Page 130 - Breeding and regulatory opportunities, Renaud

Chapter 4

All terms were assumed to be normally distributed with a proper variance. For

ease of interpretation and to allow straightforward comparisons between traits,

the variance components were reported as coeicients of variation, which is

standard deviations as a percentage of the trait mean, i.e.,
̅

CV=100 V ⁄ ,

with V the variance for a particular model term and x x the trait mean. Repeatability

(analogous to broad sense heritability, but for unrelated genotypes) was

calculated from the variance components as:

H = V/ (V+ V/nL + V/nS + V/nM + V/(nL.nS) + V/(nL.nM)
G G GLGS GM GLS GLM
+ V/(nS.nM) + V/(nL.nS.nM) + V/(nL.nS.nM.nR),

GSM GLSM e

where the variance components correspond to the terms in the mixed model

above. The terms nL, nS, nM and nR represent the number of locations (2: Maine

and Oregon), number of ‘seasons’ (4: Fall 2006, Spring 2007, Fall 2007, Spring

2008), management (2; organic and conventional), and replicates (2 or 3).

The above model and repeatability was simpliied when performing analyses

per location, or per management regime. For the irst situation, analysis for
x
Oregon and Maine separately, we omitted all terms that contain L. For the second

situation, analysis for organic and conventional management separately, we

omitted all terms containing M.

To calculate genotypic means across all conditions, the general model deined

above was used, but the genotypic main efect was assumed to be ixed instead

of random. Similarly, genotypic means per location and management system

were obtained by assuming ixed genotypic main efects as well as the relevant

environmental efects (L, M) and their interactions (G × L, G × M). These means

were used to calculate correlations and for box plots and biplots (procedure

dbiplot in GenStat). Pairwise comparisons between means were performed using

GenStat procedure VMCOMPARISON. Correlations on the basis of genotypic

means were referred to as genetic correlations.

112