Analysis of aflatoxins in traditional Chinese medicines: Classification of analytical method on the basis of matrix variations

A classification system for analytical methods was developed for the first time to determine the presence of aflatoxins B1, B2, G1 and G2 in traditional Chinese medicines (TCMs) based on different matrix types using ultra-performance liquid chromatography–tandem mass spectrometry. A useful characteristic of the approach was that the TCMs could be systematically divided into four categories (i.e., volatile oils, proteins, polysaccharides and fatty oils) depending on the matrix types. The approach concluded that different types of TCMs required different optimal sample preparation procedures. Based on the optimized analytical conditions, the limits of detection and quantification, average recoveries and linearity of four aflatoxins were determined and conformed to research limits. Of 22 TCMs samples, 14 samples were contaminated with at least one type aflatoxin at concentrations ranging from 0.2 to 7.5 μg/kg, and the average contents of aflatoxins were significantly different for the different matrix types. Moreover, we found a potential link between the contamination levels of aflatoxins and matrix types. TCMs containing fatty oils were the most susceptible to contamination by aflatoxins and followed by TCMs containing polysaccharides and proteins; TCMs containing abundant amounts of volatile oils were less prone to contamination.

Due to the complexity of TCMs, the matrix effect has become a main factor that has affected the accuracy of detecting AFs in TCMs 9,23 . Thus, the methods used for sample pre-treatment are very important for the accurate detection of AFs in TCMs. Sample pre-treatment mainly includes extraction and purification processes, and existing literature reports have shown that samples of different matrix types can be adopted using appropriate sample pre-treatment methods 9,18,23,24 . For instance, samples with fatty oils have high proportions of fatty oil contents; Huang B et al. adopted an extraction method using homogenization and a reliable solid phase extraction-based clean-up method to process such samples 25 . For cereal samples with high protein and polysaccharide (starch) contents, the extraction methods for AF samples frequently used ultrasonography 12,26 and clean-up methods for AFs employed solid-phase extraction (SPE) methodologies 27,28 . The studies described above mostly utilized complex sample processing methods to accurately determine the presence of AFs in one or several samples. However, to accommodate the large number and variety of TCMs, much work is required to develop corresponding sample pre-treatment methods. If classified sample pre-treatment mode were to be established, the accuracy of the measurements and efficiency of work would improve.
Scholars have adopted classification methods 18,29 , and TCMs have always been divided into different medical parts, i.e., rhizomes, roots, seeds, flowers, grasses and leaves, which were extracted and purified using the same procedures depending on the medical parts. However, this classification method has some defects. For example, there may be a major difference in the matrix of the same medicinal parts of different medicinal materials. The same sample pre-treatment methods developed by such researchers were not suitable for extracting or detecting AFs in TCMs.
The aim of this work was to develop a novel classification of analytical method to detect aflatoxins B 1 , B 2 , G 1 and G 2 in widely applied TCMs based on different matrix types. Research efforts have focused on the influence of different sample pre-treatment methods on the samples and optimization of UPLC-MS/MS parameters. This research can offer a reference for systematically establishing analytical methods for the detection of AFs in TCMs. Meanwhile, data on the contamination levels of AFs and the contents of different matrix types were processed and analyzed using statistics software, and an inner relationship was found, which could be used to infer the susceptibility of fungus contamination based on the matrix types of TCMs and to provide a reference point for the safety of TCMs.

