Field Crops Research 64 (1999) 153±176
Rainfed lowland rice breeding strategies for Northeast Thailand II. Comparison of intrastation and interstation selection M. Coopera,*, S. Rajatasereekulb, B. Somrithc, S. Sriwisutd, S. Immarke, C. Boonwitef, A. Suwanwongseg, S. Ruangsookh, P. Hanviriyapanti, P. Romyenj, P. Porn-uraisanitj, E. Skulkhub, S. Fukaia, J. Basnayakea, D.W. Podlicha a School of Land and Food, The University of Queensland, Brisbane, Queensland 4072, Australia Rice Research Institute, Thai Department of Agriculture, Chum Phae Rice Experiment Station, Thailand c Rice Research Institute, Thai Department of Agriculture, Thailand d Rice Research Institute, Thai Department of Agriculture, Surin Rice Experiment Station, Thailand e Rice Research Institute, Thai Department of Agriculture, Phitsanulok Rice Research Center, Thailand f Rice Research Institute, Thai Department of Agriculture, Ubon Rice Research Center, Thailand g Rice Research Institute, Thai Department of Agriculture, Khon Kaen Rice Experiment Station, Thailand h Rice Research Institute, Thai Department of Agriculture, Sakon Nakhon Rice Research Center, Thailand i Rice Research Institute, Thai Department of Agriculture, Sanpatong Rice Experiment Station, Thailand j Rice Research Institute, Thai Department of Agriculture, Phimai Rice Experiment Station, Thailand
b
Accepted 3 September 1999
Abstract There has been limited progress for grain yield of rainfed lowland rice in Northeast Thailand since the 1960s. The current breeding strategy operates as a series of six semi-independent pedigree programs, each at a different site. Each program has three major phases of selection: (1) intrastation selection, (2) interstation selection, and (3) on-farm selection. The expected selection response for grain yield based on intrastation and interstation selection was examined using a combination of experimental results, prediction equation theory and computer simulation. Experiments were conducted to estimate genetic, genotype-by-environment interaction and error components of variance as inputs for estimation of heritability on a number of bases and also to obtain estimates of realised response from selection. Estimates of line-mean heritability for grain yield based on intrastation evaluation of lines suggest that it is low, ranging from 0.07 to 0.13, for one to four replicates, respectively, at a single site in 1 year. Line-mean heritability for intrastation evaluation based on two replicates and 2 years was estimated to be 0.18, only slightly higher than for 1 year and four replicates. In contrast, estimates of line-mean heritability for interstation testing were intermediate, e.g. 0.32 and 0.48 for two replicates at six sites for 1 year and 2 years, respectively. Estimates of realised selection response for grain yield from intrastation and interstation selection were consistent with the low to intermediate heritability estimates. Interstation selection, based on two replicates, eight sites and 1 year, showed an advantage over intrastation selection, based on two replicates and 1 year, when response was measured as the mean yield of selected lines across environments. The present breeding strategy applies intense selection during the intrastation phase of the breeding * Corresponding author. E-mail address:
[email protected] (M. Cooper)
0378-4290/99/$ ± see front matter # 1999 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 8 - 4 2 9 0 ( 9 9 ) 0 0 0 5 7 - X
154
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
programs. Consequently, only a small number of lines (ca. 70 lines from all stations) are advanced from the intrastation selection phase to the interstation selection phase. Therefore, for most lines generated by the breeding program there is limited opportunity to evaluate the contributions of broad and speci®c adaptation to higher yield. The presence of large genotype-byenvironment interactions, in combination with limited yield evaluation of lines in multi-environment trials (until the ®nal stages of testing), is identi®ed as a major factor contributing to the slow genetic progress for grain yield. The proposed breeding strategy replaces the intrastation testing phase with a coordinated early generation interstation testing based on F4 bulks. Evaluation of the proposed breeding strategy by computer simulation demonstrated an advantage from modifying the current breeding strategy to give greater emphasis to interstation selection in place of intrastation selection. # 1999 Elsevier Science B.V. All rights reserved. Keywords: Yield; Genotype-by-environment interactions; Drought; Breeding strategy; Simulation
1. Introduction There has been limited genetic progress for grain yield of rainfed lowland rice in the highly droughtprone farming system of Northeast Thailand. Studies based on the use of multi-environment trials (METs) commenced in 1992 to examine possible reasons for the slow progress for improvement of grain yield. The in¯uence of genotype-by-environment (G E) interactions was identi®ed as one of the key factors contributing to the slow rate of genetic improvement (Fukai et al., 1997; Fukai and Cooper, 1999; Cooper et al., 1999a). Large and complex G E interactions have been identi®ed for grain yield of rainfed lowland rice in Northeast Thailand (Henderson et al., 1996; Cooper and Somrith, 1997; Cooper et al., 1999a). These studies have been based on both selected lines (Henderson et al., 1996; Cooper and Somrith, 1997) and random breeding lines sampled from crosses relevant to the Thai breeding program (Cooper et al., 1999a). Comparable results were observed for both types of germplasm. For the study based on the random breeding lines, the genotype-by-site-by-year (G S Y) interaction was found to be the largest interaction component, and was estimated to be around four times the size of the genotypic component of variance. Signi®cant genotype-by-year (G Y) interactions were identi®ed, and these were comparable in size to the genotypic component of variance, but there was little evidence of important genotype-by-site (G S) interactions. Investigations of the causes of the G S Y interactions strongly indicate that a major source of these interactions can be attributed to genetic variation for
the control of ¯owering time in combination with environmental variability in the timing and intensity of water-de®cit (Cooper et al., 1999a). In Northeast Thailand there has been widespread use of cultivars with a strong photoperiod requirement for determination of the time of ¯owering. Most cultivars ¯ower towards the end of the main wet-season and are often subjected to drought periods during the reproductive and grain ®lling phases of development. Many of the crosses made in the Thai breeding program segregate for genes that in¯uence the regulation of plant development, in particular through the effects of the photoperiod sensitivity genes (Immark et al., 1997). The wide range of phenology types generated among these progeny lines in the breeding program creates a series of genotypes that can differentially interact with the range of drought-prone environments sampled across the region. This form of G E interaction introduces a strong in¯uence of drought escape on the relative yield of the genotypes in the breeding trials, which can mask differences in adaptation associated with drought resistance (tolerance and avoidance). The contribution of this mixture of drought escape and drought resistance to high yield in the breeding trials complicates the selection of genotypes with improved adaptation and grain yield for the drought-prone environments. The structure of the breeding program in Northeast Thailand is based around a series of six semi-independent breeding programs operating from the four Rice Research Stations and two Centers distributed across Northeast Thailand; Ubon Ratchathani (UBN), Sakon Nakhon (SKN), Chum Phae (CPA), Phimai (PMI), Surin (SRN) and Khon Kaen (KKN). The programs are semi-independent rather than indepen-
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
dent since they can, and do, interchange breeding lines for crossing as well as advanced stage testing. Each station operates a pedigree breeding strategy (Table 1). In each of the pedigree programs there are three major phases of selection: (1) intrastation selection, (2)
155
interstation selection, and (3) on-farm selection. Intrastation selection is the selection conducted within an individual station. Interstation selection is the selection based on the results of experiments conducted across the stations. On-farm selection, under
Table 1 A schematic outline of the current breeding strategy, including areas required in each yeara Year
Activity
Description of activity
Maximum area per stationb
1
A/B A/B//C or A/2*B (30±50 crosses/station) 4 stations
Cross
Glasshouse
F2 spaced plants (4000±7000/cross) 6 stations
Selection among and within crosses
2
3
F3 rows (200±400/cross) 6 stations
4
F4 rows (50±100/cross) 6 stations
5
F5 rows (150 rows) 6 stations
6
F6 (50±100 lines) 6 stations
7
F7 (44 lines) 6 stations
8±10
F8±F10 (a) (70 lines) 6 stations
(b) Fertiliser trial 8 lines 6-8 N levels 3 replicates 3 stations (c) Disease-insect trials 3 stations 11±13
a b
F11±F13 (6±8 lines) 18 sites (3 irrigated)
Intrastation testing activities are given in italics. 6.25 rais 1 ha.
