Peter Koehler, 1 Theresa Schwalb, 1 Ulrike Immer, 2 Markus Lacorn, 2 Paul Wehling, 3 and Clyde Don 4

AACCI Approved Methods Technical Committee Report: Collaborative Study on the Immunochemical Determination of Partially Hydrolyzed Gluten Using an R5 Competitive ELISA Peter Koehler, 1 Theresa Schwalb, 1 Ulrike Immer, 2 Markus Lacorn, 2 Paul Wehling, 3 and Clyde Don 4 ABSTRACT In 2008, the AACC International Protein Technical Committee (now the Protein and Enzymes Technical Committee) initiated a collaborative study of a method for determining gluten in fermented products using an R5 competitive ELISA system. The method has been approved as AACCI Approved Method 38-55.01. The new method has been validated for testing fermented foods and beverages to determine whether they conform to the newly defined Codex threshold of 20 mg of gluten/kg in total for gluten-free products. Gluten is a protein fraction found in wheat, rye, barley, oats, and their crossbred varieties and derivatives thereof to which some people are intolerant; it is insoluble in water and 0.5 mol NaCl/L (4,5). Prolamins are gluten fractions that can be extracted with 40 70% ethanol. The prolamins gliadin, secalin, hordein, and avenin are found in wheat, rye, barley, and oats, respectively (4). The prolamin content of gluten is generally taken as 50% (4). The celiac activity of gluten protein in oats is still under discussion, and the Codex standard notes that the allowable level for oats in foods not contaminated with wheat, rye, or barley may be determined at the national level. In foods labeled as gluten-free, the gluten level must not exceed 20 mg/ kg of food (4,5). Foods processed to reduce their gluten content to a level ranging from 20 to 100 mg/kg may not be labeled as gluten-free ; in such cases, labeling is regulated on a national level (e.g., these foods could be labeled as very low gluten ). From these regulations, it is obvious that effective test methods are needed to determine the gluten concentration in foods, beverages and raw materials. The Working Group on Prolamin Analysis and Toxicity (PWG) focused on improving the ELISA methodology for gluten analysis because the existing methods were inadequate with respect to sensitivity and reliability (3). Collaboration between the PWG and the research group headed by Enrique Méndez at the University of Madrid led to improved ELISA methods that use both sandwich and competitive assay systems and are based on the monoclonal R5 antibody. This antibody raised against the -type of rye prolamins ( -secalins) is directed toward the epitope glutamine-glutamine-proline-phenylalanine-proline (QQPFP) in gliadins, hordeins, and secalins. R5 ELISA is commercially available in two versions, as a sandwich ELISA for intact gluten proteins with at least two binding epitopes and as a competitive ELISA for partially hydrolyzed gluten (gluten peptides) that needs only one epitope for binding. Although sandwich ELISA has been studied extensively (8,9), leading to its 1 German Research Center for Food Chemistry, Freising, Germany. 2 R-Biopharm AG, Darmstadt, Germany. 3 General Mills, Inc., Minneapolis, MN, U.S.A. 4 Foodphysica, Driel, The Netherlands. http://dx.doi.org/10.1094/cfw-58-3-0402 2013 AACC International, Inc. approval as AACCI Approved Method 38-50.01 (1), the competitive R5 ELISA method has not been validated to date. The R5 sandwich ELISA method is not as sensitive as the competitive ELISA format for detecting partially hydrolyzed prolamins due to the fact that the sandwich ELISA needs two binding sites (7). The competitive assay is the method of choice for measuring partially hydrolyzed gluten in foods. Following the guidelines of the AACC International (AACCI) Approved Methods Technical Committee (11), an international collaborative study was set up to validate R5 competitive ELISA (RIDASCREEN Gliadin competitive R7021, R-Biopharm) as an AACCI Approved Method for gluten/prolamin quantitation in fermented foods and beverages. The study was carried out as a collaboration between the PWG and AACCI. It was coordinated by Peter Koehler (German Research Center for Food Chemistry, chair of the PWG, and member of the AACCI Protein and Enzymes Technical Committee) in