Occurrence of Aflatoxin M1 (AFM1) in Donkey Milk Collected in Northern Italy

Aflatoxin M1 (AFM1) is a well-known mycotoxin that can be found in the milk of animals that have ingested feed contaminated with aflatoxin B1 (AFB1). In Italy, the development of donkey farms is mainly due to growing request of donkey milk, which is considered an incomparable substitute for human mother’s milk for its chemical composition and organoleptic characteristics. The aim of this study was to assess the occurrence of AFM1 in donkey milk produced in a farm in Northern Italy, also in view of the few data available about the presence of this mycotoxin in this type of milk. Therefore, 63 milk samples were collected and analyzed using a fast and sensitive HPLC and fluorescence detection (FLD) method previously optimized and validated. None of the milk samples collected were found to be contaminated at a level above the limit of quantification (LOQ) (0.0125 ng/mL), while only one sample showed traces of the mycotoxin at a concentration between the limit of detection (LOD) and LOQ (0.0044 ng/mL), well below the legal limit established for infant milk and follow-on milk (0.025 ng/mL). These results are in line with those of the few similar surveys carried out on donkey milk and seem to indicate a low risk of AFM1 contamination for this food.


Introduction
Donkey's milk is considered the most similar to mother's milk for its chemical composition and organoleptic characteristics. For this reason, it has become a staple food in the diet of infants affected by cow's milk protein allergy (CMPA). The incidence of this disease of infancy and early childhood seems to be prevalent in the developed countries, where it ranges from 0.5 to 3% in the first year of life [1]. Caseins are one of the main allergenic components of milk. Donkey milk has a casein concentration lower than cow's milk, and it is characterized by a greater digestibility, clinical tolerability, palatability, and nutritional adequacy. Clinical studies showed that donkeys' milk is well tolerated (82.6-88%) by infants [2]. Furthermore, it can provide additional physiological functions, such as antibacterial substances, digestive activity molecules, growth factors, and hormones [3,4]. For its properties, several authors envisage a role of this food in the osteogenesis process, in treatment of atherosclerosis, in the post-heart-attack recovery period, in cases of premature senescence, and in hypocholesterolemic diets [5]. Moreover, the use of donkey milk in external-dermal application for cosmetic and medical purposes is well known, and its beneficial effects are due to milk proteins, which contain all essential amino acids, which may help restore the protein layer of cell membranes and intracellular compartments of cells [6].
The donkey (Equus asinus) is a rustic, undemanding, and easily adaptable animal, which in most cases lends itself to semi-wild breeding, and is able to enhance marginal areas [7]. In the 1950s, in Italy, there were almost a million donkeys, but their numbers have decimated over time, and it is only since such cases, subsequent controls in milk very often show levels of AFM 1 higher than normal as a result of ingestion of contaminated feed by milk-producing animals. Similar conditions occurred in Italy in 2003 [30,42] and in 2012 [43,44].
The aim of the present study was to assess the occurrence and levels of AFM 1 in donkey milk produced in a farm of the North of Italy, also in view of the few data available about the presence of this mycotoxin in this particular type of milk. For the analysis, a sample preparation procedure with immunoaffinity columns (IAC) was used, and a fast and sensitive HPLC method coupled with fluorescence detection (FLD) was optimized and validated.

Animals, Diets, and Nutrition
The milk came from donkeys of the following breeds: Martina Franca, Amiata, Sant'Andrea, San Domenico, Argentato di Sologno, Sardo, Ragusano, Poitou. The animals were raised outdoors free to move in large spaces within a rural area. The feeding was mainly the spontaneous vegetation coming during grazing integrated with feed. The composition of the feed was as follows: barley, flaked barley, wheat bran, flaked corn, flaked fava beans, flaked soybean meal, dehydrated alfalfa flour, sugar beet pulp, sugar cane molasses, fatty acids (fat of vegetable origin), dicalcium phosphate, calcium carbonate from ground calcium rocks, sodium chloride, and brewer's yeast.

