Enhancing Line Breeding 27.04.2017 Sebastian Michel 1 Selection - - PowerPoint PPT Presentation

27.04.2017 Sebastian Michel

1

Genomic Assisted Selection for Enhancing Line Breeding

Selection through the years…

2

Selection through the years…

3

• Great progress in the genetic improvement of crop plants by breeders

Selection through the years…

4

• Great progress in the genetic improvement of crop plants by breeders

>>>But: not all expectations were met

? !

Selection through the years…

5

>>> Development of new technologies & tools for breeding crop plants!

• Great progress in the genetic improvement of crop plants by breeders

>>>But: not all expectations were met and yield progress is insufficient

? !

Source: Ray et al. (2013)

6

• 960 F4:6 wheat lines with subpopulations tested in METs 2010-2016

Genomic selection across years

7

• 960 F4:6 wheat lines with subpopulations tested in METs 2010-2016
• Within-year prediction accuarcy was high

Genomic selection across years

8

• 960 F4:6 wheat lines with subpopulations tested in METs 2010-2016
• Within-year prediction accuarcy was high but overestimated

Genomic selection across years

9

• 960 F4:6 wheat lines with subpopulations tested in METs 2010-2016
• Within-year prediction accuarcy was high but overestimated
• Great promise to genomically predict major traits across years
• protein content rCV:ACROSS = 0.57
• grain yield

rCV:ACROSS = 0.37

Genomic selection across years

10

>>>Can GS support the selection of laborious to phenotype traits?

• 960 F4:6 wheat lines with subpopulations tested in METs 2010-2016
• Within-year prediction accuarcy was high but overestimated
• Great promise to genomically predict major traits across years
• protein content rCV:ACROSS = 0.57
• grain yield

rCV:ACROSS = 0.37

Genomic selection across years

Baking quality in bread wheat

• Assessment of baking quality time-consuming, labor-intensive & costly
• Too little grain available from selection candidates in early generations

11

Source: LflL (2008); modified after Laidig et al. (2016)

Baking quality in bread wheat

• Assessment of baking quality time-consuming, labor-intensive & costly
• Too little grain available from selection candidates in early generations

>>>Breeders postpone thoroughly quality testing into later generations >>>Indirect selection for quality by protein content in early generations

12

Source: LflL (2008); modified after Laidig et al. (2016)

Baking quality in bread wheat

• Assessment of baking quality time-consuming, labor-intensive & costly
• Too little grain available from selection candidates in early generations

>>>Breeders postpone thoroughly quality testing into later generations >>>Indirect selection for quality by protein content in early generations

13

Source: LflL (2008); modified after Laidig et al. (2016)

Accuracy of indirect selection?

Baking quality in bread wheat

• Assessment of baking quality time-consuming, labor-intensive & costly
• Too little grain available from selection candidates in early generations

>>>Breeders postpone thoroughly quality testing into later generations >>>Indirect selection for quality by protein content in early generations

14

Source: LflL (2008); modified after Laidig et al. (2016)

Protein content (%) Loaf volume (ml/100g)

Accuracy of indirect selection? Validation!

Baking quality in bread wheat

• Assessment of baking quality time-consuming, labor-intensive & costly
• Too little grain available from selection candidates in early generations

>>>Breeders postpone thoroughly quality testing into later generations >>>Indirect selection for quality by protein content in early generations >>> Additional assessment of the protein quality seems necessary…

15 Protein content (%) Loaf volume (ml/100g)

Accuracy of indirect selection? Validation!

Source: LflL (2008); modified after Laidig et al. (2016)

Baking quality in bread wheat

• Assessment of baking quality time-consuming, labor-intensive & costly
• Too little grain available from selection candidates in early generations

>>>Breeders postpone thoroughly quality testing into later generations >>>Indirect selection for quality by protein content in early generations >>> Additional assessment of the protein quality seems necessary…

16

Source: Brabender (2017); LflL (2008)

>>> Additional assessment of the protein quality seems necessary… >>> Rheological tests are highly informative

Baking quality in bread wheat

• Assessment of baking quality time-consuming, labor-intensive & costly
• Too little grain available from selection candidates in early generations

>>>Breeders postpone thoroughly quality testing into later generations >>>Indirect selection for quality by protein content in early generations >>> Additional assessment of the protein quality seems necessary… >>> Rheological tests are highly informative but time-consuming & costly

17

Source: Brabender (2017); LflL (2008)

• Training with rheological data from several locations 2009-2012
• Validating the trained statistical model with data from 2013-2015

18

Forward prediction of quality traits

• Training with rheological data from several locations 2009-2012
• Validating the trained statistical model with data from 2013-2015

19

Forward prediction of quality traits

• Training with rheological data from several locations 2009-2012
• Validating the trained statistical model with data from 2013-2015
• GS enables an earlier selection for quality traits with high intensity

20

Forward prediction of quality traits

21

Forward prediction of quality traits

Trait Accuracy Farinograph Development 0.43 Stability 0.36 Extensograph Resistance 0.37 Energy 0.60

• Training with rheological data from several locations 2009-2012
• Validating the trained statistical model with data from 2013-2015
• GS enables an earlier selection for quality traits with high intensity

>>>Predict labor-intensive and costly quality traits in early generations!

