| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Animal and Veterinary Science and
2 Statistical Programs, College of Agricultural and Life Sciences, University of Idaho, Moscow 83844
Corresponding author: A. N. Hristov; e-mail: ahristov{at}uidaho.edu.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
100 d in milk). A mixed model regression analysis was used to completely investigate the potential relationships among composition variables and DMI, MY, and MPY. Protein and carbohydrate fractions were the main components within the DMI models, and DMI played a dominant role in estimating MY and MPY. Inclusion of stage of lactation substantially improved the MY models but did not affect model fits or residual structure for DMI and MPY.
Key Words: diet composition • dry matter intake • milk yield
Abbreviation key: AC = all cows, BIC = Bayesian Information Criterion, CHO = carbohydrate, DIM 1 = early lactation cows, DIM 2 = mid and late lactation cows, MY = milk yield, MPY = milk protein yield, PE = milk protein N efficiency
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
). Martin and Sauvant (2002) reported a relatively high correlation (r = 0.52 to 0.85), depending on parity and feed allocation system, between DMI and MY in mid and late lactation cows, emphasizing the importance of DMI as a driving force in milk synthesis. A number of regression models designed to predict DMI based on animal factors have been developed (Blaxter et al., 1966; Yungblut et al., 1981; Kertz et al., 1991; Roseler et al., 1997a) and evaluated (Fuentes-Pila et al., 1996; Roseler et al., 1997b; Pittroff and Kothmann, 2001). Dietary ADF (Yungblut et al., 1981) and NDF and CP (Harlan et al., 1991; Holter et al., 1996, 1997; Rayburn and Fox, 1993) concentrations have been proposed as predictors of DMI. Mertens (1987) indicated the need to integrate animal and feed characteristics into the modeling process to improve intake prediction and understand the relationships among physiological and physical factors regulating DMI in ruminants.
|
The objective of this study was to determine the impact of dietary components estimated with 2 diet evaluation models on DMI, MY, and MPY in Holstein dairy cows.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
100, 309 observations; DIM 2). The earliest DIM record was 3, and the latest DIM recorded was 247. Average DIM was published for 575 observations; for 248 observations, DIM was indicated as early, mid, or late. Cows in early lactation were assigned to the DIM 1 group, and cows in mid and late lactation were assigned to the DIM 2 group. For 23 observations, DIM of the cows was not published, and these observations could not be used for structural analysis. Not all variables were present across all observations in the data set. Hence, the number of observations used for regression varied from model to model, depending on the regressor variables contained in each model.
The response variables of interest were DMI, MY, and MPY. The average (±SD) DMI of the cows involved in this study were 22.1 ± 2.67 kg/d with a minimum of 12.2 and a maximum of 29.9 kg/d. Average MY was 31.9 ± 5.61 kg/d and had a minimum of 20.1 and a maximum of 50.1 kg/d. Average milk fat and milk protein concentrations were 3.4 ± 0.38 and 3.1 ± 0.21%, respectively. Average milk protein N efficiency (PE) [(MPY ÷ 6.38) ÷ N intake] was 24.7 ± 3.99%, with a minimum and maximum of 13.7 and 39.8%, respectively. Body weight of experimental cows was published in 552 of the observations used in the analysis. Mean BW was 602 ± 46.6 kg (minimum = 476 kg; maximum = 750 kg). The dietary variables derived from the feeding programs and considered in the structural analysis are shown in Figure 2. Ratios among protein and energy variables and BW were also considered for potential relationships with the response variables.
|
for abbreviations): ME, FermNDF, FNDF, FermB1, FCHOB1, and FCHO (CPM data set) and MCP, MPBact, MPFeed, AA, DigAA, ME, UTDN, discounted TDN (DTDN) (NRC data set). Multiple linear regression was chosen as a means for investigating the structural relationship among the responses, DMI, MY, and MPY, and the nutritional attributes obtained from CPM and NRC. Because the data used in this meta-analysis were collected from multiple studies conducted over many years, the random effects of each study needed to be accounted for in the modeling process (St-Pierre, 2001). In this case, random effects were incorporated into the estimation procedure using a mixed model framework. The general form of the mixed model is as follows:
 | ([1]) |
where Y is the response being modeled, X is a matrix of the nutritional variables, and ß is a vector of the corresponding regression coefficients. These terms represent the fixed portion of the model and are equivalent to those found in a standard multiple linear regression. The additional components, Z and 
, account for the random effects caused by the various studies; Z represents either all or a portion of the variables present in X; is a vector of their associated coefficients; and 
is a random error term assumed to be normally distributed with a mean of 0 and a constant variance.
