The quantification is based on the catalytic activity of alcohol dehydrogenase, which oxidizes ethanol to acetaldehyde and converts nicotinamide-adenine dinucleotide (NAD) to its reduced form, NADH. Measurement is performed at 340 nm in the RIDA^®CUBE SCAN device (Figure 1) and calculated against a calibration function stored on a lot-specific RFID card (see Table 1).

The test kit RIDA^®CUBE Ethanol (R-Biopharm, art. no. RCS4340) consists of 32 tubes (containing 800 µL reagent 1), 32 caps (200 µL of reagent 2) and one lot-specific RFID card. For measurement, the RFID card is placed on the RIDA^®CUBE SCAN (R-Biopharm, art. no. ZRCS0546 or ZRCS0580) device as shown in Table 1. Next, information about the sample is entered into the tablet app. One tube (see also Figure 2) is opened and 20 µL of diluted/undiluted sample is added. The tube is closed with the cap and inserted in the scan device. After closing the door of the device, the analysis is finished automatically and a concentration for the analyzed sample is given on the tablet. A dilution step prior to measurement needs to be taken into account for calculation. If the result should be given as % alcohol-by-volume (%ABV) calculate as follows: % ABV = Ethanol [g/L]/7.8924 [1] .

Following AOAC guidelines, which are published as Appendix D [4] , an

international collaborative study was set up. Detailed instructions for collaborators and data return sheets were provided to all participants. In order to qualify participants, we asked each laboratory to analyze two aqueous reference materials with known ethanol concentrations (0.1 g/L and 2.0 g/L). Results were forwarded to the study director before each lab could continue with the analysis of the matrix samples.

Eight laboratories participated: One each from France, Canada, United States of America, and Argentina, two from Italy and two from Germany. To mimic the typical user of such a measurement device, it was decided not to look for the best-trained specialist in analysis but in analysts that will use the method after reading the instruction for use and a short training. The participants were advised to ensure that no alcoholic vapors (ethanol or propanol) e.g. from cleaning solutions or beverages with high ethanol content were present during sample preparation and measurement of the extracts. Samples were shipped to the laboratories in September 2020, and all results were received by the end of December 2020.

Three matrices (kombucha, orange juice, and beer) were tested at different concentrations (Table 2) up to 9 g ethanol per L (equivalent to 1.14% alcohol by volume; % ABV).

Each of the 14 different samples was analyzed as blind duplicates by each participant. For spiking of the juice and beer, EMSURE^® Ethanol (absolute for analysis; Merck, 1.00983.1000) was used. For preparation, a 100 mL volumetric flask was filled with about 50 mL distilled water. Twenty mL of absolute ethanol was pipetted into the flask, mixed and filled up with water up to 100 mL. This solution was used for spiking the matrices designated as spiked in Table 2.

^aboth from ACQ Science, Rottenburg-Hailfingen, Germany; volume is 1.5 mL.

Following the collaborative test guidelines of AOAC Appendix D [4] , a set of blind duplicates for each sample was provided to each participating laboratory. The samples were marked with a laboratory-specific letter (B to N) and a randomized number. Each laboratory obtained its own coding (different randomized numbers for each laboratory).

The method protocol was provided to each laboratory with the instruction to follow the method as written with no deviations. The final data from the laboratories were sent to the study director. The participants were advised to analyze all samples with an initial dilution of 1:20. After identification of diluted samples with concentrations lower than 20 mg/L, the participants were advised to run these samples undiluted.

A total of eight data sets from eight participants were obtained for the pre-collaborative study using two aqueous solutions with known ethanol concentrations (Table 2 and Table 3). Lab K and lab F only analyzed one of the solutions while lab N analyzed the high-concentrated solution only once.

Relative repeatability standard deviation RSD(r) and relative reproducibility standard deviation RSD(R) are within the expected range and nearly fulfil the requirement RSD(R) of ≤6% laid down in AOAC SMPR 2016.01 [1] . Consequently, all participants were allowed to proceed with the analysis of matrix samples.

Data sets were checked for outliers (Tables 4-6) according to AOAC Appendix D [4] . In total, four out of eight data sets showed no outliers at all. Six pairs of duplicates out of a total of 112 pairs were outliers according to Cochran marked with a small type c in Tables 4-6. Outliers according to the Cochran test indicate higher differences between the two replicates compared to the other labs. There is no tendency towards a specific proportion of outliers due to a participant or a sample. The pattern looks quite random.

After elimination of outliers, the performance characteristics repeatability standard deviation s(r) and reproducibility standard deviation s(R) were calculated using the AOAC International Interlaboratory Study Workbook for Blind (Unpaired) Replicates (version 2.0). Table 7 shows the results of these calculations for the kombucha samples.

Table 8 compiles these calculations for orange juice and beer. In the past, it was common to characterize each s(r) or s(R) individually for each sample. Here we would like to do an additional characterization for all kombucha sample at a time as described by Lacorn et al. [3] . In an effort to interpret observed RSD values across a wide concentration range, we have modeled reproducibility standard deviation s(R), as a function of mean observed concentration.

We analyzed the relationship by simple least squares linear regression (Figure 3). The procedures used for modeling reproducibility by mean are taken from ISO 5725-2 [5] . The coefficient of regression is 0.964, and the slope of the regression line is 0.0629, which corresponds to an overall RSD(R) of 6.29%. As expected, the estimated relative standard deviation will rise as the concentration approaches zero (Table 6). Kombucha I with a mean ethanol concentration of 0.031 g/L is close to the LoQ of the method and the higher relative standard deviation was in the expected range.

Other methods for determination of ethanol in Kombucha are OMA 2019.08 [6] and OMA 2016.12 [7] . OMA 2019.08 is also based on an enzymatic principle but was not studied collaboratively until today [6] . OMA 2016.12 is a gas-chromatographic method with RDR(R) values between 10% for low-concentrated kombucha samples and around 5% for samples with higher ethanol concentrations [7] .

Analysis of orange juice and beer showed an overall higher variation as can be seen in Table 5. RSD(R) range from 8.7% up to 15.2% depending on the concentration of the samples. The estimated overall RSD(R) value by regression is 14.4% (Figure 4). It is obvious that one sample deviates from the others for unknown reasons, but the coefficient of regression is still at 0.961.

It was not possible to calculate recovery for kombucha since all samples were naturally contaminated. Nevertheless, recovery data on reference materials is provided in Table 4.

As expected for the group of participants, especially precision of reproducibility is sometimes quite high for orange juice and beer. It should bear in mind that these samples had to be centrifuged or degassed whereas kombucha could be used directly. From our experience, more training of some participants will decrease this un-precision. Since the analysis of aqueous solutions during the pre-collaborative test resulted in much better precision in any case, it is clear that not the measurement per se is the main contributor to uncertainty but the handling and preparation of matrix samples. As stated above, pipetting ethanol-containing solutions is also a critical point that needs to be accounted for. The combination of the RIDA^®CUBE Ethanol and the RIDA^®CUBE SCAN was especially developed for users with a low sample throughput, limited laboratory facilities, and basic knowledge of analytical chemistry. If needed just one sample can be analyzed and due to high stability of the reagents in the test tubes, there is no issue if only a few samples per month have to be analyzed.