(Off-season advance F1±F2)
Seedling screen for blast resistance. Discard crosses on short grain Selection among and within crosses and among and within rows on plant height, flowering and plant type. Target to select a total of 300 lines from all crosses Within selected rows bulk seed from plants with the same height and flowering time Observation Nursery Plots (5 m 4 rows, 25 cm 25 cm plant to plant spacing) Lines compared to checks for height, flowering, grain chalkiness Intrastation trial 1 site 4 replicates (irrigated) Lines stratified for maturity and endosperm type (glutinous or non-glutinous) Select for grain yield, chalkiness, chemical quality Pure seed increase of selected lines Interstation trial (a) 6 sites 4 replicates. Often 4 trials for early and medium maturity and for glutinous and non-glutinous lines (each about 20 lines) Selection for grain yield, quality, flowering, height, disease resistance, insect resistance (b) Fertiliser responsiveness of promising lines. Eight lines are selected from F8±F13 lines (c) Trials to test resistance to gall midge, brown plant hopper and others for F8±F13 lines On-farm trials 4 replicates 16 rows 20 hills Yield and quality Consider for release after year 12
6 raisb
9 rais 2 rais 0.4 rais 0.5 rais
1.5 rais
3.5 rais
4 rais 1.5 rais 1.5 rais 3 sites
156
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
farmer paddy conditions, is the ®nal stage of evaluation and selection and is conducted on those lines that were selected following intrastation and interstation testing. In the current system there is extensive use of the intrastation testing strategy, which is conducted from the F2 to F7 generations. The interstation selection is conducted from the F8 to F10 generations. The number of lines that enter the interstation phase is small relative to the number of lines generated from the crosses, with a total of approximately 70 lines from all stations entering the interstation testing phase. Only a small number of lines, approximately 6±8, enter the on-farm testing phase. Given the incidence of large G S Y interactions for grain yield (Cooper et al., 1999a), there is concern that a consequence of the strong emphasis by the breeding programs on intrastation testing is that there is a low likelihood of identifying lines with improved adaptation and higher grain yield for the Northeast region. The counter argument has been made that the semi-independent programs, operating at the six stations, can exploit speci®c adaptations to sub-regions within the Northeast. However, the lack of evidence for any repeatable G S interactions for grain yield, associated with the sub-regions the stations are intended to represent, suggests that this argument has no merit. Further, the strong in¯uence of the G S Y interactions suggest that intense selection for grain yield at a single site conducted across a limited sample of years is unlikely to reliably identify lines with high grain yield for the target population of environments in Northeast Thailand. KDML105 and RD6 are the two most popular rainfed lowland rice cultivars in Northeast Thailand, each occupying approximately 40% of the total area under rainfed lowland rice production (Center for Agricultural Information, 1998). Results from a number of studies indicate that lines with superior grain yields to those of KDML105 and RD6 have been developed (Henderson et al., 1996; Cooper and Somrith, 1997; Cooper et al., 1999a). The argument against the use of these high yielding lines as improved cultivars is largely based on the grounds of inferior quality relative to KDML105 and RD6. A common crossing strategy that has been used for Northeast Thailand has been to cross the high yielding lines with the cultivars KDML105 and RD6. However, the outcomes from applying this strategy generally sug-
gest that it is dif®cult to generate progeny from these types of crosses that will have higher grain yield than the two benchmark cultivars. This was observed in the study of seven rainfed lowland rice crosses by Cooper et al. (1999a). They also observed that a number of crosses based on non-traditional Thai lines generated an array of progeny that had higher grain yield than both KDML105 and RD6. Therefore, yield improvement of rainfed lowland rice for Northeast Thailand is possible but the challenge is to combine the improved grain yield with the desired grain physical and chemical quality attributes. Mackill et al. (1996) reported that the important quality attributes for rainfed lowland rice are simply inherited and can therefore be manipulated relatively easily in a breeding program. If this is the case for the important quality attributes for rainfed lowland rice in Northeast Thailand then it may be possible and easier to improve the quality of the high yielding rainfed lowland rice lines, than to improve the grain yield of the high quality cultivars KDML105 and RD6. These alternative breeding strategies warrant consideration, given the lack of signi®cant progress for yield since the release of KDML105 in 1959 (Somrith, 1997), together with the experimental evidence that it is possible to develop higher yielding lines and that many of the quality attributes are simply inherited. Fukai and Cooper (1999) proposed a modi®cation of the current breeding strategy used in Northeast Thailand that changed the balance of emphasis on intrastation, interstation and on-farm testing (Table 2). They proposed a reallocation of resources within the Northeast region to enable earlier multi-environment testing of a larger number of lines in the interstation and on-farm testing phases of the breeding program. Key components of their proposed modi®cation were: (1) intrastation selection in early generations (F2±F4) for days-to-¯ower, plant height and the long grain quality trait, (2) no intrastation testing for grain yield, (3) early generation interstation testing for grain yield based on F4 bulks, (4) early generation evaluation for quality attributes based on the high yielding F4 bulks identi®ed from interstation testing, (5) selection of F6:7 lines from F6 pedigree rows based on all of the available yield and quality results, (6) additional interstation evaluation of F6:7 bulked lines, and (7) more extensive use of on-farm trials and evaluation of a larger number of advanced breeding lines in the on-
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
157
Table 2 Proposed breeding strategy for rainfed lowland rice in Northeast Thailanda Year
Activity
Description of activity
Maximum area per stationb
1
A/B A/B//C or A/2*B (40 crosses) 2 stations
Crossing
Glasshouse
WSc: F2 (4000/cross) 2 stations
WSc: Seedling screen for blast resistance. Discard on short grain. Select 100 plants per cross on average. DSc: Rapid generation advance of 2500 plants per station. WS: Select rows on plant height and flowering time. Single plant selection (1000 total) from within selected rows. Bulk remainder of selected rows for yield test. DS: Rapid generation advance of 1000 plants per station Interstation trial (F4 bulks)
2
DSc: F3 (2500 plants) 2 stations 3
WS: F4 rows (2500 lines) 2 stations
DS: F5 (1000 plants) 2 stations 4
(a) F4 bulk yield test 1000 lines 1 replicate (4-row plots) 6 stations
(b) F6 rows (1000 lines from all crosses) 2 stations 5
(a) F4 bulk yield test 500 lines 2 replicates (4-row plot) 6 stations
(b) F7 rows (1000 rows total) 6 stations 6
7
8
F8 yield test 250 lines 2 replicates (4-row plot) 6 stations
F9: 125 lines 2 replicates (4-row plot) 6 stations
(a) F10: 50 lines 3 replicates 6 stations (b) F9±F10 Disease and insect resistance 2 stations
9±10
(a) F11±F12: 20 lines 4 replicates 18 sites (6-row plot)
Coordinate among stations (Off-season advance F1±F2)
(a) Selection for grain yield, quality, flowering, height, disease resistance, insect resistance. Use of the CPAd drought screening facilities (b) Select rows from F6 nursery and F4 bulk yield results. Chemical quality testing on selected lines. Interstation trial (F4 bulks)
6 raisb 2 stations Glasshouse 3 rais 2 stations
Glasshouse (a) 4.5 rais 6 stations
(b) 1 rais 2 stations (a) 4.5 rais 6 stations
(a) Selection for grain yield. Include CPAd Drought screening nursery as one site (b) Select rows from F7 nursery and F4 bulk yield results. Bulk selected rows Interstation trial (F7 bulks)
2.5 rais 6 stations
Selection for grain yield, quality, flowering, height, disease resistance, insect resistance Interstation trial
1 rais 6 stations
Selection for grain yield, quality, flowering, height, disease resistance, insect resistance Inter-station trial (a) Selection for grain yield, quality, flowering, height (b) Selection for resistance to gall midge, brown plant hopper and other insects and diseases On-farm trials
(b) 1.0 rais 6 stations
(a) 1 rais 6 stations (b) 1.5 rais 2 stations (a) 1.5 rais 3 sites 6 stations
158
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
Table 2 (Continued ) Year
Activity
(b) Fertiliser trials 8 lines 4 replicates 3 sites 11
F13: 6±8 lines 18 sites 4 replicates (16-row plot)
Description of activity
Maximum area per stationb
(a) Selection for grain yield, quality, flowering, height, disease resistance, insect resistance. Consider for release (b) Responsiveness of 8 selected lines for N fertilizer (from F11-F13 lines) On-farm trials
(b) 4 rais 3 sites
Consider for release
1.5 rais 3 sites 6 stations
