PROTEIN AND AMINO ACID CONTENT OF FEED INGREDIENTS AND THE IMPACT OF DIETARY BALANCED PROTEIN ON MAXIMIZING ECONOMIC RETURNS FROM

Transcription

1 i PROTEIN AND AMINO ACID CONTENT OF FEED INGREDIENTS AND THE IMPACT OF DIETARY BALANCED PROTEIN ON MAXIMIZING ECONOMIC RETURNS FROM BROILER TECHNICAL RESPONSES by NUNTAWADEE SRIPERM (Under the Direction of Gene M. Pesti) ABSTRACT Dietary balanced protein (BP) has been adopted in broiler feed formulation. It not only considers the crude protein (CP) content of the diet, but more importantly the essential or indispensible amino acids (AAs) are considered and maintained in balance. The BP concept is when AAs are set relative to lysine. Since AAs play an important role in determining feed cost and bird performance, using AAs efficiently by minimizing the excess is a method of minimizing feed cost. This research was conducted to obtain the data need to understand AA content in major feed ingredients, which is useful in adjusting ingredient matrix values to meet the targeted nutrients. The CP content of ingredients is not a reliable measurement in feed formulation. An experiment was conducted to show the impact of broiler responses from different digestible lysine levels during two phases. The results were used to generate different production functions in order to determine the most profitable feeding level at various input and output prices. INDEX WORDS: Dietary Balanced Protein, Digestible Lysine, Crude Protein, Maximum Profit, Broiler Production, Corn, SBM, Amino Acid Content

2 ii PROTEIN AND AMINO ACID CONTENT OF FEED INGREDIENTS AND THE IMPACT OF DIETARY BALANCED PROTEIN ON MAXIMIZING ECONOMIC RETURNS FROM BROILER TECHNICAL RESPONSES by NUNTAWADEE SRIPERM B.S., King Mongkut Institute of Technology Lardkrabang, Thailand, 2002 A Dissertation Submitted to the Graduate Faculty of the University of Georgia in Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY ATHENS, GEORGIA 2011

3 iii 2011 Nuntawadee Sriperm All Rights Reserved

4 iv PROTEIN AND AMINO ACID CONTENT OF FEED INGREDIENTS AND THE IMPACT OF DIETARY BALANCED PROTEIN ON MAXIMIZING ECONOMIC RETURNS FROM BROILER TECHNICAL RESPONSES by NUNTAWADEE SRIPERM Major Professor: Committee: Gene M. Pesti Michael E. Wetzstein Sammy Aggrey Jack E. Houston Scott M. Russell Timothy A. Park Electronic Version Approved: Maureen Grasso Dean of the Graduate School The University of Georgia August 2011

5 v ACKNOWLEDGEMENTS I would like to thank all the faculty, staff, colleagues, and friends who helped me with my research. Without their help, my work would have been more difficult. I owe my deepest gratitude to my committee members: Dr. Michael E. Wetzstein, Dr. Jack E. Houston, Dr. Timothy A. Park, Dr. Sammy Aggrey, and Dr. Scott M. Russell for their support. A special thanks to Dr. Wetzstein for the encouragement, guidance and support from the beginning till the end. It was an honor for me to work with Dr. Houston and Dr. Park as their guidance and advice helped me in my research and classes. I appreciate Dr. Aggrey for letting me borrow useful books and his advice and Dr. Russell for being part of my committee and his valuable suggestions. Lastly, I offer my regards and blessings to all of those who supported me during my research trial; Jessie, Chris, Liz, Clayton, Dean, Carl, Paula, Danny, Dewey and Paul. It was a pleasurable time to spend with those friends. Thanks to my family and friends for their love and encouragement which pushed me through to the end and helped me reach my goal.

6 vi TABLE OF CONTENTS Page ACKNOWLEDGEMENTS. v LIST OF TABLES. vii LIST OF FIGURES... xi CHAPTER 1 INTRODUCTION LITERATURE REVIEW 7 3 THE DISTRIBUTION OF CRUDE PROTEIN AND AMINO ACID CONTENT IN MAIZE GRAIN AND SOYBEAN MEAL EVALUATION OF THE FIXED NITROGEN-TO-PROTEIN (N:P) CONVERSION FACTOR (6.25) VERSUS INGREDIENT SPECIFIC N:P CONVERSION FACTORS IN FEEDSTUFFS.74 5 RESPONSE SURFACE ANALYSIS OF DIETARY BALANCED PROTEIN RESPONSES DURING GROWER AND FINISHER PHASES OF ROSS 708 MALE BROILERS 91 6 OPTIMIZING BROILERS DIETARY BALANCED PROTEIN AND ECONOMIC RETURNS CONCLUSION 165 APPENDICES. 170

7 vii LIST OF TABLES TABLE PAGE 2.1 University of Illinois ideal ratios for selected amino acids at three growth periods Ideal amino acid ratios (%) of essential amino acids for broiler chicks Example of Growth and Production Functions used to determine growth or responses in animal Descriptive statistics and Shapiro-Wilk test of crude protein and amino acids in maize grain samples Descriptive statistics and Shapiro-Wilk test of crude protein and amino acids in soybean meal samples Linear and quadratic relationships between crude protein and amino acids and in maize grain and soybean meal Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in Descriptive statistics of crude protein and amino acids in soybean meal samples

8 viii analyzed in Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in Total amino acid and ammonia composition, DM content and sample size (g kg -1 ) of 5 feedstuffs Nitrogen to protein (N:P) conversion factors, total nitrogen content and nitrogen recovery of 5 feedstuffs Linear regressions of total anhydrous amino acid residues ( E i ) or total nitrogen from amino acid and NH 3 ( D i ) as a function of total nitrogen content from the Dumas method (N L ) for 5 feedstuffs Specific crude protein, crude protein and true protein content of 5 feedstuffs Composition of the experimental diets (%, as-fed basis) The formulated and analyzed nutrient composition of the experimental diets and relative amino acid ratio Analyzed nutrient composition of 9 treatments Growth performance of Ross x Ross 708 male broilers fed different digestible lysine (dlys) levels during 15 to 34 d of age Processing characteristics of Ross x Ross 708 male broilers fed different digestible lysine (dlys) levels during 15 to 34 d of age Growth performance of Ross x Ross 708 male broilers fed different digestible lysine (dlys) levels during 35 to 49 d of age.. 113

9 ix 5.7 Processing characteristics of Ross x Ross 708 male broilers fed different digestible lysine (dlys) levels during 35 to 49 d of age Growth performance of Ross x Ross 708 male broilers fed different digestible lysine level (dlys) levels during the grower (15-34 d) and finisher phases (35 to 42 d) Processing characteristics of Ross x Ross 708 male broilers fed different digestible lysine level (dlys) levels during the grower (15-34 d) and finisher phases (35 to 42 d) Growth performance of Ross x Ross 708 male broilers fed different digestible lysine level (dlys) levels during the grower (15-34 d) and finisher phases (35 to 49 d) Processing characteristics of Ross x Ross 708 male broilers fed different dlys levels during the grower (15-34 d) and finisher phases (35 to 49 d) Digestible lysine (dlys) response levels of Ross x Ross 708 male broilers based on quadratic model at 15 to 35 d and quadratic response surface model at 15 to 42 d and 15 to 49 d Analyzed nutrient composition of the experimental diets Comparison of the experimental results for BW using the Contrast procedure of SAS (2004) Regression coefficients and statistics for BW and cumulative feed intake (CFI) of broilers at 49 d of age Residuals analysis of the Cobb - Douglas production function, quadratic polynomial, ascending quadratic with plateau and Chen-Clayton models for BW of broilers at

10 x 49 d of age Residuals analysis of the Cobb - Douglas production function, quadratic polynomial, ascending quadratic with plateau and Chen-Clayton models for feed intake of broilers at 49 d of age Optimization analysis of changing feed ingredient and broiler prices on dlys levels during the grower (15 to 34 d) and finisher (35 to 49 d) phases using Cobb-Douglas production functions Optimization analysis of changing feed ingredient and broiler prices on dlys levels during the grower (15 to 34 d) and finisher (35 to 49 d) phases using quadratic polynomial and ascending quadratic with plateau functions Optimization analysis of changing feed ingredient and broiler prices on dlys levels during the grower (15 to 34 d) and finisher (35 to 49 d) phases using Chen-Clayton functions Advantage and disadvantage of applying models to nutrient dose-response data

11 xi LIST OF FIGURES FIGURE PAGE 2.1 Graphical illustration of central composite rotatable design for two factors with the combination of 9 treatments Histograms with probability density function of lysine (normal distribution) and arginine (non-normal distribution) of maize grain samples analyzed between 2002 and Quantile quantile plot of lysine (normal distribution) and arginine (non-normal distribution) of maize grain samples analyzed between 2002 and Histograms with probability density function of valine (normal distribution) and lysine (non-normal distribution) of soybean meal samples analyzed between 2002 and Quantile quantile plot of valine (normal distribution) and lysine (non-normal distribution) of soybean meal samples analyzed between 2002 and A plot of total anhydrous amino acid residues ( E i ) on total nitrogen from the Dumas method (N L ) for soybean meal A plot of total anhydrous amino acid residues ( E i ) on total nitrogen from the Dumas method (N L ) for soybean meal Nine diet combinations of grower dlys levels and finisher dlys levels

12 xii used in Experimental Phase Growth performance of grower phase (15 to 34d) Processing characteristics of grower phase (15 to 34d) Growth performance of finisher phase (35 to 49d) Response surface BW gain from 15 to 49 d of age Response surface of feed intake from 15 to 49 d of age Response surface of FCR from 15 to 49 d of age Three-dimensional surfaces plot of predicted values of BW (kg) from the growth function of broilers fed diets with different levels of dlys during the grower and finisher phases at 49 d of age using the Cobb-Douglas production function Three-dimensional surfaces plot of predicted values of BW (kg) from the growth function of broilers fed diets with different levels of dlys during the grower and finisher phases at 49 d of age using the Chen-Clayton model Three-dimensional surfaces plot of predicted values of BW (kg) from the growth function of broilers fed diets with different levels of dlys during the grower and finisher phases at 49 d of age using the Quadratic Polynomial function Three-dimensional surfaces plot of profit maximizing over feed cost from the profit function using Cobb-Douglas production functions for broilers fed diets with different levels of dlys during the grower and finisher phases up to 49 d of age, based on scenario when corn, SBM, and broiler prices

13 xiii were at $309 and $424 per metric ton, and $1.42 per kg live bird, respectively Graphical illustration of the impact of variation in feed costs and live broiler prices on net returns based on fifty combinations of selected price scenarios using the Cobb-Douglas production functions A 1.1 Average relative humidity of rooms 1 and 2 recorded during the experiment of Chapter A 1.2 Average temperature of rooms 1 and 2 recorded during the experiment of Chapter A 1.3 Daily low and high temperature of rooms 1 and 2 recorded during the experiment of Chapter B 1.1 SAS codes for analyzing central composited rotatable design, residuals test and nonlinear models

14 1 CHAPTER 1 INTRODUCTION The primary goal of feed formulation within a poultry company should be to maximize profits and not to just minimize feed costs. It has been noted by several researchers (Eits, et al., 2005; Guevara, 2004; Sterling, et al., 2005; Vedenov and Pesti, 2010), formulating diets to maximize profitability, rather than to maximize body weight gain or breast meat yield can increase the profitability of a broiler production system. Similarly, formulating diets to minimize feed utilization or to reduce feed cost per ton via lowering nutrient density, might not be the most cost effective means of maximizing profitability if it adversely affects body weight. The idea behind maximizing profitability through nutrition is to formulate to the optimal nutrient density, which accounts for variation in feed price, production cost and broiler market price. Simply lowering nutrient density in an attempt to save on feed costs may not generate more revenue or economic return for the company. Determining the optimal economic level of nutrition, rather than just the lowest cost of feed is the sensible means of setting nutrient requirements. In broiler feed formulation, about one-third of the diet cost comes from ingredients (soybean meal (SBM), animal by product meal, supplemented amino acids) used to meet the crude protein (CP) and amino acid (AA) requirement of broilers. Thus, decreasing excess CP and nonessential (dispensable) AA contents based on the bird s requirement should improve feed formulation efficiency, which reduces the cost of production. The process of minimizing excess CP and AA contents starts with properly knowing the nutrient content of feed ingredients.

15 2 Soybean meal is the main feed ingredient supplying protein and AA in broiler diets, accounting for approximately 25% or more of the grower diet. Thus, knowledge of CP and AA nutrient composition in SBM plays a key role in feed formulation. SBM is normally analyzed by wet chemistry methodology or by using predicted values such as those estimated by near-infrared spectroscopy. These analyzed or predicted nutrient levels should be a close estimate of the true values of a received batch of SBM, compared with standard book values (NRC, 1994). However, there is no perfect method; thus, the decision has to be made based on knowledge of biology and statistics. Replacing a portion of SBM with supplemental AA is a way to minimize the excess CP and nonessential AA contents. When SBM is partially replaced with supplemental AA, a portion of the AAs (essential and non essential AA) are reduced, which also lower the CP levels. Specific AA levels that may be removed, while maintaining or improving overall broiler performance and feed utilization, and reducing nitrogen excreta has been widely studied (Aletor et al., 2000; Rezaei et al., 2004; Si et al., 2004). Initially, emphasis was placed only on CP, but over time, formulating diets to specific AAs that could be economically supplemented became a favorable approach. Additional knowledge was gained in regards to the essential AAs which were not being supplemented. With the evolution of AA supplementation, this led to research focused on a more balanced solution by using the optimal proportion of other AAs to just one AA, in most studies that being lysine (Baker, et al., 2002; Emmert and Baker, 1997; Mack, et al., 1999; Wijtten, et al., 2004). It has been shown that the modern high yield broiler genotype, which has a large capacity for breast meat deposition, has an increased lysine requirement. Thus today, nutritionists employ the dietary balanced protein (BP) or ideal protein concept or ideal amino acid ratios as all essential AAs held in ratios to lysine. Lysine has been selected because it

16 3 is the second-limiting AA in broiler diets, is only used for protein synthesis, and is relatively easy to assay. Moreover, dietary lysine does not interact metabolically with any other amino acids and is used primarily for protein accretion, not as a precursor for other functions as is methionine. It has proved to be more applicable to define the lysine requirement based on empirical evidence and calculate the other AA requirements based on the proper ratio to lysine based upon gender, age, environment and body composition (D'Mello, 2003; Garcia, et al., 2006). When modeling the response to BP, the key AAs are kept proportional to the dietary lysine level such as the responses measured by Baker, et al. (2002) and Lemme et al. (2008). Moreover, the broiler s needs for both the essential and the non-essential amino acids as well, should be accounted for. Diets low in dietary lysine, especially during early development, result in reduced breast meat formation because protein accretion from protein synthesis declines (Tesseraud et al., 1996). Thus, dietary lysine is more precise, and has become accepted, to use as a target in feed formulation compared with crude protein, which is an indirect calculation from the lab, based on the determination of the nitrogen level. Digestible lysine level (dlys) is discussed within this dissertation. It does not mean that only lysine level was considered, but the term is used to represent the nutrient density of the diet. Hence the "optimal nutrient density" or the "derived recommendation" is really the point of maximum economic efficiency. Because it is an economic measure, it will change with changing economic conditions. The research within this dissertation was conducted to determine the maximum broiler technical performance and optimum broiler profitability based on BP for a particular commercial broiler strain. Emphasis was placed on determining the optimal economic dlys level during two

17 4 feeding periods; 1) grower (15 to 34 d), and 2) finisher (35 to 42 and 49 d) phases, based on variations in production costs and meat prices. The growth and processing responses were studied to determine whether there is an interaction between feeding different dlys levels during the grower and finisher phases. This information will be presented in the following chapters. The third chapter is focused on determining the distribution of CP and AA contents of the two primary feed ingredients (corn and soybean meal) used in the United States. Knowing the nutrient composition of these ingredients before formulating the feed is important. Several methods, such as laboratory analysis involving the average of the collected samples and using regression equations, are generated in order to obtain those values. According to the study, the linear or quadratic equation provides poor prediction for AAs as a function of CP. In addition, CP and some AA contents of each feed ingredient are not normally distributed. Thus, other model fitting procedures beside linear or quadratic regression equations are needed to improve model predictions. The average value does not represent the sample mean if the sample distribution is not normally distributed. This could lead to formulating to an actual level lower or higher than the targeted level. As such, actual nutrient levels in the feed would not reflect what was expected in the formulation process. The fourth chapter evaluated the ingredient specific nitrogen to protein (N:P) conversion factor (ISNF) in feedstuffs, which should be applied in feed formulations rather than crude protein (percentage of nitrogen from wet chemistry times 6.25). Because each feed ingredient contains different AA levels and thus different nitrogen levels, it should not be assumed that all ingredients contain 16% N (100/16 = 6.25). Using an ISNF is more realistic and adequately represents protein values of the diet. However, true protein (the summation of all the amino acid content from the amino acid analysis) is perhaps the best approach to estimate the actual protein

18 5 in feedstuffs. When the amino acid laboratory test or an AA profile is not available, using the proposed ISNFs to calculate the specific crude protein based upon a simple N analysis, will closely estimate the true protein in feedstuffs. In the fifth chapter, the effect of BP on technical performance, body weight, body weight gain, feed intake, feed per gain and carcass composition, of a commercial broiler strain during both the grower (15 to 34 d) and finisher (35 to 42 and 49 d) was determined. These data were generated across 9 dose-response treatment diets defined by different dlys levels, using the Central Composite Rotatable Design (CCRD). For this study, large growing birds having a final body weight of approximately 4 kg were measured. The dlys level dictated the BP since key essential amino acids were set as minimums relative to dlys. The results showed that during the grower phase the dlys levels yielding maximum responses ranged from to 1.388% for growth performance and various processing characteristics. Evaluating the combination of both phases up to 42 d, the dlys levels yielding maximum responses for growth performance and processing characteristics were mostly higher than the dlys levels used up to 49 d. There were no interactions for any of the responses, which meant that responses to BP during the finisher phase were independent of those used in the grower phase. In the sixth chapter, an alternative model called the Cobb-Douglas production function (CD), ascending quadratic with plateau (QPP) and Chen-Clayton (CC) models, that were compared with the quadratic polynomial (QP) using the nutrient dose-response data from Chapter 5. The profit function that provides the maximum profit, under any pricing scenario, as well as the dlys levels which maximize profit can be determined. The variation in feed ingredient prices (corn and soybean meal) and broiler prices were used in profit functions to demonstrate the prediction capability of the model and elasticity. Theoretically, the CD, CC and QPP are better than QP

19 6 model because QP is hyperbolic and reaches the maximum point and then declines, while the CD closely resembles a plateau (asymototic), that is observed before AA or crude protein becomes toxic and reduce bird performance. The results from this study demonstrate the potential of using the Cobb-Douglas function in decision making in broiler feed formulation. Feeding broilers at optimum dlys levels based on these models showed that profitability in all scenarios were higher than the profit using the recommended requirement levels by Ross (Aviagen, 2007).

20 7 CHAPTER 2 LITERATURE REVIEW Crude Protein in Feedstuffs Crude protein (CP) has been widely accepted as an indicator of protein content of feedstuffs or formulated feed for monogastric animals. Crude protein is calculated by multiplying the total nitrogen determined by methods such as Kjeldahl or Dumas by a standard nitrogen conversion factor of The 6.25 factor does not relate to any specific feedstuff (i.e. it is a non-ingredient specific nitrogen-to-protein (N:P) conversion factor, which assumes one kg of plant or animal protein contains 160 g N (1000/160 = 6.25, Jones, 1931)). The N:P conversion factor in feedstuffs can vary depending upon the amino acid composition, as the nitrogen content of each amino acid is different. For example, the variation in N:P conversion factors, such as in meat products, differs depending upon the collagen level. Since the nitrogen content of collagen is 180 g per kg, it is clear that the level of collagen can affect the N:P factor in meat products. Moreover, the nitrogen content of feedstuffs is not just derived from protein and amino acids, so other nitrogenous organic compounds need to be taken into account. These include nucleic acids, urea, ammonia, phospholipids, nitrates, purine derivatives, etc (Mariotti et al., 2008). Thus, using a traditional fixed factor of 6.25 to calculate the determined total organic N content of a protein is not a realistic approach. Alternative N:P conversion factors other than 6.25 have been proposed by many researchers (Mosse, 1990; Sosulski and Imafidon, 1990; Mariotti et al., 2008).

21 8 True Protein in Feedstuffs Protein consists of at least 20 different amino acids bound together by peptide bonds between the carboxyl and amino groups of adjacent amino acids. The process of protein formation eliminates one molecule of water, from dehydration synthesis, for each peptide bond formed between the adjacent amino acids (Garrett and Grisham, 2007). Salo- väänänen and Koivistoinen (1996) proposed a calculation of net (or true) protein to be used instead of CP. True protein (TP) is the summation of the total amino acid residues from the amino acid analysis. As protein consists of these amino acids, TP closer estimates protein in feedstuffs instead of using a factor 6.25 when calculating CP from nitrogen content in feedstuffs. Mosse' et al. (1990) suggested crude protein calculation using the best estimate of TP. Estimation of Amino Acid (AA) Contents as a Function of Crude Protein in Feedstuffs Knowing AA contents in feedstuffs is an important step in feed formulation. Several research studies have been done in an attempt to evaluate the relationships between CP and AA using linear and quadratic regressions, along with other mathematical model approaches (Roush and Cravener, 1997; Cravener and Roush, 1999; Cravener and Roush, 2001). Roush and Cravener (1997) studied linear regression (LR) and artificial neural networks (ANN), which are biological programs used in regression analysis for complex data. The LR and AAN were used for predicting amino acid levels (methionine, methionine + cystine, lysine, threonine, tyrosine, tryptophan, and arginine) in corn, wheat, soybean meal (SBM), meat and bone meal (MBM), and fish meal (FM) using proximate analysis input (crude protein, fat fiber, moisture, ash and AA levels). The results showed that ANN provided higher R 2 values, a measurement of the amount of variability explained by the model, compared with LR. The authors concluded that ANN was more effectively identified the complex relationship between nutrients and feed ingredients than

22 9 did LR. Cravener and Roush (1999) studied three types of ANN (NeuroShell 2): three-layer backpropagation (BP3), Ward Backpropagation (WBP), a general regression neural network (GRNN) and LR using SAS PROC GLM. They found that GRNN provided the highest R 2 in all nutrients and feed ingredients. The GRNN provided much better R 2 compared with LR, for example, for methionine in corn, R 2 from GRNN was 0.95; while, LR was Cravener and Roush (2001) showed that the maximum R 2 value was highest when a GRNN with iterative calibration (GRNNIT) compared with LR and two other types of ANN. This study investigated the potential of a new method of calibration using the genetic algorithm (GA) to optimize GRNN smoothing values. A GRNN architecture (NeuroShell 2 Software) with GA calibration (GRNNGA) was used to train an ANN to predict AA levels in maize, SBM, MBM, FM and wheat, based on proximate analysis input. The GA calibration gave maximum R 2 values in all nutrients and ingredients compared with LR. However, GA calibration was not always provided the highest R 2 values among all types of ANN. These studies showed that ANN can be used as a successful alternative to statistical regression analysis for predicting AA levels in feed ingredients. However, the ANN program is quite costly and uncertain with regard to repeatability. Dietary Balanced Protein (BP) or Ideal Protein (IP) or Ideal Amino Acid (IAA) ratios in Broilers The Ideal Amino Acid (IAA) ratios were first implemented for practical diet formulation of pig by the British Agricultural Research Council (ARC, 1981). Today, IAA ratios, using lysine as the reference amino acid, are used all over the world for diet formulation of pigs (D'Mello, 2003). In broiler feed formulation, CP had been widely used as an indicator of protein content in the diets. Earlier studies focused on CP, energy and body composition (Fraps, 1943;

23 10 Hill and Dansky, 1950; Pesti et al., 1986; Bartov and Plavnik, 1998). The conclusions were that carcass fat content declined when feeding low dietary energy level in isonitrogenous diets, feed consumption decreased while carcass protein, carcass yield and breast meat yield increased when feeding high dietary CP level in isocaloric diets. Later researches (Baker, 1978; D'Mello, 1994; Morris et al., 1999) put more emphasis on AA composition in the diet. Baker (1978) concluded that lysine utilization in soybean meal (SBM) decreased because of excess amino acid in SBM. D Mello (1994) suggested that the excess AAs decreased the lysine utilization because chickens put more emphasis on metabolizing the excess AAs, resulting in less lysine utilization efficiency. Morris et al. (1999) reported AA imbalance at higher CP levels reduced the effectiveness of utilization of the first-limiting AA, causing an increase in the requirement and a decrease in growth. The authors stated that the difficulty of applying amino acid supplements in chicken diets was pricing, transportation, and government policies in order to be able to meet the minimum recommended values for growth. However, today there are plenty of AA supplements available for chicken. DL-Methionine provides methionine which is the first-limiting AA. L- lysine provides lysine which is the second-limiting AA. L-threonine provides threonine which is the third-limiting AA, L-valine provides valine which is the forth-limiting AA in corn-soybean meal diets without animal by-products (Corzo, 2008). L-tryptophan provides tryptophan which is the forth-limiting AA in corn-soybean meal diets with meat blend of poultry meal, meat and bone meal, and feather meal (Kidd and Hackenhaar, 2005). The earlier approach of listing broiler requirements for all essential AA at various stages of growth and maintenance received many arguments. These led to more research focused on a better solution by using the optimal proportion of the other AA to just one AA (AA ratios), in most studies were lysine. As new broiler strains are developed to lysine requirements needed.

