LPGC-MS/MS Analysis
In previous LPGC-MS/MS studies in our group, we used a 7000A MS/MS instrument [10, 21], which was upgraded to a 7000B [13, 18], and now to a 7010 system. In regulatory monitoring of pesticide residues in foods, the need of analysis calls for LOQ <10 ng g−1 [22]. The injection of the least amount of sample equivalent to meet the desired LOQ improves the ruggedness of the method and reduces instrument maintenance demands. Originally with the 7000A instrument, 10 mg sample equivalent (1 mg µL−1 QuEChERS extracts in MeCN) was injected [10, 21], which was reduced to 2.5 mg when using the 7000B [18]. In this study, the 7010 upgrade enabled injection of 1 mg (1 µL) sample equivalent to still achieve <5 ng g−1 LOQs for all analytes in the fruit and vegetable matrices. Cold inlet conditions using programmable temperature vaporization were used in each method thus far, but in the near future, we plan to investigate hot split mode injections to further speed the analysis, possibly improve performance, and reduce the amount of matrix components being introduced into the instrument [23].
We started to estimate LOQs from the final method in this study, but recognized quickly that the calculations were inaccurate due to ultra-trace carry over or artifacts that infiltrated the background for nearly all analytes in MeCN, reagent, and matrix blanks. The carry over averaged 0.4 % of the previous injection, and likely originated from the injection syringe needle contaminating the 100 mL wash bottles on the autosampler, not from the mini-SPE procedure which prepared 14 blank extracts prior to the reagent blank in its sequence.
Peak integration was conducted using the summation function in the Agilent MassHunter software, which simply drew a baseline at the lowest point between analyst-defined start and stop times to cover the t
R and peak width for each analyte. The integration start and stop times were set by ensuring that they fully covered the analyte peaks for all 325 injections over the course of all five sequences in the validation study. The peak area was always positive using this software’s summation integration function, which led to positive responses even when only noise was present. However, inspection of the MRM chromatograms and ion ratios clearly indicated that the background peaks in reagent and solvent blanks originated from the analytes in most cases, not simply electronic or chemical noise. The equivalent background concentrations of analytes in RO blanks averaged 0.5 ng mL−1 for the pesticides, FRs, and PAHs, and 0.05 ng mL−1 for PCBs, which did not affect the results of the study, but also did not permit an accurate estimation of LOQs. The likely source of carryover can be eliminated using a newer type of flowing solvent wash station that rinses the analytes coating the needle to waste rather than allowing them to contaminate the fixed volume wash solutions. LOQs can be better estimated after this issue is resolved, but the 1 µL injection clearly met analytical purposes in this application. We can conclude that the LPGC-MS/MS method yielded LOQs <0.5 ng g−1 (<0.05 ng g−1 for PCBs) for all but a few analytes, such as fenpropathrin, endosulfans, and BDE 183 which gave LOQs ≈1 ng g−1.
Automated Mini-SPE
Morris and Schriner [12] reported that mini-SPE cartridges containing 20 mg anh. MgSO4, 12 mg PSA and C18 each, and 1 mg CarbonX (45 mg total) provided satisfactory cleanup and recoveries for a wide range of pesticides and matrices in GC-MS/MS analysis. For LC-MS/MS, they chose to use cartridges containing 20.7 mg C18, 8.3 mg Z-Sep, and 1 mg CarbonX (30 mg total), which provided acceptable cleanup and pesticide recoveries for monitoring purposes [12]. In their method, they pre-conditioned the mini-cartridge with 150 µL MeCN, added 150 µL QuEChERS extract, then eluted with another 150 µL elution solvent.
In d-SPE of QuEChERS extracts, no pre-conditioning or extra solvent elution steps are done, and the MeCN extract itself serves as the elution solvent. Ideally, the sorbents only adsorb co-extracted matrix components and not analytes. We decided to streamline and speed the method of Morris and Schriner [12] by eliminating the MeCN pre-conditioning and solvent elution steps for the same mini-cartridges they developed, which are commercially available. They conducted very thorough studies to set the sorbent combinations, and we had previously found similar sorbent mixtures work well for cleanup and analysis pesticides and environmental contaminants in seafood matrices using filter-vial d-SPE [13]. We have since extended the filter-vial d-SPE method to other animal tissues (cattle, pork, and chicken muscle), but found that only filtering of initial extracts was needed for analysis of the 99 LC-amenable pesticides in our recent approach [18].
