Advances have been made to provide people with celiac disease (CD) access to a diverse diet through an increase in the availability of gluten-free food products and regulations designed to increase label reliability. Despite advances in our knowledge regarding CD and analytical methods to detect gluten, little is known about the effects of fermentation on gluten detection. The enzyme-linked immunosorbent assay (ELISA) and lateral flow devices routinely used by analytical laboratories and regulatory agencies to test for the presence of gluten in food were examined for their ability to detect gluten during the fermentation processes leading to the production of soy sauce, as well as in finished products. Similar results were observed irrespective of whether the soy sauce was produced using pilot-plant facilities or according to a homemade protocol. In both cases, gluten was not detected after moromi (brine-based) fermentation, which is the second stage of fermentation. The inability to detect gluten after moromi fermentation was irrespective of whether the assay used a sandwich configuration that required two epitopes or a competitive configuration that required only one epitope. Consistent with these results was the observation that ELISA, lateral flow devices, and Western immunoblot analyses were unable to detect gluten in commercial soy sauce, teriyaki sauce, and Worcestershire sauce. Although reports are lacking on problems associated with the consumption of fermented soy-containing sauces by consumers with CD, additional research is needed to determine whether all immunopathogenic elements in gluten are hydrolyzed during soy sauce production.
Approximately three million Americans suffer from celiac disease (CD) (7, 15, 23). CD is an immunopathogenic response to the consumption of gluten or gluten-source grains (e.g., wheat, barley, and rye). A subset of the population sensitive to gluten is also sensitive to avenin derived from oats, but the etiology of this immunopathology is not fully understood (9, 10). The U.S. Food and Drug Administration (FDA) issued a regulation in 2013 defining the term “gluten-free” to provide a uniform standard for manufacturers who choose to label their products as gluten free and, at the same time, to help consumers with CD identify products safe for consumption (25). In developing the 2013 regulation, the FDA identified uncertainty with respect to interpreting the results of methods to detect and quantify gluten in fermented and hydrolyzed foods. In 2015, the FDA issued a proposed rule regarding the use of the label term gluten free on fermented, hydrolyzed, and distilled foods (26).
Fermentation reduces the presence of detectable intact gluten through associated proteolysis reactions. Little is known about the effects of proteolysis on the biological activity of gluten to cause a pathogenic response in people with CD and (nonceliac) gluten sensitivities. Further, proteolytic standards for gluten do not exist. Thus, interpretation of the gluten content following proteolysis in terms of content, biological activity, and regulatory thresholds based on intact gluten is problematic (13, 16, 21). Research studies have been published on the effects of fermentation involved in the production of beer on the detection of gluten (20–22, 24). However, little is known about the ability of commonly used antibody-based assays to detect gluten during other fermentation processes, such as those involved in the production of soy sauce.
Traditional Japanese soy sauce (shoyu) is made from soybeans and wheat in a two-stage fermentation process. First, a mold-covered mixture of soybeans and wheat, called koji, is generated. The koji is mixed with brine (salt water) to make moromi (mash) that is allowed to age for several months, during which salt-tolerant, lactic acid bacteria and yeast catalyze further fermentation of the mash. Traditional Japanese soy sauce, shoyu, typically involves the use of Aspergillus sp. to generate the koji. There are five types of shoyu products in Japan: koikuchi, usukuchi, tamari, shiro, and saishikomi (11). Based on ingredients, shoyu can be classified into three types: koikuchi (dark color), usukuchi (light color), and saishikomi (double fermented). These have a widely variable combination or ratio of soy to wheat; tamari has little-to-no wheat, and shiro has a high amount of wheat relative to soy.
Using immunoglobulin (Ig) E antisera from five wheat-allergic children, Kobayashi et al. in 2004 (14) tracked the presence of antigenic elements in soy sauce during production. Analyses using a direct enzyme-linked immunosorbent assay (ELISA) format indicated an 80% reduction in antigenicity upon initiation of the moromi stage, with virtually no detectable antigenicity after 48 days of moromi fermentation. An inhibition ELISA configuration using the IgE antisera provided a more sensitive profile of antigenicity in the salt-soluble fraction during fermentation. Antigenic elements were not detected in the soy sauce upon completion of the fermentation processes. Examination of 10 commercial soy sauce products failed to display antigenicity with the IgE antisera, consistent with the production studies (14).
