Introduction

The compound feed industry reached an estimated worldwide production of 954 million tonnes in 2012. The European Union (EU) is one of the leading compound feed manufacturers and accounts for 20 % of global production [1]. The potential for the rapid spread of diseases among livestock has made the routine use of veterinary drugs integral to maintain the viability of the growing animal food industry. Oral drug administration via feed is the most common practice [2]. Hence, about 3–7 % of the feeds that are produced correspond to medicated feeds.

Antimicrobials are by far the largest authorized group in the production of medicated premixes. Tetracyclines are the most commonly used antimicrobials, followed by sulfonamides, trimethoprim and macrolides [3]. However, the manufacture of medicated feeds in the same production lines as non-medicated ones has raised concerns about unavoidable cross-contamination, which may occur in all stages of feed production and processing, as well as during storage and transport [4].

Contaminated feeds may be given unknowingly to non-target animals. This could harm certain species and lead to residues in end products for human consumption, which may cause serious public health problems (e.g. the development and transfer of resistant bacterial strains, and the onset of allergies in individuals with hypersensitivity) [5].

In order to ensure a high level of human and animal protection, the EU adopted harmonized community rules on this matter. Following the As Low As Reasonably Achievable principle, Directive 2009/8/EC [4] established maximum limits for unavoidable carry-over of coccidiostats and histomonostats, which are used as feed additives. However, this regulation does not give legal cover to the carry-over of other authorized antimicrobials prescribed in medicated feeds. Zero-tolerance levels should therefore be applied to antimicrobials other than coccidiostats and histomonostats in non-target feeds.

Notifications of the improper presence of residues of antimicrobials in animal feeds can be found in the Rapid Alert System for Food and Feed portal database [6]. Since 2000, 51 notifications about feeds have been listed. Most of them refer to the presence of coccidiostats and, to a lesser extent, tetracyclines and other antimicrobials (e.g. sulfadiazine, amoxicillin, colistin, chloramphenicol and bacitracin). Thus, the availability of multiclass analytical methods to monitor a variety of antimicrobial families is of great interest for laboratories involved in the official control of feeds.

In the last few years, the use of multiresidue and especially multiclass methods for the analysis of antimicrobials in foodstuff matrices has become an emerging analytical trend. However, this approach has not been extensively applied in the analysis of animal feeds.

Most of the methods reported in this field in the last 10 years, compiled in a previous review [7], have ranged from single-analyte to multiresidue methods dealing with a few analytes from the same or, at most, three different classes of antimicrobials. Real multiclass approaches are still scarce in the literature [8–10].

Moreover, few methods have reported the use of mass spectrometry in feed analysis. To date, only one such method makes use of high-resolution mass spectrometry instruments [11]. Matrix effects caused by coeluting interfering matrix components, which reduce (ion suppression) or increase (ion enhancement) the analyte response, are the main problems to overcome. In the case of feed analysis, these phenomena were found to take place extensively in comparison with other kinds of matrices (maize, honey, meat, eggs or milk) [8]. However, matrix effects often do not get the attention they deserve [12]. Due to the high complexity and variability of the feed matrix composition, ignoring these detrimental effects may adversely affect the accuracy and reliability of a method [7].

The aim of the present study was to develop a multiclass LC-MS/MS method, based on a triple quadrupole, for the simultaneous analysis of 50 antimicrobials in animal feeds. The procedure was designed to cover the 12 methods that were in use at the Laboratori Agroalimentari of the Generalitat de Catalunya (LAC) for the official controls of feeds and also to include 25 new compounds. The target compounds, which included amphenicols (2), benzimidazoles (3), coccidiostats (9), diaminopyrimidines (1), lincosamides (1), macrolides (3), nitrofurans (2), pleuromutilins (2), polypeptides (1), quinolones (9), quinoxalines (2), sulfonamides (11) and tetracyclines (4), as well as their abbreviations, are listed in Table 1. Various clean-up strategies, based on either solid-phase extraction (SPE) or dispersive solid-phase extraction (d-SPE), were assessed and compared using different kinds of feeds. Special focus was placed on the evaluation of matrix effects, both during the development of the method and the validation studies. Matrix effects are comprehensively reported in the paper. The proposed approach was finally applied to the analysis of real feed samples.

