Dynamic predictive models of five alkaloids in Coptis during the process of stir-frying with wine using near-infrared spectroscopy

ABSTRACT

The goal of this research was to develop dynamic predictive models for five alkaloids in Coptis during the process of stir-frying with wine by near-infrared spectroscopy (NIR). Six batches of samples were collected and then processed on the basis of orthogonal design. With high-performance liquid chromatography (HPLC) analysis as reference, calibration models were generated by a partial least squares (PLS) regression. The root mean square errors of prediction (RMSEP) for the PLS models of jatrorrhizine, epiberberine, coptisine, palmatine, and berberine were 0.0030, 0.0387, 0.0206, 0.0095, and 0.0516, and the correlation coefficients (R) were 99.43%, 98.41%, 99.29%, 99.23%, and 99.87%, respectively. Accordingly, these results demonstrated that the models could be efficiently used to develop robust methods of online analysis and quality control for Coptis.

KEYWORDS:

• Alkaloids
• Coptis
• Stir-frying with Wine
• Near-infrared Spectroscopy (NIR)
• Partial Least Squares (PLS)

Introduction

Currently, traditional Chinese medicines (TCMs) still play an important role in the treatment and prevention of human diseases. The safety, effectiveness, and quality of processed TCMs have attracted worldwide attention.^[1] As a type of TCM, Coptis has many medicinal values and is widely used in clinical care.^[2] Coptis is derived from the dried roots of Coptis chinensis Franch., Coptis deltoidea C. Y. Cheng et Hsiao and Coptis teeta Wall., which are called “Wei Lian”, “Ya Lian”, and “Yun Lian” in Chinese, respectively.^[3] Modern pharmacological studies have shown that Coptis has multiple pharmacological effects such as dissipating heat and drying dampness, easing disturbed states of mind, and purging fire to eliminate toxins.^[4-6] Generally, Coptis needs to be processed before use because Coptis is strongly bitter and cold, which can be alleviated in processed Coptis. At present, there are several methods to process Coptis. One of the most popular is stir-frying with wine, which means soaking the Coptis in rice wine for several minutes and then stir-frying it to dryness.^[2,3]

Along with the process of stir-frying with wine, the main chemical constituents of Coptis are in constant flux. As has been reported, Coptis is rich in alkaloids, including jatrorrhizine, epiberberine, coptisine, palmatine, and berberine, which are considered the major active constituents and quality control targets for Coptis.^[7,8] To date, several conventional methods, for example, high-performance liquid chromatography (HPLC), have been reported for the qualitative and quantitative analysis of extracts and their active components in Coptis.^[9,10] However, in these methods, Coptis must be removed from the process of stir-frying with wine, and the samples require tedious pretreatment and time-consuming analysis.^[11] In addition, Coptis in different processed states has different contents of the five alkaloids. Therefore, it is necessary to determine a method that can rapidly online monitor the variation of the contents during the stir-frying process.

Near-infrared spectroscopy (NIR) coupled with chemometrics is a quick, accurate, and non-destructive analytical tool, which can be considered as a replacement for traditional chemical analysis. As it possesses advantages of fast acquisition, no sample preparation, noninvasiveness, nondestructiveness, and simultaneous determination, NIR spectroscopy with fiberoptic probes has been widely used in qualitative discrimination, quantitative determination, identifying counterfeits and forecasting the stability of TCMs, among other applications.^[12–19] This is the first study using NIR to simultaneously analyse these five alkaloids in Coptis during the process of stir-frying with wine.

The purpose of this study was to evaluate the possibilities of rapid dynamic quantitative analysis of jatrorrhizine, epiberberine, coptisine, palmatine, and berberine in Coptis during the process of stir-frying with wine by NIR. Firstly, 155 samples of raw Coptis were processed with rice wine as per the results of an orthogonal design. Secondly, the processed Coptis samples were detected by NIR. Then, calibration models generated by partial least squares (PLS) of the five alkaloids were established based on the NIR spectra and the reference data collected from HPLC. Finally, the validation set was used to validate the predictive ability of the optimized PLS models. This study was performed as a laboratory simulation intended for subsequent use in rapid online analysis and quality control to monitor the industrial process of stir-frying Coptis with wine.