24 11 Thus, today nutritionists employ the BP, IP, or IAA concept as all essential amino acids are held in ratios to lysine. Lysine is an essential amino acid or indispensable amino acid out of 12 essential amino acids required by the chicken. Because it cannot be synthesized by the chicken, the chicken can only obtain it from the diet. Protein synthesis is primary function of lysine in the body. Lysine has been selected because it is the second-limiting AA in corn-soybean meal diets for poultry. It is only used for protein synthesis. It is relatively easy to assay. Moreover, it is not involved in any other biochemical pathways, unlike methionine. Many AA requirements were neither unknown nor they were known by feeding variety of protein levels, energy levels and stages of animal. It would be more applicable to define the lysine requirement based on empirical evidence and calculate the other AA requirements based on the proper ratio to lysine at a given set of gender, age, raring environment, dietary and body composition parameters (D'Mello, 2003; Garcia et al., 2006). Emmert and Baker (1997) suggested IAA ratios for older birds based on best empirical estimates of lysine, sulphur AA (SAA), and threonine requirements of broiler chicks during 0 to 21, 21 to 42, and 42 to 56 d of age. Using the new knowledge of maintenance contributions to the total requirement for these amino acids, the IAA ratio of SAA and threonine to lysine change very little as birds approached 56 d of age as shown in Table 2.1. They concluded that the IAA ratios of tryptophan and SAA had not changed at all during 6 to 10 weeks, while Thr:Lys ratio was slightly increased with age. D'Mello (2003) summarized the estimated IAA ratios based on Han and Baker (1994) and NRC (1994) as shown in Table 2.2. The IAA ratios suggested by Emmert and Baker (1997) were calculated on a true digestible basis based on several requirement studies with chicks at 2 to 3 weels of age fed diets using supplemental AA. The IAA

25 12 ratios suggested by NRC (1994) were total AA basis for chicks fed a corn-sbm diet from 0 to 3 weeks old using the lysine requirement corrected to 12 g per kg. Si et al., (2001) evaluated the relationship of dietary Lys and other essential amino acids (EAA) in diets for broilers based upon NRC (1994) recommendations. The diets were formulated by addition of crystalline amino acids (L-LysHCl, DL-Met, L-Arg, L-Thr and L-Trp) at 100%, 110%, 120% and 130%, which were provided to maintain minimum or minimize excess levels of amino acids across the diets as crude protein was decreased during the starter phase (0 to 21 d), grower phase (21 to 42 d), and finisher phase (42 to 56 d), based on the NRC (1994) recommendation (100%). Lysine levels were set at the NRC recommendation +0.1%, +0.2% and +0.3%. The result at 21 d showed that addition of 0.1% Lys to diets containing NRC recommended levels of Lys improved FCR but the higher levels gave no improvement. This was because when adding only lysine as a feed additive into the diet, at some point, the other essential AA (EAA) will become limited and produced a broiler with reduced body weight and feed conversion. This was confirmed by the significant improvement in FCR at 21, 42 and 56 d when the authors increased the level of other EAA, which provided balance AA profile for protein synthesis. They concluded that NRC (1994) levels of Lys and other EAA are adequate for optimum performance of male broilers processed at 56 d but may be less than adequate at younger ages (Si et al., 2001). Quentin et. al. (2005) studied the effect of early nutrition (during the first 7 d) on subsequent response of male broilers to increasing levels of EAA in their diet containing 85, 95, 105, and 115% of NRC from 21 to 42 d of age were evaluated under 2 light intensities (80 and 5 lux). They found that EAA levels were driven by the level of lysine between 21 to 42 d. A starter diet contained more amino acid nutrient density stimulated bodyweight gain and breast meat

26 13 development, but reduced the resistance of birds to stresses. It was noted that the EAA requirements of finishing broilers is likely in excess of 115% of the NRC (1994) recommendation and light intensity did not interact significantly with nutritional factors. Leclercq (1998) reported that threonine and valine do not have an observable effect on body composition. The difference in effects of lysine, threonine, and valine can be used in setting different amino acid requirement profiles depending upon the desired responses. They also suggested mathematical model including broken line and monomolecular models to calculate amino acid requirements that can also influence amino acid requirement profiles. Lysine Requirement for Growth, Feed Conversion and Carcass Composition A number of studies on AA requirements have been done; however variability of the discrepancy of the method used provided a variable outcome, which is caused by difference in age groups of the studied birds, sources of ingredients used in the diets, bird strains, rearing conditions, energy level in the diets and statistical evaluation methods (Han and Baker, 1994; Leclercq, 1998; Kerr et al., 1999; Sterling et al., 2003; Wijtten et al., 2004; Corzo et al., 2005; Corzo et al., 2006; Garcia et al., 2006; Dozier et al., 2008). The NRC (1994) requirement for broilers 3 to 6 weeks of age is 1.00% total lysine. Han and Baker (1994) studied Ross x Ross at the same age and found that the digestible lysine (dlys) requirement for maximum body weight gain was 0.85% for males and 0.78% for females. The requirement for optimum feed efficiency was 0.89% for males and 0.85% for female broilers. Garcia et al. (2006) studied Cobb500 birds from 21 to 38 d of age and found that the dlys requirement of males was higher than that of females, based on BWG (0.97 vs. 0.93%), but was similar for both sexes (0.96%) for gain per feed. Dozier et al. (2008) evaluated Ross x Ross 708 from 49 to 63 d of age and reported that the dlys requirements for growth performance were 0.87% for males and 0.81% for females. These

27 14 studies suggested that male broilers require higher dlys level compared with female broilers and modern strains require higher dlys level compared with the older ones. Kerr et al. (1999) demonstrated that feeding starting broilers diets containing higher levels of lysine (113% for BW and 121% for breast meat yield) than recommended by the NRC improves broiler performance. The Lys requirement for growth was less than required for feed conversion and breast meat yield according to Tesseraud et al. (1996). These authors examined the effects of lysine level on the growth of different muscles and protein renewal. They found that deficiency of lysine in the diet exclusively reduces Pectoralis major growth, which contains Type IIb muscle fibers. On the contrary, lysine level in the feed has less influence on the growth of Sartorius and Anterior latissimus dorsi, which contain both Type I and IIb muscle fibers at 2, 3, and 4 wk of age. Lysine deficiency increased turnover constants (protein synthesis and degradation, percent per day) of proteins in the three muscles. However, the Pectoralis major showed the highest increase in rate of renewal. As a result, Pectoralis major seems to be more sensitive to lysine intake than other muscles. Leclercq (1998) also reported that lysine requirement for maximum gain is lower than that for breast meat yield, which is lower than the requirement for FCR. Lysine Requirement for Phase Feeding The NRC (1994) recommended a feeding program of 3 phases but at considerably lower levels (0 to 21d (1.10% total Lys), 21 to 42d (1.00% total Lys), 42 to 56d (0.85% total Lys). The modern Ross x Ross 708 nutrition specification (Aviagen, 2007) recommends a feeding program consisting of 3 phases (0 to 10d (1.43% total Lys; 1.27% dlys), 11 to 24d (1.24% total Lys; 1.10% dlys), 25 to slaughter (1.09% total Lys; 0.97% dlys), in order to meet the requirements of this modern genetic strain. Many different phase feeding programs have been proposed by researchers. Pesti and Fletcher (1984) found that broilers fed low protein diets from 21 to 42 d

28 15 were able to compensate in growth performance during the finisher phase (371 versus 331 g gained) and feed utilization was more efficient (0.383 versus g of gain per g of feed intake). Wijtten et al. (2004) suggested that enhanced dietary ideal protein (IP) levels in the starter diet increased BW gain in the starter phase and later in grower phase. Besides, it was shown that a delay in BW gain due to suboptimal IP levels in the starter diet could only be partly recompensed in later phases of life. These results reveal the need for a reevaluation of IP levels used in practical starter diets. BW gain and FCR responses to increased IP levels in the grower and finisher diets were less pronounced when high compared with adequate IP levels were fed in the preceding phase. The authors did not detect a statistically significant difference in responses but was consistent between experiments and phases. Thus, they suggested that this phenomenon should not simply be neglected. Warren and Emmert (2000) compared 3 different feeding programs (Phase-feeding (PF), feeding Illinois ideal chick protein (IICP) and NRC (1994)). From 0 to 21d, there were no differences in BW gain, feed intake, feed efficiency, digestible AA intake, or gain per unit digestible AA intake among chicks fed PF, IICP, or NRC diets. From 40 to 61d, no differences were observed in BW gain or feed intake among chicks fed PF, IICP, or NRC diets. However, feed efficiency (gain per feed) of birds fed IICP diet was significantly lower than the other two programs. There were no differences in processing characteristics observed among the feeding programs evaluated. In this study, phase feeding reduced feed costs without giving up either growth performance or carcass yield. Kidd et al. (2004) reported on a BP feeding program, consisting of 4 phases from 0 to 49 d, using male and female Ross x Ross 508 broilers. They found that the highest nutrient density throughout the grow-out period resulted in maximizing breast meat yield. The highest total Lys levels used were 1.38, 1.19, 1.11, and 1.06 % for 0 to 14 d, 15 to 28 d, 29 to 35 d and 36 to 49 d, respectively. Feeding 23.4% CP

29 16 or 1.38% total Lys for 0 to 14 d and 21.8% CP or 1.19% total Lys for 15 to 28 d gave the maximum saleable white meat. Corzo et al. (2005) reported on a BP feeding program consisting of 4 phases from 0 to 56 d in 2 broiler strains. Digestible Lys levels fed from low to high ranged from 1.09 to 1.22% for 1 to 14 d, 0.97 to 1.07% for 14 to 28 d, 0.84 to 0.92% for 28 to 42 d and 0.79 to 0.81% for 42 to 56 d. They found that the higher dlys levels resulted in lower abdominal fat percentage, while also maximizing fillet and tender yield at 42 and 56 d of age. Kidd et al. (2005) reported on a BP feeding program consisting of 5 phases from 0 to 55 d using male and female Ross x Ross 708 broilers. Birds performed best when fed high amino acid diets during the first 3 phases. Total dlys levels fed were 1.38, 1.36, and 1.23 % for 1 to 5 d, 6 to 14 d and 15 to 35 d, respectively. They concluded that the Ross 708 strain requires high amino acid density from 1 to 35 d in order to maximize performance. Dozier et al. (2006) using BP, fed straight-run Ross 708 broilers with a 4 phase feeding program to 59 d (1 to 17 (High (H) dlys level= 1.36%), 18 to 35 d (H = 1.18%), 36 to 47d (H = 0.93%, Medium (M) = 0.84%, Low (L) = 0.75%), 48 to 59 d (H = 0.89%, M = 0.80%, L = 0.72%)). The H amino acid densities throughout the grow-out period provided the best cumulative feed conversion. Decreasing amino acid density (H (0.89%) to L (0.72%) for 48 to 59 d or H (0.93%) to L (0.75%) for 36 to 47 d and H (0.89%) to L (0.72%) for 48 to 59 d) limited growth of breast fillets, tenders and total white meat yields, when compared with the H amino acid density. Carcass and breast meat yield was not affected by feeding the M dlys density (0.84%) from 36 to 47 d and from feeding either the H (0.89%) to M (0.80%) or L (0.72%) dlys for 48 to 59 d compared with H amino acid density. The authors concluded that the H amino acid level was the most economical due to the improvement in cumulative feed conversion. In another experiment, Dozier et al. (2007) fed straight-run Ross 708 broiler diets up to 35 d. Digestible Lys levels were 1.21 and 1.06% for 1 to 17 d and 18 to 35

30 17 d, respectively. Birds were fed diets containing one of the three levels of dlys (H, 0.92%, M, 0.85% or L, 0.78%) from 36 to 47d followed by 2 levels of dlys H (0.88%) and L (0.75%) from 48 to 60 d. The H amino acid density program improved cumulative feed consumption by 4 points over the M and L density programs and increased breast meat yield by 0.6% compared with the L density program. Overall, these studies agree with the importance of feeding H nutrient densities in early stages to improve growth performance and carcass characteristic. Later in life, birds may not require the H density and reduced levels can be fed. Lemme et al. (2008) reported on a BP feeding program consisting of 3 phases from 0 to 49 d using male and female Ross x Ross 308 and 708 broilers. Total Lys levels from low to high ranged from 1.03 to 1.72% for 1 to 10 d, 0.87 to 1.44% for 11 to 28 d and 0.78 to 1.28% for 29 to 49 d. They found that the highest balanced protein levels optimized BW, processing performance and FCR. There was no difference in biological response to balance protein between the Ross strains tested. Using Response Surface Methodology to Determine Response Levels Response surface methodology (RSM) is a collection of statistical and mathematical techniques useful for developing, improving and optimizing processes (Myers and Montgomery, 2002). The RSM is applied in case there is several input variables influencing some performance measure or often is called response. The response equation can be shown as a surface when the effect of two quantitative factors is investigated in the experiment. Then a quadratic polynomial for two quantitative factors is used to estimate the experimental responses. A response surface can be used to investigate the response over a region of interesting factor levels and to examine the sensitivity of the response to the treatment factors. It also provides an optimum condition which maximizes the experimental response (Kuehl, 2000).

31 18 RSM assumes the mean of the response variable (y) is a function of quantitative factor levels (x 1,x 2,.x n ). The polynomial models are used as practical approximation to the true response function as it is usually unknown. Thus, the polynomial functions provide good approximations in relatively small regions of the quantitative factor levels (Kuehl, 2000). The most first-order or linear model and the second-order or quadratic model are commonly polynomial models used for response surface analysis. The first-order model for two factors or independent variables is (1) where y is the response variable;,, are regression coefficients for the intercept, slope of variable x 1 and slope of variable x 2, respectively, and is the error term. The second-order model for two factors is (2) In the case of the second-order model, the surface contains linear terms in x 1 and x 2, quadratic terms in and, and the cross-product or interaction term. Box and Wilson (1951) were the first to used RSM to determine the optimum operating conditions for an industrial process. The general approach was to use 2 n factorial treatment designs to identify factors that influence the process. They developed a design, called composite design, for fitting second order response surfaces. The total number of treatment combinations can be calculated by using 2 n +2n+1. The central composite design starts with an exploratory 2 n factorial to which a linear response surface is fitted. If the center of the experiment is close to a maximum, the response surface is curved. This also can be indicated by the lack of fit test. The 2n+1 treatment combination is added to create a central composite design (Cochran and Cox, 1957). A rotatable property was developed for central composite designs because the estimated

32 19 regression equation will not be constant over the entire experimental region (Kuehl, 2000). The rotatability requires the variance of estimated values at points equally distant from the center of the design to be constant. The central composite design can be made rotatable by setting the axial point values as α =. The two factor design has α = = = Graphical illustration of central composite rotatable design for two factors (dlys levels of grower and finisher phases) with the combination of 9 treatments is shown in Figure 2.1. The axial points are treatment number 1, 4, 6 and 9. The center point is treatment 5 and the 2 n factorials are treatment number 2, 3, 7 and 8. The RSM has been widely applied in many areas of research, including poultry nutrition. Roush (1979) described an application of RSM to research in poultry nutrition. Pesti (1983) reported protein and energy effects on performance and carcasses of broilers using RSM. Edwards et al. (1983) used RSM to study the role of calcium and phosphorus in the etiology of tibial dyschondroplasia in young chicks. Liem et al. (2009) used the RSM to optimize phytate phosphorus utilization by broiler chickens. Faria et al. (2008) used RSM to estimate body weight gain (BWG) and feed conversion (FC) using protein levels, environmental temperature and age from previous published studies between 1995 and The results showed that protein levels, environmental temperature and age have significant linear and quadratic effects on BWG and FC. However, there was no interaction between protein level and environmental temperature. Age and crude protein showed interaction for weight gain and feed conversion; whereas, interaction between age and temperature was detected only for weight gain. The authors performed economic analyses to determine maximum profit as a function of the variables that were included in the model. They concluded that the response surface models are effective to

33 20 predict the performance of broiler chickens and allow the elaboration of economic analyses to optimize profit (Faria et al., 2008). Modeling Approach: Growth, Production and Nutritional Response Functions A variety of growth functions including logistic, Gompertz, Richards, Michaelis-Menten, logistic nonlinear mixed effect, spine regression, exponential, and log linear models have been reported as models for consideration in poultry production (Lopez et al., 2000; Aggrey, 2002; Guevara, 2004; Eits et al., 2005a; Eits et al., 2005b; Zuidhof, 2005; Aggrey, 2009; Pesti et al., 2009; Darmani Kuhi et al., 2010, Table 2.3). These growth functions have been commonly used to estimate the optimum growth, nutrient responses and time required to produce birds of desired weights. Normally, the functional form of the Growth Function used for final weight or mature size is determined as a function of time or age of the bird. However, the models with two inputs have not been developed. Lopez et al. (2000) recommended a Michaelis-Menten-type equation to model growth in a wide variety of animals because it described diminishing returns behavior. Aggrey (2002) suggested using the logistic, Gompertz and Richards models over the spine linear regression model in chickens. Darmani Kuhi et al. (2010) demonstrated that the Richards model was the best fitting model and the Gompertz model fitted better than the logistic model in a data set with poultry. However, if initial BW was unknown, using the Richards equation can lead to optimization problems. Aggrey (2009) applied the logistic model with nonlinear mixed effects to model growth in Japanese quail. The model accounted for the variation between measurement units and heteroscedasticity. Zuidhof (2005) evaluated 8 dynamic nonlinear broiler carcass and carcass part yield models for predicting weights of carcass parts and studied their ability to predict carcass part weights with maximal accuracy and minimal bias. The nonlinear models consisted of four sigmoidal (S) models (Gompertz, modified Gompertz, Richards, and Lopez)

34 21 which were used to describe carcass part weight as a function of age. There were three diminishing returns (DR) models (Lopez, Mitscherlich, and log linear), and a log-linear proportional yield (PY) model, which were used to describe carcass part yield and weight, respectively. He studied parameters for nonlinear equations that could be used to predict carcass part weights of broilers at different processing ages up to 112 d. The study was conducted to develop strain-specific mathematical descriptions of carcass part yields of males and females from 6 commercial strain crosses in order to support decisions about the selection of strains and marketing weights. He found that the Richards model, with inflexion points that can vary depending on age, gender, strain and species, were better to predict carcass part weights as a function of age when compared with the Gompertz model, with fixed points of inflexion. This knowledge can be used in order to optimize broiler production economic and processing decisions. Hruby et al., (1994) showed the fit of the Gompertz, logistic and polynomial functions to body protein composition of male broilers as a function of age (week). The Gompertz function was the best fit according to residuals and coefficients of determination during the 19 week broiler study. Hancock et al. (1995) evaluated the differences between six commercial broiler genotypes and two sexes on potential growth, nutrient and environment requirements. The Gompertz model was used to describe the growth potential of broilers up to 11 weeks. They found that the rates of growth among strains and sexes were significantly different, while the rates of maturing were not significantly different either among strains or sexes. Emmans (1995) applied the Gompertz model to describe the growth curves of males and females of a single strain of broilers. The model was used to determine weight as a function of time. The requirement of protein or AA was stated as the sum of the separate requirements for maintenance

35 22 and the growth of protein, assuming the retentions of lipid, ash and water have no protein requirement. Besides those growth functions, several models have been applied in order to estimate production function of broiler production. Clark et al. (1982) applied a Reading model, which was described by Fisher et al. (1973) and Curnow (1973), to fit protein response curves of their experimental data. The model explained the dietary protein requirement as a function of broilers mean body weight and live weight gain. The model described that a group of broilers showed diminishing incremental responses as the dietary protein increased and reached a plateau when broilers were adequately supplied with the limiting nutrient. Zuidhof (2009) studied the nonlinear effect of diet on broiler production up to 56 d of age using a Cobb x Avian cross. A nonlinear model based on a Cobb-Douglas functional form and a stepwise procedure to estimate feed intake as a function of BW, ME, Lys, gain, and sex, provided reasonable accuracy of predicted ME. The author focused on the effect of dietary balanced protein (DBP) and dietary energy level on feed intake, growth curves, and yield dynamics. As feed consumption is a major cost in broiler production, estimating feed intake becomes an important factor in evaluating cost. The result of this study showed that at higher BW, broiler females grew at a slower rate and consumed less ME per unit of BW than males. The model predicted lower intake with higher dietary ME levels and showed that ME intake was slightly higher at low BW. Analysis of variance indicated that feed intake was maximized at recommended DBP levels. The author also found that sex, prestarter nutrition, and subsequent dietary energy and protein levels had significant nonlinear effects on broiler feed intake, growth rates, and yield dynamics. Many researchers (Kidd et al., 2004; Kidd et al., 2005; Lemme et al., 2008; Corzo et al., 2005; Dozier

36 23 et al., 2006; Dozier et al., 2007) agreed with the importance of feeding high nutrient density using the DBP concept to improve growth performance and carcass characteristics. Oviedo-Rondón et al. (2002) estimated amino acid requirements of broiler chicken using the Omnipro II growth model, which was introduced by Novus International, Inc.,. The model has been calibrated with growth data from breeder companies, genders, rearing environment (temperature, density, feeding schedule, etc.). The Omnipro II growth model is a semiempirical, deterministic and dynamic growth model that has been designed to optimize growth performance and nutritional changes. The model was used to estimate CP and AA levels, which were used to formulate the experimental diets. The diets formulated based on the model were compared with diets formulated based on NRC (1994) recommendations. The results demonstrated that BW of broilers fed 100% or 110% of the Omnipro II estimation level was similar to those broilers fed the NRC recommended level. The BW of broilers fed 90% of the Omnipro II estimation level was the lowest. In addition, FCR of broilers fed the Omnipro II estimation level was significantly better than those fed the NRC recommended level. They concluded that Omnipro II can be used to estimate nutrient level which performed as good as the NRC recommended level or better (Oviedo-Rondón et al., 2002). For nutrient dose-response type data, several models have also been proposed. The nutrient dose-response data is normally used to fit a function that explains the response to nutrient dose across the experimental treatment dose levels. In general, nutrient dose-response curves such as linear and quadratic broken-lines (Vedenov and Pesti, 2007), saturation kinetics (Phillips, 1982), Michaelis-Menten (Morgan et al., 1975), and exponential (Eits et al., 2005), are widely used in fitting nutrient-response curves with one independent variable. The function is then is used to estimate the nutritional requirement of the animal. Morgan et al. (1975) stated that

37 24 linear, semi-logarithmic, and quadratic models can only estimate within narrow ranges of nutrient intake. Thus, the authors suggested a saturation kinetic (SK) model, because many translocations and transformations (organisms absorb and utilize nutrients via sequences of them) of intermediary metabolism follow saturation kinetics. Phillips (1982) stated that the relationship between the response and nutrient intake of animals is not linear. Thus, a SK model that explained the efficiency of nutrient utilization decreases as the requirement is reached based on some rate limiting step was suggested. Phillips (1982) also stated that the SK and logistic fitted experimental data quite well, but both failed to converge to meaningful solutions when deficient curvature of the response and nutrient intake was examined. Robbins et al. (2006) explained that the linear broken-line model can be used to fit the nutrient dose-response data, which assumed the response to nutrient dose is linear. In general, the rate of change with nutrient dose-response level decreases as the nutrient dose-response level moves toward the broiler s maximum requirement. Thus, other models that include a non-linear component must be considered. Lamberson and Firman (2002) proposed a segmented regression or breakpoint analysis over a QP for estimating nutrient requirements. The authors found that SR provided the closest estimation to the true nutrient requirement in 73 of 100 replicates and that the 0.90 of quadratic maximum was closer in the remaining 27 replicates. The estimate obtained from 0.95 of quadratic maximum was never closer to the true requirement. The authors reported that the QP overestimated the true requirement, especially when the experimental diets were not equally distributed above and below the requirement. Eits et al. (2005a) used an exponential model to predict growth rates and feed conversion, along with carcass and breast meat yield of broilers, as a function of dietary balanced protein content. The exponential assumptions consisted of: 1) continuous reproduction or no