An additional factor for consideration was the precision of each study. To account for this, each regression analysis was weighted by the reciprocal of the squared response standard errors, as reported in each study (St-Pierre, 2001). Weighting in this manner can improve the overall precision of the regression estimates and ensure homogeneity of variance for the model.
To identify a final model for each response, the backward elimination technique described by Oldick et al. (1999) and Firkins et al. (2001) was used for variable selection. Initially only main (linear) effects were considered in the process because of the large number of variables involved and stability problems created by interaction and quadratic terms. At each stage of elimination, the P value and variance inflation factor of each variable were evaluated as described in Oldick et al. (1999). Variables with P > 0.10 or a variance inflation factor >30 were removed from the model. Once the selection process had identified a reduced model, all possible two-way interactions and quadratic terms derived from the remaining variables were added to the model. The backward elimination procedure was then repeated with the exception that main effects remained in the model, even when nonsignificant, if they were involved in any significant interaction or squared term.
A final factor considered for the modeling process was the effect of DIM. For each model identified by the process outlined previously, a second model was constructed. In these latter models, DIM entered the model as a dummy variable representing the 2 DIM classes: DIM 1 and DIM 2. The DIM variable allowed for estimation of separate coefficients for each variable in the model including intercept terms. These coefficients were then tested for equivalency across DIM classes using single degree of freedom contrasts. When no significant difference was indicated, the dummy variable designation was removed for that model component.
Estimation for all models was carried out using the method of REML, assuming a diagonal variance-covariance (variance component) structure among the regressors. All estimated models were assessed, and a final model for each response was selected based on the Bayesian Information Criterion (BIC), model parsimony, parameter significance, residual analysis, and biological relevance. In addition, mean and linear biases of the selected models were assessed following St-Pierre (2003). All statistical computations were carried out using the PROC MIXED, PROC REG, and PROC GPLOT procedures of the SAS software system (SAS, 1999).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
. Corn silage, alfalfa silage, and hay were the predominant forage components of the diets. The data set included diets that varied widely in forage-to-concentrate ratio. Corn and barley grains were the main energy concentrates. Solvent-extracted soybean meal was the main protein supplement. Corn gluten meal, fishmeal, blood meal, and meat-bone meal were the sources of RUP. Animal fat and urea were used in 124 and 131 diets, respectively.
|
Concentration and ruminal fermentability of protein and CHO dietary fractions were estimated using CPM Dairy. The diets surveyed encompassed a significant range of soluble protein and fraction A, B1, and B2 proteins. Protein fraction B3 was not studied because of its low dietary concentration and unlikely effect on ruminal microbial protein synthesis and MY and composition. Similar to the protein fraction, diets varied significantly in CHO composition.
Efficiency of N utilization for milk protein (PE) varied considerably among diets. In general, the diets that produced high PE contained more corn and cereal silages and concentrate and less alfalfa forage. For example, 29% of the diets with PE >30% (77 diets; average PE = 33%) contained alfalfa silage compared with 50% for all diets in the study. Corn silage was fed with 74% of the high PE compared with 57% of all diets; average concentration of corn silage in dietary DM was similar: 35 and 32%, respectively. The high PE diets more often contained corn (77% of the diets compared with 61% for all diets) and barley grain (29% vs. 15%, respectively). Average CP, NEL, discounted TDN (DTDN), and NFC concentration (NRC, 2001) of the high PE diets was 15.8%, 1.64 Mcal/kg, 66.7%, and 43.4%, respectively, and the average for all diets was 17.8%, 1.61 Mcal/kg, 65.2%, and 41.5%, respectively. Compared with the average MY per cow from all diets, cows produced more milk in trials where PE was high: 35.2 vs. 31.9 kg/d, but milk protein concentration was not different, 3.10 and 3.11%, respectively.
Modeling
For each response, the backward selection procedure was used to identify a model that had acceptable residual structure, BIC, and parsimony. All final models showed good prediction with residuals that had no discernible trend or pattern. All model residuals were tested for significant mean and linear biases. In all cases, the biases were either non-significant or small relative to the measured standard errors based on analyses similar to St-Pierre (2003). Residual plots for the models with the smallest BIC value (AC or DIM models) are shown in Figures 3 through 5.
|
|
(the reduced DIM model had identical structure to the AC model). The residual plot is shown in Figure 3. Variables included from the CPM data set were CP, NEL, metabolizable protein, metabolizable protein from bacterial protein, and protein fraction B1. The model also contained significant quadratic effects for metabolizable protein from bacterial protein and NEL. The significant random component for metabolizable protein indicated an interaction of metabolizable protein with the random effect of study.
|
MY models.
