Optimization of MS and MS/MS parameters
The optimization of the electrospray ionization interface and the tandem-mass-spectrometric parameters were achieved by infusions of reference material solutions (individual analytes in MeOH, 5 μg/mL) using a syringe pump. To mimic actual infusion conditions, a flow rate of 3–10 μL/min tune solution was combined with an LC flow of 0.3 mL/min using a T-piece before entering the ESI source. During the infusion optimization, the ratio of eluent A and B was selected according to the polarity of a certain analyte (i.e., highly polar analytes were optimized at between 10 and 40% eluent B and lipophilic analytes between 60 and 90% eluent B). MS and MS/MS parameters were optimized in both polarities using the instruments’ tuning software. First, the most appropriate precursor ions were selected by maximum intensity in full scan mode. Interestingly, negative ionization led to higher signals for all 17 target analytes. The eight most abundant product ions and their corresponding collision energies were determined for each analyte by comparing signal intensities with changing parameters. To achieve highest sensitivity and selectivity in the three food matrices, S/N ratios for all eight transitions were evaluated by independent LC-MS injections of spiked blank matrix samples. The two ions with the highest relative S/N value, which were fortunately identical in all matrices, were selected as quantifier and qualifier ion (Table 1). ESI parameters (spray voltage, vaporizer temperature, sheath gas, ion sweep gas, auxiliary gas, and capillary temperature) were examined manually and set to provide the best overall performance. It was suggested that charging effects in an ESI interface may result in signal suppression of certain challenging analytes including TeA . Hence, the ionization polarity was switched from negative to positive mode for the final 2.5 min of each run.
Development of the chromatographic method
An important aim of the developed chromatographic method was to baseline-separate the isomeric analytes of interest in reasonable run times resulting in reproducible, sharp, and symmetrical signals. The selected reversed phase column was described before to be favorable for the separation of the five Alternaria toxins, namely AOH, AME, ALT, TeA, and TEN . By optimization of eluents, temperature, and the multi-step gradient, we were able to separate our 17 target analytes of highly diverse polarity. A basic eluent system (eluent A, 5 mM NH4Ac in water, pH 8.7) was crucial for a symmetric peak shape of the polar TeA, which is typically measured after derivatization or alternatively exhibits very broad peaks and peak tailing in acidic eluents. Also, the AOH-3-S and AME-3-S showed peak tailing with acidic eluents, which could be resolved using the basic conditions. Importantly, also, the most lipophilic toxins (AME, TEN, and the perylen quinones) showed a favorable behavior with very narrow peak shapes allowing for enhanced signal intensities. Due to the optimized multi-step gradient, it was possible to baseline-separate the isomers ALT and isoALT, as well as the glucosides of AOH for the first time (Fig. 2). Purging of the column after finishing measurement sequences at 95% MeOH ensured an acceptable column lifetime despite the applied basic eluents. The flow rate and column temperature were optimized to yield the overall best signal-to-noise ratios and shortest run time. In general, retention times of the target analytes were stable and reproducible (see Table 1). Only the retention times of TeA, AA-III, and AOH and its modified forms were prone to minor pH changes observed after preparing fresh eluent A. Maximum shifts have been observed for TeA (0.1 min), AOH (0.3 min), and AA-III (0.3 min). However, since signals derived from reference standards and unknown samples behave equally, this was not an issue. No relevant carry-over between injections was observed. However, the absence was constantly verified by monitoring solvent and matrix blank samples. The injection needle of the LC autosampler was washed with 100 μL isopropanol/water (75/25, v/v) before and after each sample injection.
Optimization of the sample preparation protocol
The sample preparation protocol was intended on one hand to be as generic as possible to prevent the discrimination of any of the chemically diverse analytes and, on the other hand, to be time- and cost-effective. Hence, sample extracts were centrifuged and diluted by a factor of two resulting in an overall dilution of 1:10 (w/v). Since the LC-MS/MS method was thoroughly optimized and allows for highly sensitive and selective quantitation, no further derivatization  or solid-phase extraction [5, 27] steps were required. This makes the method attractive for large-scale food-monitoring programs as suggested by the EFSA in their recent scientific report . Due to fine particles suspended in the wheat flour extracts, an additional filtration prior to analysis was required. This ensured reproducible pressure conditions of the LC system even after a high number of injections. The addition of n-hexane to the sunflower seed oil samples simplified their handling and led to enhanced extraction efficiency. This was not necessary for tomato sauce and wheat flower samples.
