Estimation of the net energy value of barley for finishing beef steers

Abstract

The objective of this study was to identify barley grain characteristics measured by laboratory procedures that could be used to predict barley energy content for finishing beef steers. Twenty-eight different barley genotypes were evaluated including 18 cultivars and 10 experimental lines. Laboratory analysis of barley samples included bulk density, particle size, N, ADF, starch, and ISDMD (in situ DM disappearance after 3 h of ruminal incubation). Animal performance data (BW, DMI, ADG, steer NE_m, and NE_g requirements) were collected from 26 feedlot experiments conducted in Montana and Idaho during a 10-yr period and were used to estimate barley NE_m and NE_g content. A total of 80 experimental units were available with each experimental unit being a diet mean from an individual feedlot experiment. Fifty-eight of the 80 experimental units were randomly selected and used in the development data set and the remaining 22 experimental units were used in the validation data set. Forward, backward, and stepwise selection methods were used to identify variables to be included in regression equations for NE_m using PROC REG of SAS. Barley samples in the model development data set represented a wide range in concentrations (DM basis): N (1.6% to 2.8%), ISDMD (25.7% to 58.7%), ADF (3.6% to 8.0%), starch (44.1% to 62.4%), particle size (1,100 to 2,814 µm), and bulk density (50.8 to 69.4 kg/hL). The barley grain characteristics of particle size, ISDMD, starch, and ADF were the most important variables in six successful models (R² = 0.48 to 0.60; P = 0.001). The six prediction equations gave mean predicted values for NE_m ranging from 1.99 to 2.05 Mcal/kg (average 2.04 Mcal/kg; 0.45% CV). The mean actual NE_m values from animal performance trials ranged from 1.75 to 2.48 Mcal/kg (average 2.03 Mcal/kg; 6.5% CV). The mean bias or difference in predicted vs. actual values ranged from −0.001 to 0.005 Mcal/kg. Barley NE_g values calculated from animal performance ranged from 1.13 to 1.78 Mcal/kg (average 1.39 Mcal/kg; 8.4% CV). Average predicted barley NE_m and NE_g were 0.02 and 0.01 Mcal/kg less, respectively, than the 2.06 Mcal/kg NE_m and 1.40 Mcal/kg NE_g reported by NRC. Barley NE can be predicted from simple laboratory procedures which will aid plant breeders developing new feed varieties and nutritionists formulating finishing rations for beef cattle.

INTRODUCTION

Barley is a major grain crop and source of feed for livestock in Canada and the northwestern United States; however, criteria for identifying feed quality characteristics of barley for beef cattle have not been clearly defined (Bowman et al., 2001; Fox et al., 2008). Any barley that does not meet the strictly defined malting quality criteria is relegated as feed barley, which generally commands a lesser price than malting barley (Fox et al., 2009). As a result, there is substantial variation in chemical composition of barley available for feeding (Hunt, 1996; Bowman et al., 2001) which leads to disparity in animal performance (Boss and Bowman, 1996a; Ovenell-Roy et al., 1998b). Equations have been reported for using chemical composition to predict the feeding value of hay (Reid et al., 1988) and energy values of corn (Smith et al., 2015) and barley (Fairbairn et al., 1999; Zijlstra et al., 2011; Sol et al., 2017) for swine. Identifying laboratory procedures that could be used to accurately predict barley energy value would facilitate the process of developing new feed varieties of barley (Surber et al., 2000) and aid nutritionists in formulating finishing rations for feedlot cattle. The objective of this research was to evaluate if barley energy value for beef cattle could be predicted by grain characteristics measured by laboratory procedures.

MATERIALS AND METHODS

The experimental protocol involving animals used in Montana was approved by the Montana State University Agricultural Animal Care and Use Committee, and the animal experimental protocol used in Idaho was approved by the University of Idaho Institutional Animal Care and Use Committee.

Data Set for Development of Prediction Equations

Animal performance data used for model development were from 19 feedlot experiments conducted in Montana (13 conducted at the Montana Agricultural Experiment Station, Bozeman, MT; and 6 conducted at the Northern Agricultural Research Center, Havre, MT) and 7 feedlot experiments conducted at the University of Idaho, Caldwell Research and Extension Center, Caldwell, ID, during a 10-yr period. A total of 80 experimental units were available, with each experimental unit being a diet mean from an individual feedlot experiment. Each diet mean was based on data from a range of 4 to 8 pens of cattle. Fifty-eight of the 80 experimental units were randomly selected and used in the development dataset, and the remaining 22 experimental units were used in the validation data set. The feedlot experiments were from a 10-yr project comparing finishing steer performance when fed current barley varieties and experimental barley lines under development. Experimental protocol was as similar as possible for all the feedlot experiments. Cattle used in the experiments were steer calves of primarily Angus and Angus-cross breeding. Steers were implanted with Ralgro (Schering-Plough Animal Health, Union, NJ) at the beginning of each experiment. A 28-d adaptation period was used to adjust steers to their respective high-concentrate diets. At the end of the 28-d adaptation period, initial unshrunk weights were obtained on two consecutive days and averaged. Final unshrunk weights were also obtained on two consecutive days and averaged. Initial and final weights were taken prior to that day’s feeding. Initial BW of the steers averaged 354 kg with a range of 287 to 462 kg. Steers at each of the three experimental locations (Bozeman, Havre, and Caldwell) were obtained from the same genetic source each year. Steers were fed for an average of 108 d, with a range of 70 to 168 d. Average final weights were 547 kg with a range of 497 to 580 kg.