38 25 seasonality, 2) the organisms are identical or there is no age structure, and 3) the environment is constant in space and time or resources are unlimited (Sharov, 1997). Vedenov and Pesti (2010) applied the economic concept of profit maximization in a meta-analysis of experiments using the exponential model as reported by Sauer et al. (2008) in evaluating the relative efficiency of two potential sources of methionine for broilers. The exponential model estimated the average daily gain or gain per feed as a function of supplemental methionine sources. They found that when the cost of either the basal diet or live broilers increased, then the inclusion of the methionine source being used also increased. Pesti et al. (2009) compared various methods of estimating nutritional requirements from experimental data. Using data from a previously published lysine requirement paper, the brokenline quadratic (BLQ) model or the Saturation Kinetic (SK) model provided the best fit, based upon minimizing residual values. While it was noted the BLQ model best describes the concept of a nutritional requirement, at least in terms typically viewed by nutritionists, it can be difficult to fit. On the other hand, the SK model fit economic data quite well - particularly so when it came to applying the law of diminishing marginal productivity. In addition, it was reported that the SK model offered the concept of a most economical feeding level versus the concept of a break-point or requirement per se. Pack et al. (2003) described this concept in similar terms, suggesting the terminology of a derived recommendation versus a true requirement per se. Profit Maximization in Broiler Production The major aim of any production system is to maximize profit. In broiler production, feed cost represents about 60-70% of the total production cost. Thus, least cost feed formulation has been a major tool of broiler production companies to decide their profits for many years. The NRC (1994) stated that, It would be desirable to have mathematical models available that would

39 26 facilitate the selection of the most economical combinations of dietary concentrations of protein/amino acids (and other nutrients) and energy to achieve poultry production goals. Since then, the poultry industry has paid more attention to a mean for feed formulation to maximize profit rather than just minimize feed cost. The main concept is to estimate the highest profitability of broiler production at a given feed ingredient and broiler prices, feeding levels of nutrients, and bird performance. While the least cost feed formulation concept maintains the constant feeding level of nutrients, it always provides constant profit or constant marginal productivity. In the early 1940 s, the linear programming (LP) was generated. The technique and simplex method were developed between 1947 and After the development of computer systems in the 1950 s and early 1960 s, the LP software for solving general linear problems both in nutrition and economics was widely used. The LP was commonly applied in feed formulations for the industry and for research in the 1960 s. This method was adopted in education in the 1970 s, which increased the level of research of this area (Black and Hlubik, 1980). The main assumptions of LP are: 1) divisibility, 2) proportionality, 3) independence and 4) no alternative activities and deterministic or nonstochastic. Allison and Baird (1974) reported that LP can be used when the production function in estimating animal performance is limited. The main concept was to minimize feed ingredient costs which provided maximum performance regardless of feed ingredient prices. Since feed nutrients such as AA were set at a minimum constraint in order to provide maximum animal performance. Brown and Arscott (1960) studied the relationship between crude protein (CP) and metabolizable energy (ME) using corn-soybean meal based diets for male and female broilers (Delaware x New Hampshire). Body weight (BW) and feed consumption (FC) were used as production functions in calculating the optimum ration

40 27 specification (CP and ME). A quadratic model was fitted to the average data of 24 pens to generate predicted BW as a function of CP and ME consumed. The quadratic model, one of various algebraic functional forms, was selected because it was the best model to describe growth performance of the experimental data. The model provided highly significant regression coefficients according to t-test and high coefficient of determination. The predicted pounds of FC per bird were measured as a function of time, CP and ME content per pound of feed. The time variable was the feeding period which interacted with feed composition terms, CP and ME contents, because time is required for feed consumption regardless of feed composition. Thus, the FC model was used to estimate the pounds of feed consumed per bird at given CP and ME contents for a variety of feeding periods. The LP was applied to calculate least-cost feed mixtures for various CP and ME specifications. The margins over feed cost for these various points on the production surface were calculated. Then, the highest profit was selected at the ration specification. Pesti et al. (1986) proposed a quadratic response surface model of energy and protein to estimate growth responses. The quadratic programing was used to evaluate the optimum operation points in broiler production. The optimum operation points were defined as maximizing production or live body weight at a given fixed level of cost (feed cost per bird) and a set of inequality constraints on nutrients and feed ingredients. Economic theory was used to illustrate that the model was able to estimate the cost per pound of broiler meat produced within a specific time interval and broiler quality (measured by carcass fat). The authors applied the law of diminising returns, which means, as nutrient levels increased, the bird performance increased at a decreasing rate. Least cost feed formulation was studied using varying prices of corn and SBM, which affected CP and energy levels and minimized cost per pound of meat. Pesti et al.

41 28 (1986) concluded that the quadratic programming can be used to determine the most profitable CP and energy levels at different lecels of feed ingredient prices. Talpaz et al. (1988) conducted a dynamic model to select the economically optimal growth trajectory of broilers. They also computed the feeding schedule that was designed to satisfy the nutritional requirements along this trajectory. The authors used nonlinear programming to determine the optimum growth path for a broiler and considered the essential AA profile for maintenance and requirement as major variables to evaluate optimum growth. Dynamic least-cost rations for the potential growth rate, subject to the nutritional requirement, were determined using their methodology. The optimal trajectory was solved for by a reducedgradient nonlinear programming algorithm. The model estimated the daily optimal growth rates along with the corresponding requirements of total protein, amino acids, and energy in obtaining the optimal diets. The results of the study showed that, as feed ingredient prices increased, more feed restrictions reduced the corresponding optimal growth trajectory. Thus, a substantial increase in profits can be achieved by following their methodology. Gonzalez-Alcorta et al. (1994) used nonlinear programming techniques to determine the precise energy and protein levels that maximize profits. The BW and cumulative FI functions were generated as a quadratic function of energy and protein levels and age of the birds at time of processing. The authors found that as the price of corn increased, the energy level decreased and protein level increased. In addition, as the price of SBM increased, the protein level decreased while the energy level increased. Gonzalez-Alcorta et al. (1994) concluded that setting CP and energy levels at various input and output prices could increase company profits compared with fixed levels of CP and energy based on a nutritional guideline.

42 29 Costa et al. (2001) developed a two step profit maximization model based on minimizing feed cost while maximizing revenue in broiler production. The minimizing feed cost was determined at the optimal feed consume, feed cost, overall production cost, which included cost of the growing broiler, optimal length of time that the broilers stay in the house, and the interest rate paid for growing the broilers. The maximum revenue was estimated at various broiler prices, either whole carcass or cut-up part prices, and optimum live or processed BW of the birds. Profit maximization was estimated at the optimal protein levels, which provided minimization of feed cost while maximizing revenue. The authors compared peanut meal as an alternative protein source for SBM and concluded that using peanut meal could generate more profit for growing broilers when compared with SBM. Guevara (2004) proposed nonlinear programming over conventional linear programming to optimize broiler performance responses to energy density in feed formulations because the energy level does not need to be set. In this study, BW and FC were fit to quadratic equations in terms of energy density. The optimal ME level and bird performance were estimated using Excel solver nonlinear programming. The variation in corn, SBM, fish meal and broiler prices were used. Nonlinear programming showed that when the protein ingredient prices decreased, the energy density increased compared with the linear programming least cost formulation. The increased broiler price had a positive impact to BW and feed conversion and also increased energy density. The conclusion was that nonlinear programming can be used to define the optimal feed mix, which maximizes margin over feed cost. Eits et al. (2005b) focused on evaluating margin over feed costs (revenue minus feed costs). This return over investment concept as shown by Eits et al. (2005b) is the difference between increasing feed costs with increasing nutrient density and the decreasing incremental

43 30 technical performance response from that of just increasing nutrient density. The authors model indicated the effect of dietary balanced protein on revenue and feed costs, and determined the difference of the two margins over feed costs. The idea behind maximizing profitability through nutrition is to formulate to the optimal nutrient density, which maximizes profit. Sterling et al. (2005) applied a quadratic growth response equation to estimate BW gain as a function of dietary lysine and CP intake using a quadratic programming model workbook called WUPP EM. The program was used to estimate maximum profit feed formulation and provided a working tool to demonstrate the interdependencies of costs, technical response functions and meat prices. Based on the quadratic programming model, increasing the price of SBM decreased CP and Lys level, which gave maximum BW gain. Sterling et al. (2005) concluded that using the maximum profit model instead of the least cost model could generate more profit for broiler production. Cerrate and Waldroup (2009a) proposed a maximum profit feed formulation (MPFF, Maximum Profit Programming 3.0) as an alternative for least cost feed formulation (LCFF). Based on Ross male performance, BW and cut-up parts were used to determine changes in dietary nutrient density, which was determined as the level of metabolizable energy (ME). The models accounted for livability, temperature, processing cost, ingredient and broiler prices, starting and ending broiler prices. The relative BW and feed consumption (FC) were estimated using a quadratic function of ME at 49 d. The absolute BW was estimated from the final day of feeding (49 d) using a Gompertz equation while the absolute FC was predicted from the absolute BW using a quadratic equation. Carcass weight was calculated from the actual BW and yield (as a quadratic function of ME at 63 d). Cut-up parts were calculated by the multiplication of carcass weight and the constant of each cut-up part. The authors found that, as the price of

44 31 poultry fat increased, the ME level tended to decrease drastically, which reduced the usage of poultry fat and SBM in the diet and increased the usage of corn. The MPFF had higher profit compared with LCFF and it provided the best profit when poultry oil price increased by 150%. Cerrate and Waldroup (2009b) compared four different economic nutritional models for maximum profit feed formulation of broilers. As there are many methods of feed formulation, either consider the ratio of energy and some nutrients such as protein (Gonzalez-Alcorta et al., 1994), increase protein and AA levels while maintaining energy levels at a constant (Eits et al. (2005a,b) or increase energy levels while maintaining AA and CP (Dozier III et al., 2006). The different feed formulation methods listed certainly provided different growth performance. The four models, which represented different methods of feed formulation, were a constant calorienutrient ratio (C-E:P: Model 1), a variable calorie-protein ratio (V-E:Pg: Model 2), a constant protein-amino acid ratio (DBP: Model 3) and a variable calorie-protein ratio for the finisher period (V-E:Pd: Model 4). Using relative performance, economic nutrient requirements, and profitability to compare the four models, the authors found that changing feed ingredient prices impacted the energy and protein content based on the four models. For example, as corn or broiler price increased, the energy and protein contents of model 1-3 increased, except for the energy content of model 2, which decreased. The opposite was observed when SBM or poultry oil prices increased. The authors concluded that model 1 is the best method to be used in feed formulation because it provided the maximum performance and profitability. Model 4 provided the best profit as well but with a narrow range of price changes and inconsistency of growth responses. Model 3 can be used at low corn or high SBM prices. The data set of model 1 came from the past ten years that contained new strain of birds and response of increasing all the nutrients. Thus, the predicted BW from model 1 was higher than model 2 and3, which resulted in

45 32 the most profitable model. The authors commented that modern broilers with rapid growth rates do not adjust feed consumption to meet a fixed energy need. This caused the birds to eat more energy as the energy content increased, especially when AA and CP increased along with the energy. However, when AA and CP were kept constant, the increase in energy content reduced feed consumption to normalize the energy intake.

46 33 Figure 2.1 Graphical illustration of central composite rotatable design for two factors with the combination of 9 treatments X dlys levels of finisher phase, ξ 2 (%) dlys levels of grower phase, ξ 1 (%) X 1

47 34 Table 2.1 University of Illinois ideal ratios 1 for selected amino acids at three growth periods (Emmert and Baker, 1997). Amino Acids Ideal Amino Acid Ratio 0 to 21 d 21 to 42 d 42 to 56 d Lysine Methionine + Cystine Threonine Valine Arginine The ideal ratios were calculated based on projected maintenance requirements and maintenance contributions to the total requirement.

48 35 Table 2.2 Ideal amino acid ratios (%) of essential amino acids for broiler chicks (D'Mello, 2003). Amino acid Baker (1997) NRC (1994) Lysine SAA Threonine Valine Isoleucine Leucine Tryptophan Arginine Histidine Phe+Tyr Sulphur amino acids.

49 36 Table 2.3 Example of Growth and Production Functions used to determine growth or responses in animal. Functions Equation 1 References Logistic Darmani Kuhi et al. (2010) Gompertz Richards Emmans (1981) Vedenov and Pesti (2007) Log linear Zuidhof (2005) Satuation Kinetics Vedenov and Pesti (2010) Exponential Eits et al. (2005); Vedenov and Pesti (2010) Linear broken line Vedenov and Pesti (2010) Broken line with ascending quadratic line Vedenov and Pesti (2010) Quadratic, where a = intercept, b = coefficient for the linear term, c = coefficient for the quadratic term. Guevara (2004) Cobb-Douglas, where K and L are two inputs Zuidhof and Rogers (2009) 1 W o is initial weight, W f is final weight, the parameters a,b,c are positive entities, n>-1 (Darmani Kuhi et al., 2010); x is nutrient level, A is the theoretical maximum, B is the intercept, C is the nutrient rate constant and D is the kinetic order when x= 0 (Vedenov and Pesti, 2010).

50 37 REFERENCES Aggrey, S Comparison of three nonlinear and spline regression models for describing chicken growth curves. Poultry Science, 81(12): Aggrey, S Logistic nonlinear mixed effects model for estimating growth parameters. Poultry Science, 88(2): Agricultural Research Council The nutrient Requirements of Pigs. Commonwealth Agricultural Bureaux, Farnham Royal, UK. Allison, J.R. and D. Baird, Least-cost livestock production rations. Southern journal of agricultural economics, 6(2): Aviagen Ross 708 broiler performance objectives. Aviagen, Newbridge, Scotland. Baker, D Nutrient bioavailability in feedstuffs: Methodology for determining amino acid and b-vitamin availability in cereal grains and soybean meal. pp: Bartov, I. and I. Plavnik Moderate excess of dietary protein increases breast meat yield of broiler chicks. Poultry Science, 77(5): Black, J.R. and J. Hlubik, Basics of coputerized linear programs for ration formulation. Journal of Dairy Science, 63(8): Box, G.E.P, and K.B. Wilson On the experimental attainment of optimum condition (with discussion). J. R. Statist. Soc. B13: Brown, W. G., and G. Arscott Animal production functions and optimum ration specifications. Journal of Farm Economics 42: Cerrate, S. and P. Waldroup. 2009a. Maximum profit feed formulation of broilers: 1. Development of a feeding program model to predict profitability using non linear programming. International Journal of Poultry Science, 8(3):

51 38 Cerrate, S. and P. Waldroup. 2009b. Maximum profit feed formulation of broilers: 2. Comparison among different nutritional models. International Journal of Poultry Science, 8(3): Clark, F. A., R. M. Gous, and T. R. Morris Response of broiler chickens to well-balanced protein mixtures. British Poultry Science 23: Cochran, W.G. and G.M. Cox Some methods for the study of response surfaces. Experimental Designs, 2nd ed. Wiley, New York, pp Corzo, A., M. T. Kidd, D. J. Burnham, E. R. Miller, S. L. Branton, and R. Gonzalez-Esquerra Dietary Amino Acid Density Effects on Growth and Carcass of Broilers Differing in Strain Cross and Sex The Journal of Applied Poultry Research, 14:1-9. Corzo, A., W. Dozier 3rd and M. Kidd Dietary lysine needs of late-developing heavy broilers. Poultry Science, 85(3): Corzo, A Maximizing production efficiency by a better understanding of the 4th limiting amino acid in broiler formulation. In: 6th Mid-Atlantic Nutrition Conference. Timonium, Maryland: pp: Costa, E.F., B.R. Miller, J.E. Houston and G.M. Pesti Production and profitability responses to alternative protein sources and levels in broiler rations. Journal of Agricultural and Applied Economics, 33(3): Cravener, T. and W. Roush Improving neural network prediction of amino acid levels in feed ingredients. Poultry Science, 78(7): Cravener, T. and W. Roush Prediction of amino acid profiles in feed ingredients: Genetic algorithm calibration of artificial neural networks. Animal Feed Science and Technology, 90(3-4):

52 39 Curnow, R. N A smooth population response curve based on an abrupt threshold and plateau model for individuals. Biometrics 29:1-10. D'Mello, J Responses of growing poultry to amino acids. D'Mello, J Amino acids in animal nutrition. CABI Publishing, Cambridge, MA. Darmani Kuhi, H., T. Porter, S. Lopez, E. Kebreab, A. Strathe, A. Dumas, J. Dijkstra and J. France A review of mathematical functions for the analysis of growth in poultry. World's Poultry Science Journal, 66(02): Dozier, W. A., III, M. T. Kidd, A. Corzo, J. Anderson, and S. L. Branton Growth, meat yield, and economic responses of Ross Ross 708 broilers provided diets varying in amino acid density from 36 to 59 days of age. The Journal of Applied Poultry Research, 15: Dozier, W. A. III, M. T. Kidd, A. Corzo, J. Anderson, and S. L. Branton Dietary Amino Acid Responses of Mixed-Sex Broiler Chickens from Two to Four Kilograms. The Journal of Applied Poultry Research, 16: Dozier, W.A., A. Corzo, M.T. Kidd and M.W. Schilling Dietary digestible lysine requirements of male and female broilers from forty-nine to sixty-three days of age. Poultry Science, 87(7): Edwards, H. M., Jr., and J. R. Veltmann, Jr The Role of Calcium and Phosphorus in the Etiology of Tibial Dyschondroplasia in Young Chicks. Journal of Nutrition 113(8): Eits, R., R. Kwakkel, M. Verstegen and L. Hartog, 2005a. Dietary balanced protein in broiler chickens. 1. A flexible and practical tool to predict dose response curves. British Poultry Science, 46(3):

53 40 Eits, R., G. Giesen, R. Kwakkel, M. Verstegen and L. Den Hartog, 2005b. Dietary balanced protein in broiler chickens. 2. An economic analysis. British Poultry Science, 46(3): Emmans, G Problems in modeling the growth of poultry. World's Poultry Science Journal 51: Emmert, J. and D. Baker Use of the ideal protein concept for precision formulation of amino acid levels in broiler diets. The Journal of Applied Poultry Research, 6(4): Faria, F.D., P. Rosa, K. Torres, M. Macari and R. Furlan Response surface models to predict broiler performance and applications for economic analysis. Brazilian Journal of Poultry Science, 10(2): Fisher, C., T. Morris, and R. Jennings A model for the description and prediction of the response of laying hens to amino acid intake. British Poultry Science 14: Fraps, G Relation of the protein, fat, and energy of the ration to the composition of chickens. Poultry Science, 22: Garcia, A., A. Batal and D. Baker Variations in the digestible lysine requirement of broiler chickens due to sex, performance parameters, rearing environment, and processing yield characteristics. Poultry Science, 85(3): Garrett, H.R., and M.C. Grisham Biochemistry. Thomson Learning, CA. Gonzalez-Alcorta, M.J., J.H. Dorfman and G.M. Pesti Maximizing profit in broiler production as prices change: A simple approximation with practical value. Agribusiness, 10(5):

54 41 Guevara, V Use of nonlinear programming to optimize performance response to energy density in broiler feed formulation. Poultry Science, 83(2): Han, Y. and D. Baker Digestible lysine requirement of male and female broiler chicks during the period three to six weeks posthatching. Poultry Science, 73(11): Hancock, C., G. Bradford, G. Emmans, and R. Gous The evaluation of the growth parameters of six strains of commercial broiler chickens. British Poultry Science 36: Hill, F. and L. Dansky Studies of the protein requirement of chicks and its relation to dietary energy level. Poultry Science, 29: Hruby, M., M. L. Hamre, and C. N. Coon Growth modelling as a tool for predicting amino acid requirements of broilers. The Journal of Applied Poultry Research 3: Jones, D.B Factors for converting percentages of nitrogen in foods and feeds into percentages of proteins. United States Department of Agriculture, 183: Kerr, B., M. Kidd, K. Halpin, G. McWard and C. Quarles Lysine level increases live performance and breast yield in male broilers. The Journal of Applied Poultry Research, 8(4): Kidd, M.T., C. D. McDaniel, S. L. Branton, E. R. Miller, B. B. Boren, and B. I. Fancher Increasing Amino Acid Density Improves Live Performance and Carcass Yields of Commercial Broilers. The Journal of Apply Poultry Research, 13: Kidd, M.T., A. Corzo, D. Hoehler, E.R. Miler, and W.A. Dozier Broiler responsiveness (Ross x 708) to diets varying in amino acid density. Poultry Science, 84: Kidd, M.T., and L. Hackenhaar Dietary threonine for broilers: dietary interactions and feed additive supplement use. CAB Reviews. 1: No. 005.

55 42 Kuehl R.O Design of Experiments: Statistical Principles of Research Design and Analysis. 2 nd ed. Brooks/Cole Publishing Company, Pacific Grove, CA. Leclercq, B Lysine: Specific effects of lysine on broiler production: Comparison with threonine and valine. Poultry Science, 77(1): Lemme, A., C. Kemp, E. Fleming, and C. Fisher Impact of increasing levels of balanced protein on biological and economic parameters of two different commercial broiler lines: Ross 308 and Ross 708. World Poultry Congress, Brisbane, Australia. Liem A., G. M. Pesti, A. Atencio, and H. M. Edwards, Jr Experimental approach to optimize phytate phosphorus utilization by broiler chickens by addition of supplements. Poultry Science, 88: Lopez, S., J. France, W. Gerrits, M. Dhanoa, D. Humphries and J. Dijkstra A generalized michaelis-menten equation for the analysis of growth. Journal of Animal Science, 78(7): Mariotti, F., D. Tome, and P.P. Mirand Converting nitrogen into protein-beyond 6.25 and Jones factors. Food Science and Nutrition, 48: Morgan, P. H., L. P. Mercer, and N. W. Flodin General model for nutritional responses of higher organisms. Proceedings of the National Academy of Sciences of the United States of America 72: Morris, T., R. Gous and C. Fisher An analysis of the hypothesis that amino acid requirements for chicks should be stated as a proportion of dietary protein. World's Poultry Science Journal, 55(01): 7-22.

56 43 Mosse, J Nitrogen to protein conversion factor for ten cereals and six legumes or oilseeds. A reappraisal of its definition and determination. Variation according to species and to seed protein content. Journal of Agricultural Food Chemistry, 38: Myers RH, and Montgomery DC Response Surface Methodology: Process and Product Optimization Using Designed Experiments. 2 nd ed. John Wiley&Sons, Inc., NY. NRC Nutrient Requirements of Poultry. 9th Rev. ed. Natl. Acad. Press, Washington DC. Oviedo-Rondón, E., A. Fritts and P. Waldroup Accuracy of omnipro ll estimations for amino acid requirements of broilers1. International Journal of Poultry Science, 1(5): Pack, M., D. Hoehler and A. Lemme Economic assessment of amino acid responses in growing poultry. Amino Acids in Animal Nutrition, 2nd ed. JP F. D Mello, ed. CABI Publishers, Wallingford, UK: Pesti G.M Protein and Energy Effects on Performance and Carcasses of Broilers. Page in Proc. of the 1983 GA. Nutrition Conference for the feed industry, GA. Pesti, G. M., and D. L. Fletcher The response of male broiler chickens to diets with various protein contents during the growing and finisher phases. British Poultry Science, 25: Pesti, G., R. Arraes and B. Miller Use of the quadratic growth response to dietary protein and energy concentrations in least-cost feed formulation. Poultry science 65: Pesti, G., D. Vedenov, J. Cason and L. Billard A comparison of methods to estimate nutritional requirements from experimental data. British Poultry Science, 50(1): Phillips, R. D Modification of the saturation kinetics model to produce a more versatile protein quality assay. The Journal of Nutrition 112:

57 44 Quentin, M., I. Bouvarel and M. Picard Effects of the starter diet, light intensity, and essential amino acids level on growth and carcass composition of broilers. The Journal of Applied Poultry Research, 14(1): Robbins, K., A. Saxton, and L. Southern Estimation of nutrient requirements using broken-line regression analysis. Journal of Animal Science 84:E155-E165. Roush, W.B., R.G. Peterson, G.H. Arscott An Application of Response Surface Methodology to Research in Poultry Nutrition. Poultry Science, 58: Roush, W. and T. Cravener, Artificial neural network prediction of amino acid levels in feed ingredients. Poultry Science, 76(5): Salo-Vaananen, P.P., and P.E. Koivistoinen Determination of protein in feeds: comparison of net protein and crude protein (Nx6.25) values. Food Chemistry, 57: Sauer, N., K. Emrich, H. Piepho, A. Lemme, M. Redshaw and R. Mosenthin Metaanalysis of the relative efficiency of methionine-hydroxy-analogue-free-acid compared with dl-methionine in broilers using nonlinear mixed models. Poultry Science 87(10): Sharov, A Exponential Model. Si, J., C. Fritts, D. Burnham and P. Waldroup Relationship of dietary lysine level to the concentration of all essential amino acids in broiler diets. Poultry Science, 80(10): Sosulski, F.W., and G.I. Imafidon Amino acid composition and nitrogen-to-protein conversion factors for animal and plant foods. J. Agric. Food Chem., 38:

58 45 Sterling, K., G. Pesti and R. Bakalli Performance of broiler chicks fed various levels of dietary lysine and crude protein. Poult Science, 82(12): Sterling, K., D. Vedenov, G. Pesti and R. Bakalli Economically optimal dietary crude protein and lysine levels for starting broiler chicks. Poultry Science, 84(1): Tesseraud, S., N. Maaa, R. Peresson and A. Chagneau Relative responses of protein turnover in three different skeletal muscles to dietary lysine deficiency in chicks. British Poultry Science, 37(3): Vedenov, D. and G. Pesti An economic analysis of a methionine source comparison response model. Poultry Science, 89(11): Wijtten, P., A. Lemme and D. Langhout Effects of different dietary ideal protein levels on male and female broiler performance during different phases of life: Single phase effects, carryover effects, and interactions between phases. Poultry Science, 83(12): Warren, W.A., and J.L. Emmert Efficacy of Phase-Feeding in Supporting Growth Performance of Broiler Chicks during the Starter and Finisher Phases. Poultry Science, 79: Zellner, A., J. Kmenta and J. Dreze Specification and estimation of cobb-douglas production function models. Econometrica: Journal of the Econometric Society, 34(4): Zuidhof, M Mathematical characterization of broiler carcass yield dynamics. Poultry Science, 84(7): Zuidhof, M.J Nonlinear effects of diet on broiler production. In: Proceedings of the 2009 Arkansas Nutrition Conference. Rogers, Arkansas.