Results and Discussion
Analysis results of sample matrix types. The matrix types of all 22 TCMs were divided according to the contents of the basic components. The results are shown in Table 1. By comparing the content ratios of the four types of components in each sample, all 22 samples were divided into four matrix types, i.e., volatile oils, proteins, polysaccharides and fatty oils. In our work, the polysaccharide content was determined to be 2.2-31.8% in the 22 samples, among which the content of polysaccharides of five samples was larger than 20%; these five samples were eventually categorized into the polysaccharides group. Of the TCMs, 6 out of the 22 samples were classified as volatile oils, in which the range of the volatile oil content was between 1.6% and 2.9%. Similarly, fatty oil and protein contents ranged from 23.8% to 50.5% and from 15.9% to 21.3%, respectively, and these samples were categorized as the fatty oils and proteins, respectively.
Moisture contents of samples were tested based on the Chinese Pharmacopoeia (2015). The results were shown in Supplementary Table 1. For volatile oils, proteins, polysaccharides and fatty oils, the moisture content was 7.52-8.92%, 7.49-10.44%, 7.16-10.54%, and 5.37-7.13%, respectively. Moisture content results met the requirements of Chinese Pharmacopoeia.
Optimization of the extraction procedure. For the TCMs samples of four different matrix types, the effectiveness of various extraction methods was investigated. Four duplicate samples of the four types of matrices were extracted through shaking, homogenizing and ultrasonicating the samples. By comparing the extraction efficiencies of three methods, each sample of the four types of matrices required its own extraction methods (Fig. 1). Based on the results, ultrasonic extraction was selected as the best extraction method for the protein and volatile oil samples. Shaking extraction methods were determined to be the optimal methods for the samples of polysaccharides, and homogenization extraction was chosen for fatty oils. Because TCMs with high contents of fatty oils and polysaccharides were more viscous, an ultrasonography extraction method was prone to aggregating the extracts, and its use to extract AFs was not conducive to dissolution of the compounds.
In addition, to allow for higher extraction efficiencies, the extraction solvents and time were optimized. Five ratios of extraction solvents were investigated: 65%, 70%, 75%, 80%, and 85% aqueous methanol solutions were used for the samples of each type, and the samples were also subjected to different extraction times. The results of this optimization study are shown in Supplementary Fig. 1. For volatile oils, the samples were extracted in 75% aqueous methanol using ultrasonography for 45 min. The samples containing proteins were sonicated in 85% aqueous methanol for 45 min. For the samples with polysaccharides, they were extracted in 70% aqueous methanol for 3 h with shaking, and the samples of fatty oils were homogenized in 70% aqueous methanol for 4 min.
Because the samples had different matrices, each category of the samples required the use of a different extraction method. The obtained results were consistent with observations reported in previously published articles. A.S. Luna et al. conducted research on peanuts with more oil and used a homogenization extraction method to process the samples 30 . Wen J. et al. adopted an extraction procedure using ultrasonication to extract AFs from ginger and products related to volatile oils 31     mycotoxins in Coix seeds 32 . However, in our work, shaking extraction was an optimal extraction method for samples containing polysaccharides.
Optimization of the clean-up procedure. To optimize extraction efficiencies and the recovery of materials, different methods were tested and compared. In our study, the use of Welchrom C18E columns and silica gel columns for the clean-up procedures after extraction was evaluated. The first two methods were compared to samples that were not subjected to purification methods, which showed that the recovery no purification > the recovery C18 columns > the recovery silicagel columns (Supplementary Table 2). Because the fatty samples contained more nonpolar and weakly polar compounds which could pollute and damage the UPLC column and consequently shorten the service life of the column upon purification, the samples needed to be processed after being subjected to a clean-up procedure. In general, the three types of TCMs mentioned above were extracted without purification, which resulted in a higher recovery rate and lower loss rate. Samples of fatty oils were purified by C18-SPE columns to protect the columns against damage, and the obtained recovery was 70-110% using the clean-up method and matched the recovery amount of the standard.
Method validation. The ranges of linearity, the coefficients of determination and correlation, as well as the limits of detection (LOD) and quantification (LOQ) for each aflatoxin were determined. The working standard solutions of AFs were diluted immediately with methanol from the original stock solutions every weekday and which were used to make the mixed working standards. A set of four standard solutions containing different concentrations in the range of 0.0502-10.4 ng/mL for AFB 1 , 0.0350-7.0 ng/mL for AFB 2 , 0.0295-11.8 ng/mL for AFG 1 and 0.0295-11.8 ng/mL for AFG 2 , which were prepared in methanol and were used for method calibration. These solutions were kept at − 20 °C and were renewed weekly. The linearities obtained for all the analytes were good, and the correlation coefficients (R 2 ) ranged from 0.9985 to 0.9996. LOD and LOQ values were 0.008-0.022 μ g/kg and 0.011-0.029 μ g/kg, respectively, which showed that the method developed, met the EU legislative requirements of 2 and 4 μ g/kg for AFB 1 and total AFs contents. The relative standard deviation (RSD) of precision at the middle concentration of the AFs mixture was 2.9-6.7% (n = 6). The data are shown in Table 2.
