NIR assignment of isopsoralen by 2 D-COS technology and model application in Yunkang Oral Liquid

Near infrared (NIR) assignment of Isopsoralen was performed using deuterated chloroform solvent and two-dimensional correlation spectroscopy (2D-COS) technology. Yunkang Oral Liquid was applied to study Isopsoralen, the characteristic bands by spectral assignment as well as the bands by interval partial least squares (iPLS) and synergy interval partial least squares (siPLS) were used to establish partial least squares (PLS) model. The coe±cient of determination in calibration (Rcal) were 0.9987, 0.9970 and 0.9982. The coe±cient of determination in cross validation (RvalÞ were 0.9985, 0.9921 and 0.9982. The coe±cient of determination in prediction (RpreÞ were 0.9987, 0.9955 and 0.9988. The root mean square error of calibration (RMSEC) were 0.27, 0.40 and 0.31 ppm. The root mean square error of cross validation (RMSECV) were 0.30, 0.67 and 0.32 ppm. The root mean square error of prediction (RMSEP) were 0.23, 0.43 and 0.22 ppm. The residual predictive deviation (RPD) were 31.00, 16.58 and 32.41. It turned out that the characteristic bands by spectral assignment had the same results with the chemometrics methods in PLS model. It provided guidance for NIR spectral assignment of chemical compositions in Chinese Materia Medica (CMM).


Introduction
Critical quality attribute (CQA) assessment has been highly regarded by International Conference on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH) and Food and Drug Administration (FDA) in pharmaceutical manufacture process.Near infrared (NIR) is an excellent and rapid technique for evaluating the CQA because of its fast, nondestructive and real-time advantages.The NIR wavelength is from 780 to 2500 nm.In this area, the main vibrations of functional groups including a hydrogen atom are observed, such as -CH, -NH, -OH and -SH bonds. 1 For the past few years, NIR has become a powerful analytical technique in quantitative and qualitative analysis. 2It provided useful functions across a range of application areas, such as pharmacy, chemical industry, food and petroleum ¯eld. 3,4owever, NIR has overlapping and broad absorption bands due to the overtones and combination tones.It is di±cult to identify the characteristic peaks in NIR wavelength.There are few researches on NIR spectral assignment.Meanwhile, there is also no report on the NIR assignment of Chinese Materia Medica (CMM).Spectral assignment is signi¯cant with regard to accurate quantitative and qualitative analysis in various ¯elds.Thus, NIR spectral assignment is very important in its application.The relationship between spectra and molecular structure enables NIR spectral assignment to achieve accurate quantitative and qualitative analysis.
Two-dimensional correlation spectroscopy (2D-COS) is a powerful and versatile technique to assign the NIR spectra.It was ¯rstly proposed in the ¯eld of nuclear magnetic resonance.Then generalized 2D-COS technology expanded a variety of applications, which was proposed by Noda et al. 5,6 It mainly focused on the external disturbances about spectral variation.It could be applied to ¯nd the characteristic bands in complicated spectra.The external disturbances included temperature, pressure and concentration. 7,8iu illuminated the NIR band assignments of methanol and ethanol by generalized 2D-COS. 9enkawa et al. demonstrated the IR/NIR absorption bands of water and liquid oleic acid by 2D-COS technology, in which the temperature was used as external disturbance. 10According to the structure and spectra, 2D-COS technology was applied to assign the variation of band position shift by Shinzawa et al.It revealed a phenomenon that the cellulose crystalline phase disorganized in accordance with the grinding. 11Other researches have also made signi¯cant trials on band assignments of NIR using 2D-COS technology in the references cited. 12ith a history of 2000 to 3000 years, CMM has formed a unique and complex system to diagnose and cure illness.NIR is a rapid technique for CQA assessment in CMM to guarantee the safety, e®ectiveness, stability and uniformity.However, when NIR technology has been used in CMM, overlapping absorption, wide peaks and low absorption would appear to limit the application.
To obtain a robust model, variable selection methods were chosen to ¯lter variables, such as interval partial least squares (iPLS), moving window partial least squares (mwPLS), backward interval partial least squares (biPLS) and synergy interval partial least squares (siPLS).Wu reported that iPLS was regarded as the variable selection method for chlorogenic acid in the extraction process of Lonicera japonica.The result demonstrated that NIR allowed the detection of a minor analyte (50 ppm) in CMM. 13 Wang et al. illuminated that the best spectral band was selected to detect alcohol concentration using siPLS analysis.It showed that the model had good predictive ability and reliability. 14Jiang et al. demonstrated that siPLS was used to predict process variables in solid-state fermentation of wheat straw.The results showed that optimal model was achieved by siPLS with selected spectral subintervals. 15Chen et al. reported that NIR could determine total polyphenols content in green tealeaves.It showed that siPLS and iPLS were used to establish NIR model, and the siPLS method achieved good predictive ability. 16owever, when variable selection methods of Chemometrics were used to establish NIR model, the same substance often has di®erent bands due to di®erent methods. 17,18In this paper, the expectation about NIR research focused on model interpretation using NIR spectral assignment coupled with the relationship between spectra and molecular structure.Yunkang Oral Liquid is a Chinese herbal preparation, including Codonopsis pilosula Nannf., Poria cocos Wolf., Psoralea corylifolia L., etc. 19 Yunkang Oral Liquid has the obvious therapeutic e®ects on enlivening the spleen, stabilizing the kidney, nourishing blood and calming abortion.It is used for treating the threatened abortion and habitual abortion. 20n this study, 2D-COS technology was introduced and concentration was selected as the external disturbance.NIR spectral assignment of Isopsoralen was performed with deuterated chloroform and the quantitative NIR model about CQA (Isopsoralen) of Yunkang Oral Liquid as the example.The characteristic bands from chemometrics and spectral assignment from 2D-NIR were used to establish quantitative model in Yunkang Oral Liquid.In chemometrics application, siPLS and iPLS methods were used to select the variable of Isopsoralen in Yunkang Oral Liquid.Finally, model interpretation in Yunkang Oral Liquid was demonstrated using NIR spectral assignment coupled with the relationship between spectra and molecular structure.

