lable at ScienceDirect

Engineering in Agriculture, Environment and Food 7 (2014) 139e142

Contents lists avai

Engineering in Agriculture, Environment and Food

journal homepage: http: / /www.sciencedirect .com/eaef

Research paper

Chlorogenic acid (CGA) determination in roasted coffee beans by NearInfrared (NIR) spectroscopy*

Jiajia Shan a, *, Tetsuhito Suzuki a, Diding Suhandy b, Yuichi Ogawa a, Naoshi Kondo a

a Graduate School of Agriculture, Kyoto University, Kitashirakawa-Oiwakecho, Sakyo-ku, Kyoto 606-8502, Japanb Department of Agricultural Engineering, Lampung University, Jl. Soemantri Brojonegoro No. 1 Gedongmeneng Bandar Lampung, Lampung, Indonesia

a r t i c l e i n f o

Article history:Available online 27 August 2014

Keywords:NIR spectroscopyPLSChlorogenic acidDifferent roasting degreesCoffee

* Partly presented at the Sensing Technologies forculture Conference in April 2013.* Corresponding author.

E-mail address: sjjkyoto@gmail.com (J. Shan).

http://dx.doi.org/10.1016/j.eaef.2014.08.0031881-8366/© 2014, Asian Agricultural and Biological

a b s t r a c t

Roasting can potentially affect the functional properties of the resulting coffee. For instance, antioxidantphenolic chlorogenic acids (CGA) are known to be thermally labile. In this study, we develop a new NIRspectroscopic method to measure CGA concentration found in the coffee beans. High performance liquidchromatography (HPLC) was used to determine standard reference values of CGA concentration. A partialleast squares regression (PLSR) with full cross-validation method was employed for the development andthe validation of the regression model. The resultant model accuracy of RCV2 ¼ 0.76 and RMSECV ¼ 1.16%with preprocessing of the spectra by MSC and SNV is thought to be accurate enough for the rapid andnon-destructive determination of CGA concentration in roasted coffee beans.© 2014, Asian Agricultural and Biological Engineering Association. Published by Elsevier B.V. All rights

reserved.

1. Introduction

In international trade, coffee is the second most important rawmaterial traded, and themost important foreignexchange earner formany developing countries. For this reason coffee is an attractivesource of tax revenue. As a consequence, in this highly competitivemarket, ensuring premium quality and customer satisfaction is ofutmost importance to the continued success of the coffee business.One of the functional quality characteristics of coffee which areattracting increasing interest is its antioxidant properties.

Within this class of antioxidant phenolic compounds, chloro-genic acid (CGA) has been proved to play an important role in hu-man health, as well as contribute to the acidity and overall qualityof the final cup of coffee, such as bitterness (Tfouni et al., 2012).However, CGA is thermally unstable, which means it can undergothermal degradation during the roasting stage. As a result, thecontent of CGA in roasted coffee beans could be several times lessthan that in the green coffee beans (Perrone et al., 2010). The effectof roasting on the CGA composition of coffee has been studied(Trugo and Macrae, 1984b) with it being well-established that CGAprogressively degrades with increasing roasting severity. Thesestudies, however, used time-consuming and laborious methods, so

Biomaterial, food and Agri-

Engineering Association. Published

it is necessary to develop a method that can detect CGA concen-tration quickly.

At present, high-performance liquid chromatography (HPLC) isthe standard method used to quantify CGA concentration in instantcoffee (Trugo and Macrae, 1984a), as well as in green and roastedcoffee beans (Trugo and Macrae, 1984b). However, this HPLCmethod has a number of limitations; it requires a complex samplepreparation, it is time consuming, and difficult to set up for onlinedetection.

In contrast, spectroscopy has been increasingly used for rapiddetermination of chemical composition of agricultural productsrecently. Compared to traditional methods, such as HPLC, spectro-scopic methods usually require little sample preparation and lessmeasurement time. While, UV-Vis spectrometry has been used todetermine CGA concentration in aqueous solutions of coffee (Belayand Gholap, 2004), the caffeine, however, has to be extracted withdichloromethane before CGA concentration can be determined inthe solution. A direct nondestructive determination of CGA in coffeebeans without extraction is impossible, due to its spectral overlapwith caffeine in theUV-Vis region. Absorption in theUV-Vis region isa result of electronic transition motions. UV-Vis spectroscopy iscommonly used for special compound (such as a conjugated systemwith special functional groups) identification because most com-pounds don't have absorption in the UV-Vis region.

Compared to the UV-Vis region, near-infrared spectroscopy(NIRS) records more information about sample characteristics, suchas basic molecular bonds of chemical components, such as CeH,

by Elsevier B.V. All rights reserved.

Fig. 1. NIR absorbance spectra of the same green coffee sample with different particlesizes.

J. Shan et al. / Engineering in Agriculture, Environment and Food 7 (2014) 139e142140

NeH, and OeH bonds, generating a spectrum that presents a moredetailed characterization of the sample. In addition, it also has thepotential for online processing. Thus, if NIR spectroscopy could bedeveloped for CGA analysis, it would offer a number of advantages.

To date, NIRS has been used to determine caffeine, theobromineand theophylline in coffee, and CGA in plant samples with wavelettransform preprocessing (Huck et al., 2005; Shao and Zhuang,2004). In addition, it has been used to evaluate the quality ofEspresso sensory properties (Esteban-Diez et al., 2004). FT-NIRspectroscopy has been used to discriminate roasted green tea ac-cording to its geographical origin (Chen et al. 2009). The FT-IR/ATRmethod has been applied to analyze brewed coffee characteristics(Hashimoto et al., 2004). However, to date, there has been no re-ported research on quantifying CGA concentration of coffees basedon NIR spectroscopy.

