| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
* Institute of Animal Breeding and Genetics, University of Göttingen, 37075 Göttingen, Germany
Section of Epidemiology and Biostatistics, Leeds Institute of Molecular Medicine, Cancer Genetic Building, St James’s University Hospital, Leeds, LS9 7TF, United Kingdom
Department of Animal and Poultry Science, University of Guelph, Kemptville, Ontario, K0G 1J0, Canada
Department of Animal Sciences, University of Wisconsin, Madison 53076
1 Corresponding author: skoenig2{at}gwdg.de
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Key Words: milk urea nitrogen • fertility • fertility indicator • genetic parameter
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
A main problem in the genetic evaluation of fertility traits is inadequate data quality. Traits describing pregnancy status (e.g., nonreturn rates measured at 56 or 90 d after a first insemination) strongly depend on voluntary data transfer of milk producers for genetic evaluation when using natural-service bulls. Statistical analysis of fertility in dairy cattle may produce different results when information on true pregnancy status (e.g., rectal palpation done by veterinarians) is available (König et al., 2006, 2007a). In addition, the impact of possible maternal and paternal components and the non-Gaussian distribution of many fertility traits increase the complexity of statistical models and make genetic evaluation more difficult (e.g., König et al., 2007a).
A favorably correlated selection response in reproductive performance through direct selection on closely related traits with greater heritabilities than fertility could overcome some of these obstacles. For example, levels of MUN are routinely recorded at official test days in German Holstein dairy cattle. They show a Gaussian distribution, and they can be used as indicators of the level of metabolic stress, especially in early lactation (Eicher et al., 1999). Hence, a negative correlation between MUN and success of insemination can be expected and was found to be strong in some previous studies (e.g., Canfield et al., 1990; Butler et al., 1996). Evidence of an association between MUN and fertility at the phenotypic level leads to the conjecture that genetic evaluation of MUN could be used to improve reproductive performance as a correlated response, as indicated by Mitchell et al. (2005). A first insemination after calving is usually done in the first third of lactation. Several recent results suggest MUN as an adequate indicator of metabolic status, especially in the early lactation of dairy cows. For example, Wattiaux and Karg (2004) found that MUN concentration was negatively correlated with DMI at 3 wk of lactation, and this correlation was larger in cows with a greater milk yield relative to DMI. Wattiaux and Karg (2004) concluded that MUN may be more reflective of energy and protein balance in the early lactation of dairy cows than the adequacy of dietary inputs would be.
The aims of the present study were 1) to assess the effect of MUN and of classes of MUN-protein percentages (UPC) measured on the first 2 test days after calving on nonreturn rates at the phenotypic level when applying logistic models; and 2) to estimate variance and covariance components among MUN, fertility traits, and milk yield by using bivariate threshold-threshold, linear-threshold, and linear-linear models in a Bayesian framework. Results obtained from the first aim can provide management tools for optimizing feeding strategies toward improvements of dairy cow fertility. Results from the second aim would answer whether MUN and the trait interval from calving to first service (CTFS) could be combined into a breeding goal, with the ultimate perspective of improving fertility in dairy cows by selection.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
.
|
 | [1] |
where 
qrst is the probability of a cow being pregnant 56 d after first service; 
is the overall mean effect; Sq is the fixed effect of season of insemination (January to March, April to June, July to September, October to December); Pr is the fixed effect of parity of the cow (parity 1, parity 2, parity 3, >parity 3); UPCs is the fixed effect of UPC; herdt is the random herd effect; CTFSqrstu is the interval from calving to first service, in days, of cow u; a1, a2, and a3 are the linear, quadratic, and cubic regressions on CTFS; (CTFSqrstu x Pr) is the interaction between CTFS and parity; MYqrstu is milk yield (average of test-days 1 and 2 after calving); and β1 is the linear regression on MY.
