Aristolochic acid A

Interference-free spectrofluorometric quantification of aristolochic acid I and aristololactam I in five Chinese herbal medicines using chemical derivatization enhancement and second-order calibration methods

Abstract
A rapid interference-free spectrofluorometric method combined with the excitation-emission matrix fluorescence and the second-order calibration methods based on the alternating penalty trilinear decomposition (APTLD) and the self-weighted alternating trilinear decomposition (SWATLD) algorithms, was proposed for the simultaneous determination of nephrotoxic aristolochic acid I (AA-I) and aristololactam I (AL-I) in five Chinese herbal medicines. The method was based on a chemical derivatization that converts the non-fluorescent AA-I to high-fluorescent AL-I, achieving a high sensitive and simultaneous quantification of the analytes. The variables of the derivatization reaction that conducted by using zinc powder in acetose methanol aqueous solution, were studied and optimized for best quantification results of AA-I and AL-I. The satisfactory results of AA-I and AL-I for the spiked recovery assay were achieved with average recoveries in the range of 100.4- 103.8% and RMSEPs less than 0.78 ng mL-1, which validate the accuracy and reliability of the proposed method. The contents of AA-I and AL-I in five herbal medicines obtained from the proposed method were also in good accordance with those of the validated LC-MS/MS method. In light of high sensitive fluorescence detection, the limits of detection (LODs) of AA-I and AL-I for the proposed method compare favorably with that of the LC-MS/MS method, with the LODs less than 0.35 and 0.29 ng mL-1, respectively. The proposed strategy based on the APTLD and SWATLD algorithms by virtue of the “second-order advantage”, can be considered as an attractive and green alternative for the quantification of AA-I and AL-I in complex herbal medicine matrices without any prior separations and clear-up processes.

1.Introduction
Nephrotoxic aristolochic acids (AAs) are a type of nitrophenanthrene carboxylic acids derivatives that root in Aristolochiaceae plants including Aristolochia and Asarum genera[1]. Most of the components have been revealed to be nephrotoxic[2-4], carcinogenic[5, 6] and mutagenic[7, 8]. Among them, AA-I, a major compound in AA-containing herbs, was found to be the most toxic and attracted the most attention[9]. The metabolites of AA analogues in organisms seem to include the relative aristololactam (AL), N-OH AL, 7-OH AL and DNA adducts. Some studies have revealed that the majority of AA-I and AA-II were reduced to their corresponding aristololactams[5, 10]. AL-I and AL-II were also found to be the main metabolites of AA-I and AA-II after oral administration in experimental mice[11]. What is more, aristolochic acids (AAs) are often accompanied by aristololactams (ALs) in Aristolochiaceae plants including Aristolochia and Asarum genera. These AA-containing herbs had been widely used as Chinese herbal medicines to treat tumors, snake bites, small pox, rheumatism and pneumonia, until AAs were confirmed to be severe human nephrotoxins, carcinogens and mutagens. Therefore, the use of AAs-containing herbs has already been banned all over the world. Regardless of the prohibition of using AA-containing plants, these herbs are still been sold via the internet and the misuse of AA-containing herbs also exists[12].

In fact, as the main active components in plants of the aristolochia species, AA-I and AL-I also possess severe nephrotoxicity, potent carcinogenicity and mutagenicity. There is few studies simultaneously analyzed AA-I and AL-I, since the considerable discrepancy of their concentration levels in the AA-containing plants and the lack of high-sensitive approaches. Therefore, it is of great significance to develop a high sensitive and specific method for the trace analysis of AA-I and AL-I in herbal medicines and botanical products.
In the past years, many attempts have been made to detect the AA analogues in AA-containing herbs using various methods including thin-layer chromatography(TLC)[13], high-performance liquid chromatography with ultraviolet detection (HPLC-UV)[14-16], high-performance liquid chromatography with mass spectrometry (HPLC-MS)[17-20], capillary electrophoresis with electrochemical detection (CE-ECD)[21-23] and enzyme-linked immuno sorbent assay (ELISA)[24]. Very recently, several reports have been published for the determination of AAs by converting the non-fluorescent AAs into the corresponding substances that intensively fluoresce via high-performance liquid chromatography with fluorescence detection (HPLC-FLD)[14, 25-27]. As one of the most sensitive and selective analytical tools, HPLC-FLD is highly selective and robust for the substances that fluoresce. Considering that complete separation of analytes and complex matrix is time-consuming and solvent-wasting, it is necessary to develop a high-performance method within the framework of green analytical chemistry.
By virtue of being fast, high signal-to-noise, ease of use and simple equipment of instruments, fluorescence spectroscopy is an attractive tool for the determination of fluorescent substances with high sensitivity.

