
input traits is largely unknown. For example, in designing an ideotype by modeling, it is assumed, either tacitly
or explicitly, that these traits can be combined at will
in a single genotype. Such an assumption ignores the
possible existence of constraints and correlations among
the traits. Constraints might be imposed simply by the
fact that little genetic variation exists in the genetic
material available for selection. Thus, models may not
identify those traits for which gain via breeding may be
easiest (Jackson et al., 1996). Correlations between the
traits, due either to a tight linkage between QTL or to
a single QTL that affects multiple traits (pleiotropy),
may seriously hamper the realization of an ideotype.
After all, plant breeding is genetic improvement; knowledge of the genetic basis of phenotypic variation,
whether described in terms of conventional agronomic
traits or model input traits, is crucial for a successful
breeding program (Stam, 1998). To assist the development of efficient breeding strategies, crop modeling requires understanding of the inheritance of the factors
that determine crop growth (Shorter et al., 1991).
White and Hoogenboom (1996) presented a model
for bean (Phaseolus vulgaris L.) in which the genetic
control of model parameters was considered. They applied linear regression to estimate values of more than
20 model input traits from information about alleles
(variants at a gene locus) of seven known genes in the
cultivars studied. This approach, however, assumes that
all the traits were controlled by pleiotropic effects of the
seven genes, ignoring possible additional trait-specific
genes. Advances in quantitative genetics, by mapping
trait-specific QTL, can help to broaden insight in the
genetic basis of crop traits.
QTL MAPPING AND ITS APPLICATIONS
In genetics, the distance between genes on the genome is assessed on the basis of the frequency of recombination of the genes, estimated from scoring genotypes
of progeny of a cross (Kearsey and Pooni, 1996). Mapping quantitative traits is difficult because the genotype
is never unambiguously inferred from the phenotype.
Classical quantitative genetics pursues a different approach, using statistical concepts such as means, variances, correlations, heritabilities (h2), built on assumptions, e.g., that effects of individual genes on a trait are
small and additive. This assumption sheds little light,
if any, on the individual genes themselves (Prioul et
al., 1997).
To map quantitative traits, supplementary information from recognizable single-gene loci is required. First
attempts to link a quantitative trait to a major gene
92
AGRONOMY JOURNAL, VOL. 95, JANUARY–FEBRUARY 2003
Fig. 1. Seed weight of three F2 seed-color genotypes of a cross between
two lines of bean (data of Sax, 1923).
locus in plants date back to Sax (1923), who studied
seed weight and color in an F2 of a cross in bean. Seed
color involved the segregation of a single gene, P/p.
Seed weight differed among the three color genotypes
(Fig. 1), indicating that either the P/p locus had a pleiotropic effect on seed weight or there was a QTL for
weight closely linked to the P/p locus.
Major gene mutants, however, are scarce and may
not exist in a population under study. Because QTL may
occur throughout the genome, a large number of gene
markers are required to locate them. Early studies of
quantitative traits suffered from the lack of major-gene
markers that could make a complete map. This problem
was overcome with the realization that maps could be
constructed using pieces of chromosomal DNA as markers (Botstein et al., 1980). The first DNA-based molecular markers were fragments produced by restriction enzyme digestion, designated as restriction fragment length
polymorphism (RFLP). The RFLP markers are naturally occurring, abundant in most species, and simply
inherited Mendelian characters. Unlike major-gene mutant (or morphological) markers, those of RFLP are not
true genes because a gene codes for a gene product,
whereas an RFLP does not because it is possibly the
result of a single base change in a noncoding genome
region (Kearsey and Pooni, 1996). There is a growing
number of alternative markers, e.g., amplification fragment length polymorphism (AFLP), based on small differences in base sequence; information about them is
now widely available (e.g., Staub and Serquen, 1996;
Jones et al., 1997).
Until the mid-1980s, most of the mapped genes were
mutant genes with clear phenotypes; mapping gradually
progressed by looking at progeny from crosses between
the carrier of the new gene and genotypes carrying already mapped gene(s) (e.g., Woodward, 1957). With the
advent of DNA markers, we are in the position of analyzing a large number of recognizable loci segregating
simultaneously in the same cross. To handle the large
number of loci, a variety of software (e.g., Stam, 1993)
has been developed to establish the overall map that
gives the best fit to the combined data. An example of
such a map, using AFLP produced from a cross between
two barley cultivars, is illustrated in Fig. 2 for chromosome 3. Needless to say, information about the position
Fig. 2. Amplification fragment length polymorphism (AFLP) marker
map for chromosome 3, established by the use of software JoinMap
(Stam, 1993), for Apex ⫻ Prisma recombinant inbred line population of barley (redrawn from Yin et al., 1999b). The AFLP markers
are labeled E45M55-408, E44M58-196, etc., and their map positions
are counted in genetic map units, cM (1 cM corresponds to 1%
recombination per meiosis), from the top terminal marker. The
position of the denso dwarfing gene is highlighted.
