APPLICABILITY OF ELISA METHODS FOR HIGH GLUTEN-CONTAINING SAMPLES

Quantitation of gluten in gluten-free products is a great challenge as it is hindered by several factors including the lack of certi ﬁ ed reference materials. To resolve this problem, our research group, in cooperation with other international experts, started a series of experiments with the goal of the production of a suitable gluten reference material. As a part of this research, several different wheat cultivars and their isolated gluten proteins were characterized by different methods, including enzyme-linked immunosorbent assay (ELISA). However, we need to know the performance of the ELISA methods used for this special area of research. During the present work we investigated the accuracy and precision of two different ELISA methods for our own laboratory conditions and special sample matrices (wheat ﬂ ours and gliadin isolate). We have found that the tested performance characteristics of the methods seem to be appropriate on a case-by-case basis, but the long-term measurement uncertainty is higher, which makes it dif ﬁ cult to evaluate the results obtained with the ELISA method for these types of samples.

Beside their key role in the formation of unique technological behaviour of wheat-based fl ours and bakery products, gluten proteins of wheat (and also rye and barley) can trigger celiac disease in about 1% of the population. The only treatment is a lifelong gluten-free diet (LUDVIGSSON et al., 2013). According to Codex Alimentarius, gluten free products cannot contain more than 20 mg kg -1 gluten (CODEX STANDARD 118-1979, 2015. Various analytical methods are available for determination or checking the gluten content, such as enzymelinked immunosorbent assay (ELISA), liquid chromatography with mass spectrometry detection, or molecular biological techniques (e.g. polymerase chain reaction). However, the most commonly used method in routine analysis is ELISA. These methods are based on immunochemical reactions, in which antibodies recognize one or more short peptide sequences (epitopes) of gluten proteins. This reaction results in a very specifi c and sensitive determination (SCHERF & POMS, 2016). Commercially available ELISA tests provide partially different strategies for determining gluten concentration as they apply different antibodies, extraction procedures, and calibrating materials (DIAZ-AMIGO & POPPING, 2013). Consequently, ELISA tests can give different results for the same sample (BUGYI et al., 2013;SCHERF, 2017). The determination of gluten is further complicated, because the protein content and composition of cereals vary depending on genetic (species, varieties) and environmental factors (harvest year and agricultural practices). As a result, these factors also affect the relative amount of epitopes that are recognized by antibodies commonly used in gluten ELISA test kits (HAJAS et al., 2018;SCHOPF & SCHERF, 2018). Complexity of the food matrices and the effects of food processing on the target proteins are also restraining factors of gluten analysis (BESLER et al., 2001;TÖRÖK et al., 2015).
Another major diffi culty of this fi eld is the lack of certifi ed reference materials (DIAZ-AMIGO & POPPING, 2013;HAJAS et al., 2018). Today, the widely used standard-like material for gluten analysis is the PWG-gliadin. It is an isolated protein fraction from 28 European wheat cultivars with good solubility, homogeneity, and stability properties ( VAN ECKERT et al., 2006). It was proposed for approval as a certifi ed reference material (RM), but it does not meet some of the RM requirements, such as reproducibility of production (DIAZ-AMIGO & POPPING, 2013).
In an international framework, our group has put a lot of effort into gluten RM development (HAJAS et al., 2016(HAJAS et al., , 2018. The speciality of our RM development work is that we had to investigate different matrices with high gluten protein contents, like wheat fl ours, gluten or gliadin isolates. As these types of samples are not included in the intended use of the ELISA methods, it is very important to understand their performance characteristics for this application. The presented work can be considered as an in-house validation of the methods for the examined sample matrices. By establishing the performance, it is possible to include ELISA methods for gluten RM development purposes.

Experimental design
Several factors were studied to assess the performance of ELISA methods. The effect of the applied methods (including different sample preparations, epitopes, and reagents) was examined by using two different ELISA methods. The matrix effect was investigated using pure protein isolate, which contains just the target protein fractions (gliadins) including proteins with target epitopes, and two wheat fl ours containing all protein fractions (gluten proteins, albumins, and globulins), other macromolecules (starch, lipids, and fi bres), and minor ingredients. The effect of kit stability was studied by the utilisation of kits with different LOT numbers. The effect of plate homogeneity was investigated by placing the samples on the same strips and randomly on the ELISA plates. The in-house repeatability was investigated by the evaluation of the results of two operators (marked with "X" and "Y"). For the determination of accuracy, gliadin recovery values were calculated with the evaluation of results obtained by reversed-phase high-performance liquid chromatography (RP-HPLC) method. The stability of samples was studied by using different results, which were performed on a total of 20 measurement days for a period of approximately 2.5 years.

