Neutral detergent fiber rather than other dietary fiber types as an independent variable increases the accuracy of prediction equation for digestible energy in feeds for growing pigs

Objective The objectives were to investigate correlations between energy digestibility (digestible energy [DE]:gross energy [GE]) and various fiber types including crude fiber (CF), total dietary fiber (TDF), soluble dietary fiber (SDF), insoluble dietary fiber (IDF), neutral detergent fiber (NDF), and acid detergent fiber (ADF), and to develop prediction equations for estimating DE in feed ingredients and diets for growing pigs. Methods A total of 289 data with DE values and chemical composition of feeds from 39 studies were used to develop prediction equations for DE. The equations were validated using values provided by the National Research Council. Results The DE values in feed ingredients ranged from 2,011 to 4,590 kcal/kg dry matter (DM) and those in diets ranged from 2,801 to 4,203 kcal/kg DM. In feed ingredients, DE:GE was negatively correlated (p<0.001) with NDF (r = −0.84), IDF (r = −0.83), TDF (r = −0.82), ADF (r = −0.78), and CF (r = −0.72). A best-fitting model for DE (kcal/kg) in feed ingredients was: 1,356 + (0.704 × GE, kcal/kg) − (60.3 × ash, %) − (27.7 × NDF, %) with R2 = 0.80 and p<0.001. In diets, DE:GE was negatively correlated (p<0.01) with NDF (r = −0.72), IDF (r = −0.61), TDF (r = −0.52), CF (r = −0.45), and ADF (r = −0.34). A best-fitting model for DE (kcal/kg) in diets was: 1,551 + (0.606 × GE, kcal/kg) − (22.1 × ash, %) − (25.6 × NDF, %) with R2 = 0.62 and p<0.001. All variables are expressed as DM basis. The equation developed for DE in feed ingredients had greater accuracy than a published equation for DE. Conclusion All fiber types are reasonably good independent variables for predicting DE of swine feeds. The best-fitting model for predicting DE of feeds employed neutral detergent fiber as an independent variable.


INTRODUCTION
The energy supply to animals using feed ingredients accounts for the largest portion of total feed cost. To determine available energy concentrations in feed ingredients fed to pigs, in vivo experiments have been widely conducted. However, animal experiments to determine the energy values of feed ingredients are timeconsuming and expensive. Therefore, alternative methods to determine energy values of feedstuffs have been developed. As one of the alter native methods, prediction equations have been developed to determine digestible energy (DE) values using the chemical composition of feed ingredients [14] and diets [5,6] for pigs.
In previously reported DE predicting equations for swine feeds, dietary fiber was used as a negative independent variable [1,2,4,6] as the dietary fiber is less digestible than starch, protein, and fat [6]. Several fiber analysis procedures are available including the crude fiber (CF) analysis [7], the detergent fiber procedure [8], and total dietary fiber (TDF) procedure [9]. Among the fiber analyz ing procedures, the TDF procedure may provide an accurate estimate of fiber because TDF procedure takes the soluble dietary fiber (SDF) into account [10]. However, to our knowledge, there has been very limited effort to employ TDF as an independent variable for predicting DE in swine feeds. Therefore, the objectives of the present study were to investi gate correlations between energy digestibility and various fiber types including CF, TDF, insoluble dietary fiber (IDF), SDF, neutral detergent fiber (NDF), and acid detergent fiber (ADF) and to develop and validate prediction equations for estimating DE using a fiber type as an independent variable for swine feeds.

Data collection
A total of 289 data (105 feed ingredients and 184 diets) with DE values and chemical composition of feeds from 39 re search papers were used to develop prediction equations for DE concentration. For the literature search in PubMed and Google Scholar, keywords used were DE, energy digestibility, nutrient digestibility, fiber, and pigs. The papers found were manually screened based on the title and the experimental procedures. During this screening process, data from nursery pigs or sows were removed. When TDF values for an ingre dient is not available, the data were not used in the present work. The database consisted of crude protein (CP), ether extract (EE), ash, CF, NDF, ADF, TDF, IDF, SDF, and gross energy (GE) in the feeds (% or kcal/kg of DM basis). When an analyzed fiber concentration was not provided in the liter ature, the dietary fiber concentration was calculated based on the inclusion rate of feed ingredients and the fiber con centration of each ingredient (Table 1). Additionally, TDF concentration less than NDF concentration in feed ingre dient was excluded from the database.

RESULTS
Most nutrient and energy concentrations in feed ingredients were more variable than those in diet based on coefficients of variation ( Table 2). The NDF concentrations in feed ingre dients ranged from 7.2% to 63.2% while those in diets ranged from 5.1% to 34.4% on DM basis. The DE values in feed in gredients ranged from 2,011 to 4,590 kcal/kg DM, and those in diets ranged from 2,801 to 4,203 kcal/kg DM. Dietary fibers including CF, TDF, IDF, SDF, NDF, and ADF were positively correlated with each other in feed ingredients (r = 0.26 to 0.99; p<0.01; Table 3) and diets (r = 0.32 to 0.93; p<0.001; Table 4). The DE values in feed ingredients were    Table 5). All nutrient vari ables are expressed as DM basis.
The DE values in diets were also positively correlated (p<     Table 5). All nutrient variables are expressed as DM basis.
The determined DE values of feed ingredients presented by the NRC [10] were plotted against a calculated DE values using an equation developed in the present work employing GE, NDF, and ash as independent variables and using an equa tion suggested by Noblet and Perez [6] (Figure 1). When the equation developed in the present work was tested using the NRC [10] data, the intercept representing a mean bias was not different from 0 but the slope representing a linear bias  was different from 0 (p<0.001; Figure 1a). For the equation suggested by Noblet and Perez [6], both intercept and slope were different from 0 (p<0.001; Figure 1b) in the model vali dation results.