Twenty-eight different barley genotypes were evaluated, including 18 barley cultivars (Baronesse, Busch 1202, Chinook, Colter, Eslick, Gallatin, Gunhilde, H3, Harrington, Haxby, Lewis, Logan, Medallion, Merlin, Morex, Steptoe, Valier, and Westbred 501) and 10 experimental lines (MTLB2, MTLB5, MTLB6, MTLB13, MTLB32, MTLB48, MTLB57, MT960228, MTSM3, and MTSM5). Dry rolled barley was the only grain source in the diets, and it made up 80% to 86% of the diet DM. Roughage ranged from 6% to 12.7% of the diet DM, and roughage sources included alfalfa hay, oatlage, wheat straw, and grass hay. Urea was added to make diets isonitrogenous within an experiment, and made up from 0.10% to 1.5% of diet DM. Other ingredients and their proportions in the diet DM included soybean oil (3.00%), wheat middlings (0 to 2.04%), calcium carbonate (0.50 to 2.87%), sodium bicarbonate (1.30%), dicalcium phosphate (0 to 3.33%), sodium chloride (0.50%), potassium chloride (0.50%), trace mineral premix (0.25%), vitamin premix (0.05%), monensin premix (0.024%; 132 g/kg monensin), and tylosin premix (0.013%; 88 g/kg tylosin). The trace mineral premix contained 20.0% Mg, 6.0% Mn, 5.0% Fe, 2.7% S, 1.5% Cu, 0.11% I, 0.01% Se, and 0.01% Co, and the vitamin premix contained 30,000 IU/g vitamin A, 6,000 IU/g vitamin D, and 7.5 IU/g vitamin E.

Chemical Composition of Barley Diets

Descriptive simple statistics for ADF, starch, and N content of the 58 finishing diets used in prediction equation development and the 22 finishing diets used in prediction equation validation are presented in Table 1. Diets used for model development ranged from 5.6% to 12.5% ADF, 36.1% to 51.3% starch, and 1.8% to 2.7% N, whereas those in the validation data set ranged from 7.5% to 11.1% ADF, 37.0% to 48.3% starch, and 1.8% to 2.4% N.

Table 1.

Simple statistics for analyzed chemical composition (DM basis) of barley diets fed to beef steers and used in prediction equation development (n = 58) and in validation of prediction equations (n = 22) for barley energy content

Item	Mean	SD	Minimum	Maximum
Database for prediction equation development
ADF, %	8.8	1.34	5.6	12.5
Starch, %	44.9	3.32	36.1	51.3
N, %	2.2	0.20	1.8	2.7
Database for validation of prediction equations
ADF, %	9.1	0.80	7.5	11.1
Starch, %	44.5	2.74	37.0	48.3
N, %	2.3	0.14	1.8	2.4

Physical Properties, Chemical Composition, and Nutritional Characteristics of Barley Samples

Whole barley grain samples from the feedlot experiments were each divided into two aliquots and the physical properties, chemical composition, and nutritional characteristics were measured at the time the specific feedlot experiment was being conducted. The first aliquot was ground through a 0.5-mm screen using a Udy-Cyclone mill (Udy Corp., Ft. Collins, CO) prior to determining the starch content by an amyloglucosidase/alpha-amylase method (AOAC, 1997) and ADF (Van Soest et al., 1991). The second aliquot was cracked through a Buehler mill to simulate the dry rolling processing done prior to feeding the barley and evaluated for DM (AOAC, 1997). Particle size was determined on the cracked grain aliquot by a dry sieving technique (Fisher et al., 1988). Two ruminally cannulated beef cows, each consuming low-quality grass hay ad libitum and 3.6 kg/d of barley, were used to determine in situ DM disappearance (ISDMD) of the cracked barley samples. Duplicate 5-g samples of each barley genotype were weighed and placed in 10 cm × 20 cm, 50-mm pore size polyester bags (Ankom Technology, Fairport, NY). Two blank bags in addition to the sample bags were placed in each rumen (each barley genotype was incubated in the rumen of both cows) and removed after 3 h. The bags were rinsed with cold water until the wash water ran clear. The bags were dried at 60°C for 48 h, and then weighed. In situ DM disappearance after 3 h of incubation was calculated.

Descriptive simple statistics of the candidate variables for the 58 barley samples used in model development, and the 22 barley samples used in model validation are shown in Table 2. Barley samples in the model development data set represented a wide range in concentrations (DM basis) of N (1.6% to 2.8%), ISDMD (25.7% to 58.7%), ADF (3.6% to 8.0%), starch (44.1% to 62.4%), particle size (1,100 to 2,814 µm), and bulk density (50.8 to 69.4 kg/hL). Barley samples in the validation data set also varied greatly in N (1.7% to 2.8%), ISDMD (26.8% to 57.0%), ADF (4.2% to 7.8%), starch (39.3% to 61.3%), particle size (1,123 to 1,956 µm), and bulk density (50.8 to 69.1 kg/hL).

