**2. Results**

#### *2.1. Optimization of LC-MS*/*MS Conditions*

In this study, basic (pH 9, 10 mM ammonium bicarbonate (NH4HCO3) with ammonia (NH3)) and acidic (pH 3, 10 mM ammonium formate (HCOONH4) with formic acid (HCOOH)) aqueous mobile phases in combination with acetonitrile (ACN) as an organic phase were used to optimize the MS ionization of target CGs. As the initial step of optimization, the flow injection analysis was performed in electrospray ionization positive (ESI(+)) and negative (ESI(-)) modes. In the acidic mobile phase, the presence of intense [M + Na]+ and [M + K]+ adducts, which did not further fragment, was observed for all CGs; therefore, further optimizations were carried out with the basic mobile phase. Figure 2a,b demonstrates a full ESI(+)-MS spectrum of CON in the mobile phase at pH 3, in which an abundant presence of [M + Na]+ and [M + K]+ adducts can be clearly observed, and a full ESI(+)-MS spectrum in the basic mobile phase, in which the abundant presence of the molecular ion is apparent. As opposed to [M + Na]+ and [M + K]+ ions, fragmentation of the molecular ion provided several abundant product ions (Figure 2c), usable for defining selected reaction monitoring (SRM) transitions. While in the ESI(+) mode [M + H]+ or [M + NH4] + ions were abundant in the spectrum, the full MS scan in the ESI(-) mode revealed su fficient abundance of [M-H]- ions for all CGs. The final MS and MS/MS conditions (Table 1) were optimized in ESI(+), because of the higher intensity of MS signals in this mode.

For LC separation of the target analytes, the suitability of a UHPLC column with the C18 BEH stationary phase in combination with the basic mobile phase consisting of ACN and 10 mM NH4HCO3 at pH 9, was tested. Using a gradient elution (see Section 5.2), good separation of the CGs was achieved (Figure 3).

**Figure 2.** Full ESI(+)-MS spectra obtained through the flow injection analysis of a 1 μg/mL solution of convallatoxin (CON) in H2O + 10 mM HCOONH4 (pH 3):acetonitrile (ACN) (50:50, v/v) (**a**) and H2O + 10 mM NH4HCO3 (pH 9):ACN (50:50, v/v) (**b**) and ESI(+)-MS/MS spectrum in H2O + 10 mM NH4HCO3 (pH 9):ACN (50:50, v/v) (**c**). The vertical axes represent relative peak intensity (normalized to 100%), while the horizontal axes display measured *m*/*z* (mass-to-charge ratio) values. The MS setup is given in Section 5.3.


**Table 1.** Electrospray ionization positive (ESI)(+)-MS/MS parameters for detection of cardiac glycosides.

1Highlighted in bold: Most abundant product ion, 2 internal standard.

**Figure 3.** LC-MS/MS selected reaction monitoring (SRM) chromatograms of a single injection of a standard mixture of DIGI (**a**), DIGO (**c**), CON (**d**), and ouabain (OUB) (**e**) at a concentration of 2.5 ng/mL and OLE (**b**) at a concentration of 0.25 ng/mL, dissolved in H2O:ACN (80:20, v/v). For each cardiac glycoside (CG) the most abundant SRM transition is displayed. The vertical axes represent relative peak intensity (normalized to 100%), while the horizontal axes display retention time (in min). The chromatographic conditions applied are given in Section 5.2.

#### *2.2. Optimization of Sample Preparation*

For the evaluation of extraction and clean-up recovery, analyte-free herbal and urine samples spiked with the target CGs before and after extraction and/or clean-up step were prepared.

A herbal mixture (herbes de Provence) was used as a test mixture for the extraction experiments. ACN, methanol (MeOH), and H2O, as single solvents or as mixtures, were tested. It was found that good extraction recoveries (>70%) were obtained with ACN, MeOH, ACN:H2O (50:50, v/v), and MeOH:H2O (50:50, v/v), with slightly better results for OUB if MeOH was used in the extraction solvent. This method was aimed at achieving as low as possible limits of quantification (LOQs). It was apparent that to accomplish that a further clean-up and concentration step of the extract was necessary. The widely-used QuEChERS method [30], which combines extraction of a sample with ACN, salting-out, and subsequent dispersive solid phase extraction (SPE) clean-up, was tried, however, it demonstrated low extraction recoveries and poor clean-up e fficiency. ENVI-Carb ™ SPE with a graphite sorbent, which is very suitable for elimination of pigments that are abundantly present in herbs, resulted in no recovery of the target CGs. Other SPE cartridges, such as Discovery ® DSC-18, and Oasis ® HLB, provided acceptable recoveries but matrix e ffects were pronounced, a ffecting the sensitivity of the method. Oasis ® MAX SPE was found to be the most suitable for clean-up of the herb samples, as it showed reduced matrix e ffects, good recoveries, and improved estimated LOQs. Among the tested extraction solvents, ACN was best compatible with the required setup of the Oasis ® MAX protocol. Other solvents in combination with the herb matrix caused blockage of SPE cartridges or of the filter prior to the SPE. It should be mentioned that none of the tested protocols provided a su fficiently low LOQ and reproducible results for OUB, therefore, this compound was not included in the final method for herbal samples.

