| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a CSIRO Plant Industry, Long Pocket Lab., 120 Meiers Rd., Indooroopilly, 4068, QLD, Australia
b School of Land and Food Sci., The University of Queensland, Brisbane, 4072, QLD, Australia
c Agric. Prod. Syst. Res. Unit, Queensland Dep. of Primary Industries, P.O. Box 102, Toowoomba, QLD, 4350, Australia
* Corresponding author (scott.chapman{at}csiro.au)
Received for publication May 1, 2001.
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Abbreviations: G x E, genotype x environment • MET, multienvironment trial • OA, osmotic adjustment • PH, flowering time • QTLs, quantitative trait loci • SG, stay-green • TE, transpiration efficiency • TPE, target population of environments
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
In quantitative genetics, computer simulation is commonly used to evaluate alternative plant breeding strategies on the basis of stochastic descriptions of gene action and interaction (e.g., Hospital et al., 1997; Podlich and Cooper, 1998). Our aim is to demonstrate how linkages between gene action and crop performance in dryland environments can be investigated by combining the biophysical response simulation of the of crops to the moisture environment with the quantitative genetics simulation of plant breeding programs. A review of present research will reveal that substantial resources are being invested into the cellular and molecular basis for adaptation to dry environments while plant breeding companies and public programs continue to make advances in yield through conventional means. Exploitation of the investment in the former requires the integration of knowledge from agronomy and cellular, plant, and crop physiology as well as plant breeding and quantitative genetics. For those less familiar with the breeding and quantitative genetics, there are some useful background texts (e.g., Hallauer et al., 1988; Falconer and Mackay, 1996) in addition to Podlich and Cooper (1998) and Chapman et al. (2002a).
In combining gene sequencing, gene cloning, and plant transformation with biochemical and genome databases, scientists in private and public industry have identified and constructed genes that control relatively linear pathways like herbicide tolerance, disease resistance, and product quality (Somerville and Somerville, 1999; Mazur et al., 1999). Molecular biology is beginning to investigate the role of the other genes that relate to adaptation to the abiotic environment. For these genotype–environment systems, thousands of genes interact in complex ways to generate crop responses to the environment via mediation of responses over both short time scales (e.g., cellular response to environment shocks like frost) and long time scales (e.g., morphological growth responses of crop development and morphology). Some pathways for abiotic adaptation are comparatively straightforward, e.g., direct cellular tolerance of salt stresses (see review by Hasegawa et al., 2000). However, it will be some time (>20 yr?) before we understand how the interactions of developmental and signaling genes control yield of crops as a function of responses at the biochemical, cellular, plant, and canopy or crop levels of organization. Until then, we need to deal with adaptive traits at a more integrative level (i.e., traits observable at the plant or crop level) while beginning to connect the tools and databases that are developing in all of the research areas (molecular biology, plant breeding, and plant and crop physiology) to understand both the effects of genes on pathways and how these are mediated in the responses of crops to the environment. Complexity per se has become an area of serious research (Gell-Mann, 1994), and the process of plant breeding is an example of the challenges to be faced in understanding the interactions of genes with each other and with environments.
|
Simulation of Crop Response to Environment and Gene Flow through Plant Breeding Programs |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Not all crop simulation models are suitable for use in these genetic frameworks. Hammer (1998) revisited the concept of emergent properties (de Wit and Penning de Vries, 1983). This concept suggests that modelers should attempt to define the rules that set the boundary conditions for simulation processes rather than applying a descriptive structure. The model needs to be able to handle perturbations to any process and self-correct, as do plants under hormonal control when growing in the field. This philosophy of parameterization and modeling of the principles of response and feedbacks, cf. description of response, infers that models should be able to express complex behavior of the type observed in the field, even given simple operational rules at a functional crop physiological level. The sorghum [Sorghum bicolor (L.) Moench] crop module (APSIM-Sorg) within the APSIM cropping systems model (McCown et al., 1996) contains several deliberate parameterizations to address genetic variation using a boundary conditions approach (Hammer and Muchow, 1994; Hammer et al., 1999; Chapman et al., 2002a), e.g., the model employs a switching method to estimate crop growth rate when limited by either radiation or water (Chapman et al., 1993) to utilize our ability to characterize the crop level efficiencies of radiation or water use for different genotypes.