Sampling
A total of 63 milk samples were collected between June and November 2015 from a farm located in Northern Italy. The sampling was carried out mainly in autumn (52 samples collected) and, to a lesser extent, in summer (11 samples collected). Each milk sample was from a single milking of an animal, except for one sample, which was collected from the farm bulk milk tank. Milk samples were stored in 50 mL Falcon tubes at −20 • C until analysis, without first having been subjected to any type of treatment.

Solvents and Reagents
The reference standard of AFM 1 was purchased from Sigma-Aldrich Co. (St Louis, MO, USA). Water, acetonitrile, and methyl alcohol used to prepare AFM 1 standard solutions, for the sample clean-up procedure, and to formulate the HPLC mobile phase, were HPLC grade. All solvents were obtained from Mallinckrodt Baker B.V. (Deventer, The Netherlands). Immunoaffinity columns used for samples purification (Afla M 1 TM HPLC) were purchased from Vicam ® (Milford, MA, USA).

Chromatographic Apparatus and Conditions
AFM 1 analysis was performed by HPLC-FLD on a chromatographic apparatus consisting of a System Gold Programmable Solvent Module 126 Pump (Beckman, San Ramon, CA, USA) equipped with an HT 800 L autosampler (HTA, Brescia, Italy) fitted with a 100 µL loop. Chromatographic separation was achieved in isocratic elution mode and at room temperature using an analytical column Zorbax Eclipse Plus C18 150 × 4.6 mm 5 µm (Agilent, Santa Clara, CA, USA). The mobile phase consisted of 61% of deionized water, and 39% of acetonitrile. The flow rate was set at 0.7 mL/min, and the injection volume was 100 µL. The FLD detection was obtained by means of a 821 FP fluorescence detector (Jasco, Tokyo, Japan) set at the excitation wavelength of 360 nm and emission wavelength of 440 nm. The system was computer-controlled by a Beckman Coulter 32 Karat Software.

Sample Preparation and Immunoaffinity Clean-Up Procedures
The procedure was set up on the basis of the instructions provided by the immunoaffinity column manufacturer with a few modifications. Forty milliliters of milk were measured into a 50 mL cylinder, transferred into a 50 mL Falcon tube, and centrifuged in a refrigerated centrifuge at 5 • C for 15 min (1540× g). The fat layer was then removed and the skim portion, once at ambient temperature, was loaded onto a IAC (Afla M 1 TM HPLC) and passed at a flow rate of about 1-2 drops per second.
After washing the column two times with 10 mL of deionized water, AFM 1 was eluted in a glass tube with 1.25 mL of acetonitrile-methyl alcohol (3:2) and 1.25 mL of deionized water at a rate of 1 drop per 2-3 s. The eluate collected was reduced to dryness in a rotational vacuum concentrator (Univapo Martinsried/Munich, Germany). The residue was redissolved in 1 mL of acetonitrile-methyl alcohol (3:2 v/v) and diluted with 1 mL of deionized water. Finally, the resulting solution was filtered through a 0.45 µm nylon syringe filter (Phenomenex, Torrance, CA, USA) and transferred to a HPLC vial for analysis (AFM 1 , if present in milk sample, will become concentrated 20 times in the final analysis solution).