The ability to detect gluten in soy sauce during fermentation was explored using commercial assays that are employed by industry and regulatory agencies. Five commercial ELISA test kits, four commercial lateral flow devices (LFDs) or dipsticks, and Western blot analyses were used to detect the presence of gluten during soy sauce production. Among the ELISAs used in this study were the MIoBS ELISA, used by FDA and the Government of Japan, and the R5 RIDASCREEN ELISA, endorsed by CODEX and used by the FDA (8, 25). The presence of gluten was examined in 15 commercial soy sauces (including one that listed wheat as the primary ingredient after water), as well as during homemade soy sauce production, pilot-scale production of a model soy sauce, and pilot-scale moromi fermentation of model wheat-free soy sauces supplemented with 0, 20, and 200 mg of wheat gluten per kg of solid ingredients (parts per million) immediately prior to the 2.78-fold dilution with brine.
MATERIALS AND METHODS
Gluten (cat no. G5004), phosphate-buffered saline (PBS; cat no. P3813), Tween 20, and all other reagents were purchased as analytical grade from Sigma-Aldrich Co. LLC (St. Louis, MO) or other suppliers, as noted. Wheat gluten stock solutions were prepared initially at 1 mg/mL in PBST (PBS supplemented with 0.1% [v/v] Tween 20) and serially diluted for use after incubation at room temperature for 2 h.
The five commercial ELISAs listed in Table 1 were used according to the manufacturer's instructions; when multiple protocols were available, the protocol associated with the multilaboratory validation was used (1, 2, 4, 18). Samples generating responses in excess of the dynamic range as defined by the lowest, nonzero and highest calibration standards supplied with the test kit were serially diluted with the extraction solution until the response was within the calibration curve. Unless stated otherwise, all analyses were done in triplicate, and the plates were read using an Infinite M200 Plate Reader (Tecan US, Inc., Morrisville, NC) or an ELX-808 plate reader (BioTek Instruments, Inc, Winooski, VT) with Gen 5 (version 2.04) software (BioTek).
LFDs or dipsticks
The RIDAQUICK Gliadin (R-Biopharm, Inc., Washington, MO), awarded AOAC OMA First action status; AgraStrip Gluten G12 (Romer Labs, Inc., Union, MO), AOAC-RI PTM certified; and the sandwich and competitive formats of the Gluten Residue Lateral Flow Test (SfGLFT and CfGLFT, ELISA Systems, U.S. distributor Pi Biologique, Seattle, WA) were used according to manufacturers' recommendations, with the modification that all devices were allowed to develop for 10 min (3, 5, 17). The LFDs and dipsticks were read using an MS1000 Immunochromato-Reader C10066-10 (Hamamatsu Photonics K.K., U.S. distributor MitoSciences, Eugene, OR) in conjunction with the Dip Type Holder (type no. A10793, Hamamatsu Corp., Bridgewater, NJ). LFDs were read using the MS1000 Immunochromato-Reader C10066-10 by immediately, upon completion of the 10 min of development, opening the LFD cassette and placing the internal strip in the strip holder (Dip Type Holder) in an identical fashion to the dipsticks. Wheat gluten calibration standards at concentrations of 0, 1, 5, 10, 25, 50, and 100 μg/mL gluten were freshly prepared on the day of use from the Sigma (intact) gluten powder (cat no. 5004) and were analyzed alongside the soy sauce samples, except when noted otherwise. LFD and dipstick analyses were typically performed in triplicate with the pilot-plant samples testing as negative for the presence of gluten. These samples were also analyzed after being supplemented with 20 μg/mL gluten to confirm that the matrix was not selectively affecting the development of the test band. In all cases, the supplementation resulted in a positive response, indicating that the LFDs and dipsticks were capable of detecting intact gluten in the samples. The larger than expected variation in the intensity of the test bands generated by the pilot-plant samples (collected at different time points) upon supplementation with 20 μg/mL gluten made quantitative comparisons difficult. This larger than expected variation in test band intensities may reflect many possibilities; one possible explanation is that changes in the composition of the matrix during the fermentation process affected test band development. Despite these variances, the qualitative responses were definitive (>50 monoclonal antibodies [MAbs], dependent on the LFD or dipstick), which indicates that if 20 μg/mL intact gluten was present, it would have been detected.