Model [2] differed from model [1] in that MUN classes were fitted instead of UPC. Seven classes of MUN were created in the range from 100 to 450 ppm in increments of 50 ppm. This was done to investigate the main effect of MUN on NR56, as opposed to the combined effect of MUN and protein percentage through UPC. The logistic model [2], used to investigate the impact of MUN on NR56, was
 | [2] |
where qrst is the probability of a cow being pregnant 56 d after first service; is the overall mean effect; Sq is the fixed effect of season of insemination (January to March, April to June, July to September, October to December); Pr is the fixed effect of parity of the cow (parity 1, parity 2, parity 3, >parity 3); MUNs is the fixed effect of classes of MUN; herdt is the random herd effect; CTFSqrstu is the interval from calving to first service in days of cow u; a1, a2, and a3 are linear, quadratic, and cubic regressions on CTFS; (CTFSqrstu x Pr) is the interaction between CTFS and parity; MYqrstu is milk yield (average of test-days 1 and 2 after calving); and β1 is the linear regression on MY.
Nonsignificant regression coefficients on different polynomial structures as well as nonsignificant interactions were removed from models beforehand by using a sum of squares type I test (Wald type test) at a significance level of 5%, which revealed similar results for effects in model [1] and model [2]. As discussed by König et al. (2005), sum of square type I tests provide sequential tests appropriate for polynomial models. Wald type III tests were used to identify significant fixed effects on NR56.
Genetic Correlations Among MUN, Milk Yield, and Fertility
For genetic analyses, a Bayesian approach using Markovchain Monte Carlo (MCMC) methods (Sorensen and Gianola, 2002) was used. Bivariate models were fitted for all pairs of traits (i.e., milk yield, MUN, CTFS, NR56, and NR90). Bivariate models with 2 binary traits, bivariate models with 1 binary and 1 Gaussian trait, or bivariate models with 2 Gaussian traits were applied.
Bivariate Model with 2 Binary Traits.
A bivariate threshold liability model was fitted, treating both NR56 and NR90 as binary traits. In a threshold liability model (Gianola, 1982; Gianola and Foulley, 1983), it is assumed that an underlying continuous variable, liability 
i exists such that the observed binary variable yi takes the value 1 if i is larger than a fixed threshold, and 0 otherwise. Hence, the binary response, yNR56 or yNR90, takes the value 0 for return after the first service if NR56 or NR90, respectively, is smaller than or equal to a conceptual threshold T, and 1 in a successful case of nonreturn (i > T). The threshold T and the residual variance of liability are not identifiable when the response is binary, so these parameters were arbitrarily set to t = 0 and 
ei2 = 1, respectively. In matrix notation, the fitted model [3] was
 | [3] |
where is a vector of unobserved liabilities for NR56 and NR90; β is a vector of systematic effects; sc is a vector of sire of cow transmitting ability effects; ss is a vector of service sire effects; h is a vector of herd effects; e is a vector of residual effects; and X, Zcs, Zss, and Zh are corresponding incidence matrices.
Vector β included parity and season effects as defined for the logistic models, and independent proper uniform priors U(–9999,9999) were assigned to each of the elements of β. A multivariate normal prior distribution for the sire of cow transmitting abilities was assumed:
 |
where 
is the co(variance) matrix between sire transmitting abilities, and A is the additive relationship matrix between sires. The service sire (i.e., the sire of the calf) and herd effects were assigned the multivariate normal prior distributions:
 |
 |
where 
is the (co)variance matrix between service sire effects, 
is the (co)variance matrix between herd effects, and I is an identity matrix of appropriate order. Because of the several small herds in our data set, and based on detailed experiences in Norway with similarly structured data (e.g., Heringstad et al., 2006), we considered herd as a random effect in the genetic analyses.
Independent inverse Wishart distributions were used as priors for the matrices SS0, G0, and H0:
 |
 |
 |
respectively, where 
ss, g and h are the degrees of freedom parameters, and Vss, Vg and Vh are the scale matrices. Residuals were assumed to follow the multivariate normal distribution e 
N(0, R0 
I), where
e1,e2 was assigned the bounded uniform prior e1e2 U(–1, 1).
Bivariate Model with 1 Binary and 1 Gaussian Trait.
