Evaluation of feeding behaviour traits to predict efficiency traits in pigs using partial least square regression

Abstract

The improvement of efficiency traits, such as protein efficiency (PE), digestible energy efficiency (EnE) and lipid gain (LipG), are relevant given their associations with environmental pollution, cost of productions, and the quality of meat. However, these traits are difficult traits to measure and usually require slaughtering of pigs. Efficiency traits are complex, and several factors, such as genetic predisposition, feed composition, but also individual feeding behaviour may contribute to efficiency. The objective of this study was therefore to evaluate the potential of using feeding behaviour traits to predict efficiency traits under dietary protein restriction. A total of 587 Swiss Large White pigs, consisting of 312 females and 275 castrated males, had ad libitum access to feed and water, and were fed a protein-reduced diet (80% of recommended digestible protein and essential amino acids) from 22.5 {+/-} 1.6 to 106.6 {+/-} 4.6 kg BW. Individual feed intake was monitored and carcass composition (lean and fat mass) at slaughter was determined by dual-energy X-ray absorptiometry (DXA). The PE and EnE were calculated as the ratio of protein or energy in the carcass (estimated by DXA) to the total protein or energy consumed. Feeding behaviour traits monitored were daily feed intake (DFI; g/day), feed intake per meal (FIM; g/meal), number of daily meals (NDM; meals/day), duration of meal (DUM; min/meal), feeding rate (FR; g/min), and feeder occupation (FO; min/day). A partial least square (PLS) regression was used to predict PE, EnE and LipG from feeding behaviour traits, while including farrowing series (for PE only), age at slaughter and body weight at slaughter. Accuracy of PLS regression was assessed based on RMSE and R2 for calibration and validation sets, and on concordance correlation coefficient (CCC), which were estimated over 100 replicates of calibration and validation sets. Models with a number of latent variables of 5, 2 and 3 were identified as optimal for PE, EnE, and LipG, which explained 34.64%, 55.42% and 82.68% of the total variation in PE, EnE, and LipG, respectively. Significant CCC were found between predicted and observed values for PE (0.50), EnE (0.70), and LipG (0.90). In conclusion, individual feeding behaviour traits can better predict EnE and LipG than for PE under dietary protein restriction when fed ad libitum. ImplicationsThis study suggests that five feeding behaviour traits, which are automatically recorded via feeder stations in large numbers with little effort, together with body weight and age, may be used to predict protein efficiency, energy efficiency and lipid gain in Swiss Large White pigs receiving a protein reduced diet with considerable accuracy. This will allow for easy collection of large amounts of data on these traits for precision feeding and genetic selection strategies, especially when additional traits are added in the future to further improve accuracy.