In-house validation was performed based on the requirements defined by the Commission Decision (EC) No. 657/2002  and the Eurachem Laboratory Guide for the validation of analytical methods . Three food commodities with diverse chemical composition and frequently contaminated by Alternaria mycotoxins [1, 3, 46,47,48] were chosen for comprehensive evaluation of the developed method and included tomato sauce (representative for aqueous matrices), sunflower seed oil (non-polar and fatty matrices), and wheat flour (carbohydrate-based matrices). The following parameters were successfully validated: selectivity linearity, matrix effects, recovery, sensitivity, repeatability, and intermediate precision. Due to a lack of certified reference materials of Alternaria toxins, the validation was based on the fortification of blank matrix samples at three concentration levels. These concentrations were based on the preliminary calculation of LOQ values (Table 2).
Selectivity was verified by the analysis of representative blank samples for each matrix. Signals from fortified blank samples and unknown samples collected from the Austrian retail marked were evaluated and no relevant co-eluting interfering signals were detected. For all analytes, suitable blank matrix samples were identified with the exception of TEN in sunflower seed oil in which all samples contained very low concentrations (see Table 3). TEN was clearly identified in these oils, seeing that this analyte proved very stable retention times (in all matrixes), reproducible narrow peak shapes, and a low background noise. The sample contaminated with the lowest amount of TEN (< LOQ) was used in the spiking experiments. The regression coefficients (R2) between 0.97 and 0.99 confirmed linearity of both solvent- and matrix-matched calibration curves over at least 3 orders of magnitude (5–7 concentration levels, Table 2).
Matrix effects varied depending on the type of matrix as reported in Table 2. Signal suppression or enhancement (SSE) for AOH, ALT, isoALT, TeA, TEN, AOH-3-S, AME-3-Glc, ATX-I, ATX-II, and ALP was between 80 and 120%. AME and its sulfate showed a signal enhancement in wheat flour of 124 and 156%, respectively. Sulfate conjugates of other mycotoxins have been described to be prone to signal enhancement before . Signals of AOH-3-Glc and AOH-9-Glc were suppressed in tomato sauces (51 and 79%) and wheat flour (68 and 72%), but enhanced in sunflower seed oil (118 and 114%). Signal enhancement in wheat flour was also found for STTX-III and AA-III. ALS showed to be susceptible to matrix effects with 144% in tomato sauce and 14 and 56% in sunflower seed oil and wheat flour, respectively, for its parent ion [M-H]− at m/z 289. Previous methods did not determine the deprotonated parent ion, but an ion at m/z 287, which may represent a ring-closing reaction product [5, 50]. This ion of unknown structure shows lower matrix effects; however, it also yielded lower signal-to-noise ratios and thus significantly higher LOD values. Therefore, we selected the [M-H]− ion for the final method.
The relative recovery (R
, extraction efficiency) of most analytes ranged between 70 and 110% in all three matrices, a range comparable with other methods published in literature [5, 27, 30]. Best results were obtained for sunflower seed oil, where the values ranged between 74 and 100% for all three spiking levels. The extraction proved to be very suitable for tomato sauce as well, only the recovery of STTX-III was below the target value. Wheat flour was a comparatively more challenging matrix. The more polar analytes AA-III, AOH-3-S, and AME-3-S exhibited recoveries between 55 and 64%, while it was even lower for ALS. Recoveries for AOH-3-S and AME-3-S in cereal-based food items published by Walravens et al.  were close to 100%, but no recoveries for AA-III and ALS in similar matrices were reported so far. Apparently, molecules holding deprotonable sulfate or carboxyl groups are less effectively extracted from wheat flour with the utilized extraction procedure. Recoveries of STTX-III, which are the first reported for any food matrix, were 94% in sunflower seed oil, but only 28–51% in tomato sauce and wheat flour. Due to the limited amounts available of the reference standard, no further investigations could be performed. Consequently, accurate quantitation of this analyte in two matrices (tomato sauces and wheat flour) is not possible but the analyte was kept in the final method for semi-quantitative assessment. Since this analyte was never determined in any food commodity before, to the best of our knowledge, it may enable first indications of the presence of this potentially potent toxin holding an epoxide group [51, 52].
The limits of detection (LODs) of the presented method were between 0.03 ng/g (AME) and 7 ng/g (ALS), whereas the limits of quantitation (LOQ) were between 0.06 ng/g (AME) and 19 ng/g (ALS). Key toxins including AOH and ATX-II can be detected down to 1 ng/g, in the case of AME and TEN even down to 0.1 and 0.5 ng/g, respectively. For TeA, which was indicated as a challenging analyte before [5, 27], but frequently occurs at higher concentrations in food stuff, an LOD of 6 ng/g in tomato sauce, 4 ng/g in sunflower seed oil, and 7 ng/g in wheat flour was achieved. Moreover, the modified mycotoxins (AOH-3-Glc, AOH-9-Glc, AOH-3-S, AME-3-Glc, and AME-3-S) can be detected as low as 0.05–6 ng/g. The repeatability (intraday precision, RSDr) and intermediate precision (interday precision, RSDR) proved to be satisfying for nearly all analytes and matrices. Even though the presented method’s sample preparation does not include derivatization or solid-phase extraction steps, compared to earlier published studies, LOD values reached a similar or even lower range for most analytes [5, 27, 30]. For AME, TEN, ATX-I, AOH-3-S, AME-3-Glc, and AME-3-S, lower LODs were achieved compared to Zwickel et al. , Walravens et al. , while for TeA and ALT, they were slightly higher. Due to the shortage of reference materials for perylene quinones (ATX-I, ATX-II, STTX-III, ALP), modified forms of AOH and AME or toxins like iso-ALT, ALS, and AA-III, there is not much data available in literature about these compounds. In conclusion, the performed validation demonstrated that the newly developed method is fit for purpose, generating valuable occurrence data of up to 17 Alternaria toxins for the first time simultaneously.