The QU-GENE simulation platform (Podlich and Cooper, 1998) simulates the stochastic properties of genes, genotypes, and environments in the operation of plant breeding programs. It can model breeding programs as search strategies that seek higher peaks on the adaptation landscape (genetic space) for a given genotype–environment system. Searches progress by creating, identifying, and selecting genotypes with improved adaptation to the TPE. The rate at which a population improves with selection is monitored by the change in grain yield of successive cycles and in the changes in the fixation (gene frequency) of both positive and negative alleles related to this yield improvement. Statistical analyses determine the effectiveness of searches in creating and finding superior combinations of alleles in the simulated populations. The superior methods of recombination and searching genetic space can then be considered for application in conventional plant breeding programs. The methods evaluated might include such things as different methods of recombining genotypes and different levels of selection pressure (the proportion of the population selected for recombination) as well as improved statistical interpretations of adaptation.
In using QU-GENE to define the genetic space to be searched, the actions of genes and their interactions with other genes (epistasis) and with environments (gene x environment interactions) are prescribed for different crop traits, as are gene associations with molecular markers. As in other genetic simulation studies (e.g., Hospital et al., 1997; Van Berloo and Stam, 1998), these actions and interactions have normally been derived from field experiments as stochastic parameters (estimates of variance components and heritability) and from direct knowledge of the allelic effects of genes on traits, yield, or both. Until now, in QU-GENE, as in other genetic simulation models, there has been no direct biophysical connection between the gene effects associated with a trait and the yield phenotype of resulting genotypes as modulated by abiotic environmental influences. Establishing this direct connection by linking QU-GENE and APSIM enables direct definition of the actions of genes on traits so that epistatic and genotype x environment (G x E) interactions for yield are emergent properties of the dynamics of the APSIM crop simulation model.
|
Improving the Efficiency of Plant Breeding for Dryland Environments |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Genomic type projects being initiated in the area of adaptation to drought or other abiotic stresses are largely focused on traits observed at the molecular and cellular levels, such as membrane stability or modified ion exchange and/or exclusion, e.g., Hasegawa et al. (2000). While these cellular traits may be essential in plant survival and, in some cases, contribute to economic yield, adaptation to variable rainfall environments is greatly mediated by traits observed at the crop level that influence the seasonal pattern and total water use of the crop, such as flowering time (PH), canopy transpiration efficiency (TE), leaf development and senescence, and repartitioning of dry matter. Richards and Belhassen (1996) have discussed examples of these effects for the adaptation of wheat (Triticum aestivum L.) to dryland environments. To a great extent, adaptation to drought as exploited through plant breeding has resulted from modifications of the normal process of growth and development (e.g., to change the pattern of water use), rather than the introduction of strong localized reactions of novel genes to a stress.
The objectives of this paper are to first demonstrate the simulation of yield resulting from definition of gene action for four physiological crop traits (additive effects with several genes and interacting states of expression) acting in different drought environment types. Second, we demonstrate how breeding progress is influenced by selection for yield given two common constraints experienced by plant breeders: restricted sampling of representative environments and the need to select within different maturity groups. The results are presented for long sequences (>10 cycles of evaluation, selection, and intermating) of recurrent S1 selection. As shall be evident, we do not (and may never) have a complete understanding of the genetic controls and physiological interactions among the targeted traits. Our objective is not to provide all of the answers, but to demonstrate tools to begin to integrate quantitative genetics with dynamic crop simulation in the investigation of the complexity of gene x gene x environment interaction.
|
MATERIALS AND METHODS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
2. For four crop traits [TE, PH, osmotic adjustment (OA), and SG], determine the yield value of all combinations of different gene expression levels for each trait (genotypes as near-isogenic lines) in all location–season combinations. For each genotype, calculate the mean yield across all of the location and year combinations that comprise each of the drought environment types. These data represent the yield genotype–environment space for the entire possible population of genotype–environment type combinations.
3. Using QU-GENE, model the processes of an example breeding program:
a. Initially sample the genotype–environment space where there was a low to moderate frequency (0.2) of favorable alleles for each gene to choose parents for intermating.
b. Evaluate S1 offspring in multienvironment trials (METs) (from a fixed or random sample of environment types) and select offspring based on mean yield.
c. Repeat the evaluation, selection, and intermating process for 12 cycles of S1 recurrent selection.
The results were then interpreted in terms of changes in the grain yield of the offspring and in the frequency of favorable alleles for the genes associated with each trait.
