Prediction of in vivo organic matter digestibility of ruminant feeds using in vitro techniques

K-J. Leeuw, D. Palić, F.K. Siebrits, H. Muller & V.A. Hindle 1 Agricultural Research Council, Private Bag X2, Irene, 0062, South Africa 2 Institute for Food Technology, Bulevar cara Lazara 1, 21000 Novi Sad, Serbia 3 TUT, Department of Animal Sciences, P. Bag X680, Pretoria 0001, South Africa UNISA Department of Statistics, P O Box 392, Unisa 0003, South Africa 5 Wageningen UR Livestock Research, PO Box 65, 8200AB Lelystad, Netherlands


Introduction
Feed cost is a major variable in raising or finishing livestock in a farming enterprise.Great emphasis is therefore placed on the quality of animal feed to improve feed efficiency and therefore reduce feed cost per production unit.An important determinant of feed quality is the digestibility of nutrients.The most accurate way of obtaining information about the digestibility of organic matter of feed for ruminants is by conducting in vivo digestibility studies.Since these methods are expensive and time consuming, and are not suited to routine analysis, reliable laboratory methods should be developed for routine prediction of the in vivo organic matter digestibility (OMD) of ruminant feeds (Beecher et al., 2015).
Although the in situ and in vitro techniques have good potential to predict in vivo OMD (Khazaal et al., 1993;Chenost et al., 2001), they have not been fully validated (Gosselink et al., 2004).Additionally, most developed techniques have been used to estimate the OMD of forages (Beecher et al., 2015;Gierus et al., 2016).There were only a few reports on the use of in vitro methods to estimate the OMD of compound feeds and complete diets (Aufrère & Michalet-Doreau, 1988).In addition, the results of OMD prediction for feedstuffs and for compound diets for ruminants are seldom reported in a single study (Dowman & Collins, 1982).
There are limitations to the use of rumen liquor for digestibility studies.There must be fistulated animals, which are not available to all laboratories, to collect fresh rumen liquor.Rumen liquor cannot be cooled down and must stay anaerobic (Stern et al., 1997).These animals must be maintained on a standard feeding regime to minimize changes in the rumen microbe population (Jones & Theodorou, 2000).
Feeds can be incubated with enzymes to predict in vivo OMD.This process aims to mimic the digestive process in the animal.The use of enzymes makes the analyses completely independent of the animal (Stern et al., 1997).Most enzymatic methods for OMD estimation were developed for forage feedstuffs, with a few being used for compound feeds (Aufrère & Michalet-Doreau, 1988;Weisbjerg & Hvelplund, 1993).Aufrère & Michalet-Doreau (1988) used two enzymatic methods (digestion by pepsincellulase, with 1 N HCl or 0.1 N HCl) to estimate the OMD of compound feeds.These were adapted from the enzymatic method, which were developed for single forages.They found that the estimation of digestibility was better with 0.1 N HCl than that of other chemical or biological methods.Weisbjerg & Hvelplund (1993) developed a pepsin-acid multi-enzymatic incubation method to estimate the enzymatic digestibility of organic matter for use on compound feeds.This procedure also showed the ability to estimate the OMD of straws (Hvelplund et al., 1999), and thus demonstrated the potential of this method to predict the in vivo OMD of both complete diets and forages.
The aim of this study was to verify the validity of the pepsin-acid multi-enzymatic procedure of Weisbjerg & Hvelplund (1993) to accurately predict in vivo OMD with a wide range of feedstuffs and compound feeds for ruminants and to compare these results with existing modified two-stage in vitro technique of Tilley & Terry (1963) and in vivo procedures.

