Comparison of alternative neutral detergent fiber methods to the AOAC definitive method

Neutral detergent fiber (NDF) is the most commonly reported metric for fiber in dairy cattle nutrition. An empirical method, NDF is defined by the procedure used to measure it. The current definitive method for NDF treated with amylase (aNDF) is AOAC Official Method 2002.04 performed on dried samples ground through the 1-mm screen of a cutting mill with refluxing and then filtration through Gooch crucibles without (AOAC−; reference method) or with (AOAC + ) a glass fiber filter filtration aid. Other methods in use include grinding materials through the 1-mm screen of an abrasion mill, using filtration through a Buch-ner funnel with a glass fiber filter (Buch), and use of the ANKOM system (ANKOM Technology, Macedon, NY) that simultaneously extracts and filters samples through filter bags with larger (F57) or smaller (F58) particle size retentions. Our objective was to compare the AOAC and alternative methods using samples ground through the 1-mm screens of cutting or abrasion mills. Materials analyzed were 2 alfalfa silages, 2 corn silages, dry ground and high-moisture corn grains, mixed grass hay, ryegrass silage, soybean hulls, calf starter, and sugar beet pulp. Samples were run in duplicate in replicate analytical runs performed on different days by experienced technicians. Compared with cutting mill–ground samples, the aNDF% of dry matter results from abrasion mill–ground samples were or tended to be lower for 8 of 11 samples. Method affected aNDF% results for all materials, with method × grind interactions for 6 of 11 samples. For ash-free aNDF% assessed with cutting mill–ground materials, a priori selected contrasts showed that the number of materials for which methods differed or tended to differ from the AOAC methods were 4 (Buch), 8 (F57), and 3 (F58); and 3 for AOAC– versus AOAC + . However, statistically different does not necessarily mean substantially different. For a given feed and grind, a positive value for the absolute difference between the


INTRODUCTION
In the Journal of Dairy Science, in other peer-reviewed publications, and in applications and publications in the field, NDF is the most commonly reported metric for fiber in dairy cattle nutrition.It is used to calculate nutrient supply to dairy cattle (NASEM, 2021).The assay for NDF is an empirical methodthe method itself defines the analyte-so it must be a well-defined assay that is strictly adhered to in the laboratory to achieve acceptable results.The current definitive reference method for amylase-treated NDF (aNDF) analysis is AOAC Official Method 2002.04 (Mertens, 2002).This method uses dried samples ground to pass the 1-mm screen of a cutting mill (e.g., Wiley mill) or the 2-mm screen of an abrasion mill (e.g., Cyclone mill) and was evaluated with filtration through coarse-porosity Gooch crucibles with no filtration aids.Test samples were refluxed with neutral detergent, Na sulfite, and heat-stable α-amylase, and soaked in hot water and acetone to obtain aNDF residues.Commercial analytical laboratories often use abrasion mills with 1-mm screens, which are needed to produce samples for which near-infrared reflectance spectroscopy (NIRS) evaluation of feeds is calibrated (Abrams, 1989).However, this grind generates smaller particles than 1-mm screens in cutting mills (Mertens, 2002).The AOAC Official Method 2002.04 [Mertens, 2002;p. 1227(e)] describes a modification that includes glass fiber filters or sand in crucibles as filtration aids for materials that contain fine particles.This modification may be appropriate for materials ground through 1-mm screens in abrasion mills, but this has not been evaluated.
An empirical analyte like NDF requires that values generated by method variants for a given material are comparable with the results of the reference method.Otherwise, correct application and interpretation of results are in doubt.Differences among methods for the same analyte affect comparability of research findings, applicability of previously developed equations that use NDF, evaluation of forages, diet formulation for herds, and development of NIRS calibrations for feed analyses that are based on the chemical analyses.Over the more than 50 years that the NDF analysis has been developed (Van Soest and Wine, 1967;Goering and Van Soest, 1970), refined (Van Soest et al., 1991), and formally defined and evaluated (Mertens, 2002), other method variants have been developed to enhance ease of use of the assay and sample throughput in laboratories.However, these method variants have not typically been formally tested and the results published to establish their reproducibility across feeds and laboratories (Fahey et al., 2018) or their comparability to the existing reference assay (AOAC Official Method 2002.04;Mertens, 2002).The performance of NDF methods has potential to differ among materials and grinding methods; therefore, a diverse set of materials and grinds of interest should be used to evaluate the methods.
The objective of this study, undertaken at the encouragement of the National Forage Testing Association, was to compare alternative variants of NDF analysis to the AOAC method performed without or with a filtration aid.Dried samples were ground to pass the 1-mm screen of a cutting mill or abrasion mill, as is commonly done in commercial laboratories.A broad range of materials was evaluated because the NDF methods are applied to both forages and nonforages used in animal diets.The number of materials was limited to 11 to allow duplicate test samples and blanks to be analyzed in each run for all methods.