Quantification
A commercial AFM 1 standard solution 10 µg/mL in acetonitrile (Sigma-Aldrich Co., St. Louis, MO, USA) was diluted to give a stock solution 1 µg/mL and subsequently divided into 1 mL aliquots, which were stored at −20 • C until use. Working solutions, obtained by diluting an aliquot of stock solution with water, were used to prepare reference standard in water-acetonitrile-methanol solution (5:3:2 v/v) at 6 concentration levels (0.25, 0.5, 1, 1.5, 2, 2.5 ng/mL), and these were analyzed by HPLC-FLD to generate reference curves in solvent. Spiked samples were prepared by adding appropriate volume of AFM 1 working solution to blank samples of milk and then processed as for the unknown samples. The resulting solutions were analyzed by HPLC-FLD to generate calibration curves in matrix. The quantities of AFM 1 to be added to milk samples were calculated so that the concentrations of the final solutions were the same of the reference standards in solvent (0.25, 0.5, 1, 1.5, 2, 2.5 ng/mL) and correspond to concentrations of 0.0125, 0.025, 0.05, 0.075, 0.1, and 0.125 ng/mL in matrix, taking into account the concentration factor resulting from the sample extraction/purification procedure (20:1). Calibration curves were obtained from least squares linear regression analysis and were used to determine AFM 1 concentrations in the unknown samples. Regression equations and determination coefficient (R 2 ) were calculated using Microsoft Excel 2013 software.
An in-house validation of the analytical method was carried out and included the following parameters: selectivity, calibration curve performance (including linearity and working range), limit of detection (LOD) and quantification (LOQ), accuracy, precision, recovery, and carry-over. Selectivity was proved using 15 blank samples, which were analyzed and evaluated for interference. Linearity was determined in the working range on the basis of the determination coefficient (R 2 ). Working range was the range of values from the LOQ value to the maximum concentration of AFM 1 in the standard solution for which the calibration curve had been plotted. LOQ was the lowest calibration standard of the 6-point calibration curve. LOD was calculated as the concentration corresponding to a signal 3 times the noise level measured in 15 blank samples at the AFM 1 retention time. The signal-concentration relationship was determined based on the lowest calibration standard (LOQ). Accuracy and precision were assessed on samples spiked with known amounts of AFM 1 and were measured at 4 different concentrations over 4 consecutive days (16 determinations in total). To evaluate accuracy, bias to nominal concentration was calculated, while coefficient of variation (CV) of the replicate measurements was taken as a measure of precision. The calculation of recovery was made by dividing peak areas of extracted standard solutions from AFM 1 -spiked samples by peak areas of AFM 1 standard solutions at the same concentration. Carry-over was assessed by injecting blank samples after the highest calibration standard. Quality control (QC) samples were processed and analyzed together with the unknown samples to monitor the performance of the extraction-purification procedure and to assess the validity of the analytical results. This study was performed according to ISO 9001 requirements.

Method Development
Compared to the extraction/purification procedure suggested by the IAC manufacturer, a smaller volume of milk was used for each sample (40 mL vs. 50 mL). The final eluate collected from IACs, instead of being analyzed directly, was reduced to dryness and subsequently redissolved (1 mL of acetonitrile-methanol solution and 1 mL of water) to obtain a more accurate measure of the volume of final solution (2 mL), and to maintain a concentration factor 20:1 in relation to the smaller amount of milk loaded onto the IAC. This is an important step of the clean-up procedure, because an inaccurate final elution can result in considerable errors in the evaluation of the concentration of the analyte. The chromatographic condition were developed starting from a method already tested in our laboratory for this type of analysis. Several parameters were optimized to reduce the run time and improve the sensitivity of the method. For the initial tests, a conventional analytical column Luna C18 250 × 4.6 mm 5 µm (Phenomenex, Torrance, CA, USA) was used, and the mobile phase composition was 70% of water-methanol (90:10 v/v) and 30% of acetonitrile, in isocratic elution. The flow rate was set at 0.9 mL/min, and the injection volume was 100 µL. At these conditions, the retention time of AFM 1 was 6.5 min and the total run time was 9 min. Since the samples were quite clean and the analyte to be quantified was only one, it was decided to test a shorter column (150 mm). Firstly a column Zorbax Eclipse plus C18 4.6 × 150 mm 3.5 µm (Agilent, Santa Clara, CA, USA) was used, and under the same conditions (flow rate, mobile phase composition, and injection volume), peak height increased by 8% and retention time decreased by 35%. Analyses using smaller injection volumes (50 and 75 µL) were also carried out, but despite the improvement of some chromatographic parameters (peak width and retention time decreased, peak shape increased), peak height decreased without a reduction in the signal-to-noise ratio. A series of experiments using linear gradients of acetonitrile from 30 to 44% with different slopes were performed, but no significant improvements were observed, also considering the re-equilibration time requested in the gradient elutions, while isocratic elution with 65% water and 35% acetonitrile seemed to lead to satisfactory analyses. However, after several injections of extracted samples, the backpressure of the column increased too much for a conventional HPLC system such as the apparatus used in this study. Consequently, the column was replaced with a similar one packed with 5 µm particles. After several runs using both linear gradients of acetonitrile from 30 to 40% and isocratic elutions with percentages of acetonitrile in the same range, the best result was obtained by applying an isocratic elution with mobile phase consisting of 61% water and 39% acetonitrile at a flow rate of 0.7 mL/min. In respect to the initial analytical conditions, the use of a shorter column together with the optimization of the chromatographic parameters has resulted in a significant reduction in AFM 1 retention time (from 6.5 to 3.8 min) and of the total analysis time (from 9 to 5 min), as well as in an improvement of the detection limit as a consequence of the narrowing of the peak width and increase in peak height.