Materials and methods

Materials and reagents

Six batches of Coptis were obtained from different provinces of China. Jatrorrhizine hydrochloride (batch no.W1900010), epiberberine (batch no.W0400010), coptisine hydrochloride (batch no.W2160010), palmatine (batch no.W0410010), and berberine hydrochloride standards (batch no.R3550010) were obtained from the ANPEL Scientific Instrument Company (Shanghai, PR China). Rice wine was purchased from the Kuaijishan Shaoxing Wine Company (Zhejiang, PR China). HPLC-grade acetonitrile was obtained from the Tianjin Kermel Chemical Reagent Company (Tianjin, PR China). Water was purified by an ultrapure water system. All the other reagents were of analytical grade.

Sample preparation

The 25 different Coptis processing methods were generated by the orthogonal design method and are shown in Table 1. Coptis slices weighing 80 g were mixed with the appropriate quantity of rice wine and stir-fried to homogeneity. After the rice wine was absorbed by Coptis for different periods, the slices were placed into a stir-fry vessel and heated at the designed temperature controlled by an infrared thermometer (GM320, BENETECH, PR China) for 6 min. Next, the processed Coptis were removed, cooled, and ground by a grinder. Finally, the small fragments were screened. In this study, six batches were used, and 155 samples were collected.

Table 1. Orthogonal design of processed Coptis.

Factors	A Contents of rice wine (%)	B Sealed moistening time (h)	C Stir-fry temperature (°C)
levels
1	0.0	0	indoor temperature
2	10.0	1	130
3	12.5	2	160
4	15.0	4	190
5	17.5	8	220

Spectra acquisition

The NIR spectra of processed Coptis samples, shown in Fig. 1, were recorded by a Quas-IR 3000 spectrometer (Vspec, USA) equipped with a PbS detector, sample cup and rotary tables. The spectra were obtained at a resolution of 8 cm⁻¹ over a wavelength range of 12000–4000 cm⁻¹ with 32 scans per spectrum, and air absorbance was recorded as the reference standard. To avoid error from the external environment, all samples were equilibrated to indoor temperature (25°C) prior to NIR analysis. In addition, the humidity was kept at the ambient level in the laboratory.

Figure 1. Near‑infrared spectra of Coptis samples.

Determination of reference values

Using the Chinese Pharmacopoeia (volume I, 2015 edition), the reference values were analysed immediately after the NIR measurements.³ An 1100 HPLC system (Agilent Technologies Inc., USA) containing an ultraviolet–visible detector was used for the quantitative determination of the alkaloids at a wavelength of 345 nm. A Kromasil (Akzo Nobel Inc., Sweden) C₁₈ column (250 mm × 4.6 mm, 5 μm) was employed at 35°C. The elution system was composed of acetonitrile–water (50:50) containing 50 mmol/L of monopotassium phosphate and 25 mmol/L of sodium dodecyl sulfate (SDS) at a flow rate of 1.0 mL/min. The sample aliquots (10 μL) were injected into the HPLC for content analysis.

All 155 samples were analysed by the above-mentioned HPLC method. Representative HPLC chromatograms of the standard solution (five alkaloids) and the processed Coptis sample solution are shown in Fig. 2. Baseline separation for all five alkaloids could be achieved successfully. The chromatographic peaks were identified by comparing their retention times against known standard peaks.

Figure 2. HPLC results of standard substance (a) and processed Coptis sample (b) Peak No.: 1. Jateorrhizine; 2. Epiberberine; 3. Coptisine; 4. Palmatine; and 5. Berberine.

Data processing

All the computations, including data acquisition, spectral preprocessing, variable selection, chemometric regression models construction and validation were performed using OPUS spectral acquisition and processing software (Operant LLC Licensed to MTG, USA).

Results and discussion

Quantitative calibration models

Division of calibration and validation set: Among the 155 samples, 1 out of 5 was selected randomly for validation analysis, and thus, 31 samples were selected into the validation set to validate the PLS models, and the remaining 124 samples were used as the calibration set to establish the calibration models.^[20]

Optimizing the calibration models using the PLS method

Spectral pretreatment methods: To gather full and effective information from the spectral data and eliminate noise, baseline shift, and matrix background interference, the spectra must be preprocessed.¹⁵ In this study, different optimization methods were implemented by the OPUS software including Constant Offset Elimination(COE), Straight Line Subtraction (SLS), Vector Normalization (VN), Min Max Normalization (MMN), Multiplicative Scatter Correction (MSC), First Derivative, Second Derivative, First Derivative + SLS, First Derivative + VN and First Derivative + MSC.