MATERIALS AND METHODS
The 11 feed materials used were 2 alfalfa silages, 2 corn silages, dry ground and high-moisture corn grains, mixed grass hay, winter ryegrass silage, soybean hulls, pelleted calf starter, and sugar beet pulp shreds (Table 1).Dried feeds were ground through the 6-mm screen of a cutting mill (Wiley mill, Arthur H. Thomas Co., Philadelphia, PA).The feeds were tumbled on brown paper to mix, as described in Mertens (2002).After tumbling, a feed was divided in half with a spackling tool, with half ground through the 1-mm screen of a cutting mill (Wiley mill) and half ground through the 1-mm screen of an abrasion mill (Cyclone mill, UDY Corp., Fort Collins, CO).Ground materials were stored in sealed containers until analysis.
Feed materials were analyzed for DM by drying samples overnight at 105°C in tared hot-weighed beakers in a forced-air oven and hot weighing the dried materials.Ash was determined on samples incinerated at 500°C  Hall, 2015).
The AOAC reference method, its modification using a filter aid, and 3 variant NDF methods were tested for measurement of heat-stable, α-amylase-treated NDF inclusive of ash as a percentage of DM (aNDF%) and corrected for ash (aNDFom%).Treatments are designated in this article as NDF method/grind treatment.AOAC Official Method 2002.04 with refluxed residues filtered through a fritted Gooch crucible only (AOAC−) is the reference method, and we also evaluated a modification of that method described for use with materials with fine particles that used a 4.25-cmdiameter Whatman GF/D glass fiber filter (Cytiva Life Sciences, Shrewsbury, MA) placed over the frit as a filter aid (AOAC+).An alternative method (Buch) filtered test samples through a 110-mm diameter Whatman 934-AH glass fiber filter (Cytiva Life Sciences) set in a Buchner funnel designed to hold 90-mm-diameter filter paper.A 10-cm (exterior diameter) polyvinylchloride tube with a beveled edge at the bottom was inserted to hold the filter against the bottom of the funnel and contain fibrous residues on the filter.For the AOAC−, AOAC+, and Buch methods that used separate refluxing and filtration, the procedures described in AOAC Official Method 2002.04 (Mertens, 2002) were followed, except for the filtration vessel used for Buch.Another alternative method (F57 method) used simultaneous extraction and filtration with agitation in a sealed, heated, pressurized vessel (ANKOM 200, ANKOM Technology, Macedon, NY) with test samples held, extracted, and filtered in 25-µm particle retention filter bags (F57; ANKOM Technology).After initial runs, an additional type of filter bag with 6-µm to 9-µm particle retention (F58; ANKOM Technology) was included in the study.Some details of filters and reagents used for the NDF methods are given in Figure 1, and details of the methods are listed below.For each NDF method, test samples for all materials of both grind methods and reagent blanks were analyzed in duplicate in at least 2 analytical runs performed on different days.The filter bag extraction units hold 24 filter bags.Test samples of a single grind were used in each filter bag analytical run, allowing for all 11 materials of that grind plus reagent blanks to be run in duplicate in a given run.Two technicians, each with more than 10 years of experience analyzing materials with the AOAC or filter bag methods, performed all analyses.
Single, common batches of neutral detergent solution, heat-stable α-amylase (Thermostable amylase HTL, ∼8,300 bacterial amylase units; Bio-Cat Inc., Troy, VA), and Na sulfite were used for all runs for all methods.Two different batches of ACS-grade acetone were used, 1 for F57 and F58 and 1 for the other analyses.Three batches, approximately 18 L each, of neutral detergent solution were prepared, and one-third of each was mixed in 18-L carboys to create equally commingled batches.Each individual batch contained 17.82 L of distilled water, 72 g of NaOH, 263 g of EDTA, 82.1 g of Na 2 HPO 4 , 122.6 g of Na 2 B 4 O 7 •10H 2 O, 540 g of sodium lauryl sulfate, and 180 mL of triethylene glycol.
The measured pH values of the commingled neutral detergent solution in 3 carboys were 7.02, 7.03, and 7.01.
In the initial plan for the experiment, AOAC−, AOAC+, Buch, and F57 were performed with test samples and reagent blanks in duplicate within 2 replicated analytical runs.Additionally, 2 other filter bag extraction units were used in single analytical runs to provide data on the comparability of results from different units.Upon analysis of the data, we discovered that aNDF% values were lower and more variable with the initial F57 extraction unit than with the other methods, and the blank filter bags increased appreciably, but variably, in weight (data not shown).After discussions with the manufacturer, we found that the distance of travel of the bag suspender in the initial extraction unit was greater than recommended by the manufacturer, possibly resulting in greater flow of reagent through the bags.The data from that unit were omitted from the study.The other 2 extraction units complied with manufacturer's specifications for travel distance of the bag suspender and temperature of the extraction vessel and were used to complete the study.Data from both units were combined for statistical analysis.Given the constraint that the same commingled batches of neutral detergent had to be used for all analyses, we were limited on how many additional runs could be made.Only aNDF% was obtained for abrasion mill-ground materials, but both aNDF% and aNDFom% were determined on cutting mill-ground materials.

