B. S. Oldick, J. L. Firkins1, and N. R. St-Pierre
The Ohio State University Department of Animal Sciences
Predictions of microbial nitrogen (N) flow to the duodenum based on the intake of net energy of lactation (NEL; megacalories per day) were improved compared with the equation in the current National Research Council’s Nutrient Requirements of Dairy Cattle (NRC, 1989). However, a procedure based on multiple regression using terms for dry-matter intake (DMI) and dietary concentrations of crude protein, neutral detergent fiber (NDF), and concentration of forage in the diet resulted in the following equation:
Microbial N flow, g/day = 16.1 + (22.9 x DMI, kg/d) - (0.365 x DMI2) - (1.74 x NDF, % of DM)
Many models that are used for estimating nutrient requirements or for ration balancing for ruminants rely on a prediction of microbial flows of protein (N x 6.25) or amino acids (protein x % amino acids in the protein) to the small intestine. After this calculation is made, then the amount of protein (or amino acids in newer models) that is needed from undegraded (bypass) protein (amino acids) can be estimated. Furthermore, microbial protein provides about 60% of the protein reaching the duodenum, and maximizing microbial protein production generally is the cheapest way to supplement protein for cattle.
Microbial protein flow to the duodenum is based on availability of energy (primarily carbohydrates from starch and fiber), ruminally degraded protein, and other factors that are difficult to quantify and that often interact with each other. Therefore, the estimation of microbial protein flow will need to be done with mechanistic models (describing the biological processes) or with empirical models (based on a statistical best fit to the data). Our approach was to use regression techniques for an empirical approach to be used under a variety of circumstances to improve the accuracy of estimation of microbial protein flow compared with current techniques.
Data from 213 treatment means from 55 published studies using duodenally cannulated cattle were selected based on criteria related to procedures used and data actually reported by the authors. Briefly, statistical procedures relied on multiple regression of measured microbial N (dependent variable) and NEL intake as the independent variable or else the following independent variables – measured DMI and concentrations (DM basis) of NEL, crude protein, NDF, forage percentage, and all two-way interactions. Nonsignificant terms were removed systematically; in some cases, significant (P < 0.05) terms were removed if they had the highest remaining probability (P) of just resulting from chance until a statistical test (variance inflation factor) suggested that autocorrelation of remaining independent variables was below an accepted value. This would reduce the likelihood of a combination of unusual dietary concentrations causing an unreasonable inaccurate prediction of microbial N flow (i.e., "overparameterized" model). Additionally, data from each published trial were weighted for the variation in that trial. That is, a mean with low variation would be given a greater emphasis in the prediction equation than a mean with a high variation. Finally, each published trial used different types of animals, different procedures, different types of diets, etc. Therefore, trial was included as a class variable. The resulting equations were adjusted to the average effect of trial.
The resulting regression equations based on NEL intake were improved compared with the equation used by the current NRC (Figure 1A), especially when data were adjusted for trial effects (Figure 1B). When dietary concentrations were included, the resulting equation (Figure 2) tended (P < 0.10) to fit better than that reported in Figure 1. The quadratic fit reaches a plateau at DMI of 29.0 kg (63.8 lb) per day, and it therefore starts predicting a declining microbial N flow as DMI exceeds 64 lb/day. For dairy cows eating more than 64 lb/day, an asymptotic model (no declining phase) may provide a safer prediction (no figure shown) until more data with cows with DMI > 60 lb/day are available:
479 (1 - 0.999 x e-0.0573 x DMI(kg/day)) - (1.72 x NDF, %).
No patterns in the residuals were detected for any models. These prediction equations provide a more accurate prediction of microbial protein flow over a wide range of conditions from the data set – DMI from 7.5 to 59.0 lb/day, forage from 0 to 100%, NDF from 17.0 to 59.4%, and crude protein from 7.8 to 24.8%. The equation in Figure 2 is recommended unless DMI exceeds the data range (59 lb/day) or at least the predicted plateau (64 lb/day), at which DMI the asymptotic model is recommended.
The equations presented are reflective of effects of DMI (the predominant effect) and dietary energy concentration (NEL or NDF; the concentration of NDF is inversely related to energy concentration). These can be used for general prediction purposes over a wide range of conditions, including for nonlactating cattle (excluding young calves consuming < 7.5 lb/day of DM). Dietary concentrations of ruminally degraded protein were removed from the regressions to prevent overparameterization. In feeding situations, however, other models or guidelines would be needed to ensure adequate ruminal pH and degraded protein. These equations should be considered for general use until more mechanistic approaches get refined and further validated for a wide range of dietary conditions.
1 For more information, contact at: The Ohio State University, 223 Animal Science Building, 2029 Fyffe Road, Columbus, OH 43210; (614) 688-3089, Fax (614) 292-1515; email:firkins.1@osu.edu