Determination of Chloramphenicol in Honey Using Salting-Out Assisted Liquid-Liquid Extraction Coupled with Liquid Chromatography-Tandem Mass Spectrometry and Validation According to 2002/657 European Commission Decision

Honey is a natural food widely consumed due to its high content in nutrients and bioactive substances. In order to prevent hive infections, xenobiotics such as pesticides and antibiotics are commonly used. Chloramphenicol (CAP) is a broad-spectrum antibiotic used to treat honeybee larvae diseases. However, CAP has toxic and nondose-dependent effects in sensitive subjects; for this reason, its use has been prohibited in food-producing animals, such as the honeybee. In this study, we proposed a rapid, simple, and cheap analytical method, based on salting-out assisted liquid-liquid extraction coupled with UHPLC MS/MS detection for the accurate determination of CAP in honey to be used in routine analyses. The parameters that influence the extraction efficiency have been optimized using an experimental design in order to maximize the recovery of the analyte by reducing the matrix effects. Therefore, the developed method was internally validated according to the 2002/657/EC Decision guidelines and applied to the analysis of 96 honey samples.


Introduction
Honey is the natural sweet substance produced by honeybee (Apis mellifera) from the nectar of plants. The chemical composition of honey is quite variable, as it depends on both the floral source from which the bees derive the nectar and on other factors, such as seasonal and environmental changes, as well as the manufacturing processes. Honey is a functional food of wide consumption particularly rich in nutritional and bioactive substances. It contains about 200 substances, including mainly sugars (75-80%), water (16-18%), and other substances such as proteins, organic acids, vitamins, minerals, pigments, phenolic compounds, and large quantities of volatile substances [1,2]. Despite its biological and therapeutic activities, honey is not free of contaminants, mainly antibiotics, whose use in beekeeping is necessary for the treatment of bacterial infections affecting hives [3]. The most common infectious diseases affecting honeybee larvae are American and European foulbrood. American foulbrood is an Table 1. Analysis of variance (ANOVA) on each factor affecting dependent variable of interest. In the ANOVA table, p-value less than 0.05 indicate the statistical significance of an effect at 95% confidence level. Furthermore, as can be clearly seen, the recovered volume of the extraction solvent (ACN) was influenced by both solvent extraction volume and honey dilution volume (p-value < 0.05) while ER and nME were statistically influenced by pH and NaCl percentage of honey-diluted solution, respectively. Regarding the interaction among the four variables, no significant interactions were found (p-value > 0.05). Therefore, their p-values were not reported in the Table 1. The main effects' plot ( Figure 1) shows that the slope of the line is proportional to the size of the effect and that the direction of the line has a positive or negative influence of each effect on each variable. As shown in the main effect plot of ER (Figure 1a), the NaCl percentage in honey solution linearly increases the recovery of CAP, while the other three effects play a slight influence on the variable. The main effect plot of nME ( Figure 1b) shows that pH has a strong influence because the suppression of CAP electro spray ionization source increases when acidic or neutral pH values were reached. This behavior may be attributed to the presence of ionizable interferences into the matrix, which could be extracted in a larger amount at low pH values. As a confirmation, after the evaporation of the extraction solvent deriving from the analysis of samples at low pH values (<7), a clearly visible yellow residue was observed. As far as the recovered SV is concerned (Figure 1c), a linear increase was observed when the ACN volume was increased, while a low amount of supernatant volume was obtained when HDV was increased.
Molecules 2020, 25, x FOR PEER REVIEW 4 of 13 recovery of CAP, while the other three effects play a slight influence on the variable. The main effect plot of nME ( Figure 1b) shows that pH has a strong influence because the suppression of CAP electro spray ionization source increases when acidic or neutral pH values were reached. This behavior may be attributed to the presence of ionizable interferences into the matrix, which could be extracted in a larger amount at low pH values. As a confirmation, after the evaporation of the extraction solvent deriving from the analysis of samples at low pH values (<7), a clearly visible yellow residue was observed. As far as the recovered SV is concerned (Figure 1c), a linear increase was observed when the ACN volume was increased, while a low amount of supernatant volume was obtained when HDV was increased. Finally, chemometric analysis indicated that the best SALLE conditions that can eliminate the matrix effect, reduce the supernatant volume, and maximize the extraction recovery are the following: 1 g of honey diluted in 5 mL of aqueous solution with a pH value of 12 and a NaCl percentage of 25%, and a volume of 2 mL of ACN as extraction solvent (degree of desirability 79%). Before using these conditions throughout the validation procedure, preliminary evaluations of matrix effect and recoveries were carried out to verify the extraction efficiency and the effective elimination of the MS signal suppression. Extraction efficiency was assessed by calculating the recoveries of fortified honey samples (acacia, chestnuts, and citrus) at 0.3 µg/kg. The recoveries were calculated by interpolating the peak areas of 35 Cl-CAP ion transitions (m/z precursor/product 321→257 and 321→194) in the calibration curves in solvent of the respective ions. As shown in Figure S1, good recoveries (>90%) were obtained for the three analyzed honey samples. The signal suppression (or enhancement) phenomenon, deriving from the co-eluted matrix interferences, was assessed by extracting three blank samples from different botanical origins in the optimized conditions and spiking them with CAP WSs before the evaporation of the ACN volume. CAP concentration levels of 0.15, 0.3, and 0.6 µg kg −1 were compared to standard working solutions at the same levels. The peak area response for both the MRM transitions of post-spiked extraction samples were comparable to those obtained in solvent (H 2 O/MeOH 70:30 v/v); this confirmed the absence of matrix effects ( Figure S2).