However, fluorescence detection cannot provide moderate selectivity, especially for mixtures and complex matrices. Therefore, analyte extraction and sample preparation are often required, which are time-costing and labor-consuming. In addition, analyte extraction may lose a portion of analyte(s), which could cause bias in the prediction of content for analyte(s). For conventional fluorescence spectroscopy, one can obtain zero-order fluorescence signal at the maximal excitation and emission wavelength. However, classical zero-order (one-way) calibration requires signal must be fully selective for the analyte of interest[28]. When measuring fluorescence excitation or emission spectra, one can acquire first-order fluorescence signal. For the first-order (two-way) calibration methods[29-31], unbiased quantification can be accomplished in the presence of spectral interferents with the prerequisite that the calibration set must consist of all interferents[32]. To circumvent aforementioned issues, excitation-emission matrix fluorescence (EEM), which is obtained by measuring emission spectra at multiple excitation wavelengths, coupling with some state-of-the-art chemometrics tools such as second-order calibration methods have been proposed[33-37]. These strategies allow the concentration and spectral information of the analytes to be extracted directly from the complicated spectroscopic measurements, even in the presence of uncalibrated compounds. This characteristic, namely “second-order advantage”[28, 38-40], has been applied in many fields such as chemistry[41-43], food safety[44, 45], environmental science[46, 47] and pharmaceuticals[48, 49].

AA analogues can be mainly classified into two groups: AAs and ALs, according to their chemical structures. Being rigid polycyclic aromatics, ALs have strongly intrinsic fluorescence. As a result of the electron-withdrawing effects by carboxyl and/or nitro group, AAs do show no fluorescence. The nitro group of AAs can be easily reduced. Thus, one can convert non-fluorescing AAs into the corresponding ALs that fluoresce by a variety of reductive agents [25-27, 50]. Herein, we choose the Zn/H+ to reduce AA-I into AL-I (in Figure 1) for the quantification of AA-I and AL-I in five Aristolochiaceae plants. In light of complex matrices and analogical properties for ALs, the strategy of EEM combining with second-order calibration methods based on APTLD and SWATLD algorithms was proposed, which is in accordance with the green analytical chemistry principles, since it implies lessening time and cost of analysis.

2.Theory
2.1.Chemometrics analysis
2.1.1. The trilinear component model for second-order calibration
In the second-order calibration field, the most commonly used model is trilinear component model (namely, PARAFAC/CANDECOMP model) [51, 52], which was proposed by Harshman, Carroll and Chang independently. When K excitation-emission matrices fluorescence (EEMs) of samples are obtained by acquiring at I excitation wavelengths and J emission wavelengths, these can form a three-way data array X with the size of I × J × K. Each element of X can be expressed as:xijk= ∑N ainbinckn+eijk, for i=1,2⋯,I; j=1,2⋯,J;k=1,2⋯,K where xijk is the fluorescence intensity of the kth sample at ith excitation wavelength and jth emission wavelength; ain, bjn and ckn are the elements of underlying excitation spectrum matrix AI×N, emission spectrum matrix BJ×N and relative concentration matrix CK×N, respectively, in which N denotes the number of detectable species, including the components of interest and the backgrounds as well as uncalibrated interferents in the system; the term eijk is the element of a three-way residual array E.