In this evaluation, we focused on 97 representative GC-amenable analytes and compared the two different mini-cartridges for their analysis in QuEChERS extracts of kale, salmon, pork, and avocado, and using extract load volumes of 200, 300, 400, 500, and 600 µL. To determine elution and dead volumes of the mini-cartridges, weights of the collection vials were compared before and after cleanup. Figure 1 shows the results in which measured dead volume (based on measured solution densities) was found to increase slightly as more extract was loaded into the mini-cartridges at 2 µL s−1. The reason for this observance was very likely because more interstitial spaces in the sorbent beds were being filled as more solution was passed through the cartridges. In any case, the dead volumes were much the same for both types of mini-cartridges, and averaged 75 ± 5 to 90 ± 8 µL from 200 to 600 µL extract load volumes, respectively. Thus, the measured extract elution volumes were 125 ± 5 to 510 ± 8 µL from 200 to 600 µL extract load volumes, respectively, which provided known and rather consistent elution volumes in the method.
In the validation study involving 235 mini-SPE cartridges, final extracts averaged 278 ± 5 μL (1.9 % RSD), in which only 4 final extracts had ≈25 μL reduced volume. The robotic liquid handler operated flawlessly throughout the study, and perhaps vial weights were transcribed improperly, bubbles materialized in the syringe, or mini-cartridge dead volumes were larger, which caused those four outliers.
Mini-SPE Cleanup
Full scan LPGC-MS chromatograms of the reagent blank, kale, avocado, salmon, and pork QuEChERS extracts were compared before and after the mini-SPE cleanup using the 200–600 µL load volumes. Results indicated that the 30 mg C18/Z-Sep/CarbonX (20.7/8.3/1, w/w/w) mini-cartridges provided little cleanup for any of the 4 matrices independent of extract load/elution volumes (see Supplemental information). The chromatograms appeared much the same in each injection. These mini-cartridges were devised for LC-MS/MS [12], and this LPGC-MS result is probably due to the lack of anh. MgSO4 in the cartridges to reduce water content in the final extracts. Density measurements of final extracts indicated that the 45 mg mini-cartridges containing 20 mg anh. MgSO4 led to cleaned extracts (before addition of AP + QC solution) of ≈2 % water (0.79 mg μL−1 density of the MeCN extracts) and the 30 mg cartridges without anh. MgSO4 maintained the same extract density (0.82 mg μL−1) as the initial extracts, which were ≈15 % moisture. Drier solutions are better for GC analyses and provide stronger adsorption properties for common sorbents (e.g. PSA, Z-Sep, silica, Florisil, Alumina).
In the case of the 45 mg mini-cartridges of 20/12/12/1 anh. MgSO4/PSA/C18/CarbonX, significant removal of matrix co-extractives was observed in the LPGC-MS chromatograms of kale, pork, and salmon, but little difference was observed for avocado (see Fig. 2). As would be expected, greater cleanup took place using smaller load/elution volumes before breakthrough began to occur. Depending on analyte recoveries, 200–300 µL would be preferably chosen to provide more cleanup and shorter time than achieved with larger volumes.
Unlike LPGC-MS, UV–Vis absorbance spectrometry of the 200–600 µL final extracts (200 µL of combined replicates in 96-well plates) showed similar cleanup efficiencies using either the 45 or 30 mg mini-cartridges (see Supplemental information). As shown in Fig. 3, chlorophyll and xanthophyll co-extractives from the kale in particular were nearly eliminated by the sorbents, mainly due to CarbonX. Salmon and pork extracts were nearly colorless, and measured UV–Vis cleanup efficiency was small compared to full scan LPGC-MS results. In the case of avocado as shown in Fig. 2, LPGC-MS showed little differences in full scan chromatograms, but UV–Vis spectra exhibited strong reduction in absorbance readings from 300 to 700 nm (see Supplemental information). Kale co-extractives were dramatically reduced in both types of measurement; especially by the 45 mg mini-cartridges (see Figs. 2, 3).