1. . Characterizing the Target Population of Environments
Chapman et al. (2002a) have described the environments (locations–season combinations) used in this experiment and their effects on the performance of 54 genotypes with the extreme (low and high) and standard levels of gene expression for the four traits considered.
Version 1.5 of APSIM was used to run the SORG (sorghum) module using weather data on a daily time step to interact with a specified soil profile and simulate the soil and plant processes associated with water and N during fallow and in-crop states (Fig. 1). We simulated an opportunity cropping system, i.e., winter or extended fallow, followed by sorghum during a summer planting window whenever minimum rainfall (25 mm in 4 d) and soil water conditions (80 mm) were achieved. Nitrogen was nonlimiting, and the crop was grown at 50 000 plants ha-1.
A genotype with all parameters set to the standard value for each of the traits (Table 1) was run using 108 yr of daily weather data at six locations (648 potential trials) across the sorghum production region of northeastern Australia. The weather data for these locations, obtained from the SILO data set (http://www.dnr.qld.gov.au/silo/; verified 21 Aug. 2002), contains actual temperature and rainfall, with actual solar radiation or, prior to 1956, solar radiation estimated from functions of temperature and cloud cover. The reference genotype was used to simulate, for each trial, the final grain yields and the average level of a water stress index (water supply/demand ratio) for each successive 100 degree days (thermal-time weeks) from emergence (Chapman et al., 2000a). This is simply to provide an objective basis (i.e., compared to sets of locations and years) for the classification of different stress environments that might be encountered in the breeding program. Individual simulations of genotypes in each environment (see next section) generate independent levels and timing of stress and therefore flow-on effects to other traits.
|
|
The TE trait and traits correlated with it (isotopic C discrimination ratio) are considered heritable as they have been selected for in plant breeding programs and backcrossed successfully into new germplasm to increase yield, e.g., wheat (Richards et al., 2002). The TE coefficient is here referred to as a trait that depends on the balance of the exchange of CO2 and water vapor at the cellular level of the leaf and is expressed at the level of the crop as the crop growth rate per unit of water transpired (absorbed by roots). The actual or realized TE depends on the vapor pressure deficit (VPD) such that: TEactual = TE/VPD. The value of TE has been shown in glasshouse and field trials of sorghum to vary across cultivars by approximately 10%, independent of vapor pressure deficit (Mortlock and Hammer, 1999). The effect of raising TE is to increase the efficiency with which water is utilized by the crop to meet the demand for potential (radiation limited) growth. Hence, when water supply is sufficient to meet demand, less soil water is extracted, leaving a larger late-season soil water reserve in seasons when rainfall is low. Alternatively, when water is insufficient to meet demand, more dry matter growth can be produced with that supply. Hammer et al. (1996) considered that cultivars with an increased TE also had a reduced radiation use efficiency. This effect was not implemented here as the data of Mortlock and Hammer (1999) did not support a clear negative association between these traits at the plant or crop level.
For trait PH, the thermal time requirement for the completion of the development stage end-juvenile to floral initiation was varied to simulate the genetic range of flowering dates (ca. 14 d) observed in the local germplasm pool. A longer thermal time requirement allowed more leaves to be initiated. Hence, increasing the trait PH increases the final number of leaves produced and therefore delays flowering because flowering only begins once all of the initiated leaves have appeared and expanded. To simplify this example, no modification of photoperiod response was introduced. Though photoperiod is believed to be a major control of PH in sorghum (Rooney and Aydin, 1999), the genotype effect can be satisfactorily mimicked using the method outlined above as the planting date and latitude variation in our example is not extreme.
Trait OA was implemented in terms of the observed crop-level effects in near-isogenic lines that differed in the level of OA under drought (P. Snell, unpublished data as summarized by Hammer et al., 1999). The crop-level effect of OA under drought conditions was to, first, reduce the amount of crop growth in the period from panicle initiation to flowering that is required to set a given number of grains and, second, increase the potential remobilization pool for the filling of grains. Increasing the genetic value of OA only has an effect under drought (Table 1), and its effect is to increase both the number of grains (sink size) and the retranslocation potential should drought continue into grain filling (Table 1).
Genetic variation in SG was simulated by modifying the target specific leaf N (g N m-2 leaf area) of new leaf. Borrell et al. (2000) reported on physiological studies of the mode of action of SG in hybrids from a cross of parents with high and low levels of SG. Increasing the target specific leaf N for new leaf allows increased N uptake during canopy development as found by Borrell et al. (2000) in SG type germplasm. Subsequently, during grain filling, depletion of N from leaves is delayed, causing the SG effect although other forms of SG are also known to exist (Thomas and Howarth, 2000).
