Application of Near-Infrared Spectroscopy (NIRS) as a Tool for Quality

Current Bioactive Compounds 2011, 7, 000-0001 Application of Near-Infrared Spectroscopy (NIRS) as a Tool for Quality Control in Traditional Chinese Medicine (TCM)L.P. Guo1, L.Q. Huang1,*, X.P. Zhang1, L. Bittner1, C. Pezzei1, J. Pallua1, S. Schönbichler1, V.A.Huck-Pezzei1, G.K. Bonn1 and C.W. Huck2,*1Institute of Chinese Materia Medica, China Academy of Chinese Medical Sciences, Beijing, China, 2Institute of Ana-lytical Chemistry and Radiochemistry, Leopold-Franzens University, Innrain 52a, 6020 Innsbruck, AustriaAbstract: Traditional Chinese Medicine (TCM) is becoming more and more popular all over the world. Novel analyticaltools for quality control are highly demanded enabling analysis starting at breeding and ending at biological fluids includ-ing urine or serum. Compared to analytical separation methods (chromatography, electrophoresis) near-infrared spectros-copy (NIRS) allows analyzing matter of interest non-invasively, fast and physical/chemical parameters simultaneously. Itcan be used for the quantitative control of certain (active) ingredients. In many cases identification can only be achievedby pattern recognition. Therefore, NIRS combined with cluster analysis offers huge potential to identify e.g. species, geo-graphic origin, special medicinal formula etc. In the present contribution the fundamentals, possibilities of NIR applied inquality control of TCM are pointed out and its ad- and disadvantages are discussed in detail by several practical examples. Keywords: Near-Infrared Spectroscopy (NIRS), Quality, Traditional Chinese Medicine (TCM), Geoherbs, Processing.1. INTRODUCTIONAuthentication of plant material and origin, identification of parts of the plant, qualitative and quantitative analysis of primary and secondary metabolites are the demanded ana-lytical key challenges to efficiently ensure quality in Tradi-tional Chinese Medicine (TCM) [1]. Traditionally applied identification, physical and chemical description follow dis-tinct and complex experience rules [1,2], which in many cases do not enable an exact, clear and objective determina-tion. Often only a (dried) part of the plant (or even animal) is present and then identification is based on personal and sub-jective operator decisions. Due to these circumstances, sev-eral new analytical techniques were established during the last three decades enabling a more objective analysis: The introduction of liquid chromatography (LC) [3] in the early 80ies and capillary electrophoresis (CE) [4] allowed a fast separation of low and high molecular weight ingredients. Due to the establishment of LC coupled to mass spectrome-try (MS) via an electrospray interface (ESI), which was hon-ored with the Nobel Prize given to Fenn in 2002, much more complex samples of interest could be put under considera-tion. Different kinds of carrier materials applied as a station-ary phase in solid-phase extraction (SPE) were designed to selectively enrich analytes of interest deriving from crude biological matrices including plant extracts, serum and urine [5]. Biochemical methods including DNA marker and coding were introduced for species identification [6]. Although, each of these methods is highly efficient, several different expensive machines controlled by special trained staff are *Address correspondence to these authors at the Institute of Chinese Materia Medica, Chinese Academy of Chinese Medical Science, Beijing 100700, China; Tel: +86 10 64014411; Fax: +86 10 6401 3996;E-mail: huangluqi@; and Institute of Analytical Chemistry and Radiochemistry, Leopold-Franzens University Innrain 52a, 6020 Innsbruck, Austria; Tel: +43 512 507 5195; Fax: +43 512 507 2965; required and finally, experiments are time consuming, being hardly suitable to high-throughput analysis [7]. Near infrared spectroscopy (NIRS), which was already introduced in 1980, offers several advantages, including fast and simultaneous determination of different physical and chemical parameters, as well as easy operation at low costs after a careful calibra-tion of the entire system. In the following chapters the fun-damental principle, potential, ad- and disadvantages of NIRS applied as a tool for quality control in TCM are pointed out and discussed in detail.2. FUNDAMENTAL PRINCIPLES OF NEAR-INFRARED SPECTROSCOPYNear-infrared spectroscopy (NIRS) is a spectroscopic method using the region of the electromagnetic spectrum from 4.000 to 12.800 cm-1 (2500 – 780 nm) [8], which was already discovered by Herschel in 1800. The NIR region covers