2.1.2. APTLD and SWATLD algorithms
In this work, the alternating penalty trilinear decomposition (APTLD) and the self-weighted alternating trilinear decomposition (SWATLD) algorithms were implemented to decompose the three-way data array consisting of K EEMs that originated from the calibration set and herbal medicine samples. Owing to their powerful property, “second-order advantage”, the spectral profiles and contents of AA-I and AL-I were retrieved from the complicated herb plant samples.The APTLD and SWATLD algorithms were developed by our group [34, 37], which were derived from the alternating trilinear decomposition (ATLD) algorithm [33]. The APTLD decomposes three-way data arrays by using the alternating least-squares principle and the alternating penalty constrains to simultaneously minimize three different alternating penalty errors. The SWATLD algorithm is unique in that it alternately minimizes three different objective functions with inherent relationship. Both of them can obtain reliable results in most cases, avoid two-factor degeneracy problem, hold fast convergence and be insensitive to the excess factors used in calculation. The APTLD and SWATLD theories were exhaustively documented in the original literatures [34, 37].

2.2.Figures of merit
Figures of merit are mainly employed to estimate the performance of the calibration method. Sensitivity (SEN) and selectivity (SEL) are important figures of merit in multi-way analysis, regularly employed for comparison of the analytical performance of the methods. Different approaches have been discussed in the literature for computing them in second-order calibration methodologies [53-55]. According to Olivieri et al. [55], the SEN and SEL for the nth analyte can be expressed as follows: where subscript ‘nn’ indicates the (n,n) diagonal element of a matrix. The parameter kn is the total signal for the analyte of interest at unit concentration. It is worth noting that the full sensitivities (kn) decrease in the presence of other sample constituents, by a degree that depends on the overlapping of their profiles.The limit of detection (LOD) and limit of quantitation (LOQ) for a given analyte are also important parameters to be reported

3.Experimental
3.1.Chemicals and reagents
Aristolochic acid I (AA-I, 99.8% purity) was purchased from the National Institute for the Control of Pharmaceutical and Biological Products (NICPBP, Changsha, China). Aristololactam I (AL-I, 98% purity) was supplied by Zhanshu Chemical Technology Co., Ltd (Shanghai, China). Methanol and acetonitrile were obtained from Oceanpak (Goteborg, Sweden). Ammonium formate (99.99%, Aladdin, China) was of analytical grade. Water of 18.2mΩ was produced by a Milli-Q Ultrapure water purification system (Millipore, MA, USA). Zinc powder (>95% purity) of 200 meshes and acetic acid (99.5% purity) were obtained from Sinopharm Chemical Reagent (Shanghai, China). Other reagents were of analytical grade, including dehydrated alcohol. Stock solutions of AA-I (50 μg mL-1) and AL-I (55 μg mL-1) were prepared by dissolving with dehydrated alcohol in brown volumetric flasks, respectively, and stored at 4 ℃in the dark for 2 months.

3.2.Materials and sample preparation
Five representative herbal medicines were investigated to validate the reliability and robustness of the proposed method, which are listed in Table 1. Prior to analysis, individual herbal medicines were powdered and sieved (80 meshes). Approximately 100 mg of the powdered herbal samples were accurately weighted and extracted by 4 mL methanol. The solution was sonicated at room temperature for 15 min, and centrifuged at 4,000 rpm for 15 min. The supernatant was collected and the residual was re-extracted in the same manner. The solvent extracts were combined and stored in a refrigerator at 4 ℃.Appropriate volume of the sample extracts were diluted to 5.00 mL with acetose methanol aqueous solution (v/v, methanol: water: acetic acid, 80: 19.5: 0.5) and the dilution factors of sample extracts were list in Table S1 (see Supplementary material).Then the chemical derivatization that convert the AA-I into AL-I was conducted by adding 10 mg zinc powder at room temperature for 20 min. The reaction mixture was filtered through a 0.45 μm pore size nylon membrane (Membrane Solutions, Texas, USA) before the filtrate was measured by excitation-emission matrix fluorescence.

3.3.Apparatus and software
EEM measurements were performed on an F-7000 fluorescence spectrophotometer (Hitachi, Japan) equipped with a Xenon lamp. The excitation wavelength varied from 240 to 440 nm (2.0 nm steps) and the emission wavelength varied from 420 to 570 nm (2.0 nm steps) with a scanning rate of 12,000 nm min-1 (4.5 min per sample), and the detector voltage was 750 V. The slit widths of excitation and emission monochromator were 5.0 nm. The spectral were saved and exported the data were exported with text format.All the programs used in this paper were written in-house under the MATLAB (Mathworks, Inc., MA, USA) environment and carried out on a personal computer equipped with a Core i7 processor with 8 GB RAM under Windows 7 operating system.

