Download - S3.1 Integrating selection for stress tolerance with selection for yield potential in maize
Integrating selection for stress tolerance
with selection for yield potential in maize
CIMMYT
Gary Atlin
Jill Cairns
Samuel Trachsel
Felix San Vicente
Cosmos Magorokosho
Peter Setimela
Dan Makumbi
Pichet Grudlyoma
PH Zaidi
…but yield is stress tolerance! -Duvick
-Zhang
Outline
1. How has CIMMYT made gains for
tolerance to severe stress?
2. What are the difficulties in using
managed-stress data for selection, and
how can we deal with them?
3. How can we increase yield “potential” in
tropical maize?
Where will the additional maize Asia
needs come from?
• Mainly from favorable rainfed environments
• …but even favorable environments have drought,
heat, cold, low sunlight, and waterlogging
• Farmers need high yield potential (YP), but high YP is
mainly tolerance to moderate stress
• Tolerance to moderate stress and high YP are easy to
integrate
• Tolerance to severe stress and high YP are much
harder to integrate
Temperate maize yield gains were due to
• Increased tolerance to high density
• Improved DT
• Enhanced capacity to extract nutrients from deeper soil layers
• Faster recovery from cold stress
• Improved stay-green.
• Faster dry-down
Gains were not due to
• Increased photosynthesis rate
• Increased harvest index
• Transgenics**
Lee EA and Tollenaar M. 2007. Physiological basis of
successful breeding strategies for maize grain yield. Crop Sci.
2007 47: S-202-215S.
How were Corn Belt stress tolerance gains achieved?
• No direct selection for yield under drought, low-N, flooding, heat, or
cold!
• Gains were achieved almost entirely from
• wide-scale multi-location testing in the TPE under rainfed
conditions
• Selection for plant density tolerance
• These selection techniques are very effective in productive
environments with moderate, intermittent stress
• Managed stress required when stress is frequent and severe
The CIMMYT approach to breeding for abiotic
stress tolerance
CIMMYT started MSS in 1975 to improve maize for drought,
low N via recurrent selection
Introduced the use of managed stress environments
■ NOT to simulate a farmers field
■ BUT to simulate a stress that is highly relevant in farmers’ fields
■ 60-80% yield reduction targeted due to stress
Gains from stress-tolerance breeding at
CIMMYT
• Early stress-tolerance breeding based on rapid-cycle
recurrent selection produced gains of about 100 kg ha-1
yr-1
• More recently, pedigree breeding has resulted in gains
in farmers’ fields, but has not led to breakthroughs in
stress tolerance
Drought stressed Well-watered
Pedigree
Yield
(t ha-`)
Days to
anthesis ASI
Yield
(t ha-1)
Days to
anthesis
DTPYC9-F46-1-2-1-2 2.66 72 0.7 7.35 73
La Posta Seq C7-F64-2-6-2-2 2.51 75 1.3 7.88 76
DTPWC9-F24-4-3-1 2.49 73 1.4 7.27 74
CML442/CML312SR (check) 2.09 77 6.0 7.52 80
CML442/CML444 (check) 2.00 80 3.7 7.19 77
Mean 2.13 74.5 4.3 6.90 76.2
LSD 0.81 2.0 3.7 1.26 2.5
Lines combining heat and drought tolerance identified from the DTMA association mapping panel as a result of screening under managed stress in 9 environments (J. Cairns)
Gains made for high-yield environments in
farmers’ fields in Eastern and Southern Africa:
Results of 26 farmer-managed strip trials in 2011
Year of first regional testing Name
Yield (t/ha)
2007 CZH0616 6.32
1995 SC513 4.75
SC627 5.05
Mean 5.37
n 26
H 0.83
LSD 0.67
Gains per year
under favorable
conditions:
• 110 kg/ha
• 2.8 %
Gains made for low-yield environments in
farmers’ fields in Eastern and Southern Africa:
Results of 19 farmer-managed strip trials in 2011
Year of first regional testing Name
Yield (t/ha)
2007 CZH0616 2.37
1995 SC513 1.60
SC627 2.03
Mean 2.00
n 19
H 0.62
LSD 0.44
Gains per year under
unfavorable
conditions:
• 66 kg/ha
• 4 %
• We are identifying some
hybrids combining high
stress tolerance and
yield potential!
• Where are these gains
coming from?