Table 2.

Simple statistics for physical properties, analyzed chemical composition (DM basis), and nutritional characteristics (DM basis) of barley fed to beef steers and used in prediction equation development (n = 58) and in validation of prediction equations (n = 22) for barley energy content

Item	Mean	SD	Minimum	Maximum
Database for prediction equation development
N, %	2.1	0.28	1.6	2.8
ISDMD1, %	42.3	8.47	25.7	58.7
ADF, %	5.5	1.02	3.6	8.0
Starch, %	54.3	4.53	44.1	62.4
Particle size, µm	1,447	357.2	1,100	2,814
Bulk density, kg/hL	62.8	4.20	50.8	69.4
Database for validation of prediction equations
N, %	2.2	0.29	1.7	2.8
ISDMD, %1	45.1	10.38	26.8	57.0
ADF, %	5.4	1.00	4.2	7.8
Starch, %	53.1	5.01	39.3	61.3
Particle size, µm	1,354	191.2	1,123	1,956
Bulk density, kg/hL	62.0	4.33	50.8	69.1

¹In situ DM disappearance after 3 h of ruminal incubation.

Animal Performance and Barley NE_m and NE_g Content

Average BW, DMI, ADG, and steer NE_m and NE_g requirements (NRC, 2000) for each pen of cattle were used to estimate barley NE_m and NE_g content (Zinn, 1993) for that pen, and the values from pens within a specific diet averaged. The diet mean from an individual feedlot experiment was considered the experimental unit. Diet NE_m and NE_g were estimated using an iterative process (NRC, 1984) to fit the relationship NE_g = (0.877 × NE_m) − 0.41. Steer requirements for NE_m and NE_g were estimated using the following equations: NE_m, Mcal/d = 0.077 × average empty BW^0.75, and NE_g, Mcal/d = 0.0635 × average equivalent empty BW^0.75 × average empty body gain^1.097 (NRC, 2000). Empty BW was determined by BW × 0.96 × 0.891, and average empty body gain by average daily gain × 0.96 × 0.956 (NRC, 2000).

Barley NE_m and NE_g were estimated using the following equations:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{d}\mathrm{i}\mathrm{e}\mathrm{t}\mathrm{N}{\mathrm{E}}_{\mathrm{m}},}\\ \mathrm{M}\mathrm{c}\mathrm{a}\mathrm{l}/\mathrm{k}\mathrm{g}=& (\mathrm{A}\times \mathrm{N}{\mathrm{E}}_{\mathrm{m}}\mathrm{b}\mathrm{a}\mathrm{r}\mathrm{l}\mathrm{e}\mathrm{y})\\ +(\mathrm{B}\times \mathrm{N}{\mathrm{E}}_{\mathrm{m}}\mathrm{r}\mathrm{o}\mathrm{u}\mathrm{g}\mathrm{h}\mathrm{a}\mathrm{g}\mathrm{e})\\ +(\mathrm{C}\times \mathrm{N}{\mathrm{E}}_{\mathrm{m}}\mathrm{s}\mathrm{u}\mathrm{p}\mathrm{p}\mathrm{l}\mathrm{e}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t})\\ +(\mathrm{D}\times \mathrm{N}{\mathrm{E}}_{\mathrm{m}}\mathrm{o}\mathrm{i}\mathrm{l})\mathrm{a}\mathrm{n}\mathrm{d}\end{array}}$$

$${\displaystyle \begin{array}{ll}{\displaystyle \mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{d}\mathrm{i}\mathrm{e}\mathrm{t}\mathrm{N}{\mathrm{E}}_{\mathrm{g}},\mathrm{M}\mathrm{c}\mathrm{a}\mathrm{l}/\mathrm{k}\mathrm{g}=}& (\mathrm{A}\times \mathrm{N}{\mathrm{E}}_{\mathrm{g}}\mathrm{b}\mathrm{a}\mathrm{r}\mathrm{l}\mathrm{e}\mathrm{y})\\ +(\mathrm{B}\times \mathrm{N}{\mathrm{E}}_{\mathrm{g}}\mathrm{r}\mathrm{o}\mathrm{u}\mathrm{g}\mathrm{h}\mathrm{a}\mathrm{g}\mathrm{e})\\ +(\mathrm{C}\times \mathrm{N}{\mathrm{E}}_{\mathrm{g}}\mathrm{s}\mathrm{u}\mathrm{p}\mathrm{p}\mathrm{l}\mathrm{e}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t})\\ +(\mathrm{D}\times \mathrm{N}{\mathrm{E}}_{\mathrm{g}}\mathrm{o}\mathrm{i}\mathrm{l}),\end{array}}$$

where the constants A, B, C, and D represent the proportion of barley, roughage, supplement, and oil, respectively, in the diet DM. Book values for NE_m and NE_g of the roughage and oil were used (NRC, 2000), and values for NE_m and NE_g of supplement were taken from the supplement formulation. Barley NE_m and NE_g were the unknowns in the above equations.

Model Development

The variables considered as candidates for independent variables in the regression equations included bulk density, barley N content, barley ISDMD, barley ADF content, barley starch content, and barley particle size. These variables were chosen as candidates due to the relative ease with which they could be obtained and used on a practical basis.