Samples
Two wheat (Triticum aestivum L.) cultivars (Capo from Austria and Glenlea from Canada) were used for producing laboratory fl our samples from the harvest year of 2014. PWG gliadin was used as a pure isolate, and it was also applied in the calibration of the RP-HPLC method (VAN ECKERT et al., 2006).

Gliadin quantifi cation with ELISA methods
The gliadin quantifi cation was performed with two commercially available ELISA test kits: the AgraQuant Gluten G12 Assay (COKAL0200, Romer Labs, Tulln, Austria) and the RIDASCREEN Gliadin Assay (R7001, R-Biopharm, Darmstadt, Germany). They apply different antibodies (monoclonal G12 and monoclonal R5, respectively) and are calibrated differently. The ELISA test kits used for analysis were randomly coded with capital letters ("A" and "B") in the article. ELISA procedures were carried out according to the kit instructions with a few modifi cations. Three parallel extractions were performed for each sample. The amount of sample to be weighed was reduced 10 times in case of fl ours and 100 times in case of PWG gliadin. Further dilution was applied with 60% (v/v) ethanol solution from the clear supernatants: 1:1000 in two steps in the case of fl ours and 1:2000 in two steps in the case of PWG gliadin. The samples were prepared and diluted independently with freshly prepared reagents for each measurement day. The absorbances were determined using a microplate reader (iMarkTM Microplate Absorbance Reader, Bio-Rad, Hercules, CA, USA). The gliadin/gluten concentration was calculated from the absorbance values by the Bio-Rad Microplate Manager 6 software (Bio-Rad, Hercules, CA, USA) using the curve fi t option recommended by the manufacturers (cubic spline in case of method "A" and point-topoint in case of method "B").

Statistical analysis
The analytical results were statistically evaluated with the investigation of means and relative standard deviations. Statistical analysis was carried out by one-sample t-test, paired-sample t-test, and analysis of variance (ANOVA) at a confi dence level of 0.95 using STATISTICA v13 software (TIBCO Software Inc). Thirty-eight independent measurements were analysed for the statistical analysis, 26 of them were measured by method "A" and 12 by method "B". The distribution of the samples and the operators is shown in Figure 1.

General assessment of the results
Based on our gliadin recovery and repeatability values, the accuracy and the precision of both methods are acceptable, because the mean of the gliadin recovery values do not differ signifi cantly from 100% and the repeatability values are suffi ciently low (  (Fig. 1). We must know if this is the uncertainty of the method, or can we identify factors that affect the performance of ELISA methods? We determined and compared the extents of these effects and tried to identify the possible causes, as well.