of mutant loci is important for building a DNA-marker
map because previously mapped mutant loci (e.g., the
denso locus in Fig. 2) can be used as anchor markers that
assign new markers to different chromosomal groups. A
growing number of map databases in plants now become
YIN ET AL.: THE COMPLEMENTARITY OF CROP MODELING AND QTL MAPPING
accessible through the web sites (e.g., http://www.nal.
usda.gov/pgdic; verified 11 Sept. 2002).
A marker linkage map can be used to localize QTL
for a quantitative trait, as first demonstrated by Paterson
et al. (1988). The basis of all QTL detection is the statistical analysis of associations between markers and trait
values (Fig. 3). Statistical techniques for using a marker
map to detect QTL have reached a fairly high level of
sophistication, but improvements are still being made
(Kearsey and Farquhar, 1998). A widely used method was
interval mapping (Lander and Botstein, 1989). Other
approaches, e.g., the multiple QTL method (Jansen,
1995), were developed to detect multiple linked QTL.
However, a QTL detected by any technique is not a
true gene, only the indicated genome region that most
likely contains gene(s) for the trait under study.
The number of research reports on QTL analysis of
specific crop traits, using the methodology outlined
above, is growing rapidly. Almost all studies, regardless
of the crop or trait to which they are applied, have come
to support a main result of the first study (Paterson et
al., 1988) that, in a given cross, a small number of QTL
explained a large part of the genetic variation, even for
highly complex traits. This result differs from the assumption of classical quantitative genetics of the effects
of many genes with small and similar actions.
Two complementary uses of the QTL approach have
emerged: the fundamental and the applied (Prioul et al.,
1997). The first use, which is of interest to physiologists,
targets QTL by determining their contribution to physiological components of macroscopic traits. Not only does
the QTL approach provide unequivocal answers to a
range of physiological questions, it also generates new
insight into the causality between components that would
have been difficult to obtain by conventional physiological approaches (e.g., Simko et al., 1997). The importance
of the QTL approach is shown in a special issue of New
Phytologist [1997, 137(1)], which was entirely devoted
to proselytizing physiologists to take a genetic approach.
The second use of the QTL studies, which is of interest
to breeders, is marker-assisted breeding (MAB). This
approach uses markers for tagging QTL of interest so
as to pyramid favorable QTL alleles and break their
linkage with undesirable alleles (Lee, 1995; Ordon et al.,
1998; Ribaut and Hoisington, 1998). An apparent use of
MAB is the marker-steered introgression with valuable
single genes from exotic donors to enhance elite breeding material (Stam, 1998), which allows faster recovery
of the recipient-parent genome than the conventional
recurrent backcrossing (Ribaut and Hoisington, 1998).
As alien species or landraces are rich in resistance genes
and resistances are simply inherited relative to yield
traits, the application of markers for tagging of resistance genes in major crops has progressed rapidly (Ordon et al., 1998).
A major challenge for MAB is to deal with traits
controlled by multiple interactive and environmentdependent QTL, such as yield and yield-relating traits
that often have a low h2. Genetic simulation studies (e.g.,
Van Berloo and Stam, 1998) have shown that MAB
can be superior to the conventional phenotype-based
approach for traits of low h2, and there is some evidence
93
Fig. 3. The QTL mapping of six traits in barley (data of the 1997 field
experiment reported by Yin et al., 1999b). The interval mapping
method (Lander and Botstein, 1989) was used. By moving the
position of the putative QTL along the genetic map (horizontal
axis), a profile of the test statistics, QTL likelihood (LOD), was
produced for each chromosome, and results are given here for
chromosome 3. The peak of the LOD profile indicates the most
likely position of a QTL affecting the trait under study. Results
indicate a major QTL at 126.4 cM on chromosome 3 (note that
this is also the position of the denso gene as shown in Fig. 2). A
in each figure refers to the additive effect of this QTL on the
trait, defined as (mean of dwarf genotypes ⫺ mean of nondwarf
genotypes)/2. DS, developmental stage.
94
AGRONOMY JOURNAL, VOL. 95, JANUARY–FEBRUARY 2003
that marker-facilitated backcrossing can be employed
to manipulate and improve grain yield in maize (Zea
mays L.) (Stuber et al., 1999). However, in most cases,
the superiority of MAB has not been convincingly demonstrated experimentally (Ribaut and Hoisington, 1998).
