NEAR~NFRARED REFLECTANCE SPECTROSCOPY IN PRECISION FEED FORMULATION'

Transcription

1 81997 Applied Pouluy Science, Inc NEAR~NFRARED REFLECTANCE SPECTROSCOPY IN PRECISION FEED FORMULATION' THE0 A.T.G. VAN KJ3MPEN2 and l? HOWARD SIMMINS3 Rh6ne-Poulenc Animal Nutrition, 42 Ave. A. Briand, BP 100, 92164Antony Ceder, France Phone: FM: Primary Audience: Nutritionists, Quality Assurance Personnel, Plant Managers INTRODUCTION Animal feeds are mixtures of feedstuffs such as maize and soybean meal. These feedstuffs are mixed in specific ratios in order to create a final feed that provides the optimum balance of amino acids and energy. In practice, however, variability in the nutritional quality of the feedstuffs may result in the production of incorrectly balanced feeds. This 1 Presented at the 1997 Poultry Science Association Informal Poultry Nutrition Symposium: "Precision Nutition for Poultry. " 2 Current address: North Carolina State University, Department of Animal Science, Box 7621, Raleigh, NC To whom correspondence should be addressed

2 472 NIRS AND FEED FORMULATION results in either energy being fed in relative excess to amino acids, leading to undesirable accretion of fat, or amino acids being fed in relative excess to energy, leading to a waste of costly amino acids. Feeding excess amino acids also augments the excretion of nitrogen, which has been identified as one of the major pollutants derived from animal production. It is thus of prime importance for the feed industry to understand both the nutritional needs of the animals and the nutritional value of feedstuffs available to formulate feeds. The nutritional needs of poultry have received substantial attention, and for the fxst limiting amino acids - methionine and lysine - reasonably accurate requirements have been set (relative to energy) for the most economically important phases of animal production. The nutritional value of feedstuffs, however, remains an important point of discussion. Numerous feedstuff evaluation systems can provide estimates of nutritional value, but each has shortcomings. Those systems that are most reliable from a biological standpoint (digestible or available amino acids) are also the most cumbersome and costly, and thus unsuitable for routine feedstuff testing. Because of these problems with feedstuff evaluation, feed manufacturers often perform inadequate nutritional testing on the feedstuffs used in manufacturing complete feeds. The extreme variation in nutritional value of feedstuffs, especially those derived from multiple sources, thus often results in the nutritional value of the final feed being left to chance. As indicated above, such haphazard formulation cannot be cost-effective. This paper examines how variation in the digestible amino acid content between batches of feedstuffs affects the quality of a final feed and discusses practical techniques that can better quantify this variation in order to produce feeds closer to the intended nutritional value. PREDICTING FORMULATION ERRORS The coefficient of variation of nutrients such as digestible amino acids [l] in a complete feed can be calculated based on the variation of these nutrients observed in the individual feedstuffs (Table 1) and the inclusion rate of these feedstuffs in the final Digestible Content I cv I % Digestible Methionine Content I cv feed, using the formula published by Fawcett and Webster [2]. For a feed containing 65% maize, 25% soybean meal, and 5% poultry by-products as the protein-containing feedstuffs, the coefficient of variation (CV) for digestible lysine was calculated to be 8.7%; for digestible methionine, 9.2%. Using these coefficients of variation, it is possible to calculate the level to which feed should be overformulated so that, for example, 80% of the produced feeds meet or exceed the animal s nutrient requirement. For the feed described above, this value came to 107.4% for digestible lysine, while for methionine, it came to 107.8%; in other words, the feed containing 65% maize, 25% soybean meal, and 5% poultry by-products as the protein-containing feedstuffs should contain on average an excess of 7.5% digestible lysine and methionine in order to assure that it meets the nutritional requirements of the animal for 80% of the feeds produced. Figure 1 shows the effect that feedstuff evaluation systems, developed with the objective to predict the nutritional value of a feedstuff, have on the required overformulation. Figure 1 presents the percentage overformulation required to ensure that 80% of feeds produced meet or exceed the calculated nutritional profile using batches of feedstuffs selected based on protein content or on digestible amino acid content when this digestible amino content is estimated with a given degree of precision [4]. The calculations are based on a complete feed containing 65% maize, 25% soybean meal, 5% poultry by-product, and 5% vitadmineral premix.

