Predicting microbial protein synthesis in beef cattle: relationship to intakes of total digestible nutrients and crude protein.

Prediction of microbial CP (MCP) synthesis in the rumen is an integral part of the MP system. For the NRC beef model, MCP is calculated as 0.13 multiplied by TDN intake (TDNI), with adjustment for physically effective NDF (peNDF) concentrations less than 20%. Despite its application for nearly 2 decades, MCP predictions using this approach have not been extensively evaluated. We assembled a database of 285 treatment means from 66 published papers using beef cattle and dairy or dairy × beef crossbred steers, fed diets with a wide range of TDN, CP, and ether extract (EE) concentrations, in which MCP synthesis was measured. Fat-free TDN (FFTDN) concentration was calculated by subtracting 2.25 × percent EE from the TDN concentration. Based on initial model selection procedures indicating that DMI and concentrations of TDN, FFTDN, and CP were significantly (P < 0.04) related to MCP synthesis, linear and quadratic effects of TDNI and FFTDN intake (FFTDNI) and CP intake (CPI) were considered as potential independent variables. Mixed model regression methods were used to fit 1-, 2-, and 3-independent-variable models based on either TDNI or FFTDNI (e.g., TDNI only, TDNI and CPI, and TDNI, CPI, and the quadratic effect of TDNI; or FFTDNI only, FFTDNI and CPI, and FFTDNI, CPI, and the quadratic effect of FFTDNI). True ruminal OM digested (TROMD; g/d) was highly related (r(2) = 0.84 using citation-adjusted data) to MCP synthesis. Similarly, both TDNI and FFTDNI were highly related to citation-adjusted TROMD (r(2) > 0.96) and MCP synthesis (r(2) > 0.89). Models with FFTDNI were slightly more precise with slightly smaller prediction errors than those with TDNI. Randomly dividing the citations into Development (60%) and Evaluation (40%) data sets indicated that models such as those derived from the overall database accounted for 46 to 56% of the variation in MCP synthesis, with neither mean nor linear bias (P ≥ 0.26). In contrast, calculating MCP as 0.13 × TDNI, with or without adjustment for peNDF concentration, resulted in overprediction of MCP (P < 0.001 for both mean and linear bias). Cross-validation using 5,000 randomly drawn training and testing data sets yielded results similar to the Development/Evaluation approach. Recommended equations are provided, but the errors of prediction associated with these empirical regression equations were on the order of 25 to 30% of the mean MCP.