In general, higher values for these different traits result in higher yields but not under all environmental conditions. This is particularly the case for PH where lower values result in early maturity, which may be an advantage to escape the effects of severe terminal drought. The background studies and references relating to the ranges of the traits described in Table 1 are given by Chapman et al. (2002a).
Defining Gene Action and Expression States for Physiological Traits
In QU-GENE, we modeled the multilocus effects for the states of expression of each trait as unlinked cumulative additive alleles across loci, with two alleles per locus, e.g., for the trait OA, we described two genes that are located at two loci, which can result in the five evenly distributed levels of expression (see below). As a convention for referring to genes and alleles, we use uppercase bold letters to refer to the gene and upper- and lowercase italic letters to refer to their alleles (e.g., gene A with alleles A and a). For each locus (position of a gene), one allele was considered to result in increased trait expression relative to the other allele. The expression state of a trait for a genotype was then determined by the total number of alleles for increased trait expression possessed by the genotype across all loci for the trait. The uppercase alleles were defined as the alleles that increased the level of a trait and are referred to as + alleles. Conversely, the lowercase alleles were defined as the alleles that decreased trait expression and are referred to as - alleles. Importantly, the + or - designation refers to their influence on the expression of a trait but not necessarily their effect on expressed traits such as leaf area, biomass, or yield. For a trait (OA) regulated by two genes [gene A with alleles A (+) and a (-) and gene B with alleles B (+) and b (-)], there were five possible expression states for the trait, based on a genotype possessing either zero, one, two, three, or four of the + alleles across genes A and B. With this gene expression model, different genotypes can have the same expression state. For example, genotypes Aabb, aAbb, aaBb, and aabB all have one + allele and therefore have the same expression state. Similarly, for the five expression states:
For the hypothesized genetic models for the four traits (Table 1), there are 315 = 14 348 907 different genotypes but a much smaller number of different expression states (11 x 7 x 5 x 11 = 4235). Alternative gene expression models could have been considered and would have created different relationships between the genotypes and phenotypes. The number of expression states (and genes) chosen for the different traits was determined either from some knowledge of gene action and recombination in breeding experiments (for PH and OA) or was suggested from the approximate number of strong molecular markers that have been found (for TE and SG). With respect to the number of gene expression states used in Table 1, we note that:
APSIM-Sorg was used as described above to generate yields for all 4235 expression states (equivalent to genotypes for our discussion) from QU-Gene for all 547 location–season combinations (Fig. 1). These yields represent a nonlinear interaction of the expression states of each trait with each other and with the growing environments. For each genotype, the mean yield in each of the three drought environment types was calculated. This was a simplification of the simulation experiment though we intend in future studies to sample environments from the three drought environment types rather than using the genotype mean across each drought environment type.
3. . Modeling the Breeding Program
The structure of the adaptation landscape for the genetic model was defined by the APSIM-Sorg crop model estimate of yield for different combinations of genes (genotypes) based on different expression levels of the four traits (Table 1). The QU-GENE software managed the creation, evaluation, and selection of genotypes within a breeding program (Fig. 1). The first stage of QU-GENE is the engine, which specifies the properties of the genetic models under investigation and hence the structure of the adaptation landscapes corresponding to the germplasm pool available. The engine creates a starting-point reference population of genotypes for investigation by second-stage application modules that simulate the structure of different plant breeding programs. In the configuration used here, grain yields were generated by APSIM for all of the genotypes and were averaged for each genotype over the three environment types.
A QU-GENE application module representing an S1 family recurrent selection breeding program was used to conduct the experiment (Fig. 1). The breeding program operated on a 4-yr cycle, with the first 2 yr used for random intermating, space plant selection, and seed increase of S1 families that are the direct offspring of the randomly intermated parents (Podlich et al., 1999). The S1 families were then evaluated in a MET over five locations and in 2 yr. Superior S1 families were selected on mean yield performance in the MET, determined from the database of APSIM runs. Reserve S1 family seed from the selected families was intercrossed to initiate the next cycle. The program was conducted for 12 cycles (equivalent to 48 yr), using a spaced plant population of 5000 individuals, with 1000 S1 families evaluated in both years of the MET. The top 100 S1 families were selected on superior performance in the MET. To simplify the analysis, plot heritability of each gene was assumed to be 1.0, i.e., there was no experimental error.