Selection environment Low-yield target
environment
Genetic correlation
Early maturity group
Optimal 0.80
Managed drought 0.64
Low-N 0.91
Late maturity group
Optimal 0.75
Managed drought 0.76
Low-N 0.90
Genetic correlations for yield between low-yield target
environments and optimal, managed drought, and low-N selection
environments: ESA 2001-9
• Yield in low-
yield trials is
most closely
related to
yield under
low N
The “standard” CIMMYT breeding pipeline
Stage Activity Screening environment
Reps
Rows/
plot
Optimal Drought Low N
----------- number of trials -------
Line development Unreplicated nursery 1 or 2
Stage 1 testcross
evaluation
Replicated yield trials 4-8 1-2 1-2 2 1
Stage 2 testcross
evaluation
Replicated yield trials 8-10 1-2 1-2 2 2
Line x tester Replicated yield trials 8-10 1-2 1-2 2 2
Advanced hybrid
testing
Replicated yield trials 8-10 1-2 1-2 3 2
Regional yield
testing
Replicated yield trials 15-30 1-2 1-2 3 2
• Replication, and therefore H, is much higher for optimal than stress
trials! • How do we combine the data from optimal and stress trials?
In combining stress and nonstress trial data
we need to consider:
• How repeatable are the stress data?
• How representative are the results of stress trials of
stress in farmers’ fields?
• Do the stress trials give information that is different from
non-stress trials
• What is the frequency of occurrence of stress and non-
stress fields in the target environment?
TPE SE
rG(SE-TPE) HSE
We select in selection environments (SE) to make gains
in the target population of environments (TPE) (farmers’
fields) via correlated response
Correlated response in farmers’
fields is a function of:
• the genetic correlation between SE
and TPE
• H in the SE
The target TPE in drought-prone regions is a
mixture of stressed and non-stressed fields
TPE
Stress
Non-stress
We use stress and non-stress selection
environments (SE) to maximize gains in the TPE
via correlated response
TPE
Stress
Non-stress
SE
Gains in the TPE depend on repeatability (H) in
the two SEs, and…
TPE
Stress
Non-stress
Hstress
Hnonstress
SE
…the genetic correlations (rG ) between SEs and
stress and non-stress components of the TPE
TPE
Stress
Non-stress
rGSS
rGSN
rGNS
rGNN
Hstress
Hnonstress
SE
…the genetic correlations (rG ) between SEs and
stress and non-stress components of the TPE
TPE
Stress
Non-stress
rGSS
rGSN
rGNS
rGNN
Hstress
Hnonstress
SE
rG(SE)
TPE
Stress
Non-stress
rGSS
rGSN
rGNS
rGNN
Hstress
Hnonstress
SE
rG(SE)
The weight should also reflect the relative
frequency of stress and non-stress fields
• Usually only H’s are known
• SE – TPE correlations are assumed
to be high
TPE
Stress
Non-stress
rGSS
rGSN
rGNS
rGNN
Hstress
Hnonstress
SE
rG(SE)
Very few of these parameters
have been measured!
What do we know about these repeatabilities
and correlations?
TPE
Stress
Non-stress
rGSS
rGSN
rGNS
rGNN
Hstress
Hnonstress
SE
rG(SE)
Hnonstress > Hstress
All of the rG’s are positive
Implications for screening systems
1. Hstress is almost always << Hnon-stress in practical
breeding programs
• Breeding programs that put too much weight on low-H non-
stress trials will reduce gains in both stress and non-stress
environments
2. rG between stress and non-stress trials is almost
always positive in adapted breeding populations
• Selection for yield under normal rainfed conditions will give
some gains in yield under severe stress.
• If rG is low, weight given to stress trials should be proportional
to H and the frequency of drought in the TPE
• If rG is high (> 0.8) managed stress is not needed
Why is H always greater in non-stress than stress
environments in cultivar development programs?
σ2G
σ2G + (σ2
GE /e) + (σ2e /re)
= H
Why is H always greater in non-stress than stress
environments in cultivar development programs?
σ2G
σ2G + (σ2
GE /e) + (σ2e /re)
= H
• Genotype x trial and within-trial variability is
almost always larger in managed stress trials
Why is H always greater in non-stress than stress
environments in cultivar development programs?
σ2G
σ2G + (σ2
GE /e) + (σ2e /re)
= H
• Genotype x trial and within-trial variability is
almost always larger in managed stress trials
• Replication across environments is almost
always lower in managed-stress than in non-
stress trials
DTMA AM set: variance components, LSD and
H from the analysis over 9 DS or 7 WW trials (2
reps per trial)
Parameter DS WW
Mean 2.12 6.88
σ2G 0.07 0.51
σ2GE 0.27 0.50
σ2E 0.31 0.57
H 0.62 0.84
LSD.05 0.81 1.16
• There is GxE in managed
stress trials
• Error in managed stress
trials is always higher than
in non-stress trials
• H in managed stress trials
is therefore lower for the
same number of trials
How many managed drought trials
does a breeding program need?
