Abstract #234

# 234
Quantifying the relation between diet branched-chain AA and production responses: A meta-analysis.
Y. Roman-Garcia*1, L. E. Moraes1, M. Socha2, J. L. Firkins1, 1Department of Animal Science, The Ohio State University, Columbus, OH, 2Zinpro Corporation, Eden Prairie, MN.

We studied the relationship between the intake of branched chain AA and the production of lactating dairy cows. A meta-analysis of 274 studies and 1036 treatment means from published dairy cow experiments was conducted using a Bayesian hierarchical modeling approach. Response variables were: milk yield (MY; kg/d), energy-corrected milk (ECM; kg/d), milk protein yield (kg/d) and milk fat yield (kg/d). Three levels of models were developed. Level 1 included the diet variables dry matter intake (DMI; kg/d), rumen degradable protein (RDP; % DM), rumen undegradable protein (RUP; % DM), starch (% DM), neutral detergent fiber (NDF; % DM), fatty acids (FA; % DM), and ratio of acid detergent fiber and NDF (ADF/NDF). Level 2 variables included variables in level 1 + Leu (% DM), Ile (%DM), and Val (% DM). Level 3 included level 1 and 2 variables + rumen NH3-N (mg/dl), rumen pH (pH), rumen isovalerate (Ival, % of VFA), and rumen isobutyrate (Ibut, % of VFA). Initially, we used an all possible models selection approach using linear mixed effects models with the random effect of study. The model with the lowest Akaike Information Criterion was then fitted as a Bayesian hierarchical model with minimally informative priors. Models were evaluated using 5-fold cross validation and variance inflation factors. In general, level 2 models always improved compared with level 1 models. For instance, the MY model (CCC = 0.61), predicts MY = 1.78 + 0.891(DMI) + 0.857(RUP) – 0.0415(RUP2) – 0.0744(NDF) + 0.449(FA) + 16.4(Ile) – 12.3(Ile2) + 2.73(Val). MY had a positive response to Ival and Ibut in the level 3. Likewise, the model for ECM (CCC = 0.68), predicted responses as ECM = – 4.26 + 1.02(DMI) + 1.16(RUP) – 0.0606(RUP2) + 4.67(ADF/NDF) + 15.4(Ile) – 8.51(Ile2). Due to high correlation between RDP and RUP they were interchangeable in model selection without much change the response of BCAA. For MY and ECM RUP yielded a better model. Overall, utilizing the diet branched chain AA as predictors improved models over other diet variables. Ile was present in all the models with a positive coefficient at a decreasing rate, suggesting an important role of Ile for dairy production.

Key Words: branched-chain VFA, Bayesian