The timing of flowering is an important consideration with respect to the occurrence of drought, e.g., earlier-flowering genotypes are able to escape the effects of a terminal drought. Further, to fit in with other operations in a cropping system, breeding programs frequently provide a suite of cultivars differing in maturity. For summer crops like sorghum, a farmer may prefer to plant a longer-season variety if an early planting opportunity arises and a shorter-season variety when planting late so that the crop does not mature into cool conditions and interfere with future crop rotations. To consider these practical issues, several selection scenarios were evaluated:
1. No constraints selection for average yield in the MET, with the 10 testing environments randomly sampled at their occurrence rates in the TPE (Table 2).
2. Selection for average yield in the MET, with the 10 environments sampled from only one drought environment type. This restricted sampling process, repeated for each drought environment type, demonstrates the effects on selection of sampling only one type of drought stress pattern, as can happen when breeding programs experience a series of wet or dry years, e.g., Chapman et al. (2000b), or repeatedly sample a particular drought environment type through managed irrigation or drought treatments, e.g., Edmeades et al. (1999).
3. Selection for yield in the MET as in each of the two scenarios above but with phenology constrained to be within a particular class. Three phenology constraints on the selection process were implemented by retaining in selection only those genotypes (and hence S1 families) that possessed allelic combinations for the PH trait genes that fell into one of the sets of expression states 1 to 3 (early maturity), 3 to 5 (medium maturity), or 5 to 7 (late maturity) (Table 1). This simulates the common process utilized when breeders are selecting for adaptation within different maturity classes.
Given that the genetic composition of the initial parents can vary, the performance of each of the defined selection scenarios was evaluated as the average of 2000 runs (10 independent sets of starting parents x 200 independent runs of each set of parents) over 12 cycles of S1 recurrent selection. The initial parents were selected such that the population had a fixed gene frequency of 0.2 for the + alleles of each of the 15 genes that described the four traits. At each cycle, the mean of the S1 families in the MET and the average gene frequency of the genes regulating expression in the four traits were tabulated for interpretation. The genotypic and G x E interaction components of variance were also calculated for each cycle of selection and selection scenario.
Running the APSIM simulations was a substantial task. APSIM was installed on the QCC (QU-GENE Computer Cluster; Micallef et al., 2001), which comprises forty-eight 400 MHz (or greater) computers, reducing the simulation time to 5 d instead of 250 d on a single computer. Report files from APSIM were assembled from the computer cluster using customized scripts written in the Tcl/Tk control language and were stored in a database before production of data files for input to QU-GENE (Micallef et al., 2001). The QU-GENE simulations (see below) were completed in about 10 h using all 48 computers in the QCC.
|
RESULTS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
For each cycle of selection, the proportion of genes fixed (gene frequency) for the + alleles was averaged over the genes for each trait (Table 1) and over the 2000 QU-GENE runs (Fig. 2). When the MET was conducted using only the Severe-Terminal drought environment type, OA and TE were the first traits fixed for + alleles (Fig. 2a). Once these genes had been fixed, the SG genes, which had changed relatively little over the first four cycles, began to be fixed rapidly. The PH genes were fixed gradually to - alleles (early maturity) over seven cycles. By Cycle 9, all genes were fixed to either + (OA, TE, and SG) or - (PH) alleles. These effects on PH genes were in contrast to selection under environments that were solely Midseason drought environment type (Fig. 2b) or Mild-Terminal drought environment type (Fig. 2c). In these two environment types, all trait genes were fixed to + alleles, with the PH trait fixed the most quickly, followed by a constant, slower rate of fixing of the + alleles for the TE and OA genes and then SG (Midseason) or the SG and then TE and OA (Mild-Terminal). When the MET evaluation was conducted using environment types sampled in proportions to mimic the TPE (Fig. 2d), the patterns of gene fixing most closely resembled those observed when the Midseason drought environment type alone was the selection environment.
|
|
The associations between the fixing of genes for + or - alleles and the mean yield of the population in the selection environment can be assessed by comparing Fig. 2 and 3 with Fig. 4. Fig. 4a shows that the maximum yield ceiling of about 3.5 t/ha in the Severe-Terminal drought environment type was reached after about nine cycles of recurrent selection. Constraining the PH genes to only the early maturing genotypes had minimal influence on the rate or end point of yield improvement as these are indeed the best-adapted genotypes. There were two phases of yield improvement in both selection scenarios: first, a linear increase from Cycles 1 to 6 (associated with the complete fixing of OA and TE genes to the + alleles, PH to - alleles, and about 0.6 of the SG genes to + alleles Fig. 2a) and second, a plateauing of yield improvement in Cycles 6 to 9 associated with the fixing of the + alleles for the remaining segregating SG genes and one PH gene. The gene fixation patterns were similar for the early-constraint scenario in this environment (data not shown).
