Genetic analysis of longitudinal height data using random regression

Abstract

Genetic analysis of forest longitudinal height data using random regression (RR) has the potential to be attractive to tree breeders because of its advantages for selection at early ages. Our study provides an example of implementation of RR to forest tree height growth data. The data set comes from the Swedish Scots pine ( Pinus sylvestris L.) breeding program with a pedigree over three generations and consists of 899 trees with reconstructed phenotypic height records for 16 years. Legendre polynomials and B-splines were used as base functions in RR models. The restricted maximum likelihood method was employed to estimate (co)variance parameters. Results show that heritability increased with age, except for early ages (years 1 to 4). In general, slightly higher heritabilities were found for the RR model than for the single-trait and paired-trait analyses for most ages. Moreover, the heritabilities obtained with B-splines as the base function tended to be somewhat higher than those obtained with Legendre polynomials. The RR method provides a promising approach for estimating genetic parameters of longitudinal data that can be used in early selection. However, application to real data from other species and to simulated data is needed before general breeding recommendations can be established.