3.4.Three-way EEMs data array
The calibration set of AL-I was constructed by taking seven concentration levels which were prepared by diluting suitable volume of AL-I stock solution to 5.00 mL with acetose methanol aqueous solution (v/v, methanol: water: acetic acid, 80: 19.5: 0.5). To obtain the calibration set of AA-I, a set of the AA-I at five concentration levels was prepared by mixing appropriate volume of AA-I stock solution with 10 mg zinc powder, reacting and performing EEM measurements as mentioned above. The concentration of AL-I and AA-I ranged from 5.5 to 148.5 ng mL-1 and 22.5 to 112.5 ng mL-1, respectively. The validation set consists of three spiked samples. Each of them was prepared as follows: volume of 25 μL madouling (MDL) extract was spiked with specific content of AL-I or AA-I within calibration range, and were diluted to 5.00 mL with acetose methanol aqueous solution. Table 2 listed the concentrations of AL-I and AA-I in the calibration set and validation set.The quantitative analysis of AA-I was performed after the AA-I was converted to AL-I. To obtain the contents of AA-I and AL-I in the herbal samples, we firstly quantified the AL-I in the samples without derivatization, then, determinated the concentration of AL-I in the samples after derivatization, and the contents of AA-I were calculated by subtracting the levels of AL-I in herbal samples before derivatization from the total AL-I levels in the corresponding derivative samples. Depending on the levels of AA-I and AL-I in each individual herbal sample, the extract might need to be diluted in order to adjust the contents of AA-I and AL-I within the calibration ranges and minimize inner filter effects (The dilution factors detailed in Table S1). The two prediction sets consist of 5 herbal samples which were used to detect the concentration of AA-I and AL-I in five types of herbal medicines, respectively. Three blank samples were prepared for estimating LOD and LOQ.

3.5. LC-MS/MS analysis for method comparison
The herbal samples were also analyzed by LC-MS/MS on an Agilent 1290 Infinity LC coupled with a 6460 triple quadrupole mass spectrometer (Agilent Technologies, Waldbronn, Germany) fitting with an electrospray ion source (ESI) for method comparison. Chromatography separation was achieved in a ZORBAX Eclipse XDB-C18 (2.1×150 mm, 3.5 μm) analytical column (Agilent Technologies, Palo Alto, USA), and eluted with the following gradient of 10 mM ammonium formate aqueous(A) and acetonitrile (B) at a flow rate of 0.4 mL min-1 and with column temperature of 50℃ and gradient program: 0-2 min, 10-25 % B; 2-5 min, 25-35 % B; 5-6 min, 35-60 % B; 6-8 min, 60 % B; 8.1-9 min, 10 % B. The LC eluate was analyzed by positive ionization mode with the following parameters: nebulizer pressure, 35 psi; capillary voltage, 4000 V; drying gas temperature, 350 ℃; drying gas flow, 12 L min-1; delta EMV, 300 V. Optimum parameters for compound-dependent were selected to achieve maximal detection sensitivity and were listed in Table S2. Quantification was performed by multiple reaction monitoring (MRM) mode using
the mass transitions (precursor ion → product ion), and fragment ions at m/z 298.0 (359.0 → 323.8 and 359.0 → 298.0) and m/z 279.2 (294.0 → 279.2) were used for the quantitative analysis of AA-I and AL-I, respectively.

4.Results and discussion
4.1.Fluorescence properties of AA-I and AL-I
AA-I, although being rigid polycyclic aromatics, shows no native fluorescence. The presence of the two electron-withdrawing groups (-COOH and –NO2) that restrain its molecular fluorescence. However, when AA-I is converted into AL-I, intense fluorescence can be observed. This is because the electro-withdrawing groups are converted into a lactam ring after the nitro reduction and intramolecular acylation process. Figure S1 shows the three-dimensional landscapes of fluorescence EEMs for the AA-I samples before (A) and after (B) derivatization, which validates the aforementioned fluorescence property of AA-I. Herein, we converted non-fluorescent AA-I into high-fluorescent AL-I for sensitive and simultaneous determination of AA-I and AL-I in herbal medicine samples.