No. of trials
Managed drought WW
1 0.14 0.39 2 0.24 0.57 3 0.32 0.66 4 0.39 0.72 5 0.44 0.76
10 0.61 0.87
Predicted H of yield under managed drought and
WW conditions, using DTMA variance components:
Mexico, Kenya, Zimbabwe, and Thailand 2009-11
It takes 3-4 managed
drought trials to
achieve same H as 1
non-stress trial.
Evaluation of commercial hybrids under moderate stress: Takfa,
Thailand 2007 (from Trial HT071) – P. Grudlyoma
Hybrid
Stress
yield
Non-
stress
yield ASI
Big 919 6.9 9.6 2.0
NK 48 5.7 9.8 3.4
Mean 6.3 9.7 2.7
LSD.05 1.9 1419 2.4
H 0.64 0.81 0.77
• Under moderate stress (yield reduction of 53%), hybrid Big919
performed well relative to stress tolerant hybrid NK48
Using managed stress trials to eliminate
very weak hybrids
Evaluation of commercial hybrids under severe stress:
Takfa, Thailand 2008 (from Trial AH8101)
Hybrid
Stress
yield
Non-
stress
yield ASI
Big 919 1.1 9.7 11
NK 48 4.5 10.5 6
Trial mean 2.2 8.8 7
LSD.05 0.4 0.3 5
H .87 .89 0.93
• Under severe stress (yield reduction of 75%), Big 919 collapsed.
P. Grudlyoma
Breeders must have mixed-model software that gives
the correct H and LSD for each trait used in selection!
• Breeders need to know H for every trait they are selecting
on in yield trials. Selecting on traits with low H is like
selecting based on random numbers
• Breeders need software that automatically calculates and
presents H from single and multi-location trials
• CIMMYT has incorporated R and SAS programs for this
into the Maize Fieldbook. We can help you implement this.
• CIMMYT will publish a set of SAS programs soon that
calculate H, LSD, and BLUP for all traits, any usual design
Entry
Optimal
yield
Drought
yield pER
(CML495 x CL-RCW54)-B-2-3//CML494 6.72 3.16 0.09
(CML495 x CL-RCW54)-B-18-1-
1//CML494 6.52 2.69 0.13
(CML495 x CL-RCW54)-B-17//CML494 6.47 1.79 0.09
(CML495 x CML254)-B-23-1//CML494 4.60 1.66 0.13
(CML503/CML492)//CML491 4.53 1.26 0.13
Trial Mean 5.60 1.91 0.11
LSD 0.88 1.65 0.06
Heritability 0.56 0.10 0.56
Entry variance 0.12 0.04 0.11
Entry x loc variance 0.26 0.02 0.33
Residual variance 0.65 0.63 0.36 Number of reps 2 2 2
Number of locs 6 1 6
Means of white lowland tropic stage 2 testcrosses
screened at 6 optimal and 1 drought location in 2008
Conclusions from CIMMYT’s experience of combining
data from stress and non-stress trials
Managed stress (MS) trials can give very important information, but
are often of low H due to high error and genotype x trial interaction
Selection decisions should be made on mean of 3-4 managed stress
trials, not 1.
We must check to see if MS trials are truly predictive of performance
under stress in the target environment
For most breeding programs, MS trials should be used like disease
screening trials – to throw out highly susceptible materials.
Putting too much weight on low-H trials is like throwing out replicates
from your good trials
Means for low-yield and high-yield trials should be reported
separately to identify specifically-adapted hybrids, and those that
work across yield levels
Breeders must have good data, and good analysis tools, to
make good decisions!
The biggest source of GEI in rainfed yield
trials is mean yield level
Often, in multi-location yield trials, we have a big range in
trial mean yield
If we analyze high- and low-yield trials together, the
information from the low-yield trials will be “hidden” by the
high-yield trials
It is best to analyze and present the means of high- and
low-yield trials separately.
This allows you to identify hybrids that are good at both
yield levels, or that should only be used by farmers in low-
or high-yield environments
Example: 2011 Southern African regional
trial
All trials High yield trials Low yield trials
PEX 501 PEX 501 CZH1033
SC535 X7A344W CZH0935
AS113 AS113 CZH1036
X7A344W SC535 CZH0928
AS115 AS115 CZH1031
Mean yield 4.81 6.51 2.17
H 0.88 0.89 0.75
Top 5 of 54 entries in 14 high-yield trials and 9 low-yield trials
All High
High 0.97
Low 0.57 0.36
Correlations
among yield
levels
Opportunities for increasing breeding gains
and yield potential in tropical maize
1. Increase density tolerance
2. Increase harvest index (HI)
3. Increase grain-filling period
and reduce dry-down time
4. Reduce breeding cycle
time.