Materials
Yunkang Oral Liquid was purchased from the market, and deposited in the key laboratory of Traditional Chinese Medicine-information Engineering of State Administration of Traditional Chinese Medicine (No. 140528).Isopsoralen reference standard (lot number: 110738-20131) was supplied by the Chinese Food and Drug Inspection Institute (Beijing, China).Deuterated chloroform was purchased from Cambridge Isotope Laboratories, Inc. (Massachusetts, USA).HPLC grade methanol was purchased from Thermo Fisher Sci-enti¯c Inc. (Massachusetts, USA).Deionized water was purchased from Hangzhou Wahaha Co., Ltd (Zhejiang, China).

Preparation of samples
For NIR assignment of Isopsoralen, the 50.08 mg of Isopsoralen was weighed accurately into a 10 mL volumetric °ask, and diluted with deuterated chloroform.The solution was obtained including 5.008 mg of Isopsoralen in each of milliliter.Series of percentages about the solution were transferred to 2 mL volumetric °ask separately, and further diluted with deuterated chloroform for 2D-COS analysis, which the concentration level ranged from 1.00 to 5.00 mg Á mL À1 . 21ile applying the NIR assignment of Isopsoralen, 50 samples of Yunkang Oral Liquid diluted with water were obtained containing series of Isopsoralen concentration levels, ranging from 2% to 100% (v/v).In order to obtain a robust NIR model, 50 samples were selected, 33 as calibration sets (validation sets, leave-one cross validation) and 17 as prediction sets according to Kennard-Stone (KS) algorithm.

NIR equipment and measurement
The NIR spectra were collected by XDS analyzer and VISION software (Metrohm NIR System, Florida, USA).The wavelength range of spectra was from 780 to 2500 nm.Each spectrum was an average of 32 scans and 3 times with a wavelength increment of 0.5 nm.Data measurement was performed by VISION software (Vision 3.5, Metrohm NIR System, Florida, USA).

