Optimization of Chromatographic and MS Conditions
All 25 target analytes presented in Table 1 were selected based on current global regulatory requirements and potential safety concerns. In consideration of the low targeted analytical range, the complexity of essential oil sample matrix, and the necessity to differentiate each coumarin derivative, LC-MS was considered as the most appropriate analytical technique. Although LC-tandem mass spectrometry (LC-MS/MS) and LC-high resolution mass spectrometry (LC-HRMS) offer the most ideal selectivity and sensitivity, a properly developed LC-MS method in single ion monitoring (SIM) mode would provide sufficient method performance for this analysis. Furthermore, it is relatively easy to operate and maintain a single-quadrupole mass spectrometer and harmonize this method in multiple laboratories for quality control testing.
Suitable chromatographic conditions are essential to achieve optimal method specificity and sensitivity by minimizing matrix interference and resolving critical pairs of target analytes. In general, reversed-phase chromatography is amenable to the reasonable retention of coumarin derivatives. However, it is a challenge to accommodate 25 targeted coumarin derivatives with acceptable selectivity in a short LC analysis. In particular, the target analytes included six groups of isobaric analytes that are not feasible to be resolved by the MS detector, hence their selective identification fully relied on the resolving power of LC separation. To determine the most ideal analytical column for this method, the separation of the target analytes using UHPLC columns of different stationary phases was investigated. As shown in Fig. 1, the chromatograms were obtained from the analyses on polar C18, phenyl-hexyl, fluoro-phenyl and biphenyl columns of the same dimensions (100 × 2.1 mm) using identical LC gradients of water and methanol. The biphenyl column clearly offered the best retention for all the target analytes, attributed to the strong pi-pi interaction between the coumarin motif and the biphenyl stationary phase. Fewer analytes coeluted using the biphenyl column, with only three groups of coeluting clusters (6-methylcoumarin and psoralen; oxyimperatorin, isobergapten and pimpinellin; isoimperatorin and phellopterin) which can be resolved by MS detector based on their m/z difference of molecular ions. In comparison, the coelution of 8 to 14 analytes were observed using the other three types of columns. With a slight adjustment to a shallower LC gradient, baseline separation of all the critical analyte pairs was achieved, including the group of four isobaric analytes 8-MOP, sphondin, bergapten and isobergapten. Therefore, the biphenyl column was chosen to perform all the subsequent optimizations. The total analysis time of one injection was 14 minutes, which significantly improves testing throughput compared to many published methods.(Govindarajan et al. 2007; Lin et al. 2009; Macmaster et al. 2012; Dugrand et al. 2013) However, it was noted that bergaptol was unable to be resolved from its isobaric isomer xanthotoxol on the biphenyl column. Bergaptol and xanthotoxol can either be quantified as a sum or will require an alternative LC separation approach.
A water-methanol mixture was used as eluent due to the incompatibility of acetonitrile with biphenyl stationary phase.(Yang et al. 2005) To improve peak shape and facilitate the ionization, modifications to the mobile phase pH were investigated. The peak shape and retention time remained constant with the increase of mobile phase acidity, attributed to the lack of ionizable functional groups in the targeted coumarin-derivatives. (Fig. 2) However, it is noteworthy that the increased acid concentration actually suppressed the ionization process, resulting in lower MS response of certain analytes. In particular, the peak area of 7-methoxycoumarin was 20% higher in pH 3.5 compared to pH 2.7. Taking this into consideration, the addition of 0.01% formic acid as the additive of mobile phase A was sufficient for improved peak shape and optimal MS response.
All of the target coumarin derivatives yielded stable protonated molecular ions [M + H]+ in positive electrospray ionization (ESI). Most of the molecular ions are well suited in the selected ion monitoring (SIM) to represent each target analyte. However, several molecular ions were less specific due to the high MS background and the susceptibility to matrix interference, including bergapten, sphondin, 8-MOP and isobergapten (m/z 217.0); phellopterin (m/z 301.0); byakangelicin (m/z 335.1); and epoxybergapmottin (m/z 355.2). (Fig. 3) In order to find the most suitable ions in SIM acquisition to represent each of the target analytes, the fragmentor voltage was ramped up in 10 V increments from 70 V to 250 V to encourage collision-induced dissociation of the molecular ions. Through this process, we were able to identify the most specific signature fragment ions of each target analyte and their respective optimal fragmentor voltage. In particular, fragment ions of bergapten, sphondin, 8-MOP and isobergapten (m/z 202.0); phellopterin (m/z 218.0); byakangelicin (m/z 317.1); and epoxybergapmottin (m/z 337.1) were identified to offer significantly improved signal-to-noise ratio (S/N) compared to their molecular ions. The top two ions of each target analyte were selected as quantifier ion and qualifier ion in the SIM acquisition, which further improved the method specificity. The complete list of the SIM ions and fragmentor voltages are summarized in Table 1. The optimized chromatography and MS conditions enabled sensitive detection of most target analytes down to 10 to 20 pg on column. 7-methoxycoumarin and epoxybergamottin were the least sensitive analytes due to the higher background of their respective SIM ions. Overall, 40 pg on column was an achievable limit of detection (LOD) for all the target analytes with corresponding S/N more than 3.