Harvest Index (%)
35 40 45 50 55
Gra
in y
ield
/ p
lan
t (g
)
40
60
80
100
120
140
160
180
HN
LN
r2
= 0.58
r2
= 0.50
Relationship between yield and HI in 23
elite hybrids, AF and Tlaltizapan, 2011
F. San Vicente, S. Trachsel
Planting Density
5 7 9
Gra
in Y
ield
/ m
2 (
g)
0
200
400
600
800
1000
1200
1400
HN
LN
* * *
ab
c
ab
b
Mean response of 4 hybrids to 3 densities
at two locations in Mexico, 2011
S. Trachsel, S. San Vicente
CM
L247
/CM
L254
CM
L448
/CM
L449
CM
L494
/CM
L495
CRCw10
5/CLW
N20
1
Ha
rve
st
Ind
ex (
%)
0
10
20
30
40
50
60
1995 2007
Harvest index of old and new
hybrids, 2 locations in Mexico, 2011
S. Trachsel, S. San Vicente
• No
improvement
in HI!
S. Trachsel, S. San Vicente
Planting Density
5 7 9
Gra
in Y
ield
/ P
lan
t (g
)
0
50
100
150
200G1
G2
G3
G4
• New hybrids should
have much better
tolerance to density!
Response of 2 older and 2 newer hybrids to plant
density: 2 locations in Mexico, 2011
Reducing the breeding cycle
• Gains per year are directly proportional to the length of
the breeding cycle
• Many breeders wait too long before using promising new
lines as parents, often testing for 7-8 years.
• The best new Stage 2 lines should be immediately used
as parents.
• Breeding cycle should be 5 years maximum. Easily
achieved with DH and 2 seasons per year
Genomic selection- a new approach to
reducing the breeding cycle in maize
• Most agronomic traits in maize are highly polygenic
• Marker index selection approaches that use the effects of many
markers (thousands) can predict performance for such quantitative
trait.s
• Modern marker prediction approaches, referred to as genomic
selection (GS), incorporate all genotyped markers into a prediction of
breeding or genotypic value (GEBV), rather than a significant subset
• Selection based on markers alone can greatly reduce cycle time, if
GEBVs are accurate and remain so for several cycles
New developments in genotyping make GS possible
• Currently, high-throughput genotyping systems based on next-gen
sequencing are generating 500,000 SNPs for around $20 per DNA
sample
• Within next 1-2 years, this service should be available in China for $10 or
less
• Cost of genotyping at high density is now no higher than testing in a 3-
rep trial at 1 location.
• All CIMMYT lines entering yield testing will be genotyped
• Historical and current performance information will be used to assign
values to haplotypes using genomic selection algorithms
• Unit of selection will be the haplotype, not the line
• Most breeding procedures will change dramatically
• Costs are only low if throughput is high
• Most large seed companies now predict performance using SNPs at
moderate density
• This is a form of genomic selection (GS)
• In GS programs, you estimate haplotype effects, then select the
lines with the best haplotype for phenotyping.
GS protocol
1. Genotype all stage 2 lines at the highest density possible
2. Estimate haplotype effects using testcross data from the lines
3. Select un-phenotyped lines of the next cohort on the basis of
summed haplotype effects (GEBVs)
4. Selection based on haplotype or marker effects alone can be
done very quickly (one or two cycles per year)
5. Gains per year will depend on accuracy of GEBVs
6. Even if GEBVs are only 25% of phenotypic estimates, gains can
be at least doubled if cycle time is reduced from 5 years to 1.
Advantages of GS?
• Will allow us to select for drought tolerance even if we can’t
phenotype in a given season (just use last season’s effects)
• Will allow us to pre-select promising DH lines, once we start
producing more than we can phenotype (next year).
• Does not require extra phenotyping of lines we would normally
discard, as does MARS. Fits well in a pedigree program
• Rapid-cycle methods can increase rates of gain
Rapid cycle GS networks
“Open-source” breeding networks could provide companies with
proprietary lines, but allow haplotype effects to be shared
Rapid-cycle
marker-only
selection
Phenotyping by
Company 3 Phenotyping by
Company 1 Phenotyping by company 2
Lines with high value confirmed
by phenotyping released
commercially by partners
Lines extracted, genotyped: untested,
proprietary DH lines provided to
companies based on GEBVs
Overall conclusions on improving yield
potential and stress tolerance
• CIMMYT is making gains in both optimal and stress-prone
environments
• The key to gains is wide-scale replicated yield testing in the target
environment
• Managed stress screening is extremely useful for identifying very
weak and very tolerant material
• Care must be taken in using managed stress (and all other) data to
avoid selecting on low-H data
• Breeders need software tools that allow them to monitor H in their
trials. CIMMYT is providing these tools
• Increasing HI and density tolerance will increase yield in the tropics
• Reducing breeding cycle time is critical to increasing gains
• High-density genotyping is now available at low cost, permitting GS
• Advantage of GS is that it permits greatly reduced cycle time, and
therefore increased gains