Bivariate linear-threshold sire models were fitted in model [4] to all combinations of 1 categorical (NR56, NR90) and 1 Gaussian trait (CTFS, MUN, milk yield). Such models were first introduced in animal breeding by Simianer and Schaeffer (1989) for genetic analysis of disease and production traits in dairy cattle. The random effect of service sire was not included in the model, because the Gaussian traits were not affected by the service sire, and the available Fortran program requires same incidence matrices for all traits. The bivariate sire model for joint analysis of the underlying liability for an unobserved nonreturn (trait 1) and observed values y for CTFS, MUN, or milk yield (trait 2) was
Vectors, matrices, and prior distributions were as described above. Residuals were also assumed to be correlated and to follow the distribution e N (0, R0 I) where
A scaled inverse 
2 prior distribution was assigned to the residual variances of Gaussian traits (e22). Bounded uniform priors
Bivariate Model with 2 Gaussian Traits.
For all combinations of the Gaussian traits milk yield, MUN, and CTFS, a bivariate linear-linear sire model was fitted. Model [5] for Gaussian traits was like model [4] but with
Genetic correlations and heritabilities for traits i and j in models [1], [2], and [3] were calculated from the sire of cow (co)variances, that is, heritability
CSij being the covariance for cow sire effects between both traits. Considering herd as a random effect enables the possibility of calculating intraherd heritabilities by excluding h2 from the calculation of the total variance p2.
Convergence Diagnostics.
The MCMC sampling procedure consists of successive iterative updates of each parameter or group of parameters. For models [3] to [5], the MCMC procedure described by Heringstad et al. (2005) was used to draw samples from marginal posterior distributions of interest. Length of burn-in and of the sampling period were assessed by the method of Raftery and Lewis (1992), as implemented in the BOA software package (Smith, 2005). Results from the first 10,000 iterations in a first run for each model of a Gibbs chain in the covariances between the sire of cow effects were used to determine the optimal number for the burn-in period and the optimal number of iterations. The covariances in matrix G0 mix more slowly than other parameters, so this assessment was deemed conservative. Based on the diagnostics and on visual inspections of trace plots in the previous analyses of 10,000 rounds, chain lengths of between 150,000 and 200,000 iterations were run for different models and trait combinations. The burn-in period was 10,000 rounds for all models and analyses.
Selection Index Calculations
The possibility of improving selection response in fertility through the inclusion of an additional indicator trait was evaluated via selection index calculations. Three different breeding scenarios were developed by using the phenotypic and genetic parameters obtained in the first part of the current study for NR56 and MUN. This was done to combine fertility traits (NR56) and an indicator trait for energy balance (MUN) in selection index procedures. The alternatives included selection based on NR56, which is the international standard at the moment (Interbull, 2008), MUN, and a combination of both. The general breeding goal is to improve NR56 within the Holstein population. Therefore, the only trait defined in the aggregate genotype was NR56. By applying the selection index procedure using the SIP computer program (Wagenaar et al., 1995) and by assuming 50 daughter records per sire as information sources in genetic evaluation for the sire, the correlation between the index and the aggregate genotype was calculated. Results were compared for the 3 different scenarios. By assuming a standardized selection intensity equal to 1.0, selection response for the trait in the aggregate genotype was calculated.
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
. Mean values for NR56 and NR90 were 0.67 and 0.60, respectively, and 93.85 d for CTFS. The averages of test-day milk yield and MUN from test-days 1 and 2 after calving were 31.06 kg and 267.11 ppm, respectively.
|
There were also significant effects of parity (P < 0.01) and of season of insemination (P < 0.05) on NR56. As a further result, first-lactation cows with high milk yield had the lowest NR56. König et al. (2006) analyzed the fertility of cows from the same region in previous years and found that the success of a first insemination was lower in the first parity than in adult cows. Still being in a period of growth, first-parity cows have distinct problems in reaching a balanced stadium of energy. In general, high milk yield in the first third of lactation increases a cow’s risk of experiencing health (Collard et al., 2000) and fertility (Van der Waaij, 2005) problems. Perhaps the energy intake is insufficient to express the production potential and to maintain overall health and fertility. High milk yield after calving modeled as a linear regression was associated with lower pregnancy rates (P < 0.001). A longer CTFS interval up to the polynomial structure of order 3 was associated with greater NR56 (P < 0.01). The results from the present study suggest that a longer CTFS circumvented metabolic stress in this early period after calving.