Manipulating these traits is difficult because of their
intrinsic complexities: polygenic control, epistasis, and
G ⫻ E. Existing QTL detection methods do not seem
to have the required precision to deal with these complexities. With traits like yield that have a low h2, many
QTL may be segregating. The QTL with major effects
are easily manipulated by empirical breeding practices
and may already be fixed in many breeding lines. It would
be more productive to use marker technology as a means
for placing greater emphasis on those QTL that show
only relative minor effects (Stuber et al., 1999). The
location of minor QTL identified by existing mapping
methods may have wide confidence intervals. The most
likely location of a useful QTL may appear to be between a pair of markers, but it could actually be as far
as 20 cM away (Kearsey and Farquhar, 1998). While
recent multiple-QTL methods (e.g., Jansen, 1995) can
reduce confidence intervals of QTL locations (Fig. 4A)
and resolve two or more linked QTL, the efficacy of
Fig. 4. Plot of QTL likelihood (LOD) over chromosome 3 for specific
leaf area (SLA) in barley measured at 27 d after emergence (DAE)
and for SLA corrected at the same developmental stage (DS) of
0.35, roughly equivalent to the time of 27 DAE (redrawn from
Yin et al., 1999a). The horizontal line at a height of 2.5 indicates
the threshold for the presence of a QTL. The dotted curve is from
the use of the interval mapping method (Lander and Botstein,
1989), and the thin continuous one is from the use of the multipleQTL mapping method (Jansen, 1995).
these methods depends on whether markers are evenly
distributed in the map. In principle, epistasis of QTL
can be included within the frame of these multiple-QTL
methods. However, the rapid increase in the number of
parameters, difficulties to decide which interactions to
include, and the computational burden force us to assume the absence of epistasis. Methods have been developed to evaluate QTL ⫻ environment interactions
(QTL ⫻ E) using multiple-environment data (e.g., Jiang
and Zeng, 1995; Van Eeuwijk et al., 2000), but the information obtained cannot be applied to predict phenotypes in independent environments (Stratton, 1998).
COMBINING CROP MODELING
AND QTL MAPPING:
AN EXPLORATIVE STUDY
The first study to explore opportunities of linking crop
modeling with QTL mapping was recently conducted for
barley (Yin et al., 1999a, 1999b, 2000a), using a SUCROS
(Goudriaan and Van Laar, 1994)-based yield prediction
model, SYP-BL (Yin et al., 2000b). Main model input
traits included preflowering duration, postflowering duration, specific leaf area (SLA), leaf N concentration,
and fraction of biomass partitioned to leaves and to
spikes. The QTL approach was applied to these traits,
using a population consisting of 94 recombinant inbred
lines (RILs) from a cross of two-row spring barley cultivars, Apex and Prisma (Yin et al., 1999b). An AFLP
marker linkage map was established for this population
(see Fig. 2 for the case of chromosome 3). By analyzing
the association between trait phenotypes and marker
genotypes of the 94 RILs, QTL were found for each of
the above model input traits (Yin et al., 1999b). Most
traits were associated, though to different extents, with
the major dwarfing gene (with the mutant dwarf allele
from Prisma), denso (also designated as sdw1), which
was mapped at 126.4 cM on chromosome 3 (Fig. 2) by
segregation analysis of the distinctive prostrate juvenile
growth habit. The importance of this gene on a number
of traits, including some model input traits, based on
QTL analysis, is highlighted in Fig. 3. The result with the
RIL population that the major QTL for so many different traits mapped at the same position as the denso
locus is in support of the pleiotropy of this gene (Yin et
al., 1999b). The additive effect of the locus (Fig. 3) indicates the direction of the gene effect on each of these
traits; for example, the dwarf allele is associated with a
prolonged flowering time. This analysis provides direct
evidence for the genetic background and the interdependence of various model input parameters, which has
received little consideration from modelers in designing
ideotypes (Aggarwal et al., 1997b; Stam, 1998).
Physiological aspects of a trait, which have so far
received little attention from geneticists in QTL analysis, were considered, using SLA as the example (Yin et
al., 1999a). The SLA was measured six times: one conducted at the same developmental stage for all RILs
(at flowering), four at specific days before flowering,
and the last one at 14 d after flowering. When the SLA of
each measurement time was directly subjected to analysis,
YIN ET AL.: THE COMPLEMENTARITY OF CROP MODELING AND QTL MAPPING
one to three QTL were detected. The denso gene was
found to affect SLA strongly at all measurement times,
e.g., 27 d after emergence (Fig. 4A), except at flowering.