Validation Protocol
The analytical methods developed to monitor residues of certain substances in food of animal origin must comply with the guidelines provided by 2002/657/CE Decision. The main purpose of this regulation, implementing the 96/23/EC Council Directive concerning the performance of analytical methods and the interpretation of the results, is to detect the illegal use of substances not allowed in the production of food from animal origin as well as to detect the improper use of unauthorized veterinary medicines in the same food products. The 96/23/EC Directive established measures that require European Member States to monitor the substances and groups of residues listed in Annex I of the Directive. These substances are grouped in two main categories: (1) Allowed substances, for which a maximum residue limit (MRL) was established, and (2) prohibited substances, for which no MRL could be established. For the last one, the European Commission set the minimum required performance limit (MRPL) for CAP at 0.3 µg/kg. Taking these assumptions into account, the proposed quantitative confirmatory method was validated according to the EC Directive.

Liquid Chromatography and Mass Spectrometry Parameters
According to the 2002/657/EC Decision guidelines, liquid chromatography techniques coupled with mass spectrometry detection (LC-MS) are considered adequate for quantitative analysis of organic contaminants in food for human consumption, as long as a sufficient number of identification points (IPs) were met. However, at least four identification points are required for substances belonging to group 2 (prohibited substances) of Commission Regulation 2010/37/UE, in which CAP is included. In our case, this requirement was fulfilled by coupling liquid chromatography-tandem mass spectrometry (LC-MS/MS) and by selecting one precursor ion and two product ions (321→257; 194). The two diagnostic ions were selected during the optimization of MS and MS/MS parameters by infusing CAP standard solution at 5 µg/mL in H 2 O/MeOH (50:50 v/v). Mass spectrum of ESI in negative mode shows the typical isotopic pattern of CAP due to the two Cl-atoms in the molecule ( Figure S3). According to theoretical CAP spectrum ( Figure S3b abundance of diagnostic ions (with a relative abundance >50% of the base peak) must not differ by more than ±20% from those observed in the QC. The result of tolerance ion ratio (CV IR ) for selected ions 194/257 in different honey matrices satisfies the permitted tolerance given by 2002/657/EC guidelines ( Table 2). After the selection of the diagnostic ions, the LC-MS method was developed; the total time of the analysis was 12 min with an analyte retention time of 3.49 ± 0.4 ( Figure 2a). This value was in agreement with the 2002/657/EC guidelines, which established that the minimum acceptable retention time for the analyte should be twice the retention time, corresponding to the void volume of the UHPLC column (0.2 min), and that the relative retention time of the analyte in matrix shall correspond to that of the calibration solution with a tolerance of ±2.5%. Furthermore, the selectivity of the method has been evaluated. According to the 2002/657/EC Decision, the selectivity is defined as the capability of the analytical method to discriminate between the target analyte molecules and isomers, degradation products, matrix interferences, and closely related substances. Selectivity has been evaluated analyzing 20 representative blank samples and checking if any interferences were noticed in the retention time region of the analyte. The chromatograms show that no interferences were eluted near the retention time of all analyzed blank samples ( Figure 2b).
