Abstract #M267

# M267
Towards the compositional prediction of the ruminal microbial community using temporal modeling in healthy and milk depressed states.
Cameron Martino*1, Grant Gogel1, James Gaffney1, Alfonso Lago2, Mallory Embree1, 1Ascus Biosciences, San Diego, CA, 2DairyExperts Inc, Tulare, CA.

Sixteen ruminally cannulated cows, 8 Holsteins and 8 Jerseys, were used in a milk fat depression (MFD) model to characterize the temporal changes of rumen bacterial populations in cows shifting between a healthy, MFD, and recovery state. The experiment consisted of a 10-d covariate period (Cov) followed by a 10-d MFD induction (Ind), and an 18-d MFD recovery (Rec). Animals were fed a common TMR (16.3% CP, 37.3% NDF, 0.67 Mcal of NEI/lb) during the Cov and Rec. During the Ind, animals were fed a low-fiber, high-starch diet that caused a 0.6% and 1.5% mean decrease in milk fat in Jersey and Holstein cows, respectively. All animals were milked and fed twice a day in addition to daily rumen sampling. Bacterial populations were characterized via 16S rRNA gene amplicon sequencing of rumen samples. MFD induced substantial transformations in the rumen bacterial populations (Cov vs. Ind vs. Rec, P = 0.001) and increased α diversity during Ind (P < 0.01). The resulting operational taxonomic unit (OTU) table was centered-log ratio (clr) transformed and bi-clustered to reveal 2 unsupervised naturally underlying group fluctuations amplified during MFD induction. Of the 2 groupings, 4 of the most universally fluctuating bacterial classes showed significant linear correlation between relative abundance and milk fat percentage during Ind. The 4 classes were Fibrobacterales (group 1, R2 = 0.64, P = 0.0072), Clostridiales (group 1, R2 = 0.57, P = 0.022), Bacteroidales (group 2, R2 = −0.66, P = 0.0056), and Selemonadales (group 2, R2 = −0.16, P = 0.55). The 2 groups’ respective combined abundunace plotted over time revealed an oscillatory nature and fit well to generative Lotka-Volterra models. The dynamics of the resulting model exhibited stable oscillatory behaviors (λ = −0.44, 0.44) with a cyclic periodicity of 6 d. Ordinary least squares regression on compositional balances were applied to the data set and results indicate that the composition of microbial communities can be accurately predicted (R2 = 0.81 MSE = 4.0) from daily environmental and milk composition data.

Key Words: microbiome, milk fat depression, sequencing