Alternative Residual Feed Intake (RFI) expressions in dairy cattle

Abstract

Residual Feed Intake (RFI) is commonly defined as residuals from linear regression of feed intake on energy sinks, expressed on the phenotypic scale.  Estimates of partial regression coefficients are obtained by Least Squares, and RFIs are subsequently used as phenotypes in a genetic evaluation model. Alternatively, regression coefficients for RFI can be derived directly from phenotypic co-variances among feed intake and the energy sinks, and EBVs for RFI can be formulated as reparameterizations of EBVs for feed intake and energy sinks from a multiple-trait (MT) model. This is equivalent to the recursive model (RM) approach, with EBVs calculated as system parameters. Using RM as operational tools, RFI can be defined and the respective parameters calculated, for overall and any individual source of random variation covered by the MT model for feed intake and energy sinks, i.e., genetic, PE, residual.  Different definitions of RFI result in independence of RFI from energy sinks on different levels of variability. These concepts are illustrated by application of the genetic evaluation model for feed efficiency of Canadian Holsteins. A six-trait MT model for Dry Matter Intake (DMI), Energy Corrected Milk (ECM) and Metabolic Body Weight (MBW) in two DIM intervals of 1^st lactation was fitted to approximately 100,000 weekly records on 5,000 cows, with 9,000 genotyped animals in the pedigree via MC-EM-REML and Single-Step GBLUP, for the purpose of co-variance component estimation and genomic evaluation. Four different expressions of RFI in 61 – 305 DIM in lactation (phenotypic = pRFI, genetic = gRFI, permanent environmental = eRFI and residual = rRFI) were defined and examined as potential selection criteria or as tools for optimizing management, with respect to estimates of genetic parameters and GEBV. Standardized regression coefficients of DMI on sinks differed among RFI definitions, but the relative impact of sinks was similar across all definitions. Heritabilities of RFIs ranged from 0.05 (gRFI) to 0.15 (rRFI). Genetic and phenotypic expressions of RFI were genetically correlated at 0.84. Genetic correlations between pRFI and energy sinks were 0.62 for ECM and 0.04 for MBW (versus 0.00 for gRFI). Genetic correlations with DMI were 0.37 and 0.59 for gRFI and pRFI, respectively. Correlations between GEBV, for official sires (N = 298), ranged from 0.64 (gRFI and pRFI) to 0.99 (pRFI and eRFI). Results clearly illustrate substantial differences among definitions of RFI in dairy cattle and important consequences of using different definitions for genetic evaluation and selection. Generalizations to other, residual or ratio, traits are straightforward.