Near and mid infrared reflectance as

tools for ingredient real time analysis:

implications for precision formulation

Presented previously at 2002 Maryland Nutrition Conference, pp. 80-87

T. van Kempen, N. Qiao,

North Carolina State University, Department of Animal Science

and S. McComas

Novus International, St. Louis, MO

Summary

Most feedmills accept that quality control can only be achieved after feed ingredients have been mixed into complete feed and that most parameters, which are assessed, are of little nutritional importance. The resulting problems with variability in feed quality are dealt with by overformulating or producing diets that will not always yield optimal performance. Such diets can reduce profit to the producer and augment nutrient losses in the animal.

With the use of modern infrared equipment, it is possible to obtain nutritionally relevant data. This paper describes calibrations for parameters such as digestible amino acids in animal meals, iodine value, energy, free fatty acid content, and other quality parameters applicable to feed-grade fats. Data presented and data available in the literature indicate that calibrations developed for such parameters can provide a wealth of information to the producer. This information can then be used not only for filing claims but also for bin selection of incoming ingredients or even for matrix adjustments.

The most appealing of infrared spectroscopy is that many diverse parameters can be quickly assessed by a single scan of the ingredient. For example, the trans fatty acid content and iodine value of fats can be obtained by a non-skilled technician in a matter of minutes after receiving the samples, thus in time to direct where a truck is unloaded. Future developments in infrared technology will even make it possible to install this equipment in line such that continuous nutritionally relevant information can be obtained on feed ingredients.

Introduction

Quality control of incoming feed ingredients and outgoing complete feeds is a major challenge for feed mills. Upon arrival of feed ingredients to a mill, decisions pertaining to accepting or rejecting ingredients or into which bin to unload have to be made virtually instantaneously based on parameters that are nutritionally relevant. Although great advances have been made in the analytical procedures available to feed mills, many still do not fit in well with the operating conditions of a feed mill.

An example is nitrogen analysis. The traditional Kjehldahl method yields results in hours, clearly not acceptable for routine analysis (although acceptable for filing claims). The newer Dumas method can yield results in 5 to 10 minutes, thus being a step in the right direction. At the same time, though, our understanding of animal nutrition has advanced. For non-ruminants knowledge of nitrogen is inadequate nutritionally as formulations are commonly based on digestible amino acids. For most ingredients, digestible amino acids correlate poorly with nitrogen (van Kempen and Simmins, 1997).

The methodology that offers most promise as a routine quality control tool for the feed industry is infrared spectroscopy. Several reviews of infrared spectroscopy have been published and the reader is referred to those for background information about these infrared methods (e.g., van Kempen, 2001). The key advantages of infrared spectroscopy are:

Typical analysis time is one to two minutes
No to minimal sample preparation is required prior to analysis
Nutritionally relevant parameters can be predicted
Many parameters can be assessed simultaneously
Minimal operator skills are required.

The disadvantages of infrared spectroscopy are:

Equipment is expensive
Infrared analyses are based on correlations derived from samples with known composition
Samples from a different source may not be predicted correctly
The quality of the reference method used to develop a calibration limits the quality of a calibration.
Not all parameters can be predicted, most importantly organic compounds present at concentrations below 1g/kg and minerals.

This paper summarizes an evaluation of infrared spectroscopy for predicting total and digestible amino acids in animal meals, and an evaluation of fat quality. Based on the results, the implications of infrared spectroscopy for precision feed manufacturing control are discussed.

Instrumentation

For near infrared analyses, a NIRSystems model 6500 spectrometer was used (Foss, Silver Spring, MD). This commonly used instrument uses a monochromator for obtaining the different wavelengths (colors) of light. This monochromator is an etched, parabolic mirror that acts like a prism. Thus, the instrument relies on mechanical positioning of this mirror for wavelength accuracy.

