Optimization of feeding systems for dairy cattle (comparative analysis)
Published: June 20, 2007
By: JR Newbold (Courtesy of AFMA Matrix)
Today, our goals are more varied: outputs of dairy farming include milk quality, cow health and environmental impact. Control of nitrogen efficiency is one example of how current national feed evaluation systems (e.g. from FR, NL, UK, USA and elsewhere) can be used to pursue these wider goals.
New mechanisms may be needed to update and internationalize existing feed evaluation systems.
Developments in science and computing make the use of complex biological models more practicable. Such models offer the prospect of better prevention of metabolic disease (for example, sub-clinical rumen acidosis, fatty liver) and more consistent control of milk quality (for example, milk fatty acid profile) and will be a key tool in worldwide efforts to improve the sustainability of milk production.
Introduction
Philosophically, there are two questions that farmers consider when deciding how to feed their animals. Firstly, when the target level of output is known, there is ‘the requirement question’. Applied to dairy cows this might be phrased as: ‘What should I feed to my cows to achieve a yield of X litres per day?’. Secondly, when the possible choices of feed inputs are known, there is ‘the response question’.
For dairy cows, this might be phrased as: ‘How will my cows respond if I offer them Xkg of feedstuff Y?’ In general, feed evaluation systems developed in various countries over the last half-century have addressed the ‘requirement question’.
This is exemplified by Vermorel and Coulon [1], who introduced a comparison of different energy systems with the words: ‘The aim of any feeding system is to meet the requirements of the animals for type and level of production using available feeds.’ It is the contention of this paper that, important though it is to answer the ‘requirement question’, more attention should be paid to the prediction of animal response.
The ‘requirement question’: comparative analysis of different energy and protein evaluation systems
Table 1: Some energy evaluation systems for dairy cows.
Energy
It is beyond the remit of this paper to describe the different energy evaluation systems in use around the world, or to document their history. A partial list, with key references, is shown in Table 1.
Although the systems differ in the currencies adopted (e.g. Metabolizable Energy, ME, versus Net Energy for Lactation, NEL) and in the methods of calculation, their similarities are generally more striking than their differences. In all cases calculation of nutrient requirement and supply is relatively simple and amenable to incorporation within least-cost linear programming on small computers.
The broadest test of a ‘whole system’ is its ability to predict animal performance (milk yield) from known intakes of defined diets. Slightly narrower tests of energy systems can be made using calorimetry data where the flow of energy from diet to milk can be tracked more precisely. Such data also allow individual components of the system, such as requirements for maintenance, to be assessed.
Animals and diets change all the time (as, indeed, do the systems themselves) so comparative evaluations have a relatively limited lifespan. For example, Europe has seen a rapid spread of Holstein genetics (leading to an increase in milk yield of approximately 62kg/lactation/year, [2]), while advances in plant breeding have allowed forage maize to march steadily northward.
The systems themselves also evolve over time. Thus, Vermorel and Coulon [1] concluded that NRC1989 [3] underestimated the amount of feed needed, especially at higher levels of yield, in comparison with French, Dutch and German systems, but this is of little relevance now that NRC1989 has been superseded by NRC2001 [4].
The NRC evaluated their 2001 model using a database of 100 different diets taken from experiments published in the Journal of Dairy Science from 1992 to 2000 [4]. Intake and utilisation of NEL were highly correlated (r2 = 0.61), with NEL use within 5% of NEL intake for 46% of observations and within 10% for 76% of observations.
Perhaps the opportunity should be taken to expand the dataset to include more non-US diets and cows and compare different systems.
Fox et al. [13] evaluated the ability of the Cornell Net Carbohydrate and Protein System (CNCPS) to predict milk yield in diets where ME was thought to be limiting.
On average, there was a mean bias of 1.3% (around 0.5 kg/d) when mobilisation or deposition of body reserves was taken into account.
The most recent evaluation in which different systems were compared is that published by Yan et al. [15] in 2003. Using data for 838 cows from 12 long-term feeding studies in Ireland these authors evaluated models of AFRC1990 [10], INRA [7], Van Es [8], SCA [5] and NRC2001 [4].