Table 1 Retention time and MS/MS parameters for all analytes and surrogate standards
Full size table

Experimental

Chemicals and reagents

Methanol (HPLC grade), formic acid (98 %) and anhydrous disodium hydrogen phosphate (Na2HPO4) were supplied by Panreac (Barcelona, Spain). Acetonitrile (HPLC grade) and sodium hydroxide 1 mol L−1 (NaOH) were obtained from J. T. Baker (Deventer, Holland). Dimethylsulfoxide (DMSO) and citric acid monohydrate were purchased from Carlo Erba (Val de Reuil, France) and ethylenediaminetetraacetic acid disodium salt (EDTA) from Merck (Darmstadt, Germany). Calcium chloride dihydrate (CaCl2) was supplied by Scharlau (Barcelona, Spain). All chemicals were of analytical grade unless stated otherwise. Double-deionized water of 18.2 MΩ cm−1 resistivity was obtained with a Milli-Q water purification system (Millipore, Molsheim, France).

Oasis HLB cartridges (200 mg, 6 mL, and 60 mg, 3 mL), used for solid-phase extraction, were purchased from Waters (Milford, MA, USA). Bondesil PSA and C18 bulk sorbents were provided by Agilent Technologies (Palo Alto, CA, USA).

The McIlvaine buffer (0.1 mol L−1, pH 4.6) containing EDTA was prepared by mixing solutions of Na2HPO4 (0.2 mol L−1), citric acid (0.1 mol L−1) and EDTA (0.1 mol L−1) in suitable proportions (46/52.5/15, v/v/v). The final pH was measured in order to check that the desired pH was attained.

Standard solutions

Veterinary drug standards, listed in Table 1, were of the highest purity available and were purchased from several commercial suppliers. Six compounds, three of them isotopically labelled, were used as surrogate standards and were added at the beginning of the analytical process: SDD-13C (for sulfonamides), NOR-d5 (for quinolones), ROX (for macrolides), DMC (for tetracyclines), RDZ-d3 (for non-ionophore coccidiostats) and NIG (for ionophore coccidiostats).

Individual stock standard solutions (1,000 mg L−1) of all compounds were prepared by weighing the appropriate amount of analyte and dissolving it in the required weight of solvent. Depending on the specific solubility properties, compounds were generally dissolved in pure methanol, with the following exceptions: FEN, FLUB, OXI, CLOP, BAC, NCZ, FTD, DCZ, CBX and NIF were dissolved in DMSO, CIP in formic acid (0.05 mol L−1) and NOR and OXO in NaOH (0.1 mol L−1). When necessary, sonication was used to ensure complete dissolution.

Six intermediate standard solutions (100 mg L−1) containing several analytes, grouped according to their sensitivity (group A, B or C as listed in Table 1), were prepared by dilution of the concentrated stock solutions with methanol. Intermediate solutions were kept at −20 ºC for up to 6 months, while stock solutions were kept at −20 ºC for up to 1 year, both in dark glass bottles.

A working standard solution was prepared daily in methanol by dilution of the intermediate solutions at suitable concentrations (2, 8 and 20 mg L−1 for groups A, B and C, respectively).

A mixed surrogate standard solution was prepared by dilution of the stock solutions with methanol at appropriate concentrations (10 mg L−1 for SDD-13C, ROX and NIG; 40 mg L−1 for DMC and 100 mg L−1 for NOR-d5 and RDZ-d3). This solution was also stored at −20 ºC in dark glass bottles for up to 1 month.

Apparatus

Extraction was performed with a JP Selecta 3000515 ultrasonic bath from Afora (Barcelona, Spain). A Sorvall RC-5B plus centrifuge (Newtown, CT, USA) and a vortex mixer SA8 from Stuart (Stone, Staffordshire, UK) were also used in the sample treatment. A turboVap LV evaporation system from Caliper (Hopkinton, MA, USA) was used for the evaporation of the extracts. Solid-phase extraction was carried out with a Visiprep solid-phase extraction unit from Supelco (Bellefonte, PA, USA). The pH was measured using a Crison GLP21 pH meter (Alella, Spain), equipped with a Crison 52-02 Ag/AgCl combined glass electrode.