For mid infrared spectroscopy, a Nicolet model Magna 760 spectrometer was used (Nicolet, Madison, WI). Because this is a research-oriented spectrometer, it is not designed for use as a routine quality control instrument, although versions for this purpose are available. Mid infrared instruments, including this one, commonly use a different mechanism than the above near infrared instrument for obtaining spectral information. The underlying principle is that the light interference obtained after combining two identical light sources which each travel a slightly different length, produces an interferogram. This interferogram is mathematically processed using a Fourier transformation. The advantage of this method is that theoretically more light energy is available to penetrate the sample and the wavelength reproducibility is excellent. The disadvantage is the mathematical transformation can ‘miss’ peaks that are very broad. This is especially an issue when working with solid samples.

Animal Meals

Animal meals are the most variable feed ingredients used routinely in the feed industry (Table 1). The reason for their variability is the non-homogeneous nature of the ingredients that make up the animal meals and the effects of processing. The objective of this study (Qiao, 2001) was to evaluate the feasibility of near and mid infrared spectroscopy as routine quality control tools for animal meals.

Table 1. Digestible lysine and methionine content (%) and coefficient of variation (CV, %) between randomly selected batches of three commonly used feedstuffs (van Kempen and Simmins, 1997).

	Dig. Lys		Dig. Met
	Content	CV	Content	CV
Corn/Maize	0.19	10	0.15	18
Soybean meal	2.44	10	0.55	11
Poultry byprod.	2.22	34	0.77	27

For this study, approximately 150 samples of animal meals were collected from the US rendering industry. These samples included typical samples but also samples with processing errors in order to maximize variability in the data. This type of sample selection is desirable for making a calibration robust. Using the ‘sample select’ procedure of the ISI software (Infrasoft International, Port Matilda, PA), 44 samples were selected that uniformly covered the variation in the original dataset. These 44 samples, together with 25 animal meals (19 meat and bone meals and 6 feather meals) assayed in vivo, were used for developing infrared calibrations for total amino acids. The 25 samples, whose true ileal amino acid digestibility coefficients were assayed with the standard caecectomized cockerel procedure (Parson 1986), were obtained from the Fats and Protein Research Foundation (Bloomington, IL).

For near infrared analysis, samples were scanned as-is using a full-size transport cell. For mid-infrared analysis, samples were ground with a 0.5 mm screen and scanned using an ATR crystal (samples are pressed onto this crystal for analysis). The calibrations were developed using partial least squares regression with full cross validation (Esbensen et al., 1996).

A synopsis of the calibration data is provided in Table 2. Good calibrations were obtained for lysine (total and digestible), independent of the infrared methodology used. For methionine, the quality of the calibrations was reasonable. Interestingly, a substantially better prediction error was obtained for digestible methionine with mid infrared spectroscopy than with near infrared spectroscopy (a similar result was obtained for cysteine, data not shown), suggesting that mid infrared was more capable to ‘see’ sulfur amino acids. This is in line with the expectations as mid infrared can detect a much wider array of molecular bonds, including sulfur bonds, than near infrared spectroscopy. The spectral data revealed, nevertheless, that the ATR used for mid infrared spectroscopy was not ideal for unground solid samples (due to poor contact between the solid samples and the crystal), indicating that a better sample presentation method is preferred.

Table 2. Cross validation results for prediction of total and in vivo digestible amino acid contents with near infrared and mid infrared spectra obtained from animal meal samples¹(Qiao and van Kempen, unpublished).

	Near infrared				Mid infrared
	Total		Digestible		Total		digestible
	R²	RMSEP	R²	RMSEP	R²	RMSEP	R²	RMSEP
Lys	0.83	0.25	0.88	0.23	0.81	0.25	0.85	0.26
Met	0.67	0.10	0.58	0.11	0.66	0.10	0.90	0.06
¹R² variation, RMSEP: root mean square error of prediction (%).