Fermented soy-containing sauces
Fifteen commercial soy, three commercial teriyaki, and two commercial Worcestershire sauces were purchased from local markets in Chicago, IL, and Washington, DC.
Model soy sauce and model wheat-free soy sauce were brewed using pilot-plant equipment at Kikkoman's R&D Division (Japan) according to the flow chart in Figure 1. The model wheat-free soy sauce did not include rice, sugar, or other wheat substitutes that are sometimes used in preparing commercial gluten-free tamari soy sauces. Batches of the model wheat-free soy sauce containing variable amounts of gluten were prepared by adding 0, 20, and 200 mg of wheat gluten per kg of solid ingredients immediately prior to the 2.78-fold dilution with brine. This resulted in moromi mash containing 0, 7.2, and 72 ppm of gluten. Samples of the model sauces brewed at the Kikkoman facility were collected for gluten analysis at the raw material stage, at the end of the koji stage, at the initial moromi stage, and during the 6-month fermentation of the moromi. Koji samples were deactivated by heat treatment (80°C, 30 min). All moromi samples were filtered, deactivated by heat treatment (80°C, 30 min), and centrifuged.
Homemade fermented soy sauces were prepared according to published recipes (19, 28) that ferment the ingredients in two stages. The soybean paste was prepared by chopping and grinding deshelled raw soybeans (Pictsweet Company, Bells, TN) that were steamed for 3 min at 1,250 W in the microwaveable bag supplied by the manufacturer. The flour used was either Organic Whole Wheat Flour (King Arthur Flour Company, Inc., Norwich, CT) or Long Grain Brown Rice Flour, labeled gluten free (Hodgson Mill, Inc., Effingham, IL), used as a substitute for the wheat flour in the recipe. The rice flour–derived mixtures with soybean paste were supplemented with wheat gluten at 0, 0.01, 0.05, 0.1, or 0.5% (0, 100, 500, 1,000, 5,000 ppm of gluten). During the first stage, four parts of soybean paste was mixed with three parts of flour. The mixture was then sandwiched (at less than a quarter-inch thick) between wet and dry paper toweling, wrapped in plastic wrap, and stored at room temperature for 28 days to allow mold growth. The second stage consisted of partially unwrapping the moldy soy flour mixtures to dry for 12 days in a chemical fume hood, after which the dry material was mixed with 6% brine (made using sea salt) at a ratio of 1 g to 20 mL of brine, followed by fermentation for 31 days. Samples were collected throughout the production process for gluten analysis. As is typical for homemade soy sauces, no effort was made to use either sterilized ingredients or standardized fungal or bacterial cultures for the fermentation stages. As a result, variability between preparations is expected.
Western immunoblot analyses of commercial soy, teriyaki, and Worcestershire sauces were performed as previously described (21). Immunoblotting was performed using the detector antibodies of the R5-Sand, G12-Sand, Skerritt, and MIoBS ELISAs diluted 1:110, 1:10, 1:50, and 1:10 (v/v), respectively. The higher dilution of the R5-Sand detector antibody conjugate reflects being supplied as an 11-fold concentrate, whereas the 50-fold dilution of the Skerritt antibody reflects its high specific activity. Chemiluminescent images were collected using a Syngene G:BOX Chemi XX9 imager (Syngene, Frederick, MD).
RESULTS AND DISCUSSION
The raw and processed ingredients used in the pilot-plant production of the model sauces were analyzed with the MIoBS ELISA, and the concentrations of gluten were as follows: raw soybeans <LOQ (0.3 ppm), cooked soybeans <LOQ, whole wheat 68,000 ± 1,000 ppm, and roasted wheat 94,500 ± 500 ppm. Detectable gluten concentration (MIoBS ELISA) in the model wheat-containing soy sauce collected 3 days after adding the Aspergillus sp. starter culture (koji) was 28,200 ± 900 ppm, 14,200 ± 500 ppm after addition of the brine solution (moromi mash), and <LOQ from 1 month through completion of the moromi fermentation process. These results indicate minimal changes in detectable gluten concentrations, other than what might be expected owing to dilution with other ingredients, until the moromi (brine-based, second-stage) fermentation.