3 ~~ ~ ~ Symposium van KEMPEN and SIMMINS 473 n 10 p~ Methionine.- U G c-4 d e In o\ \o 00 o\ o\ o\ Q) Explained variation in daa (%) PC FIGURE 1. Percentage of overformulation required such that 80% of the feeds produced meet or exceed the calculated nutritional profile PREDICTION OF VALUE NUTRITIONAL BASED ON PROXIMATE ANALYSES Determination of nitrogen traditionally has been used to predict the digestible amino acid content of a batch of a feedstuff, on the assumption that nitrogen content is correlated with digestible amino acid content. This hypothesis was tested using data from the Rhodimet Nutrition database [3], for which nitrogen analyses were performed using the Kjeldahl method. True ileal digestible protein, lysine (dig. Lys) and methionine (dig. Met) were measured in poultry as per Green et al. [5]. Linear regression equations were determined using The Unscrambler Version 6.11 (a multivariate statistical analysis program from Cam0 AS, Trondheim, Norway), using full cross-validation in order to determine the explained variation [4]. Figure 2 shows that in maize no statistically significant correlation exists between protein (nitrogen x 6.25) and digestible lysine. Summarizing similar information for soybean meal and poultry by-products as well (Table 2) shows no correlation between protein and digestible lysine or methionine content for maize, while the correlations for poultry byproducts were poor. For soybean meal, the correlation was mediocre. Thus, knowing protein content and using this information to predict the digestible amino acid content of a batch of a feedstuff would allow overformulation to be reduced to approximately 6.3%. Including fiber data (NDF, ADF, and cellulose) for the regression equations improved the explained variation for the plant products such that approximately 60% of the variation in digestible lysine and methionine could be explained (Table 3). (For poultry by-products, the regression was still based on protein only). Thus, knowing proximate analyses and using this information to predict the digestible amino acid content of a batch of a feedstuff would allow overformulation to be reduced to approximately 5.5%. However, adding these parameters also increases the time required to complete the analyses needed for the regression. Fiber analyses performed using classical chemistry require approximately 2 days, and are thus not suited for routine application. IN ~ TRO DIGESTIBILITY MEASUREMENTS FOR PREDICTING NUTRITIONAL VALUE An alternative method of obtaining information on the digestible amino acid content of a batch of a feedstuff is the in vitro digestibility assay. In such assays, either protein (nitrogen) digestibility or the digestible protein content is

4 474 JAPR NIRS AND FEED FORMULATION g 0.21 W I v CornIMaize Soybean meal Poultly by-product I I I Protein (%) TABLE 2. Variation in true ileal digestible lysine and methionine explained by protein (Nx6.25) Digestible Digestible Methionine ** * * ** * * ** * * FIGURE 2. Digestible lysine content (%) as afunction of the protein (Nx6.25) content ( A) of randomly selected batches of maize TABLE 3. Variation in true ileal digestible lysine and methionine explained by protein (Nx6.25) and fiber (ADF, NDF, cellulose) Digestible Digestible Lyxine Methionine Soybean meal measured. We tested the validity of this technique using in vivo data rather than in vitro data, on the assumption that if it failed using in vivo values, then it definitely would not work with in vitro values. The results of the regression tests are presented in Tables 4 and 5. Table 4 shows that the in vivo protein digestibility coefficients explain no variation in maize. They explain a small portion of the * variation in soybean meal and poultry byproducts, in which heat treatment affects the digestible amino acid content. The regression lines obtained thus have no practical application. Table 5 shows that digestible protein content explains little variation in maize and only some variation in soybean meal and poultry by-products. Comparing Table 5 to Table 2, in which crude protein was used to predict digestible amino acid content, shows that the digestibility of protein explains only slightly more of the variation for maize and soybean meal, but noticeably improves the variation explained for poultry by-products. The reason for the only small improvement observed is that a large portion of the variation between batches of feedstuffs is linked to the amino acid content of a feedstuff, rather than to the digestibility of protein or protein content; in vitro digestibility assays, however, do not take this into account. To obtain useful information for routine feedstuff evaluation from in vitro digestibility assays it is thus necessary to determine both amino acid content and amino acid digestibility, which renders this type of test not suitable for routine application.

