*2.3. Method Validation*

The method was validated in accordance with the Commission Decision 2002/657/EC decision [38] and CEN/TR 16059:2016 guidelines [39]. The fortification levels were 10, 20, and 30 μg/kg for TEA, ALT, and TEN, respectively; and were 5, 10, and 15 for AOH and AME, respectively. These levels were set in line with the validation ranges used in the MVS [6]. Investigations at higher concentration levels were not needed because natural contamination of oils was reported at low μg/kg levels only. Six parallel samples were analyzed at each level that are in accordance with the EU guideline (Table S2). Measurements were carried out over 3 days, and all 54 samples were analyzed (3 levels × 6 samples × 3 days). The performance characteristics were as follows: selectivity, identification, linearity, recovery, precision, and limit of quantification (LOQ).

Blank samples were spiked and analyzed using the optimized method (Sections 4.2 and 4.3). The chromatograms obtained from the blank samples were free of any interfering peak. For identification, the ion ratios (IAs) were calculated for all compounds in both neat standard solutions and samples. IAs were all within the tolerance ranges for all toxins (Table S2). The selectivity and identification met the criteria of EU guidelines [38]. Five-point calibration curves were performed to evaluate the linearity. Concentration levels, determination coefficients (R2), and equations are given in Table S2.

The requirement for recovery has been obtained between 70% and 120% at spiking levels used for validation [38,39]. The recovery varied from 73.6% to 95% at levels between 5 μg/kg and 15 μg/kg for AOH and AME. In the case of TEA, ALT, and TEN, the recovery was between 92.4% and 122% at the concentration range of 10–30 μg/kg. Only one value (122%) exceeded the acceptable ranges. Below the concentration of 100 μg/kg, the precision should be as low as possible [38], normally, RSD ≤ 30% [39]. The within-laboratory precision varied from 10.1% to 22.2% (Table S2). The LOQ was calculated from the SNR and evaluated as 10 times of SNR. The LOQ was checked by fortifying blank samples (*n* = 6) with standard solution to obtain the individual LOQ levels, and samples were analyzed. The SNR was above 10 in each sample and the IAs were in the acceptable ranges.

#### *2.4. Analysis of Sunflower Oil Samples*

Sixteen different brands and lots of sunflower oil samples were collected and analyzed for the five toxins mentioned above. Three samples were contaminated at low levels, in which only TEA and TEN were detected. One sample (cold pressed oil) contained both TEA (12.8 μg/kg) and TEN (7.1 μg/kg). The other two samples (refined oils) contained TEN at concentrations between 4.5 μg/kg and 5.0 μg/kg.

#### *2.5. Analysis of Sunflower Seed QC Samples*

In lack of sunflower oil QC samples, sunflower seed QC samples were measured. The method optimized for sunflower oil had to be modified to obtain the suitable method for sunflower seeds (Section 4.6). Both spiked (C08 SP and Q25 SP) and naturally contaminated (W52 NC) samples were tested. The samples were leftovers from MVS performed by JRC in 2018. The detected values and reference concentrations are given in Table S2. Even though the method presented herein was developed for sunflower oil samples, the concentrations detected in sunflower seed samples were not considerably different to the reference values. The method could not detect ALT at all, since the reference concentrations were all below the LOQ (10 μg/kg). Also, AOH and AME were not found in C08 SP due to the same reason. The accuracy of the method for sunflower seed samples was between 72% and 129% (Table S2).

#### **3. Discussion**

#### *3.1. Method Development for LC-MS*/*MS Separation*

The MS/MS detection of *Alternaria* toxins can be carried out in both positive and negative ionization modes (Table S1). The choice of polarization mode can be instrument dependent, but the negative mode usually results in a higher sensitivity for these toxins (Table S1) due to their weak acidic characteristics. We also observed considerable enhancement in sensitivity when negative ionization was applied. In addition to the ionization mode, the choice of ion source can also influence the sensitivity of MS/MS detection of *Alternarias*. Zwickel et al. [30] tested three ion sources (ESI, APCI, and atmospheric pressure photo ionization) for these toxins and found that the ESI was the most suitable one. In general, the ESI was employed (see Table S1), but Prelle et al. [16] reported three times higher responses for TEA when APCI was used, while the rest of the toxins had similar sensitivity in both ESI and APCI modes. Even though the ESI source of the applied LC-MS/MS instrument enabled appropriate sensitivity for the compounds other than *Alternarias*, the sensitivity for *Alternaria* toxins was quite a bit lower than those reported in earlier methods utilizing other types of instruments (Table S1). This led to the application of an APCI probe that significantly improved the instrumental LOQ for all compounds. One participant in the MVS 2018 used the same instrument as in our study and also applied the APCI source [7].