The results of this study also indicated that developing calibrations for digestible amino acids based on a database of samples with known in vivo digestibility does not necessarily result in better calibrations than using samples with known total amino acids and regression equations for converting these totals to digestible amino acids. For example, in vivo digestible lysine could be predicted from a regression equation using total lysine assayed by HPLC with Y=1.10*Lys_HPLC-0.83 (R²=0.91, prediction error (root mean square error)=0.21%). Using this approach, prediction errors (RMSEP) for digestible lysine of 0.26 and 0.28 were obtained for near and mid infrared, respectively. The major advantage of this approach is that any feedmill can develop its own calibrations based on field samples with known total amino acid contents predicted with infrared spectroscopy. The disadvantage is that this method ignores effects on digestibility, but as this explains less than 10% of the variation in digestible amino acids, this error may well be acceptable.

This study focused on animal meals, as they are the most variable feed ingredients. Predicting total and digestible amino acids in other feed ingredients, such as soybean meal and corn, may be possible as well. For soybean meal, virtually all variation in digestible amino acids is caused by variation in total amino acids (van Kempen et al., in press), and it may thus be adequate to develop calibrations for total amino acids only and convert totals to digestible amino acids using regression equations.

For corn, the challenge is that the amino acid content for nutritionally important amino acids is very close to the detection limit of the near infrared instrument (1 g/kg). Thus, calibrations will be more difficult to develop, especially if the reference data (lab amino acid data) have poor repeatability. Mid infrared spectroscopy, given its lower detection limit for certain amino acids (e.g., sulfur amino acids), may do a better job predicting the amino acids in corn. Unfortunately, the authors are not aware of data to substantiate this at this time.

Feed grade fats are commonly used in animal feeds, especially for broilers, as a means to increase the energy density of the diet. These fats though, can greatly affect the quality of the feed and the quality of the end product. The objective of this study (van Kempen and McComas, in press) was to determine the feasibility of using infrared spectroscopy as a quality control tool for feed grade fats.

The quality of fat is assessed by many parameters. Some of the more important parameters to the producer are iodine value, free fatty acid value, and measures of oxidative stability (e.g., initial peroxide value and peroxide value for the active oxygen method at 4 and 20h). Mid infrared spectroscopy is commonly used in the food industry for assessing fat quality, but in the animal feed industry fat quality is typically still assessed using slow wet chemical procedures.

To determine if infrared spectroscopy would be of use for assessing fat quality in the feed industry, feed mills in North Carolina collected 42 samples of feed grade fat from 18 renderers throughout the US. These fats were assayed commercially for their quality. By using these data, calibrations were developed for both mid and near infrared spectroscopy. The sample preparation for both methods consisted of preheating the samples to 65°C. For near infrared spectroscopy, samples were subsequently pipetted into a heated transmission cell with a path-length of 1 mm. For mid infrared spectroscopy, samples were pipetted onto an ATR crystal.

The results from this study are summarized in Table 3. Excellent calibrations were obtained for free fatty acids and iodine value, although mid infrared spectroscopy performed slightly better than near infrared spectroscopy. For moisture, unsaponifiables, and energy, the quality of the reference method limited the quality of the calibration and better reference methods are needed for developing good calibrations. Oxidative stability was difficult to predict with infrared spectroscopy, in line with observations in the literature (Dong et al., 1997).

Table 3. Measurement error and calibration statistics for fat samples analyzed chemically and using near and mid infrared spectroscopy (van Kempen and McComas, 2002).

In this study a major difference between near and mid infrared was apparent. Regression coefficients in mid infrared spectroscopy could be assigned to specific components in the fat samples. Thus, mid infrared spectra can theoretically be used for direct quantification of quality parameters in fat. In contrast, near infrared calibrations were very difficult to interpret, leaving making calibrations using samples with known reference values the only option. Although this method is easy and straightforward, it is more prone to errors when a new source of samples is used, which is chemically distinct from those already present the infrared database. Mid infrared spectra also showed that it was able to measure the parameter of trans fatty acids, whose presence in food may initiate yet another health food craze.