Samples and pretreatment

Feed samples from farms and feed mills that were received by the LAC for routine analysis were used in this study. We focused on pig, poultry and cattle feeds, as more than 90 % of the total EU compound feed production is intended for these three animal species [13].

Samples were checked by LC-MS/MS to ensure that they were free of any of the target antimicrobials. Blank feeds were used for the development of the method and for calibration and validation purposes. Feeds were minced, homogenized and stored at room temperature until analysis.

Spiked feeds were prepared by adding the appropriate amount (range 50–500 μL) of the working standard solution to each portion of the weighed samples. Samples were vortex-mixed before the extraction.

Sample preparation

Extraction was based on a previously reported procedure [10]. Feed samples weighing 4 g, placed in a 50-mL polypropylene centrifuge tube, were spiked with analytes (if required) and with 50 μL of the mixed surrogate standard solution. Then, 15 mL of the extraction solution, consisting of methanol/acetonitrile/McIlvaine buffer (37.5/37.5/25, v/v/v), was added. The mixture was manually shaken for 30 s and placed in an ultrasonic bath for 15 min. In contrast to the method proposed by Boscher et al. [10], no sample clean-up was performed.

After centrifugation at 3,000 rpm for 10 min, 3 mL of the supernatant was simply diluted to 10 mL with a 5 mmol L−1 formic acid/methanol (50/50, v/v) mixture in a volumetric flask. Finally, the extracts were filtered directly into vials through 0.22-μm membrane filters and injected into the LC system.

LC instrumentation and conditions

An Acquity UPLC system from Waters (Milford, MA, USA) was used. Chromatographic separation was achieved on a Kinetex XB-C18 column (100 mm × 2.1 mm; 1.7 μm particle size) with an installed pre-filter, both from Phenomenex (Torrance, CA, USA). The column was maintained at 35 °C, and the injection volume was set to 8 μL. Mobile phase A consisted of 5 mmol L−1 aqueous formic acid, while mobile phase B was 50 mmol L−1 aqueous formic acid/acetonitrile (10/90, v/v). Mobile phases were filtered through a 0.2-μm membrane filter unit prior to usage. The mobile phase flow rate was 0.5 mL min−1 and was directed to the mass spectrometer without splitting. The following linear binary solvent gradient was applied: 0–1 min 0 % B; 1–1.5 min 0–15 % B; 1.5–8.5 min 15–70 % B; 8.5–9 min 70–100 % B; 9–12 min 100 % B; 12–14 min 100–0 % B and finally, 14–16 min 0 % B.

MS/MS instrumentation and parameters

The instrument consisted of a Quattro Premier triple quadrupole mass spectrometer from Micromass (Waters, Milford, MA, USA), equipped with an ESI source. It was operated in positive and negative ion modes under the following working conditions: capillary voltage of 3.5 and −3.0 kV, respectively; source temperature of 120 °C; desolvation temperature of 300 °C; cone and desolvation gas (N2) flow rates of 50 and 500 L h−1, respectively; and gas (Ar) pressure in the collision cell of 7 × 10−3 mbar. Instrument control and data acquisition under time scheduled multiple reaction monitoring (MRM) conditions were achieved using MassLynx 4.0 software. Table 1 shows the quantification and confirmation transitions selected, as well as the retention time (t R), ESI mode (+/−), cone voltages (C.V.), collision energies (C.E.) and dwell times (d.t.) optimized for each compound.

Results and discussion

Sample preparation

Due to the different physicochemical properties of the large number of families considered, the extraction and clean-up procedures are the most challenging steps in a multiclass method. A sample treatment method, previously reported by Boscher et al. [10] for the analysis of 33 antimicrobials in feed (24 in common with our method), was taken as the starting point of this study. The extraction was performed successfully and consisted of a mixture of organic solvents and a McIlvaine buffer, containing EDTA, combined with sonication. Addition of EDTA was required to properly extract tetracyclines and macrolides, which have a strong tendency to form chelates with divalent metallic cations [14]. For the clean-up, two previously proposed approaches [10] based on d-SPE (method I) and SPE (method II) were initially assessed. Both strategies were compared with a generic SPE protocol (method III), reported by Waters [15]. The three methods are summarized in Table S1 (Electronic Supplementary Material).