AOAC Official Method 2002.04
The reference method in this study (AOAC−) followed AOAC Official Method 2002.04, and AOAC+ used the option with the glass fiber filter filtration aid.Dry weights of empty, coarse-porosity Gooch crucibles, with or without glass fiber filters inserted, were determined after drying at 105°C in a forced-air oven and hot weighing.Test samples (0.5 ± 0.0008 g) were weighed, weights recorded to the nearest 0.0001 g, and samples transferred into tall-form Berzelius beakers without spouts.Sodium sulfite (0.5 ± 0.1 g) was added to each test sample using a "pinch" (0.31 mL) measuring spoon.Neutral detergent solution (50 mL) was added, the test sample swirled, and the beaker placed in a preheated reflux apparatus.The test sample with reagents was heated to boiling within 4 to 5 min, and 2 mL of working amylase solution was added, the sample swirled, and replaced in the refluxing unit.After 5 to 10 min, a fine stream of neutral detergent was used to rinse particles from the sides of the beaker back into the solution.Samples were refluxed for 60 min.Once removed from the reflux unit, the sample was allowed to settle for 30 to 60 s.A crucible was preheated with boiling distilled water, which was removed under low vacuum before detergent was decanted and fibrous residues were filtered under low vacuum, with the vacuum shut off before residues became dry.A stream of boiling water was used to rinse residue particles from the beaker into the crucible, filling the crucible approximately half way.Amylase working solution (2 mL) was added to the crucible, stirred with a Teflon stir rod, and allowed to react for a minimum of 45 to 60 s.The liquid in the crucible was evacuated under vacuum, and the remaining residue in the beaker was dislodged with a rubber policeman, and rinses of the beaker were transferred to the crucible.The crucible was filled three-fourths full of boiling distilled water and allowed to soak for 3 to 5 min followed by evacuation under vacuum.If filtration was difficult, 2 mL of amylase working solution was added at this step.The residue in the crucible was subjected to 2 additional 3-to 5-min soaks in 40 to 50 mL of boiling distilled water, and two 3-to 5-min soaks in 40 to 50 mL of acetone, with each soak followed by evacuation.After the final acetone soak, the sample was evacuated to dryness under vacuum, the crucible removed from the filtration rack, and residual acetone allowed to evaporate before transfer of the crucible to a 105°C forced-air oven for drying overnight and hot weighing the next day.Samples were ashed at 500°C for 5 h, tempered at 105°C in a forced-air oven, and hot weighed to determine residual ash.The aNDF% was calculated as 100 × (Wf -We -Bf + Be)/(S × DM%), where Wf = crucible + residue, We = empty crucible weight, Bf = average of reagent blank final crucible weight, Be = average of reagent blank empty crucible weight, S = test sample portion weight, and DM% = DM percentage of the material.All crucible weights are 105°C dry weights.For AOAC+, the filter was included in the crucible weight.For aNDFom%, the We and Be weights were replaced with the hot weights after crucibles with residues or reagent blanks were ashed.

Buchner Funnel Method
Test sample extraction was identical to the AOAC procedure except for the device used for filtration.Autoclavable, polypropylene Buchner funnels designed to hold 90-mm-diameter filter papers were used.A piece of polyvinylchloride tubing with a 10-cm outside diameter and the outside edge at one end beveled at an approximately 45° angle was used to hold a 110-mm diameter Whatman 934-AH glass fiber filter in place.The undried, preweighed filter was centered on top of the funnel and pressed into place with the beveled end of the tube.The filter formed a cup at the bottom of the funnel, with approximately 1 cm of filter extending up around the outside of the tube (Figure 1).The extracted fibrous residue was decanted into the tube in the funnel.Soaks with boiling distilled water or acetone filled the funnel to approximately one-fourth of the volume.The filter with residue was transferred to a preweighed beaker for drying in a 105°C forcedair oven and subsequent ashing at 500°C for 5 h with 105°C hot weights taken at each step.Dry matters were also determined on representative filters to adjust the weight of the filters used to a dry basis.Calculations were the same as for the AOAC methods but with the weight of the glass fiber filter or filter plus residue or ash replacing crucible weights.

F57 and F58 Filter Bag Methods
The filter bag methods were performed according to recommendations of the manufacturer (ANKOM Technology).Filter bag empty weights were recorded undried and at ambient temperature.Test sample (0.5 ± 0.0004 g) was weighed into each bag, and the weight of sample and bag was recorded to the nearest 0.0001 g.Each filter bag was heat-sealed and the test sample was distributed uniformly in the bag to prevent clumping.In each analytical run, only 1 grind type and 1 bag type were evaluated with all 11 materials and reagent blanks.Working from the top and bottom trays toward the center of the bag suspender, 1 of each duplicate test sample and blank bags were placed as far apart vertically as possible in bag suspender trays so that duplicates of materials were not both located at either the top or bottom levels of the bag suspender.The bag suspender containing 24 filter bags was placed in the extraction unit and 2,000 mL of neutral detergent at ambient temperature (23 to 26°C) was added, followed by 20 g of Na sulfite and 4 mL of heatstable α-amylase.The unit heating and agitation were turned on, the lid closed tightly, and the extraction run for 1 h 15 min.The additional 15 min of extraction time compared with the AOAC method 2002.04 was used to allow time for the vessel and contents to come to temperature.Vessel temperature checked at 15 min was between 91 and 100.4°C and was between 99.9 and 100.1°C at 1 h 15 min according to the thermometers built into the extraction units.At the end of the run, pressure was carefully released from the unit and detergent drained from the unit by the hose built in for that purpose.Three 5-min soaks with the unit lid closed and heat and agitation on were performed, with 2 L of heated distilled water added each time and with 4 mL of heat-stable α-amylase added to the first 2 rinses and liquid drained after each rinse.Upon removal from the bag suspender, bags were gently squeezed to remove water and soaked twice for 5 min in acetone in a 250-mL beaker with stir rods between the bags to facilitate access to ac-etone.Bags were gently squeezed to remove excess acetone and placed on a cart for residual acetone to evaporate.Bags were examined for evidence of leakage and would have been omitted if such was obvious, but none was noted.Once dry, bags were moved to a forced-air 105°C oven to dry overnight.The next day, the 24 bags were moved to a plastic sealable bag containing a pouch of desiccant and were allowed to equilibrate to ambient temperature for 10 min before weighing to obtain neutral detergent residue + bag weights for aNDF% determination.For aNDFom% determination, the weighed bags were transferred to previously hot-weighed beakers for subsequent hot weighing at 105°C to determine residue + bag weight, ashed at 500°C for 5 h, and tempered at 105°C in a forced-air oven before hot weighing them to determine residual ash.As per manufacturer's instructions, empty filter bag weights were adjusted for changes in blank bag weights by multiplying bag weights by the average final oven-dried blank bag weight divided by the original blank bag weight; in a given analytical run, the blank bags used for correction were the 2 in that run.The aNDF% was calculated as 100 × [W3 -(W1 × C1)]/W2, where W1 = bag empty weight, W2 = test sample 105°C dry weight, W3 = dried weight of bag and residue, and C1 = blank bag correction.The aNDFom% was calculated with hot weights as 100 × [W3 -(W1 × C2) -W4]/W2, where C2 = average OM percentage of the blank filter bags after NDF processing and W4 = the weight of ash from the filter bag and residue.

