Rapid Assessment of Feed Ingredient Quality

Rapid assessment of the quality of feed ingredients is now crucial for the financial success of intensive livestock producers. A great deal is known about the nutrient requirements, factors determining feed intake and the negative impacts of anti-nutritional factors for all intensively reared livestock including poultry. Diets formulated for optimum performance and profitability need to meet, but not greatly exceed, the nutrient requirements for specific livestock classes and to contain low concentrations of anti-nutritional factors and mycotoxins.

There are large variations across and within feed ingredients in the quantity and availability of energy, amino acids and minerals such as phosphorous. For example, the apparent metabolisable energy (AME) content (MJ/kg) of wheat, barley and triticale has been shown to range between samples within each grain type by over 3 MJ/kg (Scott, 2004; Black et al., 2005). The intake of energy (MJ/day) by broilers varies by approximately 34% when different samples of wheat are incorporated into diets (Black, 2008). Similarly, the variation in nutrient content of Australian canola meal samples is high, with crude protein ranging from 316 to 425 g/kg, lysine from 17.7 to 21.4 g/kg, methionine plus cysteine from 14.0 to 17.3 g/kg, threonine from 13.7 to 16.5 g/kg and total phosphorus (P) from 7.9 to 11.9 g/kg with phytate bound P ranging from 67 to 95% of total P (Spragg and Mailer, 2007). Glucosinolate content also ranged widely from 2.4 to 8.9 mmole/g in these canola samples and can have negative effects on animal performance. Mycotoxin contamination of feed ingredients occurs sporadically, but when present can have a major impact on the performance of poultry (Yunus et al. 2011).

Traditional methods for assessing the nutritional value of feed ingredients such as mean values tabulated for specific feed types or assessing available energy content from test weight (kg/hl) and screening percentage are frequently inaccurate, whereas laboratory analyses are costly and generally provide results long after the ingredient has been used. However, in recent years near infrared spectroscopy (NIR), image analysis and immunoassay or enzymic colorimetric test strips have been developed to allow rapid and often real-time measurement of feed ingredients before being fed to animals. Ingredients can be selected for diet formulation to best meet animal requirements based on these rapid measurements. This paper discusses recent developments in NIR technology for predicting the available energy content (MJ/kg) and intake (MJ/d) of cereal grains for poultry and the available lysine content in oilseed meals. Brief reference is also made to developments of technologies for rapid assessment of other ingredient components including anti-nutritional factors and mycotoxins. 

Provided other nutrient requirements are satisfied, the rate of broiler production is driven by the intake of available energy (MJ/d), whereas the efficiency of production (kg product/kg feed) is driven primarily by the available energy content of the diet (MJ/kg). Consequently, both measures of energy (MJ/d and MJ/kg) are needed to fully describe the apparent metabolisable energy (AME) value of a grain for broilers. A series of experiments have been conducted at the South Australian Research and Development Institute, Roseworthy with 309 cereal grains to measure AME content and AME intake in broiler chickens. The cereal grains examined were wheat, barley, triticale, sorghum, oats, maize and rice. The grains included those grown under near normal environmental conditions, weather damaged grains due to drought, frost and pre-harvest germination, new or unusual cultivars such as high and low amylose grains, hard and soft cultivars, coloured and white endosperm grains and those with unusual NIR spectra. Each experiment included approximately 30% of grains that had been used in previous experiments to provide connectivity between experiments for statistical analysis needed to remove variation between experiments.