Development of Sample Preparation Procedure
The major constituents of citrus-derived essential oils are various monoterpenes, such as limonene, α-pinene, γ-terpinene and myrcene, with limonene as the most abundant constituent commonly above 50% (w/w).(Kvittingen et al. 2021) Although monoterpenes are hydrocarbons that are not ionizable by ESI, they share elution profiles similar to coumarin derivatives in reversed-phase chromatography. To understand whether the analysis of coumarin-derivatives is affected by the presence of concentrated limonene in the citrus-derived oil, distilled wild orange oil fortified with the target analytes was dissolved in methanol and analyzed by a LC-UV-MS system, with UV detection at the apex of limonene UV-Vis spectrum (206 nm). Although most of the target analytes were not affected, suppressed response and retention time shifts were observed in the elution of imperatorin (8.5 min), trioxsalen (8.6 min) and isoimperatorin (9.3 min), which happened to overlap with the elution region of limonene (8.3–8.8 min) (Fig. 4AI-AIII). Column performance and MS ionization were compromised by the overloading of an abundance of limonene in the tested sample. Thus, additional sample clean-up to remove limonene is required for the accurate analysis of all the targeted coumarin-derivatives.
QuEChERS is the most widely used sample preparation in food and environmental applications due to excellent cost-effectiveness.(Anastassiades et al. 2003; Majors et al. 2010) It is the primary solution to extract organic residues from many food matrices, such as protein, carbohydrates and lipids. However, it is ineffective at selectively extracting target analytes from small molecule matrix components with similar partition coefficients. The calculated partition coefficient of limonene (logP = 3.22) is within the range of the partition coefficients of the target analytes (logP 1.0–4.8), indicating a tendency of QuEChERS to co-extract both the analytes and limonene into the acetonitrile phase (Table 1). Thus, an alternative technique to remove the monoterpene matrix was explored for this analysis.
The use of an appropriate, selective extraction solvent turned out to be a more suitable sample preparation technique. Although limonene is soluble in medium polarity solvents such as methanol and acetonitrile, the use of water as a co-solvent effectively minimizes its solubility. On the other hand, coumarin derivatives are relatively more soluble in a methanol-water mixture. A water-methanol mixture as the extraction solvent is capable of extracting coumarin derivatives in the presence of monoterpene sample matrix. 10% water in methanol (v/v) was determined as the optimal ratio in removing abundant limonene while maintaining acceptable recovery of all target analytes. After centrifugation, most of the sample matrix precipitated while the target analytes extracted into the upper layer of the extraction solvent. Further increases of water ratio marginally improved the removal of limonene but resulted in much higher recovery loss of the least polar analytes, such as bergamottin and 5-geranoxy-7-methoxycoumarin.
For a sample containing > 90% limonene, 10% water in methanol (v/v) as the extraction solvent is not sufficient to remove all the matrix interference for LC-MS analysis (Fig. 4BI-III). To further improve limonene removal, an additional step of liquid-liquid extraction with hexane was introduced. A sample diluted with 10% methanol was mixed with hexane in a 1:20 ratio, then it was partitioned by centrifugation and the hexane layer was discarded. This enabled the removal of the matrix interference associated with residual limonene in the sample extract. As shown in Fig. 4III, the limonene level in the wild orange oil sample extract after liquid-liquid extraction was significantly reduced compared to the methanol/water extract. The retention time of trioxsalen, imperatorin and isoimperatorin in the wild orange oil sample extract also overlapped well with a solvent standard of the same concentration, indicating the successful removal of limonene interference (Fig. 4CI-II). Subsequently, the sample was incubated with C18-based dispersive solid phase extraction (dSPE) sorbent to remove residual hexane that can cause retention time shifts.
