Recipient Organization
UNIV OF WISCONSIN
21 N PARK ST STE 6401
MADISON,WI 53715-1218
Performing Department
DAIRY SCIENCE
Non Technical Summary
In states like Wisconsin, where milk is paid to producers based on solids (fat and protein) content, and butter is the most important component of milk prices, losses in milk fat production in addition to the historically low milk prices can put dairy producers out of business. Milk fat depression is a metabolic syndrome in which milk fat yield can decrease as much as 50%, without affecting the yield of milk or other solids. Producers have been aware for many decades of milk fat losses in lactating dairy cows fed diets with large proportion of soluble carbohydrates, mainly corn. However, it was not until recently that researcher identified the causes of the milk fat depression syndrome, which are related with higher rumen yield of partially biohydrogenated fatty acids. Despite the fairly clear understanding of the causes of this syndrome at cellular level, models to predict rumen yield and absorption of milk fat depressing CLA in a timely manner are still lacking, which results in continuous losses of milk fat for dairy producers.The long-term goal of this proposal is to develop those needed prediction models and to generate computational in-farm tools for early detection and prevention of milk fat depression. To achieve our long-term goal, within the scope of this proposal, we aim to use NIR spectral data from individual feed ingredients and total mixed rations to predict rumen environments that promote partial biohydrogenation of polyunsaturated fatty acids and yield of milk fat depressing CLA. Second, we will develop an optimization tool to alert nutritionists and dairy producers about potential milk fat losses at the ration formulation stage, allowing them to take preventative actions. Finally, we aim to predict rumen yield of specific CLA known to depress milk fat production from milk mid-infrared (MIR) spectra.Preventing milk fat depression syndrome-related losses will increase nutrient efficiency, enhance the quality of the milk produced, and reduce emission of greenhouse gases per unit of milk produced.
Animal Health Component
40%
Research Effort Categories
Basic
30%
Applied
40%
Developmental
30%
Goals / Objectives
The long-term goal of this proposal is to develop a farm tool to predict milk fat depression and to prevent losses related with it. Within this proposal, we aim to develop robust, infrared-based models to detect risk of milk fat depression at the ration formulation stage. We hypothesize that full spectra infrared data from feeds (NIR) and milk (MIR) can be strong predictors of ruminal yield of milk fat depressing-CLA.The specific objectives are:1. Train one Ph.D. student on the multidisciplinary field of lactating dairy cow metabolism, high throughput data management, and machine learning-based modeling.2. Coupling feed NIR data and machine learning algorithms to predict rumen environmentsand outputs that favor milk fat depression.3. Develop adequate dietary strategies, through model optimization, that prevent partial biohydrogenation of dietary fatty acids and undesired rumen yield of CLA.4. Use milk mid-infrared spectroscopy to map milk fat depressing compounds in post ruminal fluid.The complementary profile of the co-PIs (molecular metabolism and empirical, big-data based modeling) will allow this project to address three scientific aims, all coordinated to address economical losses of Wisconsin and U.S. dairy farmers related with milk fat depression in lactating dairy cows. These are:1) Use milk mid-infrared spectroscopy to map milk fat depressing compounds in post ruminal fluid;2) Coupling feed raw NIR data and machine learning methods to predict rumen environments and outputs that favor milk fat depression; and3) Develop adequate dietary strategies through model optimization to prevent partial biohydrogenation of dietary milk fatty acids
Project Methods
Aim 1: Coupling feed NIR data and machine learning algorithms to predict rumen environment profiles that increase yield of milk depressing CLA.Hypothesis: Dietary NIR spectra accurately predicts changes in rumen environment that favor partial biohydrogenation of dietary fatty acids and yield of milk fat depressing CLA.Approach: Rumen environmental factors critical for partial biohydrogenation of milk fatty acids will be predicted by raw spectra NIR of dietary ingredients. We will also investigate the association between rumen environment (i.e. pH, ammonia concentration, rumen microbial population, etc.) and omasal bioactive fatty acids. Thirty-two ruminally cannulated, peak lactating (> 60 DIM) dairy cows will be fed 8 contrasting diets expected to generate a wide range of rumen biohydrogenation conditions. Energy and protein sufficient diets will contrast on the content of starch (25 or 35%), C18:2 fatty acids (≈1 or 3%), and ruminally degradable protein (8 or 12%) in a 2x3 factorial arrangement of treatments.Cows will be grouped in 8 latin squares, with 4 thirty-day periods each. The latin squares will be repeated at different levels of RDP (n=64/treatment). Furthermore, due to facility, herd and student capacity, four of the squares (sixteen cows) will be run in year one, and four in year two of the grant. The sample size is large enough to have a nontrivial power (> 0.8) to detect important relationships among variables with a correlation of at least 0.35. Considering the model data structured by treatment, we should be able to detect, with similar power, even lower correlation levels.Variables of rumen environment that affect the yield of partially