Assay Validation
The 15 blank samples extracted and analyzed for the evaluation of selectivity did not show interfering peaks at the retention time window of AFM 1 , as well as no abnormal background noises being observed. The retention time was 3.7 min while the run time was 5 min. Figure 1 shows representative chromatograms of a blank sample of milk (a) and a blank sample of milk spiked with AFM 1 at 0.025 µg/L level (b). Both reference curves in solvent generated from analysis of pure AFM 1 standard solutions and calibration curves in matrix, obtained from analysis of standard solutions resulting from extraction of spiked milk samples, showed a satisfactory linearity in the range of concentrations tested, which went from 0.0125 to 0.125 ng/mL (working range). In fact, in all regressions, the coefficient of determination (R 2 ) was always >0.999. LOQ and LOD of the method were 0.0125 and 0.0025 ng/mL, respectively, well below the AFM 1 limits established at EU level for milk (0.05 ng/mL), infant milk, and follow-on milk (0.025 ng/mL). Accuracy and precision were determined by preparing and analyzing four samples per level at four concentration levels (LOQ, low, medium, and high) covering the working range, and the concentrations tested were 0.0125, 0.025, 0.075, 0.125 ng/mL. The bias ranged from −6 to −2.5%, while the coefficient of variation (CV) was in the range 5.1-12.8%. Recovery experiments were carried out at four spike levels within the working range, and the average recovery percentages were found to be between 86.6 and 92.2%, while the overall average recovery was 89.0%. The data about the percent recoveries and the mean recoveries for each fortification level are shown in Table 1. Injection of blank samples following the highest calibration standard did not show peaks above LOD, thus proving that no significant carry-over effects occurred. Therefore, the overall results of the method validation demonstrated the reliability of the method for the determination of AFM 1 in milk.
Vet. Sci. 2020, 7, x FOR PEER REVIEW 6 of 14 highest calibration standard did not show peaks above LOD, thus proving that no significant carryover effects occurred. Therefore, the overall results of the method validation demonstrated the reliability of the method for the determination of AFM1 in milk.  The extraction-purification technique based on the use of immunoaffinity columns together with HPLC-FLD analysis is still a valid approach for this type of determination especially when more sophisticated and expensive equipment such as HPLC-MS/MS are not available. The refinement of the method carried out in this study has allowed a general improvement of several analytical parameters that have led to lower consumption in the mobile phase, reduction in analysis time, and therefore, reduction in costs.  The extraction-purification technique based on the use of immunoaffinity columns together with HPLC-FLD analysis is still a valid approach for this type of determination especially when more sophisticated and expensive equipment such as HPLC-MS/MS are not available. The refinement of the method carried out in this study has allowed a general improvement of several analytical parameters that have led to lower consumption in the mobile phase, reduction in analysis time, and therefore, reduction in costs.