Analyte Retention in Mini-SPE
The extract elution volume not only affected the cleanup efficiency in mini-SPE, but analyte recoveries also depend on load/elution volume. LPGC-MS/MS was used to compare relative recoveries (w/o using IS) normalized to the average 600 μL result with respect to extract load volumes in triplicate for each of the four matrices. Figure 4 shows an example of typical results that were representative of nearly all of the pesticides, demonstrating that neither load/elution volume nor matrix made a significant difference in recoveries. The 30 mg 3-sorbent mixture mini-cartridges also gave similar analyte results as the 45 mg 4-sorbent mini-cartridges. Figure 5 shows the results for the IS with respect to extract load volumes. Only PAHs and a few other analytes with co-planar chemical structures gave less than complete elution, mainly due to partial retention on CarbonX. Hexachlorobenzene (HCB) is one the few pesticides that gave incomplete elution at 200 μL extract load volume. Figure 6 shows how 300 μL load volume (≈220 μL elution) yielded ≈80 % relative HCB recovery while still removing ≈95 % of co-extracted chlorophyll from initial QuEChERS extracts of kale. PBDE 183 and co-planar PCBs (126 and 169) yielded similar results as HCB, but large PAHs (≥3 rings) did not fully elute even with >500 μL load volumes. Similar results were achieved with the 30 mg mini-cartridges, indicating that 1 mg CarbonX was mostly responsible for the retention of the PAHs. As shown in Supplemental information, no significant differences were observed in the relative recoveries for the analytes when using either type of mini-cartridge. Independent of extract volumes, isotopically labeled IS of several PAHs were expected to help compensate for reduced actual recoveries in the final method (see Fig. 5).
Considering the notable cleanup efficiency observed in both UV–Vis and LPGC-MS for the 45 mg mini-cartridges, especially at smaller load/elution volumes, and lack of retention of most analytes in QuEChERS extracts, we chose to use 300 µL load volumes with those mini-cartridges in the final method.
Method Validation and Summation Integration
For validation of the automated high-throughput cleanup and LPGC-MS/MS method, 10 representative food matrices were chosen using SANTE/11945/2015 [22] as a guide. The final method was conducted for two matrices per day in the following sequence: Day (1) Gala apple and kiwi; (2) carrot and kale; (3) navel orange and canned black olive; (4) wheat grain and dried basil; and (5) pork loin and salmon. QuEChERS has been extensively validated [24] and implemented in many labs [25], and to isolate the mini-SPE cleanup step in this study, initial QuEChERS extracts were spiked (or not) with the 97 analytes at 10, 25, 50, and 100 ng mL−1 (n = 4 at each level and matrix). Each sequence began with seven blanks of each matrix, which were used for MM calibration standards, followed by the reagent blank and 16 spiked extracts of each matrix, for a total of 47 samples daily, and 235 altogether in 5 days sequentially. The LPGC-MS/MS sequences were conducted each day overnight, which included an additional 7 RO standards conducted at the beginning and end of each sequence, plus a system suitability standard as the first injection and 3 MeCN blanks injected after the 150 ng mL−1 (ng g−1) RO standards to check for carry over. A total of 65 LPGC-MS/MS analyses were conducted each night, and 325 within 5 days. The same injection liner and autotune parameters were used in each sequence.
Summation integration function start and stop times for each analyte were set after the conclusion of all 325 analyses by ensuring that all analyte peaks fell within the integration windows. The use of APs in LPGC-MS/MS led to highly consistent t
R and peak shapes for all analytes, and relatively few chemical interferences in the chromatograms for each MS/MS ion transition provided leeway to each side of the peak without significantly impacting the integrated peak areas. Figure 7 and Supplemental information show many examples of how the summation integration approach worked for different analytes and matrices. Dechlorane plus, (es)fenvalerate, permethrin, deltamethrin, cyfluthrin, and cypermethrin gave multiple peaks with the same ion transitions, several of which could have been integrated separately, but we chose to integrate them together as a single analyte. Sapozhnikova and Lehotay previously presented several examples of LPGC-MS/MS separations of different closely eluting analytes with isobaric ion transitions comparing columns from different manufacturers [17] (Figs. 4, 5, 6).