Molecules 2020, 25, x FOR PEER REVIEW 6 of 13 of 35 Cl-CAP (m/z 321), produced a base peak at m/z 257, chosen as quantitative ion, and the product ion at m/z 194, chosen as confirmatory ion. The presence of CAP in a sample would be characterized by comparing two parameters: (1) The relative retention time of the suspect peak and (2) the ratio of the relative abundance of its main diagnostic ions with those obtained by analysis of a positive control (QC). In particular, the relative abundance of diagnostic ions (with a relative abundance >50% of the base peak) must not differ by more than ±20% from those observed in the QC. The result of tolerance ion ratio (CVIR) for selected ions 194/257 in different honey matrices satisfies the permitted tolerance given by 2002/657/EC guidelines (Table 2). After the selection of the diagnostic ions, the LC-MS method was developed; the total time of the analysis was 12 min with an analyte retention time of 3.49 ± 0.4 ( Figure 2a). This value was in agreement with the 2002/657/EC guidelines, which established that the minimum acceptable retention time for the analyte should be twice the retention time, corresponding to the void volume of the UHPLC column (0.2 min), and that the relative retention time of the analyte in matrix shall correspond to that of the calibration solution with a tolerance of ±2.5%. Furthermore, the selectivity of the method has been evaluated. According to the 2002/657/EC Decision, the selectivity is defined as the capability of the analytical method to discriminate between the target analyte molecules and isomers, degradation products, matrix interferences, and closely related substances. Selectivity has been evaluated analyzing 20 representative blank samples and checking if any interferences were noticed in the retention time region of the analyte. The chromatograms show that no interferences were eluted near the retention time of all analyzed blank samples (Figure 2b).   In addition, the analysis of the 20 blank samples allowed us to calculate the average (µ N ) and the standard deviation (σ N ) of the noise of the signal amplitude, which were used in the CCα and CCβ determination. The calculated values were µ N of 235.2 and σ N of 1902.0 for precursor/product 321→257, while µ N of 165.3 and σ N of 86.1 for precursor/product 321→194.

Calibration Curves and Linearity
The guidelines establish that the linearity of the analyte confirmatory methods, expressed as regression coefficient (R 2 ), must be greater than 0.9980. Calibration curves were constructed by comparing the responses given by samples in solvent and in matrix at nine levels (≥ 5 + i; as suggested by guidelines) in the range of 0.01-0.6 µg/kg, considering µ N as intercept. Good linearity was obtained for calibration curve in solvent (α = 177419; R 2 = 0.9982) and calibration curve in matrix (α = 176014; R 2 = 0.9989) relative to m/z 321→257. The same good linearity was obtained for both the calibration curves in solvent (α = 84032; R 2 = 0.9989) and in matrix (α = 81590; R 2 = 0.9990) relative to m/z 321→194. The regression analysis, performed on the two pairs of curves (calibration curves in solvent and in matrix), provides a p-value of 1.13E-06 for 321/257 ion and a p-value of 3.2 × 10 −7 for 321/194 ion. A p-value < 0.05 for both curves indicates that there is no statistical difference between the calibration curves in solvent and in matrix (Figure 3), with a confidence level of 95%. Therefore, the quantification of CAP in honey samples was performed by using calibration curve in solvent. In addition, the analysis of the 20 blank samples allowed us to calculate the average (µN) and the standard deviation (σN) of the noise of the signal amplitude, which were used in the CCα and CCβ determination. The calculated values were µN of 235.2 and σN of 1902.0 for precursor/product 321→257, while µN of 165.3 and σN of 86.1 for precursor/product 321→194.