the overtone and combination transition vibrations of mainly the C-H, O-H and N-H groups. The molar absorbance in the NIR region is typically quite small and due to broad signals more difficult to identify in comparison to mid-IR (MIR, 2.500 – 25.000 nm) spectra because of the higher grade of overtone and combination excitations. For spectral data evaluation two methods, raw spectra interpreta-tion and chemometrics/multivariate data analysis (MVA) are commonly used to elucidate the NIR spectra [9]. Visual spectra interpretations and absorbance band assignments play an important role, especially for the comparison of pure materials but also of rather complex spectra [10]. MVA based calibration techniques are applied to combine the spec-tral data with target parameters transferred from reference techniques or to expose similarities and hidden data struc-tures in the spectra [11-13]. NIRS coupled with spectra pre-treatment methods (derivatives, smoothing, normalization, filters, and baseline- and multiplicative correction methods) and multivariate methods (e.g., principal component analy-2 Current Bioactive Compounds 2011, Vol. 7, No 2 Guo et al.ses, PCA; partial least squares, PLS; multiple linear regres-sion, MLR; principal component regression, PCR) has been successfully used for the simultaneous analysis of chemical and physical parameters in agriculture, pharmaceutical and material analysis [14-16]. One of the forerunners of modern NIR applications, Karl Norris of the U.S. Department of Ag-riculture, used the NIR wavelength region for the spectral analysis of moisture content of grain and seeds [17], which at the same time was the first application of NIR in plant analysis. Radiation in the NIR region can typically penetrate deeper into a sample than MIR. Instrumentation for NIRS is suitable either for measurement in reflection/transflection (R), transmission (T) or interaction (I) mode (Fig. 1). It has generally been described to be very useful in probing bulk material with little or no sample preparation. NIR is a non-invasive, fast analytical technique since the sample of inter-est (tissue, extract, tablet, etc.) must not be destroyed for the analytical procedure. Next to this, NIRS possess the follow-ing additional advantages over other analytical techniques: Chemical (class of plant ingredients) and physical parame-ters (solvent composition, viscosity, pH, and conductivity) can be determined simultaneously; Measurements are robust and cheap; Analyses can be carried out off-line, on-line or in-line; High suitability for automation and high-throughput screening is guaranteed and measurements do not require special trained staff.3. NIR IN QUALITY CONTROL OF TCM BY QUAN-TITATIVE MEASUREMENTSIn many cases quality control is achieved by quantitative measurement of interesting components. Therefore, it is es-sential to calibrate the NIR system with a suitable set of samples, analysed by a reference method. Reference analysis is carried out by chromatographic/electrophoretic methods including iquid chromatography (LC), LC coupled to mass spectrometry (MS), micro-liquid chromatography ( -LC), gas chromatography (GC), capillary electrophoresis (CE) and capillary electrochromatography (CEC) or wet chemical analysis (titration etc) [3,7]. Thus, Huck et al. introduced in 1999 a strategy, which enables determination of plant con-tent in a multi-plant extractive system by analysing its corre-sponding leading compound (Fig. 2) [19]. Furthermore, other parameters e.g. pH, viscosity, solvent composition can be determined simultaneously by calibrating the system with the appropriate reference method. The suitability of this strategy of analysis was successful demonstrated by simulta-neously analysing the leading compound 3´,4´,5´-trime-thoxyflavone, water and ethanol content in a huge sample set of Flos Primulae veris [19] and was found to be also appli-cable to the analysis of St. John´s Wort [20] extracts.The biggest obstacle for quality control in TCM is the fact that there are often many compounds (some time more than hundred) in a raw material and no information is present about the individual health effect, which is valid for most Traditional Chinese Medicines. In this case, the second qual-ity control approach by NIRS is the establishment of a quali-tative cluster model, which can be suitable for a fast authen-tication and identification of raw material recording to their origin and composition, respectively. In the past, this method was shown to be highly efficient for controlling the sort, origin and year of wines [21, 