The Agilent summation function does not find the deflection point between unresolved peaks, which does not always occur at the exact same time in each chromatogram. Thus, benz(a)anthracene + chrysene and PCBs 118 + 123 and 156 + 157 could be distinguished individually by t
R, but were integrated together because they closely co-eluted. We could have also integrated phenanthrene and anthracene together, but we chose to treat them separately and apply the default Agile2 integrator for purposes of comparison. Dozens of manual re-integrations each of phenanthrene and anthracene were needed, especially at the lowest levels, to correct the mistakes by the default program when it selected the wrong peak or used an inconsistent baseline. In the case of the other 107 analytes, IS, and QC standard, only spot-checking of integrations was conducted out of curiosity and no manual adjustments were made.
The 5-day validation experiment generated 66,950 data points (ion transition peaks) that would have taken an analyst several days to check that each peak was integrated (or not) appropriately and consistently. Using the summation integration function, the analyst spent only a few hours to set parameters, and all results were generated within a matter of minutes. Not only was human review of results not needed, the consistency of the summation integrations was superior to human capabilities and led to reliable quantitative results (see Supplemental information). Chemical interferences could be observed, especially in dried basil, orange, and black olive, but such interferants complicate peak integrations and worsen quality of results for any integration function.
Different means of post-run data processing batches were considered before settling on the final approach to report results calculated with and without normalization to the IS (see Table 1) using MM calibration standards for the same matrix. Potentially better results could have been obtained by narrowing the summation integration time windows to exactly accommodate the peak widths for each matrix, which were analyzed within hours of each other, but the wider windows that fully covered all peaks for all 325 injections were used in all cases. One drawback of the current software version was that the reported analyte t
R was the midpoint between the integration start and stop times rather than the actual apex of the peak, and this should be corrected in the future.
Validation Results
Table 3 lists the compiled results with normalization to the IS for all 94 analytes (including pairs) in the 10 matrices spiked at different levels (n = 40 each), as well as overall results for all levels and matrices (n = 160). Underlined font emphasizes when overall results yielded 100 % ± 10 % recoveries, and bold text indicates when <70 or >120 % recovery and ≥25 % RSD were obtained. When using the IS, 42 out of 97 analytes gave 90–110 % recoveries and ≤10 % RSD, which is exceptional considering that 230 injections of 10 complex matrices at ultra-trace spiking levels were made without manual re-integrations or instrument maintenance over 5 days sequentially. For comparison, the overall average recoveries and RSD for each analyte without normalization to the IS are also presented in Table 3. Clearly, the IS were needed to provide better precision in the results, and to account for partial retention of the larger PAHs. For nearly all pesticides and most FRs and PCBs, the IS only helped to account for volume fluctuations in the mini-SPE cleanup step. The variability of the elution volume can be observed in Fig. 1, and Table 3 shows how well the IS compensated for these fluctuations from day-to-day and matrix-to-matrix. The differing amounts of water in the different matrices and initial and final extracts also likely contributed substantially to the variations without the IS.
Table 3 Average mini-SPE % recoveries (and %RSD) of the 94 analytes (including three pairs) in 10 matrices at four spiking levels (n = 40 at each level, and n = 160 overall) in the validation study
When using the IS, only three analytes (PBEB, benz(a)anthracene + chrysene, and thiabendazole) had <70 % overall average recoveries (64, 68, and 67 %, respectively) and only six analytes (naphthalene, TPP, TBCO, fenpropathrin, deltamethrin, and thiabendazole) had overall RSDs >25 %. TPP and naphthalene were ubiquitous in all of the reagent blanks at varying concentrations averaging 7 and 75 ng mL−1, respectively, and similar levels in the final matrix extracts. Analytical conditions for fenpropathrin were not optimal, and an interferant averaging 18 ng g−1 equivalent response adversely affected its results. Deltamethrin is known to degrade in the injection inlet [20], which was the main cause of its more variable results, which was also the case for methoxychlor to a lesser extent. Last, thiabendazole and TBCO (and TBECH) gave very broad chromatographic peaks. Despite the inconsistent results for those six analytes, the chromatographic peak shapes and results were very good for other notoriously difficult analytes in GC, such as methamidophos, acephate, omethoate, dimethoate, carbofuran, chlorothalonil, imazalil, and myclobutanil (see Supplemental information). In fact, dimethoate was among those analytes that gave the best results, and cyfluthrin was another analyte that gave surprisingly high quality results.