Recovery estimations were carried out using the standard addition method, which comprised three spiked samples at different levels. Different types of TCMs were used for the recovery test to ensure that the method had broad applicability. Each sample was selected at random, and aliquots (n = 9) of the samples were spiked with the mixed standard solutions at a high concentration level (10.4 ng/mL for AFB 1 , 3.5 ng/mL for AFB 2, 11.8 ng/mL for AFG 1 and 5.9 ng/mL for AFG 2 ), a medium concentration level (4.16 ng/mL for AFB 1 , 1.4 ng/mL for AFB 2, 4.72 ng/ mL for AFG 1 and 2.36 ng/mL for AFG 2 ) and a low concentration level (1.04 ng/mL for AFB 1 , 0.35 ng/mL for AFB 2, 1.18 ng/mL for AFG 1 and 0.59 ng/mL for AFG 2 ). In general, a sample (2.0 g) was spiked with high, medium or low levels of the AF standards; and were treated and tested following the procedures outlined above. All recovery amounts ranged from 80.4% to 103.3% (Table 3). The spiked samples were extracted and analysed by UPLC-MS/ MS, as previously described.
For the four AFs the results indicated good accuracy of the method for the detection of aflatoxins B 1 , B 2, G 1 , G 2 in TCMs of different matrix types, and the recoveries were also in compliance with the requirements of the European Union (70-110%).
Method application. Following the optimization and validation of the analytical approach, it was successfully utilized to determine the contamination levels of four AFs in 22 classified TCMs. The levels of total and individual AFs are summarized in Table 1.Typical UPLC-MS/MS chromatograms of the four AFs in standard solutions (A) and in contaminated samples (B) are shown in Supplementary Fig. 2. Of the 22 samples, 14 samples were detected to be positive with four AFs at concentrations ranging from 0.2 to 7.5 μ g/kg, and 13 samples were detected to be contaminated with AFB 1 . The incidence rate was as high as 63.6%, and four positive samples (18.2%) exceeded the maximum limit set by the European Union (4 μ g/kg). With regards to individual AFs, the levels of AFB 1 , AFB 2 , AFG 1 , and AFG 2 were detected in ranges of 0.   Correlation analysis. To further analyse the contamination levels of the 22 TCMs, we compared the contents of AFB 1 and AFs in the samples of four matrix types. The effects of the matrix types on the contamination levels of AFs are thought to be due to their different abilities for breeding fungus. The average contamination levels of AFB 1 and total AFs in the samples of four matrix types are shown in Fig. 2. The content of AFs in the samples of different matrix types was varied significantly. Results showed that TCMs with an abundant of fatty oils had the highest amounts of AFB 1 and total AFs, while these contamination levels were very low for samples with an abundance of volatile oils. Furthermore, the internal relation between the contamination levels of AFs and matrix types was studied. In our study, the results obtained by Pearson correlation analysis indicated that the contents of AFB 1 and total AFs had varying degrees of influence on the different matrices. As shown in Table 4, the contents of AFB 1 and AFs were negatively correlated with the contents of volatile oils, and the correlation coefficients (r) were − 0.612 and − 0.556 (P < 0.05). respectively. The content of fatty oil exhibited a positive correlation to the contamination levels of AFB 1 (r = 0.661, P < 0.01) and AFs (r = 0.749, P < 0.01). The contents of AFB 1 and AFs were not significantly positively correlated with the contents of polysaccharides and proteins.
Our results indicate that TCMs with fatty oils may easily multiply Aspergillus flavus and A. parasiticus, resulting in the production of secondary metabolites (AFs). Polysaccharides and proteins also provided nutritional ingredients for fungus and promote their growth; the contents of AFs were relatively high in both types of TCMs. TCMs with volatile oils, such as Fructus Tsaoko, Fructus Anisi Stellati and Flos Caryophylli contain the active chemical components, known as essential oils, which possessed antifungal effects that reduced or prevented fungal infection and subsequent AFs production. The essential oils can decrease the damaged effect of aflatoxins by two different ways. Firstly, DNA binding formation of aflatoxins is reduced by essential oils. Secondly, aflatoxins cause increase of reactive oxygen species and essential oils react with reactive oxygen species. Therefore, essential  Table 3. Recovery results of AFB 1 , AFB 2 , AFG 1 and AFG 2 a (%). a Each value represents the mean ± SD of at least three measurements.  Table 4. The correlation between the contents of volatile oils, fatty oils, polysaccharides, and proteins in AFB 1 and total AFs. * * extremely significant, P < 0.01; * significant, P < 0.05. oils protect the cells from harmful impact of aflatoxins 33,34 . Similar results have been reported for studies conducted on Ocimum basilicum L. 35 , Radix Puerariae Lobatae and Semen Persicae samples 18 .