If the SLA of the different RILs was corrected for differences in physiological age at the time of measurement,
using the phenology submodel in SYP-BL, QTL were
detected for SLA at only three stages. Moreover, the effect of the denso gene was no longer significant during the
preflowering stages, e.g., at developmental stage 0.35
(Fig. 4B). The effect of the denso gene on the SLA
detected in the first instance was therefore the artifact
of its direct effect on the preflowering duration that can
be seen in Fig. 3. This result suggests potential use of
physiology and modeling in QTL analysis. Any further
roles of physiology or modeling should be explored,
especially given that any great change in the reliability
of QTL detection methods can hardly be achieved in
future (Kearsey and Farquhar, 1998).
Next, the identified QTL were coupled to the SYPBL model by replacing the original measured input trait
values with those predicted from the QTL effects (Yin
et al., 2000a). This replacement generated a QTL-based
model for barley, QTL-BL. Yields predicted by both
models correlated with the observed values, despite substantial unexplained variation (Fig. 5). The QTL-BL
model predicted yield differences slightly better than the
SYP-BL model. Similar results were obtained when the
models were applied to a season independent from the
one in which the original input traits used for QTL
analysis were measured (Yin et al., 2000a). The slightly
better performance of QTL-BL could be due to less
random noise in the QTL-based values because the random error in measured model input traits was partly
removed by QTL analysis statistics. However, this advantage of the QTL-BL model is obtained at the cost
of ignoring some genetic effects because all of the QTL
detected for a trait often do not fully explain its genetic
variation. Nonetheless, the correlation between SYPBL- and QTL-BL-predicted yields was high (r ⬎ 0.88),
indicating that QTL information can successfully replace measured parameters (Yin et al., 2000a).
EXPECTATIONS AND
FUTURE PERSPECTIVE
Potential Uses of Crop Models in Plant
Breeding and QTL Mapping
For a model to be an effective tool in breeding, it
must accurately simulate the difference in performance
among relatively similar lines in a population (McLaren,
1995). Obviously, however, there remains substantial
yield variation that was not explained by the existing
model (Fig. 5). Current crop models have to be improved in this context, in considerations of both input
traits and feedback structure. The random errors in input traits of current models are largely caused by field
samplings. The need for destructive samplings to determine input parameters is a major limitation in using
them in breeding, not only because of the required
amount of work but also because of the limited amount
95
Fig. 5. Comparisons between observed grain yields in barley and
those predicted by the two models, QTL-BL and SYP-BL (redrawn
from Yin et al., 2000a). Because the denso gene affected most
model input traits, QTL-BL predicted two clusters in yields that
match the segregation of the gene (the dwarf recombinant inbred
lines had higher yields than the tall lines).
of available material (Aggarwal et al., 1997b; Stam,
1998). Input parameters of current crop models may
vary with environment (Yin et al., 2000b). Model parameters have to be environment independent to enable the
models to extrapolate G ⫻ E, the expected advantage of
process-based crop models over any data-based genetic
G ⫻ E models (Shorter et al., 1991; Hunt et al., 1993).
While there is evidence that current crop models can
partially reproduce the observed G ⫻ E in cultivar performance trials (Mavromatis et al., 2001), developing
models that can accurately predict G ⫻ E on yields in
a population is a major challenge for modelers.
If models are capable of predicting G ⫻ E in a population, they can assist QTL analysis to resolve QTL ⫻ E, a
major problem that hampers the use of MAB in practical
breeding (Lee, 1995; Ribaut and Hoisington, 1998). The
QTL ⫻ E is commonly seen when growing a mapping
population under a range of environments. An example
of this is flowering time in Arabidopsis spp., examined
under various daylength and vernalization regimes (Jansen, 1995). It turned out that daylength and/or vernaliza-
96
AGRONOMY JOURNAL, VOL. 95, JANUARY–FEBRUARY 2003
than to resolve G ⫻ E. It could be demonstrated first
for relatively simple traits (such as time to flowering)
or in species with simple genetic makeup (such as Arabidopsis spp.) through simulating relevant biochemical
pathways.
Integration of Crop Modeling and QTL
Mapping into a Breeding Strategy
Fig. 6. Proposed framework of combining crop modeling and QTL
mapping for an integrated approach to select crop ideotype for a
specific environment. The dotted part in the figure is optional
for this framework because development of crop models that are
capable of resolving epistasis may take a long-term effort. G ⫻ E,
genotype ⫻ environment interaction; AFLP, amplification fragment length polymorphism.
tion influence the effect of some QTL, indicating QTL ⫻
E in a statistical sense. However, this information on
interaction cannot be applied to new environments (Stratton, 1998). From a physiologists’ or modelers’ point of
view, the impact of environments has to be minimized
to identify the true genetic effect. Phenology models
separate different aspects of flowering responses to photothermal environments (Atkinson and Porter, 1996).