Estimated MY models are shown in Table 3. Dry matter intake, metabolizable protein, metabolizable protein from bacterial protein, and dietary concentration of CHO fraction B1 were significant factors in the CPM-AC model. Metabolizable protein from bacteria had a negative influence in the model. Incorporation of DIM into the CPM model substantially improved model fit as measured by the 140-point decrease in BIC as well as an improved relationship between observed and predicted values and overall residual structure. Dry matter intake showed significantly different coefficients for each DIM class, and the remaining coefficients did not. In both the full and reduced models, DMI and metabolizable protein had significant interactions with the random effects of study.
|
). The significant random component for DMI indicated an interaction of DMI with the random effect of study.
|
. Dry matter intake, dietary concentrations of soluble protein, non-structural CHO, CHO fractions A and B1, protein fractions A and B1, and interactions among DMI and protein fractions and CHO fraction B1 and protein faction B1 were indicated as significant factors in the CPM model. Both CHO and protein fraction effects had negative coefficients. Inclusion of DIM into the model resulted in separate coefficients for DMI, nonstructural carbohydrates, CHO fraction B1, and protein fraction A, but did not improve the model fit or residual structure. The models also contained significant interaction of DMI and soluble protein/protein fraction A and CHO fraction B1 and protein fraction B1. In both models, DMI and protein fraction A had significant interactions with the random effects of study.
|
). Both models contained a number of significant interactions between dietary components and quadratic terms for DMI, metabolizable protein, and BW. The DIM model included separate coefficients for DMI, the interaction of DMI and BW, and the quadratics term for DMI. In both models, DMI significantly interacted with the random effects of the study. |
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Regarding chemical composition, the diets represented a large variation in protein, energy, fiber, starch, and discounted TDN concentration. Diets with CP between 17.0 and 19.5% of DM represented 63% of the total, and those with an NEL concentration of 1.56 to 1.67 Mcal/kg of DM represented 69% of the total. Rumen-degradable protein and RUP concentrations of 10.9 to 13.7 and 5.0 to 6.5% of DM represented 66 and 60%, respectively, of the total diets. Diets with NDF and forage NDF concentrations of 26.8 to 32.9 and 16.8 to 27.1% of DM were, respectively, 65 and 73% of the total, and diets with ADF concentration of 18.3 to 21.5% of DM were 52% of the total. Calculated discounted TDN concentration varied less among diets; 75% of the diets had discounted TDN concentration between 63.6 and 67.5%. A similar variation in dietary CP, NDF, and ADF concentration from a comparatively smaller data set was reported by Holter et al. (1997). Thus, composition of the diets is in line with practical feeding and should ensure that the results from this study are applicable to most diets fed to dairy cows in North America.
In this data set, extremely high PE were associated with comparatively low dietary CP concentrations and very high MY (Bach et al., 2000). Thus, high PE was more often found with diets in which corn silage rather than alfalfa silage was the main forage ingredient, CP concentration was low, and cows were producing higher MY. Not surprisingly, dairy efficiency (kg of solids-corrected MY/kg of DMI) was negatively correlated with DMI and positively correlated to MY in Holstein cows (Britt et al., 2003). Satter et al. (2002) calculated that with diets in which low DM alfalfa silage is the only forage, 20.8% dietary CP, as predicted by NRC (2001), was needed to meet the requirements of a high-producing, non-pregnant dairy cow; with diets in which the forage was alfalfa hay and corn silage (1:1), the requirements of the cow would be met at 15.8% CP. Wilkerson et al. (1997) reported proportionally greater urinary N losses and lower PE with low-producing cows (<20 kg/d) than with high-producing cows (>20 kg/d milk): 37.9% vs. 34.6% and 22.0% vs. 29.7%, respectively. It has to be noted, however, that the high-producing cows excreted a total of 143 g/d per 1000 kg BW more N than the low-producing cows. As a proportion of intake, the total excreta N between the two groups was not dramatically different: 72.6% vs. 68.9% for low- and high-producing cows, respectively.
