Comparison of 2-pool and 3-pool digestion kinetic model predictions of neutral detergent fiber digestibility of forages from commercially available data

ABSTRACT: Sizes and rates of potentially digestible (B) and undegradable (C) pools of amylase-treated neutral detergent fiber (aNDF) are used to predict ruminal aNDF digestibility (raNDFD%) in widely used dairy cattle diet formulation programs. An exponential 3-pool (3P) model has been suggested for estimating digestion kinetics parameters for this purpose; however, the approach has not been compared with using a simpler exponential 2-pool (2P) model, nor with using commercial laboratory data on which applications would rely, nor on model effect on predictions of raNDFD%, which is the aim of their application. Our objective was to determine whether the 2P or 3P model most accurately and efficiently characterizes aNDF digestion kinetics and whether the models differ in predicted raNDFD%. Dry forages and silages (6 alfalfas, 6 species of grasses) were analyzed by 2 commercial laboratories that each performed 2 in vitro incubation runs with mixed ruminal microbes, with samples and blanks in duplicate at each of 11 time points; residual aNDF (Ut) was measured at each time point. Sampling hours (t) were 0, 3, 6, 12, 18, 24, 30, 48, 72, 120, and 240 h. Outlier Ut values were removed. Pools as proportions of aNDF were B in 2P, B1 rapid and B2 slow in 3P, and C in both; B pools have digestion rates (kd, h−1; denoted as kd“Bpool”) and lag (h). Models were fit to data for each forage in each incubation with equations 2P: Ut = B × e(−kdB × z) + C and 3P: Ut = B1 × e(−kdB1 × z) + B2 × e(−kdB2 × z) + C, where z = [−(lag − t − |t − lag|)/2]. There were 48 curves for each model. Parameters were estimated with the optim function in base R. The Akaike information criterion (AIC) was used to select the model with the best fit for each forage in each incubation: 16 3P and 32 2P curves were selected. Expressed as (difference between runs)/mean, average deviations between runs for laboratories 1 and 2, respectively, were as follows: for 3P, B1 = 0.50, 0.17; B2 = 0.26, 0.33; C = 0.50, 0.06; kdB1 = 0.81, 0.32; and kdB2 = 0.93, 0.54; for 2P, B = 0.04, 0.01; C = 0.07, 0.01; and kdB = 0.17, 0.08. Estimates of raNDFD% for 2P and 3P were calculated with no lag at passage rates (kp) reported for forages of 0.02 through 0.07 h−1. T-tests determined whether differences were ≠ 0 for 2P − 3P for raNDFD% at each kp for each feed evaluated. With 2P minus 3P differences in raNDFD% listed sequentially by 0.01 h−1 from kp = 0.02 to 0.07 h−1, for 16 AIC-selected 3P curves, differences were −0.29, −0.45, −0.65, −0.87, −1.04, and −1.20%, and for 32 AIC-selected 2P curves, values were −0.15, −0.13, −0.14, −0.17, −0.19, and −0.22%. Some differences were significant, but all were quite small. With little difference between models, use of the more complex 3P conferred no advantage over 2P for prediction of raNDFD% in this dataset.