Parameters in a physiologically robust phenology model
are genetically determined and are not altered by environment but predict flowering date of genotypes in a
wide range of environments (Roberts et al., 1996). It
is therefore expected that the QTL and their effects,
detected for model parameters, will not be environment dependent.
When crop models enter a high-precision stage at
which critical processes are quantified and integrated
at the biochemical level, they could be used to resolve
epistasis, a classical difficulty in genetics. Epistasis is
often found for phenotypes that are achieved through
interactive and interrelated metabolic and ontogenetic
pathways (Lee, 1995). It might be reduced or even disappear if input traits of a model that accounts for interrelations among relevant processes are subjected to analysis.
Such possibility agrees with the awareness of geneticists
that epistasis can often be removed by a physiologically
based scaling of trait values (Kearsey and Pooni, 1996).
It should be acknowledged, however, that use of crop
models to resolve epistasis may be a more difficult task
When crop models advance to the level of reliably
predicting genotype difference, crop modeling could be
integrated into the framework of MAB for an improved
breeding approach (Fig. 6). Within this integrated approach, the crop model is evaluated if it predicts yield
differences among genotypes in a genetic population
under diverse environments; thus, G ⫻ E is interpreted
in terms of a biological, as opposite to statistical, model.
Mapping is performed on input traits of the model to
dissect their variation into individual QTL, which in
turn, will be coupled to the model. Once the physiological and genetic bases of yield responses to environments
are adequately quantified, ideotypes can be proposed
for a specific environment (Atkinson and Porter, 1996)
in terms of the allelic constitution of the QTL for model
input traits that determine yield. This approach overcomes the limitations in designing ideotypes by using
models that ignore genetic constraints and correlations
among the traits. Information obtained can be applied
to any environment because of the high ability of extrapolation of crop modeling. With this integrated approach,
epistasis may also be considered (Fig. 6), but resolving
epistasis needs a long-term strategy.
While the proposed integrated approach could potentially deal with G ⫻ E and epistasis, it cannot solve all
limiting factors in using MAB, especially nontransferability of information obtained from one cross to another. The nontransferability can be largely due to the
possibility that a QTL detected in one cross simply does
not segregate in a second cross because the parents of
the second cross carry identical alleles at that QTL. A
gene important for physiologists or modelers may be
useless for geneticists or breeders because if the gene
is physiologically crucial, its variation will have been
strongly reduced over generations of breeding (Prioul
et al., 1997); so, QTL will hardly be detected at such a
gene locus.
Traditionally, physiologists have worked with only a
few genotypes but measured many characteristics or
processes to understand crop responses to environments.
In contrast, geneticists and breeders usually score a few
traits on many genotypes (often ⬎100) of a segregating
population and rely on selection and statistics to move
the population mean in the desired direction. This fundamental difference has often led geneticists and breeders to be skeptical of using physiological knowledge.
On one hand, our proposed integrated approach does
provide an excellent opportunity for collaboration among
physiologists, modelers, geneticists, and breeders. On the
other hand, implementation of such integrated approach
needs large experiments, assessing many traits in many
genotypes. To reduce this difficulty, the crop model
YIN ET AL.: THE COMPLEMENTARITY OF CROP MODELING AND QTL MAPPING
should be developed such that its input parameters can
be quickly assessed or through the way by which tissue
can be harvested and frozen for later analysis. Options
from genetic studies such as selective mapping (Xu and
Vogl, 2000) should also be considered. Reducing the size
of a mapping experiment with little sacrifice of the power
of QTL detection, as the common interest of geneticists
and physiologists, may represent a specific research area
for their collaborations.
Use of Marker Technology in
Modeling Cultivar Difference
Using regression analysis, Virk et al. (1996) has shown
that variation of many agronomic traits in rice germplasm is associated with allelic variation of markers,
indicating that marker-trait association is present not
only in segregating populations but also across a crop
germplasm or cultivar collection. If this result turns out
to be generally true, QTL-based modeling may be applicable to a germplasm collection, for which important
markers identified by, for example, multiple regression,
are used as the surrogate of QTL. Because the chance
that a specific marker maps to different genome positions in different populations within a species is small
(e.g., Waugh et al., 1997), we could use markers identified from a germplasm collection to infer the position of
QTL controlling the trait. This opportunity is especially
true when integrated marker maps based on acrosspopulation data are becoming available (e.g., Haanstra
et al., 1999). The applicability of marker information
across germplasm or cultivar collection would allow the
genetically based crop modeling to be performed without recourse to the use of a mapping population.
