TY - JOUR
AU - Zhang,, Kai
AB - Abstract Background In the present study, we developed a novel automated sample preparation workflow for the determination of mycotoxins in foods. Objective This workflow integrates off-line devices such as a centrifuge, shaker, liquid and solid dispensing units into a unified platform to perform gravimetric and volumetric dispensing, capping/decapping, extraction, shaking, filtration, and centrifugation. Two robotic arms provide sample transportation without human assistance. Method Critical method performance attributes were characterized using spiked corn, milk and peanut butter containing aflatoxins, deoxynivalenol, fumonisins, ochratoxin A, HT-2 and T-2 toxins and zearalenone and certified reference materials. Prepared samples were analyzed by liquid chromatography mass spectrometry (LC-MS). Results Recoveries of spiked samples range 100–120% with RSD<20% and the majority of measured values of certified reference materials are consistent with certified values within ±20%. Within- and between-batch variabilities of QC samples range 5–9% and 7–12% respectively. Conclusions Our workflow introduces a straightforward and automated sample preparation procedure for LC-MS-based multimycotoxin analysis. Further, it demonstrates how individual sample preparation devices, that are conventionally used off-line, can be integrated together. Highlights This study shows automated sample preparation will replace manual operations and significantly increase the degree of automation and standardization for sample preparation. With advances in automation technologies, growing interest in replacing manual operations using robotic tools prompts the need for automation systems that can handle repetitive and high-volume tasks. In the field of food analysis, automated sample preparation is still at an early stage, though such practices have been commonly used in areas such as drug screening and routine clinical sample analysis (1, 2). In general, prior to instrumental measurements, food samples need to be processed through multiple steps such as homogenization, weighing, solid phase extraction (SPE), solvent exchange, shaking, evaporation, filtration, centrifugation, and derivatization. This makes automated sample preparation a challenge as each step requires different tools, labware, holders, and sample vessels. As an important part of food monitoring and surveillance, food and feed products are screened for mycotoxins on a regular basis. Two types of sample preparation have dominated mycotoxin analysis using LC, immunoaffinity column/SPE cleanup and dilute-and-shoot (3). The majority of the AOAC official methods or modified versions use the former approach for sample cleanup of single mycotoxin analysis, followed by LC-fluorescence detection/UV detection (4–6). The latter has been increasingly used for LC-MS-based multi-mycotoxin analysis and commonly involves the use of stable isotope internal standards during sample processing (7, 8). For immunoaffinity column/SPE cleanup, challenges still exist to automate conditioning, loading, washing and eluting. If disposable columns are used, manual replacement of the columns after each batch creates another limitation for the development of automated mycotoxin analysis (9). Automated systems developed in the 1990s suffered from inconsistent recoveries due to erratic flowrates through columns or lack of integrated gravimetrical or volumetric dispensing tools (10, 11). A newly-developed robotic system for aflatoxin analysis, however, demonstrates sufficient method performance in terms of precision and accuracy (12–14). With an online conditioning feature and reusable columns, the system can use the same columns up to 15 times, which significantly increases throughput for single mycotoxin analysis. However, prior to the automated sample preparation procedure, human operators need to extract samples and transfer extracts to the system (12–14). Since the early 2000s, mass spectrometry coupled to liquid chromatography has rapidly gained popularity among mycotoxin laboratories. LC-MS has been recognized as a viable tool for the analysis of multiple mycotoxins, promoting the simultaneous detection and identification of mycotoxins without the need for repeat confirmatory analysis (15). Moreover, due to better sensitivity, selectivity and specificity of LC-MS, sample preparation has been significantly simplified, making it more amenable to being automated (16–18). Several studies have reported the development of online SPE coupled with LC-MS/MS for the determination of mycotoxins in beers, juices, grains, feeds, and dried fruits (19–21). Reportedly, these commercial systems worked well for liquid samples, as there are fewer steps involved with extraction of mycotoxins from liquid matrixes. Human operators only need to aliquot samples and load them onto the sample tray. The rest of the procedure (SPE and LC-MS analysis) could be performed automatically using an online autodiluter, autosampler and autoinjector. For solid samples, human operators used multiple offline devices to prepare samples for the online SPE and LC-MS analysis. Certain time-consuming and laborious steps such as extraction and sample transportation are still a burden on human operators. A highly automated sample preparation workflow requires integration of independent devices and capability of transportation and storage so that samples can be processed and moved through each sub-step. In this proof of concept study, we evaluated an automation system, the Chemspeed Swing XL®, focusing on sample preparation for the determination of mycotoxins using LC-MS. This system is equipped with two robotic arms, various modular tools and storage racks, and can conduct transport, gravimetric and volumetric dispensing, shaking, capping/decapping, filtration, and centrifugation without human assistance. To assess the quality of results obtained from a range of food matrixes in terms of precision, accuracy and batch-to-batch variability, we used fortified corn, peanut butter, milk, and certified reference materials to evaluate the performance of the system. Our goal was to use integrated tools to increase the degree of automation and minimize human intervention, paving the way for a highly automated mycotoxin analysis in the future. Experimental Chemicals and Materials HPLC grade acetonitrile, methanol, and water