Method Notes
In the Buchner funnel method, the polyvinylchloride tubing tended to deform due to the heat of the reagents.Although it did not fail during this experiment, use of a tubing made from a more heat-resistant material to handle the boiling reagents is recommended.
The technician noted that the location of the seal of F58 bags was more precise than that of the F57 bags because of the narrower area in which the F58 seal could be made.Two F58 test samples that had values that were half that of their replicate were omitted.Seven samples were omitted from 1 analytical run of F58 for aNDF% due to apparent malfunction of the computer weighing program.
It has been suggested that pressure buildup in the filter bag extraction vessel expels air from bags, facilitating flow of neutral detergent reagent through them.Tests for the presence of gas bubbles in sealed F57 and F58 filter bags and the effects of pressure on them (data not shown) were made using (1) 0.5 g of mixed grass hay/cutting with bags sealed individually into tubeshaped pressure vessels with 30 mL of neutral detergent and approximately 5 mL of gas headspace and heated to 99°C in a forced-air oven to generate pressure inside the vessel; or (2) filter bags containing zirconia/silica beads with approximately the same volume as each grass hay test sample (0.92 mL) laid flat in the bottoms of separate 125-mL Erlenmeyer flasks, 50 mL of neutral detergent added to each flask, and flasks sealed with black rubber stoppers.Stoppers were held in place manually and 40 to 50 mL of air was injected from a syringe through the stopper via hypodermic needle into the approximately 101-mL headspace.Flasks were calculated to achieve a pressure averaging 146 kPa at ambient temperature, as calculated according to the ideal gas law (PV = nRT, where P = pressure, V = volume, n = amount of gas, R = ideal gas constant, and T = temperature).Air bubbles were visible in F58 but not F57 filter bags before heat or pressure treatments were applied.Upon increasing the internal vessel pressure via heating or air injection, no changes were noted in the presence of gas bubbles in either bag type; the air bubbles did not change in shape or location in the F58 bags.
With the possibility that gas bubbles generated through boiling of the neutral detergent reagent could affect test samples in filter bags if they were not held under pressure during incubation, we determined the boiling point of neutral detergent with Na sulfite added.A tube containing boiling chips, a thermometer, and neutral detergent with Na sulfite (1.15 g added to 100 mL; AOAC Official Method 2002.04,Mertens, 2002) with the tube opening covered tightly with aluminum foil was incubated in a forced-air oven set to 108°C.Gas bubbles evolved from the solution after it reached 102°C.The solutes in the reagent solution gave the reagent a boiling point greater than that of water.At the approximately 100°C temperatures reached in the filter bag extraction unit, gas bubbles related to boiling of the solution would not evolve even if pressure was not maintained and so would not affect the contents of the filter bags.Increased pressure in the extraction vessel may be a useful indicator of adequate sealing of the unit to avoid evaporative loss during extraction.Temperature of an extracting solution can affect efficacy of an extraction.Just as water shows a decline in boiling point as altitude increases (Earl, 1990), so does neutral detergent reagent.An experience in the familiarization stage of the AOAC collaborative study was that a laboratory located at a high altitude obtained higher aNDF values than other laboratories.They determined that the boiling point of the neutral detergent reagent in their laboratory was approximately 95°C (D.Mertens, unpublished data).The laboratory did not participate in the collaborative study.Although the range of acceptable temperatures of the boiling reagent is not specified in the AOAC method, the method was developed below 305 m elevation above sea level (40 m, Beltsville, MD; 254 m, Madison, WI; https: / / en -us .topographic-map .com/map -klvc57/ Beltsville/ and https: / / en -us .topographic-map .com/map -nlgp/ Madison/ ; accessed Dec. 24, 2022).Relative to the boiling point at sea level, pure water shows an approximately 1°C depression in boiling point at 305 m elevation (calculated from Earl, 1990).Analogous to the issues for achieving desired heat processing to ensure the safety of canned foods at elevations above 305 m (Hertzberg et al., 1984;Kendall, 2013), extending the time of refluxing at the lower boiling temperature or achieving a desired temperature by performing the extraction under pressure offer potential solutions for the altitude/temperature issues, but these options must be evaluated for efficacy.