It is well known that when ESI is used, the presence of matrix components can affect the ionization of the target compounds, decreasing or enhancing the analyte response. Hence, the effectiveness of the clean-up procedures was evaluated both in terms of recovery values and reduction of the ion suppression/enhancement on the resulting MS/MS signals.

Preliminary experiments were carried out in duplicate with pig feed spiked with all analytes at a concentration of 2 μg g−1. Mean recoveries (in percent) for each antimicrobial class, presented in Fig. S1 (Electronic Supplementary Material), were calculated against blank feed extracts spiked at the expected concentrations prior to injection. Benzimidazoles, coccidiostats, pleuromutilins and BAC were insufficiently recovered, or not recovered at all, when SPE cartridges (method II and III) were used. Moreover, method II resulted in low recoveries for the other families, and although a significant improvement was attained by applying the generic SPE protocol (method III), the recovery rates of some families of compounds were still low compared with those obtained using PSA (method I). The results were in accordance with those reported by Boscher et al. [10]. Therefore, d-SPE was selected as the clean-up method. However, in further experiments, the last tedious and time-consuming steps of method I (e.g. evaporation, transfer, dilution and refrigeration) were replaced by dilution of the extract (3 to 10 mL). In addition, different d-SPE sorbents (C18 and its combination with PSA) were tested to improve recoveries. As shown in Fig. S2 (Electronic Supplementary Material), the use of the C18 sorbent resulted in poor efficiency for ionophore coccidiostats (<30 % for MON, NAR, SAL and LAS), whereas with PSA, the lowest recoveries were obtained for tetracyclines (<65 %). The mixture of these sorbents proved to be an acceptable compromise for tetracyclines (>75 %), but ionophore losses were still considerable (recoveries <60 %).

At the same time, and in order to assess possible matrix effects, responses from matrix-matched standards were compared with responses from an identical analyte concentration in solvent (5 mmol L−1 formic acid/methanol; 50/50, v/v). Results revealed that, although d-SPE provided satisfactory recoveries, it was unable to remove coeluting matrix interferences, which caused intensive signal suppression/enhancement.

Several attempts were then made to overcome this problem. Different factors related to d-SPE and their combinations (see Table 2) were studied. Unfortunately, matrix effects were not reduced by using either a higher amount of sorbent or a smaller volume of extract in the d-SPE clean-up. Removal of fatty matrix components by precipitation via overnight refrigeration of the extract also led to unsatisfactory results. Matrix effects for the combination of C18 and PSA plus dilution of pig feed extracts spiked at 2 μg g−1 are depicted in Fig. 1. Some compounds, such as TRI and AMP or NIF and BAC, still showed extensive suppression or enhancement effects, respectively. Acidic quinolones (OXO and FLU) behaved in a different way, with significant signal enhancement in comparison with piperazinyl quinolones, which were slightly suppressed. However, matrix effects proved to be of little relevance for sulfonamides and coccidiostats (except for AMP). Figure 1 shows that similar matrix effects were observed for most of the studied compounds, whether the dilution was carried out with d-SPE or without it. As a consequence, d-SPE was abandoned. No further clean-up approaches were considered, since a more selective strategy would likely result in the loss of some families of analytes as well. Therefore, extracts were simply diluted before injection, without additional treatments. The sensitivity achieved by this process was sufficient to determine the analytes at the concentration levels of interest. As shown in Table 3, extraction recoveries were satisfactory (above 70 %) for most compounds and matrices tested, and despite some exceptions (CAP in cattle feed or DCZ and TRI in poultry feed) similar recoveries were also obtained for each analyte among the three investigated feeds.

Table 2 d-SPE clean-up procedure: tested factors
Full size table
Fig. 1
figure 1

Matrix effect for each antimicrobial in pig feed extracts, spiked at 2 μg g−1, after two different clean-up procedures. A value of 0 % means no matrix effect. Negative values stand for suppression of the analyte signal, whereas positive values point to matrix-induced signal enhancement

Full size image
Table 3 Extraction recovery rates from spiked feeds at 2 μg g−1
Full size table

Finally, the variability of the final matrix effects between different animal feeds was evaluated. Three matrix-matched calibration curves (n = 5, from 25 to 225 μg L−1) were performed by spiking extracts from pig, cattle and poultry blank feed samples. The slopes were graphically compared for each analyte, and no significant differences between matrices were observed. Figure S3 (Electronic Supplementary Material) shows the calibration curves for some of the studied compounds. The conclusion was that, at least in principle, matrix-fortified standard curves (blank feed samples spiked before extraction) prepared from a single feed could be used to compensate for losses in sample preparation and the matrix-related effects of any kind of feed. However, matrix effect studies during the method validation pointed to a feed-dependent effect for some of the target compounds, as presented in “Method validation”.