59 46 CHAPTER 3 THE DISTRIBUTION OF CRUDE PROTEIN AND AMINO ACID CONTENT IN MAIZE GRAIN AND SOYBEAN MEAL 1 1 N. Sriperm, G. M. Pesti, and P. B. Tillman, Animal Feed Science and Technology 159: Reprinted here with permission of publisher.

60 47 ABSTRACT This study examines the assumptions of normal distributions for crude protein (CP) and amino acid (AA) contents in feedstuffs. Data for maize grain and soybean meal (SBM) were collected from the Ajinomoto Heartland LLC laboratory analysis database between 2002 and Tests of normality for CP and selected AA were performed for both feedstuffs by using graphical methods (histogram and normal quantile quantile plot) and numerical methods (skewness and Shapiro-Wilk procedure (W)). Relationships between CP and AA were also computed using linear and quadratic regression and W were used to test for normality of the internally Studentized residuals of the regression model. Results indicated that methionine (Met) and arginine (Arg) were not normally distributed in maize grain (P < 0.05). In addition, CP, lysine (Lys), threonine (Thr), Met, isoleucine (Ile) and tryptophan (Trp) were not normally distributed in SBM (P < 0.05). There were linear relationships between CP and most of the AA in maize grain and SBM, except for the relationship between CP and Thr, and CP and Ile in maize grain and CP and total sulfur amino acids (TSAA), and CP and Arg in SBM which were found to be non-linear (significant quadratic terms at P < 0.05). The results indicate the need for normality testing of AA levels in feed ingredients prior to generating prediction equations for AA levels from CP. Assuming a normal distribution of CP and AA in critical feed ingredients may lead to an over or under estimated nutrient content in feed formulation. Even though the regression residuals are normally distributed in maize grain and SBM, other models beside linear and quadratic regression could be applied in order to accurately predict AA contents based on CP.

61 48 INTRODUCTION Maize grain and soybean meal (SBM) are the major feed components of poultry feed in the United States, often representing greater than 85% of a diet. Habitually, most nutritionists formulate diets for poultry based on average reference values of maize grain and SBM from the National Research Council (National Research Council, 1994) or some other ingredient composition tables; nevertheless, variability in all ingredients has been observed. Wet-chemistry laboratory analysis is the most accurate method to determine the nutritional content of ingredients; but the analysis cost and promptness of results are issues of concern. Therefore, many nutritionists are attempting to predict the essential amino acid (AA) contents of their ingredients, as accurately as possible, using least squares regression or near-infrared spectroscopy. When least squares regression is used to estimate AA contents, coefficients in prediction equations from historical laboratory samples relating AA composition to the CP (g. per 100 g. of nitrogen x 6.25) content of an ingredient, are usually based on linear models (National Research Council, 1994). One of the model assumptions is normality in the distribution of the error terms (or residuals). If this assumption is incorrect, then the estimation by a linear model may be inefficient or misleading. Applying an appropriate transformation to the response variable may improve the model accuracy (Box and Cox, 1964). Cravener and Roush (1999) suggested that linear regression of maize grain and soybean meal has less accurate predictions among their studied models. It provided the lowest R 2 compared with the other models and a low R 2 value indicates a low level of precise prediction. The starting point for detecting non-normality in datasets is to view the data graphically using either a histogram, stem-and leaf plot, normal probability plot or normal quantile - quantile plot (Q-Q plot) (Mendenhall and Sincich, 2003). Secondarily, one could use numerical

62 49 calculations to evaluate summary statistics such as skewness, kurtosis or use one of several statistical tests for normality (D Agostino and Stephans, 1986 and Park, 2008). A statistical analysis program such as that from the SAS Institute (SAS, 2004) offers four statistical tests for determining if a dataset or a models error terms are normally distributed; Shapiro-Wilk, Kolmogorov-Smirnov, Cramer-von Mises, and Anderson-Darling. In this study, we primarily considered the Shapiro-Wilk test (W) because Shapiro and Wilk (1965), Royston (1982, 1983) and Park (2008) agree that this is the most reliable test for non-normality. Moreover, Royston (1982) extended the test for the sample size range between 7 and 2000 from the original W statistic (Shapiro and Wilk, 1965) which had a small sample size range between 3 and 50. Therefore, the W statistic is appropriate for this dataset as the sample size of this study was less than 2,000. The W statistic, which is positive and always less than or equal to one, was determined by the ratio between the best estimator of the variance and the usual corrected sum of squares estimator of the variance (SAS, 2000). The null hypothesis for this test is that the probability distribution of the response, y, is normal. Hence a computed value of W less than W α suggests the hypothesis of normality at a specified α level, should be rejected (Shapiro, 1986). The purpose of this work were 1) to determine if using the mean value of CP and AA (Lys, Thr, Met, TSAA, Val, Ile, Arg, and Trp) in maize grain and SBM is adequate for feed formulation and 2) to test the assumption of normality of the relationships between CP and AA in those two major feed ingredients. This paper will also provide a guideline for testing whether an individual AA or CP distribution in a selected feed ingredient is normally distributed.

63 50 MATHERIALS AND METHODS Amino acid and CP contents, for maize grain and SBM samples submitted between 2002 and 2008, were determined at Ajinomoto Heartland LLC s AA laboratory. The ingredient samples tested, while having an unidentified regional origin, were collected from across the United States. The AA and CP contents were adjusted to a g. per 88 g. of dry matter based upon dry matter determination in a forced air oven at 135 C for 2 hours (AOAC, 2000; ). The stem and leaf procedure was used to analyze data for outliers by identifying extreme values. The digits of the individual values were ordered in a histogram and the extreme values, listed individually outside the scale of the histogram, were removed from the data (Statistix9, 2008). Once the determined outliers were removed, the number of analyzed samples of each component (CP and each AA) was between 290 to 340 for maize grain and 489 to 571 for SBM (Table 1 and 2, respectively). The maize grain and SBM samples were analyzed for CP (Assoc. of Official Analytical Chemists (Method , nitrogen divides by 0.16, AOAC, 2000)) and AA (AOAC, 2000; E [a, b]) composition. The AA used HPLC (HITACHI 8900) with ninhydrin in a post-column derivatization. Regarding AA recovery, internal reference samples are run with each batch of hydrolysates in order to monitor the degree of hydrolysis. In addition, nor-leucine is used as an internal standard. A known sample is used to monitor hydrolysis efficiency and HPLC conditions. Sample results are as-is, and represent the average of two analyses. They were all rerun in the event the known sample was not in line with established specifications or did not agree with each other within 3 %, samples were rerun. Performic acid oxidation (AOAC, 2000; E [b]) was conducted prior to acid hydrolysis for the determination of TSAA (Met plus cystine (Cys)); whereas, all other amino acids (Lys, Thr, Val, Ile and Arg) were determined following acid hydrolysis. Tryptophan was

64 51 hydrolyzed under alkaline conditions using barium hydroxide and subsequently acidified using chlorhydric acid. Tryptophan from the extracts was separated by reverse phase high performance liquid chromatography and determined by fluorometric detection. The AA considered are generally set as minimums in feed formulation, especially under the ideal AA or balanced protein concept. Data distributions were analyzed using the PROC UNIVARIATE of SAS 2004 procedure to provide the descriptive statistics, Shapiro-Wilk critical value, histogram and Q-Q plot. PROC UNIVARIATE was also used to investigate the Shapiro-Wilk critical value to test the assumption of normality of the relationships between CP and AA. The bimodality coefficient was computed using the SAS MODECLUS procedure, based upon a bimodal distribution consisting of a mixture of two normal distributions (SAS, 2004). PROC GLM of SAS (2004) was used to investigate simple linear and quadratic regressions of CP and AA in maize grain and soybean meal samples by considering the determined R 2 and P value at the 5 % significance level. The cubic functions were fitted to the data, but they were not significant. Each component was fit with cubic, quadratic and linear functions. Only significant coefficients were reported. RESULTS Maize grain Graphical depictions of the histogram, normal probability plot, and Q-Q plot help to visualize if non-normality might exist. The classic bell shaped distribution of a histogram indicates normally distributed data which has a mean of zero and a variance of one. A straight linear line graphical representation of a probability plot implies normally distributed data. If the Q-Q plot which the data describes is not critically departed from the fitted line, indications of a normal distribution of the data exist (Park, 2008). Such a plot is also helpful in detecting skewness of the data; as indicated by either a shift to the left or to the right of the linear line.

65 52 Figures 1 and 2 show examples of a histogram and Q-Q plot, respectively, for Lys (which represent a normal distribution) and Arg (which represent a non-normal distribution) in maize grain. Even though the histogram does not show obvious dramatic differences between normally and non-normally distributed variables for this data, it can be a method of observing whether the data appears normal (Mendenhall and Sincich, 2003). The Q-Q plot which shows both ends of the normality plot to bend above the hypothetical straight line, indicates the distribution may be skewed to the right (such as for Lys and Arg in Figure 2). Conversely, if the normal plot shows tendencies of being below the hypothetical straight line then the distribution might be skewed to the left. The decision of normality should ultimately be based upon numerical methods (W, skewness, etc.), particularly when the graphical methods are difficult to interpret. The W test, based upon the null hypothesis of the data being normally distributed, was rejected at the 5% significance level (P < 0.05) for Met, and Arg in maize grain, indicating that these two amino acids were not normally distributed in maize grain. Crude protein, Lys, Thr, TSAA, Val, Ile and Trp were normally distributed at P > 0.05 (Table 1) in maize grain although TSAA was borderline non-normal. At a 1% significant level (P < 0.01), CP Val and Trp in maize grain would have been classified as non-normally distributed with Ile being borderline normal. The skewness of Lys, Thr, TSAA, Val, and Arg are greater than zero (0.030, 0.176, 0.200, and 0.292, respectively). The positive skewness (skewed to the right) means the distribution has observations that are clustered together below the mean, but have a long tail above the mean. In contrast, the skewness of CP, Met, Ile and Trp are less than zero (-0.196, , and , respectively). The negative skewness (skewed to the left) means the distribution has observations that are clustered together above the mean, but have a long tail below the mean (D Agostino, R.B et. al., 1990). According to Oehlert (2000), skewness measures asymmetry of

66 53 the distribution and is equal to zero for a normal distribution. Therefore, the closer the skewness is to zero, the less the distribution of that variable deviates from a normal distribution, which is true for Lys and Ile. Crude protein, TSAA, and Val have large skewness values even though they are normally distributed, although being barely significant at P < 0.05 (Table 1). Soybean Meal Figures 3 and 4 show examples of a histogram and Q-Q plot, respectively, for Val in SBM, which represents a normal distribution and for Lys which represents a non-normal distribution. The bell shaped histogram of Val in SBM indicates normal distribution of the data. In contrast, the histogram of Lys (Figure 3) in SBM indicates bimodal distribution even though its bimodality coefficient is only Technically, a bimodality coefficient must be greater than to demonstrate a true bimodal distribution and it would equal 1.00 for a pure bimodal distribution (SAS, 2004). The Q-Q plot for Lys in SBM (Figure 4) shows both ends of the normality plot bending above the hypothetical straight line meaning the distribution is skewed to the right. On the contrary, the Q-Q plot for Val in SBM (Figure 4) shows one end of the normality plot bending below, while the other end barely above, the hypothetical straight line implying the distribution is skewed to the left. The W test, based upon the null hypothesis of the data being normally distributed, was rejected at the 5% significance level (P < 0.05) for CP, Lys, Thr, Met, Ile and Trp. In contrast, TSAA, Val and Arg were normally distributed at P > 0.05 (Table 2) in SBM. The skewness of Lys, Met, TSAA, Ile and Trp were greater than zero (0.543, 0.110, 0.023, and 0.382, respectively) or skewed to the right. In contrast, the skewness of CP, Thr, Val, and Arg are less than zero (-0.121, , and , respectively) or skewed to the left. The skewness

67 54 of TSAA, Val and Arg were closer to zero compared with the other examined AA or CP which indicates that skewness can also be an indicator for non-normality. Relationships between crude protein and amino acids Linear relationships between CP and AA were highly significant at P < in maize grain (Table 3). All relationships had positive linear slopes, showing that amino acid levels increased with increasing CP, except for the relationship between CP and Thr, and CP and Ile which had significant quadratic terms (P < 0.05). As with maize grain, linear relationships between CP and AA also existed in SBM and were highly significant at P < All relationships had positive linear slopes, showing that amino acid levels increased at a constant rate with increasing CP, except for the relationship between CP and TSAA, and CP and Arg which had significant quadratic terms (P < 0.05). The W test, based upon the null hypothesis of the residuals of the model being normally distributed, for all relationships between CP and AA in maize grain and SBM was normally distributed at the 5% significance level (P > 0.05). The failure of rejecting the null hypothesis indicates the assumption of normality of the residuals for the model exists. DISCUSSION The results from this data set indicate the need for normality testing of amino acid levels in feed ingredients, to assure diets meet the targeted nutrient levels at expected frequencies. Assuming a normal distribution of CP and AA in critical feed ingredients may lead to an over or under estimation of nutrient contents for feed formulation, particularly for those nutrients which are not normally distributed. This analysis indicates that CP and most of the analyzed AA are not normally distributed in maize grain and SBM. Least-cost feed formulation typically uses mean ingredient composition values and it is assumed that the resulting batches of mixed feed will

68 55 contain the specified targeted minimum levels, for those nutrients which just meet this constraint, half the time (half above and half below the mean). Because most distributions are skewed slightly (either to the high or to the low side), resulting batches of mixed feed will only contain targeted minimum or maximum values more or less 50% of the time. For non-normal distributions, the median and not the mean is the 50% quantile; whereas the mean and the median are identical for normally distributed data. However, the difference between the mean and the median for maize grain as observed from this study were very small for all AA (less than 0.004) but not for CP (0.03). The difference between the mean and the median for SBM were small for CP and most AA (less than 0.004), except Ile (0.012 and Lys (0.03). From a practical perspective, using the mean or median in feed formulation will make little difference, except for Lys and perhaps Ile in SBM. The indicated bimodal distribution of Lys in SBM could be due to several causes. There may be two sets of processing conditions of the soybeans during the production of SBM: perhaps one which under-processed the soybeans and perhaps one which over-processed the soybeans leading to the formation of lysine reaction products (Maillard / browning reaction, which are not reversed on acid hydrolysis in amino acid measurements). Another possibility is the presence of two distinct cultivars (Northern or Southern region), or families of cultivars, containing different protein profiles with different amounts of AA. Unfortunately, this additional processing and regional information was unavailable for the samples used in this analysis. If feeds are formulated using the mean value versus the median value for Lys in SBM, the resulting batches of feed will be under formulated. Changing the Lys value from (mean) to (median) in the SBM matrix of a broiler grower diet increased L-lysine HCl supplementation by 4.3% to meet the targeted minimum total Lys level (1.1%) of that diet. Of course, the impact of changing

69 56 the Lys level in SBM will alter commercial L-lysine HCl, and other commercial amino acids, use to a greater or lesser extent based upon the targeted Lys levels, nutrient matrix values of other ingredients and pricing. According to this data set, the mean Lys value is at the 60.5% quantile as opposed to the median Lys value which is at the 50% quantile. Amino acid distributions in maize grain were different than those in SBM. There would be no major impact in feed formulation of using mean or median values for the eight AA reported here for maize grain (Table 1). This suggests that the lack of normal distributions for Met and Arg in maize grain is of little practical significance as maize grain is not a large contributor of these AA, despite being a high inclusion ingredient in most poultry feeds. As for predicting AA levels from determined CP values, the residuals of model estimation are normally distributed for all AA. However, using linear and quadratic models to predict AA from CP in maize grain and SBM appear inaccurate since the R 2 values are only between 0.02 and 0.51 (Table 3). Consequently, more studies should be considered to improve the accuracy of predicting AA levels from CP. CONCLUSION Although CP in SBM and some observed AA in maize grain and SBM were not normally distributed, there may be little effect on practical feed formulation, except for the Lys content in SBM. Using the mean Lys value for SBM in a nutrient matrix will result in over-estimating the Lys content. This could lead to formulating to an actual level lower than the targeted level. Moreover, the distribution of Lys in SBM is bimodal which indicates the need of further investigating the data, not simply using the mean value. Although the assumption of the normality of the residuals of the models are presented, other model fitting procedures beside linear or quadratic regression equations are needed to improve model predictions.

70 57 ACKNOWLEDGEMENTS The authors gratefully acknowledge the Ajinomoto Heartland LLC for access to their amino acid database.

71 58 Table 3.1 Descriptive statistics and Shapiro-Wilk test of crude protein and amino acids in maize grain samples Component N MIN a MAX b Mean Median Diff c SD CV Skewness W d P e NRC (1994) f Crude protein Lysine Threonine Methionine TSAA Valine Isoleucine Arginine Tryptophan a MIN, minimum value (Unit: g. per 88 g. dry matter). b MAX, minimum value (Unit: g. per 88 g. dry matter). c Diff, Mean Median value (Unit: g. per 88 g. dry matter). d W, the Shapiro-Wilk test for normally distributed data. e P, the probability distribution of the Shapiro-Wilk test at 0.05 significance level. f CP and AA contents of the ingredient listed for poultry by National Research Council, 1994 ( Unit: g. per 100 g. ).

72 59 Table 3.2 Descriptive statistics and Shapiro-Wilk test of crude protein and amino acids in soybean meal samples Component N MIN a MAX b Mean Median Diff c SD CV Skewness W d P e NRC (1994) f Crude protein Lysine Threonine Methionine TSAA Valine Isoleucine Arginine Tryptophan a MIN, minimum value (Unit: g. per 88 g. dry matter). b MAX, minimum value (Unit: g. per 88 g. dry matter). c Diff, Mean Median value (Unit: g. per 88 g. dry matter). d W, the Shapiro-Wilk test for normally distributed data. e P, the probability distribution of the Shapiro-Wilk test at 0.05 significance level. f CP and AA contents of the ingredient listed for poultry by National Research Council, 1994 ( Unit: g. per 100 g. ).

73 60 Table 3.3 Linear and quadratic relationships between crude protein and amino acids and in maize grain and soybean meal Amino Acids Regression Equation P a P b P c R 2 W residuals d Maize grain Lys = CP < < Threonine = CP CP Met = CP < < TSAA = CP < < Valine = CP < < Isoleucine = CP CP Arginine = CP < < Tryptophan = CP < < Soybean meal Lys = CP < Threonine = CP < < Met = CP < TSAA = CP CP Valine = CP < < Isoleucine = CP < < Arginine = CP CP Tryptophan = CP < P residuals e a P, the probability distribution of the intercept term b P, the probability distribution of the coefficient of crude protein c P, the probability distribution of the coefficient of crude protein square d W residuals, the Shapiro-Wilk test for normally distributed of the internally Studentized residuals (Gray and Woodall, 1994) e P residuals, the probability distribution of the Shapiro-Wilk test at 0.05 significance level.

74 61 Figure 3.1 Histograms with probability density function of lysine (normal distribution) and arginine (non-normal distribution) of maize grain samples analyzed between 2002 and Frequency Probability Density Frequency Probability Density Lysine content, g. per 88 g. dry matter Arginine content, g. per 88 g. dry matter Solid black line represents a normal distribution based upon the mean and standard deviation.

75 62 Figure 3.2 Quantile quantile plot of lysine (normal distribution) and arginine (non-normal distribution) of maize grain samples analyzed between 2002 and Lysine content, g. per 88 g. dry matter Arginine content, g. per 88 g. dry matter Normal Quantiles Normal Quantiles

76 63 Figure 3.3 Histograms with probability density function of valine (normal distribution) and lysine (non-normal distribution) of soybean meal samples analyzed between 2002 and Frequency Probability Density Frequency Probability Density Valine content, g. per 88 g. dry matter Lysine content, g. per 88 g. dry matter Solid black line represents a normal distribution based upon the mean and standard deviation.

77 64 Figure 3.4 Quantile quantile plot of valine (normal distribution) and lysine (non-normal distribution) of soybean meal samples analyzed between 2002 and Valine content, g. per 88 g. dry matter Lysine content, g. per 88 g. dry matter Normal Quantiles Normal Quantiles

78 65 Table 3.4 Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in 2002 a Variable N Minimum Maximum Mean Median Std Coefficient of Skewness Kurtosis Dev Variation DM CP LYS THR MET CYS M+C TRP VAL ARG ILE LEU HIS ALA GLU GLY ASP PHE PRO SER TYR a Values are calculated base on g. per 88 g. dry matter.

79 66 Table 3.5 Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in 2003 a Variable N Minimum Maximum Mean Median Std Dev Coefficient of Variation Skewness Kurtosis DM CP LYS THR MET CYS M+C TRP VAL ARG ILE LEU HIS ALA GLU GLY ASP PHE PRO SER TYR a Values are calculated base on g. per 88 g. dry matter.

80 67 Table 3.6 Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in 2004 a Variable N Minimum Maximum Mean Median Std Dev Coefficient of Variation Skewness Kurtosis DM CP LYS THR MET CYS M+C TRP VAL ARG ILE LEU HIS ALA GLU GLY ASP PHE PRO SER TYR a Values are calculated base on g. per 88 g. dry matter.

81 68 Table 3.7 Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in 2005 a Variable N Minimum Maximum Mean Median Std Dev Coefficient of Variation Skewness Kurtosis DM CP LYS THR MET CYS M+C TRP VAL ARG ILE LEU HIS ALA GLU GLY ASP PHE PRO SER TYR a Values are calculated base on g. per 88 g. dry matter.

82 69 Table 3.8 Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in 2006 a Variable N Minimum Maximum Mean Median Std Coeff of Skewness Kurtosis Dev Variation DM CP LYS THR MET CYS M+C TRP VAL ARG ILE LEU HIS ALA GLU GLY ASP PHE PRO SER TYR a Values are calculated base on g. per 88 g. dry matter.

83 70 Table 3.9 Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in 2007 a Variable N Minimum Maximum Mean Median Std Coeff of Skewness Kurtosis Dev Variation DM CP LYS THR MET CYS M+C TRP VAL ARG ILE LEU HIS ALA GLU GLY ASP PHE PRO SER TYR a Values are calculated base on g. per 88 g. dry matter.

84 71 Table 3.10 Descriptive statistics of crude protein and amino acids in soybean meal samples analyzed in 2008 a Variable N Minimum Maximum Mean Median Std Dev Coefficient of Variation Skewness Kurtosis DM CP LYS THR MET CYS M+C TRP VAL ARG ILE LEU HIS ALA GLU GLY ASP PHE PRO SER TYR a Values are calculated base on g. per 88 g. dry matter.

85 72 REFERENCES AOAC, Official Methods of Analysis of AOAC International, sixteenth ed. AOAC International, Gaithersburg, MD, USA. Box, G.P., Cox, D.R., An Analysis of Transformation. J. Roy. Statistical Society, Cravener, L.T., Roush, B.W., Improving Neural Network Prediction of Amino Acid Levels in Feed Ingredients. J. of Poult. Sci. 78, D Agostino, R.B., Stephans, M.A., Goodness of Fit Techniques. Marcel Deckker, Inc., New York, NY, USA. D Agostino, R.B., Belanger, A., D Agostino Jr., R.B., A Suggestion for Using Powerful and Informative Tests of Normality. Amer. Statistician 44, Gray, J.B., Woodall, W.H., The Maximum Size of Standardized and Internally Studentized Residuals in Regression Analysis. American Statistician 48, Hitachi Hitachi High Technologies America, Inc. 10 North Martingale Road, Suite 500 Schaumburg, Illinois Mendenhall, W., Sincich, T., A second course in statistics: Regression analysis, Pearson Education, Inc. NC, USA. National Research Council, Nutrient Requirements of Poultry, 9th rev.ed. National Academy Press, Washington, DC, USA. Oehlert, G.W., A First Course in Design and Analysis of Experiment. New York, Freeman, USA.