The mobile phase pH was set at 8.8 due to the negative ionization mode and the chromatographic separation of TEA. In the existing methods, the alkaline pH was used when the detection was carried out in negative ionization mode; and the acidic or neutral eluent pH was utilized if the positive ion mode or polarity switching was employed (Table S1). Even though the alkaline mobile phase pH is not usual in LC-MS/MS separation, it is feasible for *Alternaria* toxins due to the chromatographic problem with TEA at acidic pH condition. Moreover, the alkaline pH enhanced the sensitivity in the negative ion mode. The acidic pH condition did not result in the appropriate peak shape for TEA and also lowered the sensitivity of MS/MS detection in negative ionization mode. The chemical derivatization, suggested in some papers [13,19,20], was not tested. Although this approach enabled the simultaneous separation of *Alternaria* toxins [13], it could have further increased the LOQ and the preparation time and costs.

#### *3.2. Method Development for Sample Preparation*

In this study, we focused on sunflower oil samples only and developed a LC-ID-MS/MS method involving a unique sample preparation approach suitable for this kind of lipophilic matrix. The goal was to develop a dilute-and-shoot method that is frequently used in toxin analysis by LC-MS/MS method [40]. The non-polar matrix constituents of oil were eliminated with hexane that could easily dissolve the oil. The toxins have low solubility in hexane; hence, the target compounds could be extracted into a non-miscible solvent such as water, methanol or acetonitrile. Even though some studies have reported the use of the general acetonitrile-based mycotoxin extraction solvent mixture (acetonitrile/water/acetic or formic acid) [41] for *Alternarias* (Table S1), we did not prefer the acetonitrile as a solvent due the lower solubility of *Alternarias* in acetonitrile. Methanol is a more suitable solvent for these toxins, and therefore, methanol/water/acetic acid mixture has been utilized for extraction in the candidate method for standardization [6]. In the case of cereal samples, the extraction medium should contain water due to the starch content of samples; and the aqueous methanolic solvent in our case was needed to obtain better solvent separation between the hexane layer and the extraction medium. Also, water can enhance the extraction of TEA with polar characteristics. The experimental design showed that 80% (*v*/*v*) methanol in water gave the best extraction from the naturally contaminated oil. In other types of samples (e.g., tomato, cereals, and oilseeds), ~ 80% methanol also resulted in the optimal extraction solvent composition (Table S1). The HPLC separation was carried out at alkaline pH, so acid was not added into the extraction solvent to avoid the large pH difference between the injection

solvent and the mobile phase, which could deform the chromatographic peak. The experimental design also indicated that the optimal sample-to-solvent ratio was 4.0, which is a general ratio in mycotoxin analysis based on the dilute-and-shoot approach [40,41].

High ME (mainly ion suppression) usually influences the mycotoxin analysis based on the LC-MS/MS method [40,41] that is also true for *Alternarias* [8,9,13,18,19,23,28,29,31,42]. The lower sensitivity of our instrument and the high ME for AOH and AME increased the LOQ. Hence, sample pre-concentration with evaporation and reconstitution was necessary to obtain appropriate LOQ (≤ 10 μg/kg) for all compounds. It should be pointed out that an instrument with higher sensitivity would allow further dilution of the extracts that could decrease the preparation time and the ME of analysis. The elimination of co-eluting matrix constituents was tested with SPE clean-up on mixed-mode cation exchange cartridges. Although the mixed-mode SPE and subsequent reversed-phase HPLC measurements enabled an orthogonal separation approach, considerable improvement in ME could not be seen (Table 2), and only the relative ME was enhanced. However, the sample preparation time and overall costs were also increased. In conclusion, the mixed-mode SPE clean-up did not improve the overall analytical process since it is time-consuming and more expensive compared to the dilute-and-shoot approach. The NP SPE clean-up was alternatively tested since this approach requires only sample dissolution in hexane and the dissolved samples can be directly subjected to NP SPE. The NP SPE clean-up would be a simpler clean-up approach, but using this process, we lost either the TEA or the AME, depending on the elution solvent composition. Overall, the dilute-and-shoot approach was the most suitable sample preparation method.