LC-MS/MS method

The liquid chromatography conditions were optimized by the injection of a mixed standard solution of all compounds. Two columns were tested in order to obtain a satisfactory separation. The Kinetex XB-C18 core–shell column was chosen as it provided better resolution and sensitivity than an Acquity UPLC BEH C18 column of the same length, diameter and particle size. A wide range of gradient profiles were also studied, with initial mobile phase percentages of the organic component varying between 0 and 20 %. The content of organic solvent was raised to 100 % during the gradient profile to prevent column contamination due to matrix compounds. The final selected gradient allowed the elution of all analytes to be achieved within a significantly short time of 16 min, including the cleaning and preconditioning steps. Despite the large number of targeted antimicrobials, complete separation of compounds with common mass transitions (FLU/OXO and SDX/SDM) was achieved.

The mass spectrometer parameters were optimized by the injection of individual standard solutions prepared in methanol. All analytes were ionized in positive mode, with the exception of DCZ, NCZ, CAP, FLOR and NIF, which were detected in the negative mode. The sodium adduct [M + Na]+ was selected as the precursor ion for ionophore coccidiostats (MON, NAR, SAL and LAS), whereas the protonated or deprotonated molecular ions, [M + H]+ or [M − H]−, were selected for the other compounds. The product ion spectra were then recorded at different values of collision energy to find the two most intense transitions for each analyte.

Data for quantification and confirmation were acquired in the MRM mode, in which the total ion current acquisition was split into 14 retention time-based windows to achieve better sensitivity. Two transitions were monitored for identification, although only one was used for quantification. Identification was based on retention time, while confirmation was performed according to the ion ratio criteria in Decision 2002/657/EC [16].

Figure S4 (Electronic Supplementary Material) shows a typical chromatogram corresponding to the lowest concentration level of a matrix-matched calibration curve prepared in pig feed.

Method validation

There is still no widespread agreement about whether European Commission Decision 2002/657/EC [16] should be applied to feed analysis. However, because it is extensively used in food analysis, some of its criteria were applied in the validation of the present method, together with requirements for official feed controls stated in Regulation (EC) 882/2004 [17] and Regulation (EC) 152/2009 [18].

The protocol included the processing of blank samples from pig, poultry and cattle feeds. Due to the high variability of the feed matrix composition, new feeds (not used during the development of the method) were included in the validation assays.

Depending on their sensitivity, the investigated compounds were divided into three groups. Two concentration levels were selected for the validation procedure of each group. The lowest level was chosen around the limit of quantification (LOQ), whereas the second level was four times higher to cover the relevant quantification concentrations for the different compounds. Hence, validation levels were 0.05 and 0.2 μg g−1 (group A), 0.2 and 0.8 μg g−1 (group B) and 0.5 and 2 μg g−1 (group C) for analytes with high, intermediate and low sensitivity, respectively. The distribution of compounds among the three groups is given in Table 1.

The validation parameters that were measured were specificity, linearity, trueness, precision (repeatability and within-laboratory reproducibility), limits of detection (LOD) and quantification (LOQ), decision limit (CCα) and detection capability (CCβ). The validation results are summarized in Table 4.

Table 4 Validation data for the three feed matrices
Full size table

Specificity/selectivity

For the specificity assessment, 24 representative blank samples (including ten pig feeds, seven cattle feeds and seven poultry feeds) were analysed and checked for endogenous peaks with a signal-to-noise ratio above 3 at the retention times of the compounds of interest. Feeds were carefully selected to account for the high variability and complexity of the matrix composition, which depends on factors such as animal species (pigs, cattle, poultry), age (e.g. pigs/piglets), gender (e.g. hens/chickens) or regimes (e.g. growing, fattening, pregnancy, lactation, etc.). The method provided clean, background-free mass traces for all analytes in the 24 matrices studied.

