Determination of Total Amino Acids in Infant Formulas, Adult Nutritionals, Dairy, and Cereal Matrixes by UHPLC–UV: Interlaboratory Validation Study, Final Action 2018.06

Abstract Background A method for the quantification of total amino acids (including taurine and excluding tryptophan) using ultra- HPLC separation coupled to UV detection (UHPLC–UV) was granted First Action status (AOAC 2018.06) by the AOAC INTERNATIONAL Stakeholder Program for Infant Formula and Adult Nutritionals (SPIFAN) in 2018. Objective An interlaboratory study was conducted to further assess method performance against the AOAC Standard Method Performance Requirements (AOAC SMPR® 2014.013). Dairy and cereal matrixes were added to expand the scope of the method in collaboration with IDF (International Dairy Federation), ISO (International Organization for Standardization), and AACCI (American Association of Cereal Chemists International, now Cereals & Grains Association). Methods Sixteen different matrixes were chosen to cover the requirements of AOAC, IDF/ISO, and AACCI. Blind duplicate samples were organized into specific series to ensure that each pair was analyzed on the same day. Fifteen laboratories returned results. Data from four laboratories were considered invalid and removed from the dataset. Remaining data were assessed according to the Appendix D of the AOAC Official Methods of AnalysisSM (guidelines for collaborative study procedures). Results This method generally met the requirements listed in the SMPR for infant formulas and adult nutritionals, except for taurine. Method performance was comparable in dairy and cereal matrixes. Five different UHPLC instruments were used with either commercial or in-house reagents, demonstrating that the method is not limited to a single supplier. Conclusion This method was recommended for Final Action in infant and adult/pediatric nutritional formulas by the AOAC SPIFAN Nutrients Expert Review Panel in April 2021, with the exception of taurine. The corresponding IDF/ISO Draft International Standard (DIS) was approved by national bodies in May 2022, and comments collected during the ballot were incorporated into this manuscript. Highlights AOAC Official Method 2018.06 for the determination of total amino acids in infant formulas, adult nutritionals, dairy, and cereal matrixes was successfully validated in an interlaboratory study.

To date, there is no standard method for the determination of total amino acids in infant formulas and adult nutritionals. Total amino acid methods in other food and feed matrixes such as AOAC Method 994.12 (1), AOAC Method 985.28 (2), American Association of Cereal Chemists International, now Cereals & Grains Association (AACCI) 07-01.01 (3), and International Organization for Standardization (ISO) 13903:2005 (4) are all based on acid hydrolysis followed by ion exchange separation and post-column derivatization with ninhydrin, and analysis of sulfur amino acids requires an extra overnight oxidation step.
A faster method combining cysteine/cystine conversion by 3,3'-dithiodipropionic acid (DDP) into S-2-carboxyethylthiocysteine (XCys) during the acid hydrolysis step (5) and pre-column derivatization with 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate (AQC) followed by reversed-phase ultra-HPLC (UHPLC) separation (6) was developed for the quantification of total amino acids and taurine (with the exception of tryptophan) in infant formulas and adult nutritionals (7). This method was granted First Action status by AOAC SPIFAN in September 2018 (AOAC Method 2018.06), and a collaborative study was set up to provide additional method performance data. Dairy and cereal matrixes were added to the study to expand the scope of the method in collaboration with IDF, ISO, and AACCI (now Cereals & Grains Association).

Interlaboratory Study Design
The interlaboratory protocol contained three phases: system qualification (to ensure baseline separation of all amino acids), laboratory qualification (using practice samples), and the interlaboratory study itself. For the last phase, the 32 blind duplicate samples were split into two different analytical series (four if needed) to ensure that sample pairs were analyzed within the same series. Results from the practice samples and interlaboratory study were communicated to the study director.

Matrixes
Sixteen samples were categorized into SPIFAN, dairy, and cereal matrixes (Table 1). SPIFAN powder samples were first reconstituted in water as described in SMPR 2014.013 (8), and the other samples were analyzed as is.