Systems were evaluated in terms of the mean-square prediction error (MSPE). This has three components: mean bias, slope bias and random effects. The best systems were SCA1990 [5] and INRA [7]. On average, SCA1990 [5] under predicted milk yield by 0.3 kg/d (or 1.3%) with no evidence that the model performed differently at low or high yields (Figure 1).
Calorimetry studies allow differences between observed and predicted energy balance to be investigated. In two studies undertaken as part of the ‘Feed into Milk’ research programme in the UK, Agnew and Yan [16] reviewed published calorimetry data while Kebreab et al. [17] analysed a new database derived using modern Holstein-Friesian cows fed diets based on grass silage or maize silage.
This database represents the most comprehensive and modern set of individual cow calorimetry results available: energy balance data for 652 cows with a range of milk yield from 0.9 to 59.7 kg/d. Both studies highlighted under prediction of maintenance requirements as a key factor explaining the relatively poor performance of systems such as AFRC1990 [10] and Van Es [8].
Kebreab et al. [17] estimated MEm at 0.62 MJ/kg0.75/d, some 27% higher than the figure used in AFRC1990 [10] and Van Es [8]. This difference is probably due to the relatively higher mass of protein and internal organs, and lower fat, in modern dairy cows compared with animals common several decades ago.
The efficiency of utilisation of ME for milk production (kl) can be estimated from regression of milk energy output (adjusted for liveweight change) on ME intake. Using this approach, Agnew and Yan [16] estimated kl at 0.66, within the range of 0.60 to 0.67 reported in other studies. The implication is that modern cows do not use ME for milk production more efficiently (although they may partition more of their available energy towards milk production).
Recommendations from these studies have been incorporated into the ‘Feed into Milk’ system [12] that is now replacing the AFRC system ([10], [11]) in the UK, and are also relevant to all other energy evaluation systems. These studies demonstrate that there is still scope to improve existing energy evaluation systems.
Figure 1: Example of an energy evaluation system: the Australian system of SCA1990 [5] evaluated in Ireland by Yan et al. (2003). Reproduced from [15].
Protein
As with energy, so with protein. A variety of systems have been proposed in different countries since a formal distinction between protein requirements of the animal and the rumen microorganisms was first recognised in the 1970’s. As with energy systems, these protein systems are broadly similar and a comprehensive description is beyond the remit of this paper (see Table 2 for key references).
In contrast to energy systems, all modern protein evaluation systems use the same currency to define both animal requirement and dietary supply. Thus, conceptually, the French ‘Protéines digestibles dans l’intestin’ (PDI) and the Dutch ‘Darmverteerbaar Eiwit’ (DVE) are directly comparable to the English term ‘Metabolizable Protein’. This describes the total amount of protein available to the animal from microbial protein, produced in the rumen, undegradable protein and (in some systems, such as NRC 2001) endogenous protein.
Several comparative evaluations have been made over the years. Van Straalen et al. [21] compared observed and predicted milk protein yield in 15 dairy cow production experiments and found the DVE system [18] to be superior, under Dutch conditions.
Tuori et al. [22], using data from Finnish Ayrshire and Friesian cows fed grass silage or hay-based diets, concluded that the Nordic AAT-BV system was better than the UK MP system or the French PDI system.
NRC [4] evaluated their new system by comparing MP allowable milk yield with actual milk yield. However, their database contained relatively few cases where MP was likely to have been limiting to milk production (Figure 2), leading to the conclusion that experiments designed specifically to test the new model were needed.
The ‘requirement question’: summary
In summary, although each has unique features, the different energy and protein systems used around the world share a common architecture. There is always scope for improvement and inter-system comparison or ‘benchmarking’ is part of that process. It is hard to resist the thought that it is absurd, in the globalizing world of the early 21st century, to have so many different systems, each representing a variation on the same theme.
It is important to recognise that millions of cows and thousands of farmers benefit every day as these systems are used to answer ‘the requirement question’ and formulate practical rations. The economic impact of this activity should not be underestimated and the widespread adoption of these systems represents a real example of successful technology transfer.
Table 2:Some protein evaluation systems for dairy cows.
The ‘response question’
The ‘response question’ has always been important in low input systems, where maximising the utilisation of scarce feeds is generally more important than maximising the biological efficiency of the animal.