Occurrence of AFM 1 in Milk Samples
In the present survey, 63 samples of donkey milk were analyzed, and none of them were found to be contaminated with AFM 1 at a level above the LOQ, while only one sample showed traces of the mycotoxin at a concentration between LOD and LOQ (0.0044 ng/mL). Milk samples and AFM 1 concentrations are reported in Table 2. In the evaluation of the results, the law limit established for infant milk and follow-on milk (0.025 ng/mL) was kept as a reference, as donkey milk is mainly used as infant food. The only sample in which AFM 1 was detected had a concentration much lower than the abovementioned limit. These data prove that the milk analyzed is safe for human consumption, at least regarding the presence of this toxin, and suitable for young children. Others similar surveys carried out in Italy but on ruminants' milk, generally showed a higher percentage of positive samples [26,42,[45][46][47][48][49][50][51]. Whilst considering the limited number of samples analyzed and their provenience from one single farm, some hypothesis can be formulated to explain the findings of the present study. The donkeys from which the milk was taken are kept for most of the year in the pasture and therefore the only way for the animals to take AFs is through the feed given in the evening when they returns to the stables. Given the almost total absence of the toxin in milk, it can be assumed that the feed was safe and effective controls have always been carried out. Another explanation could be that the non-detection of AFM 1 in practically all milk samples is due to the very low casein content of donkey milk compared to the cow milk [52][53][54][55]. In fact, this toxin has a high affinity for casein [50,[56][57][58][59], and therefore in the milk of animals exposed to AFs via feed, there could be a correlation between the amount of AFM 1 and the percentage of casein.
A further possibility is that the virtual absence of AFM 1 is due to a low carry-over of AFM 1 in the milk of donkey. While for lactating cows a large amount of literature exists, in the case of lactating donkeys, there is very little information on the transfer of AFM 1 into milk. The only reference study is the one reported by Tozzi et al., which evaluated the amount of AFM 1 and AFM 2 in donkey milk in response to feeding AFB 1 and AFB 2 . For 12 days, six donkeys received a diet supplemented with naturally contaminated corn containing the two toxins (202µg of AFB 1 and 11 µg of AFB 2 ). The maximum levels detected in milk were 32.25 ng/L (AFM 1 ) and 23.33 ng/L (AFM 2 ), and carry-over into milk resulted to be 0.02% and 0.31% for AFM 1 and AFM 2 , respectively [12]. By focusing on AFM 1 , the percentage of carry-over is very low as compared with that reported in other species.
Several experiments were carried out using dairy cows fed AFB 1 contaminated diets, and the ratio between the AFM 1 in milk and the intake of AFB 1 was calculated. Pettersson et al., in a study in which six cows were fed AFB 1 contaminated diets at two levels (5.3 and 10.8 µg/kg), in a change-over experiment to groups of three animals for one week, reported that the overall mean carry-over was 2.54 ± 0.76% [60]. Masoero et al. investigated the relationship between somatic cell counts, milk yield, and conversion of dietary AFB 1 into milk AFM 1 using 34 lactating dairy cows. The AFB 1 carry-over into milk calculated that the plateau was between 1.29 and 2.70%, and resulted to be affected by the milk yield [61]. In a study carried out on 18 dairy cows to evaluate the efficacy of a mycotoxin deactivating product, Pietri et al. reported that the carry-over of AFB 1 in milk was 3.85% in animals who ingested a daily dose of AFB 1 equal to 97.3 µg for seven days [62]. Sumantri et al. reported a carry-over of AFB 1 in milk equal to 0.1% in dairy cows who ingested AFB 1 at low and high doses (3.4 and 350 µg AFB 1 /cow/day) for 10 days [63]. In another experiment carried out by Britzi et al., 12 cows were used to determine AFM 1 carry-over following daily administration of feed containing 86 µg AFB 1 for seven days. The mean carry-over rate at steady-state was 5.8% and 2.5% in mid-lactation and late-lactation groups, respectively [64]. Gonçalves et al. used 20 Holstein cows in late lactation to evaluate the effect of different sources of Saccharomyces cerevisiae on AFM 1 excretion in milk from cows receiving 480 µg AFB 1 per day for six days. In the control group (animals who ingested AFB 1 contaminated feed without Saccharomyces-cerevisiae-based products), the carry-over rate was in the range 0.83-3.89% [65]. In an experiment to quantify AFM 1 carry-over rate from feed to cheese, Costamagna et al. reported that in dairy cattle receiving an average daily dose of 66.8 µg AFB 1 via natural contaminated feed, the carry-over to milk was 0.84% [58].