Fitted NR56 values by UPC are shown in Figure 1. More important than the anticipated effect of available energy fitted through UPC (i.e., the combination of MUN and protein percentage on NR56) was solely the level of MUN. All UPC containing more than 300 ppm of MUN were associated with the lowest NR56, irrespective of the protein or energy supplied in the feed. The greatest least squares means for NR56 were from UPC 1, visualizing a low level of MUN (<150 ppm) and a surplus of energy supply. The negative impact of high levels of MUN on fertility was supported by results obtained with model [2] (Figure 2). Greater levels of MUN were associated with lower NR56. A negative impact of high concentrations of MUN on reproductive performance was also reported in some other studies (e.g., Gustafsson and Carlsson, 1993; Rajala-Schulz et al., 2000), where it was found that levels >190 ppm were detrimental. In Israeli Holstein cows (greatest average production level worldwide), Hojman et al. (2004) found that the greatest pregnancy rates were at levels of MUN below 11.75 mg/dL. One possible physiological explanation for the negative correlation between MUN and fertility was given by Elrod and Butler (1993). They reported that increased BUN decreased uterine pH several days after calving, which caused greater embryo mortality.
|
|
. Variance components and ratios for traits analyzed with different models (e.g., NR56 analyzed with a bivariate threshold-threshold model including NR90, or analyzed in bivariate threshold-linear models including Gaussian traits) were nearly identical. Heritabilities of NR56 and NR90 were 0.029 and 0.028, respectively, when applying model [3] or model [4]. These values are similar to heritabilities of pregnancy status in recipients after embryo transfer, as reported by König et al. (2007a) when applying complex synergistic models. This estimate is interpreted as heritability of a maternal trait and includes the combined effects of egg quality, uterine environment, and embryo vitality (Haussmann and Heinkel, 1989). Several other studies in the past 20 years have focused on estimation of variance components of nonreturn rates (Thaller, 1997), and lower heritabilities in the range from 0 to 3% have been found. Most of these studies, conducted 15 to 20 yr ago, used linear models when analyzing categorical traits, and results are on the observable scale. Greater heritabilities on the underlying liability scale from threshold models are to be expected when considering theory for categorical traits (Dempster and Lerner, 1950). However, even in Norway, with a well-developed recording system for health and reproduction traits, heritabilities on the underlying liability scale for nonreturn rates have been below 5% (e.g., Andersen-Ranberg et al., 2003)
|
) and therefore negligible. Boichard and Manfredi (1994) found that the contribution of service sire in French Holstein dairy cattle was only 0.8% of the phenotypic variance for conception rate. Estimates of Weigel and Rekaya (2000) in US Holsteins were below 0.5%. In an analysis considering all genetic paths contributing to nonreturn after 56 d, Andersen-Ranberg et al. (2003) also found little influence of the service sire. As a consequence, the service sire effect was excluded in subsequent studies of fertility traits in Norwegian dairy cattle (e.g., Chang et al., 2006).
Heritability of CTFS was 0.073 and was therefore greater compared with the 2 binary traits NR56 or NR90 (Table 3). The interval from calving to first insemination depends on the farmer’s decision and on genetic factors affecting the beginning of a cow’s reproduction cycle after calving, as supported by some other studies (e.g., Jamrozik et al., 2005). Regardless of the distribution of observations (i.e., categorical vs. Gaussian), CTFS may be more reliable than nonreturn rate as a fertility trait. First, when using nonreturn rates as a measure for fertility, there is limited documentation. Cows are often assumed to be pregnant after a first insemination, but there is limited or even no knowledge about subsequent services done by natural-service sires. The percentage of natural services in German Holstein cows is approximately 20% and shows an increasing trend (Arbeitsgemeinschaft Deutscher Rinderzüchter, 2005). In practice, elite sires from throughout the world are used for a cow’s first insemination. However, to reduce cost and labor, further service after failure to impregnate a cow is done by natural-service bulls, which may not be recorded. The interval from CTFS is free from these problems. On the other hand, CTFS is heavily dependent on management practices, and milk producers voluntary extend the waiting period before insemination as milk yield of the cow increases (Dekkers et al., 1998). Pasman et al. (2006) considered calving interval as the "ultimate" fertility trait in dairy cows. Calving interval depends on the start of the first cycle after calving and on the success of an insemination. Pasman et al. (2006) suggested the inclusion of CTFS in addition to nonreturn rates in genetic evaluation, because cows without a nonreturn observation or censored observations can also be considered in the evaluation model. Finally, from the objective of animal breeding, traits with the greatest heritabilities should be preferred when developing combined breeding goals. The present study revealed greater heritabilities for CTFS compared with nonreturn rates, by a factor of 2.6.