Calibration Curves and Linearity
The guidelines establish that the linearity of the analyte confirmatory methods, expressed as regression coefficient (R 2 ), must be greater than 0.9980. Calibration curves were constructed by comparing the responses given by samples in solvent and in matrix at nine levels (≥ 5 + i; as suggested by guidelines) in the range of 0.01-0.6 µg/kg, considering µN as intercept. Good linearity was obtained for calibration curve in solvent (α = 177419; R 2 = 0.9982) and calibration curve in matrix (α = 176014; R 2 = 0.9989) relative to m/z 321→257. The same good linearity was obtained for both the calibration curves in solvent (α = 84032; R 2 = 0.9989) and in matrix (α = 81590; R 2 = 0.9990) relative to m/z 321→194. The regression analysis, performed on the two pairs of curves (calibration curves in solvent and in matrix), provides a p-value of 1.13E-06 for 321/257 ion and a p-value of 3.2 × 10 −7 for 321/194 ion. A p-value < 0.05 for both curves indicates that there is no statistical difference between the calibration curves in solvent and in matrix (Figure 3), with a confidence level of 95%. Therefore, the quantification of CAP in honey samples was performed by using calibration curve in solvent.

Recovery, Repeatability (within-Laboratory Reproducibility) and Accuracy
The repeatability was evaluated by the analysis of fortified honey samples at CAP concentrations of 0.1, 0.3, 0.45, and 0.6 µg/kg (including six replicates for each level), and carried out by two laboratory operators within three months. The concentration detected in each fortified honey sample as well as the average (AV), the standard deviation (SD), and the coefficient of variation (CV%) are summarized in Table 3.

Recovery, Repeatability (within-Laboratory Reproducibility) and Accuracy
The repeatability was evaluated by the analysis of fortified honey samples at CAP concentrations of 0.1, 0.3, 0.45, and 0.6 µg/kg (including six replicates for each level), and carried out by two laboratory operators within three months. The concentration detected in each fortified honey sample as well as the average (AV), the standard deviation (SD), and the coefficient of variation (CV%) are summarized in Table 3. The results show that the selected transitions provided comparable results for both recovery (%) and repeatability (SD). Since no certified reference materials (CRMs) were available in our laboratory, accuracy was assessed through the recovery of known amounts of CAP into blank samples. Three different botanical honey samples were fortified at 0.3 µg/kg and recoveries were calculated by interpolating the peak areas of 35 Cl-CAP ion transitions (m/z precursor/product 321→257 and 321→194) in solvent calibration curves of respective ions. The calculated CV% provided an estimate of the precision of the analysis ( Table 3). All data, corrected for the average of the recovery, fall within the ranges established in the 2002/657/EC guidelines; that is, the experimentally calculated concentration range must fall between −50% and +20% for MRPL ≤1 µg/kg and lowest possible CV% for concentrations <100 µg/kg.

Decision Limit (CCα) and Detection Capability (CCβ)
Undoubtedly, the most important changes that the 96/23/EC Directive introduced into the validation protocol of analytical methods were the introduction of the decision limit (CCα) and the detection capability (CCβ) criteria. The practical meaning of these two parameters can be summarized as follow: Signals less or equal to CCα are considered as noise (compliant samples), signals equal or higher to CCβ are considered as produced from forbidden substances (noncompliant samples), and signals between these two values require further investigation to take a decision. The determination of CCα and CCβ was carried out following the guidelines proposed by Antignac et al. [28] whereby a total number of 40 honey samples, of which 20 were blank and 20 were fortified samples, at six concentration levels were analyzed. The CCα and CCβ values for both the ion transitions were calculated using Equations (1) and (3), respectively. The average of the noise (µ N ) and the slope (α) of the matrix-matched calibration curve allowed us to calculate the decision limit (CCα). A concentration level that produced a signal-to-noise ratio of approximately 6 (0.01 µg/kg), during the analysis of the matrix-matched calibration curve, was chosen in order to calculate the detection capability (CCβ) through the standard deviation (σ S ) and the coefficient of variation (CV S ) of the signal (six replicates). Following the above procedure, the calculated values of CCα corresponded to 0.0025 µg/kg for m/z precursor/product 321→257 and 0.0052 µg/kg for m/z precursor/product 321→196. Therefore, the CCα calculated concentration allowed us to calculate the CCβ value, which corresponds to 0.0029 µg/kg for m/z precursor/product 321→257 and 0.0076 µg/kg for m/z precursor/product 321→194.