close collaboration with Clyde Don (chair of the AACCI Protein and Enzymes Technical Committee). The analytical performance of this method is reported in this article. ELISA Kit and Calculation Software An R5 competitive ELISA kit (RIDASCREEN Gliadin competitive R7021) designed for the quantitation of gluten in fermented foods and software (RIDASOFT Win Z9999, R-Biopharm) for constructing calibration curves (cubic spline fitting) and calculating prolamin concentrations from measured optical densities (OD) were used. A cubic spline is a curve constructed of piecewise third-order polynomials that pass through a number (m) of control points. The second derivative of each polynomial is commonly set to zero at the end points of the pieces. This provides a boundary condition that completes the system of m 2 equations. It produces a natural cubic spline and leads to a simple tridiagonal system that can be solved easily to give the coefficients of the polynomials (13). In this way a function with a continuous curvature over the entire range is obtained. The third derivative is used as a smoothing factor in the calibration curves to determine the extent of interpolation. Lower factors lead to more approximation; higher factors (>100) lead to more interpolation of the curve function. The RIDASOFT software uses a factor of 10,000. To minimize boundary effects and allow extrapolation, two additional control points are added to the set of control points as the starting and end points, where the starting point is near zero and set to x 0 = 0.001 and y 0 = OD (lowest Standard 1) and the virtual end point is determined by calculating the linear regression of the other control points by assuming that x n has the same distance to x n 1 as x 1 has to x 0. Because the cubic spline model did not provide concentration values for samples below the lowest standard, a second-order polynomial curve-fitting model was used to determine values for samples 1 and 4 and to enable calculation of the limit of detection (LOD). 154 / MAY JUNE 2013, VOL. 58, NO. 3

Participating Laboratories All laboratories participating in the collaborative study were required to be familiar with immunological tests and, if possible, with competitive ELISA tests. They were advised to use a separate test room for the collaborative study due to the low detection limit and the possibility of contamination. To check the samples, test requirements, and documentation and to identify critical points, a precollaborative study with four laboratories in Europe was completed before the full collaborative study. Encouraging results were obtained in the prestudy. Only minor changes in the study design were required, and the full collaborative study went on as scheduled. The labs were given six weeks to perform the analyses (August 1 to September 15, 2011). Sixteen labs were selected (designated A to P): one each in Argentina, Austria, Belgium, Canada, Finland, Hungary, Ireland, Italy, New Zealand, Sweden, and Switzerland; two in Germany; and three in the United States. Samples and Sample Preparation The following samples were prepared or obtained for the collaborative study: 1) Beer, gluten-free 2) Beer, 15 mg of prolamin/kg 3) Beer, 50 mg of prolamin/kg 4) Starch syrup, gluten-free 5) Starch syrup, naturally gluten-contaminated 6) Sourdough, 35 mg of prolamin/kg 7) Sourdough, 75 mg of prolamin/kg Examples of some samples are shown in Figure 1. All ingredients, except barley prolamin hydrolysate and rye sourdough, were confirmed to be free of gluten contamination before use by R5 competitive ELISA, which was also used in this collaborative study. Peptic-Tryptic Hordein Digest. Barley cv. Barke grain was milled into white flour (0.50 0.60% [dm] ash content) using a laboratory mill and a 0.2 mm sieve. Flour (200 g) was dispersed Fig. 1. Samples used in this study: A, gluten-free beer; B, naturally glutencontaminated starch syrup; and C, dried sourdough. twice in 600 ml of light petroleum (40 60 C boiling range) and stirred for 30 min at room temperature ( 20 C). The solvent was removed, and the residue was air-dried overnight on a filter sheet. The defatted flour (50 g) was extracted stepwise with 3 200 ml of buffer (0.4 mol NaCl/L and 