Using the options FORWARD SLE = 0.10, and BACKWARD, in PROC STEPWISE of SAS (SAS Institute Inc., Cary, NC) variables were identified to be included in regression equations. The criterion used to select variables was each individual variable had to have Type III sums of squares P-value < 0.10 to be included in the model. After choosing the variables to be included in the multiple linear regression model, parameter estimation was performed using PROC REG of SAS. The options INFLUENCE and R were used to provide Cook’s D statistic, the Studentized Deleted Residuals, and the Dffits statistic of each observation in the model for outlier detection. An observation was considered an outlier and removed from the model if the absolute value of the Cook’s D statistic was ≥ 2, the absolute value of the Studentized Deleted Residual was ≥ 2, or the absolute value of the Dffits statistic was ≥ 1 (Christensen, 1997).

In order to be considered a valid model, the Mallows’ C_p statistic had to be ≤ p + 1, where p is the number of independent variables in the model (Hocking, 1976). The C_p statistic was obtained using PROC REG of SAS. The VIF option in PROC REG was used to provide the Variance Inflation Factor (VIF) which tests for multicollinearity between the independent variables. If the Variance Inflation Factor was ≥ 10, then the variables were determined to be correlated, and the model was considered not appropriate (Christensen, 1997). The SPEC option in PROC REG was used to test for the distribution and dependence of error terms, and a P-value of > 0.10 allowed us to conclude that the error terms were independent and identically distributed (Christensen, 1997). Once the models were determined to be valid using the C_p statistic, the Variance Inflation Factor, and the SPEC option in PROC REG; the J_p (Hocking, 1976), S_p (Hocking, 1976), Amemiya’s Prediction Criterion (PC; Amemiya, 1980), and Akaike’s Information Criterion (AIC; Amemiya, 1980) statistics were calculated using PROC REG of SAS. Models with minimum values for these statistics were considered more desirable.

Model Evaluation

After 58 of the original 80 experimental units were randomly selected to be used in model development, the remaining 22 experimental units were used as the validation dataset. The mean square prediction error (MSPE; Rook et al., 1990), the mean prediction error (MPE; Fuentes-Pila et al., 2003), the relative prediction error (RPE; Fuentes-Pila et al., 2003), and the mean bias (Cannas et al., 2004) were used to evaluate the prediction accuracy of the models. The MSPE was calculated as

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{M}\mathrm{S}\mathrm{P}\mathrm{E}=1/n\sum {\left(\mathrm{P}\mathrm{A}\right)}^2,}\end{array}}$$

where P is the predicted NE_m; A is the actual NE_m; and n is the number of pairs of values of P and A. The MPE was calculated as the square root of the MSPE; and the RPE was calculated as the MPE divided by the mean of the actual NE_m values and expressed as a percentage (Fuentes-Pila et al., 2003). The mean bias was calculated as

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}\mathrm{b}\mathrm{i}\mathrm{a}\mathrm{s}=1/n\sum \left(\mathrm{P}\mathrm{A}\right).}\end{array}}$$

The mean bias was used to evaluate the accuracy of the prediction, the MSPE was used to estimate the magnitude of the error, and the predicted residual sum of squares (PRESS) statistic was used to summarize the fit of each model to the validation data set (Draper and Smith, 1981). Mean predicted NE_m values were used to calculate NE_g based on the aforementioned equation (NRC, 2000).

RESULTS AND DISCUSSION

Observed Barley NE_m and NE_g Content

Descriptive simple statistics for animal performance and observed barley NE_m and NE_g of the 58 barley diets used in model development and the 22 barley diets used in prediction equation validation are presented in Table 3. Substantial variation in animal performance and observed barley NE_m and NE_g were observed in both model development and validation data sets. In the database for prediction equation development, barley NE_m from animal performance trials ranged from 1.79 to 2.48 Mcal/kg, with a mean of 2.04 Mcal/kg and barley NE_g ranged from 1.17 to 1.78 Mcal/kg, with a mean of 1.40 Mcal/kg. The range of NE_m and NE_g for the validation dataset was 1.69 to 2.20 Mcal/kg NE_m and 1.09 to 1.53 Mcal/kg NE_g. Mean values for NE_m and NE_g of the validation data set were slightly lower than the means from the development data set (1.97 Mcal/kg NE_m and 1.33 Mcal/kg NE_g).

Table 3.