4.2.Optimization of derivatization conditions
In this study, Zn/H+ was used to derivatize AA-I, by reason of its facile operation and high efficiency. To achieve the best derivatization efficiency, we optimized the derivatization conditions, including solvent, amount of zinc powder, reaction time and temperature, and concentration of acetic acid. We first investigated the effect of different methanol-water mixture as the solvent for the derivatization reaction. The fluorescence intensity of AA-I derivative obviously increased with the decrease of water ratio, suggesting that the methanol enhanced the fluorescence intensity of AA-I derivative. And we also found high percentage of water as solvent would result in AA-I derivative break down. As time progresses, it became quite obvious. The results (Figure 2A) suggested that higher percentage of methanol in solvent should be applied for the sake of enhancing the detection sensitivity and stability. Consequently, we chose 80% methanol in solvent.

We next optimized the amount of zinc powder, concentration of acetic acid and reaction time for derivatization of AA-I. Our results demonstrated that the fluorescence intensity of AA-I derivative reached a plateau when the amount of zinc powders was 50 mg (Figure 2B). In addition, the optimal concentration of acetic acid for derivatization of AA-I was investigated. The results showed that the largest fluorescence intensity of AA-I derivative was achieved when the concentration of acetic acid was 0.5 % (Figure 2C). With regard to the optimization of reaction time, the results showed that the derivatization reaction proceeded rapidly. When the reaction time was more than 20 min, the fluorescence intensity increased slightly (Figure 2D). Since the derivatization reaction at room temperature was fast, the temperature effect on derivatization was not further investigated. Taken together, the optimum derivatization conditions for AA-I were added 50 mg zinc powder into acetic methanol aqueous solution (v/v, methanol: water: acetic acid, 80: 19.5: 0.5) for 20 min at room temperature.

4.3. Preprocessing of the EEMs data array
All samples are presenting serious Raman and Rayleigh scattering. These nonlinear factors would result in the three-way data array deviating from the trilinear component model which is the prerequisite of the second-order calibration algorithm to decompose the profile of each mode correctly. To avoid the presence of these signals that affect the decomposition of the obtained three-way data array, we used automated scatter identification method proposed by Bahram and Bro[57], which handled scattering via using interpolation in the identified regions of Raman and Rayleigh scattering (demonstrated in Figure S2).Figure 3 shows the 3D plots of the preprocessed EEMs spectra for the AA-I derivative (A) and xungufeng (XGF) samples with and without derivatization (B and C). As one can see, the EEMs spectra of XGF samples (before and after derivatization) are quite complicated, which may consist of many fluorescent components. There are also no clear resolutions among AA-I derivative and XGF samples. What is more, the spectral interferences of these herbal extracts are varied from each other, which certainly affect the quantification of the analytes using traditional fluorescence spectroscopy. The common strategy usually resorts to chromatography separation of the analytes and background. Whereas, second-order calibration methods in chemometrics are attractive alternative for complicated systems, which could replace or enhance “physical or chemical separation” with the smart “mathematical separation” strategy through separating the signals of the analyte(s) from the uncalibrated interferents and/or background. In this study, we quantified AA-I and AL-I in herbal samples using the APTLD and SWATLD algorithms to decompose the three-way EEMs data array.

4.4.Estimation of the number of components
The number of components, when the APTLD and SWATLD algorithms were applied, was estimated by the core consistency diagnostic (CORCONDIA)[58], which was proposed by Bro et al.. When the core consistency drops from a high value (about 60%) to a low one, this indicates that a suitable number of components has been obtained. Though, the APTLD and SWATLD algorithms both possess the advantage of being insensitive to overestimated number of components. As long as the selected number of components is no less than the underlying real number of components, the APTLD and SWATLD algorithms can provide correct decomposed profiles, and the excess components always appear in the shapes of noises which are easily recognized. Of course, an exact number of components would result in more accurate and reliable decomposition for second-order calibration methods. With the comprehensive consideration of the core consistencies, the RMSECs and the reference spectra, N =4 is an appropriate number for the quantification of AA-I and AL-I in five herbal medicine samples.