biohydrogenated fatty acids, including pH, concentration of individual volatile fatty acids and ammonia will be measured in samples collected every 6 hours on the last two days of each thirty-day period. Omasal flow will be collected at the same timepoints. Samples will be combined per period, and lipids will be extracted from the combined omasal fluid sample by the chloroformmethanol-water method (Folch et al., 1957). Targeted CLA (i.e. C18:2 trans-10, C18:2 cis-12, trans-9, C18:2 cis-11, cis-10, C18:2 trans-12; and C18:2 trans-7, cis-9) will be separated by liquid chromatography and measured by mass spectrometry using the internal standard method (Rodriguez-Alcala and Fontecha, 2007). Such bioactive fatty acids were selected based on its association with milk fat depression (Bauman et al., 2011).Individual feed ingredients, and total mixed ration and refusal from the last two days of each period will be scan with NIR, and full spectra data will be used in combination with machine learning-based algorithms to develop prediction models of rumen environment. The analysis will start with pairwise correlation analyses between each rumen environment component and the bioactive fatty acids. The Pearson correlation coefficient will be used to study the relationships involving linear (i.e. normally distributed) traits (or transformations thereof), such as rumen fatty acids. As an alternative to the linear Pearson correlation, the Spearman's Rank Correlation and the Kendall's coefficient will be used whenever the normal distribution assumption does not hold, even after variable transformation.Expected outcomes: Dietary NIR spectra will produce robust models of rumen environment and postruminal appearance of milk fat depressing CLAAim 2: Adequate dietary strategies through model optimization to prevent partial biohydrogenation of dietary milk fatty acids.Hypothesis: Diet formulation can be optimized, based on feed composition, to minimize rumen output of fat depression CLA, avoiding decreases in milk fat synthesis.Approach: The data generated in Aim 1 will be used to perform model optimization in order to adequate nutritional strategies and prevent milk fat depression. Variables related to the rumen environment profile, such as rumen pH, bioactive fatty acids, ammonia and others, will be used as input to optimize dietary compounds and hence the targeted milk fat content, or major milk fatty acids profile. To evaluate the impact of feed fatty acid composition on milk fat depression and to optimize diet formulation, we will implement a stochastic non-linear model. This mathematical programming model will be developed to identify optimal levels of the five major dietary fatty acids and to simulate milk fat depression based on the feed fatty acid composition. The objective function in our study will be to maximize milk fat content. The model will be developed using the general algebraic modeling system with the BARON solver (Bussieck et al., 1985).Expected Outcomes: The management decision-making tool generated upon completion of this aim will help scientists and nutritionists to determine optimal levels of feed fatty acids yield from the rumen and to maximize milk fat synthesis.Aim 3. Predict ruminal yield of milk fat-depression CLA from milk mid-infrared spectroscopy.Hypothesis: Milk MIR spectra accurately predicts rumen yield of partially biohydrogenated compounds known to depress milk fat production. We have previously demonstrated that non-esterified fatty acids (NEFA) in plasma can be predicted from fatty acids in milk (Dórea et al., 2017). Given that partially biohydrogenated fatty acids in the rumen are minimally metabolized by the lactating dairy cow, we expect the same will be possible with postruminal CLA. Furthermore, technologies for high-throughput phenotyping, such as milk MIR spectroscopy, have been implemented to predict milk fatty acid composition (Soyeurt et al., 2011; Wojciechowski and Barbano 2016). Our research group has combined this technology with machine learning algorithms to predict more complex traits, like feed intake (Dórea et al., 2018). Following a similar approach, we will use full milk MIR spectra and machine learning algorithms to predict rumen yield of the conjugated linoleic acids (CLA) trans-10 cis-12; trans-9, cis-11; cis-10, trans-12; and trans-7, cis-9, the four fatty acids known to have the most depressing effect on milk fat production (Bauman et al., 2011).Approach: To predict rumen output of key partially biohydrogenated CLA known to depress milk fat production from full milk MIR spectra, eight contrasting diets will be fed to 32 ruminally cannulated lactating dairy cows (see Aim 1 for details). Full milk MIR spectra will be measured by a commercial laboratory from combined samples over the morning and afternoon milkings (AgSource Cooperative Services, Verona, WI). All milk samples will be analyzed using MilkoScan FT6000 spectrometers (Foss, Hillerod, Denmark). The Foss MIR spectrum contains 1,060 data points that represent the absorption of infrared light through the milk sample at WL in the 925 to 5,008 cm−1 range. Spectral information from milk will be merged with respective omasal fatty acid data previously analyzed by LCMS (see Aim1). To develop the predictive models, we will explore the use of partial least squares (Dórea et al., 2018a), as well as machine learning algorithms such as artificial neural networks (Dórea et al., 2018; Pralle et al., 2018). Prediction accuracy will be assessed by k-fold cross-validation, such that records from a specific period are not included in both the training and testing sets (Dórea et al., 2018).