HPLC methods
Chromatographic analysis was carried out on Agilent HPLC apparatus, which was comprised of Agilent 1100 system, an auto-sampler, a column temperature controller and a diode-array detector (DAD) (Agilent Technologies, California, USA).Samples were primarily separated on an Agilent ZORBAX SB C18 column (5 m, 4:6 Â 250 mm, Agilent Columns, California, USA) at 25 C using the mobile phase of methanol and water (50:50, v/v).The detection wavelength was set at 264 nm. 225.The method of 2D-COS 2D-COS is an excellent method for NIR spectral assignment.It is performed by synchronous spectrum and asynchronous spectrum.The overlapping bands could be distinguished by the synchronous spectrum.The variable order of various bands could be elucidated by the asynchronous spectrum.The auto-peaks in synchronous spectrum represent the characteristic bands of pharmaceutical ingredients.In addition, the enhancement and debase of crosspeaks re°ect functional groups.23,24 In our previous work, the number of samples and the intervals of concentration range were illuminated in 2D-COS analysis of NIR assignment.21 In this study, Isopsoralen was chosen to obtain the NIR assignment according to the positions of the auto-peaks by 2D-COS technology.

Software and data analysis
The Unscrambler 9.6 software package (CAMO Software AS, Oslo, Norway) was applied to analyze the data.2D-COS and home-made routines program in MATLAB code (MATLAB v7.0,The MathWorks, MA) was also used.The toolbox of iPLS and siPLS used to select the informative variables were downloaded from http://www.models.kvl.dk/.In addition, the coe±cient of determination in calibration (R 2 cal ), the coe±cient of determination in cross validation (R 2 val ), the coe±cient of determination in prediction (R 2 pre Þ, the root mean square error of calibration (RMSEC), the root mean square error of cross validation (RMSECV), the root mean square error of prediction (RMSEP) and the residual predictive deviation (RPD) were applied to evaluate the PLS model.

Quantitative analysis of Isopsoralen by HPLC method
High performance liquid chromatography (HPLC) method was applied to determine Isopsoralen in Yunkang Oral Liquid.The HPLC chromatograms of reference and sample are shown in Fig. 1.The retention time of Isopsoralen in sample solution was the same as the reference solution.The calibration curve of the HPLC method was drawn before sample analysis.The calibration curve exhibited good linearity (Y ¼ 7 Â 10 6 X-6.9995, R 2 ¼ 1:0000) within the content ranging from 0.054 to 0.189 g.According to the calibration curve of Isopsoralen, each accurate concentration of Yunkang Oral Liquid was determined, in which the content of Isopsoralen ranged from 0.488 to 24.4 g.The mean content was 11.97 g.

Selection of deuterated chloroform and comparison of di®erent spectral pretreatment methods
The raw spectra of Yunkang Oral Liquid are shown in Fig. 2(a).As we can see, large °uctuations appeared in the region of combinations and ¯rst overtones, which are caused by the noise and the strong absorption of water.Meanwhile, spectra of Isopsoralen were obtained using deuterated chloroform solvent as shown in Fig. 2 was one of the simplest systems, because NIR absorption was mainly from hydrogen groups.Our results proposed that deuterated chloroform was a good solvent for NIR spectral assignment of Isopsoralen.
In addition, spectral pretreatment is very important before establishing a robust model.Di®erent pretreatment methods were used to eliminate the e®ects of noise and baseline drift and obtain an accurate model.The model performance was excellent according to the following strategies: the coef-¯cient of determination R 2 was approximately equal to 1; RMSEC, RMSECV and RMSEP values were one of the least and the RPD value was more than 3.
Baseline was mainly used to reduce the e®ects of background and drift on the spectral signal.S-G was applied to decrease noise, especially in combination and overtones of NIR.The derivation can improve the sensitivity and resolution.SNV was applied to reduce the in°uence of multiplicative interference of scatter by standardizing the absorbance at each wavelength in the spectrum. 25eanwhile, the numbers of latent factors were investigated coupled with the selection of appropriate pretreatment method.Figure 3 and Table 1 showed the relationship with latent factors and PRESS in di®erent pretreatment methods using leave-one cross validation.The results showed that baseline was the best pretreatment method, in which R 2 cal was 0.9951, R 2 val was 0.9716 and R 2 pre was 0.9923.RMSEC, RMSECV and RMSEP value were 0.51, 1.27 and 1.00 ppm, and RPD value was 7.13.