Participating Laboratories, Equipment, and Chemicals Used
Fifteen laboratories returned results. One laboratory did not analyze the two liquid samples and one laboratory only analyzed the SPIFAN powder samples. A survey was conducted among the participants to evaluate the diversity of instrument and reagents used during this study, and thirteen laboratories provided answers (see Supplemental Table 1).
Different instruments (six Waters Acquity, four Waters Acquity H-class, one Waters Acquity I-class, one Thermo 3000 RS, and one Agilent 1200 Infinity II) were used. All laboratories used the column specified in the method. For the mobile phase A, 11 laboratories prepared Eluent A from the AccQÁTag Ultra Eluent A concentrate according to the protocol, one laboratory only used 120 mL of the concentrate (instead of 150 mL), and one laboratory used the Alternative Eluent A. For the mobile phase B, seven laboratories used the Eluent B from Waters and six laboratories used the Alternative Eluent B. Twelve laboratories used the AccQÁTag Ultra derivatization kit from Waters and one laboratory used the Alternative Derivatizing Reagent. [Applicable for quantitative determination of total amino acids (AAs) including alanine, arginine, aspartic acid (combined with asparagine), cystine (dimer of cysteine, combined with cysteine), glutamic acid (combined with glutamine), glycine, histidine, isoleucine, leucine, lysine, methionine, phenylalanine, proline, serine, threonine, tyrosine, and valine in infant and adult/pediatric nutritional formulas. Method is not suitable for the determination of tryptophan and taurine.] Note: Other matrixes such as dairy products, infant cereals, and pet foods were part of the multi-laboratory study for this method, but those data have not been evaluated for AOAC approval.
Caution: Refer to Material Safety Data Sheets prior to use of chemicals. Use appropriate personal protective equipment when performing testing. Because of the use of chemical solvents, acids, and reagents, perform sample preparation under a fume hood and take appropriate safety precautions.

A. Principle
Proteins are hydrolyzed in 6M HCl for 24 h at 110 C in the presence of phenol, 3,3 0 -dithiodipropionic acid (DDP), and norvaline (Nva). Phenol is added to prevent halogenation of tyrosine. Nva is added as an internal standard. Cystine and cysteine are converted to S-2carboxyethylthiocysteine (XCys) by DDP (5). After hydrolysis and neutralization, amino acids and XCys are derivatized with 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate (AQC). Derivatized AAs are separated using reversed phase UHPLC with UV detection at 260 nm. (Fluorescence detection is also an option.) During acid hydrolysis, glutamine (Gln) and asparagine (Asn) are converted to glutamic acid (Glu) and aspartic acid (Asp), respectively. Thus, Glu values represent the combined values of Glu and Gln, and Asp values represent the combined values of Asp and Asn. Cys values represent the combined values of cysteine and cystine since both are converted to XCys by DDP. Tryptophan is degraded by acid hydrolysis and cannot be analyzed by this method.              Table 2018.06A describes how to prepare calibration standards for converted cystine at 0-10 pmol/lL and Nva at 10 pmol/lL (all are final concentrations after derivatization). Note: Phenol-HCl solution has to be added under the fume hood.

E. Sample Analysis
Sparge the tube for a minimum of 5 s with a stream of nitrogen to displace oxygen. Close tubes with screw caps and mix on a vortex mixer. Note: Make sure the caps are perfectly clean (i.e., devoid of any particles) to ensure tightness and avoid evaporation during hydrolysis. (e) AAs calibration standards preparation.-Acid hydrolysis is not necessary. Table 2018.06B shows how to prepare 0.5 mL calibration standards at 0-25 pmol/lL and Nva at 10 pmol/lL (all are final concentrations after derivatization). The AA solutions are stable for 1 week when stored at 4 6 2 C. (f) Derivatization (of samples, cystine standards, and AAs standards).-Derivatization converts free AAs into highly stable derivatives. Standards and samples are derivatized following the manufacturer's instructions as described below.
(1) Preheat a heating block to 55 C.
(3) Allow the chromatographic system to stabilize before injecting standards and samples. Make sure the system pressure and initial conditions are stable before performing injections (around 62 MPa/9000 psi/620 bar). (4) Before starting a series of analyses, inject two blanks (water) to condition the column.  calibration standards. If a peak has not been integrated correctly, call the recorded data and reintegrate.
To verify that the system is stable, inject a mid-level standard a minimum of three times (five times for U.S. Pharmacopeia requirements) and ensure that response and retention times have an RSD <2%. Check that peaks are separated with a good resolution (baseline separation). If this is not the case, adapt the chromatographic conditions (e.g., gradient, temperature, tubing length, etc.) accordingly.
Verify that the derivatization reagent was present in excess. Excess reagent hydrolyzes to yield 6-aminoquinoline (AMQ), a noninterfering byproduct present on the chromatogram as the large peak first to elute. In samples, the response of the AMQ peak should not be smaller than that observed in the 25 pmol/lL standard, or the reaction and sample should be flagged and discarded. The derivatization peak at approximately 17 min prior to lysine can be ignored.