5 van KEMPEN and SIMMINS Symposium 475 TABLE 4. Variation in true ileal digestible lysine and methionine explained by in riyn protein digestibility coefficients Cornmaize CornD4aize Soybean meal PoultIy by-product 70 Methioine Sovbean meal I I I Poultrv bv-droduct I 33 I 2 1 TABLE 5. Variation in true ileal digestible lysine and methionine explained by h riyn digestible protein content NIRS TO PREDICT THE DIGESTIBLE AMINO ACID CONTENT OF FEEDSTUFFS Current methods that yield useful nutritional data to evaluate feedstuffs are so cumbersome or costly that feedmills continue to use traditional techniques. As indicated above, however, indirect techniques that allow for quick testing are of limited value. Clearly, producers could formulate feed more reliably if rapid and reliable tools were available to assess the nutritional value of feedstuffs. One technique that appears promising is Near- Infrared Reflectance Spectroscopy (NIRS). NIRS utilizes a principle that has been recognized for over 200 yr: bonds between organic molecules absorb a specific range of wavelengths in the near-infrared region, so that the near-infrared color of a sample provides information about its composition. The mechanics and underlying principles of NIRS have previously been explained [6]. NIRS is actually a regression technique: its predictions are based on correlations between spectral information (light absorption) and reference data [g, such as in vivo measured digestibility data. An advantage of such a system is that it can be applied to materials that have received various types of processing, such as heat treatment. Heat treatment will change the organic structure of materials, and thus the near-infrared spectrum. NIRS could use this information to correctly calculate the digestibility of amino acids in heat-treated materials, provided such materials were included in the reference database. Such corrections are difficult to make when basing regression on results of proximate analyses, since processing does not necessarily affect such results. We have been developing MRS calibrations for predicting the true ileal digestible amino acid content of feedstuffs. For this purpose, feedstuff samples with known in vivo amino acid digestibility were scanned using a NIRSystems 6500 (monochromator instrument). Spectral data from 1100 to 2500 nm were then pretreated with a standard normal variate and detrending scatter correction, and derivatized twice. Calibrations were then developed based on the converted spectral data using partial least squares regression. These calibrations were tested by running a crossvalidation [8]. Because animal meals have the greatest variation in digestible amino acid composi- tion, we have given them the highest priority for calibration development. The obtained prediction errors (RMSEP) and explained variation (r2,1) for three commonly used animal meals (meat and bone meal, poultry by-products, and fish meal) for lysine and methionine are provided in Table 6. Table 6 shows that MRS can predict the digestible amino acid content of animal meals. TABLE 6. Calibration performance (g/100 g) of animal meal calibration 1.20 for poultry digestible lysine and methionine* MEAT & BONE MEAL Methionine FISH MEAL Methionine POULTRY BY-PRODUCTS *RMSEP = prediction error; r2 = explained varia. tion as determined in the validation process.