were purchased from ThermoFisher Scientific (Waltham, MA). One QC sample (T04312QC, corn) was obtained from FAPAS (Fera Science Ltd., UK). One reference material, SRM 2387 (peanut butter) and two corn samples were obtained from National Institute of Standards and Technology (NIST, Gaithersburg, MD). Four reference materials, ERM-BE375 (low-level aflatoxins in compound feeding stuff), ERM-BE376 (high-level aflatoxins in compound feeding stuff), ERM-BE283 (low-level aflatoxin M1 in milk powder) and ERM-BE 284 (high-level aflatoxin M1 in milk powder) were purchased from Analytical Reference Materials International (Golden, CO). Amicon Ultra-4 centrifugal filters (15 mL) with Ultracel-3 membrane (3 kDa) were purchased from EMD Millipore (Billerica, MA). Mycotoxin Mix I, II, and III were obtained from Romer Labs Inc. (Union, MO). Mix I contained 1.0 µg/mL each of aflatoxin B1, aflatoxin B2, aflatoxin G1, aflatoxin G2, and ochratoxin A in methanol; Mix II contained 10 µg/mL each of fumonisin B1, B2, and B3 in acetonitrile–water (50:50, v/v); Mix III contained 10 µg/mL each of deoxynivalenol, HT-2 toxin, T-2 toxin, and zearalenone in acetonitrile. Aflatoxin M1 (10 μg/mL) was purchased from Supleco (St. Louis, MO). Working solutions containing these mycotoxins at different concentrations were prepared by mixing and diluting appropriate amounts of the three stock solutions using acetonitrile–water (50:50, v/v). The following 13C-labelled internal standard (IS) stock solutions were purchased from Romer Labs, 13C17-aflatoxin B1 + B2 + G1 + G2 (0.5 µg/mL), 13C17-aflatoxin M1 (0.5 µg/mL), 13C20-ochratoxin A (10 µg/mL), 13C34-fumonisin B1 (25 µg/mL), 13C34-fumonisin B2 (10 µg/mL), 13C34-fumonisin B3 (10 µg/mL), 13C15-deoxynivalenol (25 µg/mL), 13C22-HT-2 toxin (25 µg/mL), 13C24-T-2 toxin (25 µg/mL), and 13C18-zearalenone (25 µg/mL). For corn, peanut butter and reference materials, the extraction solution containing the 12 13C-IS was prepared by mixing and diluting appropriate amounts of the stock solutions using acetonitrile–water (50:50, v/v). The concentration of each 13C-IS in the extraction solution was as follows: 13C17-aflatoxin B1, B2, G1, and G2 (0.001 µg/mL), 13C17-ochratoxin A (0.01 µg/mL), 13C34-fumonisin B1 (0.05 µg/mL), 13C34-fumonisin B2 (0.05 µg/mL), 13C34-fumonisin B3 (0.05 µg/mL), 13C15-deoxynivalenol (0.05 µg/mL), 13C22-HT-2 toxin (0.05 µg/mL), 13C24-T-2 toxin (0.05 µg/mL), and 13C18-zearalenone (0.05 µg/mL). For milk, the extraction solution that contained the 13 13C-IS was prepared by mixing and diluting appropriate amounts of the stock solutions using acetonitrile–water (50/50, v/v). The concentration of each 13C-IS in the extraction solution was as follows: 13C17-aflatoxin B1, B2, G1, G2, and M1 (0.001 µg/mL), 13C17-ochratoxin A (0.01 µg/mL), 13C34-fumonisin B1 (0.05 µg/mL), 13C34-fumonisin B2 (0.05 µg/mL), 13C34-fumonisin B3 (0.05 µg/mL), 13C15-deoxynivalenol (0.05 µg/mL), 13C22-HT-2 toxin (0.05 µg/mL), 13C24-T-2 toxin (0.05 µg/mL), and 13C18-zearalenone (0.05 µg/mL). Automated Sample Preparation System A Chemspeed Swing XL system (Chemspeed, Switzerland) was used to prepare samples throughout the study. The system consisted of a platform (2.35 × 1.92 × 0.95 m, length  ×  height  ×  width), on which the following tools, racks, and plates were installed: one overhead robotic arm, one DENSO robotic arm, one four-channel liquid-handling unit with wash station, one liquid-dispensing unit, one solid-dispensing unit, one multigripper, one screwcapper, two shakers, one vortexing station, three sample vial racks (one for 2 mL LC autosampler vials, one for 15 mL centrifuge tubes and one for 50 mL centrifuge tubes), two cap plates (one for caps of the 15 mL centrifuge tubes and the other for caps of the 50 mL centrifuge tubes), one SPE rack, one blender, one sonicator and one injection valve. Additionally, a Sigma 4-16 KL centrifuge (Sigma, Germany) was attached to the platform. The overhead robotic arm could move within the platform and pick up different tools to perform various tasks as programmed using localization and mapping features. The DENSO robotic arm was used to transfer samples to and from large devices that could not fit in the platform. In this study, the main function of the DENSO robotic arm was to transport samples in and out the centrifuge. The layout of the system is illustrated in an overhead view (Figure 1). The system was built based on a modular design, so tools, racks and holders were developed with standard interfaces or dimensions. This made it flexible and easy to choose, install and switch different modular components depending on applications. The system was controlled by Chemspeed AutoSuite software, which was used to program, simulate, run and record applications on the system. Figure 1. Open in new tabDownload slide Overhead view of the automated sample preparation system. 1: centrifuge; 2: DENSO robotic arm; 3: centrifuge bucket tray; 4/9: 4 needle-head liquid handling unit/syringes; 5: liquid-dispensing unit; 6: SPE rack; 7: vortexing station; 8: solid-dispensing unit; 10: cap plate for 15 mL vials; 11: screwcapper; 12: cap plate for 50 mL vials; 13: rack for 15 mL vials; 14: multigripper; 15: rack for 50 mL vials; 16: rack for 2 mL LC autosampler vials; 17: sonicator; 18: rack for disposable syringes; 19: blender; red crosshair: overhead robotic arm. Note: The above schematic was modified based on a simulation of the sample preparation system generated using Chemspeed AutoSuite. Figure 1. Open in new tabDownload slide Overhead view of the automated sample preparation system. 