0.067 mol HKNaPO 4 /L at ph 7.6) followed by 3 200 ml of 60% (vol/vol) aqueous ethanol by homogenizing in a centrifuge vessel for 5 min at room temperature. Each suspension was centrifuged for 30 min at 3,550 g and 4 C, and the supernatants were decanted and combined. The combined ethanol extracts were dialyzed against tap water containing 0.01 mol acetic acid/l and freeze-dried to provide the hordein fraction (equivalent to barley prolamin). Hordein (0.5 g) was suspended in 10 ml of distilled water, and the ph was adjusted to 1.8 with 1.0 mol HCl/L (6). Next, 2.5 mg of pepsin (Merck No. 7192) was added, and the suspension was stirred for 4 hr at 37 C. After adjusting the ph to 7.8 with 1.0 mol NaOH/L, 2.5 mg of trypsin (Merck No. 24579) was added. After further stirring for 4 hr at 37 C, the ph was adjusted to 4.5 with 1.0 mol HCl/L, and the suspension was centrifuged at 4,000 g for 20 min at room temperature. The supernatant was decanted and freeze-dried to provide the peptic-tryptic (PT) hordein digest. The crude protein contents (N 5.7) of hordein and the PT hordein digest were determined according to Dumas using a combustion instrument (FP-328, Leco Corp.) and ethylenediaminetetraacetic acid (N = 9.59%) for calibration. The protein contents of the hordein and PT hordein digest were 84.3 ± 0.2% and 74.0 ± 0.5%, respectively. Beer. Beer, a typical fermented product that is analyzed by R5 competitive ELISA, was chosen as a sample. A gluten-free beer (Beer-Up, Malt n more trading GmbH) made from sorghum was used as a zero sample and as base material spiked to a defined prolamin (hordein) concentration with the PT hordein digest. The advantage of this was that samples with precisely defined prolamin contents determined by an independent analytical method (Dumas analysis) were available. Based on the fact that the nitrogen contents of both the PT hordein digest and hordein had been determined, the amount of added PT hordein digest corresponded to the amount of hordein used for its preparation. This was crucial for determining the recovery rate. Briefly, a defined amount of PT hordein digest was added to the gluten-free beer and stirred for 24 hr at room temperature to guarantee homogeneous distribution in the sample. Sourdough. A sourdough with defined prolamin content was prepared by mixing dried, gluten-free quinoa sourdough with an appropriate amount of dried rye sourdough (both from Ernst Böcker GmbH & Co. KG) and shaking the mixture overhead for 3 hr. Two sourdough samples with gluten contents of 35 and 75 mg of prolamin/kg were prepared. R5 competitive ELISA was used to determine the prolamin content of the rye sourdough (1,345 mg/kg), as well as the prolamin contents of the quinoa/rye sourdough mixtures that were used as samples in this study. Starch Syrup. One starch syrup sample was a commercial gluten-free product (Stayley 300 Corn Syrup, Tate & Lyle); the other sample was a naturally gluten-contaminated starch syrup from an anonymous industrial supplier. Gluten contamination was detected by R5 competitive ELISA. Analysis determined a prolamin concentration of 5 mg/kg. Homogeneity of Samples. All samples were checked for homogeneity before they were packaged in airtight bottles and accepted for the collaborative study. This was done by taking 10 representative 1 g aliquots (1 ml for beer) from 10 different CEREAL FOODS WORLD / 155

parts of the bulk sample and then analyzing by R5 competitive ELISA. The coefficients of variation (CV) for the prolamincontaining samples were ±10.1% or less for sourdough with 35 mg/kg and ±18.0% or less for beer with 50 mg/kg. The naturally gluten-contaminated starch syrup showed higher variation (±22.3%) due to its low prolamin concentration. All samples were accepted for the collaborative study. Gluten-free samples 1 and 4 were considered homogeneous because all analyses provided values below the cut-off (<5 mg/kg). Both samples had OD values scattered around the zero calibrator provided (CVs of ODs were around ±6%; n = 10). Presentation of Samples to Labs Following the AACCI collaborative trial guidelines, two independent blinded replicates for each sample were provided to the participating laboratories. Each sample was extracted