MULTI-CLASS DETERMINATION OF UNDESIRABLES IN AQUACULTURE SAMPLES BY GAS CHROMATOGRAPHY/TANDEM MASS SPECTROMETRY WITH ATMOSPHERIC PRESSURE CHEMICAL IONIZATION: A NOVEL APPROACH FOR POLYCYCLIC AROMATIC HYDROCARBONS

In this work, a method for the analysis of 24 PAHs in 19 different matrices, including fish tissues, feeds and feed ingredients, has been developed using gas chromatography coupled to triple quadrupole tandem mass spectrometry with atmospheric pressure chemical ionization source (GC-APCI-MS/MS). The method is based on a modification of the unbuffered QuEChERS method, using freezing as an additional clean-up step and applying a 20-fold dilution factor to the final extract. The procedure was also tested for 15 pesticides and 7 polychlorinated biphenyl (PCB) congeners in order to widen the scope of the method. The excellent sensitivity and selectivity provided by GC-APCI-MS/MS allowed the dilution of the sample extracts and quantification using calibration with standards in solvent for all the 19 matrices tested. The developed method was evaluated at 2, 5 and 50ngg-1 spiking levels. LOQs were 2ngg-1 for most compounds, and LODs ranged from 0.5 to 2ngg-1. Analysis of real-world samples revealed the presence of naphthalene, fluorene, phenanthrene, fluoranthene and pyrene at concentration levels ranging from 4.8 to 187ngg-1. No PCBs, DDTs and pesticides were found in fillets from salmon and sea bream.


INTRODUCTION
Farming of carnivorous marine fish species such as Atlantic salmon (Salmo salar) and gilthead sea bream (Sparus aurata) has traditionally relied on the use of marine feed ingredients such as fish oil and meal in their formulated feeds [1][2][3]. Limited access to fish meal and fish oil for the rapidly growing aquaculture industry has led to a need to develop more sustainable aquafeeds that rely less on marine ingredients from feral fish stocks [4,5]. Vegetable oils and proteins are the main alternative to marine feed ingredients in aquafeeds [2,5].
The development of new sustainable plant based feeds for marine fish farming introduces new challenges concerning contaminants that were previously not relevant when using traditional marine feed ingredients. Unrefined plant oils obtained from oilseeds such as soybeans, rapeseeds, olive seeds, and sunflower seeds are known to contain elevated levels of polyaromatic hydrocarbons (PAHs) [6][7][8][9][10][11]. Although fish oil also contains PAHs from environmental pollution [12], the use of plant oils causes increased PAH levels in plant-based salmon feeds compared to marine-based feeds [13].
In contrast, in a study on fish oil replacement in sea bream (Sparus aurata) feeds, plant oils did not have elevated PAH levels compared to fish oil, and it was concluded that feed PAH levels did not correlate with the amount of fish or plant oil used [14]. The genotoxic and carcinogenic "heavy" PAHs (>4-6 rings), such as benzo have received special attention with regards to food safety [15]. Studies related to plant oil PAH contamination, however, mainly focus on light (2-4 rings) PAHs such as fluoranthene, naphthalene, anthracene, phenanthrene, as they are most dominantly present in unrefined plant oils. These light PAHs are also on the US EPA (Environmental Protection Agency) list for environmental relevant PAHs but are mostly not carcinogenic and genotoxic [10,16].