The PLS methods must be computed for each number of components, and the optimum factors for the calibration set were selected based on the correlation coefficients of the calibration set (R_cal), the root mean square errors of cross-validation (RMSECV), and the residual predictive deviations (RPD). The models with the best prediction ability are usually selected on the basis of the root mean square error (RMSEP) and the correlation coefficients of the validation set (R_val).^[21,22] Calibration models with the highest R and RPD as well as the lowest RMSECV are considered optimal. OPUS could determine the values of R, RPD, RMSECV and RMSEP, and then, the best spectral pretreatment methods were chosen according to these values. The results obtained with different preprocessing methods are shown in Table 2.

Table 2. Results obtained with different spectral pretreatment methods.

	Pretreatment method	Rcal (%)	RMSECV	RPD
	MSC	97.98	0.0038	6.84
	VN	99.05	0.0024	10.43
	SLS	99.23	0.0024	11.40
	COE	99.09	0.0029	10.50
	MMN	98.60	0.0030	8.68
Jatrorrhizine	First derivative	99.06	0.0030	10.35
	Second derivative	98.20	0.0039	7.48
	First derivative + VN	98.65	0.0029	9.05
	First derivative + SLS	98.85	0.0033	9.32
	First derivative + MSC	99.11	0.0023	10.60
	MSC	96.67	0.0326	6.04
	VN	93.45	0.0412	6.20
	SLS	78.92	0.0729	3.74
	COE	97.32	0.0300	6.20
	MMN	98.15	0.0259	7.43
Epiberberine	First derivative	90.11	0.0581	3.22
	Second derivative	74.55	0.0919	2.00
	First derivative + VN	70.98	0.0979	1.89
	First derivative + SLS	86.11	0.0688	2.69
	First derivative + MSC	69.38	0.1090	1.81
	MSC	98.95	0.0119	9.78
	VN	98.75	0.0148	8.93
	SLS	99.31	0.0136	12.00
	COE	99.45	0.0119	13.50
	MMN	99.05	0.0111	10.54
Coptisine	First derivative	99.37	0.0130	12.60
	Second derivative	95.25	0.0245	4.65
	First derivative + VN	98.62	0.0141	8.57
	First derivative + SLS	99.15	0.0106	10.85
	First derivative + MSC	98.98	0.0111	9.95
Palmatine	MSC	97.39	0.0100	6.23
	VN	96.70	0.0112	5.52
	SLS	98.56	0.0080	8.46
	COE	98.89	0.0068	9.53
	MMN	97.45	0.0099	6.43
	First derivative	98.63	0.0074	8.53
	Second derivative	86.48	0.0216	2.72
	First derivative + VN	96.90	0.0105	5.68
	First derivative + SLS	98.59	0.0076	8.43
	First derivative + MSC	96.50	0.0112	5.38
Berberine	MSC	98.40	0.0941	7.91
	VN	98.00	0.1060	7.07
	SLS	99.42	0.0523	13.10
	COE	99.18	0.0723	11.10
	MMN	98.77	0.0806	9.02
	First derivative	99.11	0.0743	10.60
	Second derivative	98.59	0.0743	8.44
	First derivative + VN	98.94	0.0663	9.73
	First derivative + SLS	99.43	0.0487	13.20
	First derivative + MSC	98.64	0.0768	8.59

Spectral region and factors: In the complex system, it was difficult to use a classical univariate parameter for quantitative analysis. Therefore, the spectral region and PLS factors were chosen as two other standard parameters to optimize the calibration models.

One or several wavelengths related to alkaloids cannot be found in the NIR spectra because of the interferences of other components, so a spectral region rich in chemical information is usually used to establish the calibration models. The PLS method of OPUS is a powerful multivariate calibration method to recommend the appropriate spectral region. Table 3 shows the appropriate spectral regions for the NIR models.

Table 3. Parameters of different spectral regions in PLS models.