Presented in Table 2 are side-by-side analyses using five different ELISAs of the koji, moromi mash, and monthly samples collected through the 6 months of moromi fermentation to make the model soy sauce. As expected, the R5-Sand and Skerritt ELISAs, like the MIoBS ELISA, displayed changes in detectable gluten concentrations consistent with dilution owing to the addition of brine. The G12-Sand ELISA also displayed a decrease (25%), but less than expected from the dilution. After 1 month of moromi fermentation, all four sandwich ELISAs were unable to detect gluten in the model soy sauce. In contrast to the sandwich ELISAs, the R5-Comp ELISA detected an increase in gluten upon dilution with brine that dropped to <LOQ after 1 month of moromi fermentation. Competitive ELISAs require only one epitope, whereas sandwich ELISAs require at least two epitopes. Further, it is not uncommon for the antigenic proteins in gluten to have multiple R5, G12, and Skerritt epitopes; a critical feature for the R5-Sand, G12-Sand, and Skerritt ELISAs, which use a MAb as both the capture and detector antibody. Thus, if during the koji stage of fermentation a small amount of antigenic gluten proteins were partially hydrolyzed, the competitive ELISA would detect an increase in antigenic polypeptides that might not significantly influence the performance of the sandwich ELISAs.
The model wheat-free soy sauce incurred with gluten and the homemade fermented soy sauces displayed a loss in detectable gluten similar to that of the model soy sauce. Specifically, no gluten was detectable upon completion of the fermentation processes, with little-to-no gluten detectable beyond the start of the second fermentation stage. However, a few differences between the model wheat-free soy sauce incurred with 72 mg/kg (ppm) gluten and the model soy sauce were observed. Specifically, after accounting for dilution, at the start of moromi fermentation, an average of 65% ± 7% less gluten than expected was detected (detected 19 to 31 ppm with the five ELISA test kits). In addition, the R5-Comp ELISA did not detect an increase in gluten prior to moromi fermentation. Both differences might be because of the relatively low amount of gluten incurred, altering the kinetics of gluten hydrolysis. Also, by incurring the gluten immediately prior to dilution with the brine, it was not exposed to the first (koji) fermentation stage, which might be necessary for partial protein proteolysis to generate the increase in detectable gluten by the R5-Comp ELISA. The model wheat-free soy sauce incurred with gluten at 7.2 mg/kg (ppm) was at or below the LOQ of the R5-Sand, R5-Comp, G12-Sand, and Skerritt ELISAs at the start of the moromi fermentation stage; only the MIoBS ELISA could detect gluten (1.5 ppm) owing to its low LOQ of 0.3 ppm. The model wheat-free soy sauce not supplemented with gluten consistently tested below the limit of quantitation with all five ELISAs.
The homemade fermented soy sauces also displayed losses in detectable gluten. The soybean paste–flour mixtures were prepared to contain 0, 100, 500, 1,000, or 5,000 ppm of gluten or wheat flour and were sampled immediately after preparation. These six initial samples, immediately after preparation and prior to fermentation or dilution, gave responses (using MIoBS ELISA) characteristic of 0, 90 ± 30, 320 ± 40, 700 ± 30, 5,000 ± 1,000, and 44,000 ± 6,000 ppm of gluten, respectively. Upon completion of the first stage of fermentation and throughout the second stage of fermentation in brine, no gluten was detected (LOQ < 0.3 ppm). There may be multiple reasons for the inability to detect any gluten upon completion of the first stage of fermentation for the homemade soy sauce made with wheat flour versus the pilot-plant model soy sauce, which contained detectable gluten into the first month of moromi fermentation. First, the homemade soy sauce had a much longer first fermentation stage prior to dilution with brine than the model soy sauce, namely 40 days versus only 3 days. A second difference is the repertoire and concentration of microbes present. Third, the fermentation conditions were different, which might affect microbial respiration and protein hydrolysis. Fourth, the wheat used to prepare the homemade soy sauce was wheat flour versus roasted, cracked wheat grain used to prepare the model soy sauce. Despite these differences, in both cases, the ELISAs failed to detect gluten in the final, fermented soy sauces.
Similarly, 12 commercial soy sauces, three commercial teriyaki sauces, and two commercial Worcestershire sauces that included wheat as an ingredient all tested negative for the presence of gluten using the MIoBS ELISA. Interestingly, one of the commercial soy sauces that tested negative for gluten was a sauce that listed wheat as the second ingredient after water. As a control, three commercial soy sauces that did not include wheat as an ingredient were examined and also tested negative for the presence of gluten. Being commercial products, details of how these sauces were produced were not available, but the results provide information on the detectable gluten content of products sold in two metropolitan areas.