|
Figure 4 also shows the yield improvements associated with constraining selection to either medium- or late-maturing genotypes. When selecting in a Severe-Terminal drought environment type, the yield decreased for the first two cycles of evaluation (Fig. 4a) while the PH genes were being fixed to the earliest of the late-genotype range (Fig. 3a). After this point, the + alleles for the OA and TE genes were completely fixed around Cycle 7, followed by the SG genes from Cycles 7 to 12 in a manner similar to that seen in Fig. 2a. When constrained to the late genotypes, yield progress in either the Midseason stress, Mild-Terminal stress, or the TPE was slightly slower than in the unconstrained case although the same end point was eventually achieved (Fig. 4b, 4c, and 4d). The more rapid fixing of the PH genes was associated with a delay of between one and two cycles in the fixation of the genes for the remaining traits (e.g., in Fig. 3d cf. Fig. 2d). As may be expected, both rate of progress and ultimate yield of the medium-maturing genotypes was intermediate to that of the early- and late-maturity selection methods.
The results in Fig. 4 show what happens to adaptation for specific drought environment types or the TPE when genotypes are evaluated only in the same drought environment type or TPE in which selection was conducted. For the selected proportions of genotypes under any scenario, it is possible to determine their corresponding value in other drought environment types and in the TPE, i.e., even though the selection has been done in a particular environment combination, we can evaluate the resulting changes in any other combination (Fig. 5). This demonstrates the principle of indirect selection where selection within an environment leads to improved performance in another environment. For example, when the TPE was used as the selection environment, the yield for the TPE and for each of the drought environment types (i.e., indirectly selected for) increased at similar rates in percentage terms (data not shown) although there were differences in absolute terms (Fig. 5d). If the selection environment was restricted to only the Severe-Terminal drought environment type (Fig. 5a), then the rate of yield improvement was most rapid in that drought environment type but was much lower when the selected population was tested separately in the other drought environment types or the TPE. Notice that the final yield for the TPE was lower than in Fig. 5d (i.e., broad adaptation was lower than if the TPE had been used for selection) but that the final yield for a Severe-Terminal drought environment type was greater than for selection in either of the other drought environment types (Fig. 5b and 5c) or in the TPE (Fig. 5d).
|
Under the maturity-constrained scenarios for selection in the TPE, the computed genotypic variance component had decreased by about 30% at Cycle 3, with little further change until Cycle 6, and then decreased more gradually compared with the no constraints scenario (Fig. 6a). The pattern was similar for the G x E interaction variance components (Fig. 6b), such that the ratio of G x E interaction and genotype effects was initially highest in the no constraints scenario (Fig. 6c). The G x E variance for the early maturing constraint was greater than for the other PH constraint scenarios. In the last three cycles, the ratio of G x E interaction and genotype effects increased greatly and was variable across the last few cycles of selection as both components of variance had become relatively small by this time.
|
|
DISCUSSION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Apart from the trait PH in the Severe-Terminal drought environment type, increasing the level of trait expression (by increasing number of + alleles for a trait) resulted in greater grain yield (Table 2). Higher values for the PH and TE traits had a greater effect on yield as the environments changed from Severe-Terminal stress to Midseason stress to Mild-Terminal stress. Conversely, greater values of TE and OA had a greater absolute effect on yield in the Severe-Terminal drought environment type. When the 4235 genotypes were sampled in the recurrent selection breeding program, the relative effects of the traits on yield in different drought environment types determined the rate at which genes were fixed by selection for grain yield.
Under the no constraints on maturity scenario, clear differences existed in the rates at which genes were fixed during selection under different drought environment types or under the sampled mixture of drought environment types representing the TPE. When selected in the TPE, late-maturing genotypes were favored due to their higher yields in 65% of the component environments, i.e., in both Midseason and Mild-Terminal drought environment types (Fig. 4). The pattern of fixation of + alleles in the TPE was similar to that observed in the Midseason drought environment type. Hence, this was the best surrogate of the three drought environment types to represent the variation in the real-world TPE. However, in the TPE, the rate at which the PH genes were fixed to the latest-maturing genotypes was clearly slower than that observed in the Midseason and Mild-Terminal drought environment types.