Real Samples Analysis
The developed analytical method was applied to the analysis of 96 honey samples from different botanical and geographical origins. A total number of 13 contaminated honey samples emerged from the analysis, of which one was purchased in a supermarket, one supplied by a local beekeeper, and 11 imported from intra-and extra-European countries. The contaminated honey samples contained CAP levels lower than 0.009 µg/kg but still higher than the detection capability (CCβ) and were therefore, considered noncompliant. These results confirmed the need to increase the monitoring programs of honey samples in order to guarantee the maximum quality and safety for consumers.

Samples and Standard Solutions
Honey samples from different botanical origin were supplied by beekeepers or purchased in supermarkets in Campania (Italy). Stock standard solution of CAP at 1000 µg/mL was prepared in acetonitrile and stored in glass vial at −18 • C for no more than one month; an intermediate standard solution at 10 µg/mL, obtained by dilution of the stock standard solution, was prepared weekly. Working standard solutions 1 and 2 (WS1 and WS2) at 100 ng/mL and 1 ng/mL, respectively, obtained by dilution of the intermediate standard solution, were prepared daily and used for sample spiking and preparation of calibration curves, in solvent and in matrix.

Salting-Out Assisted Liquid-Liquid Extraction Procedure
The salting-out assisted liquid-liquid extraction (SALLE), under optimized conditions, was performed on 1 g of honey, weighed into a 15-mL polypropylene centrifuge tube and diluted with 5 mL of water, previously brought to pH value of 12 (adjusted with 0.1 M NaOH) and to a concentration of 25% NaCl w/v. Afterwards, the mixture was vortexed until a homogeneous solution was obtained. Samples were extracted with 2 mL of ACN, then vortexed for 2 min in order to disperse extraction solvent into the aqueous sample solution ensuring analyte extraction. The mixture was centrifugated at 13,000 rpm for 5 min to achieve the separation between aqueous and organic phases. In order to facilitate the recovery of the extraction solvent (upper phase), water (lower phase) was removed using a Pasteur pipette. The extraction solvent (1.5 mL), containing the analyte, was transferred into a clean tube and evaporated under a gentle flow of nitrogen. Finally, the dry residue was reconstituted with 100 µL of H 2 O/MeOH (70:30 v/v), transferred into an HPLC vial and analyzed by LC-MS/MS.
Two multiple reaction monitoring (MRM) transitions of CAP were monitored; in particular, the MRM transition 35 Cl-CAP, m/z precursor/product transitions 321→257 (collision energy [CE]−14) was used for the quantification, while 35 Cl-CAP, m/z precursor/product transitions 321→194 (CE-10) was used for the identification.

Experimental Design
In order to find the best SALLE conditions, a chemometric approach was employed. For this purpose, a 2 4 -factorial design of 18 randomized experimental runs (1 block, 16 factorial design runs, 2 centerpoints) with 7 degrees of freedom was used. In the planning of the experimental design, the influence of four independent variables at low, medium, and high level was valued. The variables evaluated were: Honey-diluted volume (HDV, mL) of 3, 4, and 5 mL; extraction solvent volume (ACN, mL) of 2, 3, and 4 mL; pH value of 2, 7, and 12, and salting-out percentage (NaCl%) of 15, 20, and 25%. The three response factors considered were: Supernatant volume (SV, mL), extraction recovery (ER, %), and normalized matrix effect expressed as percentage of suppression (nME%). In order to eliminate the influence of matrix effect on the analyte response, extraction recoveries were calculated by comparing the CAP peak area resulting from the pre-spiked extraction procedure with that resulting from the post-spiked extraction procedure. The design was performed on 1 g of honey, spiked with 30 µL of WS1 (100 ng/mL), and corresponding to 0.3 µg/kg of CAP. The range of each factor used was selected by preliminary experiments. The experimental design conditions and the response factors results are shown in Table S1. Analysis of variance (ANOVA), obtained from the fitted model, was carried out to determine the statistical significance of the experimental variables and the optimum experimental conditions that minimized SV, maximized ER%, and hit the target of 0 for nME%. The experimental design was set up using the Statgraphics Centurion XVI software, Version 16.1 (Rockville, MD, USA).