6 476 NIRS AND FEED FORMULATION The NIRS predictions explained approximately 90% of the variation in digestible lysine and 80% of the variation in digestible methionine within the product categories. Predictions for digestible lysine were more accurate than predictions for methionine, probably because of the lower levels of methionine in these products and the higher assay variation for methionine. For the amino acids not shown, the calibration developed generally explained 7&W% of the variation within each of the animal meals presented. Comparing the results in Table 6 with the results presented in Tables 2-5 shows that NIRS was able to predict the digestible amino acid content of feedstuffs with greater accuracy than current feedstuff evaluation techniques. Additionally, NIRS results can be obtained in a fraction of the time it takes to determine proximate analyses or in vitro digestibility data (2-5 min with current equipment; in seconds with in-line equipment), making this technology more suitable for routine testing. If NIRS calibrations that can explain 8045% of the variation in digestible amino acid content between batches of all important feedstuffs are feasible, implementation of such NIRS technology would allow feed manufactures to reduce the overformulation of feeds from the original 107.5% of calculated requirements to 103% (Figure 3). Such a reduction in the use of expensive amino acid sources 6 g5 W a4 u M will lead to lower feed preparation costs. Small improvements in animal performance (lean tissue gain) should also result, since formulating feed with greater accuracy will cause the 20% of feeds that do not meet the set nutrient requirement to be, on average, closer to the intended nutrition level (Figure 3). The average inadequate feed fulfilled nutrient requirements at the following levels: for feed formulated with randomly selected batches of each feedstuff, 96%; for feed formulated with batches selected based on protein content, 96.9%; and for feed formulated with batches selected based on NIRS predicted digestible amino acid content, 98.5%. In theory it is also possible to calculate how reducing overformulation will affect nitrogen wastage, and thus environmental pollution from broiler production. Additional assumptions for these calculations were that the feed conversion for the randomly formulated feed was 1.8. The feed itself contained 21% protein, while the birds gained 31.7 g N/kg feed consumed. The ratio of nitrogen excretion to nitrogen accretion then dropped from 0.91 for the randomly formulated feed to 0.87 for the protein-formulated feed and to 0.79 for the NIRS-formulated feed, a drop of 13%. Reducing overformulation to zero would reduce nitrogen excretion to 0.66, a drop of 27%. Formulating feeds with more accuracy can thus yield a variety of benefits Random N analysis - NIRSi t _- 9 H Nutrient contentlnutrient requirement v M e I c? d FIGURE 3. Distribution of the nutrient (digestible amino acid) content relative to (calculated) requirement in feeds, based on 65% maize, 25% soybean meals, 5% poultry by-products, and 5% vitamin/mineral premixes, formulated with randomly selected batches of feedstuffs or with batches selected based on their protein (N) content or with batches selected based on NIRS's predicted digestible amino acid content

7 Symposium van KEMPEN and SIMMINS 477 CONCLUSIONS AND APPLICATIONS 1. NIRS calibrations may be developed which allow for the prediction of nutritionally relevant parameters, such as the digestible amino acid content in a feedstuff. 2. The accuracy of such calibrations easily exceeds the accuracy of other rapid methods currently available. NIRS with appropriate calibrations is thus a method of choice, given its rapidity and ease of use, for the evaluation of the nutritional value of feedstuffs. 3. Implementation of such technology allows a feedmill to improve the relevance of quality control exerted. Such quality control should lead to diets which better match their intended nutritional profile. 4. Producing diets which accurately match the intended nutritional profile should be cheaper, result in better animal performance, and reduce environmental pollution. 1. In this paper it is assumed that the digestible amino acid content of a feedstuff is a valid estimate of the bioavailability of amino acids, since databases on bioavailabilityare currently too limited in size for the type of calculations presented. 2. Fawcelt, Rand M. Webster, Valuingvariance reduction. Pages S m in: Proc. 8th Australian Poultry Sci. Symposium, Sydney, Australia. 3. RhBne-Poulenc, Rhodimet Nutrition Guide. 2nd Ebition. RhBne-Poulenc Animal Nutrition, Antony, France. 4. Wlained variation is the ability to correctlypredict a parameter in a sample not used to generate a regression equation, expressed as an $value. 5. Green, S., S. Bertrand, M. Duron, and R Malllard, Digestibilities of amino acids in maize, wheat, and REFERENCES AND NOTES barley meals, determined with intact and caecectomized cockerels. Br. Poultry Sci Kempen, T. van and D. Jackson, NIB may provide rapid evaluation of amino acids. Feedstuffs 68(50): Shenk, J.S. and M.O. Westerhaus, Analysis of Agriculture and Food Products by Near-Infrared Reflectance Spectroscopy. Infrasoft International, Port Matilda, PA. 8. Kempen, T. van, J-C. Bodin, P. Williams, and D. Jackson, Near Infrared Reflectance Spectroscopy (NIRS) outperforms nitrogen-based regression as a rapid tool to predict the poultry digestible amino acid content for commonly used feedstuffs. In preparation.