Data Analysis
The experiment was a completely randomized design with a 2 × 3 factorial arrangement of treatments.The aNDF% and aNDFom% responses were analyzed for outliers for each combination of feed, NDF method, and grind using the Dixon Q test (Rorabacher, 1991).Efforts at transformation failed to make the residuals normally distributed when working with the entire data set.Instead, feed materials were analyzed individually for response variables of aNDF% and aND-Fom%.The model for aNDF% included the effects of method, grind, and their interaction, and that for aNDFom% included method.Calf starter data were quartic power-transformed for analysis of aNDFom%.Adjusted P-values of the Tukey-Kramer test were used to compare AOAC−/cutting and AOAC+/cutting to other method variants/grinds.For analysis of aND-Fom% with cutting mill data only, a priori contrasts tested both AOAC variants versus Buch, F57, and F58, and AOAC− versus AOAC+.Analysis of the standard deviations (SD) between duplicate test samples run together within an analytical run were performed for aNDF% and aNDFom% with the data quartic rootand cube root-transformed, respectively.For aNDF%, the model included feed material, method, grind, and all 2-and 3-way interactions.For aNDFom%, which was analyzed only on cutting mill-ground samples, the model included feed material, method, and their interaction.Adjusted P-values of the Tukey-Kramer test used for mean separation on the effects of method were applied to the SD data when that term was significant.Analyses were performed using PROC MIXED of SAS (version 9.4, SAS Institute Inc.).No effects were treated as random.Variances of the NDF methods were assumed not to be equal, and Satterthwaite degrees of freedom were used.Normality of the residuals tested with the Shapiro-Wilk test were all P > 0.05.However, to achieve normality, some data sets required stepwise removal of individual data points designated as "extreme observations" in PROC UNIVARIATE of SAS, otherwise P-values would not be accurate.The extreme observations removed were all from F57 or F58 treatments: for aNDF% analysis, F58 (dry corn, abrasion and cutting mill; low lignin alfalfa, abrasion mill; mixed grass hay, cutting mill; sugar beet pulp, abrasion mill) and F57 (brown midrib corn silage, abrasion mill); for aNDFom% analysis, F58/cutting dry corn grain, conventional alfalfa silage, and mixed grass hay.Significance was declared at P ≤ 0.05 and tendency at 0.05 < P ≤ 0.10.
Another approach used to evaluate patterns in the data was to regress the aNDF% of all feeds for a given NDF method and grind against the mean values for those feeds determined with the reference AOAC-and cutting method.The tests applied were whether the regression slope = 1 and intercept = 0, indicating that the tested NDF method/grind did not differ from the AOAC−/cutting method.The slope, intercept, adjusted R 2 , and root mean square error, number of data points, and P-values for the test are reported.The analysis was performed with PROC REG of SAS.
We also evaluated whether it was likely that aNDF% or aNDFom% results from the NDF method/grind variants would fall within the 2-SD range observed for the AOAC/cutting analyses.For a given material and grind, 2 times the observed SD of the AOAC analysis was subtracted from the absolute value of the difference between the mean of an AOAC analysis and the mean of each other method: {Acceptance criteria for each material = [absolute value (material AOAC mean -material variant mean)] -(2 × SD of material AOAC mean)}.If the result was positive, the other method's result fell outside the range predicted to contain approximately 95% of results generated by the AOAC method in this study.No statistical evaluation was applied.