4.5.Method validation
The linearity of the fluorescence response for each analytes of interest was investigated independently at the concentration ranges of 1.1 – 302.5 ng mL-1 and 11 – 357.5 ng mL-1 for AL-I and AA-I, respectively. The values of coefficient (R2) were above 0.99. When we designed the concentrations of calibration set, the calibration ranges were selected within the linear concentration ranges.Judging whether a chemometrics model is valid or not, the chemometricians can use the calibration set to validate the calibration model internally through some statistical parameters. The results of quantitative analysis for calibration set obtained by APTLD and SWATLD were listed in Table S3. The values of coefficient (R2) are higher than 0.99, each R2 deems a good linear fit for AA-I and AL-I in their calibration range, respectively. The average recoveries for AA-I are 99.3 ± 4.8 % and 99.2 ± 4.9 %, and for AL-I are 100.6 ± 3.2 % and 100.9 ± 3.9 % obtained by APTLD and SWATLD algorithms, respectively. The RMSEC is the root mean squared error of calibration, which is used to estimate the performance of models. The RMSECs of AA-I and AL-I are 0.89 and 0.15 ng mL-1 for APTLD algorithm, 0.91 and 0.21 ng mL-1 for SWATLD algorithm, respectively. All these results clearly manifest that the trilinear component models decomposed by APTLD and SWATLD algorithms perform quite good.To check the accuracy of the proposed method and to study the interference of matrix, spiked recovery experiment was carried out by standard addition method. It is known that acceptable spiked recovery is a referential criterion for the quantitative analysis. The results listed in Table 3 showed that the spiked recoveries of AL-I and AA-I obtained by APTLD and SWATLD algorithms were very close to 100%. For the AL-I and AA-I, the root mean squared error of predictions (RMSEPs) are 0.35 and 0.69 ng mL-1 for APTLD algorithm, 0.07 and 0.78 ng mL-1 for SWATLD algorithm, respectively. The F-test was carried out among spiked results acquired by APTLD and SWATLD algorithms. As each Fcalculated (2.49 and 2.01 for AL-I and AA-I, respectively) is lower than critical value (Ftable=39.0), it shows that the null hypothesis could not be rejected. Therefore, there is no significant difference in the variances from APTLD and SWATLD algorithms, at the 95% confidence level. The t-test was performed to compare the recoveries of AL-I and AA-I with the ideal value of 100%. The results are also shown in Table 3. As tcalculated were < t30.025=4.30 in all cases, the null hypothesis appears to be valid, i.e., spiked recoveries are close to 100%. At the same time, these results highlighted that the second-order calibration methods based on APTLD and SWATLD algorithms are capable to accurately quantify AL-I and AA-I in complex herbal medicine samples. 4.6.Quantification of AL-I and AA-I in herbal medicines 4.6.1. Analysis of AL-I in herbal samples The proposed strategy was applied for the determination of AL-I in herbal medicine samples, which was determined by analyzing the herbal extracts undergoing no derivatization. The prediction set consists of five herbal medicine samples, which was used to quantify the contents of AL-I in herbal medicines. Firstly, we stacked the calibration set of AL-I and prediction set together to construct a three-way data array with the size of 76 × 101 × 12; then, APTLD and SWATLD algorithms were applied to decompose the three-way data array. According to the core consistency diagnostic (CORCONDIA), we chose N = 4 as the number of components for this three-way data array.The decomposed results of the three-way data array were depicted in Figure 4. It shows actual profiles of AL-I (star) and decomposed profiles of AL-I (blue solid line) in each mode obtained by APTLD and SWATLD algorithms. From the figure, one can easily observe that this is indeed a high collinearity and complicated system. The excitation and emission spectra of AL-I seriously overlap with the uncalibrated interferences in five herbal medicine samples. As a matter of fact, the fluorescence spectra of ALs were extremely similar, which restrains the detection of ALs using conventional fluorescence spectrometry without sample preparations. However, by virtue of "second-order advantage", the APTLD and SWATLD-based second-order calibration methods accurately retrieve the excitation and emission profiles of AL-I from the high-collinear and overlapped spectra. As can be seen, the resolved profiles of excitation and emission modes for AL-I coincide with the actual profiles obtained by the pure AL-I. It is noteworthy to mention that the resolved profiles from APTLD and SWATLD algorithms show