The utility of + alleles for different traits in improving adaptation depended on the constraints set by both the maturity times of the genotypes and the selection environment. Early genotypes were favored in the Severe-Terminal drought environment type, with most of the + alleles for the SG trait not being fixed until the - alleles of the PH genes had first been fixed (Fig. 2). When the phenology was constrained to early-maturity types, the value of the + alleles for SG was increased in all environments as the SG alleles began to be fixed earlier in the selection process. Where the Severe-Terminal drought environment type was part of the selection environment (Fig. 2a and 2d), + alleles for OA genes were fixed more rapidly than those for TE. This reflects that fact that the OA genes were implemented in the crop simulation model to only have an effect under conditions of severe stress around flowering and during grain filling. As these conditions were less frequent in the Midseason and Mild-Terminal drought environment types than in the Severe-Terminal drought environment type, OA genes were not particularly favored over TE genes, which provide a yield advantage in all environment types.
The selection criteria (maturity) and the selection environment also interacted to influence the effectiveness of selection for yield. For example, in the Severe-Terminal drought environment type, selection with a constraint to early maturity resulted in the same rate of gain as no constraints. In contrast, although late maturity was favored in the other two drought environment types, constraining the selection to late-maturing types in these drought environment types reduced the rate of improvement in grain yield. It seems that restricting the selection in these environments effectively reduced the selection pressure for yield, i.e., if only 300 of the 1000 families met the late-phenology requirement and 100 of the 1000 have to be selected, then selection pressure is lessened to the degree that some superior-yielding genotypes (of early or medium maturity but with other favorable genes) did not meet the maturity criteria. This equates to a selection bottleneck in the breeding program where favorable-yield genes can be lost from the breeding population as their value is masked by the use of a rigorous culling criteria, in this case, maturity. Another type of bottleneck arises when breeding programs experience a particular sequence of environment types though we do not attempt to evaluate that here.
When evaluated in the TPE, the no constraints selection scenario retained the greatest degree of genotypic variance for yield for a longer period (until approximately Cycle 6) of the selection process (Fig. 6a). Constraining the phenology during selection quickly eroded the genotypic variance for yield although the later-maturing genotypes actually had lower G x E interaction effects in the TPE, again due to the dominance of favorable environments (Midseason and Mild-Terminal drought environment types) for them in the TPE compared with the early maturing genotypes.
It is clear from Fig. 2 that, in an unconstrained scenario selecting in the TPE, it is difficult to retain segregation for maturity because of its strong association with yield in the better environments. If we were employing this methodology in practice to deliver a suite of cultivar maturities to the industry, the results suggest that it would be useful to set up a separate early maturing population for adaptation to Severe-Terminal stress drought environment type as this drought environment type tends to be associated with locations in the shallow soils of Central Queensland and the poorer rainfall zones of northern New South Wales (Chapman et al., 2000b, 2002a). There seems less justification to have a medium-maturing population as the Midseason drought environment type where its quality of performance is not particularly biased in any location. Nevertheless, medium-maturing cultivars would provide greater stability (i.e., less G x E interaction for yield) in these poorer water environments than late-maturing cultivars.
The process of defining cumulative additive genes for traits and then expressing them via crop simulation generated both pleiotropic (where genes affect multiple traits, e.g., TE and yield) and epistatic (where different gene combinations interact with each other) effects for yield. These effects can be illustrated if we examine the results of a QU-GENE simulation for the same breeding program employing a conventional quantitative genetic model, i.e., many genes of small effect (see Podlich and Cooper, 1998, for examples of this simulation). The model was specified with 15 genes (as we used), each having small, equal additive effects on yield in different environments but with no specified epistatic or G x E effects and no elaboration of the trait effects via a biophysical crop simulation model. Figure 7a shows that with selection, each of these genes for yield becomes fixed at a similar rate. This similarity in gene fixation contrasts with the case for the QU-GENE results when using the sorghum crop model to determine crop yield as controlled by 15 genes directly affecting four traits (Fig. 7b). The use of the crop model has modified the relative yield value of the different genes in the environment types sampled and hence has influenced the rate and timing of fixation of the + alleles for the different traits. The difference between these two approaches also indicates the importance of the state of the current germplasm as well as the selection environments in the potential for progress in the breeding program, i.e., while the relative value of each of the genes in Fig. 7a, indicated by its rate of fixation, is more or less similar, this is not the case for Fig. 7b where the genes clearly have different values in terms of yield.