Examining factors infl uencing recovery values
Comparing the average recovery values obtained with the two methods, only 4% difference is found (Table 1/Column 2). The variation between the operators within method "A" is only 1%, while in case of method "B", it is 14% (Table 1/Column 6). The comparison of the average gliadin recovery values obtained for the different samples are shown in Figure 1. The Glenlea fl our results measured by method "A" are typically less than 100%, and the Capo fl our results are higher than 100%. The values of PWG gliadin vary around 100%. In case of method "B", both fl ours fl uctuate around 100%, while PWG gliadin is higher than 100%. The uncertainty of recovery values measured by "X" is typically lower than measured by "Y" operator (Table 1/Column 11). However, the wide range of deviation for "Y" could be explained by the fact that measurements were made within a much wider time interval than in the case of "X" operator. On the other hand, operator "X" used ELISA kits with the same LOT number in case of method "A", so the difference in LOT numbers could also affect the results.  Theoretically, the relatively high fl uctuation of the recovery values may also be due to the dependence of the results on sample position. The reasons can be a small inhomogeneity of antibody covered wells on ELISA plates or the time necessary for reading the absorbance (in case of higher sample number). The results of randomly and not randomly placed samples are shown in Figure 2. Only small and not signifi cant differences are observed between the two placement modes, so the placement has no effect on the recovery values. Comparing the contribution of all factors to the uncertainty of gliadin recovery with analysis of variance, we have found that only the different samples and measuring on different days have a signifi cant effect on the results, which also interact with each other (Fig. 3). It is important to note that the effects of the other factors might not be detected, because these two effects cause very high measurement uncertainty. The different results of the samples can be explained by the presence of the amount of epitopes in them, but there is still no explanation for the differences between the measurement days. It might be caused by change in the stability of the samples, as their properties may change over time. But any positive or negative tendency in the recovery values of all samples cannot be identifi ed through the two and a half year research period (Fig. 1), so it is not likely that sample instability stands behind these observations.
Continuing the investigation of the effects of days, a critical point of the quantitative analysis can be the correct calibration procedures. Figures 4 and 5 show the calibration curves of the ELISA kit "A" and "B" on each measurement day. It is clear that the shape of the curves shows variability from day to day, resulting in slight changes in sensitivity and different correlations to calculate the concentration of the samples. Differences arising from the use of different calibrations are diffi cult to justify, so there is an urgent need for a reference to examine this and all factors infl uencing gluten determination.   : 24.02.2016; : 03.03.2016; : 08.03.2016; : 11.03.2016; : 25.05.2016; : 31.01.2017; : 28.05.2017; +: 22.06.2017; : 29.11.2017; : 30.11.2017; : 09.02.2018; : 17.07.2018 Acta  : 17.03.2016; : 19.03.2016; : 30.05.2016; : 30.01.2017; : 30.05.2017; : 15.12.2017; : 01.08.2018

Examining factors infl uencing precision of the methods
Although the repeatability values are acceptable for a general ELISA method (ABBOTT et al., 2010), the relative standard deviation (RSD) values of the parallel samples on the different measurement days move on a wide scale, from 0 to 15% (Fig. 6). In case of Capo sample, we can see a slight increase in the measurement uncertainty measured by method "A", which can be related to the LOT number, too. However, this is not found in the results of Glenlea sample measured by ELISA kits with the same LOT number. A clear increasing tendency cannot be observed in the results of the other method, in fact, in case of Glenlea, this is rather a slight decrease (Fig. 6).
The inhomogeneity of ELISA plate may have an impact on measurement uncertainty. Based on the experiment of different locations on ELISA plate, the uncertainty of the randomly placed samples is signifi cantly higher than the samples next to each other, so the placement could affect the RSD values (Fig. 2).
Since no signifi cant differences can be obtained between the methods, the operators, and the samples, these factors have no signifi cant effect on measurement uncertainty, but the inhomogeneity of the plates and the instability of the kits may affect the uncertainty of the measurements. The inhomogeneity of the samples cannot be identifi ed analysing the RSD values of the measurements (Fig. 6).
Random errors in sample preparation could affect the repeatability values of the methods. Since neither method-dependent nor person-dependent changes can be identifi ed, it is worth looking at the values of the calibration standards here, too. The standard solutions are readyfor-use in the kits, therefore, when evaluating their values, we can get information on the measurement regardless of sample preparation. We have found earlier that the repeatability of the methods is about 8%. The uncertainty of measuring the calibration solutions is about 7% for both methods, so regardless of sample preparation, 7% uncertainty can occur. This result is reassuring since it can be assumed that in our special fi eld of research, we do not cause a large measurement error during the preparation of these samples with high gluten content.

Conclusions
During our research we studied different factors infl uencing analytical performance of ELISA method on a special research fi eld, where samples with high gluten content were investigated. The recovery values for accuracy and RSD values for long-term precision are acceptable for general ELISA methods. On the other hand, we pointed out that the comparison of individual analytical results of the same sample in a long-term investigation study can show much higher diversity than the mentioned parameters of the analytical performance of both ELISA methods we used. The traditional sources of analytical errors (such as sample preparation, sample placement on plate, sample stability, etc.) have no signifi cant effects, which means both methods we used are working reliably on our samples. One of the two signifi cant effects belongs to the sample identity, which is good as we want to compare wheat varieties as one of the main goals in our RM-oriented research. The second signifi cant factor was the measurement days -including calibration procedures. These results reinforce also the necessity of a reference material, which allows to improve the calibration procedures, support the validation process, and give a chance to compare the analytical results obtained by different methods such as different ELISAs or even HPLC-MS.