The method also meets the criteria for compound identification. It scores four identification points through the measurement of a precursor ion plus two product ions. The ion ratio for each analyte in the samples matched the ion ratio for the standards within each run, since differences between the calculated ratios were lower than the maximum permitted tolerances [16].

Linearity

Quantitative analysis was based on matrix-fortified standards. Blank feed samples (n = 6) were spiked in the range of 0.025–0.25 μg g−1 (group A), 0.1–1 μg g−1 (group B) and 0.25–2.5 μg g−1 (group C) and subjected to the entire analytical procedure. Least-square regression analysis was performed by plotting the peak area versus the analyte concentration. The calibration parameters showed good linearity, with correlation coefficients (r) higher than 0.99 for all the analytes in the three feed matrices. Surrogate standards were used for internal quality control of the method performance (recovery and sensitivity).

Accuracy

Trueness and precision were determined by processing five validation series. For pig feed, three independent series were analysed on three different days to account for within- and between-day variations, while only single-day experiments were performed for cattle and poultry feed. Each series consisted of one non-spiked matrix sample, a six-point calibration curve and six replicates of blank samples spiked at the two validation levels. This resulted in 19 individual extractions, dilutions and injections per analytical run.

Trueness, expressed at each validation level as the relative difference between the mean measured and nominal concentrations (in percent), was lower than 15 % for most analytes. It increased in only a few cases up to 22 %, especially in cattle feed matrix.

Repeatability (RSDr) and within-laboratory reproducibility (RSDR) values were calculated as the relative standard deviation of six replicate measurements for each concentration level on the same day and on three different days, respectively. RSDr values below 15 % were obtained in most cases, whereas RSDR ranged between 10 and 20 %.

Sensitivity

The LOD and LOQ were determined as the concentration of the analyte at which the signal-to-noise (S/N) ratio was equal to 3 and 10, respectively. LOD values were in the range of 0.1–71.4 ng g−1, whereas LOQ values were from 0.2 to 238.1 ng g−1(see Table 4). For all compounds, it was experimentally verified that peaks with S/N ≥10 were obtained from both MS transitions at the LOQ.

The limits obtained for AMP, CBX and OLQ were more than five times lower than those reported for the LC-UV analytical methods described in Regulation (EC) 152/2009 [18] to control the illegal presence of such non-authorized additives in feed.

Finally, the decision limit (CCα) and the detection capability (CCβ) values were also determined in pig feed, as stated in Decision 2002/657/EC [16]. According to Directive 2009/8/EC [4], only maximum limits for unavoidable cross-contamination of coccidiostats (except AMP and CLOP) and histomonostats have been established. After the application of good manufacturing practices, carry-over rates of approximately 1 and 3 % of the authorized maximum content can be tolerated for sensitive and less sensitive non-target animal species, respectively. Thus, the more restrictive 1 % was taken as the maximum permitted level for these compounds (LAS, NAR, SAL, MON, DCQ, NCZ and DCZ). Target analytes that do not belong to these two classes of antimicrobials, for which a maximum permitted level has not been defined, were considered to be non-authorized drugs.

CCα, defined as the concentration at and above which it can be concluded that a sample is non-compliant with an error probability of α, was determined using the calibration curve procedure. For substances with an established permitted limit (coccidiostats), CCα was calculated as the concentration at the permitted limit plus 1.64 times the within-laboratory reproducibility standard deviation. For banned compounds, CCα was calculated as the concentration at the y-intercept plus 2.33 times its standard deviation under within-laboratory reproducibility conditions.

CCβ is defined as the smallest content of the substance that may be detected, identified and/or quantified in a sample with an error probability of β. In the case of substances with an established permitted limit, the detection capability is the concentration at which the method is able to detect permitted limit concentrations with a statistical certainty of 1 − β. For banned compounds, CCβ is the minimum concentration at which the method can detect truly contaminated samples with an error probability of β. It is usually calculated from the signal of CCα plus 1.64 times the within-laboratory reproducibility standard deviation at this level. In the present paper, the within-laboratory reproducibility standard deviation obtained at the lowest validation level was used to calculate CCβ. CCα and CCβ values were checked experimentally and are summarized in Table 4.