22

Forward prediction of quality traits

Trait Accuracy Farinograph Development 0.43 Stability 0.36 Extensograph Resistance 0.37 Energy 0.60

• Training with rheological data from several locations 2009-2012
• Validating the trained statistical model with data from 2013-2015
• GS enables an earlier selection for quality traits with high intensity

>>>Predict labor-intensive and costly quality traits in early generations! >>> How can a breeder implement GS into a line breeding program?

Line breeding with genomic selection

23

• Pedigree selection until F4:5 generation preliminary yield trials
• Genotyping and prediction of genomic breeding values

Source: Löschenberger et al. (2008); Elshire et al. (2011)

Line breeding with genomic selection

>>> How good is genomic in comparison to phenotypic selection?

24

Source: Löschenberger et al. (2008); Elshire et al. (2011)

• Pedigree selection until F4:5 generation preliminary yield trials
• Genotyping and prediction of genomic breeding values
• Selected lines are retested in multi-environment trials

Predict line performance across years

25

• Accuarcy of phenotypic selection:

>>>Correlation between observed line performance in preliminary yield trials and multi-environment trials the following year

Predict line performance across years

26

• Accuarcy of phenotypic selection:

>>>Correlation between observed line performance in preliminary yield trials and multi-environment trials the following year

• Accuarcy of genomic selection:

>>>Correlation between predicted line performance and observed performance in multi-environment trials the following year

2011 2012 2013 2014 2015 2016

Training GS PYT Validation

2010

Predict line performance across years

27

• Accuarcy of phenotypic selection:

>>>Correlation between observed line performance in preliminary yield trials and multi-environment trials the following year

• Accuarcy of genomic selection:

>>>Correlation between predicted line performance and observed performance in multi-environment trials the following year >>> Cross-validation with 10 training x 5 validation population combinations

2011 2012 2013 2014 2015 2016

Training GS PYT Validation

2010

Phenotypic versus genomic selection

28

• Genomic selection was superior to early generation phenotypic

selection based on unreplicated preliminary yield trials >>> Low quality of the derived phenotypic data (!?)

r = 0.20 r = 0.42 r = 0.38 r = 0.49

+ 20% + 90%

Phenotypic versus genomic selection

29

• Genomic selection was superior to early generation phenotypic

selection based on unreplicated preliminary yield trials >>> Low quality of the derived phenotypic data (!?)

r = 0.20 r = 0.42 r = 0.38 r = 0.49

+ 20% + 90%

>>> Utilizing marker data for enhancing phenotypic selection…

Breeding values in preliminary yield trials

30

• Modeling kinship to enhance phenotypic breeding values (KBLUP)

Breeding values in preliminary yield trials

31

Family relationships within preliminary yield trials

• Modeling kinship to enhance phenotypic breeding values (KBLUP)

Breeding values in preliminary yield trials

32

• Modeling kinship to enhance phenotypic breeding values (KBLUP)
• Reduce the challenge of predicting untested genotypes in untested

years to predicting tested genotypes in untested years

r = 0.20 r = 0.42 r = 0.38 r = 0.49 r = 0.33 r = 0.52

+ 25% + 60%

Breeding values in preliminary yield trials

33

• Modeling kinship to enhance phenotypic breeding values (KBLUP)
• Reduce the challenge of predicting untested genotypes in untested

years to predicting tested genotypes in untested years >>> Merging the enhanced phenotypic selection with genomic selection…

r = 0.20 r = 0.42 r = 0.38 r = 0.49 r = 0.33 r = 0.52

+ 25% + 60%

Genomic assisted selection

34

+ 35% + 115%

• Genomic assisted selection was more accurate for predicting across

years than either phenotypic selection or genomic selection alone

r = 0.20 r = 0.42 r = 0.43 r = 0.57 r = 0.33 r = 0.52

Genomic assisted selection

35

+ 35% + 115%

• Genomic assisted selection was more accurate for predicting across

years than either phenotypic selection or genomic selection alone >>> Improve selection decisions by using all available data

r = 0.20 r = 0.42 r = 0.43 r = 0.57 r = 0.33 r = 0.52

Genomic assisted selection

36

>>> How robust is the genomic assisted selection approach? + 35% + 115%

• Genomic assisted selection was more accurate for predicting across

years than either phenotypic selection or genomic selection alone >>> Improve selection decisions by using all available data

r = 0.20 r = 0.42 r = 0.43 r = 0.57 r = 0.33 r = 0.52

Robustness genomic assisted selection

37

38

• Genomic assisted selection was always superior to genomic selection

Global prediction

Robustness genomic assisted selection

39

• Genomic assisted selection was always superior to genomic selection
• The majority of single environments benefitted from the model (80%)
• Strong variation of the prediction accuracy for single environments

Global prediction Local prediction

Robustness genomic assisted selection

40

• Genomic assisted selection was always superior to genomic selection
• The majority of single environments benefitted from the model (80%)
• Strong variation of the prediction accuracy for single environments
• Predicting lines for the entire target population of environments

>>> How strong is the influence of the accuracy on selection decisions?