using 60% (vol/vol) ethanol and analyzed in duplicate in one analytical run. Fourteen samples were analyzed by each laboratory. The high polyphenol content in the beer samples required a different extraction. These samples were specifically labeled and were extracted with 60% (vol/vol) ethanol containing 10% (wt/vol) fish gelatin. Method The method was written in AACCI style and was provided to each lab with instructions to follow the method as written, with no deviations. Labs were directed to pay particular attention to cases where samples had to be repeated by further dilution and to how the dilutions were to be carried out. All OD values had to be recorded in a ready-to-use Excel (Microsoft Corp.) worksheet. Participants were asked to use RIDASOFT calculation Table I. Prolamin concentrations determined using R5 competitive ELISA (data from all participating laboratories [ raw data ]) a Prolamin Concentration (mg/kg) b Sample 1 Sample 2 Sample 3 Sample 4 Sample 5 Sample 6 Sample 7 Lab Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 A 1.1 2.9 11.8 10.2 55.8 46.9 2.2 3.9 3.8 4.31 23.4 23.6 76.5 85.0 B 0.73 1.3 20.4 6.9 75.7 63.7 1.5 1.1 5.3 2.6 19.4 26.5 81.8 61.4 C 2.7 5.3 17.1 41.1 96.1 53.8 3.1 1.0 6.4 6.3 23.6 33.7 90.7 71.7 D 0.37 0.88 11.9 14.3 87.6 48.8 1.7 1.7 4.9 5.5 16.5 30.1 53.2 53.8 E 3.2 10.2 36.2 25.2 12.3 102.0 11.7 8.8 10.2 14.7 34.3 36.4 125.5 122.1 F 2.7 2.0 7.3 13.5 62.0 80.0 2.6 2.8 4.6 3.4 23.5 25.7 64.4 75.8 G 3.0 2.2 16.2 16.0 108.1 104.1 1.7 1.2 7.5 7.0 23.4 42.7 96.4 101.5 H 3.5 0.78 22.2 13.1 72.8 16.4 2.9 1.6 10.3 8.1 19.4 15.5 47.3 44.5 I 0.33 0.66 11.1 6.9 50.6 32.2 0.44 0.31 2.7 2.1 17.9 22.5 59.2 37.5 J 0.75 0.57 10.6 10.0 60.9 64.4 0.37 0.81 3.7 4.0 22.8 29.1 66.4 69.6 K 8.2 7.4 25.0 22.4 108.4 154.3 10.5 4.9 16.8 11.2 43.8 40.0 174.0 15.1 L 0.84 0.16 19.9 24.5 112.4 114.4 0.9 1.7 6.6 5.8 32.0 33.6 85.8 122.3 M 0.33 2.1 9.9 9.6 64.7 66.8 1.1 0.3 5.0 4.3 18.1 19.8 80.9 60.2 N 0.02 0.38 17.1 9.2 48.5 54.3 0.9 2.2 5.4 4.6 21.7 22.3 58.8 77.2 O 0.79 0.21 9.5 8.3 55.3 68.3 0.05 0.63 5.9 4.1 25.7 23.2 76.4 82.3 P 3.0 0.42 12.7 12.4 74.7 55.6 0.77 0.67 6.3 5.4 19.1 23.1 97.4 55.6 a Calculation of the statistics for gluten containing samples (2, 3, 5, 6, and 7) was done based on a cubic spline function using RIDASOFT Win Z9999 b Sample 1, gluten-free beer; sample 2, beer containing 15 mg of prolamin/kg; sample 3, beer containing 50 mg of prolamin/kg; sample 4, gluten-free starch Table II. Prolamin concentrations determined using R5 competitive ELISA (data after elimination of noncompliant laboratories) a Prolamin concentration (mg/kg) b Sample 1 Sample 2 Sample 3 Sample 4 Sample 5 Sample 6 Sample 7 Lab Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 Expt. 1 Expt. 2 A 1.1 2.9 11.8 10.2 55.8 46.9 2.2 3.9 3.8 4.3 23.4 23.6 76.5 85.0 B 0.73 1.3 20.4 6.9 75.7 63.7 1.5 1.1 5.3 2.6 19.4 26.5 81.8 61.4 C 2.7 5.3 17.1 c 41.1 c 96.1 53.8 3.1 1.0 6.4 6.3 23.6 33.7 90.7 71.7 D 0.37 0.88 11.9 14.3 87.6 48.8 1.7 1.7 4.9 5.5 16.5 30.1 53.2 53.8 G 3.0 2.2 16.2 16.0 108.1 c,d 104.1 c,d 1.7 1.2 7.5 7.0 23.4 42.7 96.4 101.5 H 3.5 0.78 22.2 13.1 72.8 16.4 2.9 1.6 10.3 8.1 19.4 15.5 47.3 44.5 I 0.33 0.66 11.1 6.9 50.6 32.2 0.44 0.31 2.7 2.1 17.9 22.5 59.2 37.5 J 0.75 0.57 10.6 10.0 60.9 64.4 0.37 0.81 3.7 4.0 22.8 29.1 66.4 69.6 L 0.84 0.16 19.9 24.5 112.4 c,d 114.4 c,d 0.91 1.7 6.6 5.8 32.0 33.6 85.8 122.3 M 0.33 2.1 9.9 9.6 64.7 66.8 1.1 0.31 5.0 4.3 18.1 19.8 80.9 60.2 N 0.02 0.38 17.1 9.2 48.5 54.3 0.92 2.2 5.4 4.6 21.7 22.3 58.8 77.2 O 0.79 0.21 9.5 8.3 55.3 68.3 0.05 0.63 5.9 4.1 25.7 23.2 76.4 82.3 P 3.0 0.42 12.7 12.4 74.7 55.6 0.77 0.67 6.3 5.4 19.1 23.1 97.4 55.6 a Calculation of the statistics for gluten-containing samples (2, 3, 5, 6, and 7) was done based on a cubic spline function using RIDASOFT Win Z9999 b Sample 1, gluten-free beer; sample 2, beer containing 15 mg of prolamin/kg; sample 3, beer containing 50 mg of prolamin/kg; sample 4, gluten-free starch c Outlier in Cochran test. d Outlier in Grubbs test. 156 / MAY JUNE 2013, VOL. 58, NO. 3