F. Calculation and Expression of Results
Force the linear regression through zero. Check the linearity of the calibration (the correlation coefficient R 2 must be above 0.99). (b) AA calculation.-Calculate the amount of individual AAs present in the sample in picomoles per microliter from the calibration curve using the following equation: where C s ¼ concentration of individual AA in the test sample solution in picomoles per microliter; A s ¼ peak area of individual AA in the test sample solution; C is ¼ concentration of internal standard injected in picomoles per microliter; A is ¼ peak area of internal standard chromatogram; and S ¼ slope of the calibration curve (all curves are forced through zero; equation y ¼ ax). Calculate the mass fraction, w, of each AA, in mg/100 g product, using the following equation: where M A ¼ molar mass of individual AAs in grams per mole (see Table 2018.06D); V s ¼ volume of hydrolysis solution in milliliters (typically 5 mL); d 1 ¼ dilution factor in the neutralization step, E(d) (typically 4 or 10); d 2 ¼ dilution factor in the derivatization step, E(f) (typically 10); m s ¼ mass of the test portion in milligrams; and 10 ¼ combined factor to convert picograms to milligrams (10 À9 ), milliliters to microliters (10 3 ), and micrograms to 100 g (1/10 À5 ).

Laboratory Qualification
Fifteen laboratories returned data. Eight laboratories submitted data from the practice samples before proceeding to the interlaboratory study. All eight laboratories provided satisfactory results (i.e., at least 30 out of their 35 reported values were within 20% of the single-laboratory validation average value) and qualified to participate to the interlaboratory test ( Table 2). The remaining seven laboratories directly proceeded to the interlaboratory test phase and returned the results. All results are reported in Supplemental File 1.

Data Validation
The data set was first analyzed to detect laboratories that should be excluded from the study because of a global bias or   Table 2, Laboratories 1, 5, and 13 showed a global bias (their average value 6 SD did not include the calculated mean), and laboratories 1 and 7 showed an extensive spread (their RSD exceeded twice the average RSD of all laboratories).
To confirm those observations, z-scores were calculated for each individual amino acid result based on the average and SD of the 11 laboratories that did not show a global bias or extensive spread. An absolute z-score higher than 2 indicates that the result is more than 2 SDs away from the mean and is considered as questionable (9). As shown in Table 2, Laboratories 1, 5, and 13 had absolute z-scores values above 2 in 74%, 91%, and 46% of the cases, with average z-scores of 18.7, À5.7, and À2.1, respectively.
Laboratory 1 noted fluctuations of the norvaline peak in the standards and inconsistencies in sample peak areas. Laboratory 5 noted that the instrument had to be repaired during this period and that the injection volumes were too small. Nothing was reported from Laboratories 7 and 13.
In summary, data from Laboratories 1, 5, 7 and 13 showed constant and severe deviations and were considered as invalid. Similarly, tyrosine was systematically underestimated in Laboratory 2, and cysteine results were systematically underestimated in Laboratories 4 (SPIFAN powder and cereal samples) and 12 (all samples). All other values were considered as valid.

Outlier Removal
For each analyte/sample pair, Cochran, Grubbs and double Grubbs statistical tests were conducted on the valid data using the "AOAC Interlaboratory Study Workbook-Blind (Unpaired) Replicates" template (https://www.aoac.org/resources) according to AOAC Appendix D (10; see Table 3 for details). No more than 2/9 laboratories were dropped for each analyte/sample pair during this process, and at least eight entries remained for each set except in a few cases, notably when the starting number of laboratories was reduced to nine or eight for the liquid samples (see Table 3 for details). Overall, 6% (164 out of 2857) of the values were judged outliers; about one third of the outliers were caused by a single laboratory, and more than 80% of outlier values were excluded after a Cochran test. Invalid data and outliers are highlighted in Supplemental File 1.