RESULTS AND DISCUSSION
Approximately 2% of the analytical results were identified as outliers, with 10 of the 517 aNDF% values (7 high: AOAC−: high-moisture corn/cutting, sugar beet pulp/cutting; AOAC+: brown midrib corn silage/ cutting; Buch: dry corn grain/abrasion, ryegrass silage/ abrasion; F58: mixed grass hay/cutting, low-lignin alfalfa silage/abrasion; and 3 low: F58: dry corn grain/ cutting, dry corn grain/abrasion, sugar beet pulp/abrasion), and aNDFom% values of 7 of the 264 cutting mill ground samples (5 high: AOAC−: soyhulls; AOAC+: dry corn grain; F57: conventional corn silage; F58: conventional alfalfa silage, mixed grass hay; 2 low: F58: dry corn grain, mixed grass hay) flagged (P < 0.05).For aNDF%, the outliers were equally split between cutting mill (n = 5) and abrasion mill (n = 5) processed samples.The only material for which outliers were not detected was calf starter.The detected outliers could differ from the next closest point by 0.3 to 5.4% of DM (mean = 2.1%, SD = 1.27%), whereas the maximum difference for analyses not designated as outliers was as high as 2.92% of DM.Closer inspection of the data showed that 11 of the 17 outliers were in data sets where the other measurements were very similar numerically, so even a small difference could designate a point as an outlier by the Dixon Q outlier detection method.None of these measurements were omitted from subsequent analyses on the basis of outlier analysis because removal of the identified outliers would not represent actual method performance.
Analysis of feed materials for aNDF% was affected by grinding method, NDF method variant, and their interaction (Tables 2 and 3).Measurements obtained with materials ground with an abrasion mill were (P < 0.01) or tended to be (P = 0.06) lower than those ground with a cutting mill for 7 of 11 feeds, with only low-lignin alfalfa silage resulting in a slightly higher value (P < 0.01).For materials ground through the 1-mm screen of an abrasion mill, all method variants differed or tended to differ from AOAC−/cutting for at least 2 of the materials, with AOAC−/abrasion differing for 4 (Table 2).The same comparison of abrasion mill-ground materials but against AOAC+/cutting showed a difference for at least 2 materials.For these comparisons, F57/abrasion differed or tended to differ for 9 and 8 materials, respectively.For materials ground through the 1-mm screen of a cutting mill, each method variant showed differences or tendencies for difference against AOAC− or AOAC+/cutting for 0 or 1 material, except for F57/cutting, which differed or tended to differ for 6 materials.The number of detected differences may be conservative, because the mean separation technique used is less sensitive to detecting differences than are orthogonal F tests.Abrasion mills produce a finer particle size than do cutting mills when the same size screen is used (Mertens, 2002).The lower aNDF% values for the abrasion mill-ground materials could be the result of greater extraction by neutral detergent or greater escape of fine particles through the filters.Evaluation of the SD between paired replicates for aNDF detected differences among methods (P < 0.01), a tendency for an effect of method × grind (P = 0.09), but did not show an effect of grind (P = 0.54; Table 4; Table 2, numeric SD averages by method).Based on mean separation, F57 replicate SD differed from those of AOAC−, AOAC+, Buch, and F58 (P < 0.03), which did not differ from each other (P > 0.10).The tendency for a method × grind interaction (P = 0.09) could relate to the numerically higher value for F57/abrasion than for F57/cutting, whereas other NDF methods were more numerically similar between grinding methods.The potential for greater escape of particles could be the basis for the numerically greater variability in aNDF% results for the F57 filter bag method.
Method variant used for aNDF% affected results from all materials (P < 0.01), with F57 giving numerically lower values for all materials except for calf starter (Table 3).An NDF method × grind interaction was detected for 6 of 11 materials.This appears to be largely a function of the F57 method resulting in a greater decline in aNDF% between cutting and abrasion mill use for the corn grains and corn silages, but an increase for conventional alfalfa silage and calf starter when other method measurements were not changing much or were declining (Table 2).
Regression analysis of the aNDF% values for all feed materials of a specific method/grind against the mean values for AOAC−/cutting materials showed a pattern similar to the previous analysis (Table 5, Figure 2).If variants agree perfectly with the reference method, the regressions have coefficients that are not statistically different from a slope of 1 and intercept of 0. The regressions for AOAC−, AOAC+, Buch, and F58 results with cutting mill grinding and AOAC−, Buch, and F58 results with abrasion mill grinding did not differ from a slope of 1 (P > 0.11).The intercepts did not differ from 0 for AOAC− and F57 with cutting mill grinding or for any method applied to abrasion mill ground materials (P > 0.18).The differences in slope for F57 with both grinds and AOAC+/abrasion indicate that the values from these analyses were lower than that of AOAC−/ cutting as aNDF% increased.That relationship may be more a function of the physical characteristics of materials across the aNDF% range (e.g., corn grain vs. grass) than due to the percentage of aNDF itself.

