A total of 60 samples, consisting of meat (chicken) and mixed meat, mixed fish, and mixed legumes were used for validation. A total of 100 samples, consisting of meat (chicken, beef, lamb, ham) and mixed meat, fish (salmon and trout) and mixed fish, legumes (white beans, chickpeas, lentils) and mixed legumes, and vegetables (green peas and courgettes) were analyzed for monitoring purposes. All samples, listed in Table 1, were purchased in local markets and subsequently stored in their original packaging, properly labeled, and refrigerated (4 ± 1°C) until analysis. In Italy, in accordance with the European Regulation (37), the dietary guidelines suggest introducing homogenized products into infant complementary feeding, particularly during weaning, thanks to their consistency and specific composition.

Analytical grade cyclohexane, ethyl acetate, and acetonitrile were supplied by Merck Life Science s.r.l. (Darmstadt, Germany). Isooctane was purchased from Panreac Química S.L.U. (Castellar del Vallès, Barcelona, Spain). Deionized water (18.2 MΩ/cm) was obtained from a Milli-Q purification system (Millipore, Milan, Italy). Polychlorinated biphenyl (PCB) 209 (purity >99.0%; 10 mg/L in isooctane, Lab Instruments Castellana Grotte, Bari, Italy) was used as internal standard. From this stock standard solution, PCB 209 working standard solutions of 1,000 μg/L in isooctane were prepared and used to spike food samples at 100 μg/L. Within the subsequent monitoring survey of PAH levels in tested samples, PCB 209 was added to the samples before extraction, to perform reliable quantifications of target analytes with correction of errors due to ion suppression/enhancement caused by the presence of matrix co-extracts in the injector and ion source. Certified mix standard solution, indicated as IL8 constituted by benzo[a]pyrene (BAP), benz[a]anthracene (BAA), benzo[b]fluoranthene (BBF), chrysene (CHR) and their relative deuterated (PAHs-d12), that is BAP-d12, BAA-d12, BBF-d12 and CHR-d12 (purity >99.0%; 100 mg/L in toluene, Lab Instruments Castellana Grotte, Bari, Italy) was used for the method validation. Working standard solutions of 1,000 μg/L, 100 μg/L, 25 μg/L, and 10 μg/L were obtained by diluting the stock IL8 in isooctane. All standard solutions were stored at −20°C and were taken out to allow them to reach room temperature and sonicated before use. Matrix-matched calibration (MMC) curves were obtained using 0.1, 0.5, 1.0, 2.0, 5.0 μg/L working standard solutions of IL8 in the blank sample. The presence of the matrix effect was considered in the optimization and validation steps and in the monitoring study. MMCs were used because the different substances investigated had a different composition from each other, i.e., lipid, protein, carbohydrate and salt content (Table 1).

SupelMIP SPE – PAHs cartridges (50 mg/3 mL) were purchased from Merck Life Science s.r.l. (Darmstadt, Germany). Bond Elut Enhanced Matrix Removal - Lipid dispersive SPE (Bond Elut-EMR lipid dSPE) Agilent Technologies were purchased from Analitica sas (Gioia del Colle, Bari, Italia).

For QuEChERS procedures, all extraction and purification salts were listed below and were supplied by Lab Instruments (Castellana Grotte, Bari, Italy):

The optimization of PAH extractions and clean-up protocols were performed on smoked salmon samples and chicken baby foods, following what was suggested by literature and on the base of our earlier knowledge relevant to the multi-residual analysis of foodstuffs (38, 39). During the development of analytical procedures, an ultrasonic bath (LIARRE s.r.l., Casalfiumanese, Bologna, Italy), a TX4 Digital vortex mixer (Velp Scientifica, Usmate Velate, Italy) and a BKC-DL5M centrifuge (BiobaseMeihua Trading Co., Ltd., Jinan, China) were used to sonicate, vortex and centrifuge, respectively.

An Agilent 7693A Automatic Liquid Sampler and an Agilent 8,890 N gas chromatograph coupled with an Agilent 7,000 Triple Quadrupole detector (Little Falls, DE, United States) were used for GC–MS/MS analyses. Data were acquired using MSD ChemStation software (Agilent, Little Falls, DE, United States). The GC–MS/MS operating conditions are shown in Table 2. Backflushing was used to prevent contamination with compounds strongly retained in the primary column by reversing a continuous flow of carrier gas, thus avoiding, through the second column, reaching the MS detector. According to European directive (40), the number of identification points that GC–MS/MS can earn, namely three points, one for ion precursor and 1.5 for each of two daughter ions. The identification point can be five if there are two precursor ions, each with 1 daughter of five points. The list of precursor ions, both quantifiers and qualifiers, their transitions, as well as the chosen collision energies (CEs), were shown for each of the 4 PAHs and the 4 PAHs-d12, in Table 3.