LFDs and dipsticks
LFDs and dipsticks are typically less sensitive than other antibody-based assays but have the advantages of being field deployable, quick, and potentially inexpensive when only a few samples need to be analyzed. LFDs and dipsticks designed in the sandwich format (e.g., RIDAQUICK, AgraStrip Gluten G12, and the sandwich format of the Gluten Lateral Flow Test [SfGLFT]) have the potential of being highly specific and reliable by requiring two epitopes for the detection of antigens. In contrast, LFDs and dipsticks employing a competitive format (e.g., competitive format of the Gluten Lateral Flow Test [CfGLFT]) rely on the detection of a single epitope with a positive response indicated by a decrease in the intensity of the test band. This means that anything that inhibits the background binding appears as a positive response. As a result, the LOD values for competitive LFDs and dipsticks are usually higher (less sensitive) than sandwich devices owing to higher backgrounds (noise). An advantage of the competitive format, as for ELISAs, is that, by requiring the presence of a single epitope, it has the potential to better reflect celiac sprue biological activity, which also requires only a single immunopathogenic site; this makes the competitive assays less likely to generate false negatives with fermented and hydrolyzed foods.
The abilities of the RIDAQUICK, AgraStrip Gluten G12, SfGLFT, and CfGLFT to detect the regulatory threshold of 20 ppm of wheat gluten are illustrated in Figures 2 and 3. Figure 2A depicts the changes in intensity of bands obtained with the RIDAQUICK dipsticks with an increase in the amount of gluten spiked into buffered detergent (PBST) and a commercial soy sauce. Figure 2B depicts a typical trace generated by the Reader. Figure 3A compares the responses generated by RIDAQUICK, AgraStrip Gluten G12, SfGLFT, and CfGLFT with gluten spiked into PBST and soy sauce. To facilitate interpretation of the CfGLFT data, the data presented in Figure 3A were normalized to the response generated by the zero-gluten control samples. Quantitative comparison between the responses generated in PBST and soy sauce are inconclusive, as evidenced by the standard deviations (error bars in Fig. 3A) and day-to-day variation in performance, as evidenced by the RIDAQUICK data in Figures 2A and 3A. It is, therefore, recommended that calibration controls be run alongside the samples whenever LFDs or dipsticks are used for (semi-)quantitative purposes, such as whether the target analyte exceeds a regulatory threshold. Further, replicate analyses should also be conducted whenever using LFDs or dipsticks to ascertain the variance.
As indicated, a competitive design has many advantages, especially in regard to gluten detection because the biological activity is dependent on single immunopathogenic epitopes. However, competitive assays also have a disadvantage, as illustrated in Figure 3B by the responses generated by CfGLFT LFDs. Depicted in Figure 3B are the responses (not normalized) generated by samples collected during the moromi (brine) fermentation of the model soy sauce and model wheat-free soy sauces. The variances between the zero control (260 MAbs) and the responses generated by the moromi fermentation samples (presumed to have no detectable gluten) are characteristic of competitive assays, which are higher than the variances of the nondetect (zero) responses observed with sandwich assays. The RIDAQUICK, AgraStrip Gluten G12, and SfGLFT sandwich assays displayed negative overall average responses for the samples collected during moromi fermentation of 0 ± 1, 0 ± 1, and 5 ± 5 MAbs, respectively. As a result, competitive assays often have high thresholds and, at low concentrations of analyte, display larger variances. Such disadvantages can be compensated for in many ways. One solution is to design sample preparations that minimize exogenous components that may interfere with the binding process. Also, the utility of competitive assays can be improved by optimizing the concentration of antigen immobilized on the surface and the concentration of detector (labeled) antibody to better reflect the target analyte concentration of interest in the sample.