Material and Methods
The in vivo OMD of all feedstuff and compound feed samples was determined in trials with sheep.The trials were conducted at the Animal Sciences Group of Wageningen UR, Division: Animal Production (former Institute for Animal Science and Health, ID-DLO, Lelystad), The Netherlands.The Animal Ethics Committee at the Animal Sciences Group of Wageningen UR, Division, approved the trial protocol: Animal Production and TUT (ref.number: AREC2011/06/008).Rumen liquor for modified Tilley & Terry in vitro analysis was collected from cannulated sheep housed at ARC-API Irene, approved by ARC-API Animal Ethics Committee.
To develop the initial prediction equations, 17 samples of commonly used temperate feedstuffs (including maize bran, maize gluten, maize gluten silage, maize cob leaves, hominy chop, barley, wheat, wheat middlings, wheat bran, wheat straw, sunflower oilcake, soybean meal and cottonseed meal) and six complete diets (with pre-determined in vivo OMD were used.An additional 21 samples of temperate feedstuffs and 24 complete diets were used to verify the prediction equations.Therefore, 68 samples of ruminant feeds were used to compare the in vivo OMD of ruminant feedstuffs, complete diets and their combination with in vitro analysis and the PME method. All samples were analysed for OMD using the PME OMD procedure (Weisbjerg & Hvelplund, 1993) and a modified in vitro two-stage OMD technique of Tilley & Terry (1963) (MT).Multi-enzymatic OMD values were compared with the apparent OMD values determined in vivo, and with values established with MT as this is currently the preferred method for estimating OMD of ruminant feeds.
The modified two-stage in vitro OMD (Tilley & Terry, 1963) was followed.The sample (0.5 g, ground in a Wiley mill through 1 mm sieve) was placed in 100 mL Schott reagent bottles (warmed to 39 °C).To this were added 5 mL urea solution (8.6 g to 1 L distilled water, as adapted by Engels & Van der Merwe (1967), and 50 mL rumen-saliva mixture (1 L rumen fluid mixed with 2 L McDougall's artificial saliva modification done by the ARC-API nutrition laboratory to improve repeatability), while flushing bottles with CO 2 gas to maintain an anaerobic environment.The bottles were then sealed and placed in an incubator (39 °C) and swirled at regular intervals (24 hours).After 48 hours incubation the sample was removed from the incubator, the lid was removed and 5 mL HCl solution (600 mL HCl to 400 mL distilled water or 6N HCl) was added in small volumes (1 mL then swirl, 2 mL then swirl, and 2 mL and swirl), after which 5 mL pepsin solution (8 g pepsin (2500 units/g) to 1 L distilled water) was added.The bottles were then sealed and placed in an incubator at 39 °C, and swirled at regular intervals (24 hours).After another 48-hour incubation, the samples were removed from the incubator.The contents of the Schott bottles were carefully transferred to a glass centrifuge tube (100 mL) after centrifuging (1207.4relative centrifugal force (RCF) for 10 min.) the supernatant was removed.The residue was rinsed with distilled water and centrifuged again (1207.4RCF for 10 min.).The tube and contents were dried for 24 hours at 105 °C after which the tube and sample were weighed.After this the tube and sample were placed in a furnace for 6 hours at 500 °C, and then both were placed in a desiccator to allow for cooling down (30 min.)prior to being weighed.Dry matter and ash contents of feeds were determined and used to calculate in vitro digestibility of organic matter for both MT and PME OMD analysis.Organic matter digestibility of the samples determined in vitro, were calculated as: In vitro organic matter digestibility (%) =

=
x 100 Castrated adult male sheep (Texel and Texel crosses) with average bodyweight of about 75 kg were used.Animals were housed in special-purpose balance crates under controlled room conditions in the purpose-built metabolism unit.The balance crates are designed to facilitate separation of faeces and urine and allow their separate collection.Water was freely available at all times.The daily amount fed to each animal (two equal portions at two mealtimes) was standardized at 1000 g DM daily for each animal.Animals went through an adaptation period of two weeks, then an 11-day preliminary period, during which the animals were fed the trial ration (feed residues and faeces were not collected), after which the 10-day collection period followed, during which the exact amounts of feed, feed residues and the faecal production were recorded.Feed residues and faeces were collected quantitatively for each animal.At the end of the trial, total feed residues and faeces were homogenized and weighed.Subsamples of feed, feed residues and faeces were taken to determine dry matter and organic matter according to the AOAC ( 2002) methods.During the preliminary period of each trial, a composite sample of feeds was taken for OMD analysis.Collected samples were sent by courier to South Africa for the MT and PME analysis.
In vivo OMD was calculated using the following equation: In vivo organic matter digestibility (%) = (feed dry matter consumed x % organic matter feed) -(faeces dry matter produced x % organic matter faeces) = organic matter disappeared / (feed consumed x % organic matter feed) The in vivo, PME OMD, and the MT OMD values were analysed, using SAS (Statistical Analysis System) software package (SAS Institute Inc., 1989, V 8), to evaluate how well the two methods predicted in vivo measures.Regression functions and R 2 s (with associated mean square error values) for an initial population of 6 complete diets and 17 feedstuffs and for a second population of 24 complete diets and 21 feedstuffs were derived for each OMD procedure, using SAS.The OMD values of the 68 feeds, obtained by both PME OMD and MT procedures, were linearly regressed against the in vivo OMD values.Verification and improvement of the regressions were done using the first dataset versus the second dataset.The formula used for the verification regression was: where: y i represents the estimated in vivo values for OMD x 1i represents the calculated MT or PME method OMD of the second experiment (and ß 0 and ß 1 are set to the intercept and slope estimates) e i represents the random error component.
The same population test on both datasets was done using a multivariate regression approach in SAS.Further differences for slope to unity were also tested under H0:  ̂1 = 1 after lack of fit was determined.