The European union has set an upper limit for 4 PAHs (B[a]P +B[a]A+Chr+B[b]F) in
food products to protect the consumer safety [15]. No upper limits for PAH in feed ingredients or animal feeds currently exist. The PAH levels of raw ingredients for feeds for farmed terrestrial animals such as cow, chicken, and pig have been reported in only few studies [11] and little is known on PAH levels in feeds for farmed fin fish species [17,18], especially for feed ingredients used in novel plant based fish feed [13,14,19].
Traditionally, the determination of PAHs in solid fatty matrices have involved timeconsuming extractions prior to GC-MS with electron ionization (EI) analysis [20].
However, the sensitivity restrictions of the GC-EI-MS(/MS) systems [21,22] in addition to the inherent matrix effects make pre-concentration and time consuming clean-up steps, as gel permeation chromatography [23] or solid phase extraction [19,24], necessary. Recently, some methodologies have applied the QuEChERS procedure in aquaculture samples [25,26], reducing notably the time and cost of sample treatment.
However, despite all efforts to minimize the presence of interferences, the still remaining matrix effects make necessary to use matrix-matched calibration for quantification purposes [22]. In the last years, the application of atmospheric pressure chemical ionization (APCI) source in combination with last generation triple quadrupole instruments has made possible to notably increase sensitivity and selectivity of the determination of organic contaminants in food safety, environment and doping control fields [27][28][29][30] in selected reaction monitoring (SRM) based methods. The improved performance of GC-APCI-MS/MS is more than welcome in the aquaculture field, where the analysis of a wide diversity of "difficult" matrices (oils, plant ingredients, marine ingredients, terrestrial animal ingredients, fish, etc.) with this technique can be simplified by reducing the matrix content in the final extract by simple dilution which minimizes the matrix effect and avoids the need of using matrix-matched calibrations.
In this work, we explore the potential of the new GC-APCI-MS/MS technique for quantification and confirmation of PAHs, together with some selected pesticides and PCBs, in very different sample matrices from aquaculture field. Sensitivity of this new technique has been exploited in order to avoid the time-consuming and expensive purification steps commonly used in these complex matrices and with the aim of obtaining a cheap and efficient sample treatment. The developed methodology has been applied to the quantification of compounds that were previously found in wide-scope screening of aquaculture samples by combined use of GC&LC-QTOF MS. Up to our knowledge this is the first application of GC-APCI-MS/MS with triple quadrupole to the analysis of PAHs in samples from aquaculture field. were used for standard preparation (see structures for all the compounds in Table S1).

Reagents
Stock standard solutions (around 500 mg·L -1 ) were prepared in acetone. Two mixtures of pesticide standards (individual concentration of each pesticide around 50 mg·L -1 ) were prepared by dilution of stock individual solutions in acetone. Working standard solutions containing all compounds were prepared by dilution of mixtures with acetone (for sample fortification in GC) and hexane (GC injection). Stock standard solutions and working solutions were stored in a freezer at -20 ºC.

Samples
The sample material has been described before in detail [31]. A total of 76 samples (from 19 different matrices) were studied in this work. The list contains ingredients from different origin (plant, terrestrial animals and marine), and also feeds based on these ingredients (PAPs not included), as well as fillets of Atlantic salmon and gilthead sea bream reared on these feeds. Fish individuals were fed for, respectively, 7 and 18 months, and fillet samples were taken for analysis at the end of the exposure trial and additionally for sea bream at 8 months (commercial size). The same feed compositions were provided throughout the feeding trial. The quantification was performed on feed ingredients, feeds produced from the same feed ingredients, and fish fillets of fish fed on these feeds. The feed samples were analysed at the beginning of the trial (additionally after 8 months for sea bream) and no stability assessment was made by analyzing the feed during storage (below 7ºC for salmon and sea bream feeds).
Commercially available plant and marine feed ingredients were provided by Biomar (Tech Center, Brande, Denmark) feed producer and processed animal proteins (PAPs) from non-ruminants were provided by the European Fat Processors and Renderers Association (EFPRA). The ingredients selected represent the novelties in fish feed compositions to reduce the inclusion of fish derivatives. Fish feeds for feeding trials were based on plant feed ingredients, and not PAPs, as higher levels of PAHs were found in plant feed ingredients. The feeds were produced by Biomar under commercial aquafeed production techniques based on high-temperature extrusion processes, which potentially could affect pesticide residue levels.