Both CPM and NRC DMI models included energy and protein components. The CPM model contained a quadratic term for NEL, indicating a non-linear response to this variable. The NRC model included negative terms for NEL, forage NDF, and ADF. High fiber diets may physically limit intake through the so-called fill effect (Mertens, 1987), and the effect of high energy diets, or diets with high concentration of rumen-degraded starch on DMI is likely related to biochemical control (Montgomery and Baumgardt, 1965; Faverdin and Bareille, 1999). Metabolizable protein and a component related to ruminal availability of protein (protein fraction B1, true protein, or RDP) were present in both DMI models. The CPM model indicated a negative coefficient for metabolizable protein from bacteria protein. In most feeding systems, microbial protein synthesis is assumed to be energy-dependent (NKJ Protein Group, 1985; Tamminga et al., 1994; NRC, 2001; GfE, 2001) and increased production of metabolizable protein from bacteria protein would indicate increased overall digestibility of dietary energy. Inclusion of DIM did not improve the DMI models, suggesting little influence in the response because of stage of lactation. As a significant random effect, metabolizable protein (CPM) and dietary fat (NRC) appear to be a source of variability in the DMI response across studies. The difference in BIC (530 points) indicated a better fit with the NRC than with the CPM models. Body weight is routinely a factor in models for predicting DMI in dairy cows (Yungblut et al., 1981; Kertz et al., 1991; Roseler et al., 1997a; NRC, 2001); however, its effects might have been mitigated in these data when the random effect of study was accounted for. Because researchers often strive to balance the BW of animals within their studies, the random effect of BW on the responses may tend to vary more across studies than within them. Therefore, BW may well be confounded with the random effect of study, and this particular data set may not be well suited for studying its effects. In fact, BW was typically nonsignificant in the estimated models.
The models derived for MY and MPY reflected the relationships existing among the dietary variables and production parameters in this analysis. Although a variety of factors were involved in predicting MY and MPY in both nutritional programs, the major variable influencing production in this data set was DMI (largest F statistic). The energy component in the MY models was either CHO fraction B1 (starch; CPM) or dietary fat (NRC). Similar to the DMI models, metabolizable protein was important in predicting MY in both datasets. In both the CPM and NRC data, the full model incorporating DIM improved model fits for MY. Dry matter intake (and metabolizable protein for the CPM model) was a significant random component, indicating DMI contribution to MY variability over the studies. Both CPM and NRC datasets provided similar fits for MY.
Milk protein yield models derived through the CPM dataset were dominated (largest F statistic) by DMI, but a number of protein- and CHO-related dietary factors were significant in predicting MPY. In general, the MPY models were not very parsimonious and should be considered less reliable than the DMI or MY models. Dry matter intake, dietary NDF concentration, and digestibility of dietary OM were important in predicting MPY in the models of Rook et al. (1992). Moloi (1998) found that diet-related variables, such as degradable intake protein, NDF, and NFC concentration of the diet, impacted MPY in dairy cows. In the Moloi (1998) study, however, no single variable was distinguished as having major influence on MPY. Smoler et al. (1998) found generally poor prediction of milk protein concentration when diet-related variables were used in the modeling process; models included both CHO and protein variables. Inclusion of stage of lactation in the CPM-MPY model did not improve the model fit. The NRC-MPY models also included metabolizable protein, signifying the importance of the postruminal protein supply in milk protein synthesis. Concentrations of CP and NDF had negative coefficients in the NRC-MPY model. Similar results for NDF were reported by Rook et al. (1992), and these may be indicative of the negative relationship found by Emery (1978) between dietary fiber and milk protein concentration. Increased CP concentration of the diet has resulted in decreased efficiency of utilization of dietary N for milk protein synthesis (Etter et al., 2003) and lowered milk protein content (Leonardi et al., 2003). As evident from the difference in BIC (329 and 327 points, AC and DIM models, respectively), the NRC models improved the prediction of MPY relative to the CPM models.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
Received for publication November 17, 2003. Accepted for publication January 26, 2004.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. N. Hristov, W. Hazen, and J. W. Ellsworth Efficiency of use of imported nitrogen, phosphorus, and potassium and potential for reducing phosphorus imports on idaho dairy farms. J Dairy Sci, September 1, 2006; 89(9): 3702 - 3712. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
V. R. Moreira and L. D. Satter Effect of scraping frequency in a freestall barn on volatile nitrogen loss from dairy manure. J Dairy Sci, July 1, 2006; 89(7): 2579 - 2587. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. L. Firkins, A. N. Hristov, M. B. Hall, G. A. Varga, and N. R. St-Pierre Integration of Ruminal Metabolism in Dairy Cattle J Dairy Sci, March 1, 2006; 89(e_suppl_1): E31 - E51. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. N. Hristov, W. J. Price, and B. Shafii A Meta-Analysis on the Relationship Between Intake of Nutrients and Body Weight with Milk Volume and Milk Protein Yield in Dairy Cows J Dairy Sci, August 1, 2005; 88(8): 2860 - 2869. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. N. Hristov, R. P. Etter, J. K. Ropp, and K. L. Grandeen Effect of dietary crude protein level and degradability on ruminal fermentation and nitrogen utilization in lactating dairy cows J Anim Sci, November 1, 2004; 82(11): 3219 - 3229. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