The start of the first cycle after calving and CTFS strongly depend on the cow’s health and energy balance status. De Jong (2005) pointed out the strong relationship between MUN and energy and health, and recommended further research in this area. From a physiological viewpoint, an increase in MUN implies that the liver must metabolize MUN into milk protein, but this process strongly depends on the availability of energy. Milk urea concentration is routinely measured at official test days in Germany, and it is mainly used to improve feeding strategies. As shown in Table 3, MUN has moderate heritability (0.13), suggesting the potential to be included in an overall breeding goal. Few studies have focused on genetic aspects of MUN. For example, Mitchell et al. (2005) used a model for repeated measurements of MUN within lactation. Heritability was 0.22 and 0.15, for infrared-analyzed and for wet chemistry-analyzed MUN, respectively. Wood et al. (2003) used a random regression model, and heritabilities of MUN were 0.44 and 0.59 for the first and second lactation, respectively. Vallimont et al. (2002) used a random regression model for daily MUN observations, but without fitting a fixed lactation curve. Heritabilities ranged from 0.12 to 0.20 in the first lactation, and were slightly lower in the second lactation. Miglior et al. (2007) used a highly sophisticated random regression test-day model, and average daily heritabilities for MUN were in the range from 0.38 to 0.41.
Genetic Analyses: Correlations Among Traits
Studies addressing MUN, production, and fertility traits simultaneously are limited. Knowledge of genetic relationships among these traits (especially in the early period after calving) is essential when developing combined breeding goals and selection strategies. Posterior means of genetic correlations and phenotypic correlations are shown in Table 4. The posterior mean of the genetic correlation between NR56 and NR90 was 0.94, indicating that both nonreturn rates are the same trait. This high genetic correlation is important because NR90 has been the central trait when evaluating fertility in Holsteins in Germany in the last decade, whereas most other countries use NR56 (Jorjani, 2005). For the purpose of harmonizing procedures for multiple across-country evaluations, Germany’s switch from NR90 to NR56 since genetic evaluation in April 2008 (VIT, 2008) should be without apparent loss. Based on the 0.94 correlation found in this study, few rerankings of sires and greater genetic correlations between Germany and other countries are to be expected.
|
This means that high milk yield after calving is genetically associated with longer intervals for showing the first heat. Milk yield and NR56 and NR90 were negatively correlated genetically (–0.31 and –0.33, respectively). Hence, breeding for increased milk yield, especially early in lactation, would have unfavorable effects on all fertility traits and, as shown earlier, also on health traits such as claw disorders (König et al., 2005).
The posterior mean of the genetic correlation between NR56 and CTFS was –0.04, and was –0.02 between NR90 and CTFS. These correlations close to zero indicate that cows starting a new reproduction cycle early in lactation are not genetically the best cows with respect to success after a first insemination. However, there is a wide range for genetic correlations between CTFS and the nonreturn rates reported in the literature. Petersson et al. (2007) reported a genetically favorable association between the commencement of luteal activity, based on progesterone profiles, indicating short CTFS and some reproduction traits. In addition, Janson and Andreasson (1981) found a genetically negative correlation (–0.34) between non-return rate and CTFS in Swedish Red cattle. Such results support the most desirable breeding scenario: a short period without a reproduction cycle after calving leads to improved nonreturn rates, and both desirable effects in combination reduce calving intervals. In contrast, positive genetic correlations in the range from 0.04 to 0.35 were estimated by Andersen-Ranberg et al. (2005a, b), Hoekstra et al. (1994), and Jamrozik et al. (2005). On the phenotypic scale (Table 4), a longer CTFS was slightly positively correlated with nonreturn rates (0.09 for both NR56 and NR90), probably because of the favorable impact of longer voluntary waiting periods on the success of a first insemination (Dekkers et al., 1998; König et al., 2006). Both components in combination [i.e., the genetic background for starting luteal activity, and a longer period of regeneration after calving (management component)] will contribute to improved nonreturn rates after a first insemination.