The identification of BAP, BAA, BBF, and CHR in GC–MS/MS analysis was performed by comparing the retention time of each compound with that of the relevant standard and verifying the presence of precursor ions (Table 3). The ratios between the peak areas of quantifier ions and PCB 209 were used for the compound quantification.

The first two extraction and clean-up protocols were tested using MMC curves with smoked salmon samples. MMC curves of SupelMIP SPE and QuEChERS based on Bond Elut-EMR lipid dSPE showed correlation coefficients ≥0.99. The recovery percentage at 2 μg/kg was also evaluated, carrying out 3 replicates for both protocols, also comparing two different solvents, i.e., acetonitrile and toluene, for the preparation of the IL8 working standard solution used to spike the samples. The results, shown in Table 4, demonstrated that the QuEChERS approach gave higher recovery values (71.4–83.0%) compared to SupelMIP SPE (29.8–48.8%). In Figure 1, the chromatographic separation and mass spectrometric identification of the analytes were reported for protocol based on the QuEChERS procedure. Therefore, the QuEChERS method was selected as the best procedure and was further optimized, modifying the extraction and purification steps in the other 5 Protocols (III, IV, V, IV, and VII). No significant differences emerged changing the solvent for standard solutions preparation, therefore, considering the high toxicity of toluene compared to acetonitrile, the latter was selected for further experiments.

Protocols from III to VII were tested with chicken baby food samples, performing two replicates for the recovery evaluation at 0.5 μg/kg. Protocols III, IV, V, and VI provided recovery values <10%, as shown in Table 5. Protocol VII provided the best results with recovery values ranging from 70 to 90% and, for this reason, it was chosen as the extraction and cleaning procedure for subsequent validation and monitoring. In particular, concerning other tested procedures, the step at −20°C overnight, together with chosen salts, enhanced the clean-up with better removal of co-extracted fat lipids. Then, protocol VII was also tested on other infant food matrices based on meat, fish, legumes, and mixed fruits. The full dataset obtained from this study was evaluated by using Box Plot, a useful tool of graphic representation used for managing quantitative data. This tool allows the comparison of many distributions in the same graph, highlighting the most significant characteristics such as symmetry, range, variance, and possible outliers. The median value, first and third interquartile, upper and lower limits within Tukey’s limit, and outliers (if any) can be visualized in a single graphical representation. Moreover, the use of median and quartiles in describing the distribution makes the considerations about possible outliers in the dataset more robust. The dataset obtained during the method validation showed similar accuracy parameters for 4 PAHs, but different behavior in different matrices. A more detailed data analysis showed that the mean recovery percentage obtained by analysing baby food samples of mixed meats, legumes, fish products, and chicken were comparable, in the range of 80–100%, with higher variance in chicken samples, while a lower value (~60%) was obtained from homogenized mixed fruits analysis. Regarding method sensitivity, evaluated by elaborating the linear regression, the higher performance was obtained for BAA, followed by CHR, BAP, and BBF. The Box Plot, shown in Figure 2, was successfully applied for optimizing the analytical method for the detection of 4 PAHs in baby food since it allows rapid and effective visualization of all statistical parameters that characterize the procedure.

The following validation parameters, namely the linearity of MMC curves, the limit of detection (LOD), the limit of quantification (LOQ), the selectivity, the precision, the recovery percentage, and the ruggedness were determined to assure the efficiency of the proposed GC–MS/MS method and to guarantee the validity of routine analysis. The MMC curve for the 4 PAHs was evaluated from 0.1 to 5 μg/kg, considering the framework of European legislation for residues in baby foods (40). MMC curves were used for the validation method, analyzing each level in triplicate. All validation parameters, obtained using matrix chicken baby food (Table 6) resulted in compliance with EU provision No. 836/2011 and 808/2021 (41, 42).

The coefficients of determination (R² values) were in the range of 0.998–0.999, indicating good calibration linearity in investigated matrices. Mandel test was used to assess whether the data best fitted a linear function. The test was verified with a p-value <0.05. The significance of the slope (b), of the regression line obtained from the MMC curve, at α = 0.05, was verified with a t-test. The t-Student was calculated for ratios s_b/b, where s_b was the standard deviation of the slope b, obtained for each of the studied analytes. All the values resulted lower than 0.22. Figure 3 shows MMC curves for each PAHs.

LODs and LOQs were calculated, according to the following equations: LOD = 3.3s_a/b and LOQ = 10s_a/b, where s_a was the standard deviation of the intercept. The calculated values of LOD and LOQ (Table 6), indicated high sensitivity for the determination of these analytes at trace levels, reducing the risk of false-negative results. Selectivity ensures the correct identification of analytes of interest without interference from other components that could be present in the sample, such as impurities, degradants, matrix components, etc. To verify method selectivity, 20 sample blanks representative of the chosen matrices were analysed. The 20 specificity tests were carried out on the following baby food samples: chicken, beef, lamb, ham, salmon, trout, white bean, chickpea, legumes, pea and courgettes, mixed meat, and mixed fish. The absence of significant interferences in the maximum tolerance range of ±0.1 min for analyte retention times compared to a spiked sample was verified.