|
Consideration of the effects of epistasis when modeling selection scenarios is vitally important. It suggests that for different combinations of traits being tested in particular environments, the fixation of some traits is unlikely to proceed until one or more other traits have been improved and in some cases, partially fixed, e.g., SG in Fig. 2a and TE and OA in Fig. 3c. If we conducted a molecular marker experiment to identify molecular markers for SG using the Cycle 3 population from Fig. 2a, we would find a low frequency of + alleles for SG genes and possibly discover some useful molecular markers. However, even if we had perfect markers, we would not achieve any advance in yield until the genes for other traits had also become fixed, i.e., the marker experiment may appear to fail even though the newly fixed alleles would be coming into play following further phenotypic selection for yield. Similarly, if we were evaluating the same population in different environments, our assessment of the existence of markers would change. These scenario results have significant implications for the manner in which strategies such as marker-assisted selection are introduced to improve the efficiency of the breeding program for yield.
The QU-GENE software can be used to investigate the inclusion of markers and degrees of recombination and linkage among the genes in the model to answer some of the questions posed in the previous paragraph. In an extension of our current analysis, for example, we examined the effects of the precision of molecular markers on the efficiency of selection (Chapman et al., 2002b). A sophisticated enhancement would examine the issue of the difficulty of finding molecular markers for yield, per se. This difficulty is not unexpected (although many marker projects continue to attempt it) because yield is a pleiotropic effect of multiple genes controlling subprocesses of crop growth over the season. By outputting the values of other key attributes from the crop model (e.g., crop growth rate at flowering), it is feasible for a researcher using this model to find sets of attributes that control yield but are simpler in their genetic control. Investigating the interaction between the biophysical simulation and the breeding program simulation should also allow derivation of key traits and genetic networks controlling yield in different environments. Additional issues that could be addressed are the effects of variation in the genotype response within an environment type, the effects of selection within fixed sequences of environment types, and the effects of indirect selection for traits other than yield.
With the availability of high-throughput capacity to sequence genomes, it has been widely claimed (e.g., Bassett et al., 1999) that biologists have not determined how to cope with the massive amounts of DNA sequence and gene expression data that are being generated. Worse, there are few examples of linking this information to crop phenotypes in the field—most are laboratory-based phenotypes in simple controlled-environment screens, which may bear little correlation with field performance. As we expand our ability to describe and understand the structure and function of plant genomes, there is an urgent need to investigate how the effects of genes are integrated at the different organizational levels of the organism and how these organizational structures (e.g., pathways, traits, and trait combinations) interact with the biophysical properties of environments to determine the crop phenotype. While the task of understanding the relationship between gene and phenotype is a major undertaking, even for many simply inherited traits, the integration of genetic models, that can simulate the properties of gene–environment systems, with dynamic biophysical crop models, that simulate plant growth and development processes, provides a quantitative framework to support these ambitious investigations. Within the generic crop template design in APSIM, developments in improved modeling of the traits are quickly captured (Hammer, 1998). Improvements in computer speed and the use of computer clusters, as outlined here, are essential to process the large numbers of scenarios to be investigated. More importantly, it is necessary to continue the field experimentation to improve the ability of crop models to capture the interaction of crop traits with the environment.
While much basic research investment is in the cellular and molecular physiology of traits, the interaction effects and levels higher in the hierarchy of growth (organ, plant, and crop) greatly complicate the expression of these traits, and many complex crop-level traits are yet to be dissected. In most situations, experimental evaluation of conventional and molecular breeding strategies for manipulating complex yield adaptation traits will be impractical or beyond the resource base of breeding programs. Therefore, simulation tools will have an important role in the design and testing of breeding strategies. This role will become increasingly feasible as our understanding of the genetic architecture of quantitative traits improves. A critical step in developing these simulation tools is to establish the link between gene- and genome-based information and the biophysical processes that determine plant growth and development and adaptation to the biotic and abiotic stresses. In this paper, we have demonstrated the linkages that have been achieved between the QU-GENE and APSIM-Sorg modeling platforms to begin to address the issue of the complexity of interactions among gene, trait, and environment effects.
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONSimulation of Crop Response...Improving the Efficiency of...MATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES</ |
|---|