Near Versus Mid Infrared

Although the animal industry predominantly uses near infrared (NIR), the food industry predominantly relies on mid infrared for fat analysis. Near infrared has a distinct advantage in that it has less difficulty with solid samples than does mid infrared. Mid infrared, however, obtains higher quality data such as sharper peaks and relating to a wider variety of organic bonds.

In the current study, these two methods performed equally well for solid samples, although it was apparent from the spectral data that the sample presentation for solid samples used in mid infrared was not ideal. For fat analysis, mid infrared clearly had the edge over near infrared. By using mid infrared it even appeared possible to develop methods that yield quantitative data based on peak integration rather than based on calibrations.

Implications

For quality control, the last few years have likely been the easiest for the feed milling industry. For instance, only a limited number of seed varieties have been grown, while growing conditions have been standardized. Processing conditions have also been standardized as the number of companies processing ingredients, such as soy, have decreased drastically. Thus, feed mills have been receiving a rather homogeneous stream of materials.

Due to environmental pressures and the introduction of genetically modified crops, this is likely to change. The food processing industry is being pressured into finding markets for waste products, and the animal industry is in a good position to benefit from this. Genetically modified crops are likely to diversify the feedstuff market, resulting in crops that may look identical, but which have different nutritional characteristics. Thus, the need for routine quality control methods is likely to explode.

The data presented above illustrate it is possible to make infrared calibrations for nutritionally important quality parameters such as total and digestible amino acids, energy, iodine value and degree of saturation of fats. Although a limited number of feed ingredients were evaluated in this study, examples are available in the literature and in the industry, which prove that similar calibrations can be developed for other feed ingredients. Such calibrations can be used for claims, bin selection, and matrix adjustment.

The key advantages of infrared spectroscopy are its speed and ease of use. Proper implementation of infrared spectroscopy will thus make it possible to evaluate incoming ingredients prior to them being unloaded (without holding up a truck or trainload), and this evaluation can be based on parameters with nutritional relevance, such as digestible amino acids. This then opens the opportunity to start segregating different quality grades of ingredients in a feed mill. For example, animal meals can be possibly rejected or can be segregated into, for instance, less than 2, 2 to 3, and more than 3% digestible lysine. The major benefit of such segregation is that over-formulation in feeds can be reduced, resulting in cheaper diets that also result in less waste production.

Infrared equipment installed in line, already present in some feed mills, signals the next generation of quality control. With these systems, ingredient quality is assessed at the moment that they are dosed out, and instantaneous formulation adjustments are made to guarantee that the final diet is as intended. Just imagine what the possibilities are.

AmiPig. 2000. Association Francaise De Zootechnie, Ajinomoto Eurolysine, Aventis

Animal Nutrition, INRA Institute Technique De La Recherche Agronomique, and ITCF Institut Technique Des Cereales Et Des Fourrages. Paris, France.

Dong, J., K. Ma, F.R. van de Voort, and A.A. Ismail. 1997. Stoichiometric determination

of hydroperoxides in oils by Fourier transform near-infrared spectroscopy. J. AOAC International 80:345-352.

Esbensen, K.H.S., S. Schonkopf, S., and T. Midtgaard. 1996. Multivariate analysis in

Van Kempen, T.A., 2001. Infrared technology in animal production. World. Poult. Sci. J.

Van Kempen, T.A., I.B. Kim, A.J.M. Jansman, M.W.A. Verstegen, J.D. Hancock, D.J.

Lee, V.M. Gabert, D.M. Albin, G.C. Fahey-Jr., C.M. Grieshop, and D. Mahan. Regional and processor variation in the ileal digestible amino acid content of soybean meals measured in growing swine. J. Anim. Sci. (in press)

Parsons, C.M. 1986. Determination of digestible and available amino acids in meat meal

using conventional and caecectomized cockerels or chick growth assays. Br. J. Nutr. 56:227-240.

Qiao, Y. 2001. Routine techniques for monitoring the nutritional value of animal meals