Performance Data
Final performance data were computed after outlier removal and are summarized in Table 3 (see Supplemental File 2 which lists performance data prior to outlier removal). Table 3 lists for each analyte/sample pair the number of valid results, outliers, and remaining duplicates, as well as the average value, SD of repeatability (SD r ) and reproducibility (SD R ), RSD of repeatability (RSD r ) and reproducibility (RSD R ), and Horwitz ratio (HorRat). The concentrations are expressed in mg per 100 g of reconstituted product for SPIFAN powder samples and in mg per 100 g of product (liquid or powder) for the other samples. Most samples have HorRat values that are between 0.5 and 2, which indicates "Method reproducibility as normally would be expected" (10).
In SPIFAN matrixes, RSD r and RSD R values were compared to the performance criteria defined in SMPR 2014.013 (8), and exceeding values are underlined in Table 3. Repeatability requirements were met for 154 of the 161 analyte/sample pairs analyzed. Reproducibility requirements were met for 135 of the 161 analyte/      sample pairs analyzed; exceeding values were mostly observed for cysteine and taurine, with average RSD R values of 11.6% and 18.5%, respectively. For all other amino acids, the RSD R values were within the SMPR boundaries in 134 out of 144 cases, and only two values were above the RSD r threshold.
Cystine calibrants must be derivatized prior to injection, which most likely increases the variability for this analyte. However, being able to analyze sulfur amino acids at the same time as the other amino acids represents an important improvement over existing methods and the SPIFAN Nutrients Expert Review Panel recommended keeping cysteine in the final method.
Taurine levels are about 10-100 times lower than those of the other amino acids, leading to higher RSD R values. The SPIFAN Nutrients Expert Review Panel recommended the exclusion of taurine from the final method because other methods such as AOAC Method 997.05 (11) can measure low levels of taurine with higher accuracy. Taurine results are included in this manuscript for the sake of completeness, and readers can refer to the single-laboratory validation manuscript (7) for the taurine protocol.
In dairy and cereal matrixes, 78 of the 111 analyte/sample observations had an RSD r below 3%, 28 observations were between 3 and 5%, and 5 were above 5%. For RSD R , 38 of the 111 observations were below 5%, 54 between 5 and 8%, and 19 above 8%. Several laboratories used at least one in-house alternative reagent. Laboratory 6 used only in-house alternative reagents and provided accurate results, with 84% of z-scores between À1 and 1 (228/273), no z-score above 2, and an average z-score of À0.01 (see Supplemental Table 2).
Recovery had previously been assessed by spike experiments during the single-laboratory validation (7). During this interlaboratory study, trueness was assessed by comparing the average values calculated in sample S1 (NIST SRM 1869), D4 (NIST SRM 1549a), and C3 (NIST SRM 3233) with the reference values from the corresponding certificate of analysis (CoA). As shown in Table 4, 37/50 amino acid values were within the concentration range described in the CoA, and 39/50 values were within 10% of the reference values. The differences were larger for the cereal reference sample C3. This might be specific to this sample and not to the matrix type since values for samples C1 and C2 were within 10% from those obtained by a classical postcolumn derivatization method (Table 4; this comparison is purely informative since samples C1 and C2 were only analyzed once by the post-column derivatization method).

Conclusions
An interlaboratory study for method AOAC Method 2018.06 was successfully conducted in sixteen SPIFAN, dairy, and cereal matrixes by 15 different laboratories using five different UHPLC instruments. The laboratories used either commercial or inhouse reagents, demonstrating that the method is not limited to a single supplier.
This interlaboratory study demonstrates the robustness and fitness-for-purpose of this method in different matrix groups, and the method is expected to be applicable to a wider variety of matrixes based on its harsh but robust chemistry. This study can now serve as a benchmark for future method improvements such as microwave-assisted hydrolysis and detection using single or triple quadrupole mass spectrometers.
The initial data validation step was questioned by the AOAC SPIFAN Nutrient Expert Review Panel, IDF Standing Committee on Analytical Methods for Composition (SCAMC), and IDF Standing Committee on Statistics and Automation (SCSA). The main argument against the rejection of laboratories showing too much bias or spread was that the method performance might be artificially overestimated, and that it will be difficult for new laboratories to match those performances when implementing the method. We are confident that the method performances presented here are realistic and that the invalid data do not represent the true performance of the laboratories concerned (two of them reported technical issues, and two of them did not return results from the practice samples). New laboratories that will implement this method will have time to properly verify their performance using standard reference materials such as those described in this study.