22].4. APPLICATIONS OF NIR IN TCMIn Austria NIRS in the analysis of medicinal plants (“phytomics”) has been introduced at the Institute of Ana-lytical Chemistry and Radiochemistry, Universtiy of Inns-bruck, in 1999 [19]. In parallel, approximately at the same time NIRS was established as a novel tool for quality control in China. Now, NIRS is used not only for authentication, identification and quantification of raw material, but also for process quality control and/or extraction monitoring. Not only single raw materials, but also complex formulas are current subjects of investigation. TCM includes besides me-dicinal plants also medicine prepared from animals, fungus and minerals; NIRS is used not only for monitoring the sec-ondary metabolites, but also for the determination of addi-tional parameters, e.g., fiber, moisture, etc. Due to shortApplication of Near-Infrared Spectroscopy (NIRS) Current Bioactive Compounds 2011, Vol. 7 No. 2 3measurement times of only a few seconds, the use of optical fiber probe makes NIRS attractive for on-line monitoring. In Fig. 3 the main application fields of NIRS in TCM are sum-marised including control of•raw materials•production and extraction processes•preparation of medicinal formulation,which is described in detail in the following chapters.4.1. Control of Raw MaterialsCompared to chromatography, electrophoresis and MS no sample destruction is required for NIRS. Information can be gathered from the intact entire piece of sample. This cir-cumstance makes NIRS the preferred tool for pattern recog-nition of raw materials applied to•identify falsification•classify material into species, geographic regions •identify multi-originated raw materials•identify geographical provenance (“geoherbs”) and hab-its•quality control by quantitative analyses of ingredients 4.1.1. Identification of FalsificationTruth or false identification is the first step in the charac-terisation of raw materials. Xiang et al. [23] as well as Tang et al. [24] described the identification of 25 official and 27 a tailored artificial neural network (ANN). Recorded spectra were compressed by wavelength transformation (WT) and allocated into clusters. The established model allowed identi-fying true/false with a selectivity of 96 % [23]. Zhao et al. applied wavelength packet entropy and Fisher classification to identify medicinal rhubarbs [25], while Zhang et al. used vector machines [26].Zhonget al. established a cluster model of Pollen Typhae from different sources, which is used to relieve blood stasis, stop bleeding, treat stranguria and aponicas pain. The estab-lished model used the 2nd derivative spectra, vector normali-zation and factor method for classification [27].4.1.2. Classification Into Species, Geographic RegionsIn some cases original plants or animal species and origin are unconfirmed and disputable. NIRS based cluster analysis offers solutions to answer both questions. Licorice (the roots of Glycyrrhizia uralensis Fisch) is used as a medicinal herb and also as a food additive in China. The classification of licorice samples according to their growing conditions (Fig. 4a), geographic areas (Fig. 4b) and plant parts was devel-oped using fiber optic diffuse reflection NIR spectroscopy (FODR-NIR). With the use of multiplicative signal correla-tion (MSC) and Norris derivative filtering, the differences of the NIR spectra among different licorice samples were en-hanced even though the raw spectra showed only slight dif-ferences. The results showed that the NIR spectra of the samples were moderately clustered in the principle compo-nent spaces. Pattern recognition of soft independent model-ing of class analogy (SIMCA) provided satisfactory classifi-Fig. (2). Strategy of analysis to establish a calibration model in NIRS.4 Current Bioactive Compounds 2011, Vol. 7, No 2 Guo et al.Fig. (4). PC score plot of licorice samples originated from different (a) growing conditions (Tandi , Liangdi , shadi and (b) geo-graphic areas (Gansu , Inner Mongolia ). Reproduced from reference [28], with permission.method using HPLC data set as reference was constructed to predict the value of glycyrrhizic acid (GA) in licorice. The results showed that PLS models with both data normalization (DN) coupled with first derivative and MSC pretreatments provided acceptable results [28]. Liu et al. used NIRS based cluster analysis and discriminative analysis to classify Yangti (Chinese herb from Rumex patientia L., R aponicas Houtt, R chalepensis Mill and R dentatus L.). The obtained Wu et al. classified Baizhi (Angelica anomala, A