Simple statistics for animal performance and observed barley NE_m and NE_g by finishing beef steers fed barley diets used in prediction equation development (n = 58) and in validation of prediction equations (n = 22) for barley energy content

Item	Mean	SD	Minimum	Maximum
Database for prediction equation development
NEm, Mcal/kg	2.04	0.13	1.79	2.48
NEg, Mcal/kg	1.40	0.12	1.17	1.78
DM intake, kg	9.4	1.19	7.3	12.4
ADG, kg	1.50	0.18	1.10	1.99
Gain/100 kg feed	16.3	1.71	13.4	21.5
Database for validation of prediction equations
NEm, Mcal/kg	1.97	0.16	1.69	2.20
NEg, Mcal/kg	1.33	0.14	1.09	1.53
DM intake, kg	9.7	1.45	7.5	12.4
ADG, kg	1.51	0.13	1.25	1.87
Gain/100 kg feed	15.8	1.75	13.3	18.6

Model Selection for Prediction of Barley NE_m Content

Table 4 summarizes the results of model development and presents the six equations with the highest R² values, along with the VIF, AIC, J_p, PC, S_p, and PRESS values for each model. The barley grain characteristics of particle size (PS), ISDMD, starch, and ADF were included as independent variables in the successful models (P = 0.001). The model with the highest R² value (0.60) and lowest values for 3 out of 5 multiple model selection criteria (AIC, PC, and PRESS) was Eq. 1: NE_m = 1.1764 + 0.00024822(PS) − 0.00033962(ISDMD²) + 0.00050543(starch x ISDMD). The model with the lowest mean bias and second lowest PRESS evaluated using the validation data set (Table 5) was Eq. 4: NE_m = 0.99514 + 0.01757(starch) + 5.31E-8(PS²). Equation 4 was also the least expensive model, which utilized only starch and particle size. Particle size analysis is simple and easy to conduct with the proper equipment and starch analysis costs ~US$36/sample (MidWest Labs, Omaha, NE). The model with the simplest terms was Eq. 3: 0.89183 − 0.00129(ISDMD) + 0.01847(starch) + 0.0001512(PS). The model that gave values closest to NRC (1996, 2016) estimates was Eq. 5 which utilized all four of the barley grain characteristics of PS, ISDMD, starch, and ADF as predictors (mean predicted NE_m = 2.05 Mcal/kg compared with NRC of 2.06 Mcal/kg; Table 6). The six validation equations gave mean predicted values for NE_m ranging from 1.99 to 2.05 Mcal/kg with an average of 2.04 Mcal/kg (Table 5). The mean actual NE_m values from animal performance trials ranged from 1.75 to 2.48 Mcal/kg with an average of 2.03 Mcal/kg (6.5% CV). The mean bias or difference in predicted versus actual values ranged from −0.001 for Eq. 4 to −0.03 for Eq. 2. Mean calculated values for NE_g based on animal performance ranged from 1.13 to 1.78 Mcal/kg with an average of 1.39 Mcal/kg (8.4% CV).

Table 4.

Prediction equations for NE_m (Mcal/kg) based on barley grain characteristics

Eq.	Equation	VIF1	R 2	Adj. R2	P-value	AIC2	Jp3	PC4	Sp5	PRESS6
1	1.1764 + 0.00024822(PS) − 0.00033962(ISDMD2) + 0.00050543(starch × ISDMD)	PS = 1.09, ISDMD2 = 8.35, starch × ISDMD = 8.15	0.60	0.56	0.001	−143.3	0.0098	0.5158	0.000335	0.3092
2	1.6998 − 0.00651(ISDMD) + 0.0002(starch2)	ISDMD = 1.09, starch2 = 1.08	0.55	0.52	0.001	−162.2	0.0097	0.5396	0.000288	0.3471
3	0.89183 − 0.00129(ISDMD) + 0.01847(starch) + 0.0001512(PS)	ISDMD = 1.15, starch = 1.14, PS = 1.01	0.53	0.49	0.001	−161.7	0.0099	0.5891	0.000294	0.3438
4	0.99514 + 0.01757(starch) + 5.311E-8(PS2)	Starch = 1.00, PS2 = 1.00	0.51	0.47	0.001	−166.0	0.0099	0.5871	0.000286	0.3555
5	1.15272 + 0.0151(starch) + 3.26E-6(PS × ISDMD) – 0.0004458(ISDMD × ADF)	Starch = 1.08, PS × ISDMD = 1.32, ISDMD × ADF = 1.23	0.51	0.47	0.001	−173.8	0.0103	0.6011	0.000282	0.3826
6	0.944 + 0.01835(starch) + 1.54E-5(PS × ADF)	Starch = 1.00, PS × ADF = 1.00	0.48	0.45	0.001	−173.1	0.0105	0.6134	0.000286	0.3971

PS = particle size; ISDMD = in situ DM disappearance after 3 h of ruminal incubation.

¹VIF = variance inflation factor.

²AIC = Akaike’s Information Criterion (Amemiya, 1980).

³J_p = J_p statistic (Hocking, 1976).

⁴PC = Amemiya’s Prediction Criterion (Amemiya, 1980).

⁵S_p = S_p statistic (Hocking, 1976).

⁶PRESS = predicted residual sum of squares (Draper and Smith, 1981).

Table 5.

Accuracy of prediction equations for NE_m (Mcal/kg) based on barley grain characteristics evaluated with the validation data set

Eq.	Mean predicted value	Mean actual value	P	MSPE1	MPE2	RPE3	Mean bias4	PRESS5
1	2.034	2.043	0.001	0.026	0.161	7.87	−0.009	0.3193
2	1.993	2.023	0.001	0.021	0.145	7.16	−0.030	0.4314
3	2.041	2.043	0.001	0.018	0.134	6.55	−0.002	0.3083
4	2.043	2.044	0.001	0.018	0.134	6.55	−0.001	0.2865
5	2.054	2.049	0.001	0.021	0.145	7.09	0.005	0.4459
6	2.047	2.044	0.001	0.017	0.130	6.36	0.003	0.2709

¹MSPE = mean square prediction error.