a
Intrastation testing activities are given in italics (modified from Fukai and Cooper, 1999). 6.25 rais 1 ha. c WS Ð Wet season, DS Ð Dry season. d CPA Ð Chum Phae site in Northeast Thailand. b
farm trials. The evaluation of a larger number of lines for grain yield in coordinated METs, together with the availability of grain quality data on the lines, should increase the opportunity to identify lines with the desired combination of quality traits and higher grain yield than the cultivars KDML105 and RD6. The objective of this study was to examine factors that would in¯uence the expected improvement in response to selection for grain yield from a change in emphasis of testing and selection between the intrastation and interstation testing phases of the breeding program, as proposed by Fukai and Cooper (1999). 2. Material and methods 2.1. Experimental evaluation: line-mean heritability Cooper et al. (1999a) reported estimates of components of variance for grain yield from the analysis of a random sample of breeding lines that were derived from seven rainfed lowland rice crosses made for yield improvement in Thailand. These estimates were based on a MET conducted across 3 years, from 1995 to 1997. A total of eight sites were sampled, two of these were from North Thailand and six from the Northeast. The six Northeast sites were the two rice research centers and four rice experiment stations used for rainfed lowland rice breeding in Northeast Thailand, and discussed above (UBN, SKN, CPA, PMI, SRN
and KKN). The two sites form the North were Sanpatong (SPT) and Phitsanulok (PSL). The estimates of the components of variance from the analysis of this MET were used to examine the in¯uence of both intrastation and interstation evaluation of lines on the magnitude of line-mean heritability for grain yield. Since the focus of this study was on the intrastation and interstation testing strategies in Northeast Thailand, it may be considered appropriate to examine the components of variance for grain yield based on only the six sites from Northeast Thailand, excluding the two sites from North Thailand. To determine whether the inclusion of the two northern sites had a strong in¯uence on the estimates of components of variance for grain yield, analyses were conducted with (N NE) and without (NE) the northern sites (Table 3). The absolute values for the components of variance were generally smaller for the MET based on only the northeastern sites. However, the same general trend was observed for the relative sizes of the partitions of the genotypic and G E interaction components of variance. Signi®cant genotypic variance was detected for both MET data sets and the G S Y interaction component was signi®cant and large relative to the genotypic component of variance. There was no signi®cant G S interaction detected for either of the MET data sets. The one difference noted between the data sets was the lack of a signi®cant G Y interaction component for the MET based on only the northeastern sites, whereas a signi®cant G Y interaction component was detected when
M. Cooper et al. / Field Crops Research 64 (1999) 153±176 Table 3 Estimates of components of variance (standard errors) for grain yield (t haÿ1), based on 1116 rainfed lowland rice lines (checks and random progeny derived from seven crosses) evaluated in a multienvironment trial (MET) with a site-year cross classification structure for (a) combined analysis of six Northeast and two Northern sites (N NE), (b) analysis of six Northeast sites (NE) Source of variance
MET data set
G GS GY GSY Residual
N NE
NE
0.060 0.006 0.003 0.006 0.049 0.006 0.259 0.009 0.440 0.005
0.053 0.005 0.000 0.002 0.004 0.165 0.009 0.360 0.005
the northern sites were included. This difference may have occurred in part because of the limited sampling of the 1995 year when only the northeastern sites were considered; two of the three sites sampled in 1995 were from the northern region. The difference in the estimates of the components of variance between the two MET data sets were not considered suf®cient to remove the additional information associated with evaluation of the lines in the two northern sites. In addition, the studies by Henderson et al. (1996) and Cooper and Somrith (1997) suggested that many aspects relevant to the yield adaptation of rainfed lowland rice in Thailand were common to both North and Northeastern Thailand. Therefore, the components of variance based on the Northeastern and Northern sites (N NE, Table 3) were used to examine the in¯uence of MET structure on estimates of line-mean heritability. Line-mean heritability (h2LM ) was estimated as: h2LM
2g 2g
2gs s
2gy y
2gsy sy
2
syr"
;
(1)
where 2g is the genotypic component of variance, 2gs the G S interaction component of variance, 2gy the G Y interaction component of variance, 2gsy the G S Y interaction component of variance, 2" the residual (error) component of variance, and s, y, and r are the numbers of sites, years and replicates, respectively. The form of heritability given in Eq. (1) was considered appropriate for evaluating the procedures that are currently used for selection by the Thai rainfed lowland rice breeders.
159
There are a number of other bases for estimating heritability that are relevant for yield of rainfed lowland rice in Northeast Thailand. Cooper et al. (1999a) examined the partitioning of genotypic and G E interaction components of variance for grain yield, days-to-¯ower and plant height into among-cross and within-cross components. They observed that there were strong contributions from both the among-cross and within-cross components. Therefore, the amongcross and within-cross components of variance reported by Cooper et al. (1999a) were used to estimate heritability on a cross-mean basis (h2C ) and within-cross basis (h2W ). Cross-mean heritability was estimated as: h2C
2C
2Cs s
2C
2Cy y
2Csy sy
2
"0 Isyr
;
(2)
where the variance components 2C , 2Cs , 2Cy , 2Csy are the among-cross, among-cross by site interaction, among-cross by year interaction and among-cross by site by year interaction components of variance, respectively. The residual component of variance 2"0 is defined as the sum of the within-cross genetic and interaction components of variance and experimental error such that 2"0 2W 2Ws 2Wy 2Wsy 2" , I is the number of individuals sampled from a cross, s the number of sites, y the number of years and r the number of replicates. Within-cross heritability was estimated as: h2W
2W
2Ws s
2W
2Wy y
2Wsy sy
2
syr"
;
(3)
where the variance components and s, y and r were as defined above. Heritability was examined for a range of possible MET structures by substituting the components of variance obtained for yield, days-to-¯ower and plant height into Eqs. (1)±(3) and changing the number of sites, years and replicates. Varying the number of sites and years was used to represent different possible emphases on intrastation and interstation testing. Where the number of sites was set to be s 1 this was considered to provide estimates of expected heritability for intrastation testing as the number of years and replicates was varied. Where the number of sites was set to be s > 1 this provided estimates of
160
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
expected heritability for different forms of interstation testing. Biplots (Gabriel, 1971) were used to graphically display the G S Y interaction variation for grain yield on both among-cross and within-cross bases. For the combined analysis of variance that was used to estimate the components of variance for the amongcross and within-cross partition of variance, Best Linear Unbiased Predictors (BLUPs) were computed for the grain yield G S Y interaction effects of the seven crosses in each environment, and for the individual lines within each cross (Cooper et al., 1999a). Principal components analysis, using the singular value decomposition algorithm, was applied independently to the cross-mean BLUP by environment yield data and each of the line-within-cross BLUP by environment yield data sets. The biplots were constructed and interpreted following environment-standardisation (Fox and Rosielle, 1982) of the yield BLUPs using the same procedures as were applied by Cooper et al. (1999a). 2.2. Experimental evaluation: genetic components of variance The ease by which progress from selection can be achieved is in¯uenced by the relative contributions of additive and non-additive sources of genetic variation to the genotypic and G E interaction components of variance. Using the Random Homozygous Line (RHL) mating-design (Hanson and Weber, 1961) available for the rainfed-lowland rice experiments reported by Cooper et al. (1999a), the genotypic variance for grain yield, days-to-¯ower and plant height was partitioned into additive (2A ) and additive-by-additive (2AA ) epistatic components of variance. Similarly, the G E interaction was partitioned into additive-by-environment (2AE ) and epistasis-byenvironment (2AAE ) interaction components of variance. The number of F2 families available for the genetic analysis ranged from 30 for Cross 3 to 91 for Crosses 2 and 5 (Cooper et al., 1999a, Table 1). Only the data from 1995 and 1996 experiments were used for the genetic analysis for each cross. The additive and additive-by-additive epistatic genetic components of variance were expressed relative to the genetic variance in the F2 generation, following the de®nitions and notation used by Hanson and Weber (1961). The