	Spectral reign (cm^–^Citation1)	Rcal (%)	RMSECV	RPD
Jatrorrhizine	12003.5-5448.2	99.24	0.0025	11.60
	12003.5-5448.2 4600.6-4249.3	99.22	0.0024	11.40
	7502.2-6097.2 5452-4249.3	98.81	0.0030	9.16
	12003.5-7498.4 5452-4249.3	98.17	0.0037	7.38
	6101-5448.2	83.57	0.0111	2.47
Epiberberine	12003.5-4596.8	97.84	0.0293	6.81
	12003.5-6097.2 5452-4596.3	97.44	0.0319	6.25
	12003.5-6097.2 5452-4596.8	97.04	0.0343	5.82
	12003.5-6097.2 4600.6-4249.3	91.44	0.0583	3.42
	7502.2-4596.8	88.89	0.0665	3.00
Coptisine	9052.3-6097.2 5452-4249.3	99.47	0.0115	13.80
	7502.2-6097.2 5452-4596.8	99.46	0.0116	13.60
	12003.5-4596.8	99.45	0.0118	13.50
	12003.5-6097.2 5452-4249.3	99.42	0.0121	13.10
	7502.2-6097.2 5452-4249.3	99.37	0.0126	12.60
Palmatine	12003.5-7498.4 5452-4596.8	98.82	0.0073	9.22
	12003.5-7498.4 5452-4249.3	98.63	0.0078	8.56
	12003.5-7498.4 6101-4249.3	98.54	0.0081	8.28
	12003.5-6097.2 4600.6-4249.3	98.54	0.0081	8.26
	12003.5-6097.2	97.76	0.0100	6.68
Berberine	7502.2-4249.3	99.43	0.0487	13.20
	12003.5-6097.2 5452-4596.8	99.42	0.0477	13.10
	12003.5-7498.4 5452-4596.8	99.11	0.0605	12.86
	6101-4596.8	98.22	0.0829	12.53
	5452-4596.8	96.68	0.1170	10.33

In the PLS method, the number of PLS factors (F) would directly affect the predictive ability of the calibration models. Not enough information would be obtained from the spectrum when F is too small, and the actual information would be misrepresented or ‘‘over fitted’’ when F is too large. OPUS is successful in recommending the ideal F according to the criteria of the best calibration. Figure 3 shows the relation between the RMSECV values (y-axis) and F (x-axis). When F>7, the RMSECV value of jatrorrhizine became gradual and tended to be stable, so the ideal F for jatrorrhizine PLS model was 7; the ideal F for epiberberine, coptisine, palmatine and berberine were 12, 9, 10, 7, respectively. Apparently, the given ideal F in Fig. 3a and 3d are clearly “over fitted”, so the smaller 7 of jatrorrhizine and 10 of palmatine were chosen.

With PLS regression, the calibration models were finally optimized individually by spectral pretreatment methods, spectral range, and F.

Figure 3. Parameters of the number of model factors by RMSECV: (a) Jatrorrhizine,(b) Epiberberine, (c) Coptisine,(d) Palmatine, (e) Berberine.

Results and verification of the quantitative models

The validation set is used to validate the predictive ability of the optimized PLS models. Following the procedures mentioned above, the calibration models of the five alkaloids were established in this study, and the above optimized models were used to predict the contents of the 31 samples in the validation set. The results in Table 4 show that the established models gave satisfactory fitting results and predictive ability and could be applied to monitor the contents of the five alkaloids in Coptis during stir-frying with wine. Figure 4 shows the correlation between the NIR predicted values (y-axis) and the reference values (x-axis) obtained by HPLC. The RMSEP and R_val of the validation set suggested that the predicted values from NIR are very close to the values determined by HPLC. However, deviations still existed in a few samples because the pretreatment process of Coptis was quite complex and Coptis was not completely uniform during HPLC determination and NIR spectrum collection. The T and F tests for the predicted result indicated that the accuracy and precision were satisfactory with a significance level of 0.05.

Table 4. Parameters of optimal models and validation results.

	Jatrorrhizine	Epiberberine	Coptisine	Palmatine	Berberine
Spectral pretreatment method	SLS	MMN	COE	COE	First Derivative + SLS
Factor	7	12	9	10	7
Spectrum region (cm-1)	12003.5-5448.2	12003.5-4596.8	9052.3-6097.2 5452-4249.3	12003.5-7498.4 5452-4596.8	7502.2-4249.3
Rcal (%)	99.23	98.15	99.45	98.89	99.43
RMSECV	0.0024	0.0259	0.0119	0.0068	0.0487
RPD	11.40	7.43	13.50	9.53	13.20
RMSEP	0.0030	0.0387	0.0206	0.0095	0.0516
Rval (%)	99.43	98.41	99.29	99.23	99.87

Figure 4. Relations of values measured by HPLC and the five alkaloids models: (a) Jatrorrhizine, (b) Epiberberine, (c) Coptisine, (d) Palmatine, (e) Berberine.