Matrix Effect Evaluation
The matrix effect phenomena (signal suppression or enhancement) was evaluated through the analysis of blank honey samples from different botanical origins (acacia, chestnut, and citrus), under optimized conditions. After SALLE procedure, the extracts and the solvent solutions (H 2 O/MeOH 70:30 v/v) were spiked with an appropriate volume of CAP WS2 at three concentration levels (0.1, 0.3, and 0.6 µg/mL). Then, samples were evaporated under nitrogen, reconstituted in 100 µL of H 2 O/MeOH 70:30 v/v, and injected into the UHPLC-MS/MS system. The analyte peak area of the honey extracts and the solvent sample solutions were compared to evaluate the contribution of matrix interferences on the response of ion transitions.

Method Validation
The calibration curves in solvent and in matrix were constructed by setting the peak area of each MRM transition as a function of analyte concentration (nine levels), ranging from 0.01 to 0.6 µg/kg corresponding to 0.1-6 µg/L. The matrix-matched calibration curve was made by spiking a mixture of the three uncontaminated honey samples with an appropriate addition of WS volume, in order to obtain the following concentrations: 0.01, 0.025, 0.05, 0.075, 0.15, 0.20, 0.30, 0.45, and 0.60 µg/kg. The calibration curve in solvent was made in the same concentration range by diluting pure standards of CAP into H 2 O/MeOH (70:30 v/v). For both curves, the intercept, the standard deviation, and the coefficient of variation were calculated. The quantification of CAP in honey samples was carried out using the calibration curve of the 321→257 transition, whereas the 321→194 transition was used for the analyte confirmation. Linearity and sensitivity were evaluated through the regression coefficient (R 2 ) and the slope (α) of the calibration curve, respectively. The ruggedness and the specificity of the method were evaluated through the average (µ N ) and standard deviation (σ N ) of the noise amplitude of the blank samples; µ N was also used as intercept in the calibration curve. Then, σ N and α permitted us to calculate the decision limit (CCα) by the Equation (1): The repeatability was evaluated through the standard deviation of the signal amplitude (σ S ) produced by the analysis of samples at a CAP concentration inducing a signal-to-noise ratio approximately equal to 6. Finally, σ N , α, and the coefficient of variation of the signal amplitude (CV S ) permitted us to calculate the detection capability (CCβ) by the Equation (2): The use of CV S in the CCβ calculating formula is preferable in order to minimize the estimation error. Otherwise, CCβ can be calculated not taking this estimation error into account, by the Equation (3): In this validation protocol, CCβ was calculated using the Equation (3). Accuracy was evaluated by spiking uncontaminated honey samples from different botanical origin (acacia, chestnut, and citrus) with an appropriate volume of WS2 (10 ng/mL), in order to obtain the following concentrations: 0.1, 0.3, 0.45, and 0.6 µg/kg.

Conclusions
In this study, a confirmatory method for the quantitative analysis of CAP in honey and its validation according to the 2002/657/EC Decision was proposed. The use of the SALLE procedure, during the sample preparation step, allowed the extraction of the analyte so effectively that it fully met the requirements of the guidelines. In addition, the reduced use of organic solvents and disposable materials reduced the environmental impact. The selection and the optimization of the extraction parameters through the experimental design allowed us to reveal CAP at trace levels, whereas its quantification by tandem mass spectrometry (MS/MS) provides the selective confirmation of the analyte. Furthermore, the total absence of matrix effects in the optimized extraction conditions allows us to quantify CAP with good precision and accuracy, avoiding the use of an isotopic standard. All the analytical parameters were satisfactory in terms of analyte recovery, repeatability, specificity, and ruggedness. The simplicity and the rapidity of the extraction technique as well as the sensitivity and the accuracy of the detection method are the main advantages of the proposed procedure. These features allow us to increase the productivity of the samples, reducing the use of toxic solvents and making the method both economical and eco-friendly.
Supplementary Materials: The following are available online, Figure S1: Evaluation of preliminary extraction recoveries on three different botanical honey under optimized experimental design conditions, Figure S2: Evaluation of preliminary matrix effects on three different botanical honey at three concentration levels under optimized experimental design conditions. Figure S3: Theoretical (a) and experimental (b) isotopic pattern of CAP in negative ion mode. Table S1: Design matrix for the 24-Factorial design and obtained result for each run.