F2 and F3-derived family structure enabled a partition of variation among the F7 lines into among-family and within-family components for each cross. The among F2 (2F2 ) and among F3-derived lines within F2 family (2F3=F2 ) variance components were equated to their genetic expectations to estimate the additive and additive-by-additive epistatic components of genetic variance; 2F2 2A 2AA and 2F3= F2 2A 32AA . In the same way the additive-by-environment and epistasis-by-environment interaction components of variance were estimated from the G E interaction components based on the F2 families (2F2E ) and F3-derived lines within F2 families (2F3=F2E ); 2F2E 2AE 2AAE and 2F3=F2E 2AE 32AAE . 2.3. Experimental evaluation: realised response to selection An experimental evaluation was conducted to estimate the realised response to selection for grain yield from applying the intrastation and interstation selection strategies. Selections based on both strategies were made in 1997 and evaluated in 1998. The experiments on which the selections were based were those described by Cooper et al. (1999a). Grain yield data from the eight research stations (six from the Northeast CPA, KKN, PMI, SKN, SRN and UBN, and two from the North PSL, SPT) were available for 463 random inbred lines. These 463 random lines were sampled from the seven rainfed lowland rice crosses, as described by Cooper et al. (1999a). At each station the scientists responsible for conducting the experiment implemented intrastation selection using the data they collected in 1997. They identi®ed their preferred ten lines, based on a combination of grain yield at the station and their own evaluations of the suitability of the lines. While each breeder applied different selection criteria, the bases of selection were consistent with the common practices of the breeders for intrastation selection at each station. In parallel, interstation selection was implemented by conducting a combined analysis of the 1997 grain yield data across the eight stations. BLUPs were computed for the grain yield of the lines across the stations. The ten lines with the highest grain yield BLUPs from the combined analysis were selected. The mean grain yield of the ten lines selected by the intrastation and interstation selection
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
strategies in 1997 were compared for average grain yield across the eight stations in 1998 and also for grain yield at each station in 1998. A total of 120 lines, including checks, were included in the 1998 experiments. These were made up of the ®ve check cultivars KDML105, RD6, RD23, NSG19 and Chiangsaen, the high yielding check line IR57514-PMI-5-B-1-2, the parents for the seven biparental crosses studied, and 108 lines selected by intrastation and interstation strategies applied to the progeny of the seven crosses. Among the 108 selected lines were some lines selected using the results of 1995 and 1996 experiments. Only the results of the intrastation and interstation selection strategies based on the 1997 experiments are considered in this paper. A randomised complete block design with three replicates was used at all stations in 1998. The trials were established by transplanting seedlings following seed germination in paddy nurseries. The experimental plots were six rows by 5 m with 25 cm between hills and three seedlings were established per hill. A constant application of fertiliser was used for all trials in 1998, which was the same as that used for the selection trials in 1997. There was a basal application of 18.75 kg N haÿ1, 37.5 kg P2O5 haÿ1 and 37.5 kg K2O haÿ1 and a top dressing of 18.75 kg N haÿ1 prior to ¯owering. Grain yield (14% moisture content) was measured at maturity for each plot. The middle four rows of the plot were harvested to estimate grain yield. 2.4. Simulation evaluation: response to selection To complement the experimental studies, computer simulation methodology was used to further evaluate the intrastation and interstation selection strategies by comparing response to selection for the current (Table 1) and proposed (Table 2) breeding strategies. The QU-GENE simulation software was used to conduct the simulation experiment (Podlich and Cooper, 1998). A module (QU-RLR) was developed to represent the two breeding strategies summarised in Tables 1 and 2. Fig. 1 shows the structure of the QU-RLR module. The simulation of the two breeding programs and the sources of information used to construct the genotype-environment system models and their application in the simulation experiment are summarised below.
161
For the current breeding program (Table 1), 40 biparental (A/B) crosses were conducted at each of the six stations (Fig. 1). Within each station, 100 F2 plants were sampled at random from the 40 crosses and progressed by sel®ng to the F6 generation. Intrastation selection of the top ten lines was conducted at the F7 generation for each station. The 60 selected lines (ten lines six stations) were evaluated in 2 years of interstation testing with the top 30 lines selected on performance across the six stations. The 30 selected lines were evaluated in another interstation trial with the top ten lines selected on performance across the six stations. The mean of the top ten lines was used to evaluate response to selection for the current breeding program. For the simulation of the proposed breeding program (Table 2), 240 biparental (A/B) crosses were conducted across the six stations (Fig. 1). A total of 1000 F2 plants was sampled at random from the 240 crosses. 1000 F4 progeny rows were derived by selfpollination from each F2 individual. A single random plant from each F4 progeny row was continued by selfpollination to the F6 generation and F4 bulks were derived from the remaining plants in the F4 progeny rows. This generated a pedigree structure connecting each of the 1000 F4 bulks to one of the 1000 F6 progeny rows. Each of the related pairs of F4 bulks and F6 rows were derived from different individual F2 plants. An interstation trial was conducted on the 1000 F4 bulks, with 2 years of F4 bulk testing. The 1000 F4 bulks were evaluated in the ®rst year. After the ®rst year of testing, 500 F4 bulks were selected based on performance across the six stations. After the second year of testing, 60 F4 bulks were selected based on performance across the six stations. Using F6 individuals derived from the same F4 progeny row that was used to derive the selected F4 bulks, 60 F6:7 bulks, representing the selected F4 bulks, were constructed. The 60 F6:7 bulks were evaluated in 2 years of interstation testing with the top 30 lines selected on performance across the six stations. The 30 selected lines were evaluated in another interstation trial with the top ten lines selected on performance across the six stations. The mean of the top ten lines selected by each strategy after one cycle of the breeding program was used to compare the response to selection for the current (Table 1) and proposed breeding strategies (Table 2). Response to selection for the two breeding
162
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
Fig. 1. Schematic representation of the major components of the current (Table 1) and proposed (Table 2) rainfed lowland rice breeding strategies as they were simulated in the QU-RLR simulation program.
strategies was considered for four genetic models with different levels of G E interaction. The results of experimental investigations (Henderson et al., 1996; Cooper and Somrith, 1997; Cooper et al., 1999a) were used to guide the selection of appropriate genetic models. The ratios of the G E interaction component of variance relative to the genotypic component of variance for the four models were 0.5, 1.3, 1.7 and 3.8. Therefore, the models ranged from relatively simple, where G E interaction variance was less than the genotypic component of variance, to complex, where G E interaction was larger than the genotypic component. For each model the gene action was speci®ed to be a combination of additive and additive-by-environment types of G E interactions. The G E interactions were all generated as crossover interactions, where the alternative alleles of a gene were considered to confer an advantage to different environment types. Cooper (1999) discussed this form of crossover interaction in
more detail and considered its relevance to rainfed lowland rice. It is hypothesised to be relevant to the common types of crossover interactions observed for grain yield of rainfed lowland rice that are associated with genetic variation for days-to-¯ower and environmental variation for timing of water-de®cit (Cooper et al., 1999a). Three levels of heritability (H 0.05, 0.50 and 1.00 on a single plant basis in a single environment) were also examined to vary the size of the experimental error component of variance. Response to selection for the two breeding strategies, for each of the four genetic models, was calculated as a percentage of the target genotype, where the target genotype was de®ned as that genotype which resulted in the highest possible trait value (Podlich and Cooper, 1998). For each combination of the 24 treatment levels considered (two breeding strategies by four genetic models and three heritability levels) 100 replications of the simulation experiment were conducted. The variability among the 100 replications