1: centrifuge; 2: DENSO robotic arm; 3: centrifuge bucket tray; 4/9: 4 needle-head liquid handling unit/syringes; 5: liquid-dispensing unit; 6: SPE rack; 7: vortexing station; 8: solid-dispensing unit; 10: cap plate for 15 mL vials; 11: screwcapper; 12: cap plate for 50 mL vials; 13: rack for 15 mL vials; 14: multigripper; 15: rack for 50 mL vials; 16: rack for 2 mL LC autosampler vials; 17: sonicator; 18: rack for disposable syringes; 19: blender; red crosshair: overhead robotic arm. Note: The above schematic was modified based on a simulation of the sample preparation system generated using Chemspeed AutoSuite. Automated Sample Preparation As illustrated in the workflow diagram (Figure 2), initial steps such as loading consumables (e.g., vials, syringes) on racks, and filling sample-dispensing containers and solvent bottles need to be performed manually for automated sample preparation using the following steps. The first automated step was to program dispensing units to calibrate and rinse the solvent-dispensing systems, including tubing, syringes, and needles. The second step was to open sample vials using the screwcapper, dispense samples into 15 mL vials (1000 ± 25 mg) using the solid-dispensing unit, followed by the addition of extraction solvent (5 mL 50% acetonitrile containing 13C-IS) using the 4-channel liquid-handling unit, which can volumetrically deliver the extraction solvent. The solid-dispensing unit consisted of an analytical balance (readability 0.1 mg) and a sample container with an extruding device inside. Based on the reading on the analytical balance, the extruding device kept adjusting its rotation rate, amplitude and time until the target mass was reached within a predefined range (e.g., 1000 ± 25 mg). It is worth noting that the current solid-dispensing unit can only handle dry and powder matrixes such as wheat flour and corn meal. For viscous pastes, slurries and fatty foods such as peanut butter, samples needed to be weighed into sample vials and loaded on the racks prior to the automated steps. The third step was extraction by initiating the shaker at 800 rpm for 30 min under the sample rack. The shaking feature was integrated into the automation system, enabling shaking with different speeds (up to 1350 rpm) and times (up to 24 h). The fourth step was centrifugation. After shaking was over, the multigripper was programmed to transfer the sample vials from the rack to centrifuge buckets, which were moved into the centrifuge by the DENSO robotic arm as the centrifuge is attached to the Swing XL platform and could not be reached by the overhead robotic arm. Once centrifugation (4000 g, 15 min) was completed, the DENSO robotic arm carried the centrifuge buckets out of the centrifuge and the multigripper moved the vials back to the rack. The screwcapper decapped sample vials and Amicon Ultra-4 centrifugal filters (15 mL) and the 4-channel liquid-handling unit pipetted the supernatant from each sample vial to the centrifugal filters. The fifth step was centrifugation of the centrifugal filters under conditions in step four. The last step was to transfer filtrates into LC autosampler vials. To accomplish such an operation, centrifugal filters were moved back to the rack and decapped, the multigripper removed the insert out and then 1 mL of filtrate was transferred into an LC autosampler vial loaded in another rack using the 4-channel liquid-handling unit, which could be equipped with up to four syringes with different volumetric capacity ranging 1 to 25 mL and needles (flat-headed or piercing needles). The current system had an online injection valve but was not connected with any analytical instrument, so it required manual transportation of prepared samples to a LC-MS. Figure 2. Open in new tabDownload slide Workflow for mycotoxin sample preparation using integrated tools (in blue). Figure 2. Open in new tabDownload slide Workflow for mycotoxin sample preparation using integrated tools (in blue). LC-MS/MS Analysis Samples were analyzed using LC-MS/MS conditions established in a previous study (22). Solvent-only calibration standards were prepared by using the 4-channel liquid-handling unit to volumetrically mix working solutions at different concentrations (0.5 mL each) with the extraction solvent (0.5 mL). Each mycotoxin was quantitated using linear least-squares calibration of the relative response ratio of the mycotoxin and its 13 C-IS plotted versus concentrations. Results and Discussion Automated Sample Preparation Workflow Our workflow was based on a stable isotope dilution multi-mycotoxin method validated in a previous study. In brief, samples needed to be fortified with labeled internal standards, followed by extraction, centrifugation and filtration. The aliquots of the resulting filtrates were transferred into autosampler vials for LC–MS analysis (22). To automate this procedure, a workflow diagram was created to define key sample preparation tasks and tools needed (Figure 2). As one could expect, each task and tool would need a set of specific parameters. Transportation tasks (e.g., transfer vials from one rack to another) involve positioning of the initial and final racks (wells) (x, y, and z coordinates), and dimensions of the vials. To successfully cap/decap sample vials, the screwcapper needs to be programmed based on the size of the cap, threading patterns and tightness of the cap needed (e.g., caps could come loose after vigorous shaking). Centrifugation required sample vials to be placed at the correct position, balancing the buckets. For liquid dispensing, volume, dispensing rate and viscosity should be factored in when programming the 4-channel liquid-handling unit. These parameters were tested using method simulation, a troubleshooting feature which could check programmed tasks, transportation of samples, positions of storage racks, plates and sample vials, and other operating parameters. This helped operators calibrate tools and modify various parameters without running real samples. Potential collisions or inexecutable moves could be identified and corrected. In the course of method development, even certain parameters could pass simulation, they still need to be tested using real samples. Performance of the Automated Sample Preparation System Performance of the system was evaluated by programing the system to prepare calibration curves, spiked samples, QC samples, and certified reference materials for LC-MS analysis. Quantitation of mycotoxins was achieved using calibration standards and curves, so we programed the automated sample preparation system to prepare calibration standards using the 4-channel liquid-handing unit to test whether the preparation of calibration standards could be incorporated into this workflow. For each mycotoxin, three batches of calibration standards were prepared at ten different concentrations ranging from 0.1 to 100 ng/mL for aflatoxins and ochratoxin A and 1 to 1000 ng/mL for deoxynivalenol, fumonisins, HT-2 toxin, T-2 toxin and zearalenone. Resulting calibration curves were evaluated in terms of their slopes and correlation coefficients (r2). All calibration curves had an r2 >0.99. The consistence performance of the 4-channel liquid-handling