Hence, the series of experiments could be regarded as one large experiment, with statistically corrected values for individual grains changing slightly after each experiment. Chickens were allocated on day 22 of age and given 3 days to adapt to the cages and experimental diets. Each cage contained five birds and was made of wire mesh with dimensions of 60 cm x 45 cm and 38 cm high. Trays were fitted below each cage for collection of excreta. All cages had individual feed troughs and drinking nipples and were shielded to prevent cross-contamination of excreta. After the 3 day adaptation period, excreta was collected daily for 4 days and dried overnight at 80ºC in a fan-forced oven. The dried excreta from each cage were pooled over the four day collection period. The birds from each cage were weighed as a group when introduced into the cage and after 7 days. Cage feed consumption was measured on days 3 and 7, then converted to mean daily intake over each period.. Dried feed and excreta samples were finely ground and analysed for gross energy content using an isoperibol bomb calorimeter with benzoic acid standardisation. Diets were formulated to contain per kg: 807 g cereal grain, 155 g casein, 11 g dicalcium phosphate, 13 g limestone, 7 g DL methionine, 2 g mineral-vitamin mixture, 3 g sodium chloride and 2 g choline chloride (60%). Experiments for 175 grains (210 with connectivity grains) included treatments with and without the addition of xylanase and phytase enzymes#.

A sample of each grain was scanned using a FOSS 6500 instrument at the time the grains were prepared into diets to be fed to birds. Scans were conducted on whole and laboratory milled grain samples. Calibrations were developed in WINISI software (FOSS, Denmark) using modified partial least squares regression and cross-validation, with scatter correction and mathematical treatments chosen to optimise calibration statistics. Calibration results presented are following omission of grains seen by the WINISI software as outliers after two passes.

Table 1 shows statistics for the NIR calibrations based on whole grain and milled grain scans for AME (MJ/kg as fed) and AME Intake Index, which is calculated by dividing each value for AME intake (MJ/day) from all experiments by the highest value and multiplying by 100 to give values from theoretically 0 to 100. The index value was used rather than the MJ/d value because the latter changes each day as a chicken grows. The index provides an estimate of the relativity between grain samples in the amount eaten when incorporated into a diet. Values are shown for the first calibration established following experiments with 109 grains and the last calibration with 309 grains. 

Table 1 - NIR calibration statistics for AME content and AME Intake Index following whole grain and milled grain scans for the initial and final calibrations

Inclusion of an additional 200 grains improved substantially both the accuracy and robustness of the calibrations for predicting the AME content and AME Intake Index of cereal grains for broiler chickens compared to the initial calibrations. The accuracy of the calibration (SECV) for AME content based on whole grain scans improved from ± 0.48 for the initial calibration to ± 0.40 (MJ/kg as fed) for the final calibration. This result means that values are predicted with 95% confidence to be within ± 0.80 MJ/kg as fed of the actual value. Similar improvement from ± 4.85 for the initial calibration to 4.31 (% units) for the final calibration for AME Intake Index was found. The robustness of the calibrations as assessed by the Ratio of Prediction to Deviation (RPD) was also substantially improved and means that fewer grains will be seen by the calibrations as outliers. RPD for the NIR calibration predicting AME content increased substantially from 2.40 for the initial calibration to 3.19 for the latest calibration. Values over 3.0 for RPD are generally considered excellent by NIR specialists for such applications. The RPD value for AME Intake Index increased from 1.82 for the initial calibration to 2.38 for the latest calibration. The NIR calibrations predicting AME content are more accurate and robust than for those predicting AME intake, but the latter still provides useful information to separating cereal grains on the basis of likely feed intake.

The R²CALvalues shown in Table 1 of 0.93 for the regression between NIR predicted and observed AME contents of cereal grains are considerably higher than 0.35 for an AME calibration derived from 94 whole grain scanned wheat samples developed by Owens et al. (2009) or the value of 0.45 for predicting AME of wheat from a calibration based on milled grain scans of 160 samples by Garnsworthy et al. (2000). However, Hughes et al. (2014) has shown considerably greater accuracy of prediction when a 'global' calibration including several grain species is used than when within grain calibrations alone are developed. The within grain R² values ranged from 0.23-0.36 for wheat, barley and triticale, but the exception was sorghum where the within grain R² was 0.90.