delicate difference. The sample modes (C1 and C2) were used for calibration through a classical linear regression of relative concentration profile against real concentration for AL-I. Next, the concentrations of AL-I in the samples were converted to the unit of µg g-1. Table 4 summarizes the prediction results of AL-I in five herbal medicines obtained by APTLD and SWATLD algorithms. A total of five representative herbal medicine samples were analyzed by the proposed method. Among the tested samples, four herbal medicines showed positive results for AL-I and Aristolochia contorta Bunge (Madouling, MDL) was found to possess the highest AL-I content (Table 4), with the concentrations being 933.9 and 915.9 µg g-1 for APTLD and SWATLD algorithms, respectively. 4.6.2. Analysis of AA-I in herbal samples The developed methods were applied to quantify contents of AA-I in herbal medicine samples. The herbal medicine extracts which had been derived as described in section 3.2 were analyzed and the total concentration of AL-I (Ct) was determined. Then, the concentration of AA-I derivative was acquired by subtracting initial concentration of AL-I (Ci) that determined in section 4.6.1 from the total concentration of AL-I (Ct). Consequently, the AA-I contents in the herbal samples were deduced by interpolation of the concentration (Ct - Ci) in the AA-I calibration curve established by regressing the relative concentration in AA-I derivative calibration samples against their nominal concentrations. Herein, we stacked the calibration sets of AL-I and AA-I together with five herbal medicine samples, constructing a three-way data array with the size of 76 × 101 × 17. And in line with CORCONDIA, the number of components was chose N = 4 when APTLD and SWATLD algorithms were applied to decompose the three-way data array.Figure 5 shows the resolved excitation-emission spectral profiles and the relative concentration profiles using APTLD and SWATLD algorithms, respectively. In comparison with Figure 4, it seems that the excitation-emission spectra of uncalibrated interferences overlap with that of AL-I more heavily and the intensity of uncalibrated interferences become more intense. It is worth noting that the uncalibrated interferences are mainly ALs and other unknown interferents in the complex herbal medicine samples. As we mentioned, AAs can be converted to ALs by derivatization reaction. With the exception of AA-I, there are many other AAs would be reduced to corresponding ALs in the herbal medicine samples. Therefore, the concentration of other ALs would increase, which manifest as heavily overlap with fluorescence spectra of AL-I and higher intensity of uncalibrated interferences (Illustration in Figure 5). Despite of the relatively complicated matrices, the proposed strategy achieved satisfactory decomposition results based on APTLD and SWATLD algorithms. As depicted in Figure 5, the resolved spectral profiles of AL-I using APTLD and SWATLD algorithms, are good in accordance with that of actual AL-I, respectively. The sample modes (C1 and C2) were applied for calibration through a linear regression. Herein, we used the AL-I as calibration standard, for better linearity and repeatability. The total concentration of AL-I (Ct) was regressed, then the concentration (Ct - Ci) that subtracted the initial concentration of AL-I (Ci) from the total concentration (Ct), was calculated back to AA-I through the AA-I calibration curve. Table 4 lists the prediction results of AA-I using APTLD and SWATLD algorithms in five herbal medicines. From the table, Aristolochia mollissima Hance (XGF) contains the highest AA-I content among the test samples, with the concentrations being 736.6 and 726.1 µg g-1 for APTLD and SWATLD, respectively. 4.7. Quantification of AL-I and AA-I by LC-MS/MS for method comparison The results from the proposed method were compared with that obtained from LC-MS/MS method. Five herbal medicine samples were also analyzed by LC-MS/MS to determine the differences between the two methods. It shows in Figure S3 a typical extracted MRM chromatogram (Figure S3) and the LC–MS/MS retention time of AA-I (m/z 359.0 → 298.0, 3.4 min) and AL-I (m/z 294.0 → 279.2, 7.1 min) in a calibration sample. Table 4 summarizes the results from LC-MS/MS analysis of AA-I and AL-I in five herbal medicines. The comparison results demonstrated that the two methods offer similar results. Both the proposed method and the LC-MS/MS method didn't detect AA-I and AL-I in the Syephania tetrandra S. Moore (FJ). In fact, S. tetrandra (Pinyin, Han Fang Ji) and A. fangchi (Pinyin, Guang Fang Ji) belong to the same therapeutic „Fang Ji‟ family in traditional Chinese medicine, and