software for cubic spline curve fitting; the software was provided with the kit. The final data from the laboratories were sent to the study coordinator. Results of the Collaborative Study The quantitative results from all laboratories ( raw data ) are compiled in Table I. Thirteen of the sixteen laboratories provided results that were suitable for further statistical analysis and performance calculations. Lab F was excluded from the statistical evaluation because it did not run the calibrators in duplicate determinations as directed. Lab E found no difference between calibration standards S1 and S2 and, as a consequence, a high OD difference between standards S4 and S5, which led to an unusual curve shape. An interview also revealed technical problems during sample preparation. Lab K had a variation in the calibration curve that was too high, and an interview revealed the possibility of gluten contamination in the lab and incorrect pipetting. Table I contains only results of <5 mg/kg. Table II contains the results after elimination of noncompliant laboratories. For gluten-free samples 1 and 4 the software returned only a result of <5 mg/kg. To be able to use the results of the analysis of gluten-free samples 1 and 4 in the performance statistics, in particular in the calculation of LOD, concentration values for these samples were required. For this purpose, additional calibration curves were constructed using a second-order polynomial model and used to recalculate the results for samples 1 and 4. This calibration provided prolamin concentrations for the gluten-free samples (Tables I and II). Statistical Analysis and Discussion Outliers were identified using the Cochran and Grubbs tests according to AOAC International guidelines (3). The performance statistics without outliers are shown in the Table III (second-order polynomial function for gluten-free samples; cubic spline function for all other samples). The number of outliers was identified from the total number of replicates and total number of laboratories. For samples 2 and 3 this reduced the number of labs that were considered in the final calculation to 12 and 11, respectively (Table III). From the measured overall mean concentrations of the prolamin-containing samples recovery rates were calculated. The recovery values for samples 2, 3, 6, and 7 were 87, 119, 69, and 97%, respectively. The mean recovery where analyte concentrations were known was 93%. The mean and bandwidth of recoveries comply with acceptable recovery rates suggested by Abbott et al. (2) for spiked food samples, incurred samples, and/or difficult matrices. For sample 5 (naturally gluten-con-taminated starch syrup), no recovery rate could be calculated because the initial prolamin content was not known. Based on an expected mean prolamin concentration of 4.2 ± 0.94 mg/kg obtained in the homogeneity test, a recovery rate of 126% could be assumed for this sample. The repeatability relative standard variation (RSD r ) was comparable for all prolamin-containing samples, ranging between 16 and 32%. This was also the case for sample 5 (naturally gluten-contaminated starch syrup), which had an average prolamin concentration of 5.3 mg/kg, which was close to the cut-off point specified by the manufacturer. Although the reproducibility relative standard deviation (RSD R ) was somewhat higher, it was limited to a maximum RSD R of 37%. HORRAT values ranged from 2.6 to 3.6 and were in an acceptable range for ELISA tests at this low concentration range (2,12). To get an impression of the LOD of the method we followed the same guidelines as in our previous report (10). The LOD calculation (3.3 reproducibility standard deviation of the blank) requires a measurement result for the blank samples. This was done by using a second-order polynomial curve fitting instead of the cubic spline function using RIDASOFT Win Z9999 software. The resulting values for each laboratory and sample are given in Table II. As expected, the measured values fluctuated around zero. Calculating the overall mean resulted in values of 1.2 and 0.7 mg of prolamin/kg for gluten-free beer and gluten-free starch syrup (sample 4), respectively (Table III). For the gluten-free samples, S R, RSD r, and RSD R were quite high compared to gluten-containing samples, which is normal for zero samples in an ELISA test (10). With the results of the second-order polynomial, an estimation of the LOD can be