Sample treatment
Samples were thawed at room temperature and 1 g was accurately weighed and transferred to 15 mL falcon tubes and spiked with 0.2 mL of SIL-IS solution of 1 µg·mL -1 . After 30 min, 2 mL of acetonitrile were added and the tube was vigorously shaken by vortex for 30 s. Then, 0.8 g of MgSO 4 were added and the tube was immediately shaken for 30 s. Subsequently, the tube was centrifuged at 1893 rcf·g for 5 min and the upper layer of the extract was transferred to a 2 mL eppendorf tube and stored for at least two hours in a freezer to precipitate proteins and fix lipids to the tube walls (freezing cleanup). Then, a QuEChERS clean-up step was carried out prior injection in the GC-system [25]. Briefly, 1 mL of the extract was carefully transferred to (1 min); 20 ºC min -1 to 230 ºC; 5 ºC min -1 to 320 ºC (6 min). The injector was operated in pulsed splitless mode (50 psi), injecting 1 µL at 280 ºC. Helium was used as carrier gas at a ramped flow mode programmed as follows: 2 mL·min −1 (23 min); a ramp of 4 mL·min −1 to 6 mL·min −1 (9 min). In the SRM method, automatic dwell time (values ranging from 3 to 63 ms) was applied in order to obtain 15 points per peak. The interface and source temperatures were set to 320 ºC and 150 ºC respectively using N 2 as auxiliary gas at 275 L·h −1 , as make-up gas at 300 mL·min −1 and as cone gas at 200 L·h −1 . The APCI corona discharge pin was operated at 1.6 µA. The ionization process occurred within an enclosed ion volume, which enabled control over the protonation/charge transfer processes. In order to work under proton transfer conditions, an uncapped vial containing 1.25 mL of water was placed in a designed holder into the APCI source door to enhance protonation. Targetlynx (a module of MassLynx) was used to handle and process the acquired data.

Study of matrix effects
Nine out of nineteen different matrices (fish fillets, feed and PAPs) were selected for the study of matrix effects and different dilution factors were tested (crude extract, 5-fold, 10-fold and 20-fold dilution). To this aim, matrix-matched calibrations were prepared for each sample matrix by taking between 25-500 μL of the "blank" sample extract (for crude extract, 500 μL were evaporated to dryness) and adding 50 μL (125 µL for the 25 ng·mL -1 level) of the corresponding standard in hexane (between 10 and 100 ng·mL -1 ), resulting in final analyte concentrations between 1 and 25 ng·mL -1 . The final volume of each calibration point was 500 µL in all cases (30% acetone, 5 % acetonitrile and 65% hexane). Matrix effect was evaluated by comparing the calibration graphs obtained with standards in solvent and in matrix (at the different dilution factors) [32]. The accepted relative error between the slopes of calibration in solvent and matrix-matched calibration was ±25%. Once the optimum dilution factor was selected, the absence of matrix effect was also tested for the remaining 10 matrices.

Validation and recovery experiments
Eleven SIL-IS were added at the initial stage of the procedure as quality control (surrogates) in order to correct for possible losses during the overall procedure and instrumental deviations. Linearity of relative response of analytes was established by analyzing standard solutions injected in triplicate in the range of 0.01-25 ng·mL -1 for all compounds except for C[c,d]P (0.005-12.5 ng·mL -1 ). Linearity was assumed when the correlation coefficient (R 2 ) was higher than 0.99 with residuals lower than 30%.
A full validation of the method was carried out for two sample matrices, salmon and sea bream fillets. Accuracy (estimated by means of recovery experiments) was evaluated by analyzing six replicates spiked at 2 ng·g -1 (1 ng·g -1 for C[c,d]P) and 50 ng·g -1 (25 ng·g -1 for C[c,d]P) for the two matrices tested. Precision was expressed as repeatability in terms of relative standard deviation (RSD, %) (n = 6) calculated for each fortification level. The limit of quantification (LOQ) objective, for salmon and sea bream fillets, was defined as the lowest concentration level validated following SANTE/11945/2015 guide criteria [33] (recovery 70-120% and RSDs below 20%). For those compounds that could not be validated due to the high concentration in the "blank" sample a statistically calculated LOQ was estimated as the analyte concentration that produced a peak signal of ten times the background noise in the chromatogram at the lowest fortification level tested for each compound. The limit of detection (LOD) was estimated for all compounds considering a signal-to-noise (S/N) ratio of three.
The methodology was validated for the remaining 17 sample matrices by analysis of quality control (QC) samples, i.e. "blank" samples spiked at 5 and 50 ng·g -1 injected in each batch. A total of 48 QCs prepared from 24 "blank" samples (corresponding to 17 different matrices) were analyzed. Recoveries between 60 and 140% for each individual sample were considered satisfactory following the acceptance criteria for routine recoveries according to SANTE/11945/2015 guide criteria [33]. The q i /Q ratio, defined as the ratio between the intensity of the confirmation ion (q i ) and the quantification ion (Q), was used to confirm peak identity in real and spiked samples.
The experimental q i /Q value for each compound was calculated as the mean value obtained from three standard solutions injected in triplicate (RSD below 15% in all cases) Confirmation of analytes detected in samples was considered positive when the q i /Q ratio was within acceptable tolerances (30% of the experimental q i /Q value calculated from standards) according to SANTE/11945/2015 guide criteria [33].