Results and Discussion
The first dataset comprised six complete sheep diets and 17 feedstuffs (data not shown).The general equation for predicting OMD can be expressed as: where: y i represents in vivo organic matter digestibility x 1i represents the modified in vitro or multi-enzymatic method e i represents the random error component and the slope by β 1 and intercept by β 0 sample in matter organic g matter organic insoluble g sample in matter organic g − The results of the regression equations based on the digestibility values (not shown) for the first dataset are presented in Table 1.The prediction equations, based on both analytical methods, predict in vivo measurement or value reliably (the R 2 values vary between 0.75 and 0.92 on the 0.1% level of significance, Table 1).These initial equations were followed by a second experiment in which a further 21 feedstuff samples and 24 complete diet samples (45 sample sets in total) were measured OMD in vivo, MT and PME.This was done to verify and improve the prediction equations and the predictive power of the first experiment.It was argued that if this first set of predictions equations (Table 1) were reliable, accurate estimates of the measured in vivo values (determined for the second experiment) should be obtained if calculated values for the MT PME (calculated in the second experiment) were substituted in the prediction equations derived in the first experiment.Therefore, these estimated in vivo OMD, using prediction equations in Table 1 for feedstuff, combined feeds and complete diets, were then regressed against the observed in vivo values for the second dataset, which resulted in the regression equations reported in Table 2.The statistically significant R 2 coefficients reported in Table 2 (values range between 0.50 and 0.97 on the 0.1% level of significance) describe the close relationship between the new observed in vivo values and the estimated in vivo values based on the prediction equations of the first experiment.This thus served as a measure of the predictive power of the initial prediction equations.The exception was the complete diets with R 2 values of 0.50 and 0.75, which was understandable, given that the initial prediction for complete diets were based on six values.The second experiment aimed to improve the prediction equation for complete diets by including 24 new samples.
The formula used for the verification regression: where: y i represents the estimated in vivo values for OMD x 1i represents the calculated modified in vitro or PME method OMD of the second experiment (and ß 0 and ß 1 are set to the intercept and slope estimates listed in Table 1 e i represents the random error component. The R 2 values (Table 2) indicate that the previous prediction equation (Table 1) predicted the new data accurately for the feedstuffs and the combined dataset.The new dataset for complete diets (24 samples) did not fit the first dataset (6 samples), as was evident from the low R 2 values (   The argument was made that the initial in vivo prediction models, obtained with the first dataset and verified with the second, could be refined and improved on by calculating new prediction models for in vivo estimations (using either the MT or PME) based on the combined dataset of the first and second experiment of this study.The results of improved in vivo prediction models for feedstuffs, complete diets and combined feeds are reported in Table 3. Models are reported for both MT and PME.
Table 3 reports R 2 values that range between 0.82 and 0.95 on the 0.1% level of significance, which bears evidence to the strong predictive power of both the MT and PME techniques to estimate in vivo measurements.Although the combined dataset gives an improved R 2 (Table 3) value, the improved equation could be validated in future research with new datasets.3 indicate that compared with the R 2 values of Table 1 and Table 2 there is an improvement in the accuracy of the predictive equations.

The regression equations in Table
The argument further had to be addressed whether it was justified to combine the two datasets: in other words, whether data from the two datasets came from the same population.This issue was addressed by means of a multivariate regression approach (including a dummy variable): a linear in vivo regression was calculated for combined feeds (with either MT or PME values as independent variable) for the combined data of datasets one and two.Furthermore, a dummy variable was then entered into the equation to test for the effect of separate, statistically significant slopes for datasets one and two.A second dummy variable was lastly entered into the model to test for the effect of separate, statistically significant intercepts for the datasets.It was reasoned that if the effect of slope proved statistically significant, this would suggest that the two datasets came from different populations and that the improvement of the prediction equations, by combining the datasets were not justified.
To accommodate the evaluation of separate slopes and intercepts in the regression model, a 'zero/one' variable was introduced into the dataset: for the combined data, values of 0 identified the first dataset and a value of 1 the second dataset.The prediction equation when testing the same population assumption for the OMD of combined feeds (y i ) for the MT or the PME methods (x1 i ), can be expressed as: where: x1 i represents the MT or PME method observations x2 i represents the qualitative variable with values of 0 and 1 to identify an observational value as belonging to the first or second dataset β 0 ,β 2 represents intercept-component parameter estimates β 1 ,β 3 represents slope-component parameter estimates e i represents the random error component The regression model can be broken down into a prediction model for the first and second datasets by plugging in the values of x2 i .For the first dataset observations, with x2 i = 0, the prediction equation for OMD becomes: y i = β 0 + β 1 x1 i + e i ; i = 1……n; with a slope of β 1 and an intercept of β 0. when x2 i = 1, representing the new dataset values, the prediction equation becomes: y i = (β 0 + β 2 ) + (β 1 + β 3 )x1 i + e i ; i = 1……n; with a slope parameter of (β 1 + β 3 ), and an intercept parameter of (β 0 + β 2 ) (Table 4).
The determination of the validity of the same population assumption of the improved in vivo prediction equations was concluded with an investigation of the residuals of both prediction equations to ensure the normality of residuals and group-homogeneity of residual variances (assumptions of regression).The results are reported in the last row of Table 4.The tests indicated that these assumptions were complied with.
From Table 4 it can be deduced that the effect of separate slopes over the two sets of data on in vivo regressions were statistically non-significant.This confirms that a single slope is applicable in each instance, which supports the assumption that the datasets come from the same population.The improved prediction equation for MT is expressed as: Y i all samples = 56.12+ 0.92(x1 i ) (values from Table 3) The same population assumption is valid for the PME method when analysing the complete dataset.The improved prediction equation for the PME is expressed as: Y i all samples = 160.6 + 0.75(x1 i ) (values from Table 3).
Table 4 Linear regression with qualitative variable included in the model: y i = β 0 + β 1 x1 i + β 2 x2 i + β 3 x1 i x2 i + e i Same population assumption verification in the improved prediction of in vivo organic matter digestibility of combined feeds using modified in vitro and pepsin-acid multi-enzymatic methods The separate feedstuffs datasets and the complete diets datasets were also analysed to verify the same population assumption for the two instances (Table 5).It could be concluded from the results in Table 5 that the effect of separate slopes for 'old' and 'new' data for complete diets and was statistically nonsignificant.Thus, a single slope was applicable in both instances and the same population assumption was verified for both types of datasets.