Conclusions
In this study, a classification method for the simultaneous detection of AFB 1 , AFB 2 , AFG 1 and AFG 2 in TCMs based on matrix types was established by UPLC-MS/MS for the first time, and the classification approach was successfully applied to analyse a total of 22 different matrix types of TCMs. This study provides a novel research approach for establishing the use of analytical methods to detect AFs in a large number of TCMs. Furthermore, we found that there was significant relationship between matrix types and the contamination levels of AFs. The contents of fatty oils, polysaccharides and proteins to the contamination levels of AFB 1 and AFs were positively correlated, whereas the contents of AFs were negatively correlated with the contents of volatile oils. Meanwhile, a possible association between the contamination levels of AFs and the different matrix types of TCMs was presented. The possibility for AFs contamination of medicinal materials containing fatty oils and polysaccharides was high, but the possibility of those containing volatile oils was low. These results indicate that the processing and storage methods used for medicinal materials are likely associated with the matrix types of their components, especially regarding the amounts of fatty oils of TCMs.

Methods
Materials and reagents. AF standards including AFB 1 , AFB 2 , AFG 1 and AFG 2 were purchased from Sigma-Aldrich (St. Louis, MO, USA). Solid powders of each aflatoxin standard were weighed accurately, and the standards were dissolved in methanol to prepare stock standard solutions and stored at − 20 °C in a dark place. Distilled water was purified using a Milli-Q Gradient A 10 system (Millipore, Billerica, MA, USA). Acetonitrile, methanol and formic acid were of LC grade (Merck, Darmstadt, Germany). All the other solvents were of analytical grade. Welchrom C18E (500 mg/3 mL) columns were purchased from Welch (USA).
A total of 22 samples were randomly purchased from June to August 2014 from several local markets and drug stores in Chongqing China; the samples were authenticated by Professor Dan Zhang at Chongqing Medical University. All the samples were ground into powders, sieved through a 60-mesh filter and stored in sealed plastic bags below 4 °C for further analysis.
UPLC-MS/MS analysis. The UPLC chromatography system (Shimadzu Corp., Kyoto, Japan) was equipped with a solvent delivery pump (LC-30AD), an auto-sampler (SIL-30AC) and a column oven (CTO-20AC). The separations were performed on a Phenomenex Luna 3 μ C18 (2) 100A column (50 × 2.00 mm) (Phenomenex, USA). Chromatographic analyses were carried out using a gradient elution, where eluent A was an aqueous solution of ammonium formate (5 mM) and eluent B consisting of acetonitrile. The analysis started with 30% of acetonitrile, which was held for 0.5 min, and was then changed to 80% acetonitrile at 4.5 min and held 1.5 min. Then, the eluent was changed to 30% acetonitrile at 6.1 min. The column was conditioned with 30% acetonitrile for 1.9 min before the next injection. The flow rate was set at 0.35 mL/min, and the injection volume was 3 μ L. Moreover, the column temperature was maintained at 30 °C.
Electrospray mass spectrometry (ESI-MS) was carried out using an API 4000 triple-quadrupole instrument from Applied Biosystems (AB Sciex, Framingham, MA, USA), equipped with an electro-spray ionization (ESI) source. The mass spectrometer was operated in positive ESI modes with multiple reaction monitoring (MRM) at unit mass resolution. Data acquisition and processing of the ESI-MS were obtained using Analyst TF software (AB Sciex), and the accurate mass data for the molecular ions were processed by PeakView TM 1.1.1 software (AB Sciex).The source/gas conditions were as follows: the curtain gases CAD and CUR were set at 4 and 25 psi, respectively. The ion source gas 1 (GS1) and ion source gas 2 (GS2) were set at 55 psi and 55 psi, respectively. The ionization source of the MS/MS detector had a capillary voltage of 5.5 kV, and the source temperature was set to 600 °C. The compound conditions were Entrance Potential (10.0) and Collision cell potential (12.0). The MRM transitions, applied cone voltages and collision energies are summarized in Table 2.

Analysis of samples matrix types.
To determine the matrix composition of various medicinal materials, the contents of volatile oils, proteins, polysaccharides and fatty oils of 22 samples were determined. The contents of these materials were determined according to the Chinese Pharmacopoeia (2015), the Kjeldahl determination method 36 , the phenol-sulfuric acid method 37 and the Soxhlet extraction method 38 , respectively. The content ratios were then calculated to classify the samples according to the matrix types.