based on the reproducibility standard deviation (s R ) of the gluten-free samples (Table III). This results in an LOD of 5 mg of prolamin/ kg. The analysis of sample 5 revealed a mean concentration of 5.3 mg of prolamin/kg with relative standard deviations similar to other prolamin-containing samples. Together with the estimated LOD, the collaborative study results are in agreement with a cut-off of 5 mg of prolamin/kg. Figure 2 shows prolamin concentrations for the beer sample containing 15 mg of prolamin/kg (equal to 30 mg of gluten/kg) and the sourdough sample containing 35 mg of prolamin/kg (equal to 70 mg of gluten/kg) obtained by 12 and 13 laboratories, respectively. The random error of the analyses, calculated according to Youden (14), was 7.9 and 10.1 mg/kg for the beer Table III. Performance statistics for overall competitive R5 ELISA results from Table II (prolamin concentrations are shown) a Sample ID b Parameter 1 2 3 4 5 6 7 Total number of laboratories (P) 13 12 11 13 13 13 13 Total number of replicates (Sum n(l) ) 26 24 22 26 26 26 26 Overall mean of all data (grand mean; XBARBAR) (mg/kg) 1.2 13.1 59.7 0.7 5.3 24.2 72.8 Repeatability standard deviation (s r ) (mg/kg) 1.2 4.0 18.6 1.0 0.9 5.6 14.2 Reproducibility standard deviation (s R ) (mg/kg) 1.5 4.8 18.6 1.5 1.8 6.3 20.0 Repeatability relative standard deviation (RSD r ) (%) 97.9 30.2 31.2 157.2 16.3 23.1 19.5 Reproducibility relative standard deviation (RSD R ) (%) 126.1 36.9 31.2 236.1 34.4 25.9 27.5 HORRAT value 8.1 3.4 3.6 13.8 2.8 2.6 3.3 a Calculation of the statistics for gluten-containing samples (2, 3, 5, 6, and 7) was done based on a cubic spline function using RIDASOFT Win Z9999 b Sample 1, gluten-free beer; sample 2, beer containing 15 mg of prolamin/kg; sample 3, beer containing 50 mg of prolamin/kg; sample 4, gluten-free starch CEREAL FOODS WORLD / 157

and sourdough samples, respectively. These values are almost twice as high as those of the s R reported in Table III. Values that are far away from the diagonal line in the Youden plot represent random errors. Dots on the line show no random error, but the further the distance to the origin the higher the systematic error. The data show that for both samples the majority of the values was close to the diagonal line. For the beer sample, three values were outside the circle, and for the sourdough sample, two values exceeded the standard error in spite of the higher standard error of the data for this sample. Discussion The competitive R5 ELISA method is designed for the detection of gluten in syrups and fermented foods at low levels. In the test samples with known levels, gluten was present as fragments generated by partial hydrolysis due to the action of peptidases. The sandwich R5 ELISA for intact gluten (1) is not sensitive enough for low levels of gluten fragments to be detected (7). The competitive R5 ELISA method is able to detect prolamin from a 5 mg/kg level (in foods as sold). In order to represent the gluten level, the prolamin concentrations must be multiplied by two (gluten = 2 prolamin). This translates to detection of gluten fragments at a level of 10 mg of gluten/kg and higher, up to 150 mg of gluten/kg (75 mg of prolamin/kg). In the case of samples with prolamin concentrations exceed-ing 100 mg/kg a further dilution step is necessary, because normally this test will not be used for measuring at such high prolamin levels (corresponding to >200 mg of gluten/kg). According to the Codex Alimentarius (4) and EU regulation 41/2009 (5) the threshold for labeling as gluten-free is set at 20 mg of gluten/kg in the food as sold. The method evaluated here is able to detect prolamin levels starting from 5 mg/kg (10 mg of gluten/kg). When the detectable level is lower than the advised cut-off of 5 mg of prolamin/kg, thus below LOD, the level of gluten fragments can be expected to be lower than 10 mg/kg. In the case of prolamin levels between 5 and 10 mg/kg, the test will produce a value that is between 10 and the threshold of 20 mg of gluten/kg. The quantitative results show that the competitive R5 ELISA is able to reliably quantify gluten in samples exceeding 20 mg of gluten/kg, clearly revealing that they are not compliant with Codex