The need of isotope dilution for *Alternaria* toxin analysis by LC-MS/MS method has been strongly suggested by Asam and Rychlik [42] in 2015. Accordingly, isotopically labeled analogues were necessary for the analysis. In line with that, an important conclusion of the ME study was that all corresponding isotopically labeled analogues were necessary for the analysis. This study is the first in using all ISTDs for five *Alternarias* analyzed. While a moderate ion suppression influences the signal of TEA, ALT, and TEN, the ME for AOH and AME is much higher. This means that AOH-d3 and AME-d3 cannot compensate the ME of other analytes, and the ISTDs are not exchangeable. The differences in ME among the compounds analyzed can originate from the retention time differences between toxins and from the various structures of analytes. Even though ALT has similar structure to AOH, the 1.2 min of retention time difference (Figure 2) resulted in considerable difference in the ion suppression (Table 2). On the other hand, there was a significant difference in slopes of matrix-matched calibrations of AME obtained from three different oils. The relative ME was evaluated from the precision of slopes in the matrix-matched calibrations and showed that the matrix-matched calibration could strongly influence the quantification of AME. Hence, the isotope dilution method is needed for appropriate quantification. The relative ME values for AME were significantly improved with the ISTD evaluation (Table 2). In general, the relative ME values were improved for all compounds under ISTD evaluation. It means that the slopes of three different matrix-matched calibrations were close to each other, and they were nearly free of ME.

#### *3.3. Real Sample Analysis*

Chulze et al. [43] has already reported the high (30 μg/kg AME—15.796 μg/kg TEA) and frequent (85%) contamination of sunflower seeds with *Alternaria* toxins in 1995. Under processing sunflower oil from the oilseeds, the *Alternaria* toxins may appear in the oil product due to the contamination of sunflower seeds with these toxins. The high natural contamination of sunflower seeds with *Alternarias* reported recently worldwide [4,6,10–12] indicates that cross contamination with toxins can occur in the final sunflower oil products. Even though Chulze et al. [43] has described the decrease of *Alternaria* toxins during the processing of sunflower seeds to oil, the TEA and AME contamination of raw seeds were still detectable in lower concentrations in the oil after processing [43]. Due to the non-polar character of AME, the occurrence of this toxin in lipophilic oil matrix is more likely, as reported previously [43]. Accordingly, the polar characteristics of TEA inhibit its accumulation in oil that was also proven in

another study [43], while the AOH contamination of oilseeds could not be detected in the oil product at all. It should be pointed out that the method used by Chulze et al. [43] had a LOQ of 50 μg/kg (AOH), but the recent methods have much lower analytical limits. To the best of our knowledge, no other newer studies have dealt with the decrease of *Alternaria* during the process of oil from sunflower seeds or other types of oilseeds. Since there is a great consumption of sunflower oil worldwide, the need for involving this sample in toxin analysis is to support the legislation. The analysis on sunflower oil was performed by Liu and Rychlik [8] in 2013. López et al. [11] has conducted studies involving other types of foods as well (Table S1). In 2016, López et al. [11] found relatively high (up to 1350 μg/kg) and frequent (80%) contamination of sunflower seeds with TEA, but sunflower oils contaminated with TEA above LOQ (5 μg/kg) were not found [11]. Only AME was detected at 17 μg/kg in one of 11 oil samples, and other toxins were all below the LOQ [11]. Liu and Rychlik [8] also investigated several types of refined and cold-pressed oil samples like pumpkin seed oil, rapeseed oil, sunflower oil, and thistle oil. They detected TEN in three refined (up to 3.95 μg/kg) and three cold-pressed (up to 6.73 μg/kg) sunflower oils, and also in a rapeseed cold pressed oil (up to 0.64 μg/kg). Furthermore, the dihydrotentoxin could be detected (up to 4.48 μg/kg) in three cold-pressed sunflower oil [8].

We have analyzed 16 sunflower oil samples: one sample was a cold pressed sample, and the others were refined ones. In agreement with Chulze et al. [43] and López et al. [11], AOH was not detected and a low concentration of TEA was found (12.8 μg/kg), but only in the cold pressed oil. In three samples, TEN was detected between 4.5 μg/kg and 7.1 μg/kg, similar to those reported by Liu and Rychlik [8]. We have also found that the cold-pressed oil is more likely to be contaminated than the refined samples. Comparatively, Chuzle et al. [43] did not investigate the TEN and López et al. [11] did not find TEN in sunflower oil. Both TEA and TEN have the least toxicity [4] and the detected concentrations are below the validation range suggested by CEN [6], therefore, these contaminations may not cause any risk to human health.

The focus of our study was on sunflower oil since ML would be set for sunflower in near future.