The trueness and precision of measurements were assessed in accordance with Decision 2021/808/EC (40) through the analysis of spiked samples, prepared starting from blank material by additions of known amounts of the analytes. The precision of a method measures the repeatability, evaluating the degree of agreement among individual test results. The recovery percentage is used to indicate trueness and to express the distance between the mean and the reference value. Precision and trueness were determined by analyzing two sets of blank matrices (six replicates each), spiked at a concentration of 0.1 μg/kg for each analyte. Precision was expressed as deviation standard (SD) and relative deviation standard (RSD), while trueness was expressed as recovery %. The intra-day RSD values were well below the reference values of 20%, derived by the Horwitz equation, under repeatability conditions, demonstrating a good method precision (43). Calculated values of recoveries % and RDS % were in the range of 73–110% and always lower than 8.75%, respectively. Homoscedasticity of data was evaluated using ANOVA (p = 0.05), calculating the mean of recoveries %, obtained for the level at 0.1 μg/Kg. The recovery factors resulted from 0.903 to 1.368 for the 4 PAHs. In Figure 4, an example of a chromatographic separation obtained at 0.1 μg/kg was reported, showing the chromatogram of quantifier ions and relative studied transition for all 4 PAHs in chicken baby food samples. For the evaluation of the uncertainty of measurements, the metrological approach was adopted, using the validation data obtained from each step of the analytical procedure. Taking into consideration the uncertainties propagation law, the concentration relative uncertainty was calculated for 4 PAHs, as reported in a previous work (39). A relative expanded measurement uncertainty was calculated using a coverage factor k of 2, corresponding approximately to a 95% confidence level.

The robustness of this analytical method was evaluated by Youden’s test, introducing several changes at once (44). The application of the method to other matrices was considered as changes to be applied. Hence, mixed meat and fish were also spiked at 1.0 μg/Kg of each PAH and then analyzed and compared with chicken samples. The standard deviation of the differences D_i (SD_i) was calculated according to the equation reported by Karageorgou and Samanidou (45). SD_i resulted not significantly larger than the standard deviation of the method carried out on validated matrices (chicken samples), confirming that the proposed method is sufficiently robust against the chosen modification.

Moreover, other matrices such as fish and legume baby foods were considered. Further analyses on two sets of blank matrices (four replicates each) were also carried out. Fish and legume blank samples were spiked at a concentration of 0.1 and 1.0 μg/kg for each analyte, respectively. To evaluate the method reproducibility, the absolute difference between independent single test results (C₁ and C₂) was considered. The |C₁-C₂| values were not found to be higher than the repeatability limit, calculated against the expected 95% probability interval of s_r × t√2, where s_r is the repeatability standard deviation at 1.0 μg/kg, as shown in Table 7. For fish samples, values of recovery % resulted in the range of 83–105%, while for the legume matrix, they were in the range of 73–85%. Also, these data were compliant with the acceptability criteria of European regulation (41, 42).

The monitoring was performed over 2 years, by analyzing a total of 60 samples of baby food, of common brands present on the Italian market, and of different compositions (meat, fish, legumes, and vegetables). The samples were analyzed in duplicate, verifying that the concentrations of each analyte satisfied the repeatability criteria, i.e., the absolute difference of two values |C1-C2| must not exceed the repeatability limit (s_r × t√2). In Table 8 values of s_r, obtained during the validation step, were calculated at three levels (0.1, 1.0, and 1.5 μg/kg). If |C1-C2| satisfied these criteria, a mean value of concentration was calculated and reported together with its measurement uncertainty, also considering the coverage and recovery factors. The final results were expressed as the sum of 4 PAHs. The results showed that the mean level of total PAHs, for all the investigated samples, was always below the lowest LOQ. Moreover, as reported in the literature, cooking methods could influence the production of PAHs (46). Considering that microwave and steaming cooking are the most used methods to conserve nutrients and quality of baby foods, 40 samples of chicken, salmon, lentils, and mixed legumes and vegetables, were analyzed after these heat treatments. Twenty samples were heated for 5 min in a water bath at 100°C and the other 20 samples were heated for 30 s in a microwave at full power. As in the previous case, PAH contents resulted under the LODs and then they were not detected and quantified (Figure 5). However, in this regard, further studies are needed to evaluate the formation of 4 PAHs also considering other setting parameters (temperature, time, power etc…) but also different cooking treatments.