unit could be further proved by quantitative analysis. For example, Figure 3 shows calibration curves of fumonisin B1 from three batches. If they were used to quantitate same samples, the RSDs of the quantitative results generated using the three calibration curves would vary less than 6% within the calibration range (10–1000 ppb). Figure 3. Open in new tabDownload slide Comparison of quantitative results using three calibration curves of fumonisin B1 prepared by automated 4-channel liquid-handling unit (batch 1: red; batch 2: blue and batch 3: green; n = 10/batch). Figure 3. Open in new tabDownload slide Comparison of quantitative results using three calibration curves of fumonisin B1 prepared by automated 4-channel liquid-handling unit (batch 1: red; batch 2: blue and batch 3: green; n = 10/batch). Manually-spiked corn, peanut butter and milk containing aflatoxin B1, B2, G1 and G2 and ochratoxin A at 5 and 100 ng/g and fumonisin B1, B2, and B3; deoxynivalenol, HT-2/T-2 toxin and zearalenone at 50 and 1000 ng/g. Spiked milk also contained aflatoxin M1 at 0.2, 5 and 100 ng/g. These samples were processed using the automated sample preparation system, followed by LC-MS analysis. In spiked corn, peanut butter, and milk, most recoveries were between 100 and 120% with RSDs <10% with a few exceptions. In spiked corn, peanut butter, and milk, most recoveries were between 100 and 120% with RSDs <10% with a few exceptions. For example, at 50 ng/g in corn, the recovery of deoxynivalenol was 123%. At 50 ng/g in milk, the recovery of T–2 toxin was 127% (Tables 1 and 2). These recovery data indicate consistent method performance that was not affected by different matrixes and clearly show that the spiked mycotoxins can be extracted, identified and quantitated in the three representative matrixes via the automated workflow. Table 1. Average recoveries (relative standard deviation, RSD %, n = 4) in spiked milk Mycotoxins . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . AFB1 0.2 103 (2) 5 111 (1) 100 105 (4) AFB2 0.2 108 (7) 5 112 (2) 100 108 (3) AFG1 0.2 106 (5) 5 104 (3) 100 110 (5) AFG2 0.2 112 (5) 5 111 (2) 100 107 (3) DON Not spiked —a 50 109 (2) 1000 103 (1) FB1 Not spiked —a 50 111 (2) 1000 105 (3) FB2 Not spiked —a 50 118 (2) 1000 110 (1) FB3 Not spiked —a 50 111 (1) 1000 107 (1) OTA Not spiked —a 50 106 (4) 1000 103 (4) T-2 Not spiked —a 50 127 (7) 1000 116 (8) HT-2 Not spiked —a 50 114 (4) 1000 104 (5) ZON Not spiked —a 50 113 (3) 1000 116 (9) AFM1 0.2 97 (5) 50 103 (3) 100 111 (1) Mycotoxins . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . AFB1 0.2 103 (2) 5 111 (1) 100 105 (4) AFB2 0.2 108 (7) 5 112 (2) 100 108 (3) AFG1 0.2 106 (5) 5 104 (3) 100 110 (5) AFG2 0.2 112 (5) 5 111 (2) 100 107 (3) DON Not spiked —a 50 109 (2) 1000 103 (1) FB1 Not spiked —a 50 111 (2) 1000 105 (3) FB2 Not spiked —a 50 118 (2) 1000 110 (1) FB3 Not spiked —a 50 111 (1) 1000 107 (1) OTA Not spiked —a 50 106 (4) 1000 103 (4) T-2 Not spiked —a 50 127 (7) 1000 116 (8) HT-2 Not spiked —a 50 114 (4) 1000 104 (5) ZON Not spiked —a 50 113 (3) 1000 116 (9) AFM1 0.2 97 (5) 50 103 (3) 100 111 (1) a No data. Open in new tab Table 1. Average recoveries (relative standard deviation, RSD %, n = 4) in spiked milk Mycotoxins . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . AFB1 0.2 103 (2) 5 111 (1) 100 105 (4) AFB2 0.2 108 (7) 5 112 (2) 100 108 (3) AFG1 0.2 106 (5) 5 104 (3) 100 110 (5) AFG2 0.2 112 (5) 5 111 (2) 100 107 (3) DON Not spiked —a 50 109 (2) 1000 103 (1) FB1 Not spiked —a 50 111 (2) 1000 105 (3) FB2 Not spiked —a 50 118 (2) 1000 110 (1) FB3 Not spiked —a 50 111 (1) 1000 107 (1) OTA Not spiked —a 50 106 (4) 1000 103 (4) T-2 Not spiked —a 50 127 (7) 1000 116 (8) HT-2 Not spiked —a 50 114 (4) 1000 104 (5) ZON Not spiked —a 50 113 (3) 1000 116 (9) AFM1 0.2 97 (5) 50 103 (3) 100 111 (1) Mycotoxins . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . AFB1 0.2 103 (2) 5 111 (1) 100 105 (4) AFB2 0.2 108 (7) 5 112 (2) 100 108 (3) AFG1 0.2 106 (5) 5 104 (3) 100 110 (5) AFG2 0.2 112 (5) 5 111 (2) 100 107 (3) DON Not spiked —a 50 109 (2) 1000 103 (1) FB1 Not spiked —a 50 111 (2) 1000 105 (3) FB2 Not spiked —a 50 118 (2) 1000 110 (1) FB3 Not spiked —a 50 111 (1) 1000 107 (1) OTA Not spiked —a 50 106 (4) 1000 103 (4) T-2 Not spiked —a 50 127 (7) 1000 116 (8) HT-2 Not spiked —a 50 114 (4) 1000 104 (5) ZON Not spiked —a 50 113 (3) 1000 116 (9) AFM1 0.2 97 (5) 50 103 (3) 100 111 (1) a No data. Open in new tab Table 2. Average recoveries (relative standard deviation, RSD, %; n = 4) in spiked corn and peanut butter Mycotoxins . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . Corn . Peanut butter . Corn . Peanut butter . AFB1 5 110 (1) 118 (8) 100 108 (3) 111 (3) AFB2 5 113 (2) 113 (1) 100 110 (2) 118 (2) AFG1 5 107 (1) 112 (2) 100 105 (2) 110 (1) AFG2 5 115 (0.2) 97 (7) 100 112 (1) 115 (4) DON 50 123 (9) 116 (8) 1000 107 (2) 112 (2) FB1 50 106 (9) 104 (3) 1000 110 (4) 113 (6) FB2 50 112 (3) 121 (4) 1000 113 (3) 119 (4) FB3 50 119 (7) 112 (4) 1000 111 (2) 110 (2) OTA 50 103 (4) 116 (5) 1000 104 (5) 103 (2) T-2 50 116 (2) 114 (2) 1000 113 (3) 114 (3) HT-2 50 109 (1) 111 (2) 1000 111 (2) 112 (6) ZON 50 109 (5) 114 (3) 1000 110 (2) 112 (3) Mycotoxins . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . Corn . Peanut butter . Corn . Peanut butter . AFB1 5 110 (1) 118 (8) 100 108 (3) 111 (3) AFB2 5 113 (2) 113 (1) 100 110 (2) 118 (2) AFG1 5 107 (1) 112 (2) 100 105 (2) 110 (1) AFG2 5 115 (0.2) 97 (7) 100 112 (1) 115 (4) DON 50 123 (9) 116 (8) 1000 107 (2) 112 (2) FB1 50 106 (9) 104 (3) 1000 110 (4) 113 (6) FB2 50 112 (3) 121 (4) 1000 113 (3) 119 (4) FB3 50 119 (7) 112 (4) 1000 111 (2) 110 (2) OTA 50 103 (4) 116 (5) 1000 104 (5) 103 (2) T-2 50 116 (2) 114 (2) 1000 113 (3) 114 (3) HT-2 50 109 (1) 111 (2) 1000 111 (2) 112 (6) ZON 50 109 (5) 114 (3) 1000 110 (2) 112 (3) Open in new tab Table 2. Average recoveries (relative standard deviation, RSD, %; n = 4) in spiked corn and peanut butter Mycotoxins . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . Corn . Peanut butter . Corn . Peanut butter . AFB1 5 110 (1) 118 (8) 100 108 (3) 111 (3) AFB2 5 113 (2) 113 (1) 100 110 (2) 118 (2) AFG1 5 107 (1) 112 (2) 100 105 (2) 110 (1) AFG2 5 115 (0.2) 97 (7) 100 112 (1) 115 (4) DON 50 123 (9) 116 (8) 1000 107 (2) 112 (2) FB1 50 106 (9) 104 (3) 1000 110 (4) 113 (6) FB2 50 112 (3) 121 (4) 1000 113 (3) 119 (4) FB3 50 119 (7) 112 (4) 1000 111 (2) 110 (2) OTA 50 103 (4) 116 (5) 1000 104 (5) 103 (2) T-2 50 116 (2) 114 (2) 1000 113 (3) 114 (3) HT-2 50 109 (1) 111 (2) 1000 111 (2) 112 (6) ZON 50 109 (5) 114 (3) 1000 110 (2) 112 (3) Mycotoxins . Spiking concn, ng/g . Recovery (RSD, %) . Spiking concn, ng/g . Recovery (RSD, %) . Corn . Peanut butter . Corn . Peanut butter . AFB1 5 110 (1) 118 (8) 100 108 (3) 111 (3) AFB2 5 113 (2) 113 (1) 100 110 (2) 118 (2) AFG1 5 107 (1) 112 (2) 100 105 (2) 110 (1) AFG2 5 115 (0.2) 97 (7) 100 112 (1) 115 (4) DON 50 123 (9) 116 (8) 1000 107 (2) 112 (2) FB1 50 106 (9) 104 (3) 1000 110 (4) 113 (6) FB2 50 112 (3) 121 (4) 1000 113 (3) 119 (4) FB3 50 119 (7) 112 (4) 1000 111 (2) 110 (2) OTA 50 103 (4) 116 (5) 1000 104 (5) 103 (2) T-2 50 116 (2) 114 (2) 1000 113 (3) 114 (3) HT-2 50 109 (1) 111 (2) 1000 111 (2) 112 (6) ZON 50 109 (5) 114 (3) 1000 110 (2) 112 (3) Open in new tab We further compared the recovery data generated using this automated workflow to the data generated using manual procedures from a previous study. In Figure 4, overlapping the two sets of data clearly shows that the automated sample preparation system provides comparable or even better consistency in these matrixes with fewer outliers (>120% recoveries or >20% RSDs) as the green data points (automated) are more tightly clustered than the blue data points (manual). One interesting observation is the recoveries generated using the automated sample preparation system seem biased high with all but two of the recovery values above 100%. Although the manual data ranged from 80 to 120%, excluding outliers. This is probably due to measurement variability associated with the syringes used to dispense IS solution. For calibration curves, we used a 1 mL syringe, whereas for recovery samples we used a 5 mL syringe. We are investigating the root cause, making sure there is no systematic bias of the system. In future studies, to further assess the performance of the system, we will compare the automated sample preparation system and manual operations side-by-side, generating paired datasets in more challenging matrixes (e.g., dry fruits and spices). Figure 4. Open in new tabDownload slide Comparison of recovery data generated using automated sample preparation workflow and manual procedures. Blue: recovery data generated using manual procedures (22). Green: recovery data generated using the automated sample preparation system. Figure 4. Open in new tabDownload slide Comparison of recovery data generated using automated sample preparation workflow and manual procedures. Blue: recovery data generated using manual procedures (22). Green: recovery data generated using the automated sample preparation system. Short-term performance of the automated system was evaluated by analyzing a FAPAS QC sample in nine batches (eight replicates in each batch). The QC sample (corn) contains aflatoxin B1, deoxynivalenol, fumonisin B1, and B2, HT-2 and T-2 toxin, ochratoxin A, and zearalenone so it was chosen as the testing material for multi-mycotoxin analysis. The measured values of these mycotoxins were compared to assigned values and Z score (-2 ≤Z  ≤2) concentration ranges of the QC sample. Table 3 summarizes the average, relative standard derivation (RSD) of each mycotoxin in each batch, and grand mean. Within- and between-batch variabilities of each mycotoxin were also calculated following an ISO protocol (23). In general, >90% (65/72) of the batch averages are between 80 and 120% of the assigned values with RSDs <15%. The averages of fumonisin B1 in batches 6, 7, 8, and 9 and those of zearalenone in batches 7, 8 and 9 are between 120 and 130% of the assigned values. No batch average fell outside of the concentration range corresponding to -2 <Z <2. For the eight measured mycotoxins, within- and between-batch variabilities ranged from 5% (deoxynivalenol) to 9% (fumonisin B1) and 7% (deoxynivalenol) to 11% (ochratoxin A), suggesting consistent method performance of the multi-mycotoxin analysis using the automated sample preparation system and LC-MS. Table 3. Average (ng/g), relative standard deviation (RSD, %), grand mean, repeatability and reproducibility of nine batches in a FABAS QC sample (corn) Batch number . AFB1 . DON . FB1 . FB2 . OTA . T-2 . HT-2 . ZON . 1 Avg. (RSD, %), n = 8 2.83 (4) 629 (4) 530 (6) 107 (7) 1.96 (13) 95 (3) 118 (3) 144 (6) 2 Avg. (RSD, %), n = 8 2.81 (5) 637 (4) 504 (7) 104 (8) 1.76 (9) 96 (5) 116 (4) 149 95) 3 Avg. (RSD, %), n = 8 2.73 (5) 618 (5) 521 (6) 93 (8) 1.74 (7) 90 (6) 113 (10) 143 (7) 4 Avg. (RSD, %), n = 8 2.86 (6) 692 (5) 592 (10) 104 (8) 1.80 (12) 108 (5) 118 (9) 167 (7) 5 Avg. (RSD, %), n = 8 2.83 (5) 656 (5) 469 (6) 111 (7) 1.66 (8) 101 (6) 120 (10) 161 (7) 6 Avg. (RSD, %), n = 8 2.78 (5) 657 (5) 567 (8) 102 (8) 1.75 (8) 110 (8) 121 (8) 167 (7) 7 Avg. (RSD, %), n = 8 2.62 (7) 655 (7) 577 (15) 84 (13) 1.79 (9) 106 (8) 125 (9) 170 (9) 8 Avg. (RSD, %), n = 8 2.97 (9) 706 (6) 582 (8) 103 (9) 2.03 (7) 113 (8) 132 (8) 173 (9) 9 Avg. (RSD, %), n = 8 2.90 (9) 675 (6) 580 (10) 105 (9) 2.02 (9) 104 (8) 118 (8) 175 (10) Grand mean (RSD, %), n = 9 2.82 (4) 658 (5) 542 (8) 101 (9) 1.85 (8) 102 (8) 120 (5) 160 (8) Within-batch variability, % 6 5 9 8 9 7 8 8 Between-batch variability, % 7 7 12 11 11 10 9 11 Assigned values 3.07 731 463 108 2.18 109 120 141 (Ranges of −2 <Z <2) (1.72–4.42) (486–976) (297–629) (61–156) (1.22–3.14) (61–157) (67–173) (80–201) Batch number . AFB1 . DON . FB1 . FB2 . OTA . T-2 . HT-2 . ZON . 1 Avg. (RSD, %), n = 8 2.83 (4) 629 (4) 530 (6) 107 (7) 1.96 (13) 95 (3) 118 (3) 144 (6) 2 Avg. (RSD, %), n = 8 2.81 (5) 637 (4) 504 (7) 104 (8) 1.76 (9) 96 (5) 116 (4) 149 95) 3 Avg. (RSD, %), n = 8 2.73 (5) 618 (5) 521 (6) 93 (8) 1.74 (7) 90 (6) 113 (10) 143 (7) 4 Avg. (RSD, %), n = 8 2.86 (6) 692 (5) 592 (10) 104 (8) 1.80 (12) 108 (5) 118 (9) 167 (7) 5 Avg. (RSD, %), n = 8 2.83 (5) 656 (5) 469 (6) 111 (7) 1.66 (8) 101 (6) 120 (10) 161 (7) 6 Avg. (RSD, %), n = 8 2.78 (5) 657 (5) 567 (8) 102 (8) 1.75 (8) 110 (8) 121 (8) 167 (7) 7 Avg. (RSD, %), n = 8 2.62 (7) 655 (7) 577 (15) 84 (13) 1.79 (9) 106 (8) 125 (9) 170 (9) 8 Avg. (RSD, %), n = 8 2.97 (9) 706 (6) 582 (8) 103 (9) 2.03 (7) 113 (8) 132 (8) 173 (9) 9 Avg. (RSD, %), n = 8 2.90 (9) 675 (6) 580 (10) 105 (9) 2.02 (9) 104 (8) 118 (8) 175 (10) Grand mean (RSD, %), n = 9 2.82 (4) 658 (5) 542 (8) 101 (9) 1.85 (8) 102 (8) 120 (5) 160 (8) Within-batch variability, % 6 5 9 8 9 7 8 8 Between-batch variability, % 7 7 12 11 11 10 9 11 Assigned values 3.07 731 463 108 2.18 109 120 141 (Ranges of −2 <Z <2) (1.72–4.42) (486–976) (297–629) (61–156) (1.22–3.14) (61–157) (67–173) (80–201) Open in new tab Table 3. Average (ng/g), relative standard deviation (RSD, %), grand mean, repeatability and reproducibility of nine batches in a FABAS QC sample (corn) Batch number . AFB1 . DON . FB1 . FB2 . OTA . T-2 . HT-2 . ZON . 