Global prediction Local prediction

Robustness genomic assisted selection

Selection descision inferences

41

• No difference in correctly identifying lines from either distribution tail
• Correct identification more frequently by genomic assisted selection

Selection descision inferences

42

A selection example:

• Assume a population of 2000 lines
• Select 200 of these lines (10%)
• No difference in correctly identifying lines from either distribution tail
• Correct identification more frequently by genomic assisted selection

Selection descision inferences

43

A selection example:

• Assume a population of 2000 lines
• Select 200 of these lines (10%)
• Correctly selected lines
• 40 (20%) by phenotypic selection
• 70 (35%) by genomic assisted selection
• Higher chance of selecting the

highest performing lines

• No difference in correctly identifying lines from either distribution tail
• Correct identification more frequently by genomic assisted selection

Selection descision inferences

44

>>> Genomic assisted selection is accurate, robust and easy to implement

A selection example:

• Assume a population of 2000 lines
• Select 200 of these lines (10%)
• Correctly selected lines
• 40 (20%) by phenotypic selection
• 70 (35%) by genomic assisted selection
• Higher chance of selecting the

highest performing lines

• No difference in correctly identifying lines from either distribution tail
• Correct identification more frequently by genomic assisted selection

Planning of crosses

45

PTested PTested PUntested PUntested T2 T1 T0 T1 SPV = µParents + iσPop

• Predicting the superior progeny value for planning of crosses

Planning of crosses

46

• Predicting the superior progeny value for planning of crosses

PTested PTested PUntested PUntested T2 T1 T0 T1 SPV = µParents + iσPop

Planning of crosses

47

SPV = µParents + iσPop Var(µParents) >> Var(σPop) PTested PTested PUntested PUntested T2 T1 T0 T1

• Predicting the superior progeny value for planning of crosses

Planning of crosses

48

• Predicting the mid-parent value for planning of crosses
• Validation with mid-parent value based on MET phenotypic data

SPV = µParents + iσPop Var(µParents) >> Var(σPop) PTested PTested PUntested PUntested T2 T1 T0 T1

Planning of crosses

49

• Predicting the mid-parent value for planning of crosses
• Validation with mid-parent value based on MET phenotypic data

>>> Higher accuracy of an early T2 in comparison to a T0 scenario SPV = µParents + iσPop Var(µParents) >> Var(σPop) PTested PTested PUntested PUntested T2 T1 T0 T1

Planning of crosses

50

• Predicting the mid-parent value for planning of crosses
• Validation with mid-parent value based on MET phenotypic data

>>> Higher accuracy of an early T2 in comparison to a T0 scenario >>> Genomic assisted selection could support in the planning of crosses SPV = µParents + iσPop Var(µParents) >> Var(σPop) PTested PTested PUntested PUntested T2 T1 T0 T1

Line breeding…

51

• Information about variety candidates becomes gradually available

Plant type Maturity Height Protein content Grain yield Resistance Tolerance Protein quality Dough rheology Baking volume P M H

52

• Information about variety candidates becomes gradually available

Plant type Maturity Height Protein content Grain yield Resistance Tolerance Protein quality Dough rheology Baking volume

Cross

P M H

Line breeding…

Line breeding with genomic selection

53

• Information about variety candidates becomes gradually available
• Genomic selection provides information earlier and of higher quality

Plant type Maturity Height Protein content Grain yield Resistance Tolerance Protein quality Dough rheology Baking volume

Cross

P G R T P M H

Line breeding with genomic selection

54

• Information about variety candidates becomes gradually available
• Genomic selection provides information earlier and of higher quality

Plant type Maturity Height Protein content Grain yield Resistance Tolerance Protein quality Dough rheology Baking volume

Cross Population improvement Product development

P G R T P M H

Line breeding with genomic selection

55

>>> Genomic selection could be a method for breeding acceleration!

• Information about variety candidates becomes gradually available
• Genomic selection provides information earlier and of higher quality

Plant type Maturity Height Protein content Grain yield Resistance Tolerance Protein quality Dough rheology Baking volume

Cross

P G R T P M H

Population improvement Product development

Acknowledgements

Hermann Bürstmayr Heinrich Grausgruber Maria Bürstmayr Barbara Steiner Franziska Löschenberger Christian Ametz Herbert Hetzendorfer Johann Birschitzky Christian Kummer Martin Gallee Batuhan Akgöl Ali Özbugday

56

Doru Epure Marius Becheritu Christiane Koselek Ralf Horbach Bertholt Bauer Helene Cochet Phillipe Lemaire