Despite the much greater variability, the overall recoveries without normalization to the IS were the same as when using the IS except PBEB and PCB 169 were 67 and 63 %, respectively, and 8 PAHs had 14–63 % uncompensated recoveries. As stated earlier, the 1 mg CarbonX was probably the cause for the lower recoveries of structurally (co-)planar analytes, but the isotopically labeled IS worked well to improve accuracy in all cases, albeit an ideal IS was not obtained for benz(a)anthracene + chrysene or benzo(b + k)fluoranthene.
In addition to its high quality of results, the automated mini-SPE + LPGC-MS/MS method provided excellent ruggedness in the analysis of all 325 samples. Due to the clean extracts, small injection volumes, mega-bore analytical column, and use of APs, the same injection liner and septum was used for all 325 injections in 10 matrices over the course of 5 days. A picture of the used liner and septum appears in Supplemental information. Even the fatty salmon matrix, which was analyzed last, showed very good peak shapes and perfectly linear calibration with the same high quality results as the other matrices (see Supplemental information).
Matrix Effects (MEs)
MEs were measured both with and without normalization of integrated peak areas to the IS. As in the case of recoveries, the IS helped to greatly reduce imprecision in the measurements. As shown in Supplemental information, the PCBs, PAHs, and nearly all FRs did not undergo MEs (<|±20 %|), mainly due to their nonpolar nature (relatively polar analytes are more susceptible to MEs). Figure 8 shows MEs in the case of pesticides in each matrix plotted vs. increasing t
R. When MEs <|±20 %| occur (shown by the box in the figure), then RO standards can be used to yield similar results as MM calibration standards. Despite the common practice in residue analysis to use MM standards, they require storage of many blank matrices and take more time for sample preparation, plus the extra materials, labor, and costs involved, than simply preparing RO calibration standards. Use of MM standards also causes more matrixes to be introduced to the instrument during analytical batches, as well as complications to match given matrices when different commodities are analyzed in the same sequence. Calibration using RO standards is more advantageous in practice, provided that MEs do not lead to unacceptable accuracy.
Unlike LC–MS/MS in which ion suppression occurs due to MEs [26], signal enhancement commonly occurs to susceptible analytes in GC analysis [27]. However, as shown in Fig. 8, basil induced a severe response diminishment effect, presumably due to the sheer amount of co-extractives from the dried herb even after mini-SPE cleanup. Herbs and spices are unique matrices that require MM calibration even after extensive cleanup in any GC-MS or LC–MS method. With few exceptions between 2.5 and 5.5 min t
R, MEs for the analytes fell within |±20 %| (see Fig. 8). As described previously [19], the APs can overcome MEs for the analytes with t
R less than about permethrin, but a good AP for the late-eluting pyrethroids has not been found yet. The current mixture of APs (and their breakdown products) is too volatile to co-elute with the late-eluting analytes, which is shown clearly in Fig. 8. In this study, 10 pesticides underwent >20 % MEs for all matrices except the least complex fruits and vegetables (apple, kiwi, and carrot). These analytes consist of coumaphos, pyridaben, methoxychlor, and the pyrethroids: bifenthrin, fenpropathrin, permethrin, cyfluthrin, cypermethrin, esfenvalerate, and deltamethrin). Similarly, late-eluting PCBs, PAHs, and FRs yielded insignificant MEs, especially when using IS with t
R near the analytes. Despite the nonpolar nature of these PCBs, PAHs, and FRs, MEs were observed to a greater extent without normalization to the IS, which leads us to hypothesize that use of an appropriate IS for pyrethroids may solve this problem with excessive MEs.