1 Avg. (RSD, %), n = 8 2.83 (4) 629 (4) 530 (6) 107 (7) 1.96 (13) 95 (3) 118 (3) 144 (6) 2 Avg. (RSD, %), n = 8 2.81 (5) 637 (4) 504 (7) 104 (8) 1.76 (9) 96 (5) 116 (4) 149 95) 3 Avg. (RSD, %), n = 8 2.73 (5) 618 (5) 521 (6) 93 (8) 1.74 (7) 90 (6) 113 (10) 143 (7) 4 Avg. (RSD, %), n = 8 2.86 (6) 692 (5) 592 (10) 104 (8) 1.80 (12) 108 (5) 118 (9) 167 (7) 5 Avg. (RSD, %), n = 8 2.83 (5) 656 (5) 469 (6) 111 (7) 1.66 (8) 101 (6) 120 (10) 161 (7) 6 Avg. (RSD, %), n = 8 2.78 (5) 657 (5) 567 (8) 102 (8) 1.75 (8) 110 (8) 121 (8) 167 (7) 7 Avg. (RSD, %), n = 8 2.62 (7) 655 (7) 577 (15) 84 (13) 1.79 (9) 106 (8) 125 (9) 170 (9) 8 Avg. (RSD, %), n = 8 2.97 (9) 706 (6) 582 (8) 103 (9) 2.03 (7) 113 (8) 132 (8) 173 (9) 9 Avg. (RSD, %), n = 8 2.90 (9) 675 (6) 580 (10) 105 (9) 2.02 (9) 104 (8) 118 (8) 175 (10) Grand mean (RSD, %), n = 9 2.82 (4) 658 (5) 542 (8) 101 (9) 1.85 (8) 102 (8) 120 (5) 160 (8) Within-batch variability, % 6 5 9 8 9 7 8 8 Between-batch variability, % 7 7 12 11 11 10 9 11 Assigned values 3.07 731 463 108 2.18 109 120 141 (Ranges of −2 <Z <2) (1.72–4.42) (486–976) (297–629) (61–156) (1.22–3.14) (61–157) (67–173) (80–201) Batch number . AFB1 . DON . FB1 . FB2 . OTA . T-2 . HT-2 . ZON . 1 Avg. (RSD, %), n = 8 2.83 (4) 629 (4) 530 (6) 107 (7) 1.96 (13) 95 (3) 118 (3) 144 (6) 2 Avg. (RSD, %), n = 8 2.81 (5) 637 (4) 504 (7) 104 (8) 1.76 (9) 96 (5) 116 (4) 149 95) 3 Avg. (RSD, %), n = 8 2.73 (5) 618 (5) 521 (6) 93 (8) 1.74 (7) 90 (6) 113 (10) 143 (7) 4 Avg. (RSD, %), n = 8 2.86 (6) 692 (5) 592 (10) 104 (8) 1.80 (12) 108 (5) 118 (9) 167 (7) 5 Avg. (RSD, %), n = 8 2.83 (5) 656 (5) 469 (6) 111 (7) 1.66 (8) 101 (6) 120 (10) 161 (7) 6 Avg. (RSD, %), n = 8 2.78 (5) 657 (5) 567 (8) 102 (8) 1.75 (8) 110 (8) 121 (8) 167 (7) 7 Avg. (RSD, %), n = 8 2.62 (7) 655 (7) 577 (15) 84 (13) 1.79 (9) 106 (8) 125 (9) 170 (9) 8 Avg. (RSD, %), n = 8 2.97 (9) 706 (6) 582 (8) 103 (9) 2.03 (7) 113 (8) 132 (8) 173 (9) 9 Avg. (RSD, %), n = 8 2.90 (9) 675 (6) 580 (10) 105 (9) 2.02 (9) 104 (8) 118 (8) 175 (10) Grand mean (RSD, %), n = 9 2.82 (4) 658 (5) 542 (8) 101 (9) 1.85 (8) 102 (8) 120 (5) 160 (8) Within-batch variability, % 6 5 9 8 9 7 8 8 Between-batch variability, % 7 7 12 11 11 10 9 11 Assigned values 3.07 731 463 108 2.18 109 120 141 (Ranges of −2 <Z <2) (1.72–4.42) (486–976) (297–629) (61–156) (1.22–3.14) (61–157) (67–173) (80–201) Open in new tab To further evaluate the performance of the sample preparation system in terms of accuracy, we analyzed five certified reference materials prepared using the system and the results are summarized in Table 4. In NIST 2387 (peanut butter), the measured concentrations of aflatoxin B1 and B2 were compared to the certified values. The difference between each measured concentration and the corresponding certified value was within ±20% of the certified value. The average measured concentrations and SDs of aflatoxin B1 and B2 were 4.7 ± 0.2 and 0.9 ± 0.1 ng/g, respectively. Both values were within ±20% of the certified values, 4.2 ± 0.9 and 0.7 ± 0.3 ng/g. In BE375 (feed), the measured concentrations of aflatoxin B1 and B2 were 2.9 ± 0.1 and 0.26 ± 0.01 ng/g. Compared to certified values, 2.6 ± 0.4 and 0.20 ± 0.04 ng/g, these results demonstrate sufficient accuracy of the analysis. In BE376, the measured concentrations of aflatoxin B1 and G1 were 13.3 ± 0.6 and 5.8 ± 0.4 ng/g, respectively, which were comparable to the certified values, 12.9 ± 1.8 and 5.2 ± 0.8 ng/g. The difference between the measured concentration (0.94 ± 0.05 ng/g) of aflatoxin B2 and the certified value, 0.68 ± 0.10 ng/g, is around 40% of the certified value but it is worth mentioning that at concentrations <1 ng/g, such difference is not uncommon (24). In BE283 and 284 (milk powder), the measured concentrations (0.116 and 0.45 ng/g) of aflatoxin M1 are in good agreement with the certified values (0.111 and 0.44 ng/g). Table 4. Analysis of certified reference materials (Avg. ± SD, n = 4) Certified materials, ng/g . AFB1 . AFB2 . AFG1 . AFG2 . Total . Certified materials, ng/g . AFM1 . NIST 2387 measured 4.7 ± 0.2 0.9 ± 0.1 NDb NDb 5.6 ± 0.2 BE283 measured 0.116 ± 0.006 Reference value 4.2 ± 0.9 0.7 ± 0.3 NDb NDb 5.0 ± 0.5 Reference value 0.111 ± 0.018 BE375 measured 2.9 ± 0.1 0.26 ± 0.01 NDb NDb BE284 measured 0.45 ± 0.01 Reference value 4.6 ± 0.4 0.20 ± 0.04 0.4 ± 0.1 Not reported —a Reference material 0.44 ± 0.06 BE376 measured 13.3 ± 0.6 0.94 ± 0.05 5.8 ± 0.4 NDb —a —a —a Reference value 12.9 ± 1.8 0.68 ± 0.10 5.2 ± 0.8 Not reported 4.7 ± 0.6 —a —a Certified materials, ng/g . AFB1 . AFB2 . AFG1 . AFG2 . Total . Certified materials, ng/g . AFM1 . NIST 2387 measured 4.7 ± 0.2 0.9 ± 0.1 NDb NDb 5.6 ± 0.2 BE283 measured 0.116 ± 0.006 Reference value 4.2 ± 0.9 0.7 ± 0.3 NDb NDb 5.0 ± 0.5 Reference value 0.111 ± 0.018 BE375 measured 2.9 ± 0.1 0.26 ± 0.01 NDb NDb BE284 measured 0.45 ± 0.01 Reference value 4.6 ± 0.4 0.20 ± 0.04 0.4 ± 0.1 Not reported —a Reference material 0.44 ± 0.06 BE376 measured 13.3 ± 0.6 0.94 ± 0.05 5.8 ± 0.4 NDb —a —a —a Reference value 12.9 ± 1.8 0.68 ± 0.10 5.2 ± 0.8 Not reported 4.7 ± 0.6 —a —a a No data. b ND = Not detected. Open in new tab Table 4. Analysis of certified reference materials (Avg. ± SD, n = 4) Certified materials, ng/g . AFB1 . AFB2 . AFG1 . AFG2 . Total . Certified materials, ng/g . AFM1 . NIST 2387 measured 4.7 ± 0.2 0.9 ± 0.1 NDb NDb 5.6 ± 0.2 BE283 measured 0.116 ± 0.006 Reference value 4.2 ± 0.9 0.7 ± 0.3 NDb NDb 5.0 ± 0.5 Reference value 0.111 ± 0.018 BE375 measured 2.9 ± 0.1 0.26 ± 0.01 NDb NDb BE284 measured 0.45 ± 0.01 Reference value 4.6 ± 0.4 0.20 ± 0.04 0.4 ± 0.1 Not reported —a Reference material 0.44 ± 0.06 BE376 measured 13.3 ± 0.6 0.94 ± 0.05 5.8 ± 0.4 NDb —a —a —a Reference value 12.9 ± 1.8 0.68 ± 0.10 5.2 ± 0.8 Not reported 4.7 ± 0.6 —a —a Certified materials, ng/g . AFB1 . AFB2 . AFG1 . AFG2 . Total . Certified materials, ng/g . AFM1 . NIST 2387 measured 4.7 ± 0.2 0.9 ± 0.1 NDb NDb 5.6 ± 0.2 BE283 measured 0.116 ± 0.006 Reference value 4.2 ± 0.9 0.7 ± 0.3 NDb