²MPE = mean prediction error (Mcal/kg).

³RPE = relative prediction error (%).

⁴Mean bias = the difference between the mean of the predicted values and the mean of the observed values (Mcal/kg).

⁵PRESS = predicted residual sum of squares.

Table 6.

Energy values of barley grain from current study and NRC

	NEm	NEg
Current study, predicted1
Eq. 1	2.03	1.37
Eq. 2	1.99	1.34
Eq. 3	2.04	1.38
Eq. 4	2.04	1.38
Eq. 5	2.05	1.39
Eq. 6	2.04	1.38
NRC, 1982
Grain	2.00	1.35
Grain, light 46.3 kg/hL	1.79	1.17
Grain, Pacific Coast	2.06	1.40
NRC, 1984
Grain	2.06	1.40
Grain, Pacific Coast	2.12	1.45
NRC, 1996 and 2000 update
Grain, heavy	2.06	1.40
Grain, light	1.85	1.22
NRC, 2016
Barley grain	2.06	1.40
Barley grain, steam flaked	2.06	1.40

¹ Mean predicted NE_m values were used to calculate NE_g based on the equation of NRC (1996), where NE_g, Mcal/kg = (0.877 × NE_m) − 0.41.

Comparison of Energy Values to NRC

Our observed mean NE_m and NE_g values from feedlot trials for dry rolled barley grain (2.03 and 1.39 Mcal/kg, respectively) were similar to the estimates reported by the NRC (2016; 2.06 and 1.40 Mcal/kg for NE_m and NE_g, respectively). In addition, our mean predicted barley NE_m (2.04 Mcal/kg; Table 5) and NE_g (1.37 Mcal/kg; Table 6) were very close to NRC (2016) estimates. Conversely, Boss and Bowman (1996a), Zinn et al. (1996), and Ovenell-Roy et al. (1998b) reported NE_m values for barley grain that were 10% to 22% greater (2.27 to 2.51 Mcal/kg NE_m) than NRC estimates. The NE_g values for barley grain from these studies and the report of Yaremcio et al. (1991) were 12% to 29% greater (1.57 to 1.80 Mcal/kg NE_g) than NRC estimates. Zinn et al. (1993, 1996), Ovenell-Roy et al. (1998b), and Nelson et al. (2000) reported NE_m and NE_g values that were closer to NRC (2016) estimates (2.14 and 2.18 Mcal/kg NE_m and 1.43 to 1.51 Mcal/kg NE_g). Interestingly, the NE_m and NE_g values reported in these studies conducted in California and Washington were similar to the NRC (1984) energy values for “Pacific Coast” barley grain (2.12 and 1.45 Mcal/kg NE_m and NE_g, respectively; Table 6) which are greater than NRC (1984, 1996, 2016) estimates for “Barley Grain” (2.06 and 1.40 Mcal/kg NE_m and NE_g, respectively). Fox et al. (2009) used the equations of Surber et al. (2000) as discussed by Bowman et al. (2001) to calculate NE for barley feed and malting cultivars from Australia and Europe. They observed NE_m values ranging from 2.36 to 2.55 Mcal/kg which would be on the high end of barley energy values reported in any of the above studies.

Other researchers have reported the ME or DE content of barley grain. Owens et al. (1997) reported the observed body weight-adjusted ME value for dry-rolled barley was 17% greater than the estimate provided by the 1996 NRC (3.55 vs. 3.04 Mcal/kg DM). In contrast, published values of DE for barley grain have been 15% lower to 9% greater than the NRC (1996, 2000, 2016) estimate of 3.71 Mcal/kg (Bradshaw et al., 1996; Ovenell-Roy et al., 1998a; Sanford et al., 2003). National Research Council (1984) estimates for barley grain were based on the equation: NE_m = 0.0305TDN − 0.5058, where the TDN for barley is 86% (Zinn et al., 1993). The NRC (2016) notes that values were obtained from three laboratories and lab analysis of TDN does not take into consideration energy losses due to urine, combustible gases, and heat (Jurgens, 1988). In addition, the studies which reported the lowest energy estimates (DE) used marker technology (chromic oxide; Bradshaw et al., 1996; Sanford et al., 2003) rather than equations based on animal performance (NRC, 1996). The DE/ME used by NRC (1996, 2016) may be low and ME should be directly measured (Ovenell-Roy et al., 1998a). The NE content of barley grain calculated from DE in digestion trials ranged from 1.63 to 1.94 Mcal/kg NE_m and 1.03 to 1.29 Mcal/kg NE_g and were lower than NE values for the same barley varieties obtained from animal performance in feedlot trials (Ovenell-Roy et al., 1998a, 1998b). The variability among barley varieties for feed quality is well established (Hunt, 1996) and differences in results among studies presented here could be due to the varieties evaluated; however, energy estimates based on TDN or marker estimates may be less accurate than NE obtained directly from animal feeding trials. Wide variability in barley energy content and the type of energy values estimated suggest that average values for NE may not be appropriate for balancing rations and emphasizes the need for, and importance of, a rapid, simple method for estimating the NE content of barley grain.