86 73 Park, H.M., Univariate Analysis and Normality Test Using SAS, Stata, and SPSS. Technical Working Paper. The University Information Technology Services (UITS) Center for Statistical and Mathematical Computing, Indiana University, USA. Royston, J.P., An Extension of Shapiro and Wilk's W Test for Normality to Large Samples. J. Appl. Statist. 31, Royston, J.P., Some Techniques for Assessing Multivariate Normality Based on the Shapiro - Wilk W. J. Appl. Statist. 32, SAS, SAS User s Guide. Statistics. Version 9.1 ed. SAS Institute Inc., Cary, NC, USA. Shapiro, S.S., Wilk, M.B., An Analysis of Variance Test for Normality. Biometrika 52, Shapiro, S.S., How to test normality and other distribution assumptions. American Society for Quality Control, WI, USA. Statistix9, User Manual. Analytical Software, PO Box 12185, Tallahassee FL 32317, USA.

87 74 CHAPTER 4 EVALUATION OF THE FIXED NITROGEN-TO-PROTEIN (N:P) CONVERSION FACTOR (6.25) VERSUS INGREDIENT SPECIFIC N:P CONVERSION FACTORS IN FEEDSTUFFS 1 1 N. Sriperm, G. M. Pesti, and P. B. Tillman, Journal of the Science of Food and Agriculture 91: Reprinted here with permission of publisher.

88 75 ABSTRACT Background: The crude protein (CP) of feedstuffs is important as an indicator of essential and non-essential amino acids for livestock. The protein (P) level needs to be known accurately, to minimize the feeding of excess nitrogen (N) and to reduce N pollution. Laboratory methods for determining N content report N from amino acids, but also N from ammonia and from nonamino acid sources. The determined CP based on 6.25 x N level typically over-estimates the true protein of feedstuffs. Result: Determined ingredient specific N:P conversion factors k A, k P and k were not equal to the standard 6.25 factor. The k A had the highest value in all ingredients, which leads to the estimation of specific crude protein (SCP), which is closer to true protein (the summation of the total amino acid residues from amino acid analyses). The SCP(k A ) was lower than CP and true protein in all ingredients, demonstrating that CP might over-estimate the actual protein in feedstuffs. Conclusion: Based on data from 677 feedstuff samples from 2009, it is concluded that the mean k A, should be 5.68 for corn, 5.64 for soybean meal, 5.74 for corn DDGS, 5.45 for poultry byproduct meal and 5.37 for meat and bone meal. INTRODUCTION In animal agriculture, crude protein (CP) has been widely accepted as an indicator of protein content of feedstuffs or formulated feed and CP has historically been used in the formulation of feeds for monogastric animals. Crude protein is calculated by multiplying the total nitrogen determined by methods such as Kjeldahl or Dumas methods by a standard nitrogen conversion factor of The 6.25 factor does not relate to any specific feedstuff i.e. it is a non ingredient specific nitrogen-to-protein (N:P) conversion factor which assumes one kg of plant or

89 76 animal protein contains 160 g N (1000/160 = 6.25). 1 However, Jones 1 reported the nitrogen content of different proteins isolated from plants and animals ranged from 130 to more than 190 g N per kg. Yeoh and Wee 2 recommended a N:P conversion factor for plants of The N:P conversion factor in feedstuffs can vary depending upon the amino acid composition, as the nitrogen content of each amino acid is different. For example, the variation in N:P conversion factors, such as in meat products, differs depending upon the collagen level. Since the nitrogen content of collagen is 180 g kg -1, it is clear the level of collagen can affect the N:P factor in meat products. 3 Moreover, the nitrogen content of feedstuffs is not just derived from protein and amino acids, so other nitrogenous organic compounds also need to be accounted for. These include nucleic acids, urea, ammonia, phospholipids, nitrates, purine derivatives, etc. 3 Thus, using a traditional fixed factor of 6.25 to calculate the determined total organic N content of a protein is not a realistic approach. Protein consists of at least 20 different amino acids bound together by peptide bonds between the carboxyl and amino groups of adjacent amino acids. The process of protein formation eliminates one molecule of water, from dehydration synthesis, for each peptide bond formed between the adjacent amino acids. 4 It has been proposed that a calculation of Net (or True) protein be used. 5 True protein (TP) is the summation of the total amino acid residues from an amino acid analysis. Gatel and Grosjean 6 found that reducing the CP content of a growing pig diet from approximately 195 to 178 g kg -1 DM basis, and of a finishing diet from 167 to 155 g kg -1 DM basis, decreased the urinary N excretion between 15 to 20%. This research supports our objective to use a specific N:P conversion factor instead of the standard 6.25 to reduce the CP content in diets. In so doing, the possibility exists to reduce N excretion into the environment. There are many suggested procedures to define the N:P conversion factor for an ingredient, of which one is using amino acid composition. 3 In this paper,

90 77 we define the N:P conversion factors, as proposed and outlined by Mosse' et al., 7 of common monogastric feed ingredients. The objectives of the present paper were to: 1) show the amino acids composition of five common feedstuffs, 2) determine ingredient specific N:P conversion factors, 3) determine a linear regression using the total N to predict ingredient specific N:P conversion factors, 4) compare TP values with ingredient specific crude protein (SCP) values. MATERIALS AND METHODS Five major feedstuffs: ground corn, soybean meal (dehulled solvent extracted (SBM)), corn distillers dried grains with solubles (corn DDGS), meat and bone meal (MBM), and poultry by-product meal (PBM), as defined by AAFCO 8, were analyzed in 2009 at Ajinomoto Heartland LLC s amino acid laboratory 9 (Table 1.). The ingredient samples tested, while having an unidentified regional origin, were collected from across the United States. Dry matter (DM) was determined in a forced air oven at 135 o C for 2 hours (AOAC, ). 10 Total N contents (N L, Table 2) were determined by the Dumas N combustion procedure (AOAC, ) 10 and were also reported as CP (N L x 6.25, Table 1). Ammonia (NH 3 ) and 17 amino acids, excluding tryptophan (Trp), were determined by HPLC (HITACHI 8900) with Ninhydrin in a post-column derivatization (AOAC, ). 10 Performic acid oxidation (AOAC, E [b]) 10 was conducted prior to acid hydrolysis for the determination of methionine and cysteine. Tryptophan was hydrolyzed under alkaline conditions using barium hydroxide and subsequently acidified using hydrochloric acid. Tryptophan from the extracts was separated by reverse phase HPLC and determined by fluorometric detection.

91 78 Statistical analysis The means and standard deviations of total amino acids and ammonia were computed by the PROC MEANS procedure of SAS 11. PROC GLM of SAS 11 was used to compare mean differences using Duncan's new multiple range test when differences were found at the 5% significance level. The residual outliers were eliminated and regression analysis assumptions of 1) normality, 2) linearity, 3) independence and 4) homoscedasticity were tested to assure a valid regression model. 12 Normality of residuals was determined using PROC UNIVARIATE of SAS 11 and by using the Shapiro-Wilk test as outlined by Sriperm et al. 13 The assumption of linearity was determined by plotting dependent against independent variables. Two regression analysis assumptions, independence and homoscedasticity, of the residuals was determined by a plot of residuals versus predicted values. 12 Since the regression analysis assumptions were satisfied, linear regressions of total anhydrous amino acid residues ( E i ) or total N from amino acids and NH 3 ( D i ), as a function of total N content from the Dumas method (N L ) for 5 feedstuffs were computed using the PROC REG procedure of SAS 11. Calculation of nitrogen-to-protein conversion factors Mosse' et al. 7 reported several methods of calculating ingredient specific N:P conversion factors k A, k P and k from E i, D i and N L. E i is the sum of anhydrous amino acid residues where E i is the grams of anhydrous AA residue of the i th amino acid kg -1 DM. Anhydrous amino acid residues were calculated by reducing the assay weight to account for the loss of 1 molecule of H 2 O which occurs when amino acids are joined to form dipeptide linkages, and ultimately proteins. This was done by multiplying individual amino acid residues (AA i, g kg -1 DM) by the ratio of each individual amino acid s molecular weight minus a water molecule (AA i(mw) -H 2 0) to the amino acid s molecular weight (AA i(mw) ) (Equation 1).

92 79 (1) D i is the sum of the total N content from each of the 18 amino acids reported (AA i(n atom) ) including the amide N (NH 3(N atom) ) of asparagine (Asn) and glutamine (Gln) where D i is grams of N of the i th amino acid kg -1 DM (Equation 2). During acid hydrolysis the and amide linkages in Asn and Gln are acid labile. Consequently, all of the Asn and Gln residues are converted to aspartic acid and glutamic acid respectively, while the released amino nitrogen became free ammonium. 4 Due to a lack of precisely defined amide N, Mosse' et al. 7 suggested using total NH 3 instead of amide N. Thus, D i for this study is the total N determined from both amino acids and NH 3 (mainly from Asn and Gln). (2) The k A is the ratio of E i and D i (Equation 3). k A (3) The k P is the ratio of E i and N L (Equation 4). k P (4) As defined by Mosse' et al., 14 the k is an average value of k A and k P (Equation 5). k (5) N recovery was calculated as the ratio D i to N L or with a ratio of k P to k A. The N recovery can be used to identify how well the N was recovered from protein. 15 RESULTS AND DISCUSSION Analyzed total amino acid and ammonia concentration means and standard deviations (SD) are described in Table 1 for the five common ingredients tested. Among protein sources,

93 80 almost all of the mean amino acids values in PBM were higher than MBM except for glycine, but MBM had more variation (higher SD). Soybean meal and PBM had similar phenylalanine values but significant differences existed for most of the AAs. However, PBM had noticeably higher variation for most of the amino acids compared with SBM. As expected, corn DDGS had higher amino acids than corn, since the starch fraction has been removed. Proline values in corn DDGS and SBM were not significantly different (P>0.05). Data for the calculated ingredient specific N:P conversion factors (k A, k P and k), E i, D i, N L and N recovery for 5 feedstuffs are described in Table 2. For a given ingredient specific N:P conversion factors, corn and corn DDGS have somewhat similar values as SBM. The corn samples have ingredient specific N:P conversion factor values of 5.06 (k P ) and 5.68 (k A ) which is similar to the range (5.59 to 5.72) previously reported. 3,16 Soybean meal samples have an average ingredient specific N:P conversion factor values of 5.13 (k P ) and 5.64 (k A ) in this data set which is similar to the range (5.38 to 5.67) previously reported. 3,14 Corn DDGS, PBM and MBM have average values of 4.99 and 5.74, 4.81 and 5.45 and 4.77 and 5.37 for k P and k A, respectively. Basically no N:P conversion factor have been reported in the literatures for corn DDGS, PBM and MBM. Mariotti et al. 3 reported a N:P conversion factor for chicken at 5.53 and reported a range for beef of 5.38 to Sosulski and Imafidon 16 reported a N:P conversion factor for meat and meat products at SD. From this analysis, SBM and corn DDGS have higher factors compared to PBM and MBM. Thus, plant protein sources have higher N:P conversion factors than animal based protein sources, which differs from previous studies. 3,14 Sosulski and Imafidon 16 suggested using a N:P conversion factor of 5.70 for animal and plant foods. Lourenco et al. 17 reported N:P conversion factors using k A and Mosse' et al. 14 mentioned that k A has been used outside of the food industry and is thus reasonable to be used for purified protein fractions

94 81 such as soybean isolate. In general, k A is larger than k P because the denominator ( D i ) in the k A equation (Equation 3) is only calculated from total N, estimated from both amino acids and NH 3, while the denominator (N L ) in the k P equation (Equation 4) includes total N from amino acids, NH 3 plus nonprotein N. Consequently, since D i is smaller than N L, the smaller denominator ( D i ) at the same constant numerator ( E i ) gives a larger value, which is k A. Therefore, the average of k A and k P gives a close approximation of the true protein conversion factor (k). 14 However, Salo-Vaananen and Koivistoinen 5 reported the general conversion factor of 5.33 for various food groups which is close to k values of corn, SBM and corn DDGS but is closer to the k A values of PBM and MBM in the present study (Table 2). The N L and N recovery are also reported and the N recovery values (Table 2) were less than one as previously reported. 7 The N recovered from the total amino acid analysis ranged from 0.87 to 0.91 suggesting that the N recovery from total amino acid analysis (total N from amino acids and NH 3 ) is relatively high compared with the total N from the Dumas method which accounts for all the N in the sample whether from protein or other sources. The N L can be used to predict the total anhydrous amino acid residues (Predicted E i, Equation 6) and total N from amino acids and NH 3 (Predicted D i, Equation 7), for all 5 ingredients, as shown in Table 3. Predicted E i = b 0 + b 1 N L i Predicted D i = b 0 + b 1 N L i (6) (7) In the prediction equation for predicted E i and predicted D i, b 0 is the intercept and b 1 is the slope of both linear regressions. An example of the plots of E i and D i against N L are shown in Figures 1 and 2. The correlation of determination (R 2 ) was used to evaluate how well the dependent variables ( E i and D i ) were estimated by the linear model where R 2 is calculated as

95 82 the square of the correlation coefficient. 12 The results showed that most of the ingredients tested here had R 2 values over 60% except SBM (approximately 45%). These high R 2 values imply over 60% of the sample variability in the dependent variables can be explained by N L using a linear model. The standard errors (SE), presented in parentheses (Table 3), explain how well the sample represents the population. The smaller the SE, the better the result from the regression equation will represent the sample. The assumptions of normality, linearity, independence and homoscedasticity were found to be valid for all the relationships investigated. Since the normal probability plots of the residuals for all 5 ingredients showed a straight line, normality is implied. Moreover, the Shapiro-Wilk critical values failed to reject the null hypothesis, of the residuals being normally distributed at the 5% significance level (P > 0.05), indicating the residuals were normally distributed for all the relationships. Comparing all 5 feedstuffs tested here, CP levels were higher than TP (Table 4), which means using CP instead of TP would overestimate the protein level in feedstuffs. Table 4 shows the calculated specific crude protein (SCP: k A, k P and k) using the respective mean values for k A, k P and k for each feedstuff. The results indicate that SCP gives lower values compared with CP or TP. This was because the ingredient specific N:P conversion factors were calculated from anhydrous amino acid residues (amino acid residues after subtracting one water molecule) while TP is the summation of the amino acid residues. Mosse' et al. 14 suggested that SCP (k) is the best estimate of TP; whereas this data shows using k A to calculate SCP (k A ) gives the closest estimate of TP. CONCLUSION From this analysis, using an ingredient specific N:P conversion factor should be considered as the appropriate approach to determine the SCP and SCP(k A ) closely estimates TP. Based on data from 677 feedstuff samples used by the American feed industry in 2009, it is

96 83 concluded that the k A are for corn, for SBM, for corn DDGS, for PBM and for MBM. Determined ingredient specific N:P conversion factors k A, k P and k were not equal to the standard 6.25 factor. The k A had the highest value in all ingredients, which leads to the estimation of SCP, which is closer to true protein (the summation of the total amino acid residues from amino acid analyses). The SCP(k A ) was lower than CP and true protein in all ingredients, demonstrating that CP over-estimate the actual protein in feedstuffs.

97 84 Table 4.1 Total amino acid and ammonia composition, DM content and sample size (g kg -1 ) of 5 feedstuffs (mean + SD). Component a Corn Soybean Meal Corn DDGS Poultry By-Product Meal Meat & Bone Meal Dry matter ± 18.60d ± 8.02b ± 13.73c ± 12.01a ± 16.67a NH 3 a 1.72 ± 0.26e 9.07 ± 1.35a 5.87 ± 0.97d 8.22 ± 0.87b 7.10 ± 1.52c Lysine 2.62 ± 0.24e ± 1.11b 9.78 ± 0.94d ± 5.27a ± 4.13c Threonine 2.86 ± 0.23e ± 0.74b ± 0.57d ± 2.24a ± 3.47c Methionine 1.77 ± 0.19d 7.66 ± 0.38b 6.02 ± 0.36c 9.91 ± 2.44a 7.78 ± 1.61b Cysteine 1.72 ± 0.13d 6.85 ± 0.38b 5.29 ± 0.30c 8.81 ± 4.20a 7.16 ± 4.43b Arginine 4.15 ± 0.38e ± 1.91b ± 1.17d ± 3.43a ± 3.93c Isoleucine 2.71 ± 0.26e ± 0.93a ± 0.66d ± 2.41b ± 4.05c Leucine 9.39 ± 1.00e ± 1.55b ± 2.07d ± 3.79a ± 7.01c Valine 3.79 ± 0.34d ± 0.93b ± 0.89c ± 3.58a ± 5.89b Histidine 2.29 ± 0.19e ± 0.73a 7.86 ± 0.42d ± 2.07b 9.56 ± 1.92c Alanine 5.88 ± 0.57e ± 1.07c ± 1.37d ± 2.95a ± 3.81b Glutamic Acid ± 1.41e ± 3.61a ± 3.42d ± 6.63b ± 6.66c Glycine 3.22 ± 0.32e ± 0.95c ± 0.65d ± 5.75b ± 11.25a Aspartic Acid 5.22 ± 0.65e ± 2.24a ± 1.50d ± 4.62b ± 4.92c Phenylalanine 4.00 ± 0.42d ± 1.87a ± 1.24c ± 2.00a ± 5.51b Proline 6.66 ± 0.98c ± 3.55b ± 2.89b ± 7.66a ± 7.80a Serine 3.82 ± 0.32d ± 0.92b ± 0.74c ± 7.51a ± 8.15b Tyrosine 2.38 ± 0.37d ± 1.93a 9.01 ± 1.19c ± 2.52b 8.91 ± 2.92c Tryptophan 0.65 ± 0.07e 6.84 ± 0.40a 2.39 ± 0.18d 4.95 ± 0.92b 3.68 ± 0.96c Sample size a Ammonia consists of N recovered from N sources other than those shown. Means in a row with different letters (Duncan's new multiple range test) differ significantly (P < 0.05).

98 85 Table 4.2 Nitrogen to protein (N:P) conversion factors, total nitrogen content and nitrogen recovery of 5 feedstuffs (mean + SD) Feedstuffs Total N ( E (N L ) i ) a ( D i ) a Nitrogen Ingredient Specific N:P Conversion Factors c Recovery b (g kg -1 ) (g kg -1 ) (g kg -1 ) k A k P k Corn 13.1 ± ± ± ± ± ± ± 0.11 Soybean Meal 83.6 ± ± ± ± ± ± ± 0.08 Corn DDGS 47.2 ± ± ± ± ± ± ± 0.11 Poultry By-Product 99.9 ± ± ± ± ± ± ± 0.10 Meat & Bone Meal 87.7 ± ± ± ± ± ± ± 0.14 a E i is the sum of anhydrous amino acid residues, D i is the sum of the total N from amino acids and NH 3. b the ratio of total N from amino acids and NH 3 to total N from the Dumas method ( D i /N L ), or k P / k A c k A is the ratio of E i and D i, k P is the ratio of E i and N L. k is the average of k A and k P.

99 86 Table 4.3 Linear regressions of total anhydrous amino acid residues ( E i ) or total nitrogen from amino acid and NH 3 ( D i ) as a function of total nitrogen content from the Dumas method (N L ) a for 5 feedstuffs. b Feedstuffs Predicted E c i d Predicted D i b 0 b 1 R 2 b 0 b 1 R 2 Corn (2.949) (0.223) (0.547) (0.041) Soybean meal (25.608) (0.306) (4.661) (0.056) Corn DDGS (11.315) (0.240) (2.105) (0.045) Poultry By-Product (37.989) (0.379) (6.434) (0.064) Meat & Bone Meal (12.944) (0.147) (2.046) (0.023) a unit g kg -1 of dry matter. b The standard errors are presented in parentheses. c Equation (6): Predicted E i = b 0 +b 1 N L. d Equation (7): Predicted D i = b 0 +b 1 N L.

100 87 Table 4.4 Specific crude protein, crude protein and true protein content of 5 feedstuffs. Feedstuffs Specific Crude Protein (SCP, g kg -1 ) Crude protein (CP) d True protein (TP) e SCP(k A ) a SCP(k P ) b SCP(k) c (g kg -1 ) (g kg -1 ) Corn ± 8.5 f 76.7 ± 7.1 Soybean Meal ± ± 17.6 Corn DDGS ± ± 14.6 Poultry By-Product ± ± 41.5 Meat & Bone Meal ± ± 55.3 a SCP(k A )= N L x k A. b SCP(k P )= N L x k P. c SCP(k) = N L x k. d CP = N L x e TP is sum of the total amino acid residues from the amino acid analysis. f mean + SD

101 88 Figure 4.1 A plot of total anhydrous amino acid residues ( E i ) on total nitrogen from the Dumas method (N L ) for soybean meal. Soybean meal E i, g kg N L, g kg -1 E i = N L

102 89 Figure 4.2 A plot of total nitrogen from amino acids and NH 3 ( D i ) and N L for corn DDGS Corn DDGS D i, g kg N L, g kg -1 D i = N L

103 89 REFERENCES 1. Jones DB, Factors for converting percentages of nitrogen in foods and feeds into percentages of proteins. United States Department of Agriculture 183: 1-22 (1931). 2. Yeoh HH and Wee YC, Leaf protein contents and nitrogen-to-protein conversion factors for 90 plant species. Food Chemistry 49: (1994). 3. Mariotti F, Tome D and Mirand PP, Converting nitrogen into protein-beyond 6.25 and Jones factors. Food Science and Nutrition 48: (2008). 4. Garrett HR and Grisham MC, Biochemistry. Thomson Learning, CA. (2007). 5. Salo-Vaananen PP and Koivistoinen PE, Determination of protein in feeds:comparison of net protein and crude protein (Nx6.25) values. Food Chemistry 57: (1996). 6. Gatel F and Grosjean F, Effect of protein content of the diet on nitrogen excretion by pigs. Livestock Production Science 31: (1992). 7. Mosse J, Huet JC and Baudet J, The amino acid composition of wheat grain as a function of nitrogen content. J. Cereal Sci. 3: (1985). 8. AAFCO, Association of American Feed Control Officials Incorporated Official Publication. Association of American Feed Control Officials, Inc. Sacramento, CA. USA. ISBN: (2010). 9. Ajinomoto Heartland LLC. 1 Heartland Drive Eddyville, IA AOAC, Official Methods of Analysis of AOAC International, sixteenth ed. Association of official analytical chemists international, Gaithersburg, MD, (2000). 11. SAS, SAS User s Guide. Statistics. Version 9.1 ed. SAS Institute Inc., Cary, NC, (2004). 12. Mendenhall W and Sincich T, A second course in statistics: Regression analysis. Pearson Education, New York (2003).

104 Sriperm N, Pesti GM and Tillman PB, The distribution of crude protein and amino acid content in maize grain and soybean meal. Anim Feed Sci Technol. 159: (2010). 14. Mosse J, Nitrogen to protein conversion factor for ten cereals and six legumes or oilseeds. A reappraisal of its definition and determination. Variation according to species and to seed protein content. J. Agric. Food Chem. 38: (1990). 15. Yeoh HH and Truong VD, Protein contents, amino acid compositions and nitrogen-to-protein conversion factors for cassava roots. J. Sci. Food Agric. 70: (1996). 16. Sosulski FW and Imafidon GI, Amino acid composition and nitrogen-to-protein conversion factors for animal and plant foods. J. Agric. Food Chem. 38: (1990). 17. Lourenco SO, Barbarino E, De-Paula JC, S.Pereira LOD and Marquez UML, Amino acid composition, protein content and calculation of nitrogen-to-protein conversion factors for 19 tropical seaweeds. Phycological Research 50: (2002).

105 91 CHAPTER 5 RESPONSE SURFACE ANALYSIS OF DIETARY BALANCED PROTEIN RESPONSES DURING GROWER AND FINISHER PHASES OF ROSS 708 MALE BROILERS 1 1 N. Sriperm and G. M. Pesti. To be submitted to Journal of Applied Poultry Research

106 92 ABSTRACT Various broiler phase feeding programs have been evaluated in an attempt to maximum technical performance responses. In this study, a titration of dietary balanced protein (BP) during the grower (15 to 34d) and finisher (35 to 42d or 49d) phases of Ross x Ross 708 broilers was evaluated. A five-level-two-factor central composite rotatable design providing 5 levels of BP during the 2 phases, generating 9 treatment combinations was used. The digestible Lys (dlys) level dictated the BP level since key essential amino acids were set as minimums relative to dlys. Corn, soybean meal and meat & bone meal based diets were formulated to dlys levels of 0.75, 0.82, 1.00, 1.18 and 1.25 % during the grower and 0.69, 0.76, 0.92, 1.08 and 1.15% during the finisher phases. Body weight, BW gain, feed intake, dlys intake, FCR, mortality, live bird weight, carcass, breast meat, and leg quarter weights were measured at 35, 42 and 49d and yields were calculated. During the grower phase, the dlys levels yielding maximum responses were 1.126, 0.984, 1.388, and 1.135% for BW gain, feed intake, FCR, breast meat weight and breast meat yield, respectively. During the finisher phase, the dlys levels yielding maximum responses were and 1.076% for BW gain and FCR. Evaluating the combination of both phases up to 42d, the dlys levels yielding maximum responses were and 0.996% for BW gain, and 0.778% for feed intake, and % for live bird weight, and 0.994% for carcass weight, and 0.981% for breast meat weight and and 0.920% for breast meat yield, respectively. Evaluating the combination of both phases up to 49d, the dlys levels yielding maximum responses were: and 0.972% for BW gain, and 0.802% for feed intake, and % for live bird weight, and 0.988% for carcass weight, and 0.995% for breast meat weight and and 1.004% for breast meat yield, respectively.