Sample Preparation
Extraction. To optimize the extraction procedure of AFs in TCMs, the influence of different extraction methods and factor levels based on the classification results of different matrix was investigated. (1) Extraction methods: For the TCMs of four matrix types, duplicate samples of each type were extracted through shaking, homogenization and ultrasonication. (2) Extraction solution: Five different ratios of extraction solvents were investigated: 65%, 70%, 75%, 80%, and 85% aqueous methanol solutions were used for samples of each type. (3) Extraction time: Samples of four matrix types were extracted for four different periods of time. Four different extraction procedures were then used for samples of different matrix types, which are described below: Volatile oils: A 2 g portion of ground sample was soaked in 10 mL of a methanol/water (75:25, v/v) solution for 1 h and was sonicated for 45 min. The sample was then centrifuged at 3000 rpm for 5 min, and1 mL of the supernatant was filtered through a 0.22 μ m syringe filter prior to analysis.
Proteins: A 2 g portion of ground samples was soaked in 10 mL of a methanol/water (85:15, v/v) solution for 1 h and was sonicated for 45 min, The following procedure was the same as that used for the extraction procedures for volatile oils.
Scientific RepoRts | 6:30822 | DOI: 10.1038/srep30822 Polysaccharides: A 2 g portion of a ground sample was extracted in 10 mL of a methanol/water (70:30, v/v) solution for 3 h by shaking the sample. The sample was then centrifuged at 3000 rpm for 5 min, and 1 mL of the supernatant was filtered through a 0.22 μ m syringe filter prior to analysis.
Fatty oils: A 2 g portion of a ground sample was homogenized in 10 mL of a methanol/water (70:30, v/v) solution for 4 min and was centrifuged at 3000 rpm for 5 min. Then, 2 mL of the supernatant was subjected to the for clean-up procedure.

Clean-up.
To evaluate the efficiency of the clean-up procedure, results obtained using Welchrom C18E columns and silica gel columns were compared to samples that were not subjected to a purification method.
Samples of fatty oils were purified using the following procedure. A 2 mL aliquot of the final filtrate was passed through a Welchrom C18E column. The C18E column was pre-treated with 6 mL methanol before washing it with 6 mL distilled water. After the sample was loaded into the column, the column was first washed with 6 mL distilled water, and then the C18E column was rinsed with 4 mL methanol. The obtained elutes were completely evaporated under a steam of nitrogen gas at 30 °C, and the sample was re-dissolved in 1 mL methanol. The solution containing the AFs was vortexed for 30 s, and approximately 50 μ L of the solution was filtered through a 0.22 μ m filter. A 3 μ L aliquot of the filtrate was injected into the UPLC-MS/MS system.

Method validation.
Quantification of the AFs in TCMs followed testing for linearity, recovery, LOD and LOQ. To check the linearity of the method, calibration curves based on the peak area were constructed in the range of 0.0295-11.8 ng/mL. To interpolate the results, concentrations outside the calibration range were performed with proper dilutions. Quantification was performed by plotting concentration versus peak area, and the regression curve was evaluated by using variance (ANOVA) analysis.
The LODs were obtained using a signal-to-noise ratio of S/N = 3:1, and the LOQ was considered the lowest point of the calibration curve that was adopted when the concentration of a compound resulted in S/N = 10:1.
Recovery analysis was performed by testing replicate spiked samples at three different concentrations (low, medium and high levels). The recovery values were estimated by relating the concentration of the AFs found to the expected concentration.
Statistical treatment of data. To obtain further details of the differences, the UPLC-MS/MS datasets of the four groups were subjected to correlate analyses. The contents of AFB 1, total AFs (AFB 1 , AFB 2 , AFG 1 and AFG 2 ) and the content ratios of the four types of matrices were expressed as mean ± standard deviation of three replicates. The significance of each group was checked by a one-way analysis of Variance (ANOVA) followed by a Pearson correlation. A bivariate correlate analysis was used to determine the relationship between AFB 1 contents, total AF contents and the content ratios of the four different types of matrices. Correlate analysis was conducted using SPSS 19.0 statistical software (SPSS Inc., Chicago, IL, USA). The significant value was set at P < 0.05.