Barley Feed Quality Characteristics

A variety of chemical constituents have been proposed as important for barley feed quality for beef cattle: bulk density and kernel weight (Khorasani et al., 2000; Bleidere and Gaile, 2012), grain hardness (Fox et al., 2009; Bleidere and Gaile, 2012; Ding et al., 2015), particle size (Mathison, 1996; Du and Yu, 2011; Abdel-Haleem et al., 2012), ISDMD (Khorasani et al., 2000; Bowman et al., 2001; Fox et al., 2009), starch (Hunt, 1996; Khorasani et al., 2000; Bowman et al., 2001), ADF (Engstrom et al., 1992; Bowman et al., 2001; Fox et al., 2009), NDF content and digestibility (Hunt, 1996; Ovenell-Roy et al., 1998b; Fife et al., 2008), chemical composition of the hull (Grove et al., 2003; Du and Yu, 2011), and protein (Bleidere and Gaile, 2012). Some researchers have recommended a combination of desirable characteristics for barley feed grain: “A good feed barley variety should have these traits: high in nutrients, good nutrient availability, slow rate of rumen starch fermentation and maintaining large particle size after mechanical processing” (Du and Yu, 2011). Bleidere and Gaile (2012) provided a profile of desirable grain characteristics for ruminants, pigs, and poultry recommending barley for ruminants have hard endosperm with high volume weight, starch, β-glucan, vitamins, and minerals; average to high levels of protein; and low to average hull and fiber content. Previous research from our lab identified high starch content, low ADF, low ruminal dry matter digestibility, and large particle size after dry rolling as desirable barley feed-quality characteristics (Bowman et al., 2001). In the current study, NE_m and NE_g of barley for finishing cattle was predicted using particle size, ISDMD, starch, and ADF.

Starch content.

All six of our prediction equations contained starch as a predictor variable which had positive regression coefficients. Starch is an important component for evaluating barley feed quality as high starch content of grain provides available energy to the animal. One reason barley may be assigned a lower energy value than corn is because of its lower starch content (Hunt, 1996). Dry rolled corn grain has 72% starch compared to 56.7% starch in barley grain and was assigned greater energy values (2.17 Mcal/kg NE_m and 1.49 Mcal/kg NE_g for dry-rolled corn grain; NRC, 2016). Sanford et al. (2003) reported that barley DE was correlated with starch content (r = 0.71) and the best prediction equation for DE contained only starch (R² = 0.41).

Particle size.

Five out of six regression equations utilized particle size as a predictor variable which had positive regression coefficients confirming that large particle size is a desirable feed quality trait for barley fed to beef cattle. Barley needs to be processed rather than fed whole in order to breach the hull and improve digestibility (Dehghan-banadaky et al., 2007; Du and Yu, 2011); however, the degree of processing must be managed in order to control ruminal DM and starch digestion (Galyean et al., 1981; Wang et al., 2003). Larger particle size was associated with lower in situ DM disappearance (Du and Yu, 2011; Abdel-Haleem et al., 2012; Tagawa et al., 2017) due to decreased surface area available to microbes (Galyean et al., 1981). Improvements in starch digestion, feed efficiency, and NE_g have been reported with greater processing (Ribeiro et al., 2016), but extensive breakage of grain kernels during processing results in fine particles (Dehghan-banadaky et al., 2007) which can increase ruminal starch digestion (Ribeiro et al., 2016) causing ruminal acidosis (Beauchemin et al., 2001; Koenig and Beauchemin, 2011) and bloat (Ramsey et al., 2002). Larger particle size would lower ruminal degradability of DM and mitigate the negative effects of starch digestion on ruminal acidosis potentially improving animal health and performance.

In situ DM disappearance.

In situ DMD after 3 h of incubation was a predictor variable for NE_m in four of our prediction equations. Khorasani et al. (2000) also emphasized the importance of ruminal digestion in selecting barley cultivars for ruminant animals and Fox et al. (2009) concluded that ISDMD could be used as the primary quality trait for feed barley. Negative regression coefficients confirm an inverse relationship between ruminal in situ DM disappearance after 3 h of incubation and NE of barley grain and underscore the importance of slower ruminal digestion for barley feed quality. In situ DMD measured after a very short time of only 3 h is interpreted to be an estimate of rate of barley digestion. More slowly digesting barley could reduce the incidence of acidosis and/or bloat and improve animal performance. Slowly degrading cultivars of barley could supply more bypass starch to the small intestine providing more energy to the animal than starch digestion in the rumen (Owens et al., 1986). Others have suggested that high ruminal digestibility of starch may improve animal performance by supplying more energy to the microbes thereby improving microbial growth and N synthesis (Orskov, 1986; Boss and Bowman, 1996b). Increasing the amount of starch bypassing the rumen may not be advantageous due to physiological limits in post-ruminal starch digestion (Orskov, 1986). The importance of large particle size and lower ISDMD in the NE_m prediction equations from the current study supports the concept that shifting starch digestion to the small intestine would provide more energy to the animal.

Acid detergent fiber content.