dahurica, A dahurica cv. Hangbaizhi, A dahurica cv qibaizhi, A. porphy-rocaulis and A formosana) by NIRS coupled with pattern recognition. The results showed that the elaborated NIRS method can provide information for identifying species of these herbal medicines [30].4.1.3. Identification of Multi-Originated Raw MaterialsRaw materials composited of several plants or animal species deriving from different origins are called “multi-originated”. In most cases they belong to the same genus and have similarity in morphology, secondary metabolites and others. To indentify each component and proof for falsifica-tion is much more complicated than in case of a single raw material. According to TCM, the combination of several different methods is necessary for the identification of multi-original materials.NIRS allows to simplify this procedure applying cluster analysis. For example, 7 certified Fritillaria species, i.e., F. przewalskii Matim, F. cirrhosa D. Don, F. unibraacteate Hsiaaoet. K. C. Hsia, F. thunbergii Mig, F. pallidiflora, F. ussuriensis Matin and F. kupehensis Hsiaet K. C. Hsia, and 3 fake species, i.e., F. thunbergii Varcheki-ugensis Hsiaet K. C. Hsia, Tulipa edulis and Iphigeniaindica were dried, grinded, sieved and studied by NIR applying cluster analysis, convolution and transform-visualization-similarity analysis. In the frist step, this method enabled dif-ferentiation between true and false species but no assignment of individual species. Finally, convolution transform –visualization-similarity analysis allowed to magnify and also quantify the minute differences between the certified Fritil-laria species [31]. Achillea millefolium and 3 of its related species, namely, A. clypeolata, A. collina and A. nobilis were discriminated by principle component analysis (PCA) [32].42 different Cnidium monnieri (L ) Cusson species and their origins were identified by Cai et al. [33].4.1.4. Identification of Geoherbs and HabitsThe term “geoherblizm” was introduced in ancient Chi-Fig. (3).Flow diagram of NIRS application fields in TCM.Application of Near-Infrared Spectroscopy (NIRS) Current Bioactive Compounds 2011, Vol. 7 No. 2 5 used as a synonym for good quality. A top-geoherb is geo-herblizm with superior quality originating from preferredgeographical areas. This “geoherblizm” principle is appliedas a quality standard controlling method of TCM raw mate-rials. The traditional identification strategy is based on expe-rience and in many cases difficult underlying subjective de-cisions. Via NIRS geoherblizm can be characterised usingcluster analysis.Wang et al. collected 102 samples of Cordyceps fromTibet and Qinghai province (Cordyceps is the top-geoherb ofCordyceps according to TCM growing in Tibet), and usedNIR in diffuse reflection and transmission mode for theanalysis of both its milled and extracted form, respectively.The result showed that all Cordyceps samples can be allo-cated according to their source [34]. Wang et al. investigated57 samples of Panax gensing from Jilin province and 60from Liaoning province (p. gensing from Jilin province is thetop-geoherb of P. gensing) by NIR diffuse reflection spec-troscopy. After performing Savitzky-Golay filtering, calcula-tion of first and second derivative spectra, they found moreintense NIRS absorption in Jilin P. gensing than Liaoning P.gensing due to the higher content of ingredients accompa-nied by smaller scattering and shifting effects [35]. Zhang et al. identified Forsythia suspense from 5 different habits by NIRS using pattern recognition based on SIMCA. 5 predic-tive models were built separately, only 1 sample among 10 prediction samples could not be indentified correctly [36]. Xing et al. successfully identified Red Kojic made by M. purpureus fermentation(Fermentum Rubrum)from 18 dif-ferent habits by NIR diffuse reflection spectroscopy and cluster analysis [37]. Yi et al. used a NIRS approach to dis-criminate Ganoderma lucidum according to its cultivation area. Raw, first, and second derivative NIR spectra were compared to develop a robust classification rule. The chemi-cal properties of G. lucidum samples were also investigated to find out the difference between samples from six different origins. It could be found that the amount of polysaccharides and triterpenoid saponins in G. lucidum samples was consid-erably different based on cultivation area. Principal compo-nent analysis (PCA), discriminant partial least-squares (DPLS) and discriminant analysis (DA) were applied to clas-sify the geographical origins of those samples. For the dis-crimination of samples from