TASK 6.3 Modelling and data analysis support

Transcription

1 Wheat and barley Legacy for Breeding Improvement TASK 6.3 Modelling and data analysis support FP7 European Project

2 Task 6.3: How can statistical models contribute to pre-breeding? Daniela Bustos-Korts and Fred van Eeuwijk, WU Jacob Lage and Nicholas Bird, KWS FP7 European Project

3 The goal of pre-breeding To provide genotypes (bridging germplasm) carrying favourable alleles that could be used as genetic resource to enrich elite germplasm. Challenge: - Landraces have lower performance than elite genotypes doe to negative genetic load. - If selection is applied on pre-bridging germplasm, the risk is to select elite alleles. How to generate good-performing genotypes, that also contribute to genetic diversity? Gorjanc et al. BMC Genomics (2016) 17:30 3

4 A Whealbi example: NAM population under selection 4

5 A Whealbi example: NAM population under selection P18 P19 P02 P03 P04 Some genomic regions show strong evidence of selection (deviate from the expected 0.5) P17 (n=26) P16 P15 BC1F5 P01 P05 P06 (n=31) P07 P14 P08 P13 P12 P11 P10 P09 (n=54) Exotic allele frequency (n=8) 5

6 Questions Which exotic parents have the largest pre-breeding value? (i.e. the ability to increase yield when crossed with an elite line, but with a high proportion of exotic alleles) Which genotypes in the progeny look most promising to be used for further breeding? Which genomic regions are responsible for the increased prebreeding value? 6

7 Challenges Multiple traits of interest Small and heterogeneous populations (low power for QTL detection) Identify good donor landraces, and good genotypes in the progeny Keep genetic diversity (avoid selecting too much for the elite) 7

8 Phenotypic variation Most of the progeny lower yielding and taller than elite parent. 8

9 Question 1: Parent pre-breeding value Advantage of genomic prediction: Use phenotypes in the progeny to predict the value of a parent Pre-breeding has trials with little replication, predictions are more precise estimate of the genotypic breeding value phenotype i(j) = μ + parent j + genotype i(j) + e i(j) genotype i(j) ~MVN(0, Kσ g 2 ) 9

10 Which exotic parents have the largest pre-breeding value? use phenotypes in the progeny to predict the value of a parent family predicted s.e. P P P P P P P P P P P P P P P P P P P04 P10 10

11 Question 2:Which genotypes in the progeny look most promising to be used for further breeding? Extreme genotypic values are shrunken towards the mean Original phenotypes Genomic predictions 11

12 Question 3 Which genomic regions are responsible for the increased pre-breeding value? For which QTLs is advantageous to have the allele of the exotic genotype? y i(j ) = μ + P j + x i(j) β + ε i(j) ε i(j ) ~N(0, I σ 2 e ) Red= exotic is better Blue= recurrent is better 12

13 QTL models: other options Model Fixed founder Fixed QTL M1a Yes Main effect M1b Yes Main effect Equation y i(j ) = μ + P j + x i(j) β + ε i(j) y i(j ) = μ + P j + x i(j) β + ε i(j) Residual ε i(j ) ~N(0, I σ 2 e ) ε i(j ) ~N(0, Diag σ 2 e ) M2a No Main effect M2b No Main effect y i(j ) = μ + x i(j) β + ε i(j) ε i(j ) ~N(0, I σ 2 e ) y ε i(j ) ~N(0, Iσ 2 i(j ) = μ + x i(j) β + ε e ) i(j) ε i(j ) ~N(0, Diag σ 2 e ) M3a Yes Nested M3b Yes Nested M4a No Nested M4b No Nested y i(j ) = μ + P j + x i(j) β j + ε i(j) y i(j ) = μ + P j + x i(j) β j + ε i(j) y i(j ) = μ + x i(j) β j + ε i(j) y i(j ) = μ + x i(j) β j + ε i(j) ε i(j ) ~N(0, I σ 2 e ) ε i(j ) ~N(0, Diag σ 2 e ) ε i(j ) ~N(0, I σ 2 e ) ε i(j ) ~N(0, Diag σ 2 e ) 13

14 Question 4 Which individuals in the progeny carry trait-increasing alleles, and contribute to genetic diversity? (we don t want to select for the recurrent/elite parent) Contribute to yield and diversity High yield, but similar to the recurrent parent Less alleles shared with recurrent parent More alleles shared with recurrent parent 14

15 Summary Genomic prediction models can help pre-breeders to: predict the performance of parents from their progeny, helping to identify useful material for further crosses. Identify which genotypes in the progeny have high yield, and also contribute to enrich the genetic diversity QTL models can help pre-breeders to identify genomic regions in which the exotic allele contributes to increase yield (helping a more targeted selection) 15