Rice breeding is both challenged and benefited by the fact that a successful varietal improvement program must embrace both the integration single genes that segregate in a simple Mendelian fashion as well as complex traits that are inherited in more quantitative ways. For decades the rice genetics community has produced a wealth of knowledge about these single genes and has developed markers that allow a breeder to track them in a population. However, marker assisted selection (MAS) alone is insufficient to drive the rates of genetic gain for more complex traits that are equally necessary. This presentation will describe the attempts made in the Favorable Environments Breeding program at IRRI to integrate the selection for single genes appropriate for MAS into a more complex population improvement strategy designed to improve quantitatively inherited traits.
Similar to Our PAG XXVI Presentations: Integrating Marker-Assisted Selection into a Population Improvement Based Rice Breeding Program - Joshua Cobb, IRRI
Similar to Our PAG XXVI Presentations: Integrating Marker-Assisted Selection into a Population Improvement Based Rice Breeding Program - Joshua Cobb, IRRI (20)
3. The landscape keeps changing and that
impacts the breeding strategy.
Big Data and HTP
Phenotyping
Statistics and
Experimental Design
Genomics and Next-
gen Sequencing
Quantitative Genetics
and Molecular Biology
QTL Mapping and
Gene cloning
Locally Adapted
Modern Germplasm
1960’s
1970’s
1980’s
1990’s
2000’s
2010’s
Consumer Preferences
4. Biology dictates two trait deployment strategies that work in tandem to create an integrated breeding
system capable of delivering on genetic gain for all traits of interest.
Revisiting the “Modern Synthesis”
Genotypes
Environments
Phenotypes
Trait Markers/QTL Fingerprinting
5. Varieties in farmer’s fields are old!
Species Period
Rate of genetic gain
(kg ha-1 yr-1)
Maize (Pioneer) 1930-2010 89 (1.2%)
Rice (IRRI) 1960-2014 13 (0.3%)
Variety name
Year of
release
Total area (x
1000 ha)
Proportion of
total area
under rice (%)
Swarna 1980 3,808 27.7
Pooja 1999 998 7.3
Lalat 1989 898 6.5
Moti 1989 277 2
Mahsuri 1975 1,208 8.8
Swarna-Sub1 2009 367 2.7
Sambha
Mahsuri
1989 220 1.6
ARIZE 6444 1990 681 4.9
Sarju-52 1982 350 2.5
MTU1001 1997 523 3.8
MTU1010 2000 346 2.5
Sahbhagi Dhan 2012 35 0.3
Samba-Sub1 2012 30 0.2
Other hybrid 232 1.7
Other improved 1,358 9.9
Other traditional 622 4.5
Unknown 1,80 13.1
Total 13,758 100
(Duvick 2005; IRRI Unpublished data)
Area-weighted avgerage age of
varieties = 28 yrs
Area and age of rice varieties grown in rainfed
eastern India: 2014 wet season (T. Yamano, IRRI)
Modified from G. Atlin, PAG 2017, BMGF
6. 1977 – 1993
Slope: –8.8 kg/ha/yr
SE Slope: +/– 27.4 kg/ha/yr
Adjusted R-squared: -0.0886
95% Confint:
-70.05 kg/ha/yr – 52.3 kg/ha/yr
1960– 1977
Slope: 53.1 kg/ha/yr
SE Slope: +/– 35.2 kg/ha/yr
Adjusted R-squared: 0.1539
95% Confint:
-33.1 kg/ha/yr – 139.0 kg/ha/yr
1994 – 2004
Slope: 83.1 kg/ha/yr
SE Slope: +/– 22.9 kg/ha/yr
Adjusted R-squared: .6026
95% Confint:
28.9 kg/ha/yr – 137.3 kg/ha/yr
2005– 2014
Slope: – 59.6 kg/ha/yr
SE Slope: +/– 16.7 kg/ha/yr
Adjusted R-squared: .6277
95% Confint:
–100.3 kg/ha/yr – –18.8
kg/ha/yr
N=513
Genetic trend for yield 1960-2014 (Pre-MET breeding material)
Dataset: Mean pedigree BLUPs for all breeding lines tested OYT, PYT, and AYT in the
last 5 years with reliabilities > 0.6 regressed on the cross year for each line
7. Human height (h2 ≈ 0.80)
QTL study <10,000 individuals = 50 significant QTLs
Explained only 5% of the pheno variance for height
Yang et al. used a Genomic Selection model
Explained 45% of the pheno variance for height
Sir Francis Galton (1885)
Most of the time, the QTL don’t matter
Yang et al. (2010) Nat. Genetics 42:565-569.