three different provinces, DPLS provided 100% correct classifications. Moreover, for sam-ples from six different locations, the correct classifications of the calibration as well as the validation data set were 96.6% using the DA method after the SNV first derivative spectral pre-treatment (Fig. 5) [38].4.1.5. Quality Control by Quantitative Analyses of Ingredi-entsOne important point in quality control is the presence of certain compounds at specific concentration levels. There-fore, ingredients of interest in raw materials or deriving ex-tracts can be determined directly and quickly by applying quantitative regression analysis, which is established by cali-brating NIR values against true values deriving from refer-ence (such as gas chromatography, GC; high performance liquid chromatography, HPLC; electrophoresis, EP; super-critical fluid chromatography, SFC etc). Fig. (5). Three-dimensional score plot using PC1, PC2, and PC3 for discriminationg six Ganoderma lucidum origins, class 1, Jiaxiang; class 2, Huangshan; class 3, Taishan; class 4, Longquan; class 5, Jinzhai; class 6, Jingdangpu. Reproduced from reference [37], with permission.Yang et al. applied NIRDS and back propagation basedartificial neural network (BP-ANN) to realize fast determina-tion of the mannitol content in a range from 8.08% to14.55% in fermented Cordyceps sinensis powder. For mod-elling, first derivative NIR spectra of fermented powder,three different multivariate calibration strategies were em-ployed, including principal component regression (PCR),partial least square regression (PLSR) and BP-ANN. Fur-thermore, the root mean square error of cross validation(RMSECV) and the root mean square error of prediction(RMSEP) were selected as the indices for evaluating andcomparing the performance of calibration models. Within the wavenumber ranges of 7.501.7, 6.097.8 cm-1 and 5.453.7,4.246.5 cm-1, the obtained lower values of RMSECV (0.475) and RMSEP (0. 608) indicated that BP-ANN was thesuperior utilisation tool [39]. Ye et al. used NIR reflectancespectroscopy for quantifying isorhamnetin between 0.1%-0.8% in Hippophae rhamnoides Linn from West Sichuan plateau. Calibration models were established using the PLS(partial least squares) within the range of 12.000 – 4.000cm-1. Different spectra pre-treatments methods were compared. The study showed that spectral information can be extractedthoroughly by constant off set elimination (COE) pre-treatments method with the correlation coefficient r2 of 0.7398 , SEC of 0.107 and SEP of 0.073 ( standard deviationof the prediction sets) [40]. Bai et al. determined ecdyster-one’s content in Radix Achyranthis Bidentalae applying(PLS). The result showed that the correlation coefficient ofthe quantitative mathematics model between the predictionand the true values was 0.9489 [41]. Yang et al. detected the content of flavonoids in ginkgo leaves. The result indicated6 Current Bioactive Compounds 2011, Vol. 7, No 2 Guo et al.to real if precision is high by chemical reference analyses [42]. Liu et al. determined arteannuin in Artemisia annua L. The corresponding model was established by PLS. RMSEP and r2 of the validation set samples of the model based on 6 coefficients were 0.544‰ and 0.998, respectively [43].In some cases it is necessary to detect several independ-ent active ingredients in a formulation simultaneously. Ana-lyzing five components (total isoflavones, puerarin, daidzin, starch, crude protein) components of P. Lobata(Pueraria lobata Willd Ohwi) demonstrated that the correlation be-tween the chemical value (true value) of the five components of sample set and the NIR predicated value was 0.9752, 0.9839, 0.9659, 0.9628 and 0.9829, respectively. The corre-lation between the calibration and test set was 0.9818, 0.9752, 0.9772, 0.9737 and 0.9798, respectively [44]. 14 relevant compounds in the medicinal plant Achillea millefo-lium were detected by NIRS using gas chromatography-mass spectrometry (GC-MS) as a reference technique. PLSR was used to create 14 single-compound models (SCM, one re-gression model for each compound) on one hand and one multi-compound model (MCM, one regression model for 14 compounds) on the other hand. SEP was 0.49 % for SCM and 0.62 % for MCM. Paired t-test and one way analysis of variance (ANOVA) showed both SCM and the MCM work well in quantities of the compounds in A. millefolium. Pear-son bivariate correlation, principle component analysis (PCA) and hierarchical cluster analysis were conducted to uncover the significant relationship between the 14 com-pounds [45].4.2. Processing