Ionization and fragmentation behaviour of PAHs in GC-APCI
In this work, the ionization behavior of the APCI interface was tested using PAH Once these modifiers were discarded, the repeatability of the response (n=10 at 10 ng·mL -1 ) was also evaluated under the remaining working conditions (without water, with water and with HCOOH 0.5%) ( Figure S1)

Selectivity of the SRM transitions for PAHs
Selectivity of the selected SRM transitions, especially in cases where coelution occurred, was also studied during the optimization process. In such cases, the possibility

Study of matrix effects
Matrix effects observed under GC-APCI-MS come from the GC liner (normally signal enhancement) and from the APCI source (normally signal suppression) [27]. As a summary of the matrix effect study for 9 out of 19 matrices (where different dilution factors were tested: crude extract, 5-fold, 10-fold and 20-fold dilution), signal enhancement/suppression was observed for: i) 36% of the combinations matrix/PAHs, 58% for matrix/pesticides and 52% for matrix/PCBs (crude extract); ii) 35% for matrix/PAHs, 46% for matrix/pesticides and 17% for matrix/PCBs (5-fold dilution factor); iii) 26% of matrix/PAHs, 33% for matrix/pesticides and 14% for matrix/PCBs (10-fold dilution factor); and iv) 11% for matrix/pesticides (20-fold dilution factor). As an illustrative example, Figure 3 shows how the matrix effect for PAHs in sea bream fillet decreases when the different dilution factors studied are applied. Thus, 20-fold dilution of the final extract was finally selected for further experiments.
As a summary, for a 20-fold dilution, signal enhancement was observed only for ethoxyquin and ethoxyquin-dimer, whereas signal suppression was observed for boscalid and azoxystrobin (mainly in plant and marine ingredients). No significant matrix effect was observed for PAHs, PCBs and the rest of pesticides (Figure S3).
According to our data, 20-fold dilution of the final sample extract eliminated matrix effect for most analyte/matrix combinations, and this allowed us to perform the quantification in the analysis of the 19 aquaculture matrices using calibration curve in solvent. The remaining matrix effects for boscalid and azoxystrobin could be corrected with the use of SIL-IS. Matrix effect for ethoxyquin and ethoxyquin-dimer could not be properly evaluated in feed, krill meal and fishmeal matrices because the analyte concentration in these samples was typically higher than the spiking levels assayed.