Two of our six prediction equations included ADF as significant predictor variable with the regression coefficients positive when it was associated with particle size and negative with ISDMD. The positive coefficient when ADF was associated with particle size is an extremely small number, and therefore may not make a substantial contribution to barley NE. Fiber and fiber digestibility are primary factors having a negative effect on energy density of barley (Hunt, 1996) partly due to the inverse relationship between starch and ADF content (Hunt, 1996; Abdel-Haleem, et al., 2012). In addition, ADF was negatively correlated to 18-h in situ DM disappearance (r = −0.44; Reynolds et al., 1992) and total tract DM digestibility (r = −0.76, Sanford et al., 2003). Barley energy values were negatively correlated with ADF content (r = −0.78 and −0.71 for ME and DE, respectively; Doornbos and Newman, 1989; Sanford et al., 2003). Engstrom et al. (1992) reported a strong relationship between ADF content and feed efficiency (R² = 0.90) suggesting it would be a useful tool for determining barley feed quality. Sanford et al. (2003) reported a significant regression equation for predicting barley grain DE using ADF (R² = 0.31).

Bulk density.

Traditionally, feed quality of barley has been based on bulk density or test weight (Hunt, 1996; Khorasani et al., 2000; Brophy et al., 2012); however, bulk density was not a significant variable in any of our NE_m prediction equations. Bulk density in our feedlot dataset ranged from 50.8 to 69.4 kg/hL with an average of 62.5 kg/hL. Bulk density was not correlated (P > 0.16, R² < 0.15; data not shown) with NE_m, NE_g, or feed efficiency in our data set. Bulk density has been correlated with in situ DM disappearance and starch content (r = 0.72 and 0.47; Reynolds et al., 1992); however, others did not find this to be true. Doornbos and Newman (1989) reported no correlation between bulk density and in vitro DM digestibility or soluble carbohydrates (r = 0.10 and −0.02, respectively). In monogastrics, bulk density was related to barley DE content for swine (R² = 0.64; Fairbairn et al., 1999), but not when measured using mice (Christison and Bell, 1975). There has been no consistent relationship between bulk density and beef cattle growth performance (Grimson et al., 1987; Mathison et al., 1991; Hunt, 1996) and no relationship between bulk density and barley grain DE (Sanford et al., 2003) or NE content (McDonnell et al., 2003) for cattle. These and our results suggest that there are more important characteristics than bulk density for evaluating barley feed quality and predicting energy content of barley grain for beef cattle.

Nitrogen.

We evaluated N content of barley grain in the current study; however, it was not a significant variable in any of the prediction equations. In agreement with our results, Sanford et al. (2003) reported no correlation between CP and barley DE in cattle and CP was also a poor predictor of barley DE in swine (R² = 0.07; Fairbairn et al., 1999). In contrast, protein was correlated with barley DE in mice (r = 0.41; Bhatty et al., 1974). Hunt (1996) suggested that CP and barley energy values were negatively related due to differences in starch content between barleys. Starch is enclosed within a protein matrix which could affect starch digestion from barley (McAllister et al., 1993) and may explain the negative correlation between CP and ISDMD of cracked barley (Abdel-Haleem et al., 2012). Khorasani et al. (2000) also reported that the degradable fraction of DM was correlated to CP content and that its inclusion slightly improved prediction equations for effective DM degradability.

Comparison to Equations Developed For Pigs

Equations based on chemical composition have been used to predict the DE, ME, or NE content of diets (Noblet and Perez, 1993), ingredients (Liu et al., 2015; Sol et al., 2017), barley (Fairbairn et al., 1999; Zijlstra et al., 2011; Bolarinwa and Adeola, 2016), and corn (Smith et al., 2015) fed to pigs. Similar to our results, several equations utilized ADF or digestibility as predictors. Acid detergent fiber accounted for 73% to 85% of the variation in DE content of barley for pigs (Fairbairn et al., 1999; Wang et al., 2017) and the prediction of NE content of various feed ingredients fed to growing pigs also utilized ADF (R² = 0.94; Liu et al., 2015). Bolarinwa and Adeola (2016) reported differences between actual and predicted NE of barley larger than what we observed in the current study (0.05 to 0.08 Mcal/kg) and concluded that regression methods gave similar energy estimates for barley as those obtained using pigs. Smith et al. (2015) developed regression equations for predicting DE of corn for swine using chemical composition and reported similar R² values to the ones we observed (R² = 0.76 to 0.79 for training data set), and similar differences between actual and predicted energy values (−0.005 to −0.0076 Mcal/kg). In vitro OM digestibility was the best predictor of energy value of conventional byproducts for pigs (R² = 0.97 for both ME and DE and CV of 3.10 and 2.68, respectively; Sol et al., 2017). Their R² values were greater than ours, but their CV between actual and predicted values were similar. The prediction equations developed for barley fed to beef cattle in the current study have similar accuracy to published equations developed for estimating energy content of feed ingredients for swine.

This is the first peer-reviewed publication based on laboratory procedures that could be used to predict energy content of barley for beef cattle. Our NE_m prediction equations utilized particle size, starch content, ISDMD, and ADF as predictor variables with regression coefficients being positive for particle size and starch and negative for ISDMD and ADF. These equations would aid nutritionists in formulating rations for finishing cattle and barley breeders in developing improved varieties of feed barley.