Positive intercepts for methods indicate that aNDF%
Hall and Mertens: EVALUATION OF NEUTRAL DETERGENT FIBER METHODS recovery is increased consistently across all materials compared with AOAC−/cutting grind.The smaller particle retention sizes of AOAC+, Buch, and F58 filters compared with AOAC− are the likely basis for the positive intercepts with cutting mill-ground materials (P < 0.10).The failure to see similar responses with abrasion mill-ground materials and the negative, nonsignificant intercept values suggest greater loss of fine particles through the filters with that grind.
The aNDFom% analyses on cutting mill-ground materials showed a pattern similar to that seen with aNDF% (Table 6).The NDF method used affected aNDFom% results for 10 of the 11 materials.Method variants differed or tended to differ from the AOAC methods for 4 materials for Buch, 8 for F57, and 3 for F58.However, even the AOAC− and AOAC+ methods differed or tended to differ from each other for high-moisture corn grain, ryegrass silage, and calf starter pelleted feed, with AOAC+ resulting in higher aNDFom%.These materials had potential to have a finer particle size upon grinding, which would warrant the use of a filtration aid to capture particles and facilitate filtration.Unlike aNDF%, the method SD for aNDFom% with cutting mill processing did not differ  b Values differ from AOAC+/cutting for that material (P < 0.05). x Values tend to differ from AOAC−/cutting for that material (0.05 ≤ P < 0.10). z Values tend to differ from AOAC+/cutting for that material (0.05 ≤ P < 0.10).
1 AOAC− = AOAC method without filtration aid filter; AOAC+ = AOAC method with filtration aid filter; Buch = Buchner funnel with filter; F57 = filter bag method with 25-µm retention filter bags; F58 = filter bag method with 6-to 9-µm retention filter bags.Test samples were cold weighed with desiccant for filter bag methods, and hot weighed for other methods.
by method (P = 0.29) but were numerically greatest for F57 (Table 4).That aNDF% and aNDFom% were statistically different between variants and AOAC NDF methods does not necessarily mean that the methods were substantially different, in the same way that statistical differences in an animal study may not translate to physiologically important differences.If the absolute difference between means of AOAC/cutting (the NDF definitive methods) and NDF method/grind variants did not exceed 2 times the SD of the AOAC method, the variant measurements could be deemed acceptable as falling within bounds of 95% of the results that could be generated by the AOAC method.The acceptability of variant method aNDF% measurements varied by method and grind (Table 7).Relative to AOAC−/cutting, the number of materials exceeding the 2-SD limit for aNDF% increased numerically with abrasion compared with cutting mill grinding, even with AOAC−, and was numerically largest for F57 with either grind.The same pattern was seen for aNDF% with AOAC+/ cutting used as the standard.The pattern of responses among variants did not change appreciably for aN-DFom% performed on cutting mill ground materials (Table 8).
The primary technique differences among the NDF variants tested relate to filtration.In the filtration systems applied to NDF, fiber particles are captured in a matrix of other extracted particles on the filter or in the pores of the filter, which can block passage of other particles, or if small enough, may pass through the filter and escape (Ripperger et al., 2013).Retention or escape of particles is affected by the filter's pore size and thickness, with thicker filters having potential to retain more solids (Xiao et al., 2018).The amount of time that particles suspended in liquid can interact with the filter may also play a role.In the AOAC−, AOAC+, and Buch methods, test samples are refluxed  1 AOAC− = AOAC method without filtration aid filter; AOAC+ = AOAC method with filtration aid filter; Buch = Buchner funnel with filter; F57 = filter bag method with 25-µm retention filter bags; F58 = filter bag method with 6-to 9-µm retention filter bags; SED = standard error of the difference.AOAC− = AOAC method without filtration aid filter; AOAC+ = AOAC method with filtration aid filter; Buch = Buchner funnel with filter; F57 = filter bag method with 25-µm retention filter bags; F58 = filter bag method with 6-to 9-µm retention filter bags.
for 1 h before being decanted and rinsed into crucibles or a Buchner funnel with filter, soaked twice each with boiling distilled water and acetone, with vacuum used to remove liquid after each step.When not under vacuum, little or no liquid passes through the filters.The extracted particles thus have 6 chances to be drawn through the filter with the liquid under vacuum and may be blocked by particles already in the pores of the filter or by a layer of particles on top of the filter (Figure 3a).Use of filter mats in Gooch crucibles (AOAC+) was listed as an option in the AOAC reference method as a filtration aid for work with finely particulate materials.This approach sometimes resulted in slightly higher aNDF% or aNDFom% values than the reference method with finely ground materials, but that is, in part, what this modification was designed to do.Use of filters with smaller particle retention as in AOAC+, Buch, and F58 also appeared to improve agreement of aNDF% results for abrasion mill-ground materials with the reference method.The smaller particle retention sizes of the filters in Buch, despite being a thinner filter, and in F58 gave the methods good agreement with AOAC+ for aNDF% and aNDFom%, particularly with cutting mill-ground samples (Tables 2, 6, 7, and 8).
In filter bag methods F57 and F58, in which test samples are simultaneously extracted and filtered, particles interact with the filter bag and liquid during the 1.25-h extraction and subsequent water and acetone rinses.This may allow particles time to reorient and have greater potential to escape the bag with  Values differ from AOAC+/cutting for that material (P < 0.05). x Values tend to differ from AOAC-/cutting for that material (0.05 ≤ P < 0.10).
1 AOAC− = AOAC method without filtration aid filter; AOAC+ = AOAC method with filtration aid filter; Buch = Buchner funnel with filter; F57 = filter bag method with 25-µm retention filter bags; F58 = filter bag method with 6-to 9-µm retention filter bags; aNDFom% = ash-free aNDF as a % of DM; M = method; SED = standard error of the difference.the liquid flow or to block other particles from leaving (Figure 3b).The form of the particles, such as needlelike from grasses or shorter and more cuboidal from legumes (Van Soest, 1994), will alter their narrowest cross-sectional diameter and their potential to escape, given their spatial orientation relative to the filter pore.
Hall and Mertens: EVALUATION OF NEUTRAL DETERGENT FIBER METHODS If pores are blocked by particles, passage of reagent through the bag for extraction could also be limited.
The larger particle retention size of F57 filter bags and their overall lower aNDF% and aNDFom% and greater aNDF% variability relative to other methods suggest that escape of particles is the basis for the lower values.Following the same logic, the better comparability to AOAC methods of the F58 filter bags should relate to its smaller particle size retention.It is possible that use of a thicker material for the filter bags combined with the smaller particle retention and pore size of F58 could improve agreement with AOAC+ for cutting mill-processed materials, but this remains to be tested.We searched for comparable NDF method comparison studies and found they were either performed before the AOAC NDF method was released (e.g., Vogel et al., 1999) or tested modified versions of the filter bag (e.g., Barbosa et al., 2015) or of the AOAC and filter bag methods (e.g., Schlau et al., 2021).Testing unmodified reference or variant NDF methods was the objective of the present study.Although we selected a variety of materials to represent different types of feeds (grains, by-products, and forages) that varied in aNDF from 6 to 60% of DM, the material set was far from complete to encompass the variety of substances ana-lyzed for aNDF.Our findings that NDF method and feed grinding variants interacted with material to affect NDF results suggest that a greater number and type of materials should be used to evaluate these and other variants of NDF methods to determine whether their performance is comparable to the reference method or if improvements in the variants are needed.A greater variety of the feeds tested, other forages, concentrates, feces, in vitro fermentation residues, and more would be useful to include in further method evaluation, with choices depending on the scope of application of the results.Given the extensive use of abrasion mills for material grinding, the effects of the particle size of different materials and grinding methods on aNDF results should be further addressed if such application of abrasion mills is continued.
The decision on whether to use the filtration aid option of the AOAC reference method (Mertens, 2002) requires knowledge of the ground material.The filtration aid captures small fiber particles and facilitates filtration with samples that grind to or are already composed of very fine particles.A practical consideration for whether or not to use a filtration aid with the AOAC reference method is that samples delivered to a commercial laboratory may be accompanied by limited or no useful information with which to decide in advance whether use of a filtration aid would be advised.With limited information on whether a sample requires a filtration aid, the use of a glass fiber filter, as in AOAC+, may be worthy of consideration for general use.
The ongoing maintenance and testing of Gooch crucibles to give the filtration performance required for the AOAC NDF method is a challenge.With use, crucibles can become clogged and filter more slowly, or degrade, lose frit material, and filter too rapidly, with both situations suggesting a change in the particle retention of the crucibles.Single-use filters such as the filter bags and the glass fiber filter in the Buch method are attractive alternatives if the NDF analysis results are comparable to a properly performed AOAC method for materials of interest.Less time is required for their maintenance, and they have increased ease of use having a larger surface area (Buch) or not requiring manual filtration (filter bags).However, attention must be paid to the quality control that manufacturers apply to the filters.Changes in weights of individual filters and of results from control materials run with new batches of filters would provide laboratories with quality control metrics to monitor filter performance.