Method
The improved prediction equation for feedstuffs OMD can be stated as follows: Y i MT all samples = 49.86 + 0.94(x1 i ) (values from Table 3) Y i PME all samples = 132.26+ 0.80(x1 i ) (values from Table 3) The prediction equations of the first dataset of complete feeds (Table 1) did not predict the second dataset accurately (Table 2) and the small sample available in the first experiment, n 1 = 6) when considering the R 2 values, especially for the MT, which led to an unsatisfactory validation.From Table 3 it can be stated that the combined dataset (n = 30) resulted in an improved regression equation for complete feeds.This new improved equation should be tested against a new dataset to validate it.
The improved prediction equation for complete diet OMD can be stated as follows: Y i MT = 8.80 + 0.97(x1 i ) (values from Table 3) Y i PME = 224.30+ 0.66(x1 i ) (values from Table 3) Homogeneity for β 1 was established and proved not to be different. 1 = 1 was determined and for MT this was true and did not differ significantly from 1, as the values were within the 95% confidence intervals.

Table 1
Organic matter digestibility means and regression equations for complete diets selected feedstuffs and combined for modified in vitro and pepsin-acid multi-enzymatic technique for the first dataset 1 Dependent means for modified Tilley & Terry (MT) and pepsin-acid multi-enzymatic (PME) methods, organic matter digestibility (OMD), in vivo mean is true mean; 'Regression F-prob' refers to the probability associated with the overall regression F-statistic; Significance legend: *5%; **1%; ***0.1% level of significance; ns: not significant Table 2) at 0.49 and 0.73 for the MT and PME techniques, respectively.The low number of samples may have contributed to the low R 2value.

Table 2
Predicted in vivo organic matter digestibility mean values, predicted with estimates of modified in vitro or pepsin-acid multi-enzymatic values of first dataset, regressed against observed in vivo values second dataset: complete diets, selected feedstuffs and combined feeds

Table 3
Improved organic matter digestibility means and regression equations to estimate in vivo measurement in complete diets, selected feedstuffs and combined, for modified in vitro and multi-enzymatic technique method 1 Dependent means for modified Tilley & Terry (MT) and pepsin-acid multi-enzymatic (PME) methods, organic matter digestibility (OMD), in vivo mean is true mean; 'Regression F-prob' refers to the probability associated with the overall regression F-statistic; Significance legend: *5%; **1%; ***0.1% level of significance; ns: not significant

Table 5
Linear regression with qualitative variable included in the model:y i = β 0 + β 1 x1 i + β 2 x2 i + β 3 x1 i x2 i + e iSame population assumption verification in the improved prediction of in vivo organic matter digestibility of feedstuffs and complete diets using modified in vitro Tilly & Terry and pepsin-acid multi-enzymatic methods 1 root mean square error; 'Regr.F prob' refers to the probability associated with the overall regression F-statistic 'separate slope' refers to separate slope-effect; 'separate intercept' refers to separate intercept-effect ns: represents non significance; *represents 5% level of significance; **represents 1% level of significance ***represents 0.1% level of significance;2 Shapiro-Wilks indicates normally distributed residuals, which is confirmed by normal probability plot (not included);3 modified Tilley & Terry (MT); 4 pepsin-acid multi-enzymatic (PME) method