Analytical parameters
Validation of the analytical procedure was carried out in salmon and sea bream fillets considering the parameters of linearity, accuracy, precision, LODs and LOQs. Relative responses to the selected SIL-IS were used (see Table 1).
The study of the linearity revealed that correlation coefficients (R 2 ) were higher than 0.99, with residuals lower than 30% for most compounds in the range 0.01-25 ng·mL -1 .
Some exceptions were: 0.0125-12. Trueness and precision data for salmon and sea bream fillets are shown in Table 2 which shows satisfactory recoveries and precision for the wide majority of compounds at the two concentrations studied, 2 and 50 ng·g - and between 92 and 116% at 50 ng·g -1 . Among the studied matrices, PAPs showed higher complexity giving individual recoveries out of 70-120% range without a clear tendency for specific compounds. Data from above illustrate that the methodology applied was robust and satisfactory for the determination of very low PAH concentrations in different sample matrices, some of them of high complexity. The excellent sensitivity of GC-APCI-MS/MS allowed the extra dilution (x20) of the sample extracts, minimizing matrix effects but still reaching low LOQs, comparable and, in some cases, even lower than those reported in the literature [21,22] with the great advantage of using solvent calibration for quantification of the compounds in all matrices tested.

Analysis of samples
In order to demonstrate the applicability of the methodology developed in this paper, it was applied to the analysis of a notable number of samples from aquaculture production: 12 diets, 12 fish fillets, 19 terrestrial animal ingredients, 6 marine ingredients, 19 plant ingredients and 8 plant oils. A reagent blank and two spiked samples (5 and 50 ng·g-1) for each matrix analyzed were included in each batch. The results, in Table S2, are expressed in a wet weight basis. The q i /Q ratios obtained for all positive samples were within the range of the tolerance accepted (30%) around the experimental q i /Q value obtained from reference standards.
LMW-PAHs together with HMW-PAHs as fluoranthene and pyrene were found in most of the samples. Regarding diets, fish fillets, plant ingredients and PAPs, concentrations ranged from <5 ng·g -1 for acenaphthene to 126.6, 77.1, 88.6 and 177.5 ng·g -1 for naphthalene, respectively. In marine ingredients, they ranged from <5 ng·g -1 for acenaphthene and acenaphthylene to 129.0 ng·g -1 for phenanthrene. For plant oil ingredients, concentrations ranged from <5 ng·g -1 for acenaphthylene to 186.9 ng·g -1 for phenanthrene. In the present study only two fish oils were analysed, and clearly a large variation exist in the environmental background levels of PAHs in fish oils [13,14]. Figure S3 shows the identification of 5-MC in one fish meal. This PAH was below the LOQ in all samples analyzed. As can be seen, three transitions were acquired and q i /Q ratios were in agreement with SANTE/11945/2015 identification criteria followed in the current research [33].
Regarding pesticides, the non-organochlorine pesticides that are currently used on crop such as chlorpiriphos-methyl and pirimiphos-methyl were only detected in the novel plant ingredients and not in marine feed ingredients such as fish oil and meal [25].
Malathion, imazalil and tebuconazole were only found in some ingredients while boscalid and azoxystrobin were indeed found in feed. It is worth to mention that studied PCBs, DDTs and pesticides were not found in fillets from salmon and sea bream. where normally other antioxidant such as BHT and BHA are used [37]. All other feed ingredients, including terrestrial and plant, also contained ethoxyquin, albeit at l00-150 fold lower levels than the marine ingredients (Table S2). PCB congeners were not detected in any sample.

CONCLUSIONS
The use of APCI has been evaluated for GC-MS/MS analysis of PAHs. The high sensitivity of this technique allowed the simultaneous quantification of 19 different complex matrices (from aquaculture field) using solvent calibration. The key aspect for this aim was the elimination of matrix effect, without the need of time-consuming purification steps, only by a 20-fold dilution factor of the final QuEChERs extract, being this aspect a great contribution of the present work. Despite this, LOQs of the developed method were 2 ng·g -1 for most analytes, in the same order or better than those reported in previously published methods for similar matrices [21,22] showing higher efficiency [19]. Also, the use of both M +• or [M+H] + gave an additional value to the selectivity in the determination/identification capabilities.      Table 2. Mean recoveries (%) and precision (RSD, %) for PAHs, PCBs and pesticides after application of overall analytical procedure to salmon fillet and gilthead fillet (n = 6), LOQs objective of the method.   (not colour necessary)