CONCLUSIONS
Crop modeling and quantitative genetics are independently evolving disciplines. Crop models are now increasingly being used to assist plant breeding, in particular, to define crop ideotypes for different environments.
The advent of DNA-based molecular markers and the
development in quantitative genetics of using marker
linkage maps to identify QTL provide a new perspective
for determining crop model input parameters, allowing
the biological meaning of model input parameters to
be more explicit. On the other hand, crop models can
potentially assist QTL mapping, especially in extrapolating information to a new environment and in dissecting
yield into physiological components that are more likely
related directly to gene expression. Based on the complementary aspects of modeling and mapping, an approach was proposed, from the future perspective, to
integrate marker-assisted technology into model-based
ideotype framework as a breeding strategy to increase
crop yield. Developing crop models, which can be sufficiently accurate to model G ⫻ E of a genetic population,
is the major challenge to enable this integrated approach
to be fruitfully used in a breeding program.
97
REFERENCES
Aggarwal, P.K., M.J. Kropff, K.G. Cassman, and H.F.M. ten Berge.
1997a. Simulating genotypic strategies for increasing rice yield potential in irrigated, tropical environments. Field Crops Res. 51:5–17.
Aggarwal, P.K., M.J. Kropff, P.S. Teng, and G.S. Khush. 1997b. The
challenge of integrating systems approaches in plant breeding: Opportunities, accomplishments and limitations. p. 1–23. In M.J.
Kropff et al. (ed.) Applications of systems approaches at the field
level. Kluwer Academic Publ., Dordrecht, the Netherlands.
Atkinson, D., and J.R. Porter. 1996. Temperature, plant development
and crop yields. Trends Plant Sci. 1:119–124.
Bindraban, P.S. 1997. Bridging the gap between plant physiology
and breeding: Identifying traits to increase wheat yield potential
using systems approaches. Ph.D. diss. Wageningen Agric. Univ.,
the Netherlands.
Boote, K.J., J.W. Jones, and N.B. Pickering. 1996. Potential uses and
limitations of crop models. Agron. J. 88:704–716.
Boote, K.J., and M. Tollenaar. 1994. Modeling genetic yield potential.
p. 533–565. In K.J. Boote et al. (ed.) Physiology and determination
of crop yield. ASA, CSSA, and SSSA, Madison, WI.
Botstein, D., R.L. White, M. Skolnick, and R.W. Davis. 1980. Construction of a genetic linkage map in man using restriction fragment
length polymorphisms. Am. J. Hum. Genet. 32:314–331.
Dingkuhn, M., F.W.T. Penning de Vries, and K.M. Miezan. 1993.
Improvement of rice plant type concepts: Systems research enables
interaction of physiology and breeding. p. 19–35. In F.W.T. Penning
de Vries et al. (ed.) Systems approaches for agricultural development. Kluwer Academic Publ., Dordrecht, the Netherlands.
Donald, C.M. 1968. The breeding of crop ideotypes. Euphytica 17:
385–403.
Dua, A.B., F.W.T. Penning de Vries, and D.V. Seshu. 1990. Simulation
to support evaluation of the production potential of rice varieties
in tropical climates. Trans. ASAE 33:1185–1194.
Goudriaan, J., and H.H. Van Laar. 1994. Modelling potential crop
growth processes. Kluwer Academic Publ., Dordrecht, the Netherlands.
Haanstra, J.P.W., C. Wye, H. Verbakel, F. Meijer-Dekens, P. Van
den Berg, P. Odinot, A.W. Van Heusden, S. Tanksley, P. Lindhout,
and J. Peleman. 1999. An integrated high density RFLP-AFLP map
of tomato based on two Lycopersicon esculentum ⫻ L. pennellii F2
populations. Theor. Appl. Genet. 99:254–271.
Haverkort, A.J., and P.L. Kooman. 1997. The use of systems analysis
and modeling of growth and development in potato ideotyping
under conditions affecting yields. Euphytica 94:191–200.
Hunt, L.A., S. Pararajasingham, J.W. Jones, G. Hoogenboom, D.T.
Imamura, and R.M. Ogoshi. 1993. GENCALC: Software to facilitate the use of crop models for analyzing field experiments. Agron.