Application to naturally contaminated food samples
To gain first insights on contamination levels and patterns of Alternaria toxins including modified forms, samples from the Austrian market (n = 28) were analyzed in a small-scale survey. Three independent measurements of tomato sauce (n = 12), sunflower seed oil (n = 7), and wheat flower (n = 9) samples were performed and average values are reported in Table 3.
Overall, nine of the 17 toxins included in the developed method have been determined in products intended for human consumption. This is intriguing given the rather small sample number analyzed in this preliminary study. In future large-scale occurrence surveys, or when analyzing visually mold-infested samples, it is likely to observe even a greater number of these emerging contaminants.
Tomato sauce was the commodity with both the highest number of detected analytes and generally the highest concentrations. This is in line with literature suggesting tomato-based products to be often contaminated by comparatively high levels of the four to six typically reported Alternaria toxins AOH, AME, TeA, TEN, ALT, and ATX-I [2, 4, 20, 27, 29, 31, 37]. Interestingly, organic products seem to be slightly more contaminated than conventionally farmed samples. However, this should not be over-interpreted due to the limited sample size but investigated in more detail in further studies.
As expected, TeA concentrations were higher than the other Alternaria toxins and reached concentrations of 300 ng/g. Compounds with genotoxic properties, AOH and AME, were found in about half of the tomato sauce samples. The concentrations determined are in a similar range as published in other recent studies [29, 31, 53]. To the best of our knowledge, the masked mycotoxin AOH-9-Glc was identified and quantified for the first time in any food matrix. Interestingly, only the C9 isomer conjugate was detected, despite the threefold lower LOD of AOH-3-Glc. This indicates that, at least in tomatoes, AOH-9-Glc is the prevalent formed metabolite of AOH. This is in line with in vitro studies where AOH-9-Glc was reported to be the major metabolite in tobacco suspension cell culture experiments after 48 h of AOH incubation . Other recently published methods also included glucosides but did not detect them in naturally contaminated samples [29, 30, 33, 53]. Furthermore, sulfate conjugates of both AOH and AME have not been described in literature as food contaminants before. We were able to detect these compounds in naturally contaminated tomato sauce samples. For confirmation purpose, selected samples contaminated at low concentrations were enriched by a factor of five and re-measured. MRM chromatograms showing quantifier and qualifier ion transitions of AOH-3-S and AME-3-S in a naturally contaminated tomato sauce sample (sample #5) are illustrated in Fig. 3. Direct comparison to a spiked blank matrix sample allowed for unambiguous identification. Surprisingly and of relevance for risk assessment, in some samples, the modified mycotoxins were present in similar concentrations as their parent toxins (Table 3, sample #5 and #9) [21, 22]. However, these first insights suggest that glycosylation is preferred for AOH, while its monomethyl ether (AME) tends to form a sulfate conjugate. It is also possible that the sulfates are not produced by the plant but by the fungus as reported by Soukup et al. .
All sunflower seed oil samples were contaminated by minor amounts of TEN (< LOQ-3.4 ng/g). Major Alternaria toxins were detected in fewer samples and lower concentrations when compared to tomato samples or to other studies [31, 33, 37, 53]. Conjugates have not been detected in any of the sunflower seed oil and wheat flour samples. The latter matrix was generally less contaminated; only AOH and TEN were detected at levels < LOQ, and surprisingly, no TeA was detected. Other studies from China reported higher Alternaria toxin concentrations in wheat [38, 39]. TeA, TEN, AOH, and AME were found in 100, 97, 7, and 97% of 181 wheat flour samples, in the range of 1.76–520 ng/g, 2.72–129 ng/g, 16–98.7 ng/g and 0.32–61.8 ng/g, respectively .
Maximum contamination levels of Alternaria toxins in food and feed are currently not defined, monitored, or regulated in the European Union. According to the recent EFSA report [19, 20], this is caused by a substantial lack of occurrence and toxicity data. The method presented here clearly fulfills the requirements for contributing important information on Alternaria toxin contamination patterns and levels. Since the method was successfully validated and is also comparatively time- and cost-effective, it proved to be fit for the intended purpose.