Genetic correlations between MUN and fertility traits were low (Table 4). A greater level of MUN was genetically associated with a longer CTFS (0.29) and lower nonreturn rates (i.e., –0.13 for NR56, and –0.12 for NR90). This indicates that, although selection for low levels of MUN would improve reproduction performance, genetic correlations were too weak to justify inclusion of MUN as an indicator trait for nonreturn rates in a combined breeding goal. Some improvements in the fertility complex are to be expected, based on the moderate correlation with CTFS. Similarly, Mitchell et al. (2005) found that genetic correlations between MUN and CTFS and interval from first service to conception were close to zero. However, some other studies reported opposite trends (e.g., Rensing et al., 2005). Hence, the relationship between MUN and fertility strongly depends on the definition of traits (e.g., test-day records vs. lactation records), on the period of lactation (e.g., test days after calving vs. the nearest test day before a first insemination), and on the production conditions (e.g., pasture-based systems vs. TMR). This implies that the impact of MUN on reproduction traits should be assessed separately for distinct regions or production systems.
The genetic correlation between MUN and milk yield was 0.44 and was in agreement with the report of Stoop et al. (2007), who applied animal models for repeated measurements. This suggests that selection for increased milk production also increases levels of MUN. Following the whole lactation and applying a random regression model for test-day observations, Miglior et al. (2007) found that MUN was not genetically correlated with milk yield. The phenotypic correlation between MUN and milk yield was much lower (0.13) than the genetic correlation, and Stoop et al. (2007) even found negative phenotypic relationships. A positive correlation between milk yield and MUN implies that more energy goes to milk and less goes to protein production, resulting in an energy shortage for protein production and, consequently, increased levels of MUN.
Selection Index Calculations
Results from selection index calculations to assess the impact of MUN observations on selection response and correlations between index and aggregate genotype in NR56 are shown in Table 5. The ultimate trait in the breeding goal was NR56, and the 3 different scenarios varied the available index sources for genetic evaluation of young sires, that is, 50 daughter records for NR56 only (scenario A), 50 daughter records for MUN only (scenario B), and 50 daughter records for each trait (scenario C). However, the overall breeding goal in Holstein dairy cattle in Germany is much more diverse (e.g., König et al., 2007b), but to determine the impact of MUN on the selection response in NR56, the chosen approximation revealed clear results. Scenario A led to a correlation between index and aggregate genotype of rTI = 0.52 and to a selection response of +4.1% in NR56 per generation. The inclusion of MUN (scenario C) as an index source gave only minor improvements. The genetic correlation between MUN and NR56, and the heritability of MUN were too low for improvements in a reproduction breeding goal. Exclusion of NR56 and using only MUN in the defined index led to a substantial decrease of both the correlation between index and aggregate genotype, and the selection response in NR56 (0.10, and 0.01, respectively). Genetic progress toward improvements in NR56 cannot be achieved via selection strategies on the MUN indicator trait. Optimizations have to consider direct selection strategies on NR56 via improvements in data quality (e.g., recording of the true status of pregnancy), greater quantities of daughter records for NR56 (more than the 50 daughters used in the present study), greater economic weights for fertility in the overall breeding goal, or their combination. However, further research including more traits in a multivariate analysis, as is done in Canada (Jamrozik et al., 2005), could reveal closer correlations with nonreturn rates, which will contribute to greater reliabilities and genetic gain for the complex of fertility.
|
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Received for publication April 4, 2008. Accepted for publication June 23, 2008.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  C. Bastin, L. Laloux, A. Gillon, F. Miglior, H. Soyeurt, H. Hammami, C. Bertozzi, and N. Gengler Modeling milk urea of Walloon dairy cows in management perspectives J Dairy Sci, July 1, 2009; 92(7): 3529 - 3540. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