The optimized extraction procedure was evaluated through the spiked recovery of target analytes in various essential oils at 5 µg/g, which is the intended LOQ of this analysis. Nineteen essential oils were screened as representative of herbal extracts derived from various citrus peels, leaves and flowers, as well as from the similar botanical family of Rutaceae and Apiaceae with reported phototoxicity. In general, the spiked recoveries were mostly within the acceptable range (75–120%), as represented by the green color in Fig. 5. Several essential oils produced from citrus peels contain measurable levels of targeted coumarin-derivatives that hinder spiked recovery determination at 5 µg/g fortification level. Thus, their recoveries were not determined (ND). Out of these 25 targeted compounds, epoxybergamottin, bergamottin, 8-geranyloxypsoralen and 5-geranoxy-7-methoxycoumarin showed lower spiked recovery. The high lipophilicity of epoxybergamottin (LogP = 3.8), bergamottin (LogP = 4.8), 8-geranyloxypsoralen (logP = 4.8) and 5-geranoxy-7-methoxycoumarin (LogP = 4.5) was presumably causing their recovery loss during the sample extraction. Overall, this optimized sample extraction procedure was demonstrated to be applicable for various types of essential oils.
Method Validation
An inter-laboratory method validation study was conducted in accordance with ICH guidelines on validation of analytical procedures.(International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use 2005) Based on ICH guidelines, this method is considered as a category II Quantitative Impurity Assay, thus the method performance characteristics evaluated as part of the validation study included specificity, linearity, range, limit of quantitation (LOQ), accuracy and precision.
The method specificity was investigated by comparing a blank sample with the same blank sample spiked with the target analytes. The selection of a blank citrus essential oil free of coumarin derivatives was rather limited because of naturally occurring coumarin derivatives inherent in most of citrus oils. Through our initial sample screening, yuzu oil was chosen as a representative blank sample without detectable amounts of target analytes. All the primary SIMs were shown to be highly selective with no chromatographic interference observed in yuzu oil. The chromatograms are presented in supporting information (Figure S1). The secondary SIMs were less selective and used as qualifier ions for confirmation purpose. Overall, successful analyte identification was demonstrated at concentrations equal to or above the method LOQ.
The analytical range evaluated in this study was from 0.04 µg/mL to 1 µg/mL, which was equivalent to the range of 5 µg/g to 125 µg/g in the essential oil sample. Calibration curves were established by a set of five calibration standards using quadratic regression and 1/x weighting. The quadratic standard curves provided a better fit due to a non-linear response near the upper end of the calibration curves, presumably related to saturation during ionization.(Yuan et al. 2012) Over the course of the validation study, the correlation coefficient (r) of all target analytes were consistently above 0.99, which demonstrates acceptable linearity of this method (Table S1).
Due to the absence of citrus-derived oils with certified levels of coumarin derivatives, the method accuracy was evaluated by the marginal recovery% of the blank sample fortified with target analytes at four concentration levels (5 µg/g, 20 µg/g, 40 µg/g and 48 µg/g). Repeatability (RSDr) was determined from the analysis of the fortified sample in three to six replicates, and the inter-laboratory reproducibility (RSDR) was determined from the repeatability results from two independent laboratories. As shown in Fig. 6A and Table S2, acceptable recoveries and RSDs were obtained in the validation study. The mean recoveries were within 80–115%, except bergamottin and 5-geranoxy-7-methoxycoumarin. The lower mean recoveries in the range of 60–80% were presumably due to recovery loss during the extraction because of their high lipophilicity. The RSDr of all the analytes at 5 µg/g fortification level was between 4–10%. In higher fortification levels, the analyses were more precise with RSDr of all 25 targeted analytes below 5%. The precision results obtained from two laboratories was also aligned with the single-laboratory results, with RSDR below 6% for all the analytes (Fig. 6B). The narrow distribution of RSDr and RSDR demonstrates the high consistency and precision of this analytical method.
The LOQs were determined as the lowest fortification level that met the identification criteria and showed acceptable recovery and precision. Yuzu oil fortified at 5 µg/g had a S/N consistently above 10 for all 25 analytes in the two laboratories, as shown in Table S3. The spike recovery and precision also met their respective acceptance criteria. Therefore, the LOQ of this analytical method is set as 5 µg/g.