107 93 There were no interactions for any of the responses which meant responses to BP during the finisher phase are independent of the grower phase. INTRODUCTION Considerable research has been conducted to evaluate the protein and/or amino acid requirements of broiler chickens. Smith and Pesti [1] found that different broiler strains have different protein requirements to maximize BW versus feed efficiency. Smith et al. [2] applied quadratic response models to determine growth performance to protein and bird age in two different broiler strains (Ross x Ross 208 and Peterson x Arbor Acres). Genetic selection plays an important role in altering broiler nutrient requirements. The modern Ross x Ross 708 nutrition specification [3] recommends a feeding program consisting of 3 phases (0 to 10d (1.43% total Lys; 1.27% digestible Lys (dlys)), 11 to 24d (1.24% total Lys; 1.10% dlys), 25 to slaughter (1.09% total Lys; 0.97% dlys), in order to meet the requirements of this modern genetic strain. The NRC [4] recommended 3 phases but at considerably lower levels (0 to 21d (1.10% total Lys), 21 to 42d (1.00% total Lys), 42 to 56d (0.85% total Lys), perhaps because it was published over 16 years ago. Many different phase feeding programs have been proposed by researchers. Warren and Emmert [5] compared 3 different feeding programs (Phase-feeding (PF), feeding Illinois ideal chick protein (IICP) and NRC [4]). From 0 to 21d, there were no differences in BW gain, feed intake, feed efficiency, digestible amino acid (AA) intake, or gain per unit digestible AA intake among chicks fed PF, IICP, or NRC diets. From 40 to 61d, no differences were observed in BW gain or feed intake among chicks fed PF, IICP, or NRC diets. However, feed efficiency (gain per feed) of birds fed IICP diet was significantly lower than the other two programs. There were no differences in processing characteristics observed among feeding programs. Phase feeding reduced feed costs without giving up either growth performance or

108 94 carcass yield. Kidd et al. [6] reported on a BP feeding program consisting of 4 phases from 0 to 49 d using male and female Ross x Ross 508 broilers. They found that the highest AA density throughout the grow-out period resulted in maximizing breast meat yield. The highest total Lys levels were 1.38, 1.19, 1.11, and 1.06 % for 0 to 14 d, 15 to 28 d, 29 to 35 d and 36 to 49 d, respectively. Feeding 23.4% CP or 1.38% total Lys for 0 to 14 d and 21.8% CP or 1.19% total Lys for 15 to 28 d gave the maximum saleable white meat. Kidd et al. [7] reported on a BP feeding program consisting of 5 phases from 0 to 55 d using male and female Ross x Ross 708 broilers. Birds performed the best when fed high AA density diets during the first 3 phases. Total dlys levels fed were 1.38, 1.36, and 1.23 % for 1 to 5 d, 6 to 14 d and 15 to 35 d, respectively. They concluded that the Ross 708 strain broilers critically need high AA density from 1 to 35 d in order to maximize performance. Lemme et al. [8] reported on a BP feeding program consisting of 3 phases from 0 to 49 d using male and female Ross x Ross 308 and 708 broilers. Total Lys levels from low to high ranged from 1.03 to 1.72% for 1 to 10 d, 0.87 to 1.44% for 11 to 28 d and 0.78 to 1.28% for 29 to 49 d. They found that the highest balanced protein levels optimized BW, processing performance and FCR. There was no difference in biological response to balance protein between the Ross strains tested. Corzo et al. [9] reported on a BP feeding program consisting of 4 phases from 0 to 56 d in 2 broiler strains. Digestible Lys levels fed from low to high ranged from 1.09 to 1.22% for 1 to 14 d, 0.97 to 1.07% for 14 to 28 d, 0.84 to 0.92% for 28 to 42 d and 0.79 to 0.81% for 42 to 56 d. They found that the higher dlys levels resulted in lower abdominal fat percentage while also maximizing fillet and tender yield at 42 and 56 d of age. Dozier et al. [10] using BP, fed straight-run Ross 708 broilers with a 4 phase feeding program to 59 d (1 to 17 (High (H) dlys level= 1.36%), 18 to 35 d (H = 1.18%), 36 to 47d (H = 0.93%, Medium (M) = 0.84%, Low (L) = 0.75%), 48 to 59 d (H = 0.89%, M = 0.80%, L =

109 %)). Birds were fed HH diet to 35d of age, and then experimental diets were started. The H AA densities throughout the grow-out period provided the best cumulative feed conversion. Decreasing AA density (H (0.89%) to L (0.72%) for 48 to 59 d or H (0.93%) to L (0.75%) for 36 to 47 d and H (0.89%) to L (0.72%) for 48 to 59 d) limited breast fillets, tender and total white meat yields when compared with the H AA density. Carcass and breast meat yield was not affected by feeding the M dlys density (0.84%) from 36 to 47 d and from feeding either the H (0.89%) to M (0.80%) or L (0.72%) dlys for 48 to 59 d compared with H AA density. The H AAs density was the most economical due to the improvement in cumulative feed conversion. In another experiment, Dozier et al. [11] fed straight-run Ross 708 broilers diets up to 35 d. Digestible Lys levels were 1.21 and 1.06% for 1 to 17 d and 18 to 35 d, respectively. Birds were fed diets containing one of the three levels of dlys (H, 0.92%, M, 0.85% or L, 0.78%) from 36 to 47d followed by 2 levels of dlys H (0.88%) and L (0.75%) from 48 to 60 d. The H AA density program improved cumulative feed consumption over the M and L density programs and increased breast meat yield by 0.6% compared with the L density program. Overall, these studies agree with the importance of feeding H AA densities in early stages (1 to 35 d) to improve growth performance and carcass characteristic. Later in life (upto 59d), birds might not need the high AA density and reduced levels can be fed without losing bird performance significantly [10]. Pesti and Fletcher [12] found that broilers fed low protein diets from 21 to 42 d were able to compensate in growth performance during the finisher phase (371 versus 331 g gained) and feed utilization was more efficient (0.383 versus g of gain per g of feed intake). The purpose of this research was to investigate the relationship between feeding different nutrient densities to Ross x Ross 708 male broilers during grower (15 to 34 d) and finisher (35 to 42 d and 35 to 49 d) phases using the central composite rotatable design to determine the

110 96 probability that interactions exist. This comprehensive design has been widely applied in many areas of research, including poultry nutrition [13-16] and is commonly used for fitting a secondorder response surface problems [17]. MATERIALS AND METHODS Bird Husbandry Ross x Ross 708 broiler chicks were obtained from a commercial hatchery (Columbia Farm of Georgia, Lavonia, Georgia) where they were vaccinated for Marek s disease, Newcastle disease, and infectious bronchitis. Chicks were vent sexed and 35 male birds (1,683 birds) were randomly assigned to each of 48 pens within 2 rooms, 24 pens per room. Stocking density was 0.11 m 2 / bird. Each pen had clean wood shavings litter, and was equipped with a hanging feeder and a nipple water drink line. Feed and water were provided ad libitum and experimental diets were pelleted. Feeder trays were provided from day 1 to 7 to ensure chicks had access to feed. Room temperatures were set at 33 0 C for the first 3 d and were decreased by 2 0 C every 3 d until 21 0 C was reached, which was maintained for the duration of the experiment. Average ambient temperature during the experiment was 24.3 o C (+ 3.8 standard deviation) for Room 1 and 24.4 o C (+ 3.5 standard deviation) for Room 2. The lighting program used consisted of a lighting intensity of 30 lux for 24 hours from 0 to 3 d of age, 30 lux for 23 hours from day 4 to 10 d of age, 10 lux for 20 hours from 11 to 22 d of age, 5 lux for 20 hours from 23 to 35 d of age, and 3 lux for 20 hours from 36 to 49 d of age. Light intensity settings were measured at bird level (30 cm from litter level) using a Light ProbeMeter (403125, Extech Instruments, Waltham, MA). Average relative humidity was 42.1 o C (+ 6.6 standard deviation) for Room 1 and 41.1 o C (+ 6.7 standard deviation) for Room 2 during the 49 d trial.

111 97 Measurements Pen body weights and feed consumption were measured at 0, 14, 35, 42 and 49 d of age to calculate avearge body weight gain and feed consumption. Average feed consumption was adjusted for mortality based on number of bird days. Feed conversion was calculated using the adjusted average feed consumption divided by total weight gain for the period. Digestible Lys intake per d and dlys intake per BW gain were calculated. Twelve randomly selected birds for processing were wing banded at 3d of age. Four of the previously wing banded birds per pen were slaughtered at 35, 42 and 49 d of age. The selected birds were moved to a separate room by treatment for overnight withdrawal of feed (approximately 12 hours); water was provided for ad libitum consumption. The following morning, birds were randomly selected from each treatment to ensure the processing error was distributed across the treatments. At processing, birds were individually weighed then electrically stunned, bled, scalded, mechanically picked, and mechanically eviscerated. Whole carcasses (with abdominal fat, without tails) were weighed and chilled in ice water slurry in a walk-in cooler at 5 o C for 4 hours. The chilled carcasses were drained of any excess water, weighed and deboned to obtain weights of breast meat (pectoralis majors and minors muscles), and leg quarters. Percentage carcass yield, breast meat yield, and leg quarter yield were calculated based on live BW (before processing). Diets Corn, soybean meal and meat and bone meal based diets were formulated based on balanced AA recommendations from Ajinomoto Heartland LLC [18], using apparent AA digestibility coefficients of feed ingredients from their data base. A safety margin (mean minus 50% SD) on the digestibility coefficients for each AA was used. Key diets are shown in Table 1. From day 0 to 14, all chicks received a starter diet with 23.5% CP and 1.28 % dlys (Table 2).

112 98 Treatment diets were assigned at random to 48 floor pens in two environmentally controlled rooms. Isoenergetic grower treatment diets formulated on targeted dlys levels of 0.75, 0.82, 1.00, 1.18 and 1.25 % were offered during the grower phase (15 to 34 d, Tables 1 and 2). Isoenergetic finisher treatment diets formulated on targeted dlys levels of 0.69, 0.76, 0.92, 1.08 and 1.15% were offered during the finisher phase (35 to 49 d, Tables 1 and 2). Corn, soybean meal, meat and bone meal and complete diets were analyzed for total AAs (AOAC method , ; [19]), and crude protein (AOAC method ; [19]), Ca, Na, total P concentration (ICAP using a wet digest procedure; [20]) and phytase activity [21] to ensure the diets contained levels as formulated (Table 2). Digestible essential AAs (Lys, Met, TSAA, Thr, Val, Ile, Trp and Arg) were recalculated using the analyzed total AAs of the feed ingredients and diets (Table 2). The recalculated values of dlys in the diets were 0.77, 0.85, 1.04, 1.23 and 1.30% during the grower and 0.71, 0.78, 0.95, 1.13 and 1.20% during the finisher phase (Table 2). The analyzed and recalculated formula values were used throughout this paper. The grower basal diets (Grower Low dlys (0.77%) and Grower High dlys (1.30%)) were mixed and blended so as to create 3 intermediate diets (0.85, 1.04 and 1.23%). The 0.85% dlys treatment diet was created by blending 85.4% of the low dlys diet and with 14.6% of the high dlys diet. The 1.04% dlys treatment diet was created by blending 50% of the low dlys diet with 50% of the high dlys diet. The 1.23% dlys treatment diet was created by blending 14.6% of the low dlys diet with 85.4% of the high dlys diet. The same procedure applied to finisher diet mixing. Experimental Design A five-level-two-factor central composite rotatable design (CCRD) was used in this study [22]. Experimental Phase 1 (15-34 d) consisted of 5 dlys levels (the grower phase). The 5 treatments from 0.77, 0.85, 1.04, 1.23 and 1.30% dlys level had 5, 10, 18, 10 and 5 replications,

113 99 respectively. Experiment Phase 2 (35-42 d or d) consisted of 9 treatment combinations as shown in Table 3 and Figure 1. Each of the 9 treatments contained 5 replications except for the center point (Treatment 5) which had 8 replications. The center point of the response surface is designed to be as close to the theoretical optimum as possible, based on current Ross 708 nutrient recommendations [3]. The design variables were dlys levels at grower phase (x 1, %) and finisher phase (x 2, %) while response variables were BW, BW gain, feed intake, dlys intake per d, dlys intake per BW gain, FCR, mortality, carcass weight, carcass yield, breast meat weight and yield and leg quarter weight and yield. Statistical Analysis The general linear model procedure, PROC GLM of SAS [23], was used to analyze experimental data from 15 to 34d (Experiment Phase 1) and 35 to 49d (Experimental Phase 2). The effect of dlys levels to responses (growth performance and processing characteristics) was determined using orthogonal polynomials. The null hypothesis of orthogonal polynomials stated that there was no significant linear or quadratic effect on the responses from birds fed different dlys levels [24]. The least square estimation procedure, PROC REG of SAS [23], was used to fit a quadratic regression model to calculate the dlys level yielding maximum technical performance during the grower phase. The maximum dlys levels were calculated using partial derivatives of the quadratic response models. The Overall Phases data from both 15 to 42 d and 15 to 49 d were analyzed by the response surface regression procedure, PROC RSREG of SAS [23], to fit the following secondorder polynomial equation: (1)

114 100 where y i is the response variable, 0, i, ii, and ij are regression coefficients for intercept, linear, quadratic and interaction terms, respectively, x i and x j are the uncoded independent variables and is the error term. The maximum or minimum dlys levels (stationary point), which provided the optimum estimated response, were reported. The RIDGE MAX option was used to compute the estimated ridge of the optimum response when the stationary point was a saddle point [25]. A Microsoft Excel [26] spreadsheet was used to generate response surface depictions of the predicted responses. Validation of the model In both experimental phases, the Shapiro-Wilk test was used for testing normality of the residuals using PROC UNIVARIATE of SAS [23]. The data were found to be normally distributed before generating the regressions [27]. In Overall Phases the lack-of-fit test, which is the F statistic of the ratio of the lack-of-fit to the pure error sum of squares [23], was performed. This determined whether the additional terms were needed in the model, in case that replicated observations exist [22]. RESULTS Experiment Phases 1 and 2 The results of growth performance during the grower phase (Experimental Phase 1, 15 to 34d) showed that the null hypothesis is rejected for linear and quadratic terms of the model for 34-d BW, BW gain, feed intake, dlys intake, and FCR (P<0.05, Table 4). These results suggested that broilers fed higher dlys levels during grower phase showed significantly improve in BW and BW gain, decrease feed intake which improved FCR. Digestible lysine intake and dlys intake per BW gain increased linearly as higher dlys levels were fed (Table 4).

115 101 The results of processing characteristics during the grower phase showed that the null hypothesis is rejected for linear and quadratic terms of the model for breast meat weight and yield at 5% level (P<0.05, Table 5) and chilled carcass at 10% level (P<0.10). These results suggested that broilers fed higher dlys levels during grower phase showed significantly improve in breast meat weight and yield. However, the null hypothesis is rejected for only linear terms of the model for 35-d BW (live bird weight) and chilled carcass weight (P<0.05). These results suggested that there was significant difference in the mean 35-d BW and chilled carcass weight when broilers fed higher dlys levels. However, there was no significant difference in chilled carcass yield, and leg quarters weight and yield when broilers fed higher dlys levels (Table 5). The results of growth performance during the finisher phase (Experimental Phase 2, 35 to 49d) showed that the null hypothesis is rejected for linear and quadratic terms of the model for FCR (P<0.05, Table 6). The null hypothesis is rejected for only quadratic term of the model for BW gain. These results suggested that broilers fed higher dlys levels during finsher phase showed significantly improve in FCR while BW gain tended to decline. Digestible lysine intake and dlys intake per BW gain increased linearly as higher dlys levels were fed (Table 6). The results of processing characteristics during the finisher phase showed that the null hypothesis is fail to rejected for linear and quadratic terms of the model for most measurements at 5% level (P>0.05, Table 7). These results suggested that broilers fed higher dlys levels during finisher phase did not significantly improve processing characteristics. The dlys levels resulting in maximum responses and the calculated maximum responses using the quadratic model for both Experimental Phases are shown in Table 12 for BW gain, feed intake, FCR, live bird weight, carcass weight, carcass weight yield, breast meat and breast meat yield. However, only responses with significant quadratic effect were evaluated. During the

116 102 Experimental Phase 1, the quadratic response curves for 34-d BW, feed intake and FCR are shown in Figure 2, even though the dlys level minimizing FCR was out of the experimental range. This indicated the FCR didn t reach the minimum point in this experiment. The quadratic response curves for breast meat weight and yield are shown in Figure 3. During Experimental Phase 2, the quadratic response curve for BW gain and FCR are shown in Figure 4. The graphical method validates the adequacy of the quadratic model in describing the relationship between dlys levels and growth and processing variables measured. With five levels of dlys, it is possible to fit a forth-order model; however a second-order model is simpler and less complex in finding the maximum or minimum dlys level which optimizes responses. During Experimental Phase 1, there was no quadratic effect of dlys levels on dlys intake per BW gain, mortality, 35-d BW, carcass weight, carcass yield, leg quarter weight and yield. In addition, there was no linear effect of dlys level on mortality, carcass yield or leg quarters weight and yield. These conclusions suggest the increase in dlys levels have little or no effect on mortality, carcass yield, leg quarter weight and yield but highly effect on 34-d BW, BW gain, feed intake, dlys intake, FCR, 35-d BW, chilled carcass weight, and breast meat weight and yield. However, in case that linear effect is significant but the quadratic effect is not. Other models besides the quadratic model such as the linear or quadratic broken-line models can be applied to obtain the maximum response [28]. During Experimental Phase 1 (grower phase: 15 to 34 d), the dlys levels which maximized BW gain and feed intake were and %, respectively (Table 12). During Experimental Phase 2 (finisher phase: 35 to 49 d), the dlys levels which maximized BW gain and FCR were and %, respectively (Table 12).

117 103 Overall Phases: From 15 to 42 d From 15 to 42 d, the overall models were highly significant (P<0.001) for 42-d BW, BW gain, feed intake, dlys intake, dlys intake per BW gain and FCR (Table 8). At d of slaughter, dlys level was a significant contributor to variation in 43-d BW, breast meat weight, breast meat yield and leg quarter yield (Table 9) at the 5% level (P< 0.05). The linear and quadratic terms were significant at the 5% level for 42-d BW, BW gain, feed intake, dlys intake, and breast meat weight (Table 8 and 9), indicating both terms are needed in the models. However, the interaction terms in all responses were not significantly different, except leg quarter yield. This indicated that there were no interactions relative to feeding difference dlys levels in the grower or finisher phases. This means the response to dlys level in the grower phase was similar across the dlys levels in the finisher phases. The dlys levels which maximized growth performance and processing characteristics are shown in Table 12. The dlys levels, which maximized BW gain, were 1.113% for grower phase and 0.996% for finisher phase. The dlys levels, which maximized feed intake, were 1.025% for grower phase and 0.778% for finisher phase. Overall Phases: From 15 to 49 d From 15 to 49 d (Table 10), the overall models were highly significant (P<0.001) for feed intake, dlys intake, dlys intake per BW gain and FCR and significant at 1% level (P<0.01) for 49-d BW and BW gain. At d of slaughter, dlys level was a significant contributor to variation in 50 d BW, breast meat weight, and yield (Table 11) at the 5% level (P<0.05). The linear and quadratic terms were significant at 5% level for 49-d BW, BW gain, feed intake, dlys intake, dlys intake per BW gain, FCR and breast meat weight (Table 10 and 11), indicating both terms are needed in the models. However, the interaction terms in all responses were not significantly

118 104 different except for carcass yield. This indicated that there were no interactions relative to feeding difference dlys levels in the grower or finisher phases. This means the response to dlys level in the grower phase was similar across the dlys levels in the finisher phases. The dlys levels which maximized growth performance and processing characteristics are shown in Table 12. The dlys levels, which maximized BW gain, were 1.081% for grower phase and 0.972% for finisher phase (Figure 5). The dlys levels, which maximized feed intake, were 1.043% for grower phase and 0.802% for finisher phase (Figure 6). Validation of the model from 15 to 42 d and 15 to 49 d From 15 to 42d, the lack-of-fit test for 42-d BW, BW gain, feed intake, dlys intake, dlys intake per BW gain, FCR, 43-d fasted BW, carcass weight and yield, breast meat yield, and leg quarter weight and yield (Table 8 and 9) were not significant at the 5% level (P > 0.05), except breast meat weight which was barely significant at the 5% level (P = 0.044). Thus, additional terms should not be considered for all the response models [26]. The canonical analysis shows the response surface has a saddle point for FCR (Figure 7), carcass yield, and leg quarter yield, which shows there is no point at which the response is either maximized (Figures 5 and 6) or minimized. The RIDGE MAX option was used to explain the direction of the responses and factors. It showed the estimated ridge of the optimum FCR improved when the dlys levels of both the grower and finisher phases increased. The optimum carcass yield increased when the dlys levels of both the grower and finisher phases increased. The optimum leg quarter yield increased when the dlys levels of the grower phase decreased and the dlys levels of the finisher phase increased.

119 105 DISCUSSION This research diets were formulated by concerning a combination of protein and amino acid sources to provide as complete as possible in amino acids at a minimum total percentage of CP as mentioned by Fancher and Jensen [29]. This formulation technique, called Balanced Amino Acids [29] or Ideal Protein Concept [30] or Dietary Balanced Protein (BP) [8], provided the most economically feasible as surplus CP was eliminated while essential and nonessential AA were not deficient. When comparing this research with previous researches which were performed using Ross 708 male broilers, the dlys levels which maximized BW gain during grower phase of this research (15 to 34 d, Table 12, Figure 2) were slightly higher than those reported by Kidd et al. [7] (1.13 % dlys or 1.29 vs % total Lys) but lower than those reported by Lemme et al. [8] (1.29 vs % total Lys). This was because broilers of this research were fed starter diet that contained higher Lys content compared with Kidd et al. [7] (1.44 vs % total Lys), but lower Lys content compared with Lemme et al. [8] (1.44 vs % total Lys). Thus, this research concurred with the previous researched that the Ross 708 strain critically needs high amino acid density from 1 to 35d in order to maximize performance. On the other hand, BW gain was not affected by decreasing dlys density during the finisher phase (35 to 49 d) as there is not significant dlys linear effects of BW gain during that phase (Table 6). However, when broilers fed the higher dlys level, feed consumption declined and FCR improved (Table 6). This finding agreed with the previous research of Dozier et al. [10-11], which reported that high AA densities throughout the grow-out period provided the best feed conversion. Thus, the higher the dlys level, the better the feed efficiency was performed during the finisher phase. Carcass and breast meat yield was not affected by decreasing dlys density at 42 d, which also agreed with Dozier et al. [10]. In brief, this research

120 106 agreed with the previous researches that feeding high BP diets in early stages (1 to 35d) of broiler production are importantly improve growth performance and carcass characteristic while later in life (35 to 49d), feeding high BP diets only improved feed efficiency. CONCLUSIONS AND APPLICATIONS 1. The dlys level which maximized growth performances and processing characteristics depended upon the targeted grow-out days. 2. Raising broilers up to 35 d, (grower phase was 15 to 35d), the dlys level which maximized responses were 1.126, 0.984, 1.388, and 1.135% for BW gain, feed intake, FCR, breast meat weight, and breast meat yield, respectively. 3. Raising broilers up to 42 d (grower and finisher phases were 15 to 34d and 35 to 42d, respectively), the dlys levels which maximized responses were and 0.996% for BW gain, and 0.778% for feed intake, and % for live bird weight, and 0.994% for carcass weight, and 0.981% for breast meat weight and and 0.920% for breast meat yield, respectively. 4. Raising broilers up to 49 d (grower and finisher phases were 15 to 34d and 35 to 49d, respectively), the dlys levels which maximized responses were and 0.972% for BW gain, and 0.802% for feed intake, and % for live bird weight, and 0.988% for carcass weight, and 0.995% for breast meat weight and and 1.004% for breast meat yield, respectively. 5. There were no interactions for any of the responses, responses to dlys in the finisher phase were not dependent on what was fed in the grower phase.