J. 85:1090–1094.
Jackson, P., M. Robertson, M. Cooper, and G. Hammer. 1996. The role
of physiological understanding in plant breeding, from a breeding
perspective. Field Crops Res. 49:11–37.
Jansen, R.C. 1995. Genetic mapping quantitative trait loci in plants—a
novel statistical approach. Ph.D. diss. Wageningen Agric. Univ.,
the Netherlands.
Jiang, C.J., and Z.B. Zeng. 1995. Multiple trait analysis of genetic
mapping for quantitative trait loci. Genetics 140:1111–1127.
Jones, N., H. Ougham, and H. Thomas. 1997. Markers and mapping:
We are all geneticists now. New Phytol. 137:165–177.
Kearsey, M.J., and A.G.L. Farquhar. 1998. QTL analysis in plants:
Where are we now? Heredity 80:137–142.
Kearsey, M.J., and H.S. Pooni. 1996. The genetical analysis of quantitative traits. Chapman and Hall, London.
Kropff, M.J., A.J. Haverkort, P.K. Aggarwal, and P.L. Kooman. 1995.
Using systems approaches to design and evaluate ideotypes for
specific environments. p. 417–435. In J. Bouma et al. (ed.) Ecoregional approaches for sustainable land use and food production.
Kluwer Academic Publ., Dordrecht, the Netherlands.
Lander, E.S., and D. Botstein. 1989. Mapping Mendelian factors underlying quantitative traits using RFLP linkage maps. Genetics
121:185–199.
Lee, M. 1995. DNA markers and plant breeding programs. Adv.
Agron. 55:265–344.
98
AGRONOMY JOURNAL, VOL. 95, JANUARY–FEBRUARY 2003
Loomis, R.S., R. Rabbinge, and E. Ng. 1979. Explanatory models in
crop physiology. Annu. Rev. Plant Physiol. 30:339–367.
Mavromatis, T., K.J. Boote, J.W. Jones, A. Irmak, D. Shinde, and
G. Hoogenboom. 2001. Developing genetic coefficients for crop
simulation models with data from crop performance trials. Crop
Sci. 41:40–51.
McLaren, C.G. 1995. Combining statistics and crop models for improved plant breeding strategies. p. 41–47. In P.K. Aggarwal et al.
(ed.) Applications of systems approaches in plant breeding. IRRI,
Los Baños, Philippines.
Miflin, B. 2000. Crop improvement in the 21st century. J. Exp. Bot.
51:1–8.
Normile, D. 1999. Crossing rice strains to keep Asia’s rice bowls
brimming. Science 283:313.
Ordon, F., W. Wenzel, and W. Friedt. 1998. Recombination: Molecular
markers for resistance genes in major grain crops. Prog. Bot. 59:
49–79.
Paterson, A.H., E.S. Lander, J.D. Hewitt, S. Peterson, S.E. Lincoln,
and S.D. Tanksley. 1988. Resolution of quantitative factors by using
a complete linkage map of restriction fragment length polymorphisms. Nature (London) 335:721–726.
Peng, S., K.G. Cassman, S.S. Virmani, J. Sheehy, and G.S. Khush.
1999. Yield potential trends of tropical rice since the release of
IR8 and the challenge of increasing rice yield potential. Crop
Sci. 39:1552–1559.
Penning de Vries, F.W.T. 1991. Improving yields: Designing and testing VHYVs. p. 13–19. In F.W.T. Penning de Vries et al. (ed.)
Systems simulations at IRRI. IRRI Res. Paper 151. IRRI, Los
Baños, Philippines.
Prioul, J.L., S. Quarrie, M. Causse, and D. de Vienne. 1997. Dissecting
complex physiological functions through the use of molecular quantitative genetics. J. Exp. Bot. 48:1151–1163.
Rasmusson, D.C. 1987. An evaluation of ideotype breeding. Crop
Sci. 27:1140–1146.
Rasmusson, D.C. 1991. A plant breeder’s experience with ideotype
breeding. Field Crops Res. 26:191–200.
Ribaut, J.-M., and D. Hoisington. 1998. Marker-assisted selection:
New tools and strategies. Trends Plant Sci. 3:236–238.
Roberts, E.H., A. Qi, R.H. Ellis, R.J. Summerfield, R.J. Lawn, and S.
Shanmugasundaram. 1996. Use of field observations to characterise
genotypic flowering responses to photoperiod and temperature: A
soyabean exemplar. Theor. Appl. Genet. 93:519–533.
Sax, K. 1923. The association of size differences with seed-coat pattern
and pigmentation in Phaseolus vulgaris. Genetics 8:552–560.