Immunoblot analysis of sodium dodecyl sulfate–polyacrylamide gel electrophoresis fractionated proteins, such as competitive antibody-based assays, requires only one antigenic epitope. However, unlike ELISAs and LFDs or dipsticks, Western analysis also provides information regarding the size of the antigen and, when compared with standards, can support a preliminary identification. Figure 4 depicts the immunoblot analysis of gluten and commercial soy, teriyaki, and Worcestershire sauces. The detector antibodies from the R5-Sand, G12-Sand, Skerritt, and MIoBS ELISAs were used to visualize the proteins. As shown in the picture after 30 s of exposure (Fig. 4A), the only bands visualized were from the 10- and 2.5-μg/mL gluten standards, and no other bands were detected for all commercial sauces (lanes 1 and 3). Specifically, as expected, the R5 and G12 antibodies recognized gliadins, the Skerritt recognized glutenins, and the MIoBS polyclonal antibodies recognized both gliadin and glutenins. Upon overexposure (for 2 min), many faint smear bands were detected (Fig. 4B). However, we could conclude that gluten was not detected because smear bands were also detected in the lanes without sample (Fig. 4B). Western immunoblot analysis of the model soy sauce moromi samples, collected throughout the fermentation process, also failed to display antigenic bands using the detector antibody from the MIoBS ELISA (data not shown). Though these results are consistent with extensive hydrolysis of the gluten during the fermentation processes, there may still be immunopathogenic peptides present.
The various antibody-based methods routinely used for analytical and regulatory testing for the presence of gluten were unable to detect gluten subjected to the fermentation processes associated with the production of soy sauce and related products. Analyses of samples collected during the fermentation processes indicated that a minimal reduction in the amount of detectable gluten was associated with the first fermentation stage, koji, with no gluten detectable after the first month of the second (brine-based, moromi) fermentation stage. Further, the addition of wheat gluten immediately prior to the addition of brine demonstrated that moromi fermentation alone was sufficient to ferment gluten such that it was not detectable with immunochemical assays such as ELISA and LFD or dipstick tests. Western blot analyses of commercial sauces using four different antibodies were also negative for the presence of gluten. These results are consistent with the fermentation processes extensively hydrolyzing the gluten. However, the inability to detect gluten-derived proteins does not necessarily mean that soy sauce is free of gluten peptides that could elicit responses in individuals with CD or (nonceliac) gluten sensitivities. Additional research is required before concluding that the fermentation processes investigated eliminated all pathogenic elements. Lacking such information, alternative approaches include either establishing that nothing of sufficient mass to elicit a gluten-based pathogenic response is present or clinical studies that demonstrate no adverse biological activity present.
In addition, how similar the fermentation processes associated with soy sauce production are to other fermentation processes used in food production is impossible to reliably gauge. This is because the enzymatic processes and fermentation conditions are not always known or standardized, either within or across a food type. Thus, whereas studies of the fermentation process associated with the production of a model beer indicated the presence of glutenins and certain gliadin-derived peptides in the final product, as measured using immunodiagnostic methods and mass spectrometry (21), other beer products might have a different composition. Indeed, surveys support variability in gluten content, as measured by immunodiagnostic and mass spectrometric methods, across the beer industry (24). This limitation also probably applies to soy sauce as well, and although in this study the 15 commercial soy sauces made with wheat had no detectable gluten, it is impossible to generalize the results without knowing the details of the fermentation and other manufacturing processes used. That is, methods whereby the ingredients are subjected to hydrolysis or a heated extrusion process or are processed in a bioreactor to expedite production have been used (6, 12, 27). In addition, the blending of fermented soy sauce with acid hydrolyzed vegetable protein has been used to augment production (29). Thus, unless demonstrated otherwise, variability in the gluten-derived proteinaceous materials present must be expected.
Appreciation is expressed for the technical support provided by the Kikkoman Corporation in Japan for their work in producing the pilot-plant model soy sauce and model wheat-free soy sauce used in this study. Thanks are also expressed to colleagues at the Center for Food Safety and Applied Nutrition, FDA, who accommodated the homemade soy sauce production and to Lora Benoit, Ph.D., and Steven M. Gendel, Ph.D., of IEH Laboratories & Consulting Group for their openness in making available LFDs. Gratitude is also expressed to Stefano Luccioli, M.D., Shaun MacMahon, Ph.D., Lanlan Yin, Ph.D., and George C. Ziobro, Ph.D., of the FDA and L. L. B. Rust, Ph.D. (NIH) for insightful discussions, without which this work would not have been possible. This work was supported in part by the FDA collaborative grant 5U01FD003801.