The sample preparation for urine was more straightforward and consisted of a sample dilution with H2O containing 2% HCOOH and clean-up with Oasis ® HLB SPE. The subsequent extract concentration was necessary to achieve a higher method sensitivity. In urine, this protocol was able to provide reproducible results for OUB and thus all selected plant toxins were included in the final method for this matrix.

The detailed protocols for extraction and clean-up of herb and urine samples are given in Section 5.4.

## *2.3. Method Validation*

The method validation data for herbs and urine are summarized in Tables 2 and 3, respectively. The method LOQs were calculated based on a signal-to-noise ratio (S/N) approach and are reported in order to simplify the comparison with other methods for CGs described in the literature. The LOQs were in the range from 1.5 to 15 ng/g for herbs and from 0.025 to 1 ng/mL for urine, with OLE showing the highest sensitivity among the target CGs (Table S1). The analysis of blank herb and urine samples demonstrated that no peak with a S/N of at least 3 was detected at the expected retention time of the CGs, pointing out good specificity of the method. The matrix e ffect experiments revealed that the calculated *t*-value for the target CGs in herbs and urine were much greater than the tabulated *t*-value at the 95% confidence level indicating a significant di fference between the slopes of calibration curves in the solvent and matrix, i.e., the presence of matrix e ffects. All CGs in both matrices su ffered from a signal suppression, with the strongest suppression for OUB and the smallest e ffect for OLE and DIGI. The calibration curves were prepared in matrix extracts (spiked post clean-up) by plotting the concentration of the analyte in the calibration standards against the ratio of peak area of the analyte to the internal standard, digoxin-d3 (DIGO-D), for all target CGs. Calibration curves for herb and urine samples were linear over the validated concentration range with coe fficients of determination (R2) >0.997. The lowest calibration level (LCL) is used as a reporting limit for quantification of CGs in herbs and urine. The mean (apparent) recovery data obtained for three concentration levels in herbs were in the range from 83% to 115% for OLE, DIGO, and DIGI, and 55% for CON. The mean recoveries for all CGs in urine ranged from 80% to 96%. The method precision was expressed as a relative

standard deviation (RSD) of replicate measurements. For herbs, the repeatability (RSDr) of the method ranged from 6% to 14% and the within-laboratory reproducibility (RSDwR) was from 7% to 17%, while these parameters for urine ranged from 1% to 7% and from 5% to 19%, respectively. The expanded measurement uncertainty (MU) was not higher than 28% and 37% at the lowest concentration levels validated for herbs and urine, respectively. The uncertainty at higher concentration levels did not exceed 31% and 16% for herbs and urine, respectively.

#### *2.4. Method Application for Analysis of Culinary Herbs*

The validated LC-MS/MS method was subsequently used to investigate the contamination of culinary herbs and spices that are available on the Belgian food market. In total, 65 samples were acquired in supermarkets and organic food shops and comprised the culinary herbs and herb/spice mixtures containing bay leaves (*Laurus nobilis*). For the majority of samples, the country of production was not specified. About 20% of samples originated from organic farming. The detailed information on ingredients of the samples is given in Table S2.

Quality control samples, namely a standard mixture of CGs in a neat solvent and herb mixture fortified with CGs at the concentrations corresponding to the middle validated level, were included in each sample sequence. Identification of CGs in samples was completed following the Commission Decision 2002/657/EC [31]. This implied the presence of a peak of the target analyte with a S/N ratio of at least 3 for each ion transition, compliance of relative retention times, and conformity of deviations of relative ion intensities with regards to the matrix-matched calibration standards. The most abundant product ion was used for quantification, while the second product ion was used for confirmation of the analytes. The analysis demonstrated that none of the collected samples contained CGs above the LCL.


**Table 2.** Validation data for cardiac glycosides in culinary herbs.



#### *Toxins* **2020**, *12*, 243