A comparison with the maximum permitted levels for coccidiostats indicated that the method is sensitive enough to prove the possible illicit presence of these drugs in animal feeds.

Matrix effects

From the development of the method, the preliminary conclusion on matrix effects was that matrix-fortified standard curves (blank feed samples spiked before extraction) prepared from a single feed could be used to compensate for the matrix-related effects of any kind of feed. However, a comparison of the sets of calibration curves obtained in the accuracy study revealed significant differences in the slopes among feeds. Thus, unacceptable errors were obtained when cattle and poultry feeds were quantified against the pig feed matrix-fortified calibration curve instead of the curve prepared in the corresponding matrix. Consequently, the initial assumption that a single feed could be suitable for quantifying other feeds, even those intended for other animal species, was questioned. To address this issue, the variability in the signal due to matrix effects within and between animal feed matrices was evaluated in depth. Three different feeds for each species (three for pigs, three for cattle and three for poultry) were spiked at 2 μg g−1, and the relative standard deviation (RSD) of the measured peak areas was calculated for each feed matrix combination (e.g. pig/pig, pig/cattle, pig/poultry, cattle/cattle, cattle/poultry, etc.). The results of this study are summarized in Table 4. For some compounds, results are also depicted in Fig. S5 (Electronic Supplementary Material). Acceptable intra- and inter-species differences (<20 %) were obtained for 33 of the 50 target compounds. For another seven antimicrobials, variations of over 20 % were only noticed inter-species. The ten remaining analytes showed differences greater than 20 % both within and between animal feed species. Moreover, it is interesting to note that for some compounds strongly affected by matrix effects (OXO, FLU and NIF), relatively low variability among different feeds was obtained (<10 %). Conversely, other compounds (TRI and AMP) exhibited both significant matrix effects and high variability both intra- and inter-species (>20 %). A third situation, less critical, was the case of antimicrobials (e.g. NAR, OLQ and LIN) showing a strong influence of the feed nature but with low matrix effects. Thus, it can be concluded that the variable composition of some individual feeds, even within those intended for the same animal species, can lead to sample extracts with high variations in matrix components and thus to different extents of ion suppression/enhancement for some of the target analytes.

Application to real samples

To evaluate the applicability of the method, 21 feed samples (nine for pigs, six for cattle and six for poultry) were analysed, which were received by the LAC from farms and feed mills. Results revealed that only three samples were free of any of the target compounds, whereas nine feeds were found to contain at least five antimicrobials, the most frequent being TIA, TYL and DC. Moreover, TRI, whose combination with sulfonamides results in a synergistic antibacterial effect, was detected in all feeds containing SDZ. All samples contaminated with coccidiostats were found to be compliant with the maximum permitted limits established by Directive 2009/8/EC. The antimicrobials detected, as well as their concentration levels, are summarized in Table S2 (Electronic Supplementary Material).

Conclusions

A fast and easy LC-MS/MS multiclass method was developed for the analysis of 50 antimicrobials in animal feeds. Dealing with the extensive matrix-related signal suppression or enhancement proved to be quite a challenge, since none of the clean-up strategies tested (SPE or d-SPE) minimized such detrimental effects. Thus, no clean-up was performed. For most of the target analytes, low matrix effects or no differences in matrix effects, either intra- or inter-species, were observed. In such cases, matrix-fortified calibration curves prepared from a single feed can be used for the quantification of any kind of feed samples. In contrast, for the few compounds that showed significant matrix effects with high intra- and inter-species variation, the more tedious, but accurate, standard addition method must be applied to ensure correct quantification.

The proposed method presents several advantages over previously reported multiclass approaches. A significant reduction in time and cost of analysis is achieved by avoiding a clean-up step, which allows for high sample throughput in routine laboratories involved in official controls. Moreover, a special effort was made to adjust the acquisition time windows, to allow the quantification of both positively and negatively ionized compounds in a single injection. Lastly, a limited survey of real feeds demonstrated that positive feed samples were mainly contaminated by more than one class of antimicrobials. This highlights the usefulness of a multiclass approach to ensure the required quality of feeds reaching the market.