Variable selection of iPLS and siPLS
Chemometric methods of iPLS and siPLS were applied to select variables of Yunkang Oral Liquid.
The results of iPLS are shown in Fig. 4. The least RMSEP value of interval number was 12, in which the wavelength ranged from 1726 to 1871 nm.For siPLS method, the best combinations of interval numbers were 5, 11 and 13, in which the wavelength ranges were 1124.5-1209.5, 1640-1725 and 1811.5-1892.5 nm (Fig. 5).

2D-COS assignment of Isopsoralen
The 2D-COS method was applied to identify the characteristic peaks of Isopsoralen.The synchronous  spectrum of deuterated chloroform is shown in Fig. 6(a), the strong peaks of deuterated chloroform including 2180-2320 and 2433-2500 nm.Before assigning the characteristic peaks of Isopsoralen, the strong peaks were eliminated.Therefore, the synchronous spectrum of raw spectra was shown in Fig. 6(b).The NIR of Isopsoralen was analyzed by 2D-COS method, the characteristic bands including 1376-1429, 1596-1692 and 2107-2143 nm.According to the structure of Isopsoralen, the band of 1376-1429 nm is the combination frequency absorption of aromatics C-H, 1596-1692 nm was the absorption of vinyl which connected with oxygen and primary frequency of stretching vibration of C-H in aromatic ring, 2107-2143 nm was C-H absorption of benzene ring.These bands are assigned to the characteristics of Isopsoralen. 26Because of strong absorption of water in combinations and ¯rst overtones, 1376-1429 and 1596-1692 nm were chosen to establish the PLS model.

The PLS model of Isopsoralen in
Yunkang Oral Liquid The bands of Chemometrics and 2D-COS were chosen to establish PLS model, respectively.Meanwhile, the numbers of latent factors were investigated coupled with the selection of appropriate bands.Figure 7 and Table 2 showed the relationship with latent factors and PRESS in di®erent variable selection methods using leave-one cross validation.The R 2 cal ; R 2 val and R 2 pre of 2D-COS spectral assignment were 0.9982, 0.9982 and 0.9988, RMSEC, RMSECV, RMSEP and RPD of 2D-COS spectral assignment were 0.31 ppm, 0.32 ppm, 0.22 ppm and 32.41.The result is good as siPLS and iPLS.However, the bands of 2D-COS spectral assignment are more signi¯cant and the PLS model was stable and robust.The phenomena in which di®erent bands were chosen by di®erent chemometric methods for the same substance was overcome.It provided a reference method and guidance for the variable selection of NIR application.

Conclusion
In this study, NIR assignment of Isopsoralen was selected using deuterated chloroform solvent and 2D-COS technology.According to the auto-peaks of 2D-COS synchronous spectrum, the characteristic bands of Isopsoralen included 1376-1429, 1596-1692 and 2107-2143 nm.Combined with the structure of Isopsoralen, the band 1376-1429 nm is the combination frequency absorption of aromatics C-H.Band 1596-1692 nm was the absorption of vinyl which connected with oxygen and primary frequency of stretching vibration of C-H in aromatic ring.Band 2107-2143 nm was the C-H absorption of benzene ring.NIR collection of Isopsoralen was used as the CQA in di®erent concentrations of Yunkang Oral Liquid, the PLS model using 2D-COS bands, iPLS and siPLS bands obtained good results.Therefore, it provides a reference for NIR spectral assignment of chemical compositions in CMM, and provides guidance for the stable and robust variable selection of NIR application.

Table 1 .
Di®erent pretreatment methods of PLS model.
Note: The unit of RMSEC, RMSECV and RMSEP is ppm.

Table 2 .
The evaluation parameters of PLS model.Note: 1376-1429 and 1596-1692 nm were the spectral assignment of Isopsoralen The unit of RMSEC, RMSECV and RMSEP is ppm.J. Innov.Opt.Health Sci.Downloaded from www.worldscientific.comby FLINDERS UNIVERSITY LIBRARY on 01/23/15.For personal use only.