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
was used to compute a standard error of the mean response to selection. 3. Results 3.1. Experimental conditions and environmental characterisation The seeding and transplanting dates of the experiments differed among the site-year combinations (Table 4). There was a higher frequency of trials seeded late in 1998 than in 1997. In a number of the later seeded trials the photoperiod insensitive check cultivar RD23 ¯owered later than some of the photoperiod sensitive checks (e.g. KKN, 1998). The incidence of drought and the effects of waterstress on mean plant height and grain yield of the check lines were highly variable among the environments. For some environments there was little evidence of strong drought effects on grain yield (e.g. SPT, 1997 and 1998, PMI, 1997 and SRN, 1997). When drought effects were detected, terminal drought, which tended to have a more severe negative effect on the grain yield of later ¯owering lines, was most frequently observed with intermittent or severe drought throughout the season less common. At some sites the pattern of occurrence of water-de®cit and the effects of drought on mean grain yield appeared to be consistent across 1997 and 1998 (e.g. CPA and KKN). Whereas at other sites there was no apparent consistency of drought occurrence or effects on grain yield across years (e.g. PSL). 3.2. Experimental evaluation: line-mean heritability Using the estimates of the components of variance for grain yield from Table 3 (N NE), the predicted line-mean heritability increased with an increase in the number of sites, years and replicates (Fig. 2). For the combinations of sites, years and replicates examined, there was evidence of a diminishing increase in the line-mean heritability as the level of each of these variables in the MET was increased. Line-mean heritability for a range of intrastation selection strategies was examined by setting the number of sites to 1 (s 1) in Eq. (1) and varying the number of replicates and years (Fig. 2a). The intrasta-
163
tion testing strategy that is used in the current breeding strategy is based on 1 year of testing at an individual site (Table 1). Line-mean heritability for intrastation evaluation based on 1 year of testing was predicted to be low regardless of the number of replicates, ranging from 0.07 to 0.13 for 1±4 replicates, respectively (Fig. 2a). These low heritability values indicate that improvement in grain yield will be slow for intrastation selection. The intrastation line-mean heritability increased as the number of years was increased from 1 to 4. The increase in heritability with additional levels of replication, from one to four replicates, was greater as the number of years increased. However, the advantage from more than two replicates was small. Interstation testing (s > 1) within a year increased line-mean heritability relative to that achieved for intrastation testing in a year (Fig. 2b). There was again only a small increase in line-mean heritability predicted from increasing replication beyond two replicates. For interstation testing based on two replicates, the line-mean heritability ranged from 0.17 to 0.32 for a MET based on 2±6 stations, respectively. The predicted values for line-mean heritability suggest that within a year it is possible to achieve an increase in line-mean heritability from the low value of 0.13, predicted for the current intrastation testing strategy based on four replicates at one site, to 0.21 by conducting interstation testing based on a single replicate at four sites (Fig. 2b). This interstation testing strategy would not require any additional seed relative to that required for the current intrastation testing strategy. Increasing the number of sites beyond four would require additional seed, but would result in a further increase in heritability (Fig. 2b). For yield evaluation based on two replicates, increasing the number of years was predicted to contribute to a substantial increase in line-mean heritability (Fig. 2c). For interstation testing based on two sites and 2 years the line-mean heritability was predicted to be 0.29. Increasing the number of sites to six, in combination with 2 years, gave a predicted heritability of 0.48, which is higher than that predicted for six sites and 1 year of testing (0.32), and substantially higher than the current intrastation testing strategy (0.13). Increasing the number of years of testing from 2 to 3, in combination with six sites, gave a predicted increase in heritability from 0.48 to 0.58. For interstation testing based on six sites and 2 years there was
164 Table 4 Experimental conditions and characterisation of the rainfed lowland rice environments sampled in the multi-environment trials conducted in 1997 and 1998; seeding date (SD), transplanting date (TD), mean grain yield and plant height of the six check lines and days-to-flower for each check linea Year
Site
SDb
TDb
Days-to-flower
KDc (SS)d
RD6c (SS)d
NSc (MS)d
IRc (MS)d
Yield (t haÿ1)
Drought stress characterisation: Description of how drought and plant stress symptoms coincided with the timing of flowering of the experimental lines
Severe terminal drought affecting late flowering lines Post-flowering drought: low-stress for early flowering and mild-stress for late flowering lines Post-flowering drought with some early flowering stress Severe terminal drought affecting all lines and greater affect on late flowering lines Mild terminal drought affecting late flowering lines No drought: mild-stress for early flowering stress Post-flowering drought: low-stress for early and late flowering lines Severe intermittent drought: early flowering stress Moderate early stress Low-stress Post-flowering drought Moderate terminal drought with a severe affect on late flowering lines Intermittent drought affecting early and late flowering lines Mild intermittent drought at flowering Post-flowering drought Severe drought throughout season
RD23c (IS)d
1997 1997
PSL SPT
20-7 25-7
16-8 19-8
103 101
99 97
99 98
87 84
104 99
93 93
147 132
2.48 3.66
1997
UBN
2-7
6-8
119
115
116
105
112
102
127
1.85
1997
CPA
7-7
31-7
109
107
108
93
98
89
117
2.14
1997 1997 1997
SKN PMI SRN
30-6 26-6 2-7
5-8 29-7 31-7
118 112 116
115 109 109
114 111 112
103 103 97
112 110 105
99 99 95
122 126 139
2.95 3.59 3.52
1997 1998 1998 1998 1998
KKN PSL SPT UBN CPA
20-6 17-7 22-7 7-7 3-8
23-7 16-8 22-8 6-8 3-9
129 120 104 117 122
130 116 99 108 102
131 118 102 111 109
117 104 89 93 101
127 119 103 107 95
119 105 97 92 102
125 135 123 117 130
2.61 2.61 3.33 1.38 3.00
1998
SKN
26-7
30-8
117
102
107
102
96
109
115
2.13
1998 1998 1998
PMI SRN KKN
23-7 6-7 21-7
31-8 14-8 31-8
101 109 107
94 105 104
96 108 106
88 94 101
107 107 113
95 93 111
103 105 102
2.16 2.01 0.89
a Drought stress characterisation of each environment was based on a combination genotype response patterns, the incidence of drought stress symptoms, together with temporal patterns of rainfall and measures of standing soil water when available (data not presented). b Environments (site-year combinations) highlighted in italics indicate incidences of late seeding. c Six check lines: CS Ð Chiangsaen, KD Ð KDML105, RD6 Ð RD6, NS Ð NSG19, IR Ð IR57514-PMI-5-B-1-2, RD23 Ð RD23. d Photoperiod sensitivity characterisation of the check lines: SS Ð Strongly sensitive, MS Ð Moderately sensitive, IS Ð Insensitive.
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
CSc (SS)d
Height (cm)
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
165
Fig. 2. Line-mean heritability for grain yield of rainfed lowland rice in Northeast Thailand estimated using Eq. (1) and the components of variance presented in Table 3: (a) intrastation evaluation where the number of sites s 1, (b) interstation evaluation where the number of years y 1, (c) interstation evaluation where the number of replicates r 2, (d) interstation evaluation where the number of years y 2.
little advantage detected in line-mean heritability from increasing the level of replication beyond two replicates (Fig. 2d). 3.3. Experimental evaluation: among-cross and within-cross heritability Using the estimates of among-cross and withincross components of variance reported by Cooper et
al. (1999a), heritability was estimated on both amongcross and within-cross bases for combinations of sites and years (Fig. 3). Among-cross heritability (assuming a sample of 100 lines per cross) was low to intermediate for grain yield (Fig. 3a) and was consistently high for days-to-¯ower (Fig. 3c) and plant height (Fig. 3e). For each trait, the estimates of among-cross heritability were always higher than the respective estimates of within-cross heritability
166
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
Fig. 3. Predicted heritability on cross-mean (a,c,e) and line-mean (b,d,f) bases for grain yield (a,b), days-to-flower (c,d), and plant height (e,f) as the number of sites and years of testing in multi-environment trials are changed. Assumes two replicates for each site-year combination and 100 lines are evaluated for each cross. Based on the components of variance for the among-cross and within-cross partition of variation given in Table 5 of Cooper et al. (1999a).