NDb 5.0 ± 0.5 Reference value 0.111 ± 0.018 BE375 measured 2.9 ± 0.1 0.26 ± 0.01 NDb NDb BE284 measured 0.45 ± 0.01 Reference value 4.6 ± 0.4 0.20 ± 0.04 0.4 ± 0.1 Not reported —a Reference material 0.44 ± 0.06 BE376 measured 13.3 ± 0.6 0.94 ± 0.05 5.8 ± 0.4 NDb —a —a —a Reference value 12.9 ± 1.8 0.68 ± 0.10 5.2 ± 0.8 Not reported 4.7 ± 0.6 —a —a a No data. b ND = Not detected. Open in new tab Table 5. Analysis of NIST corn samples Certified material, ng/g . NIST corn-1 measured . Reference value . NIST corn-2 measured . Reference value . AFB1 7.80 ± 2.20 7.50 ± 1.70 NDb —a AFB2 1.52 ± 0.23 1.43 ± 0.34 NDb —a AFG1 0.84 ± 0.10 0.98 ± 0.19 NDb —a AFG2 NDb 0.87 ± 0.24 NDb —a Total AFs 10.2 ± 2.4 10.2 ± 2.9 NDb —a DON 444 ± 23 467 ± 67 118 ± 14 142 ± 36 FB1 730 ± 93 805 ± 190 8.2 ± 0.9 10.4 ± 3.9 FB2 208 ± 26 217 ± 30 NDb —a FB3 84 ± 13 99 ± 8 NDb —a Total FBs 1122 ± 128 1150 ± 170 NDb —a OTA 10.0 ± 2.3 9.4 ± 1.2 NDb —a T-2 14.1 ± 4.5 18.4 ± 4.2 NDb —a HT-2 33.2 ± 1.8 38.2 ± 6.0 NDb —a ZON 58 ± 6 61 ± 36 NDb —a Certified material, ng/g . NIST corn-1 measured . Reference value . NIST corn-2 measured . Reference value . AFB1 7.80 ± 2.20 7.50 ± 1.70 NDb —a AFB2 1.52 ± 0.23 1.43 ± 0.34 NDb —a AFG1 0.84 ± 0.10 0.98 ± 0.19 NDb —a AFG2 NDb 0.87 ± 0.24 NDb —a Total AFs 10.2 ± 2.4 10.2 ± 2.9 NDb —a DON 444 ± 23 467 ± 67 118 ± 14 142 ± 36 FB1 730 ± 93 805 ± 190 8.2 ± 0.9 10.4 ± 3.9 FB2 208 ± 26 217 ± 30 NDb —a FB3 84 ± 13 99 ± 8 NDb —a Total FBs 1122 ± 128 1150 ± 170 NDb —a OTA 10.0 ± 2.3 9.4 ± 1.2 NDb —a T-2 14.1 ± 4.5 18.4 ± 4.2 NDb —a HT-2 33.2 ± 1.8 38.2 ± 6.0 NDb —a ZON 58 ± 6 61 ± 36 NDb —a a No data. b ND = Not detected. Open in new tab Table 5. Analysis of NIST corn samples Certified material, ng/g . NIST corn-1 measured . Reference value . NIST corn-2 measured . Reference value . AFB1 7.80 ± 2.20 7.50 ± 1.70 NDb —a AFB2 1.52 ± 0.23 1.43 ± 0.34 NDb —a AFG1 0.84 ± 0.10 0.98 ± 0.19 NDb —a AFG2 NDb 0.87 ± 0.24 NDb —a Total AFs 10.2 ± 2.4 10.2 ± 2.9 NDb —a DON 444 ± 23 467 ± 67 118 ± 14 142 ± 36 FB1 730 ± 93 805 ± 190 8.2 ± 0.9 10.4 ± 3.9 FB2 208 ± 26 217 ± 30 NDb —a FB3 84 ± 13 99 ± 8 NDb —a Total FBs 1122 ± 128 1150 ± 170 NDb —a OTA 10.0 ± 2.3 9.4 ± 1.2 NDb —a T-2 14.1 ± 4.5 18.4 ± 4.2 NDb —a HT-2 33.2 ± 1.8 38.2 ± 6.0 NDb —a ZON 58 ± 6 61 ± 36 NDb —a Certified material, ng/g . NIST corn-1 measured . Reference value . NIST corn-2 measured . Reference value . AFB1 7.80 ± 2.20 7.50 ± 1.70 NDb —a AFB2 1.52 ± 0.23 1.43 ± 0.34 NDb —a AFG1 0.84 ± 0.10 0.98 ± 0.19 NDb —a AFG2 NDb 0.87 ± 0.24 NDb —a Total AFs 10.2 ± 2.4 10.2 ± 2.9 NDb —a DON 444 ± 23 467 ± 67 118 ± 14 142 ± 36 FB1 730 ± 93 805 ± 190 8.2 ± 0.9 10.4 ± 3.9 FB2 208 ± 26 217 ± 30 NDb —a FB3 84 ± 13 99 ± 8 NDb —a Total FBs 1122 ± 128 1150 ± 170 NDb —a OTA 10.0 ± 2.3 9.4 ± 1.2 NDb —a T-2 14.1 ± 4.5 18.4 ± 4.2 NDb —a HT-2 33.2 ± 1.8 38.2 ± 6.0 NDb —a ZON 58 ± 6 61 ± 36 NDb —a a No data. b ND = Not detected. Open in new tab Additionally, we analyzed two corn samples (corn-1 and corn-2) provide by the National Institute of Standards and Technology (NIST). The corn samples have been certified by the NIST as a standard reference material (25). Results are summarized in Table 5. Corn-1 contains aflatoxin B1, B2, G1 and G2; fumonisin B1, B2 and B3; deoxynivalenol, ochratoxin A, HT-2/T-2 toxin and zearalenone at various concentrations. Corn-2 only contains deoxynivalenol and fumonisin B1 at 142 and 10.4 ng/g, respectively. Except for aflatoxin G2 in Corn-1, our results are consistent with the reference values provided by NIST. The difference between each measured concentration and the corresponding reference value is with ±20% of the reference value. In summary, our results suggest that automation is a promising alternative to manual operation and has potential for food analysis. Furthermore, automation could assist laboratories to achieve a high degree of traceability and standardization for routine mycotoxin analysis. Future Studies and Limitations The above workflow successfully replaced key manual steps in mycotoxin sample preparation and in principle would help minimize human errors such as inconsistent pipetting/weighing, improve lab safety, and relieve operators from repeating laborious operations. The implementation of automation also could provide a better QC system as operations are performed in a highly standardized manner with digitized traceability. Although the current system can be programmed to prepare mycotoxin samples, the degree of automation could be further increased if an analytical instrument is coupled with the automated sample preparation system. This would enable real-time sample analysis, without waiting for the completion of the whole batch. Due to limited laboratory space, the LC-MS instrument was not connected with Chemspeed Swing XL. The prepared samples were manually transferred to the autosampler on the LC-MS. Had the instrument been integrated with the automated sample preparation system, the prepared samples could have been directly injected into the LC-MS using the online injection valve or the samples could have been loaded on the autosampler. Either way the sample preparation and instrumental measurement could be streamlined in a more automated manner. With an online analytical instrument, the system also could be used to automate method development studies. Based on the analytical results (e.g., recoveries) generated using automated data processing from the instrument, the automated system could be programed to modify sample preparation parameters (e.g., testing different extraction solvents, extraction times, etc.) accordingly at run-time, saving researchers time manually testing and comparing various parameters. These potential applications are worth exploring in the future. A common expectation of automated sample preparation is that the process is fully unattended. However, it is no easy task to ensure that sufficient starting materials and consumables for a large quantity of samples could be loaded in an automated way. For the current system, human operators need to estimate the usage of consumables and load them on the system prior to sample preparation. If sample preparation needs to be scaled up for continuous operations, having storage units attached to the system and loading consumables with the robotic arms could be a possible approach to address this issue in the future. Another notable limitation of automated systems is