The accuracy of prediction from the NIR calibration established to predict the AME content of cereal grains for broilers as shown by the SECV of ± 0.40 MJ/kg as fed is considerably less than when the same grains are fed to pigs where a value of ± 0.27MJ/kg for digestible energy (DE) content has been obtained. The number of grains used to establish the calibrations in both species of animals was similar, but the average standard error for measured AME values in broilers was 0.20 MJ/kg compared with 0.11 MJ/kg in pigs. This variation in measurement means that the accuracy of the multigrain NIR calibration for broilers cannot be greater than twice the standard error. Thus, the current accuracy of ± 0.40 MJ/kg as fed is unlikely to be improved unless the experimental variation can be reduced. The lower variation in measurements of DE in pigs compared with measurements of AME in poultry may be due to individual pigs being accounted for in the experimental design, but this is not possible with the broiler experiments where measurements of AME were for a cage of five birds. In addition, differences in gut microbial population between birds and variation in mean retention time of digesta in the gastrointestinal tract of broilers are known to have an effect on AME values (Torok et al., 2008; Hughes, 2008). There appear to be large differences between individual birds in the extent of antiperistaltic waves, which is likely to alter mean retention time of digesta.

Comparison of the calibration statistics between whole grain and milled grain scans (Table 1) suggests there would be little difference in accuracy between the calibrations when used to predict either AME content or AME Intake Index. The accuracy of the final calibration for AME content based on whole grain scans was slightly better than that based on milled grain scans with values for standard error of cross validation being ± 0.40 and ± 0.42, respectively. Robustness of the calibration for AME using whole grain was also slightly better than when the calibration was based on milled grain (RPD, 3.19 compared with 3.10). Similar, slightly better results were found for the calibrations predicting AME Intake Index when using whole grain rather than milled grain.

The mean difference in AME content between whole grain and milled grain across all grains was minimal at 0.016 MJ/kg as fed. However, the regression equation relating predicted AME content (MJ/kg as fed) from calibrations based on milled grain (Y) or whole grain (X) was: Y=0.97+0.33. The R² for the regression was only 0.92, which means there are considerable differences in the predicted values for some individual grains. The greatest difference in predicted AME content between whole grain and milled grain was 1.19 MJ/kg as fed for one of the rice samples. Approximately 18% of all samples had differences in predicted values greater than the standard error of cross validation of ± 0.40 MJ/kg as fed for the whole grain calibration.

Similarly, the mean difference in AME Intake Index between whole grain and milled grain scans across all grains was minimal at 0.101 % units. However, the R² between the values predicted using whole grain or milled grain is only 0.88. The greatest difference in predicted AME Intake Index between whole grain and milled grain was 11.01 % units and approximately 16% of all samples had differences in predicted values greater than the standard error of cross validation of ± 4.2 % units for the whole grain calibration. This comparison highlights one of the difficulties with NIR technology in statistically relating spectra obtained from different preparations of the same grain to measured values. It would be expected that the differences between predicted values from calibrations based on whole or milled grain scans would decrease as the number and range of samples was increased.

The response in AME content (MJ/kg) and AME intake (MJ/d) to the addition of xylanase and phytase enzymes varied greatly within and between grain types (Table 2). The greatest mean response in AME content of 0.74 MJ/kg as fed was for wheat samples, with a substantially lower mean response of 0.29 MJ/kg as fed for triticale and 0.22 MJ/kg as fed for barley. The mean response to enzymes included in sorghum or maize diets was low, with mean enzyme response values of 0.04 and 0.02 MJ/kg as fed, respectively. Six of the 35 barley samples, 20 of the 66 sorghum samples and four of the ten maize samples gave a negative response to enzymes. A similar pattern of response to enzymes was found for AME intake, except the response for maize was high in some samples. 

Table 2 - Measured mean, minimum and maximum responses to xylanase and phytase enzymes for AME content (MJ/kg as fed) and AME intake (MJ/d) in broiler chickens fed diets with different grain types

Results from the responses to enzymes in AME content and AME intake were used to develop NIR calibrations for all grains including connectivity grains where enzymes were included in the diet. Statistics for calibrations developed using whole scans are shown in Table 3. A satisfactory NIR calibration could not be developed for the response in AME intake to enzymes, whereas the calibration for AME content was of marginal value. The response in AME content to enzymes could be predicted with a 95% probability to ± 0.38 MJ/kg as fed. The robustness of the calibration, as illustrated by the RPD of 1.81, was considered to be sufficient for distinguishing between high and low responders. However, the addition of more samples should strengthen the calibration. 