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
(Fig. 3a, cf. Fig. 3b and c, cf. Fig. 3d and e, cf. Fig. 3f). However, the relative magnitudes of within-cross heritability were similar to those for among-cross heritability, with that for grain yield low to intermediate (Fig. 3b) and for days-to-¯ower (Fig. 3d) and plant height (Fig. 3e) higher than for yield. Among-cross heritability for grain yield increased with number of sites, with some evidence for a diminishing rate of increase for more than ®ve sites (Fig. 3a). There was a substantial increase in the crossmean heritability associated with an increase in the number of years, but particularly for 1±2 years. Similar trends were observed for the within-cross heritability (Fig. 3b). Given the higher heritability values for days-to-¯ower (Fig. 3c and d) and plant height (Fig. 3e and f), in comparison to grain yield (Fig. 3a and b), a MET structure that was optimised for selection on grain yield (e.g. ®ve sites and 2 years) would enable a reliable characterisation of crosses and lines on variation for the traits days-to-¯ower and plant height. Biplots were constructed to display the amongcross (Fig. 4a) and within-cross (Fig. 4b±h) G S Y interaction variation for grain yield. For the among-cross BLUPs 64% (Component 1 43% and Component 2 21%) of the total standardised G S Y interaction variation was retained in the ®rst two principal components (Fig. 4a). The taller, later ¯owering Crosses 4, 6 and 7 were contrasted with the shorter, earlier ¯owering Crosses 1±3 and 5 on Component 1. On Component 1 there was also a contrast between environments where there was limited or no occurrence of drought, which tended to have positive scores on Component 1, and those environments where the characterisation indicated the occurrence of drought, which tended to have negative scores on Component 1. The contrasts on Component 2 were associated with different ways in which drought coincided with the average timing of ¯owering of the crosses. For example, at KKN97 (which had a high positive score on Component 2) there was severe intermittent drought, with severe effects of waterstress throughout ¯owering. Whereas at SPT95 and SPT97 (which had a high negative score on Component 2) there was an indication of post-¯owering drought. The taller, later ¯owering crosses tended to have higher grain yield in those environments where there was a lower incidence of drought, while the
167
shorter, earlier ¯owering crosses had higher yield in those environments where some form of drought occurred. For the line-mean G S Y interaction BLUPs within each cross (Fig. 4b±h), the representation of the variation for yield differed among the crosses, ranging from 24% for Cross 6 (Component 1 14% and Component 2 10%; Fig. 4g) to 39% for Cross 4 (Component 1 23% and Component 2 16%; Fig. 4e). The relationships between the environments, as depicted by the angles between the environment vectors on the biplots, differed among the crosses (Fig. 4b±h). This was to a large extent in¯uenced by the variation for days-to-¯ower both among and within the crosses. For those crosses that were either on an average later ¯owering, i.e. Crosses 4 (Fig. 4e), 6 (Fig. 4g) and 7 (Fig. 4h), or had a large proportion of late ¯owering lines, i.e. Cross 3 (Fig. 4d), there was an association (relatively small angles between the vectors) between environments where terminal drought had a signi®cant negative effect on grain yield, particularly CPA97 and PSL97. Whereas for the earlier ¯owering crosses, i.e. Cross 1 (Fig. 4b), Cross 2 (Fig. 4c) and Cross 5 (Fig. 4f), there was less impact of terminal drought on the relative yield of the lines, and less tendency for the environments where terminal drought was detected to be associated (i.e. angles between the vectors approaching 908). The biplots for the individual crosses can be used to identify particular lines of interest from within promising crosses. For example, Cross 3 was identi®ed to have higher mean grain yield than the other crosses (Cooper et al., 1999a). The cross-mean G S Y interaction BLUPs for Cross 3 across the individual environments place it relatively close to the origin on the among-cross biplot (Fig. 4a). Therefore, given the broad adaptation of this cross and the structure of this particular biplot this cross was on an average less likely to have low grain yield in individual environments. This was in part because there was a wide range in the days-to-¯ower of the individual lines within this cross and there were early ¯owering lines that had high yield when there was terminal drought as well as later ¯owering lines with intermediate height that had high yield within those environments where there was little or no drought.
168
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
Fig. 4. Biplots for Components 1 and 2 for the ordination of grain yield Genotype-by Site-by-Year interaction BLUPs of rainfed lowland rice genotypes derived from seven crosses and evaluated in 18 environments evaluated on an (a) among-cross basis and (b±h) within-cross basis.
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
169
Table 5 Estimates of components of variance (standard errors) for the analysis of variance of grain yield (GY) (t haÿ1), days-to-flower (FL) (days) and plant height (HT) (cm) measured on seven rainfed lowland rice crosses, partitioned into additive (2A ), additive-by-additive epistasis (2AA ), additive-by-environment interaction (2AE ) and epistasis-by-environment interaction (2AAE ) components for the multi-environment trial without considering the site-year cross classification structure Cross parents
Trait
Source 2A
1. IR66321 IR43506-UBN-520-1-3-1-1/IR43342-10-1-1-3-3 2. IR66322 IR43506-UBN-520-1-3-1-1/IR49804-UBN-7-B-1-4-1 3. IR66327 IR46331-PMI-32-2-1-1/IR53466-B-118-B-B-20 4. IR66331 IR46331-PMI-32-2-1-1/NR15013-40-10-7 5. IR66364 KDML105/IR51952-B-12-1-1-1 6. IR66368 RD6/IR46331-PMI-32-2-1-1 7. IR66369 RD6/IR49804-UBN-7-B-1-4-1 a
GY FL HT GY FL HT GY FL HT GY FL HT GY FL HT GY FL HT GY FL HT
a
NS NS 116.4 14.6 0.022 0.021 6.6 2.4 65.7 33.0 NS 14.9 12.4 42.7 15.1 NS 28.6 7.6 64.9 41.7 0.022 0.014 12.9 3.4 34.1 5.5 0.025 0.012 2.0 1.2 203.0 30.3 NS 6.9 2.0 3.2 4.1
2AA
2AE
2AAE
0.047 0.014 5.9 1.4 NS 0.024 0.012 2.5 1.3 49.4 18.4 0.021 0.022 6.2 6.9 NS 0.014 0.016 NS NS 0.009 0.008 1.3 1.8 NS NS 1.4 0.7 NS 0.015 0.007 0.8 1.1 5.8 2.4
NS NS NS NS NS NS 0.111 0.050 NS 18.2 4.9 0.105 0.052 5.6 4.5 NS 0.025 0.019 NS NS NS NS NS NS 1.1 0.6 NS
0.072 0.021 1.9 0.8 14.0 3.6 0.097 0.014 2.1 0.5 19.4 4.9 0.049 0.029 14.1 2.5 NS 0.032 0.031 6.5 2.7 71.3 15.6 0.018 0.013 3.1 1.0 9.1 2.7 0.137 0.018 4.2 0.5 81.3 8.7 0.093 0.012 0.9 0.3 4.2 1.8
NS Ð not significant (components of variance estimated to be negative or positive values small relative to their standard error).
3.4. Experimental evaluation: genetic components of variance A number of the genetic and G E interaction components of variance were estimated to be negative or small positive values relative to their standard errors and these were assumed to be non-signi®cant sources of variance (Table 5). The genetic components of variance detected as signi®cant, and their relative sizes, differed for the three traits among the seven crosses. The additive and epistasis-by-environment interaction components were the most commonly detected sources of genetic variance. No additive genetic variance was detected for grain yield for four of the seven crosses (Crosses 1, 3, 4 and 7). For these crosses additive-by-additive epistasis sources of genetic variance were detected for grain yield. The additive genetic variance for plant height was large for Crosses 1 and 6, intermediate for Crosses 2±4 and 5, and was small for Cross 7. Additive-by-
additive epistasis for plant height was only detected for Crosses 2 and 7. The additive component of variance for days-to-¯ower was large for Crosses 3±5, small for Crosses 2, 6 and 7, and non-signi®cant for Cross 1. Signi®cant additive-by-additive epistasis was detected for days-to-¯ower in Cross 1, where no additive variance was detected. For Crosses 2, 3, 5 and 7 there was some evidence for additive-by-additive epistasis for days-to-¯ower, however this component of variance was always smaller than the additive component. The additive-by-environment interaction component of variance was an important source of genetic variance for grain yield in Crosses 3±5. It was also detected as a small source of variance for days-to¯ower in Crosses 4 and 7, and for plant height in Cross 3. A signi®cant epistasis-by-environment interaction component of variance was detected for grain yield and days-to-¯ower in all crosses and for plant height in six of the crosses. For days-to-¯ower and plant height the epistasis-by-environment interaction component
170
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
was generally smaller than the other sources of variance detected for the cross. However, in some cases it accounted for a relatively large proportion of the genetic variance, e.g. for days-to-¯ower Crosses 3 and 6, and for plant height Cross 4. For grain yield the epistasis-by-environment interaction component of variance was a relatively large source of variance for Crosses 1, 2, 6 and 7. 3.5. Experimental evaluation: realised response to selection Based on the evaluation of the 463 random breeding lines in 1997, different sets of lines were selected by the interstation and intrastation selection strategies. No two stations selected the same set of ten lines when intrastation selection was applied independently by those stations. The maximum number of lines selected in common by any two stations was one, and the maximum number of times any one line was picked by intrastation selection was three. No individual station, applying intrastation selection, selected the same set of ten lines that was identi®ed by interstation selection. Five of the ten lines selected by interstation selection were not selected by intrastation selection. For the other ®ve lines detected by interstation selection, two were also selected by intrastation selection by only one station, two were selected by two stations and one line was selected by three stations. The different sets of lines that were identi®ed by the eight intrastation selection strategies and the interstation selection resulted in different mean grain yields for the selected sets of lines across the eight sites in 1998 (Fig. 5a). The interstation selection resulted in a comparable mean yield to that achieved by intrastation selection at UBN and CPA and a higher mean yield across stations than that achieved by intrastation selection for ®ve of the eight stations (SRN, SPT, KKN, PMI and PSL). Intrastation selection conducted at SKN in 1997 identi®ed the highest yielding set of lines in 1998, which was higher yielding than the set of lines identi®ed by interstation selection. The mean grain yield of the set of ten lines identi®ed by interstation selection was compared to that of the sets selected by intrastation method at the same site used to select the lines in the previous year (Fig. 5b). There was no strong evidence of speci®c adaptations to individual sites that resulted in a substantially