the lack of self-training and self-troubleshooting capability. This is highlighted by the need for comprehensive sensory monitoring systems and intelligent data-processing algorithms. For example, the current system does not have any vision and image analysis features, so once an application is initiated, tools and robotic arms could not adjust its movement to its working environment (e.g., avoid unexpected objects or collisions). Features such as real-time obstacle/collision-avoidance would further enhance the performance of the system. With careful calibration, individual tools such as the screwcapper show amazing dexterity compared to human operators; however, the calibration process is mainly performed by experienced engineers or researchers and generated parameters are saved via hardcoding, a practice which seems to contradict to the purpose of automation. Some tools (e.g., the gripper) need to be calibrated for racks and sample vessels with different dimensions or sizes via trial and error. Very often the process is time consuming as current tools could not flexibly adjust its operating parameters according to shapes, sizes or weights of objects during run time. To some extent, using highly standardized consumables (e.g., vials, filters and solvent containers) could simplify the process; however, this might limit the selection of consumables. Obviously, it is important to develop self-training/configuration features that could enhance versatility of the system, making it more suited to automated food analysis. Although these challenges and limitations remain in the development of hardware and software for automated systems, automated sample preparation for food analysis has become self-evident as such an approach could achieve levels of throughput, reproducibility and robustness unmatched by manual operations. It is important to dedicate continuous efforts to identify specific application domains where benefits can be demonstrated in both the short and long term. Acknowledgments The author would like to thank Christine Parker (US Food and Drug Administration, Center for Food Safety and Applied Nutrition, Office of Regulatory Science) and Jaroslav Padevet (Chemspeed Technologies AG) for their support. JAOAC does not publish color figures in the print version. Color images are published online only. References 1 Schneider G. ( 2018 ) Nat. Rev. Drug Discov. 17 , 97 – 113 Crossref Search ADS PubMed 2 Armbruster D.A. , Overcash D.R., Reyes J. ( 2014 ) Clin. Biochem. Rev. 35 , 143 – 153 PubMed 3 Rahmani A. , Jinap S., Soleimany F. ( 2009 ) Compr. Rev. Food Sci. Food Saf . 8 , 202 – 251 Crossref Search ADS 4 Trucksess M.W. , Stack M.E., Nesheim S., Page S.W., Albert R.H., Hansen T.J., Donahue K.F. ( 1991 ) J. Assoc. Off. Anal. Chem. 74 , 81 – 88 PubMed 5 Weaver C.M. , Trucksess M.W. ( 2010 ) J. AOAC Int. 93 , 184 – 189 Crossref Search ADS PubMed 6 Bao L. , Liang C., Trucksess M.W., Xu Y., Lv N., Wu Z., Jing P., Fry F.S. ( 2013 ) J. AOAC Int. 96 , 1017 – 1018 Crossref Search ADS PubMed 7 Sulyok M. , Berthiller F., Krska R., Schuhmacher R. ( 2006 ) Rapid Commun. Mass Spectrom. 20 , 2649 – 2659 Crossref Search ADS PubMed 8 Häubl G. , Berthiller F., Krska R., Schuhmacher R. ( 2006 ) Anal. Bioanal. Chem. 384 , 692 – 696 Crossref Search ADS PubMed 9 Chan D. , MacDonald S.J., Boughtflower V., Brereton P. ( 2004 ) J. Chromatogr. A 1059 , 13 – 16 Crossref Search ADS PubMed 10 Sharman M. , Gilbert J. ( 1991 ) J. Chromatogr . 543 , 220 – 225 Crossref Search ADS PubMed 11 Carman A.S. Jr, Kuan S.S., Ware G.M., Umrigar P.P., Miller K.V., Guerrero H.G. ( 1996 ) J. AOAC Int. 79 , 456 – 464 . Crossref Search ADS PubMed 12 Rhemrev R. , Pazdanska M., Marley E., Biselli S., Staiger S. ( 2015 ) J. AOAC Int. 98 , 1585 – 1590 Crossref Search ADS PubMed 13 Oulkar D. , Goon A., Dhanshetty M., Khan Z., Satav S., Banerjee K. ( 2018 ) J. Environ. Sci. Health B 53 , 255 – 260 Crossref Search ADS PubMed 14 Dhanshetty M. , Banerjee K. ( 2019 ) J. AOAC Int. 102 , 1666 – 1672 Crossref Search ADS PubMed 15 Malachová A. , Sulyok M., Beltrán E., Berthiller F., Krska R. ( 2014 ) J. Chromatogr. A . 1362 , 145 – 156 Crossref Search ADS PubMed 16 Lattanzio V.M.T. , von Holst C., Lippolis V., De Girolamo A., Logrieco A.F., Mol H.G.J., Pascale M. ( 2019 ) Toxins 11 , 129 – 146 Crossref Search ADS 17 Mol H.G.J. , Plaza-Bolaños P., Zomer P., de Rijk T.C., Stolker A.A.M., Mulder P.P.J. ( 2008 ) Anal. Chem. 80 , 9450 – 9459 Crossref Search ADS PubMed 18 Berthiller F. , Brera C., Iha M.H., Krska R., Lattanzio V.M.T., MacDonald S., Malone R.J., Maragos C., Solfrizzo M., Stranska-Zachariasova M., Stroka J., Tittlemier S.A. ( 2017 ) World Mycotox. J . 10 , 5 – 29 Crossref Search ADS 19 Ates E. , Mittendorf K., Stroka J., Senyuva H. ( 2013 ) Food Addit. Contam. Part A Chem. Anal. Control. Expo. Risk Assess. 30 , 156 – 165 Crossref Search ADS PubMed 20 Lhotská I. , Šatínský D., Havlíková L., Solich P. ( 2016 ) Anal. Bioanal. Chem. 408 , 3319 – 3329 Crossref Search ADS PubMed 21 Campone L. , Piccinelli A.L., Celano R., Russo M., Valdés A., Ibáñez C., Rastrelli L. ( 2015 ) Anal. Bioanal. Chem. 407 , 2899 – 2911 Crossref Search ADS PubMed 22 Zhang K. , Schaab M.R., Southwood G., Tor E.R., Aston L.S., Song W.L., Eitzer B., Majumdar S., Lapainis T., Mai H., Tran K., El-Demerdash A., Vega V., Cai Y., Wong J.W., Krynitsky A.J., Begley T.H. ( 2017 ) J. Agric. Food Chem. 65 , 7138 – 7152 Crossref Search ADS PubMed 23 ISO 5725-2:1994 Accuracy (trueness and precision) of measurement methods and results–Part 2: Basic method for the determination of repeatability and reproducibility of a standard measurement method. International Organization for Standardization. https://www.iso.org/standard/69419.html. 24 Horwitz W. , Albert R. ( 2006 ) J. AOAC Int. 89 , 1095 – 1109 Crossref Search ADS PubMed 25 Phillips M.M. , Lopez T., Ness J., Zhang K. ( 2019 ) J. AOAC Int. 102, 1642–1650 AOAC INTERNATIONAL 2020. This work is written by a US Government employee and is in the public domain in the US. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Evaluation of Automated Sample Preparation for Mycotoxin Analysis in Foods
JF - Journal of AOAC International
DO - 10.1093/jaoacint/qsz044
DA - 2020-07-01
UR - https://www.deepdyve.com/lp/oxford-university-press/evaluation-of-automated-sample-preparation-for-mycotoxin-analysis-in-iYXINVtuCo
SP - 1052
EP - 1059
VL - 103
IS - 4
DP - DeepDyve
ER -