Boyle, Li, and Pritchard (2017)
8. Genetic Gain through population improvement:
IRRI’s elite breeding galaxy
Mean: 4.4 t/ha
StDev: 0.40 t/ha
h2: 0.44
Ne = 33
Elite Founders
TIME à
YIELDà
YLD.FLW
RICE CRP Target:
1.3% rate of genetic gain/year (57 kg/ha/yr)
10 breeding cycles (40 years)
A “breeding galaxy” has been selected from an initial
analysis using 5 years historical breeding data.
-17,216 lines
-23 environments (locations/seasons)
Mean:
5.26 t/ha
86 Parents
9. 0
5
10
15
20
0 5 10 15 20
Population ID
Geneticimprovement
scheme
sim10Ne
sim20Ne
sim50Ne
sim100Ne
Breeding schemes
GeneticStandardDeviation(1SD=0.4t/ha)
Breeding Cycles (1 cycle = 4 years)
Simulated Genetic Gain using New Breeding Strategy
Assumptions: No introductions, Percent Selected = 2.5%
Ne = 33
10. A Tale of Two Galaxies…IRRI’s unique
germplasm resources
PCA of sequenced breeding lines (White) and diverse
indica lines (Yellow) from the 3K genomes.
The elite and diverse galaxies share a
relationship, but are (and should be!) distinct.
• Primary driver of genetic variation:
Diverse = Mutation
Elite = Recombination
• Genetic variation over time:
Diverse = Static
Elite = Dynamic
• Value capture:
Diverse = Rare alleles with large
effect
Elite = Enrichment of elite haplotypes
through recombination.
11. High Efficacy within selected breeding zones
Broad Efficacy across all breeding zones
Sometimes the QTL do matter!
xa5
xa13
xa5
xa13
Xa21
xa5
xa13
Xa7
Xa21
Xa7
Xa23
Xa23
• “Trait Packages” of key genes identified
that are capable of delivering on
phenotypic targets.
• Priority markers deployed at Intertek
($2/sample)
• Locus validation indicates broad application across geographies.
• Marker validation indicates good informativeness within IRRI elite germplasm.]
• Most markers available on Intertek “Ten SNP set”
Broad Efficacy across all breeding zones
Pi9
Pita-2
Pi9
Pik-m
Pita-2
Pik-m
Pi9
Pish
Pita-2
Pish
Pish
Pik-m
SNP Panel Gene/QTL Trait
Panel 2 Pi9 Blast
Panel 2 Pita Blast
Panel 2 xa5 Bacterial leaf blight
Panel 2 xa13 Bacterial leaf blight
Panel 2 Pi54 Blast
Panel 2 Xa21 Bacterial leaf blight
Panel 2 sub1 Submergence
Panel 2 Chalk5 Grain chalkiness
Panel 2 BPH17 Brown planthopper
Panel 2 BADH2 Fragrance
12. x
C AT C
SNP
1
SNP
2
SNP
1
SNP
2
C/A T/C
SNP
1
SNP
2
x
Linkage Disequilibrium: gene pool
C T
A C
492
488
~5
~4
~5
~6
1000
SNP 1 and SNP2
are 2cM apart
C T
SNP
1
SNP
2
A C
SNP
1
SNP
2
C C
SNP
1
SNP
2
A T
SNP
1
SNP
2
250
250
250
250
SNP 1 and SNP2 are NOT in LD in
the breeding pool
Don’t deploy the marker you mapped with…
Linkage: Segregating Population
13. Trait Deployment Strategies for MAS (ex: BLB, Blast)
SNP set Locus Marker ID Utility FPR FNR Okay for QTL profiling?