121 107 Table 5.1 Composition of the experimental diets (%, as-fed basis) Ingredient Starter 4 Grower Low dlys 5 Diets Grower High dlys 6 Finisher Low dlys 7 Finisher High dlys 8 Ground yellow corn Soybean meal (dehulled) Meat and bone meal Poultry Fat Limestone Salt L-Lys HCl DL-Met L-Thr Defluorinated Phosphate S-Carb (Sodium Bicarbonate) Choline chloride Vitamin Premix Trace Mineral Premix Copper Sulfate Quantum 2,500 FTU / gram BMD Coban 90 (Monensin Na) Total Vitamin premix provided the following (per kilogram of diet): vitamin A (retinyl acetate), 5,511 IU; vitamin D 3 (cholecalciferol), 1,102 IU; vitamin E (DL-alpha tocopherol acetate), IU; vitamin K (menadione sodium bisulfite complex), 1.1 mg; vitamin B 1 (thiamin mononitrate), 2.2 mg; vitamin B 2 (riboflavin), 4.41 mg; vitamin B 6 (pyridoxine hydrochloride), 2.2 mg; vitamin B 12 (cyanocobalamin), 12.0 μg; niacin, mg; pantothenic acid (calcium pantothenate), mg; folic acid, 0.55 mg; biotin, 0.11 mg; choline chloride, mg; ethoxyquin 5.1%, mg. 2 Trace mineral premix provides the following (per kilogram of diet): Mn (manganese sulfate), mg; Zn (zinc sulfate), mg; Fe (ferrous sulfate), mg; Cu (copper sulfate), 3 mg; I (calcium iodate), 0.75 mg; Se (sodium selenite), 0.3 mg; Mg (magnesium oxide), 20.1 mg. 3 Quantum Phytase activities (FTU / Kg) were 500 for starter and all grower diets, and range from 390 to 295 for F1 to F5 diets. 4 Starter = starter diet (0 to 14d), 5 Grower Low dlys= grower (15 to 34d) diet with low dlys level (0.75%), 6 Grower High dlys= grower (15 to 34d) diet with high dlys level (1.25%), 7 Finisher Low dlys= finisher (35 to 42 and 49d) diet with low dlys level (0.69%), 8 Finisher High dlys= finisher (35 to 42 and 49d) diet with high dlys level (1.15%).

122 108 Table 5.2 The formulated and analyzed nutrient composition of the experimental diets and relative amino acid ratio Item Starter 6 Grower Low dlys 7 Diets Grower High dlys 8 Finisher Low dlys 9 Finisher High dlys 10 ME, kcal/kg 3, , ,164 3,064 3,164 3,042 3,208 3,086 3,208 3,086 CP, % Lys, % Met, % TSAA, % Thr, % Val, % Ile, % Trp, % Arg, % dlys, % dmet, % dtsaa, % dthr, % dval, % dile, % dtrp, % darg, % TEAA, % TNEAA, % TAA, % TEAA/TAA, % TNEAA/TAA, % Calcium, % Available phosphorus, % Total phosphorus, % Sodium, % Choline, mg/kg DEB, meq/kg Amino acid ratios dlys dmet/dlys dtsaa/dlys dthr/dlys dval/dlys dile/dlys dtrp/dlys darg/dlys Formulated values were in the first column of each diet. 2 TEAA = Lys+Met+Thr+Val+Ile+Trp+Arg+Leu+His+Phe.

123 3 TNEAA = Cys+Ala+Glu+Gly+Asp+Pro+Ser+Tyr. 4 TAA = TEAA+TNEAA. 5 Dietary electrolyte balance represents dietary Na + K Cl in meq/kg of diet. 6 Starter = starter diet (0 to 14d), 7 Grower Low dlys= grower (15 to 34d) diet with low dlys level (0.75%), 8 Grower High dlys= grower (15 to 34d) diet with high dlys level (1.25%), 9 Finisher Low dlys= finisher (35 to 42 and 49d) diet with low dlys level (0.69%), 10 Finisher High dlys = finisher (35 to 42 and 49d) diet with high dlys level (1.15%). 11Analyzed values were in the second column of each diet; amino acids and CP values (%, as-fed basis) were analyzed from Ajinomoto Heartland LLC [17]; ME and minerals values were analyzed from Midwest Laboratory [19]. 12 Analyzed digestible amino acids were calculated from the analyzed total amino acids. 109

124 110 Table 5.3 Analyzed nutrient composition of 9 treatments Treatment Digestible lysine level, % CP level, % Finisher Starter Grower Starter Grower 35 to 42d, 0 to 14d 15 to 34d 0 to 14d 15 to 34d 35 to 49d Finisher 35 to 42d, 35 to 49d

125 111 Table 5.4 Growth performance of Ross x Ross 708 male broilers fed different digestible lysine (dlys) levels during 15 to 34 d of age 1 dlys levels, % 34-d BW (kg) BW Gain (kg) Feed Intake (kg) dlys Intake (g/d) dlys Intake/BW FCR 2 Mortality Gain (g/kg) (%) P-value for contrasts 3 dlys linear effect <.0001 < <.0001 <.0001 < dlys quadratic effect < < Main effect means + SE of 5, 10, 18, 10 and 5 replications of 35 birds per pen of treatments contained 0.77, 0.85, 1.04, 1.23 and 1.30% dlys level, respectively. 2 Feed Conversion Ratio corrected for mortality (kg of feed per kg of BW gain). 3 Growth performance = dlys levels.

126 112 Table 5.5 Processing characteristics of Ross x Ross 708 male broilers fed different digestible lysine (dlys) levels during 15 to 34 d of age 1 dlys levels, % 35-d Fasted BW (kg) Chilled Carcass Breast Meat Leg quarter Weight (kg) Yield (%) Weight (kg) Yield (%) Weight (kg) Yield (%) P-value for contrasts 2 dlys linear effect < dlys quadratic effect Main effect means + SE of 5, 10, 18, 10 and 5 replications of 35 birds per pen of treatments contained 0.77, 0.85, 1.04, 1.23 and 1.30% dlys level, respectively. 2 Processing characteristics = dlys levels.

127 113 Table 5.6 Growth performance of Ross x Ross 708 male broilers fed different digestible lysine (dlys) levels during 35 to 49 d of age 1 dlys levels, % 49-d BW (kg) BW Gain (kg) Feed Intake (kg) dlys Intake (g/d) dlys Intake/BW FCR 2 Mortality Gain (g/kg) (%) P-value for contrasts 3 dlys linear effect <.0001 <.0001 <.0001 < dlys quadratic effect < Main effect means + SE of 5, 10, 18, 10 and 5 replications of 35 birds per pen of treatments contained 0.71, 0.78, 0.95, 1.13 and 1.20% dlys level, respectively. 2 Feed Conversion Ratio corrected for mortality (kg of feed per kg of BW gain). 3 Growth performance = dlys levels.

128 114 Table 5.7 Processing characteristics of Ross x Ross 708 male broilers fed different digestible lysine (dlys) levels during 35 to 49 d of age 1 dlys levels, % 50-d Fasted BW (kg) Chilled Carcass Breast Meat Leg quarters Weight (kg) Yield (%) Weight (kg) Yield (%) Weight (kg) Yield (%) P-value for contrasts 2 dlys linear effect dlys quadratic effect Main effect means + SE of 5, 10, 18, 10 and 5 replications of 35 birds per pen of treatments contained 0.71, 0.78, 0.95, 1.13 and 1.20% dlys level, respectively. 2 Processing characteristics = dlys levels.

129 115 Table 5.8 Growth performance of Ross x Ross 708 male broilers fed different digestible lysine level (dlys) levels during the grower (15-34 d) and finisher phases (35 to 42 d) 1 Treatment Grower dlys level (15-34d), % Finisher dlys level (35-42d), % 42-d BW (kg) 15 to 42 d BW Gain (kg) Feed Intake (kg) dlys Intake (g/d) dlys Intake/ BW Gain (g/kg) FCR 2 Mortality (%) Regression 3 P-values Linear terms <.0001 <.0001 < Quadratic terms Interaction terms Overall Model < <.0001 <.0001 < Lack of Fit R 2 of the overall model Root MSE Main effect means + SE of 5 replications of 35 birds per pen, except 8 replications of 35 birds per pen for treatment 5. 2 Feed Conversion Ratio corrected for mortality (kg of feed per kg of BW gain). 3 The quadratic response surface model is y = b 0 + b 1 x 1 + b 2 x 2 + b 11 x b 22 x 2 2 +b 12 x 1 x 2 where b 0 is intercept; b 1, b 2, b 11, b 22, b 12 are parameter coefficients; x 1 is the dlys level at d; x 2 is the dlys level at d; and y is growth performance.

130 116 Table 5.9 Processing characteristics of Ross x Ross 708 male broilers fed different digestible lysine level (dlys) levels during the grower (15-34d) and finisher phases (35 to 42d) 1 Treatment Grower dlys level (15-34d), % Finisher dlys level (35-42d), % 43-d Fasted BW (kg) Chilled Carcass Breast Meat Leg quarter Weight (kg) Yield (%) Weight (kg) Yield (%) Weight (kg) Yield (%) Regression 2 P-values Linear terms Quadratic terms Interaction terms Overall Model Lack of Fit R 2 of the overall model Root MSE Main effect means + SE of 5 replications of 35 birds per pen, except 8 replications of 35 birds per pen for treatment 5. 2 The quadratic response surface model is y = b 0 + b 1 x 1 + b 2 x 2 + b 11 x b 22 x 2 2 +b 12 x 1 x 2 where b 0 is intercept; b 1, b 2, b 11, b 22, b 12 are parameter coefficients; x 1 is the dlys level at d; x 2 is the dlys level at d; and y is growth performance.

131 117 Table 5.10 Growth performance of Ross x Ross 708 male broilers fed different digestible lysine level (dlys) levels during the grower (15-34d) and finisher phases (35 to 49d) 1 Treatment Grower dlys level (15-34d), % Finisher dlys level (35-49d), % 49-d BW (kg) BW Gain (kg) Feed Intake (kg) dlys Intake (g/d) 15 to 49 d dlys Intake/ BW Gain (g/kg) FCR 2 Mortality (%) Regression 3 P-values Linear terms <.0001 <.0001 < Quadratic terms < Interaction terms Overall Model <.0001 <.0001 <.0001 < Lack of Fit R 2 of the overall model Root MSE Main effect means + SE of 5 replications of 35 birds per pen, except 8 replications of 35 birds per pen for treatment 5. 2 Feed Conversion Ratio corrected for mortality (kg of feed per kg of BW gain). 3 The quadratic response surface model is y = b 0 + b 1 x 1 + b 2 x 2 + b 11 x b 22 x 2 2 +b 12 x 1 x 2 where b 0 is intercept; b 1, b 2, b 11, b 22, b 12 are parameter coefficients; x 1 is the dlys level at d; x 2 is the dlys level at d; and y is growth performance.

132 118 Table 5.11 Processing characteristics of Ross x Ross 708 male broilers fed different dlys levels during the grower (15-34d) and finisher phases (35 to 49d) 1 Treatment Grower dlys level (15-34d), % Finisher dlys level (35-49d), % 50-d Fasted BW (kg) Chilled Carcass Breast Meat Leg quarter Weight (kg) Yield (%) Weight (kg) Yield (%) Weight (kg) Yield (%) Regression 1 P-values Linear terms Quadratic terms Interaction terms Overall Model Lack of Fit R 2 of the overall model Root MSE Main effect means + SE of 5 replications of 35 birds per pen, except 8 replications of 35 birds per pen for treatment 5. 2 The quadratic response surface model is y = b 0 + b 1 x 1 + b 2 x 2 + b 11 x b 22 x 2 2 +b 12 x 1 x 2 where b 0 is intercept; b 1, b 2, b 11, b 22, b 12 are parameter coefficients; x 1 is the dlys level at d; x 2 is the dlys level at d; and y is processing characteristics.

133 119 Table 5.12 Digestible lysine (dlys) response levels of Ross x Ross 708 male broilers based on quadratic model 1 at 15 to 35 d and quadratic response surface model 2 at 15 to 42 d and 15 to 49 d Response From 15 to 34 d dlys level Maximum/ Minimum Response From 35 to 49 d dlys level Maximum/ Minimum Response From 15 to 42 d 15 to 34 d 35 to 42d Maximum dlys level dlys level Response dlys level From 15 to 49 d 15 to 34 d 35 to 49d dlys level Maximum Response Growth Performance Body weight gain Feed intake NA 3 NA Feed conversion ratio Processing characteristics Live bird weight NA NA NA NA Carcass weight NA NA NA NA Carcass yield NA NA NA NA Breast meat weight NA NA Breast meat yield NA NA The quadratic model is y = b 0 + b 1 x 1 + b 2 x 2 1 where b 0 is intercept; b 1, b 2 are parameter coefficients; x 1 is the digestible lysine level; and y is growth performance and processing characteristics. 2 The quadratic response surface model is y = b 0 + b 1 x 1 + b 2 x 2 + b 11 x b 22 x 2 2 +b 12 x 1 x 2 where b 0 is intercept; b 1, b 2, b 11, b 22, b 12 are parameter coefficients; x 1 is the dlys level at d; x 2 is the dlys level at d and d. 3 There values did not be able to specified because of the dlys quadratic effects were not significant difference at 5% level (P>0.05). 4,5 There was no point at which the responses of FCR and carcass yield were either maximum or minimum.

134 120 Figure 5.1 Nine diet combinations of grower dlys levels and finisher dlys levels used in Experimental Phase dlys levels of Finisher, % dlys levels of Grower, %

135 121 Figure 5.2 Growth performance of grower phase (15 to 34d) Body weight _ Feed intake.. Feed conversion ratio BW and FI, kg FCR Digestible lysine level, % Arrows indicate the dlys which maximized BW and feed intake.

136 122 Figure 5.3 Processing characteristics of grower phase (15 to 34d) Breast meat weight _ Breast meat yield Breast meat Weight, kg Breast meat yield, % Digestible lysine level, % Arrows indicate the dlys levels during the grower and finisher phases which maximized breast meat weight and yield.

137 123 Figure 5.4 Growth performance of finisher phase (35 to 49d) 1.70 Body weight gain _ FCR 2.10 Body weigt gain, kg FCR, kg:kg Digestible lysine level, % Arrows indicate the dlys levels during the grower and finisher phases which maximized BW gain and FCR.

138 124 Figure 5.5 Response surface BW gain from 15 to 49 d of age BWG, kg Arrows indicate the dlys levels during the grower and finisher phases which maximized BW gain

139 125 Figure 5.6 Response surface of feed intake from 15 to 49 d of age FI, kg Arrows indicate the dlys levels during the grower and finisher phases which maximized feed intake

140 126 Figure 5.7 Response surface of FCR from 15 to 49 d of age FCR, kg/kg The saddle shape indicated that there was no point at which the response of FCR was either maximum or minimum.

141 127 REFERENCES 1. Smith, E. R., and G. M. Pesti Influence of broiler strain cross and dietary protein on the performance of broilers. Poult. Sci. 77: Smith, E. R., G. M. Pesti, R. I. Bakalli, G. O. Ware, and J.F.M. Menten Further Studies on the Influence of Genotype and Dietary Protein on the Performance of Broilers. Poult. Sci. 77: Aviagen Ross 708 broiler performance objectives. Aviagen, Newbridge, Scotland. 4. NRC Nutrient Requirements of Poultry. 9th Rev. ed. Natl. Acad. Press, Washington, DC. 5. Warren, W.A., and J.L. Emmert Efficacy of Phase-Feeding in Supporting Growth Performance of Broiler Chicks during the Starter and Finisher Phases. Poult. Sci. 79: Kidd, M.T., C. D. McDaniel, S. L. Branton, E. R. Miller, B. B. Boren, and B. I. Fancher Increasing Amino Acid Density Improves Live Performance and Carcass Yields of Commercial Broilers. J. Appl. Poult. Res. 13: Kidd, M.T., A. Corzo, D. Hoehler, E.R. Miler, and W.A. Dozier Broiler responsiveness (Ross x 708) to diets varying in amino acid density. Poult. Sci. 84: Lemme, A., C. Kemp, E. Fleming, and C. Fisher Impact of increasing levels of balanced protein on biological and economic parameters of two different commercial broiler lines: Ross 308 and Ross 708. World Poultry Congress, Brisbane, Australia. 9. Corzo, A., M. T. Kidd, D. J. Burnham, E. R. Miller, S. L. Branton, and R. Gonzalez-Esquerra Dietary Amino Acid Density Effects on Growth and Carcass of Broilers Differing in Strain Cross and Sex. J. Appl. Poult. Res. 14:1.

142 DozierIII, W.A., M.T. Kidd, A. Corzo, J. Anderson and S.L. Branton, Growth performance, meat yield, and economic responses of broilers provided diets varying in amino acid density from thirty-six to fifty-nine days of age. Journal of Applied Poultry Research, 15: Dozier, W. A. III, M. T. Kidd, A. Corzo, J. Anderson, and S. L. Branton Dietary Amino Acid Responses of Mixed-Sex Broiler Chickens from Two to Four Kilograms. J. Appl. Poult. Res. 16: Pesti, G. M., and D. L. Fletcher The response of male broiler chickens to diets with various protein contents during the growing and finisher phases. Br. Poult. Sci. 25: Roush, W.B., R.G. Peterson, and G.H. Arscott An Application of Response Surface Methodology to Research in Poultry Nutrition. Poult. Sci. 58: Pesti G.M Protein and Energy Effects on Performance and Carcasses of Broilers. Page in Proc. of the 1983 GA. Nutr. Conf. for the feed industry, GA. 15. Edwards, H. M., Jr., and J. R. Veltmann, Jr The Role of Calcium and Phosphorus in the Etiology of Tibial Dyschondroplasia in Young Chicks. J. Nutr. 113(8): Liem A., G. M. Pesti, A. Atencio, and H. M. Edwards, Jr Experimental approach to optimize phytate phosphorus utilization by broiler chickens by addition of supplements. Poult. Sci. 88: Myers R.H., and Montgomery DC Response Surface Methodology: Process and Product Optimization Using Designed Experiments. 2 nd ed. John Wiley&Sons, Inc., New York, NY. 18. Ajinomoto Heartland LLC True Digestibility of Essential Amino Acids for Poultry.

143 129 Ajinomoto Heartland LLC, Chicago, IL. 19. AOAC International Official Methods of Analysis of AOAC International. 18th ed. Rev.2. AOAC Int., Gaithersburg, MD. 20. Midwest Laboratory B Street Omaha, Nebraska. 21. AB Vista Feed Ingredient. Suite 550, 1350 Timberlake Manor Parkway, Chesterfield, MO. 22. Cochran, W.G., and G.M. Cox Some methods for the study of response surfaces. Experimental Designs, 2nd ed. Wiley, New York, pp SAS User s Guide Version 9.1 ed. SAS Inst. Inc., Cary, NC. 24. Kuehl R.O Design of Experiments: Statistical Principles of Research Design and Analysis. 2 nd ed. Brooks/Cole Publishing Company, Pacific Grove, CA. 25. Freund R.J., and Little R.C SAS System for regression. 3 rd edition. SAS Institute Inc., Cary, NC. 26. Microsoft Excel Microsoft Corp., Redmond, WA. 27. Draper N.R., and Smith H Applied Regression Analysis. 2 nd Edition. John Wiley&Sons, Inc., New York, NY. 28. Vedenov D., and G. M. Pesti A comparison of methods of fitting several models to nutritional response data. Journal of Animal Science 86: Fancher, B., and L. Jensen Influence of varying dietary protein content while satisfying essential amino acid requirements upon broiler performance from three to six weeks of age. Poultry Sci 68: Emmert, J. and D. Baker Use of the ideal protein concept for precision formulation of amino acid levels in broiler diets. The Journal of Applied Poultry Research, 6(4):

144 130 CHAPTER 6 OPTIMIZING BROILERS DIETARY BALANCED PROTEIN AND ECONOMIC RETURNS 1 1 N. Sriperm, G. M. Pesti and M. E. Wetzstein. To be submitted to International Journal of Poultry Science

145 131 ABSTRACT The overriding goal of poultry-feed formulation should be to maximize profits and not to just minimize costs. Formulating diets to maximize profitability, rather than to maximize body weight gain or breast meat yield can positively impact the profitability of a broiler production system (Guevara, 2004; Eits, et al., 2005; Sterling, et al., 2005; Vedenov and Pesti, 2010). Similarly, formulating diets to reduce feed cost per ton, via lowering amino acid (AA) density, will not be the most cost-effective means of maximizing profitability if it adversely affects body weight (BW). The maximizing profitability concept with nutrition is to formulate considering both cost and body weight or meat yield relationships and not necessarily the maximum BW or minimum AA density. This study demonstrated alternative models to generate dose-response curves for a modern broiler strain and calculated the economically efficient feeding levels (or optimal nutrient density) during the grower and finisher phases. The Cobb-Douglas production function (CD), quadratic polynomial (QP), ascending quadratic with plateau (QPP) and Chen- Clayton (CC) models were fitted to the data from a growth trial with Ross x Ross 708 male broilers for BW and cumulative feed intake as functions of digestible lysine (dlys) levels in both phases at 49 d of age. The CD at 49 d can be explained as and CFI. The QP can be explained as BW = and CFI = The QPP can be explained as a quadratic model with a plateau beyond the maximum response point. The CC model can be explained as BW = and CFI =. The profit function can then provide the maximum profit, under any pricing scenario, as well as the dlys levels which maximize profit. The variation in feed ingredient prices (corn and soybean meal) and

146 132 broiler prices were used in profit functions to demonstrate the prediction capability of the model and elasticity. The model suggested that when broiler price increased, bigger birds need to be produced and higher dlys levels need to be applied to maximize profits. When feed ingredient prices increased, the maximum profit condition was to reduce feed consumption by increasing dlys level in the diet in order to optimize bird performance. Statistically, the CD and CC did not fit the data better than the QP and QPP according to the adjusted multiple coefficients of determination (Adj.R 2 ). However, theoretically, the CD, CC and QPP are better than QP model because QP is hyperbolic and reachs the maximum point and then declines or becomes toxic and reduce bird performance. The CD and CC closely resemble a plateau (asymototic) and QPP reaches a plateau. However, CC provided less flexible in estimating BW compared with CD. Thus, the results from this study demonstrate the potential of using the CD and QPP for profit maximization analysis in decision making of broiler feed formulation. Feeding broilers at optimum dlys levels based on these models showed that profitability in all scenarios were higher than the profit using the recommended requirement levels by Ross (Aviagen, 2007). INTRODUCTION In broiler production, knowledge of the most economically efficient feeding levels based on current production costs and market prices is very important in today s competitive marketplace. For determining economically efficient feeding levels, data from growth performance studies are required. Ideally, these data should be generated by individual companies using their strain, gender, feed formulation and manufacturing procedures and growing environment under different nutrient levels. With the availability of such information, a mathematical Production Function can be generated to determine the profit function which can be used to calculate the maximum profitability of raising broilers.