and ExtractionIn the following two chapters the suitability of NIRS to monitor the quality during the production and extraction pro-cedure are summarized.4.2.1. Processing Quality Control (QC)Most raw materials are further processed prior to clinical use or manufacturing according to TCM theories and clinic practices. The commonly used processing steps include cleaning, freezing, boiling, steaming, etc. Sometimes, vine-gar, honey, salt is added. Thereby, processing ensures dosage and removement by purification on one hand and improve-ment or change of the effect on the other hand. In many cases, the toxicity of the used raw materials is reduced. For example, heads and legs of Chantui (periostracum cicadae, Cryptotympana pustulata Fabricius) is removed; the small hair on leaves loquat (Eriobotrya japonica) are brushed off in order to avoid itch of throats, Dahuang (Rhizoma et Radix Rheum palmatum, R Tangutici and R Officinalis) is be steamed with wine to avoid abdominal pain or receive loose bowels. It is obviously that both physical and chemical pa-rameters are changed. Processed and unprocessed raw mate-rials are declared as two different kinds of medicine in TCM practice, i.e. dried Rehmannia root and streamed Rehmannia root (radix Rehmanniae preparata).The commonly used identification methods include expe-rienced identification, and some simple physical or chemical identification procedures. NIRS provides a very fast, user-friendly identification of processed materials. American ginseng (Panax quinaquefolia) and Asiatic ginseng (Panax ginseng C. A. Mey) as well as Asiatic ginseng processed products were analyzed by NIRS in diffuse reflection mode. The result show that American ginseng and Asiatic ginseng are more similar than Asiatic ginseng processed products, which indicated that processing can change the Asiatic gin-seng [46]. Bai et al. determined the reducing sugar content in powder of decoction pieces of Shu Dihuang (radix rehman-niae preparata) stewed with wine by FT-NIRS and data analysis. Cross validation and test samples determination showed that the correlation coefficient of the prediction model were 89.02 and 88.47, the RMSECV were 0.962 and 0.887, respectively. t-Test confirmed that there was no sig-nificant difference between the true data and predictive value [47].4.2.2. Extraction Quality ControlDuring recent years, extracts have even become more popular in TCM, because they are easy to prepare and are known to have a good patience compliance. The collected powder samples of Ginkgo biloba extracts were analyzed by HPLC and NIR, followed by chemometrical data treatment. The results showed that MPLS (modified partial least squares) regression model gave the best result. The multi-correlation coefficient of the prediction samples was 0.973, the recovery 96.15%-102.0%, with RSD of 1.1%. The elabo-rated method indicated that the total flavones in powder of G. biloba extract could be determined directly with high ac-curacy [48].NIRS was used for measurement of water content in 5 kinds of TCM extracts, namely Radix Scutellariae, For-sythia suspense, Flos Lonicerae, Cornu gorais and Bear gall powder. Spectra were recorded with the average of 64 scans over the spectral region 4.000 – 10.000 cm-1 at 8 cm-1 inter-val. The wavenumber bands containing 3 characteristic wavelengths of water (6.900, 5.200, 5.180 cm-1) were se-lected and pretreated applying multiplicative scatter correc-tion, Savitzky - Golay filtering, and first derivative. The calibration was obtained applying PLSR and optimized via inner cross validation and external validation. Results showed that the coefficients of correlation of inner cross validation and external validation were both above 0.90, and both RMSECV and RMSEP below 0.05. The calibration models were used for testing a new set of unknown samples, and the results were highly satisfying. The presented method is timesaving and accurate, which indicates potential to be widely used in the water content determination of TCM ex-tracts [49].Panax notoginseng herb extracts were investigated by NIR, HPLC and colorimetric method to monitor the content of ginsenosides Rg1, Rb1, Rd and saponins. The NIRS cali-bration models of ginsenoside Rg1, Rb1, Rd were built by using support vector regression (SVR). This method was compared with partial least square regression (PLSR) and radial-basis function neural network (RBFNN) modeling methods. The results showed that the predictive accuracy of NIR calibration models built by SVR was much better than that of the models built by PLSR and RBFNN [50].。