Setter, T.L., E.A. Conocono, J.A. Egdane, and M.J. Kropff. 1995.
Possibility of increasing yield potential of rice by reducing panicle
height in the canopy: I. Effects of panicles on light interception
and canopy photosynthesis. Aust. J. Plant Physiol. 22:441–451.
Shorter, R., R.J. Lawn, and G.L. Hammer. 1991. Improving genotypic
adaptation in crops—a role for breeders, physiologists and modelers. Exp. Agric. 27:155–175.
Simko, I., S. McMurry, H.M. Yang, A. Manschot, P.J. Davies, and
E.E. Ewing. 1997. Evidence from polygene mapping for a causal
relationship between potato tuber dormancy and abscisic acid content. Plant Physiol. 115:1453–1459.
Stam, P. 1993. Construction of integrated genetic linkage maps by
means of a new computer package: JoinMap. Plant J. 3:739–744.
Stam, P. 1998. Crop physiology, QTL analysis and plant breeding. p.
429–440. In H. Lambers et al. (ed.) Inherent variation in plant
growth: Physiological mechanisms and ecological consequences.
Backhuys Publ., Leiden, the Netherlands.
Staub, J.E., and F.C. Serquen. 1996. Genetic markers, map construction, and their application in plant breeding. HortScience 31:729–741.
Stratton, D.A. 1998. Reaction norm functions and QTL-environment
interactions for flowering time in Arabidopsis thaliana. Heredity
81:144–155.
Stuber, C.W., M. Polacco, and M.L. Senior. 1999. Synergy of empirical
breeding, marker-assisted selection, and genomics to increase crop
yield potential. Crop Sci. 39:1571–1583.
Van Berloo, R., and P. Stam. 1998. Marker-assisted selection in autogamous RIL population: A simulation study. Theor. Appl. Genet.
96:147–154.
Van Eeuwijk, F.A., J. Crossa, M. Vargas, and J.-M. Ribaut. 2000.
Variants of factorial regression for analysing QTL by environment
interaction. p. 107–116. In A. Gallais et al. (ed.) Quantitative genetics and breeding methods. Proc. Meet. of the EUCARPIA Section
Biometr. in Plant Breeding, 11th, Paris. 30 Aug.–1 Sept. 2000.
INRA Editions, Versailles Cedex, France.
Virk, P.S., B.V. Ford-lloyd, M.T. Jackson, H.S. Pooni, T.P. Clemeno,
and H.J. Newbury. 1996. Predicting quantitative variation within
rice germplasm using molecular markers. Heredity 76:296–304.
Waugh, R., N. Bonar, E. Baird, B. Thomas, A. Graner, P. Hayes, and
W. Powell. 1997. Homology of AFLP products in three mapping
populations of barley. Mol. Gen. Genet. 255:311–321.
Whisler, F.D., B. Acock, D.N. Baker, R.E. Fye, H.F. Hodges, J.R.
Lambert, H.E. Lemmon, J.M. McKinion, and V.R. Reddy. 1986.
Crop simulation models in agronomic systems. Adv. Agron. 40:
141–208.
White, J.W., and G. Hoogenboom. 1996. Simulating effects of genes
for physiological traits in a process-oriented crop model. Agron.
J. 88:416–422.
Woodward, R.W. 1957. Linkages in barley. Agron. J. 49:28–32.
Xu, S.Z., and C. Vogl. 2000. Maximum likelihood analysis of quantitative trait loci under selective genotyping. Heredity 84:525–537.
Yin, X., S.C. Chasalow, C.J. Dourleijn, P. Stam, and M.J. Kropff.
2000a. Coupling estimated effects of QTLs for physiological traits
to a crop growth model: Predicting yield variation among recombinant inbred lines in barley. Heredity 85:539–549.
Yin, X., M.J. Kropff, P.K. Aggarwal, S. Peng, and T. Horie. 1997.
Optimal preflowering phenology of irrigated rice for high yield
potential in three Asian environments: A simulation study. Field
Crops Res. 51:19–27.
Yin, X., M.J. Kropff, J. Goudriaan, and P. Stam. 2000b. A model
analysis of yield differences among recombinant inbred lines in
barley. Agron. J. 92:114–120.
Yin, X., M.J. Kropff, and P. Stam. 1999a. The role of ecophysiological
models in QTL analysis: The example of specific leaf area in barley.
Heredity 82:415–421.
Yin, X., P. Stam, C.J. Dourleijn, and M.J. Kropff. 1999b. AFLP mapping of quantitative trait loci for yield-determining physiological
characters in spring barley. Theor. Appl. Genet. 99:244–253.