greater response to selection from intrastation selection at a site relative to that achieved by interstation selection. Trends in the mean yields of the sets of lines suggested that interstation selection had some advantage over intrastation selection at three sites (UBN, SPT and PSL), no advantage at three sites (SRN, KKN, and PMI) and a slight disadvantage at two sites (SKN and CPA). However, the standard errors of these differences suggested that the trends should be treated with some degree of caution. 3.6. Simulation evaluation: response to selection The genetic models that were implemented in the simulation experiment were relatively complex and the target genotype was not selected after one cycle of the breeding program for any of the combinations of breeding strategy, genetic model and heritability examined (Fig. 6). However, for all 12 combinations of genetic model and heritability, the proposed breeding strategy resulted in a greater simulated response to selection than the current breeding strategy and therefore moved the selected genotypes closer to the target genotype (Fig. 6). As expected, progress from selection for both the current and modi®ed selection strategies was generally higher as the heritability was increased from 0.05 (Fig. 6a) to 0.50 (Fig. 6b) and to 1.00 (Fig. 6c). Also as the complexity of the form of the G E interaction was increased from G E interaction variance that was 0.5 times that of the genotypic variance to 3.8 times the genotypic variance, there was a reduction in the response to selection. For the three levels of heritability, the general reduction in selection response, which was observed as the level of G E interaction increased, was greater for the current strategy in comparison to the proposed strategy. Therefore, the advantage of the proposed strategy over the current strategy increased as the size and complexity of the G E interaction variance increased. The advantage of the proposed strategy was a consequence of its greater use of multi-environment testing to determine the value of the genotypes. 4. Discussion Large G E interactions and experimental error for grain yield of rainfed lowland rice in Northeast Thai-
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
171
Fig. 5. Mean grain yield of 10 rainfed lowland rice lines selected by interstation and intrastation selection strategies on the results of experiments in 1997 and evaluated in 1998: (a) mean yield across eight sites in 1998 for lines identified by interstation selection on mean yield across sites in 1997 (inter) and intrastation selection at individual sites in 1997 (SKN, UBN, CPA, SRN, SPT, KKN, PMI, and PSL), (b) mean yield at individual sites in 1998 for sets of lines identified by interstation selection on mean yield across sites in 1997 and sets of lines identified by intrastation selection at the individual site in 1997. Standard errors of the differences between the means of the sets of lines shown.
land, both contribute to the low to intermediate linemean heritability for grain yield. This study has focused on strategies to accommodate the effects of the G E interactions in selection for broad adaptation and investigation of the possible reasons for speci®c adaptation. The presence of large G E
interactions, in particular G S Y interactions, strongly suggests that extensive testing of breeding lines in METs will be required to adequately evaluate their grain yield potential and performance stability for the complex drought-prone target population of environments in Northeast Thailand. The structure of
172
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
Fig. 6. Simulated response to selection expressed as the mean performance of the 10 selected lines relative to the target genotype for the current (Table 1) and proposed (Table 2) breeding strategies (see Fig. 1) for four genotype-environment system models and three levels of broad sense heritability (H) on an individual site basis in the base population: (a) H 0.05, (b) H 0.50, (c) H 1.00.
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
the rainfed lowland rice breeding program currently used for Northeast Thailand was examined to evaluate its capacity to deal with the complicating effects of the G E interactions for grain yield. The current breeding strategy emphasises intense intrastation selection based on quality traits and visual selection for plant type, prior to any interstation testing for yield performance of the lines. For the six pedigree breeding programs, each operating at a single station in Northeast Thailand, usually less than 50 lines are evaluated each year in the intrastation testing phase for grain yield performance. Of these 50 lines, only a small number are entered into the interstation testing phase. The line-mean heritability for intrastation testing in a single year was low (h2LM 0.13 for 1 year and four replicates). Interstation testing based on six sites and 2 years, with two replicates for a site-year combination was required to raise the line-mean heritability to an intermediate level (h2LM 0.48). With the present breeding strategy there is currently no attempt to coordinate the analysis and interpretation of the results of the interstation trials in order to facilitate assessment of the broad and speci®c adaptation of the breeding lines or compare their G E interaction patterns of performance relative to those of the check cultivars. There is scope to implement selection on an amongcross basis prior to selection on a line-mean basis within crosses. For grain yield, the estimates of among-cross heritability were higher than those for within-cross line-mean heritability. The among-cross and within-cross variation for grain yield was examined by inspecting the BLUPs for cross-means and line-means within the crosses. To facilitate inspection of the G E interaction variation for grain yield, biplots were used to summarise the G S Y interaction BLUPs. This provided a useful graphical display of the results of the combined analysis of this relatively large MET and could be used as a basis for implementing an among-cross and within-cross selection strategy. Cross 3 was highlighted as a potential source of high yielding lines with superior yield to that of the popular Thai cultivars KDML105 and RD6. This particular cross is based on two semi-dwarf parents, with one parent photoperiod sensitive and the other insensitive. Among the progeny lines there were a number of early and late ¯owering lines with high yield. Therefore, the high yield of the lines was
173
not solely a result of drought escape under terminal drought conditions. There was transgressive segregation for plant height in this cross. The high yielding progeny lines generally had intermediate height between the semi-dwarf type represented by the parents, and the check line RD23, and the taller plant type represented by the traditional Thai cultivars KDML105 and RD6. Clearly, the lines derived from Cross 3 should be examined further for their potential to increase yield and improve the broad and speci®c adaptation of rainfed lowland rice in Northeast Thailand. Fukai and Cooper (1999) proposed a modi®ed breeding strategy (Table 2) to deal with some of the weaknesses identi®ed for the current strategy (Table 1). The proposed strategy emphasised: (1) greater coordination of crossing among the six research stations in the Northeast region, (2) early generation selection for the simply inherited grain quality traits and the agronomic traits days-to-¯ower and plant height, known to affect yield, (3) discard the intrastation selection phase of the program and implement F4bulk yield evaluation in interstation trials to obtain information on the yield adaptation of a larger number of lines than is currently achieved. While it is dif®cult to experimentally evaluate the effectiveness of alternative breeding strategies, the estimates of line-mean heritability for the intrastation and interstation testing strategies, and the results from the experimental selection study, suggested that a greater response to selection should be expected from the proposed breeding strategy. The computer simulation analysis supported the hypothesis that the proposed breeding strategy would more effectively accommodate the effects of G E interaction and contribute to a greater response to selection for broad adaptation than the current strategy. As additional information becomes available on the genetic and environmental bases of the G E interactions for grain yield and the traits contributing to broad and speci®c adaptation (e.g. Fukai and Cooper, 1995; Nyguen et al., 1997; Mackill et al., 1999), the QU-RLR simulation model developed for this study can be used to re-evaluate this result. The quantitative analyses of the genetic control of the variation for the three traits in this study, and the companion study by Cooper et al. (1999a), indicated that in general plant height and days-to-¯ower were under more simple genetic control than grain yield.