147 133 As an aid for developing a production function, a variety of growth functions including Gompertz, Modified Gompertz, Richards, Lopez, Michaelis-Menten, Logistic, Logistic Nonlinear Mixed Effect, and Spine Linear Regression models have been reported as models for consideration as poultry production functions (Talpaz et al., 1986; Lopez et al., 2000; Aggrey, 2002; Zuidhof, 2005; Aggrey, 2009; Darmani Kuhi et al., 2010). Lopez et al. (2000) stated that growth functions are frequently used in empirical works. The form of the function is chosen by its ability to fit the data. These growth functions have been commonly used to estimate the optimum growth, nutrient responses and time required to produce birds of desired weights. Talpaz et al. (1986) applied Gompertz models to estimate BW, body fat and feather fractions over time (days). In general, the functional form of the growth function used for final weight or mature size is determined as a function of time or age of the bird. Furthermore, production functions can be generated by using nutrient dose-response data from titration experiments. The data are normally used to generate a nutrient dose-response curve that determined the nutrient requirement, the nutrient level that maximized the response. In general, nutrient dose-response curves, such as Broken-Line Linear and Broken-Line Quadratic models (Vedenov and Pesti, 2007), Saturation Kinetics (Morgan et al., 1975; Phillips, 1982; Pesti et al., 2009), Logistic (Phillips, 1982; Pesti et al., 2009), Michaelis-Menten (Morgan et al., 1975), Exponential (Eits et al., 2005; Liebert, 2008) and Monomolecular (Kuhi et al., 2009) models, were commonly used in fitting responses with one input variable. These nutrient doseresponse curves have a similarity as they tend to plateau or have a limiting response at high nutrient intakes (Broken-Line Linear and Broken-Line Quadratic models) or approach an asymptotic (Saturation Kinetics, Michaelis-Menten, Exponential and Monomolecular models, Morgan et al., 1975). Clark et al. (1982) applied a Reading model to fit protein response curves

148 134 of their experimental data. The model explained the dietary protein requirement as a function of broilers mean body weight and live weight gain. The model described that a group of broilers showed diminishing increments of responses as the dietary protein increased and reached a plateau when broilers were adequately supplied with the limiting nutrient. However, these models with two or more input variables have not yet been developed. Phillips (1982) stated that the relationship between the response and nutrient intake of animals is not linear. Thus, he found that Saturation Kinetic and Logistic models fitted his experimental data quite well. This finding was confirmed by a study by Pesti et al. (2009). Robbins et al. (2006) explained that the linear broken-line model can be used to fit nutrient dose-response data, which assumed that the response to nutrient dose is linear. In general, the rate of change with nutrient dose-response level decreases as the nutrient dose-response level moves toward the broiler s maximum requirement. Thus, other models that include a non-linear component must be considered. Pesti et al. (2009) reported that a Broken-Line Quadratic model was clearly the best approach in finding nutrient requirements for their data. However, they stated that the best feeding level of a nutrient should be defined by the law of diminishing returns, where the response increases at a decreasing rate with respect to increasing nutrient level, to evaluate optimum economic levels because dietary protein, amino acids (AA), and energy follow diminishing returns phenomena (Almquist, 1953; Kuhi et al., 2009; Pesti et al. 2009), rather than just the fixed requirement level which maximizes biological performance. Thus, the ascending quadratic polynomial with plateau (QPP) was developed within this study. A quadratic polynomial (QP) model has been often applied to nutrient dose-response data to estimate maximum BW and feed intake because of its function contained two or more nutritional input variables (Miller et al., 1986; Pesti et al., 1986; Gonzalez-Alcorta et al., 1994;

149 135 Guevara, 2004). However, QP responses increase at a decreasing rate until reaching a single maximum, not a plateau out, and further increase nutrient intake causes the curve to trend downward, which results in a reduced predicted response. Morgan et al. (1975) stated that quadratic models can only estimate within narrow ranges of nutrient intake; moreover, higher organisms catalytic and transport processes are known to follow saturation kinetic patterns. This implies that QP might not be the best model to fit nutrient dose-response data because most nutritional responses are considered to contain a plateau which separates the requirement level at maximum response from those levels that are toxic (Pesti et al., 2009). However, some researchers observed a decline in response at very high levels of a nutrient (Lamberson and Firman, 2002; Robbins et al., 2006). Lamberson and Firman (2002) proposed a segmented regression or breakpoint analysis over a QP for estimating nutrient requirements because segmented regression provided more precise estimates compared with quadratic regression. In addition, segmented regression was unbiased when experimental diets are not uniformly distributed around the true mean, while QP was biased. Unfortunately, the segmented regression with two or more variables has not been reported in the previous literature. Thus, it cannot be applied to experimental data with two input variables of this study. Therefore, a Cobb-Douglas production function (CD), a nonlinear deterministic model, was proposed in this study as an alternative model for two or more input variables. The potentials of CD are 1) the CD is asymptotic; 2) the CD follows the law of diminishing returns similar to a Monomolecular (Kuhi et al., 2009), Satuation Kinetic and Logistic models (Pesti et al. 2009); and 3) it is widely used by many researchers (Heady, 1957; Walter, 1963; Zellner, 1966; Kmenta, 1967; Douglas, 1976; Kennedy et al., 1976; Romero et al., 2009; Zuidhof, 2009). In broiler production, Heady (1957) applied CD function to determine the least-cost ration for different weight ranges

150 136 based on the ration of corn and soybean oilmeal (two input variables). Kennedy et al., (1976) used the CD form to determine broiler production models to estimate daily weight gain or daily energy intake as a function of ages, phases, BW and energy density (three input variables), and mortality as a function of age and BW (two input variables). Zuidhof (2009) applied a nonlinear model based on a Cobb-Douglas form and a stepwise procedure to estimate feed intake as a function of BW, ME, Lys, gain, and sex (five input variables). These factors provided reasonable accuracy of predicted ME. Romero et al., (2009) studied metabolizable energy utilization in broiler breeder hens and applied CD function to the interaction between BW and average daily gain or egg mass. A general form of the CD is where K and L are two inputs (capital and labor, respectively), thus the function can be applied with two or more inputs as well as the QP. The functional form of the CD can be exposed as a linear in logs: (Beattie and Taylor, 1985). Sun (1999) proposed several nonlinear models with two variable inputs (Chen-Clayton, Modified-Chung-Pfost, Modified-Halsey, Modified-Henderson, Modified-Oswin, and Strohman- Yoerger). These models are commonly used to describe the equilibrium moisture content and equilibrium relative humidity sorption behavior of grain and oilseed (Sun, D.W., 1999). The proposed models by Sun (1999) were considered in this study, only the models that fit this experimental data better than the QP or QPP or CD were reported. The purposes of this study were 1) to develop nonlinear deterministic models as an alternative procedure for two inputs to estimate AA requirement for the modern broiler strain (Ross x Ross 708); 2) To apply these alternative models to calculate the economically efficient feeding levels during grower and finisher phases; 3) To compare the economically efficient feeding levels between selected models at various simulated feed ingredient and broiler market prices; and 4) To

151 137 compare the economically efficient feeding levels based on these models with the recommended requirement levels by Ross (Aviagen, 2007). While other nonlinear models (e.g. Saturation Kinetic) were applied to a profit maximization procedure with one input variable, the results from this study demonstrated a potential for using the other models besides QP to estimate production functions. This finding can be used in decision making in broiler feed formulation and how it can be applied to a profit maximization procedure. MATERIALS AND METHODS Determination of Production Functions Data of a dose-responses trial with Ross x Ross 708 male broilers were used (Sriperm and Pesti, 2011). The study was conducted to evaluate the digestible lysine (dlys) responses to bird performance (body weights (BW) and cumulative feed consumption (CFI)) during grower (15 to 34 d) and finisher (35 to 49 d) phases. The dlys levels were maintained in a constant ratio to other essential AAs (Dietary Balanced Protein) across five experimental diets for each phase according to Ajinomoto Heartland LLC (2009) recommendations. The five treatment diets for each phase were formulated to contain the constant ME, sodium, calcium and phosphorus levels. There were 9 treatment combinations of dlys and crude protein levels for grower and finisher phases according to the central composite rotatable design used in the experiment (Table 1). The BW data of broilers fed the higher three dlys level treatment diets of each phase were evaluated using the Contrast procedure of PROC GLM (SAS, 2004) to determine whether toxicity exists at the higher dlys levels (1.04 vs or 1.30% during the grower and 0.95 vs or 1.20% during the finisher phases). The data of bird performance at the end of day 42 and 49 were used to fit to the CD and QP models using PROC REG of SAS (2004). The non-linear α function of CD,, can be explained as a linear in log: α

152 138. The QP model can be explained as α α α, where q is a production function for BW or CFI, A, α, and are coefficients of the model, is the residuals, x 1 is a set of dlys levels fed during the grower phase and x 2 is a set of dlys levels fed during the finisher phase, d is equal to 1 if broilers were slaughtered at 42 d and equal to 0 otherwise. The QPP can be explained as: q = A + α α α if and q max = K if or where K is the maximum response at and, x 1 is a set of dlys levels fed during the grower phase and x 2 is a set of dlys levels fed during the finisher phase. The maximum was calculated by differentiating the regression equation and solving for the level of gain at which the first derivative was equal to zero. Nonlinear models according to Sun (1999) were used to fit the experimental data using PROC NLIN of SAS (2004). The results showed that only the Chen-Clayton model (CC) fitted the data better than the CD. Thus, the CC was included in the study. The CC model can be explained as Y = where a, b, c and d are coefficients of the model. Adjusted multiple coefficients of determination (adj. R 2 ) and sum of squared residuals (SSR) were used to measure the prediction reliability of the models (Mendenhall and Sincich, 2003). The residual analysis of nonlinear functions during the grower and finisher phases were reported. The plot of residuals versus predicted values was fitted to ensure there was no abnormality (Draper and Smith, 1981). Linear Programming Procedure Linear programming (LP) was used to generate a least-cost feed formula when ingredient prices were changed. In general, the LP model was explained by minimizing the objective function of, subject to AX B, X 0 where C is a (n 1) matrix of unit

153 139 prices of the objective function, X is a (n 1) matrix of the quantities of feed ingredients, A is a (m n) matrix of technical coefficients (e.g. protein and AA compositions of feed), B is (m 1) matrix of nutrient requirements for the specified growth performance (e.g. ME). Profit Maximizing Procedure over Feed Cost using Non-Linear Programming Procedure The relationship between the cost and revenue functions can be explained by the profit function using the production function for BW and CFI. Thus, profit equals revenue from output minus costs of using input. Equation (1) can be explained as maximizing profit by minimizing cost for a given BW and then selecting the optimal BW given the broiler price. This implies a firm which maximizes profit also maximizes revenue and minimizes costs for a given level of BW (Coelli, 2005). The BW and CFI functions were non-linear functions of dlys levels during grower and finisher phases using functional forms of selected models. where was broiler price ($/kg), r was feed prices for grower and finisher diets ($/kg), CFI was cumulative feed intake for grower and finisher phases (kg), which average feed consumption among treatments of 46 and 45%, respectively, of the total feed consumption up to day 49, x 1 is a set of dlys levels during the grower phase and x 2 is a set of dlys levels during the finisher phase. Price Scenarios Simulation The broiler prices were ranged from $1.40 to $1.71 per kg live bird or $1.50 to $2.25 per kg carcass at 76% carcass yield. Feed prices were simulated based on changes in major feed ingredient prices: corn and soybean meal (SBM) prices ranged from $186 to $433 per metric ton and $255 to $594 per metric ton, respectively. At the given set of simulated prices, the profit maximizing input-output combination over feed cost during grower and finisher phases can be calculated using profit function (Equation 1). The optimum feeding levels or levels which

154 140 maximized profit (dlys levels) were estimated by taking the first derivative of the profit function and set it equal to zero. This can also be done by using Excel (Microsoft, Seattle, WA) with Solver nonlinear programming (Frontline System, Inc., 1999) and Microsoft Visual Basic (Microsoft, Seattle, WA). The generated spreadsheet in Excel using Microsoft Visual Basic (Microsoft, Seattle, WA) was used to compute fifty combinations of selected price scenarios in order to investigate the impact of net returns at various feed costs and broiler prices RESULTS AND DISCUSSION The BW of broilers fed the higher three dlys levels of each phase were compared using contrast procedure of SAS (2004) as shown in Table 2. The BW of broilers fed the higher three dlys level diets were not significantly different (P > 0.05) for either phase (1.04 vs or 1.30% during the grower and 0.95 vs or 1.20% during the finisher phases). This indicated that no toxicity existed at higher dlys levels in this experiment, so the model should plateau and not have a single maximum. The fitted results of experimental data with CD, QP, QPP and CC models for BW and CFI were shown in Table 3. The adj. R 2 of CD and QP models for predicting BW and CFI were similar (24.3 vs % for BW and 42.6 vs % for CFI for CD and QP, respectively) which indicated that either CD or QP models provided similar predictions. However, the adj. R 2 of QPP for predicting BW and CFI were the highest among three models (39.0 and 55.8 %, respectively). The CC fit this experimental data slightly better than the CD but not better than QP and QPP. In addition, the small SSR model shows the best fit to the data. The SSR was smallest for the QPP in predicting BW and CFI compared with CD, QP and CC. However, the SSR values of the four models were quite similar. The estimated responses of BW and CFI at the maximum dlys levels of QP and QPP were also shown in Table 3. The QPP has the same ascending quadratic portion as the

155 141 QP but plateau after the maximum response was observed, thus both models provided the same maximum values. The estimated dlys requirements, which maximized BW, based on the QP and QPP models were 1.08 and 0.97% during grower and finisher phases, respectively, up to 49 d while the CD and CC could not provide the requirements, per se, (Figures 1 and 2) because it was accessing a maximum response but never reaching it. Thus, CD and CC are asymptotic as similar to Saturation Kinetics, Michaelis-Menten, Exponential and Monomolecular models. Moreover, further increase dlys levels using the CD, QPP and CC models do not predict toxicity, as does the QP (Figure 3). Thus, the CD and CC are not able to define requirements but adequately represented biological responses. Tables 4 and 5 showed the residual analysis for BW and feed intake upto 49 d of age, respectively. The QP and QPP models have similar residual values and smaller compared with CD and CC, especially at the requirement where dlys level were at 1.04 and 0.95% during grower and finisher phases, respectively. Thus, the QP and QPP models statistically fit this experimental data better than the CD and CC. Table 6 demonstrated the optimization analysis of changing feed ingredients (corn and SBM) and broiler prices on feeding levels of dlys during grower and finisher phases using the CD production function. The results explained economic theory that if the output price (broiler price) increased, production should be increased (produce bigger broilers), consequently, the amount of input should be increased (feeding higher dlys levels). In the other words, the model explained that when broiler price increased, bigger birds need to be produced which implied higher dlys levels need to be fed to the birds. Table 7 demonstrated the same optimization analysis using both QP and QPP models because the optimum condition will not occur beyond the maximum response (plateau part of QPP), as profit declined when an additional cost occurred

156 142 but revenue did not change. However, the estimated dlys levels of both phases using the QP and QPP models showed narrower ranges (0.77 to 1.27 % and 0.83 to 1.06 % dlys during grower and finisher phases, respectively) compared with those using CD (0.77 to 1.30 % and 0.71 to 1.20 % dlys during grower and finisher phases, respectively) and the estimated BWs of QP and QPP were, in general, higher than those of the CD. The results agreed with Morgan et al. (1975) that quadratic models can only estimate within narrow ranges of nutrient intake. The estimated feed consumptions which maximized profit were lower when feed ingredient prices increased. This implied that when feed prices increased, the maximum profit condition was to reduce feed consumption while increasing dlys level so that to achieve the optimum solution. At the high feed ingredient prices, the model suggested that feeding higher dlys levels that would reduce FCR. Table 8 demonstrated the optimization analysis using Chen-Clayton functions. The estimated dlys level of both phases using the model ranged from 0.77 to 1.30 % during the grower phase and 0.71 to 1.20 % during the finisher phases, which were similar to the results of CD. The estimated BWs of CC were, in general, in between the results of QP and CD. However, the CC models (Figure 2) provided less flexible in estimating BW compared with the CD (Figure 1). The maximum profit condition (or marginal profit) was observed when the revenue is maximized and the cost was minimized for a given level of BW (or marginal revenue equal to marginal cost). When feed ingredient prices increased, the profit declined. Figure 4 illustrates the dlys levels that maximized profit during grower and finisher phases based on changed prices of corn, SBM and broiler prices at $309 and $424 per metric ton, and $1.42 per kg live bird, respectively and using the CD production functions. The dlys levels that maximized profit were 1.09 and 0.90 % during the grower and finisher phases, respectively. Notice that these levels

157 143 were higher than the levels which maximized growth performance during grower phase but lower during finisher phase (1.08 and 0.97%, respectively) based on the QP model. This can be explained by the steeper ascent of the CD model during the grower phase compared with finisher phase. The calculated profits using Ross 708 nutrient recommendation (Aviagen, 2007) at 1.10 % dlys during grower phase and 0.97% during finisher phase showed that profitability in all scenarios were lower than the maximum profit using the models. This showed that the unnecessary cost occurred when following the technical requirement. Because the breeder guide makes static recommendations, this level might produce the maximum BW and the best feed efficiency, but it might not be the most profitable. The economic efficiency (or cost of feed per gain BW) suggested that when the corn price is high and the SBM price is low, the higher nutrient density should be considered in order to reduce feed consumption. Variation in broiler prices has a higher impact on net returns compared with variation in feed costs because of a steeper slope as shown in Figure 5. Even though, the fitted CD and CC are not statistically better than the QP based on the adj. R 2 ; however, CD and CC models are theoretically better than the QP model because the QP is hyperbolic, which reaches the maximum point and then decline (Pesti et al., 2009), while the CD and CC closely resemble a plateau (asymototic). Moreover, higher organisms catalytic and transport processes are known to follow a kinetic pattern (Morgan et al., 1975), thus CD and CC should be considered over QP in fitting a biological response. The advantage and disadvantage of the four models and other growth functions were listed in Table 9. In short, the CD is quite easy to estimate because it is linear in parameters in logarithmic form (Beattie and Taylor, 1985). Thus, it is easier to estimate compared with the CC. The CD has advantages over a broken-line model. The broken-line model always provides the optimal solution at the break point and does not

158 144 change after the plateau (Vedenov and Pesti, 2007) while the CD does not contain a plateau. Moreover, the CD is more flexible than a broken-line model as the slopes of the model changes across different levels of nutrient. For this reason, the CD is likely better than the QPP. However, for this data set the CD did not provide a better fit compared with QP and QPP, other data set might possibly fit the CD better than QP and QPP. Much research exists to support the belief that nutrient dose-response relationships reveal a saturation phenomenon (Almquist, 1953; Schulz, 1973; Morgan et al., 1975; Clark et al., 1982; Kuhi et al., 2009; Pesti et al., 2009). The results from this study demonstrated potential for applying the form of CD for two input variables of nutrient dose-response data, which never have been applied to two input variables in the previous literature (Heady, 1957; Kennedy et al., 1976; Romero et al., 2009; Zuidhof, 2009).The CD follows the law of diminishing returns as is similar to Monomolecular (Kuhi et al., 2009), Satuation Kinetic, and Logistic models(morgan et al., 1975; Phillips, 1982; Pesti et al., 2009). The CD has an advantage over the growth functions (Talpaz et al., 1986; Lopez et al., 2000; Aggrey, 2002; Zuidhof, 2005; Aggrey, 2009; Darmani Kuhi et al., 2010, Table 9) because it can be applied to two or more input variables, while growth functions have been commonly used to estimate the optimum growth, nutrient responses as a function of time (only one input variable). The QPP was a modified version of QP, developed by setting a constraint where the response was maximized. The QPP also followed the law of diminishing returns and it is asymptotic. From this study, the QPP provided similar fit compared with QP, but both models were better than CD. In summary, using this experimental data, the CD can adequately predict the maximum BW, based upon the dlys levels fed during the grower and finisher phases, while QPP provided the best fit among three models.

159 145 The price scenarios used here were historical values for study purposes, not to identify a specific profit or feeding levels for the future. As profit maximizing input levels of broiler production may change tomorrow based on new price scenarios and bird performance, this study demonstrated that the nutrient levels which maximize profit needs to be considered as dynamic levels, not static levels as suggested by breeder guides. The concept of requirement is simply defined, but less effective as it does not consider changing in economic environment.

160 146 Table 6.1 Analyzed nutrient composition of the experimental diets Digestible lysine level a, % CP level, % Treatment Starter 0 to 14d Grower 15 to 34d Finisher 35 to 49d Starter 0 to 14d Grower 15 to 34d Finisher 35 to 49d a Analyzed digestible lysine levels were calculated from the analyzed total lysine levels.

161 147 Table 6.2 Comparison of the experimental results for BW using the Contrast procedure of SAS (2004) Experiment data at 34 d Experiment data at 49 d Digestible Lysine Level of Digestible Lysine Level of Pr > F Treatment Diets, % Treatment Diets, % Pr > F 1.04 vs vs vs vs vs 1.23, vs. 1.13,

162 148 Table 6.3 Regression coefficients and statistics for BW and cumulative feed intake (CFI) of broilers at 49 d of age Variable a Cobb-Douglas b Quadratic Models c Ascending Quadratic with Plateau Chen-Clayton f Variable BW CFI BW CFI BW CFI BW CFI A Coefficient 1.158*** 1.597*** a Standard Error X 1 Coefficient 0.068*** ** 8.369*** 4.055** 8.369*** b Standard Error X 2 Coefficient *** 3.655** ** c Standard Error X 1 Coefficient ** *** ** *** d Standard Error X 2 Coefficient ** * ** * Standard Error X 1 X 2 Coefficient Standard Error D Coefficient 0.233*** 0.268*** Standard Error Adj.R SSE d Pr > F < < < < < Maximum dlys Level (%) e Grower Phase Finisher Phase Maximum Responses (kg) a A is intercept, x 1 is dlys levels at grower phase, and x 2 is dlys levels at finisher phase. b Cobb-Douglas Production Function of estimated BW or CFI= e A x 1 α x 2 e d* c Quadratic Polynomial of estimated BW or CFI= A + d Sum of squared residuals, a smaller SSR indicated a better fit of the model to the data.

163 149 e Maximum dlys Levels during grower and finisher were calculated by solving the first derivative of the response function at which it was equal to zero. f Chen-Clayton BW or CFI (x 1,x 2 ) = where a, b, c and d were coefficients of the model. * Statistically significant at the 0.10 level. ** Statistically significant at the 0.05 level. *** Statistically significant at the level.

164 150 Table 6.4 Residuals analysis of the Cobb - Douglas production function, quadratic polynomial, ascending quadratic with plateau and Chen-Clayton models for BW of broilers at 49 d of age Grower dlys level (%) Finisher dlys level (%) Actual BW (kg) Cobb-Douglas Predicted BW (kg) Quadratic Model Ascending Quadratic with Plateau Chen-Clayton Cobb-Douglas Quadratic Model Residuals a Ascending Quadratic with Plateau Chen-Clayton a the difference between actual and predicted value.

165 151 Table 6.5 Residuals analysis of the Cobb - Douglas production function, quadratic polynomial, ascending quadratic with plateau and Chen-Clayton models for feed intake of broilers at 49 d of age Grower dlys level (%) Finisher dlys level (%) Actual CFI (kg) Cobb-Douglas Predicted CFI (kg) Quadratic Model Ascending Quadratic with Plateau Chen- Clayton Cobb-Douglas Residuals a Quadratic Model Ascending Quadratic with Plateau Chen- Clayton a the difference between actual and predicted value.

166 152 Table 6.6 Optimization analysis of changing feed ingredient and broiler prices on dlys levels during the grower (15 to 34 d) and finisher (35 to 49 d) phases using Cobb-Douglas production functions Broiler Grower Finisher Returns Ross Cost of Economic Feed Corn SBM Carcass dlys dlys Live BW Revenue over Net Wrong Efficiency FI (kg) Cost ($/MT) ($/MT) Price level level (LBW,kg) ($/bird) Feedcost Returns Decision ($ feed/kg ($/bird) ($/kg) (%) (%) ($/bird) ($/bird) a ($/bird) b LBW) a Calculated profit from Ross 708 recommended feeding program of 1.10% dlys during 11 to 24 d and 0.97% dlys during d 25 to market. b The difference between feeding the dlys levels which maximized profit and the dlys levels which recommended by the breeder.

167 153 Table 6.7 Optimization analysis of changing feed ingredient and broiler prices on dlys levels during the grower (15 to 34 d) and finisher (35 to 49 d) phases using Quadratic Polynomial and Ascending Quadratic with Plateau functions Broiler Grower Finisher Returns Ross Cost of Economic Feed Corn SBM Carcass dlys dlys Live BW Revenue over Net Wrong Efficiency FI (kg) Cost ($/MT) ($/MT) Price level level (LBW,kg) ($/bird) Feedcost Returns Decision ($ feed/kg ($/bird) ($/kg) (%) (%) ($/bird) ($/bird) a ($/bird) b LBW) a Calculated profit from Ross708 recommended feeding program of 1.10% dlys during 11 to 24 d and 0.97% dlys during d 25 to market. b The difference between feeding the dlys levels which maximized profit and the dlys levels which recommended by the breeder.

168 154 Table 6.8 Optimization analysis of changing feed ingredient and broiler prices on dlys levels during the grower (15 to 34 d) and finisher (35 to 49 d) phases using Chen-Clayton functions Broiler Grower Finisher Returns Ross Cost of Economic Feed Corn SBM Carcass dlys dlys Live BW Revenue over Net Wrong Efficiency FI (kg) Cost ($/MT) ($/MT) Price level level (LBW,kg) ($/bird) Feedcost Returns Decision ($ feed/kg ($/bird) ($/kg) (%) (%) ($/bird) ($/bird) a ($/bird) b LBW) a Calculated profit from Ross708 recommended feeding program of 1.10% dlys during 11 to 24 d and 0.97% dlys during d 25 to market. b The difference between feeding the dlys levels which maximized profit and the dlys levels which recommended by the breeder.

169 155 Table 6.9 Advantage and disadvantage of applying models to nutrient dose-response data Models Ascending Cobb- Quadratic Quadratic with Douglas Polynomial Plateau Chen-Clayton Other Models a Fit two or more variables Yes Yes Yes Two variables No Approach an asymptotic Yes No Yes Yes Yes Difficulty to fit the model Easy Easy Moderate Difficult Moderate Mathematical form of the Addition with Multiplication Addition model for two linear constraints Complex N/A variables a Gompertz, Modified Gompertz, Richards, Lopez, Michaelis-Menten, Logistic, Saturation Kinetics, Exponential and Monomolecular models.

170 156 Figure 6.1 Three-dimensional surfaces plot of predicted values of BW (kg) from the growth function of broilers fed diets with different levels of dlys during the grower and finisher phases at 49 d of age using the Cobb-Douglas production function BW, kg

171 157 Figure 6.2 Three-dimensional surfaces plot of predicted values of BW (kg) from the growth function of broilers fed diets with different levels of dlys during the grower and finisher phases at 49 d of age using the Chen-Clayton model BW, kg

172 158 Figure 6.3 Three-dimensional surfaces plot of predicted values of BW (kg) from the growth function of broilers fed diets with different levels of dlys during the grower and finisher phases at 49 d of age using the Quadratic Polynomial function dlys level during the grower phase = 1.08% dlys level during the finisher phase = 0.97% BW, kg The arrow indicated the dlys levels which maximized profit during the grower and finisher phases.

173 159 Figure 6.4 Three-dimensional surfaces plot of profit maximizing over feed cost from the profit function using Cobb-Douglas production functions for broilers fed diets with different levels of dlys during the grower and finisher phases up to 49 d of age, based on scenario when corn, SBM, and broiler prices were at $309 and $424 per metric ton, and $1.42 per kg live bird, respectively dlys level during the grower phase = 1.09% dlys level during the finisher phase = 0.90% Profit, $/bird The arrow indicated the dlys levels which maximized profit during the grower and finisher phases.