Table 3 - NIR calibration statistics for AME content and AME intake using whole grain scans for the response to xylanase and phytase enzymes (¹Refer to Table 1 for abbreviations)

Variation in processing methods and temperature applied to oilseeds during the oil extraction process can markedly affect the availability for animals of essential amino acids, particularly lysine (Newkirk and Classen, 2002). Wide variation in amino acid availability between specific batches of oilseed meals can have marked effects on the efficiency of production, time taken for animals to reach market specifications and profitability of enterprises. The heating of proteins, particularly in the presence of reducing sugars, is known to induce the Maillard reaction, rendering the lysine incorporated into reaction products unavailable for metabolism by animals. However, a proportion of early Maillard or Amadori products can be reconverted to lysine during conventional amino acid analyses and is falsely assumed to be available to animals. True lysine can be converted to homoarginine through a guanidination reaction for the measurement of lysine available to animals for metabolism. Lysine measure by this process has been called reactive lysine (Moughan and Rutherfurd, 1996). Heat treatment of oilseed meals reduces the measured total lysine content, but reactive lysine as a proportion of total lysine can vary widely depending on the amount and time of heat treatment.

Two projects in Australia have been designed to develop NIR calibrations to predict the total and reactive lysine content of canola meal and soybean meal. Canola samples were collected from six commercial oilseed crushing plants using expeller and solvent oil extraction procedures (Spragg, 2011). The cooking temperatures and times for processing some of the samples were either increased or decreased compared with normal procedures. Total and reactive lysine content was measured on 129 samples, with total lysine ranging from 15.8 to 24.4 g/kg and reactive lysine as a proportion of total lysine ranging from 0.66 to near 1.0. NIR calibration statistics for the total and reactive lysine content of canola meal are shown in Table 4. Although these calibrations are promising, RPD values are relatively low at 2.14 for total lysine and 2.00 for reactive lysine. Consequently, an additional 60 samples, some of which have been autoclaved for different times, are being added to increase the range in values and to strengthen the NIR calibrations.

Soybean meal samples imported into Australia from major soybean meal producing countries (USA, Brazil, China, Argentina, India) have been collected with 216 being analysed for total and reactive lysine (Kim and Mullan, 2013). In addition, a further 68 soybean meal samples and 25 soy protein concentrate samples were subjected to autoclaving at 134°C and 217 Kpa for periods from 5 to 30 minutes prior to analysis for total lysine, reactive lysine and other amino acid contents. There were significant negative, linear effects of time of heat treatment on the measured total lysine, reactive lysine, arginine and cysteine concentrations of the meal. NIR statistics for total and reactive lysine content of combined soybean meal and soy protein concentrate are given in Table 4 and are highly positive. The standard error of cross validation was ± 1.02 g/kg and 0.96 g/kg (as fed) and R² values of 0.94 and 0.95, respectively, for total and reactive lysine. These values mean that total and reactive lysine content of soybean meal and soy protein concentrate should be predicted with 95% confidence to be within 2.04 and 1.92 g/kg, respectively, of the actual value. The RPD values of 3.35 and 3.94, respectively, for total and reactive lysine, indicate that the calibrations have high robustness for predicting accurately values for unknown soybean meal and protein concentrate samples. Amino acid digestibility assays were also conducted during the experiments and NIR calibrations were developed for apparent, standardised and true ileal digestible total lysine and reactive lysine. RPD values for each calibration were between 3.4 and 4.9 (results not shown), indicating robustness in prediction. 