DISCUSSION
The data of CF, NDF, and ADF used in the present study were in good agreement with NRC [10] and Sauvant et al [12]. When a TDF concentration was less than an NDF concen tration in a feed ingredient, the data were not used for equation development because theoretically TDF includes SDF such as pectin, βglucan, and gum [9,10].
When collecting data to develop an accurate prediction equation for DE, 2 factors were considered. First, only data of DE:GE were collected from pigs fed mashform diets be cause feed processing may affect DE:GE [13]. Second, data derived from less than 20 kg of initial body weight of pigs were excluded. This is because the energy digestibility of feed in gredients [14] and diet [15] of nursery pigs would be less than that of growing and finishing pigs due to the immature di gestive capacity and relatively small intestine size of nursery pigs [16].
Energy or nutrient digestibility is dependent on physico chemical characteristics of dietary fiber in feeds [1719]. Even though TDF, IDF, and SDF are regarded as dietary fiber, the impact of each dietary fiber on digestibility differs. The energy digestibility coefficients were greater in growing pigs and sows fed highSDF diets compared with pigs fed highIDF diets [17,18]. In the same manner to in vivo studies, in vitro total tract disappearance of DM and organic matter had greater correlation with IDF than TDF [20]. These results indicate that the TDF, IDF, and SDF may differently affect energy di gestibility due to the different physicochemical properties. Generally, IDF is less fermentable than SDF and the passage rate of digesta is most likely to be increased by IDF rather than SDF due to greater fecal bulk inducing intestinal motility and peristaltic wave in the gastrointestinal tract [21]. Also, the SDF is lesslignified than IDF [22] leading to increased digesta viscosity [23] and enzymatic digestion compared with IDF [24]. For these reasons, energy digestibility of a highSDF diet is greater than that in a highIDF diet. As the influence of IDF on digestibility is largely different from that of SDF, the TDF which is the sum of IDF and SDF would be less corre lated with DE:GE compared with IDF in diets (r = -0.52 vs -0.61; Table 4). However, the DE:GE and DE were much more correlated with TDF than IDF in the work of Navarro et al [25]. The reason for this inconsistency may be the specific ingredient composition in the experiment by Navarro et al [25] who used synthetic cellulose and pectin to represent high IDF and highSDF source, respectively. In the present study, however, the data employing synthetic cellulose or pectin were not used.
The DE was calculated by multiplying GE concentration by DE:GE. In the present work, TDF and IDF were negatively correlated with DE:GE in feed ingredients whereas those fiber components were positively correlated with GE resulting in weakened negative correlation between those fiber compo nents and DE (Table 3). These results are supported by a recent study [4]. For this reason, TDF may have shown less accuracy in predicting DE compared with NDF in feed in gredients. However, Anderson et al [26] and Kerr et al [2] reported that TDF had greater R 2 than NDF to predict DE values of feed ingredients. The reason for this inconsistency may be due to the differences in analyzed TDF concentrations in cornbyproducts. In the present database, TDF concen trations in cornbyproducts such as distillers dried grains with solubles was greater than NDF concentration whereas TDF was less than NDF in Anderson et al [26] and Kerr et al [2]. Feed ingredients used to develop equations would be an important factor for the inconsistent results. In contrast to the previous studies [2,26], highSDF feed ingredients such as barley and sugar beet pulp were used to develope equations in the present work.
The NDF and IDF concentrations in the same ingredients have a similar range except for a highIDF ingredient (cellu lose) and a highSDF ingredient (pectin) in Navarro et al [25]. Although NDF and IDF values were comparable in most of feed ingredients or diets, NDF was the most accurate inde pendent variable compared with other dietary fibers in the current study. This was unexpected because TDF more ac curately represents the sum of fibers in a feed ingredient or diet compared with other dietary fibers including CF, SDF, IDF, NDF, and ADF. This result may be attributed to the analy sis errors of the TDF procedure. The TDF procedure (TDF, IDF, and SDF) had less reproducibility and repeatability than the detergent fiber procedure [7]. Additionally, the IDF and SDF had different physicochemical characteristics which may decrease correlation between TDF and GE [20]. The charac teristics of dietary fiber may contribute to the accuracy of DE estimation using TDF as an independent variable. In present work, however, NDF showed the greatest accuracy for esti mating DE values perhaps because NDF had no significant correlation with GE. Therefore, further research is warranted to compare the detergent fiber procedure and TDF procedure as an independent variable on estimating DE values.
The bestfitting model for DE of feed ingredients in the present work had a better accuracy than an equation from Noblet and Perez [6] who used NDF as an independent vari able ( Figure 1). When developing a prediction equation, a wide range of chemical composition is desirable for high ap plicability [5]. The chemical compositions in the work by Noblet and Perez [6] had a relatively narrower range than those of the present work.
A limitation of the present work is that only DEpredict ing equations are reported. The relationship between energy digestibility and fiber types was mainly addressed. When collecting data from the literature, quite a few experiments employed an index method and did not report metabolizable energy values. Further research is warranted to develop pre diction models for metabolizable energy and net energy.

CONCLUSION
The energy digestibility may be less affected by SDF than IDF. The GE is an important factor for predicting DE. The DE in swine feed ingredients and diets can be fairly accurately esti mated using equations with NDF compared with other fiber types.