Under these conditions, it is important to quantify marginal responses to changes in diet. In high input systems, the focus has traditionally been on maximising biological efficiency within the cow. In practice, this has usually meant maximising output per cow in the short term, to dilute the costs of maintenance. Here, answering the‘requirement question’ has been of paramount importance (especially under a milk quota regime, where the target outputs are very clearly defined).
However, responses other than milk yield are increasingly important. Long-term sustainability demands that we understand– and can quantify - effects of nutrition on cow health, welfare and fertility.
Real pressure to minimise environmental pollution (for example, to comply with the EC Nitrate Directive or the US Clean Water Act) highlights the need to know how efficiencies of nutrient utilisation respond to changes in diet.
Focus on product quality (for example, the physical properties of butter) means we must understand how milk composition responds when we introduce new feedstuffs or change nutritional parameters. It’s not just about milk yield, and hasn’t been for some years.
To deal with these issues, we need models that predict more comprehensively the way cows respond to diet. For example, will an extra 500g of long hay stabilise rumen pH and milk fat concentration? If sunflower meal is cheap, can I add another kilo without elevating rumen ammonia production and putting pressure on fertility?
Figure 2:Example of a protein evaluation system: the US system of NRC (2001). Reproduced from [4].
However, this does not mean we should throw away our existing ‘energy’ and‘protein’ systems as concepts such as ME and MP can also be used to predict responses other than milk yield.
Use of current concepts to predict cow response to diet: protein:energy ratio
As an example, consider the ratio between the protein and energy made available to the cow. This ratio can be derived from any national system: in the UK it would be expressed as MP:ME, in the USA as MP:NEL, in France as PDI:UFL, etc. The terms MP and ME are used here to illustrate the broader principle.
Classically, a nutrient ‘requirement’ is set by fitting a ‘broken-stick’ function to a plot of an animal response to that nutrient. For MP, the ‘requirement’ is better described as an optimum MP:ME ratio. The broken stick model is a very imperfect representation of an underlying biological response curve.
This is apparent in the real data reviewed by NRC [4], Figure 2. Biological explanations for the continued, but diminishing response to increases in MP:ME above the optimum include mobilization of body energy reserves and use of surplus protein as an energy source, as discussed by Newbold [23].
Newbold [24], using relative costs of MP and ME relevant in the UK in the mid- 1990s, showed that in some circumstances, a narrow assessment of feed costs and milk outputs suggested that it was better to increase the MP:ME ratio to exploit marginal responses to MP, rather than to increase overall energy supply.
However, such a narrow analysis ignores other responses to MP:ME ratio, such as reduced efficiency of N utilisation. Some of these responses were elegantly demonstrated, using real data, by Verite and Delaby ([25], Figure 3). Here, an increase in protein:energy ratio (as PDI:UFL) leads to increased milk yield, but is it worth the increase in N pollution as urinary N and the associated risk of metabolic disorders in the cow due incomplete conversion of ammonia to urea in the liver?
Figure 3: Use of protein:energy ratio (in this case, PDI:UFL) to predict range of responses in dairy cows. Reproduced from [25].
Thus, existing nutritional currencies, such as MP and ME, calculated using existing feed evaluation models, can be used to predict a variety of responses to diet, helping farmers and advisors identify the optimum strategy for each farm.
Use of nutrient-based models to predict dairy cow responses to diet
Traditional feed evaluation systems aggregate a large amount of biochemistry into single concepts such as ME and MP. By analogy, a diverse and complex range of dairy cow responses to diet are‘aggregated’ into the single figure of milk yield (or milk energy or protein yield). We now need to ‘disaggregate’ the cow’s response and consider separately responses in milk quality, cow health and environmental impact, as well as milk yield.
To do this, we need less aggregated models than those described above.
So-called ‘nutrient-based’ feed evaluation systems, representing feed and animal biochemistry at a less-aggregated level than traditional energy and protein systems, have been under development for several years for use in research, teaching and, increasingly, practical field application.
Developments in computer technology mean that the old objection – that they are too complex and cumbersome for field use– is largely redundant.
The Cornell Net Carbohydrate and Protein System (CNCPS, first published in 1992 but comprehensively described more recently by Fox et al. [13]) represents an excellent example of a practical model that predicts a range of cow responses to diet. It combines a relatively disaggregated model of rumen metabolism with a relatively aggregated model of animal metabolism.