Alimentarius (4) and EU regulation 41/2009 (5). Fig. 2. Youden plot of results obtained by analysis of the beer sample containing 15 mg of prolamin/kg (equal to 30 mg of gluten/kg) (A) and the sourdough sample containing 35 mg of prolamin/kg (equal to 70 mg of gluten/kg) (B) using R5 competitive ELISA after calibration with a cubic spline model and using RIDASOFT Win Z9999 software. The red circle represents the random error according to Youden (14), which is considerably higher than the reproducibility standard deviation (s R ) reported in Table III. The dots are labeled with the codes of the laboratories that produced the results. Conclusions This collaborative study has shown that the competitive R5 ELISA is capable of analyzing prolamin fragments at concentrations starting from 5 mg/kg (equal to 10 mg of gluten/kg) up to 75 mg of prolamin/kg in a sourdough sample (equal to 150 mg of gluten/kg). The competitive R5 assay showed sensitivity for low levels of gluten fragments, enabling quantitation below and above 20 mg of gluten/kg. Acknowledgments We wish to thank Terry Nelsen and Greg Grahek for useful assistance in statistical evaluation of the data, as well as Michael Tilley for his support as co-chair of the AACCI Protein and Enzyme Committee. References 1. AACC International. Method 38-50.01, Immunochemical Determination of Gluten in Corn Flour and Corn-Based Products by Sandwich ELISA. Approved Methods of Analysis, 11th ed. Published online at www.aaccnet.org/approvedmethods/default.aspx. AACC International, St. Paul, MN. 2. Abbott, M., Hayward, S., Ross, W., Godefroy, S. B., Ulberth, F., et al. Validation procedures for quantitative food allergen ELISA methods: Community guidance and best practices. J. AOAC Int. 93:442, 2010. 3. AOAC International. Appendix D: Guideline for collaborative study procedures to validate characteristics of a method of analysis. AOAC Official Methods of Analysis. The Association, Gaithersburg, MD, 2002. 4. Codex Alimentarius Commission. Codex Standard 118-1979 (rev. 2008), Foods for special dietary use for persons intolerant to gluten. Codex Alimentarius. FAO/WHO, Rome, 2008. 5. European Commission. Commission Regulation (EC) No. 41/2009 of 20 January 2009 concerning the composition and labelling of foodstuffs suitable for people intolerant to gluten. Off. J. Eur. Union 16:3, 2009. 6. Frazer, A. C., Fletcher, R. F., Ross, C. A. C., Shaw, B., Sammons, H. G., and Schneider, R. Gluten-induced enteropathy: The effect of partially digested gluten. Lancet 274:252, 1959. 7. Gessendorfer, B., Koehler, P., and Wieser, H. Preparation and characterization of enzymatically hydrolyzed prolamins from wheat, rye, and barley as references for the immunochemical quantitation of partially hydrolyzed gluten. Anal. Bioanal. Chem. 395:1721, 2009. 8. Immer, U., and Haas-Lauterbach, S. Statistical evaluation of gliadin ring trial. Page 23 in: Proceedings of the 18th meeting of the Working Group on Prolamin Analysis and Toxicity. M. Stern, ed. Verlag Wissenschaftliche Scripten, Zwickau, Germany, 2004. 9. Immer, U., Vela, C., Mendez, E., and Janssen, F. PWG collaborative trial on gluten in gluten-free food through Cocktail ELISA. Page 23 in: Proceedings of the 17th Meeting of the Working Group on Prolamin Analysis and Toxicity. M. Stern, ed. Verlag Wissenschaftliche Scripten, Zwickau, Germany, 2003. 10. Koehler, P., Schwalb, T., Immer, U., Lacorn, M., Wehling, P., and Don, C. AACCI Approved Methods Technical Committee report: Collaborative study on the immunochemical determination of intact gluten using an R5 sandwich ELISA. Cereal Foods World 58:36, 2013. 11. Nelsen, T. C., and Wehling, P. Collaborative studies for quantitative chemical analytical methods. Cereal Foods World 53:285, 2008. 12. Thompson, M. Recent trends in inter-laboratory precision at ppb and sub-ppb concentrations in relation to fitness for purpose criteria in proficiency testing. Analyst 125:385, 2000. 13. Weisstein, E. W. Cubic spline. Published online at http://mathworld. wolfram.com/cubicspline.html. MathWorld, Wolfram Research, Inc., Champaign, IL, 2012. 14. Youden, W. J. Graphical diagnosis of interlaboratory test results. Ind. Qual. Control 15:24, 1959. 158 / MAY JUNE 2013, VOL. 58, NO. 3