the herbal ingredients are generally traded using their common Pin Yin name. Syephania tetrandra (Pinyin, Han Fang Ji) do not contain AA-I and AL-I, whereas A. fangchi (Pinyin, Guang Fang Ji) contain abundant AA-I and AL-I [10]. Negative results were observed for AA-I and AL-I in Syephania tetrandra S. Moore (FJ), which demonstrate that the proposed method is capable of anti false-positive result that often happen in conventional fluorescence spectrometry. The t-test was used to detect the difference between the contents obtained from the proposed method based on APTLD and SWATLD algorithms and the LC-MS/MS for both analytes in five herbal medicines (null hypothesis, p-value = 0.05). The t-test also showed that the null hypothesis could not be rejected with tcalculated < t5 (2.57). Therefore, there are no significant difference between the proposed method and the LC-MS/MS method. With the high correlation between results from the two methods, we certainly believe that the developed method based on the APTLD and SWATLD algorithms can accurately quantify AA-I and AL-I contents in complicated herbal medicine matrices. Insert Table 4 here 4.8.Figures of merit The figures of merit, such as SEN, SEL, LOD and LOQ for determination of AA-I and AL-I in the MDL samples, were investigated for further assessing performance characteristic of the proposed method based on APTLD and SWATLD algorithms. The figures of merit of AA-I and AL-I were summarized in Table 5. It is obvious that the figures of merit for AA-I and AL-I obtained by APTLD and SWATLD algorithms show minor difference, which demonstrates both of the proposed method based on APTLD and SWATLD algorithms can yield satisfactory predictive capacity for the determination of AA-I and AL-I in herbal medicines. In addition, the LODs and LOQs of AA-I and AL-I for the LC-MS/MS method were also calculated for comparison. As is shown in Table 5, the LOD of AA-I obtained from the LC-MS/MS method is slightly higher than that of the proposed method based on APTLD and SWATLD algorithms. However, the LOD of AL-I from the LC-MS/MS method is apparently higher than that of the proposed method. The main reason for which is that the relatively lower ionization efficiency of AL-I results in low MRM response. It is clearly shown in Figure S3, the chromatographic peak area of AL-I was smaller than that of AA-I with almost one order at the same concentration level.In order to further insight into the performance of the proposed method based on APTLD and SWATLD algorithms for the analytes, an elliptical joint confidence region (EJCR) test[59] was performed with estimated intercept and slope that obtained from spiked recovery assay. Figure 6 shows the results of EJCRs for both AA-I and AL-I from APTLD and SWATLD algorithms. It shows that the ideal point (0, 1), labeled with a star (*), lies in all EJCRs, indicating the reference values and the results from the proposed method do not show significant difference at the level of 95% confidence. What is more, the elliptic size corresponding to the SWATLD algorithm (solid line) is smaller than that of APTLD algorithm (dash line) for AA-I, while the exact opposite for AL-I. The elliptic size denotes precision of an analytical method, and smaller size corresponds to higher precision. All these results demonstrate the proposed method based on APTLD and SWATLD algorithms is able to provide accurate prediction, although there is somewhat distinction in the performance of predictive precision between the two algorithms. 5.Conclusions The combination of excitation-emission matrix fluorescence with second-order calibration methods based on APTLD and SWATLD algorithms allowed successful determination of AA-I and AL-I in five Chinese herbal medicines. The conversion of non-fluorescent AA-I to high-fluorescent AL-I by the derivatization process provided high sensitivity of AA-I for the fluorescence detection. By virtue of the "second-order advantage", the proposed method based on APTLD and SWATLD algorithms has accurately acquired the qualitative and quantitative information of the analytes in five Chinese herbal medicines regardless of the complex matrices. The results obtained from the developed method were compared with the validated LC-MS/MS method, which showed no significant difference. The satisfactory spiked recovery, SEN, SEL, LOD, LOQ and RMSEP for AA-I and AL-I further validated the reliability and reproducibility of the proposed method. What is more, the proposed analytical methods based on APTLD and SWATLD algorithms requires no prior Aristolochic acid A separation and clear-up process, which has advantages of being simple, fast, green and high-sensitivity.