Discussion of the process of stir-frying Coptis with wine

Based on the results calculated by the orthogonal design assistant SPSS 11.5, as shown in Table 5, we found that the sequences of every factor and level in affecting the processed results were A₃>A₄>A₅>A₂>A₁; B₃>B₁>B₄>B₂>B₅; and C₃>C₄>C₂>C₅>C₁. The results of variance analysis of each factor indicated that the stir-fried temperature was the most important factor, the rice wine content was the second most important factor and the sealed moistening time was the least important factor. Therefore, the optimized method for processing was C₃A₃B₃. In other words, the optimized method was to add 15% (v/v) rice wine to 80 g of Coptis, seal for 2 h to moisten, and stir-fry at 160°C for 6 min, which was in accordance with the Chinese Pharmacopoeia.

Table 5. Statistical table of single factors created by SPSS. Dependent Variable: Content of total alkaloids.

				95% Confidence Interval
		Mean	Std. Error	Lower Bound	Upper Bound
	1	23.132	0.295	22.549	23.714
	2	23.414	0.295	22.832	23.997
A	3	24.478	0.295	22.895	25.060
	4	23.653	0.295	23.071	24.236
	5	23.475	0.295	22.892	24.057
	1	23.486	0.295	22.903	24.068
	2	23.246	0.295	22.663	23.828
B	3	23.811	0.295	23.229	24.394
	4	23.454	0.295	22.872	24.037
	5	23.155	0.295	22.572	23.737
	1	22.153	0.295	20.571	23.736
	2	23.299	0.295	22.717	23.882
C	3	25.234	0.295	23.652	25.816
	4	23.524	0.295	22.941	24.106
	5	22.942	0.295	22.359	23.524

The results also showed that different processing extents led to different contents of alkaloids in Coptis. Comparing the alkaloids contents of raw and processed Coptis, it can be observed that the contents of palmatine and berberine increased notably after processing, while the contents of epiberberine and coptisine decreased and the content of jatrorrhizine had no clear change, which indicated that the processing method does affect the chemical components in Coptis, but the mechanism requires further study.

In addition, as shown in Fig. 5, there were two new chromatographic peaks of unknown components generated during the process of stir-frying Coptis with wine. Further study is required to determine whether these new components are transformed from epiberberine and coptisine. It also requires further study whether these new components have any relations to the traditional saying “processed Coptis can ease the cold property of raw Coptis” or have any other new pharmacological effects.

Figure 5. HPLC results of raw (a) and processed Coptis (b).

Conclusion

In this study, six batches of samples, which were processed according to the results of an orthogonal design setting different contents of rice wine, sealed moistening time, and stir-fry temperature, were surveyed by NIR spectroscopy, and models were built by means of PLS. Their main parameters were also discussed in detail. The root mean square errors of prediction (RMSEP) for the PLS models of jatrorrhizine, epiberberine, coptisine, palmatine, and berberine were 0.0030, 0.0387, 0.0206, 0.0095, and 0.0516 and their correlation coefficients (R) were 99.43%, 98.41%, 99.29%, 99.23%, and 99.87%, respectively. The overall results showed that NIR spectroscopy combined with PLS provided reliable, stable and efficient analysis for jatrorrhizine, epiberberine, coptisine, palmatine, and berberine in Coptis during the process of stir-frying with wine and showed great promise for use in industrial manufacturing processes for rapid online analysis and quality control.

Declaration of interest

The authors declare that there is no conflict of interest.

Acknowledgements

The authors thank Sichuan Vspec Technologies Company, who provided the instruments and analysis instruction.

Funding

The authors graciously acknowledge the financial support provided by the National Sciences Foundation of PR China (No.81373969) and the University Key Research Projects of Henan Province (No. 17A360026).

Additional information

Funding

The authors graciously acknowledge the financial support provided by the National Sciences Foundation of PR China (No.81373969) and the University Key Research Projects of Henan Province (No. 17A360026).