Other models include a more comprehensive representation of post-absorptive nutrient metabolism (especially the ‘Molly’ model of Baldwin and colleagues [26]) but these are not yet suitable for practical field use. The CNCPS model therefore provides a useful framework for this brief discussion.
Responses in milk composition
Milk composition is controlled by complex interactions between nutrient supply and animal metabolism. Therefore a largely rumen-based model, such as CNCPS, cannot be expected to predict milk protein or fat concentrations. Moate et al. [27] designed a sub-model for the CPM model (closely related to CNCPS) to predict the supply of long-chain fatty acids, and this should help in the prediction of milk fat. Accurate prediction of milk protein concentration remains a challenge, as it has been for many years.
Responses in cow health
The CNCPS is useful in predicting those aspects of cow health related to rumen function. For example, it includes an explicit (if largely empirical) prediction of pH and this can be used to warn of sub-clinical acidosis. Several other systems either predict pH explicitly or warn of possible acidosis as part of a ‘decision support system’ (for example, ‘Feed into Milk’, [12]).
Offner et al. [28] evaluated three models of rumen pH against published literature. In general, all models tended to overestimate pH, suggesting scope for improvement, with a bigger over prediction for high concentrate diets. The model of Lescoat and Sauvant [29] was best.
However, as with milk composition, many issues of cow health represent an interaction between nutrient supply and animal metabolism. Of particular importance are the many production diseases linked to liver disorders (see Drackley, [30], for review).
Although models of liver metabolism exist [31] there is currently no system suitable for highlighting (and avoiding or treating) these problems on-farm.
Responses in environmental impact
Developed originally to predict milk production on an individual cow level, the CNCPS has been extended to predict efficiency of nutrient use on a herd basis, accounting for soil fertility, crop rotation and manure management (Cornell Nutrient Management Planning System, Tylutki and Fox [32]). The model is now being applied, in conjunction with other management tools, to improve efficiency of N and P use in targeted areas of New York State [33] in response to an urgent need to protect water supplies.
While CNCPS is being used with such urgency, other, more mechanistic models are being developed to predict other pollutants, such as methane, as well as N and P. Particularly noteworthy is the model of Kebreab [34]. In a ‘sign of the times’, this model has the explicit objective of providing farmers with a decision support system to model output of pollutants, even though it is an extension of earlier models where the main objective was to improve prediction of milk production.
The ‘response question’: summary
These are exciting times in dairy cow nutrition. While they remain useful, we are quickly moving beyond the use of simple aggregated models to help meet requirements for energy and protein.
Developments such as CNCPS are enabling farmers, nutritionists and policymakers to quantify how responses of cows (and farms) to changes in diet, in terms of milk quality, cow health and environmental efficiency.
Significant challenges remain – perhaps the greatest of which is to develop disaggregated models of cow metabolism (gut tissues, liver, mammary gland, etc) to complement disaggregated models of rumen function. Like any model, the rule of‘rubbish in, rubbish out’ applies, and continued effort is needed to improve methods of feed characterisation.
Implementation of existing ‘response prediction models’ and their continued improvement are vital to the sustainability of dairy farming in the years ahead.
References
The references are available from the author or from AFMA.
Author: JR Newbold
Provimi Research and Technology Centre, Lenneke Marelaan 2, B1932 Sint-Stevens-Woluwe, Belgium.
 | The previous article is a special collaboration from AFMA South Africa |
Related topics:
Recommend
Comment
Share
20 de junio de 2007
Very interesting article. Mysef, I was confronted with difficulties when I needes to make a ration for dairy, and I tried to use different information sources and specifications of raw materials. In Romania, in the past especially, we used the Russian system for evaluating energy and protein, like UN for energy = energy of 1 kg barley and PBD = crude protein digestible. In the present, we use the French system, but not everybody understands it very well. If I need to use the nutritional value of main forages, I meet difficulties, because the same forages has different units of energy and protein. Unification, in an unique unit method, will help nutritionists and farmers. All who can contribute on this will be welcome.
Eng. Ianes
Recommend
Reply
Would you like to discuss another topic? Create a new post to engage with experts in the community.
Featured users in Dairy Cattle