Table 4 - NIR calibration statistics for the total and reactive lysine content (g/kg as fed) in canola meal and soybean meal - protein concentrate (¹Refer to Table 1 for abbreviations)

In addition to the energy content of cereal grains and available lysine content of oilseed meals, other information about feed ingredients is required to optimise diet formulation. Several commercial companies now provide rapid turnaround times for NIR analysis of many nutrients in feed ingredients including moisture, crude protein, essential amino acids, starch, ether extract, ash, acid detergent fibre, neutral detergent fibre, total and phytate bound phosphorus. The ingredients are scanned by the end-user and the scans uploaded via the world wide web where they are analysed using a range of NIR calibrations and the results returned promptly. Some companies also provide estimates of standardised digestible amino acid content. NIR calibrations also have been developed for several anti-nutritional factors including glucosinolates and sinapines in canola meal (Font et al., 1999; McFadden and Mailer, 2003). Image analysis or machine vision technology is now well advanced and is used in many industries for quality control. Image analysis software has been developed to identify cereal grain quality (Guevara-Hernandez and Gomez-Gil, 2011) and weed seed contamination (Pablo et al., 2002). Similarly, immunoassay and enzyme binding techniques with colorimetric analysis via test strips or image analysis have been developed to identify the presence of mycotoxins including ochratoxin A contamination in cereal grains (Bazin et al., 2010) and various of aflatoxins (Moscone et al., 2011).

The advent of many accurate and rapid methods for assessing the quality of feed ingredients before they are used by poultry producers greatly increases the opportunity to formulate diets that will optimise nutrient utilisation, reduce concentrations of anti-nutritional factors and decrease the risk of mycotoxin contamination to greatly improve productivity and enterprise profitability. 

ACKNOWLEDGEMENTS: Partial funding from the Rural Industries Research and Development Corporation Chicken Meat Program and the Cooperative Research Centre for High Integrity Pork is acknowledged. 

Black JL (2008) Premium Grains for Livestock Program: Component 1 – Coordination. Final Report, Grains R&D Corporation, Canberra, Australia.

Black JL, Hughes RJ, Nielsen SG, Tredrea AM & Flinn PC (2009) Proceedings of the Australian Poultry Science Symposium 20: 31-34.

Black JL, Hughes RJ, Nielsen SG, Tredrea AM, MacAlpine R & van Barneveld RJ (2005) Proceedings of the Australian Poultry Science Symposium 17: 21-29.

Bazin I, Nabais E & Lopez-Ferber M (2010) Toxins (Basel) 9: 2230-2241.

Font R, Del Rio M, Dominguez J, Fernandez-Martinez JM & De Haro A (1999) 10th International Rapeseed Congress.

Garnsworthy PC, Wiseman J & Fegeros K (2000) Journal of Agricultural Science, Cambridge 135: 409-417.

Guevara-Hernandez F & Gomez-Gil J (2011) Spanish Journal of Agricultural Research 9: 672-680.

Hughes RJ (2008) British Poultry Science 49: 716-720.

Hughes RJ, Black JL, Flinn PC, Tredrea AM & Diffey S (2014) Proceedings of the Australian Poultry Science Symposium 24: (this publication)

Kim JC & Mullan BP (2013) Further development of a reactive lysine NIR calibration for soybean meal, Final Report Pork CRC, Roseworthy, Australia.

McFadden A & Mailer RJ (2003) 13th Australian Research Assembly on Brassicas pp. 98-100.

Moscone D, Fabiana A & Aiz A (2011) Methods in Molecular Biology 739: 217-235.

Moughan PJ & Rutherfurd SM (1996) Journal of Agriculture and Food Chemistry 44: 2202-2209.

Newkirk RW & Classen HL (2002) Poultry Science 81: 815-825.

Owens B, McCann MEE, McCracken KJ & Park RS (2009) British Poultry Science 50: 103-122.

Pablo MG, Hugo DN, Pablo FV & Ceccatto HA (2002) Computers and Electronics in Agriculture 33: 91-103.

Torok VA, Ophel-Keller K, Loo M & Hughes RJ (2008) Applied and Environmental Microbiology 74: 783-791.

Scott TA (2004) Proceedings of the Australian Poultry Science Symposium 16: 9-16.

Spragg JC (2011) Canola meal value chain quality improvement-Stage 3, Report, Pork CRC, Roseworthy, Australia.

Spragg JC & Mailer R (2007) Canola meal value chain quality improvement, Final Report, Pork CRC, Roseworthy, Australia.

Yunus AW, Razzazi-Fazeil E & Bohm J (2011) Toxins (Basel) 3: 566-590.