CONCLUSIONS
As an empirical test, analyses for aNDF must agree with those generated by the reference method, AOAC Official Method 2002.04, to be considered accurate.For aNDF analysis of materials ground through 1-mm screens in cutting mills, the AOAC option that used filter mats in Gooch crucibles and the variants using filter mats in Buchner funnels or F58 filter bags generated aNDF% values with the fewest differences compared with the reference method with the materials tested.The F57 filter bags differed from the reference method and filtration aid option for most of the materials tested with either grinding method.Given the limited number of materials evaluated in the present study, a greater number and more diverse materials should be used to assess agreement of the methods with the reference method.Compared with materials ground through the 1-mm screen of a cutting mill, materials ground through the 1-mm screen of an abrasion mill produced more aNDF% results that were lower than the reference method, but with fewer differences noted when filter particle retention size was smaller.If use of the 1-mm abrasion mill grind is desired, additional research is needed to develop suitable NDF method modifications, possibly with filtration membranes that retain finer particles, to improve comparability with the reference method.The variability of responses among feed materials to variants in NDF and grinding methods would seem to preclude development of adjustment factors to attain agreement with the reference method.
Hall and Mertens: EVALUATION OF NEUTRAL DETERGENT FIBER METHODS Hall and Mertens: EVALUATION OF NEUTRAL DETERGENT FIBER METHODS Hall and Mertens: EVALUATION OF NEUTRAL DETERGENT FIBER METHODS
Hall and Mertens: EVALUATION OF NEUTRAL DETERGENT FIBER METHODS

Figure 3 .
Figure 3. Flow and filtration of liquid and particles in (a) a filter or crucible into which residue and extractant are poured to be filtered under vacuum, and (b) a filter bag submerged in extractant with vertical agitation in which extraction and filtration occur simultaneously.

Table 1 .
Hall and Mertens: EVALUATION OF NEUTRAL DETERGENT FIBER METHODS Compositional analysis of dried feed materials by grinding method 5 h followed by tempering at 105°C in a forced-air oven and hot weighing.Nitrogen was determined by combustion analysis (TruMac CN, Leco Corp., St. Joseph, MI) with CP calculated as N × 6.25.Starch was analyzed according to AOAC Official Method 2014.10 ( 1 HM = high moisture; conv.=conventional; BMR = brown midrib; LL = low lignin.2DM is % of air dry material, other analytes are % of DM.for

Table 2 .
Arithmetic means, SD, and number of analyses (N) for amylase-treated NDF inclusive of ash as a percentage of DM (aNDF%) determinations with 1-mm cutting or abrasion mill grinding 1 a Values differ from AOAC−/cutting for that material (P < 0.05).

Table 3 .
Effects of grinding and analysis methods (values are least squares means) and the interaction on amylase-treated NDF inclusive of ash as a percentage of DM (aNDF%) measures 1

Table 4 .
Comparison among NDF and grinding methods (values are least squares means) of standard deviations between duplicates within analytical run for aNDF% and aNDFom% analyses 1

Table 6 .
Effect of analysis method on measures of aNDFom% for 1-mm cutting mill-processed samples with arithmetic means, SD, and number of analyses performed (N) 1 b

Table 7 .
Absolute difference between means of AOAC methods determined with 1-mm cutting mill-ground samples and AOAC and alternative methods for aNDF% of DM determined with 1-mm cutting mill-or abrasion mill-processed samples minus 2 times the standard deviation of the AOAC−/cutting mill methods 1 2 HM = high moisture; conv.= conventional; BMR = brown midrib; LL = low lignin.

Table 8 .
Absolute difference between means of AOAC and AOAC alternative methods for aNDFom% (ash-free aNDF) of DM minus 2 times the standard deviation of the AOAC methods for 1-mm cutting mill ground samples 1