The abovementioned experiments show that, in dairy cows, AFB 1 carry-over to milk ranged from 0.1 to 5.8%. These studies also show that factors influencing carry-over include milk yield, feed consumption, and the ratio of concentrated feed included in the diet, as reported also by Völkel et al. [66]. Furthermore, carry-over rate can vary widely among individual animals, from day to day and from one milking to the next, as it is influenced by various (patho-)physiological factors [67].
Fewer similar studies were carried out on ruminants of other species. Battacone et al. reported two experiments carried out on dairy ewes. In the first one, four animals received a single dose of AFB 1 (2 mg), and the mean rate of transfer of AFB 1 into AFM 1 in milk was 0.032%. In the second experiment, 12 ewes divided into three groups received different daily doses of AFB 1 (32, 64, and 128 µg) for 14 days, and AFB 1 carry-over in milk was 0.082, 0.133, and 0.119% for each group, respectively [68]. In a later similar study carried out using 12 ewes divided into three groups receiving 32, 64, and 128 µg AFB 1 per day for seven days, higher carry-over values were found (0.33, 0.29, 0.26% for each experimental group, respectively) [69].
Scientific literature reports that these type of evaluations were carried out also on goats. In an experiment in which three goats received a single dose of AFB 1 (0.8 mg), Mazzette et al. found that the mean rate of transfer of AFB 1 into AFM 1 in milk was 0.26%, with a high individual variability [70]. In a very similar study carried out by Battacone et al., five dairy goats were administered 0.8 mg of AFB 1 orally, and about 0.17% of the amount of the toxin administered was detected as AFM 1 in milk [71]. Nageswara Rao et al. used nine lactating goats in early lactation to evaluate the effect of two inert adsorbents on AFM 1 excretion in milk from animals receiving AFB 1 at the rate of 100 µg/kg in the diet for 14 days. In the control group (animals who ingested AFB 1 contaminated feed without adsorbents), the carry-over rate measured after three and 14 days was 0.140% ± 0.049 and 0.397% ± 0.124, respectively [72]. Similar results were obtained in an almost identical experiment carried out by Mugerwa et al. In the goats of the control group, the carry-over rate measured after three and 14 days was 0.11 and 0.30%, respectively [73].
The surveys on the presence of AFM 1 in donkey milk are really very few and in many cases, the number of samples analyzed is low. The findings of such studies show no or few positive samples and low concentrations of mycotoxin, as reported in Table 3. With respect to the table above, in the survey carried out in Serbia in 2013, the authors suggest that the high contamination frequency observed could be consequence of the hot and dry weather conditions of 2012 that lead to the development of aflatoxigenic molds in maize and feed [74]. The study carried out in Croatia in 2013 does not report the percentage of positive samples, but it specifies that AFM 1 levels exceeded the LOQ value (23.2 ng/L) in only 59 samples of cow milk and two samples of goat milk. The authors state that these positive samples were linked to the use of contaminated supplementary feedstuff [75]. The survey carried out in Sicily from 2013 to 2016 reports the complete absence of AFM 1 in donkey milk, and that such milk comes from animals that lived freely [57]. Since the feed of grazing animals consists mainly of grass and hay, which are not major sources of intake for toxins, the possibility that these animals assume AFs through the feed is very low. The findings of the survey carried out in Greece in 2015 are in line with those of the studies mentioned above (few positive samples and low concentrations detected). According to the authors, these data can be due to the type of feed (the animals were reared on pasture), and to the very low carry-over of aflatoxin B 1 to M 1 that has been reported for donkeys [76]. Finally, also in the case of the study carried out in Germany, the authors suggest that the absence of AFM 1 in the milk samples can be explained by the feeding of the animals that consisted of grass and hay, supplemented by pelleted feed and oat grains, which were tested negative for AFB 1 [77].

Conclusions
In conclusion, the data of the present study together with those of the other surveys carried out on donkey milk seem to be reassuring about the safety of this type of milk. However, as the main consumers of this food are infants and young children, it is worth stressing the importance of regularly monitoring donkey milk, although it is a niche product, including checks on the feed administered to the animals in order to ensure its healthiness. The present study is only a survey focusing on one set of samples from one single farm. Considering the few data on the mycotoxin contamination of this type of milk (to date only one survey carried out in Italy), it would be interesting to continue the research extending it to the milk from other donkey farms to have a more complete picture of the AFM 1 risk in this food.