BPH17 S4_4649295 95% 40% 0% No
xa5 Xa5_S1_SKEP 80% 0% 0% Yes
chalk5 chalk5_576 46% 0% 0% Yes
Pi9 Pi9-1b 100% 19% 0% No
frg-1 BADH2.1-7 100% 0% 0% Yes
xa13 S8_27520607 100% 27% 0% No
Sub1 s9_6774928 91% 0% 0% Linkage not great
Xa21 Xa21_SKEP 100% 0% 0% Yes
Pi54 Pi54-4 86% 0% 50% Yes
Pi-ta Pi-ta 24% 0% 0% Yes
DTY1.1 38474037 100% 0% 25% Maybe
DTY1.1 38867490 100% 0% 25% Maybe
DTY3.1 MSU7_3_30047928_[G/A] 97% 0% 0% Yes
DTY3.1 MSU7_3_31323659_[C/G] 97% 0% 0%
BPH3 S6_1362133 89% 0% 50% No
BPH3 S6_1780414 89% 0% 0% No
TSV1 RTSV1-1* 87% 0% 80% No
xa4 S11_27603799_v2 3% 0% 0% No
DTY12.1 S12_15310036 64% 0% 50% No
DTY12.1 S12_17576492 64% 0% 50% No
TSS-1
TSS-2
Hybridisation
OYT
AYT
LSTF5:6
Trait Introgression pipeline
Haplotype embedding
Trait Development Pipeline
Line augmentation
QTL pyramiding
Coupling-phase linkage
VarietyDevelopmentPipeline
RGA
F2
F3
F4
F5
MAFB
Parental selection QTL profiling
QTL deployment
14. Platten, Cobb, and Zantua (2018) Criteria for evaluating molecular markers: Comprehensive quality metrics to
improve marker-assisted selection, submitted
Comparison of quality
metrics for different markers
for salinity tolerance QTL,
qNa1L, between 37 and
41Mb on chr 1.
Variation in utility for
various QTL in rice.
QTL were selected that have
diagnostic markers or
markers scoring 100% on
both TPR and TNR
15. Integrated and Modernized Breeding System
Estimation Set
Genomic Prediction (384 lines)
• Unselected estimation set deployed to Br. Zones to estimate
performance of OYT lines in regional environments.
PhilippinesBangladesh Breeding Zone X
Product Cycle (from cross o MET): 5 years
6144 lines (197 families) evaluated for MAS in 2017
MB Activities highlighted RED
AYT (384)
AYT (384)
WS
DS
Year 1
WS
DS
Year 2
WS
DS
Year 3
WS
DS
Year 4
Year 5
WS
DS
Est Set grow out; SHU
SHU
LST (5K)
Crosses
Line Fixation
Fingerprint Lines
Prediciton DS performance
Superior yield in Multi-loc trial (Philippines)
Grain Quality profile
Parental Selection
AYT (384)
AYT (384)
AYT (384)
AYT (384)
F1 Verification
Elite x Elite Parents
RGA
MAS
Maturity, height; grain type
Superior yield; Appropriate GQ; Locally disease resistant
Selected based on predicted performance in BZ of interest
Bangladesh Philippines Br. Zone X
OYT (2K)
1920 OYT lines fingerprinted with 822 SNPs in 2017
Breeding Cycle (from cross to cross): 3.5 years
17. Three key messages from your friendly
neighborhood rice breeder:
1) Most of the time, knowing the QTL doesn’t help
– They’re useful for molecular physiologists, but not for breeders
2) Sometimes QTLs do matter, but not all markers are created equal
– Every breeding program is different
3) Simply having access to the tools of the trade isn’t good enough.
– Every breeding program must combine them in ways unique to their
germplasm and objectives.
18. Acknowledgments
• Vit Lopena
• Princess De la Cruz
• Holden Verdeprado
• Partha Biswas
• Roselynne Juma
• Rafiq Islam
• Jessica Rutkoski
• Argem Flores
• Juan David Arbaelez
• Damien Platten
• Tobias Kretzschmar
• Nickolai Alexandrov
• Dmytro Chebotarov
• George Kotch
• Russell Reinke
• Gary Atlin