174
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
This is consistent with the reports by Mackill et al. (1996). Parental performance for plant height and days-to-¯ower was found to be indicative of progeny mean performance on a cross-mean basis, but not for grain yield (Cooper et al., 1999a). The predominant genetic components of variance for the three traits differed among the seven crosses examined. For both plant height and days-to-¯ower there was a strong contribution from additive sources of genetic variance. However, for grain yield there was a predominance of non-additive sources of genetic variance (additive-byadditive epistasis and also epistasis-by-environment interaction) contributing to the genotypic variance. This again suggests that genetic manipulation of grain yield will be more complicated than for plant height and days-to-¯ower. The frequency distributions for plant height and days-to-¯ower indicated that majorgenes were segregating in some crosses and not others. Breeding strategies with the objective of improving grain yield of rainfed lowland rice in Northeast Thailand should take into consideration: (1) the differences in sources of genetic variation for the traits both among and within the crosses, (2) the different associations between grain yield and both days-to-¯ower and plant height among the crosses, and (3) the low heritability for grain yield on line-mean and withincross bases and intermediate heritability for grain yield on an among-cross basis. The presence of a large component of variance associated with G E interaction is indicative of the need for particular attention to be given to the relationship between the samples of environments obtained in the on-station METs and the performance of the breeding lines in the on-farm target population of environments. To date there has been limited attention to experimental activities that could be used to quantify the genetic correlation between the results of the on-station breeding trials and line performance under on-farm conditions, for rainfed lowland rice in Thailand. To accommodate the degree of uncertainty that is associated with the magnitude of the on-station to on-farm relationship, Fukai and Cooper (1999) recommended that a larger number of the lines demonstrating superior performance in the interstation evaluation phase should enter the on-farm testing phase (Table 2). However, it is recognised that this strategy will be sub-optimal if there is a poor genetic correlation between the on-station and on-farm performance
of the breeding lines. If this relationship is found to be poor then consideration will have to be given to greater use of on-farm evaluation at all stages of the breeding program. Determining this relationship should be given a high priority for future evaluations of the effectiveness of any rainfed lowland rice breeding strategy in Northeast Thailand. The large experimental error component of variance observed in the breeding trials is another impediment to progress from selection. Randomised complete block experimental designs are routinely used for all stages of the Thai rainfed lowland rice breeding program. There have been signi®cant advances in experimental design and analysis methodology that have not yet been fully evaluated for their potential impact on accommodating the effects of experimental error in the rainfed lowland rice trials in Northeast Thailand (Basford et al., 1996). A preliminary evaluation of the rainfed lowland trials included in the study by Cooper et al. (1999a) suggested that statistical adjustment of the results for environmental trends within the experiments could be used to reduce some of the negative effects of experimental error on selection (Cooper et al., 1999b). The potential for routine application of this statistical methodology to reduce experimental error in rainfed lowland rice experiments and the impact of this on heritability and response to selection requires further consideration. From the results of the analyses of genetic variation and G E interactions for grain yield of rainfed lowland rice in Northeast Thailand (Cooper et al., 1999a; Fukai and Cooper, 1999), and the sensitivity of the line-mean heritability to the emphasis placed on intrastation and interstation selection the following conclusions can be made: 1. There is little merit in having more than two replicates within a site for on-station testing. 2. The genetic control of grain yield within crosses is likely to be complicated by non-additive sources of genetic variation. Heritability of grain yield is low on a line-mean basis within individual environments. However, it can be increased by a combination of examining genetic variation on among-cross and within-cross bases, and greater use of multi-environment testing. The six semiindependent breeding programs could take advantage of these opportunities for increasing herit-
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
ability by greater co-ordination of activities and by operating as one co-ordinated breeding program for Northeast Thailand. 3. The current intrastation selection strategy should be discontinued and replaced with an extensive interstation selection strategy. An interstation MET based on six sites, 2 years and two replicates, is recommended. The optimal interstation MET structure should be continually reevaluated as the understanding of the causes of the G E interactions, in particular the G S Y interactions, for yield improves. 4. To take full advantage of the information that is generated on the broad and speci®c adaptation of breeding lines in the interstation METs there should be co-ordination of the analysis and interpretation of the results of the interstation trials. 5. There is a need for more extensive testing of a larger number of breeding lines under on-farm conditions and evaluation of the genetic correlation between the on-station and on-farm performance of the breeding lines. Acknowledgements The Australian Centre for International Agricultural Research provided the ®nancial support for conduct of the experiments. References Basford, K.E., Williams, E.R., Cullis, B.R., Gilmour, A., 1996. Experimental design and analysis for variety trials. In: Cooper, M., Hammer, G.L. (Eds.), Plant Adaptation and Crop Improvement. CAB International, Wallingford, UK, pp. 125± 138. Center for Agricultural Information, 1998. Report on survey of main season rice cultivation, crop year 1997/98. Office of Agricultural Economics, Ministry of Agriculture and Cooperatives. Agricultural Statistics Document 14/2542. Cooper, M., 1999. Concepts and strategies for plant adaptation research in rainfed lowland rice. Field Crops Res. 64, 13±34. Cooper, M., Rajatasereekul, S., Immark, S., Fukai, S., Basnayake, J., 1999a. Rainfed lowland rice breeding strategies for Northeast Thailand. I. Genotypic variation and genotype environment interactions for grain yield. Field Crops Res. 64, 131±151. Cooper, M., Podlich, D.W., Fukai, S., 1999b. Combining information from multi-environment trials and molecular markers to select adaptive traits for yield improvement of rice in water-
175
limited environments. In: O'Toole, J.C., Ito, O., Hardy, B. (Eds.), Genetic Improvement of Rice for Water-Limited Environments. Proceedings of the Workshop on Genetic Improvement of Rice for Water-Limited Environments, 1±3 December 1998, Los BanÄos, Philippines. International Rice Research Institute, Los BanÄos, Philippines, 1999, in press. Cooper, M., Somrith, B., 1997. Implications of genotype-byenvironment interactions for yield adaptation of rainfed lowland rice: influence of flowering date on yield variation. In: Fukai, S., Cooper, M., Salisbury, J. (Eds.), Breeding Strategies for Rainfed Lowland Rice in Drought-prone Environments. Proceedings of an International Workshop held at Ubon Ratchathani, Thailand, 5±8 November 1996. ACIAR Proceedings No. 77, pp. 104±114. Fox, P.N., Rosielle, A.A., 1982. Reducing the influence of environmental main-effects on pattern analysis of plant breeding environments. Euphytica 31, 645±656. Fukai, S., Cooper, M., 1995. Development of drought-resistant cultivars using physio-morphological traits in rice. Field Crops Res. 40, 67±86. Fukai, S., Cooper, M., 1999. Plant breeding strategies for rainfed lowland rice in Northeast Thailand. In: Horie, T., Geng, S., Amano, T., Inamura, T., Shiraiwa, T. (Eds.). Proceedings of the International Symposium World Food Security and Crop Production Technologies for Tomorrow, 8±9 October, Kyoto, Japan, pp. 153±156. Fukai, S., Cooper, M., Salisbury, J. (Eds.), 1997. Breeding Strategies for Rainfed Lowland Rice in Drought-Prone Environments. Proceedings of an International Workshop, Ubon Ratchathani, Thailand, 5±8 November 1996. ACIAR Proceedings No. 77, Canberra, ACT, 248 pp. Gabriel, K.R., 1971. The biplot graphic display of matrices with application to principal component analysis. Biometrika 58, 453±467. Hanson, W.D., Weber, C.R., 1961. Resolution of genetic variability in self-pollinated species with an application to the soybean. Genet. 46, 1425±1434. Henderson, S.A., Fukai, S., Jongdee, B., Cooper, M., 1996. Comparing simulation and experimental approaches to analysing genotype by environment interactions for yield in rainfed lowland rice. In: Cooper, M., Hammer, G.L. (Eds.), Plant Adaptation and Crop Improvement. CAB International, Wallingford, UK, pp. 443±464. Immark, S., Mitchell, J.H., Jongdee, B., Boonwite, C., Somrith, B., Polvatana, A., Fukai, S., 1997. Determination of phenology development in rainfed lowland rice in Thailand and Lao PDR. In: Fukai, S., Cooper, M., Salisbury, J. (Eds.), Breeding Strategies for Rainfed Lowland Rice in Drought-prone Environments. Proceedings of an International Workshop held at Ubon Ratchathani, Thailand, 5±8 November 1996. ACIAR Proceedings No. 77, pp. 89±96. Mackill, D.J., Coffman, W.R., Garrity, D.P., 1996. Rainfed Lowland Rice Improvement. International Rice Research Institute, PO Box 933, Manila, Philippines, 242 pp. Mackill, D.J., Nguyen, H.T., Zhang, J.X., 1999. Use of molecular markers in plant improvement programs for rainfed lowland rice. Field Crops Res. 64, 177±185.
176
M. Cooper et al. / Field Crops Research 64 (1999) 153±176
Nyguen, H.T., Chandra Babu, R., Blum, A., 1997. Breeding for drought resistance in rice: Physiology and molecular genetic considerations. Crop Sci. 37, 1426±1434. Podlich, D.W., Cooper, M., 1998. QU-GENE: a simulation platform for quantitative analysis of genetic models. Bioinformatics 14, 632±653.
Somrith, B., 1997. Cultivar improvement for rainfed lowland rice in Thailand. In: Fukai, S., Cooper, M., Salisbury, J. (Eds.), Breeding Strategies for Rainfed Lowland Rice in Droughtprone Environments. Proceedings of an International Workshop held at Ubon Ratchathani, Thailand, 5±8 November 1996. ACIAR Proceedings No. 77, pp. 36±42.