This paper investigates the potential for using NIR spectroscopyto measure CGA concentration in roasted coffee beans. Variousroasting temperatures and times were used to obtain differentdegrees of roasted coffee. NIR diffuse reflectance spectra ofdifferent roasting degrees of coffee were obtained from groundcoffee powder. Diffuse reflectance spectroscopy measures not onlythe reflectance from the coffee surface, but also information frombelow the coffee surface.

Firstly, the effect of different particle sizes on the obtaineddiffuse reflectance spectra was evaluated. Prior to the NIR spectralanalysis, the CGA concentration of all samples was determined byhigh performance liquid chromatography (HPLC). In this report,partial least squares regression (PLSR) was used to calibrate theconcentration of CGA model with NIR diffuse reflectance spectra.Multiple scattering correction (MSC) was used to remove baselineshift, and full cross validation was used to develop and validate theperformance of the calibration and validation models.

2. Materials and methods

2.1. Sample preparation and chemicals

A batch of Arabica green coffee originating from Antigua Islandwas purchased from a supermarket. In order to obtain a wide rangeof CGA concentration to develop a universal model, the Arabicagreen coffee beans were divided into 65 sub-samples and eachsample was roasted at a different time and temperature. Morespecifically, 38 of these green coffee bean samples were roasted in ahome coffee roaster (Gene Caf�e, CBR-101A) at 5 �C intervals from atemperature of 65 �C to 250 �C for 15min; and 22 of the other greencoffee bean samples were roasted at 5 min intervals from 10 min to60 min at 170 �C and 200 �C, respectively. Another 5 referencegreen coffee bean samples were not processed. After being roasted,each sample was ground in an electric coffee grinder (Cuisinart,DBM-8). Particle sizes were not uniform in the ground coffeepowder. In order to check the effect of particle sizes on diffusereflectance spectra, coffee powder from a green coffee sample wasseparated into different particle sizes (200 mm, 300 mm, 500 mm,600 mm, 700 mm, 1000 mm) by sieving through successively finersieves in ascending order of particle size. The sieving conditionswere the same for every sample class. These experiments wereperformed at room temperature (around 20 �C).

2.2. Near infrared spectroscopy (NIRS)

The NIR diffuse reflectance spectra were recorded by a UV-NIRspectrometer (V-670, JASCO Co.), equipped with an integratingsphere beside the detector (HISN-729, JASCO Co.) which is con-nected to the spectrometer by an optical fiber. Light is firstdispersed by a grating, and then guided by optical fiber to the

integrating sphere. Diffuse reflection light from the coffee powdersample is then integrated and detected. An integrating sphere is anoptical component consisting of a hollow spherical cavity with itsinterior covered with a diffuse white reflective coating, with smallholes at the entrance and exit ports. As a consequence, the inte-grating sphere has a uniform scattering or diffuse effect.

The NIR instrument was controlled by a compatible WindowsXP system, and Spectra Manager (V-670, JASCO Co.) was used toacquire spectra data. The size of the light spot in the instrumentwas 4 cm in diameter. A 4 cm by 4 cm square sample holder wasused; with care taken to ensure every sample (weight of approx.2.45 g) was placed into the sample holder and the sample surfacekept flat. Each spectrumwasmeasured from 1000 to 2700 nm, witha bandwidth of 8.0 nm, and a scan speed of 400 nm/min. All ex-periments were conducted at room temperature (around 20 �C),and the baseline was repeatedly measured every five samples toguarantee that the device was operating correctly.

2.3. High performance liquid chromatography analysis

After NIR diffuse reflectance spectra were taken, the actualreference concentrations of CGA in the coffee beans weremeasuredby a standard high performance liquid chromatography (HPLC)procedure. The HPLC system was equipped with an autosampler(Model SIL-20AC, SHIMADZU, Japan), a UV-Vis detector (ModelSPD-20A, SHIMADZU, Japan), and a shim-pack VP-ODS column(SHIMADZU, Japan). HPLC sample preparation consisted ofextracting the ground coffee with a water and methanol mixturesolution (6:4 volume ratio). The mixture containing the groundcoffee powder was stirred, and then allowed to stand for 5 min.After that, the supernatant was used for subsequent HPLC analysis.The mobile phase was 70% methanol and the flow rate was 0.5 mL/min. The UV detector was set to 324 nm and the volume of sampleinjected was 10 ml. Quantification was achieved by peak areameasurement and compared to the calibration graph, which wasplotted from four pure CGA solutions with a correlation coefficientof 0.999. The main CGA groups in the green coffee beans included:

Table 1CGA concentration in the sampled coffee beans.

The number of sample Range (%) Mean (%) Standard deviation (%)

65 1.7e10.3 4.78 2.06

Fig. 2. NIR absorbance spectra of roasted coffee at different roasting temperatures.

J. Shan et al. / Engineering in Agriculture, Environment and Food 7 (2014) 139e142 141

caffeoylquinic acid, dicaffeoylquinic acid and feruloylquinic acid. Inthis study, the total isomers of CGA were used.

2.4. Data analysis

The whole spectra data (within the wavelength range1000e2700 nm) and the CGA concentration variables were enteredinto the Unscramble software v9.8 (CAMO, Norway), and sequen-tially treated mathematically and finally analyzed statistically.Several pretreatments were applied to find the best relationshipbetween the NIR spectra and CGA concentration. Partial leastsquares regression (PLSR) was employed to develop a calibrationmodel. Full cross-validation was then used to test the performanceof the calibration model, taking into account the number of sam-ples. The signal to noise ratios (SNR) of the spectra decreased after2200 nm due to the instrument properties. Consequently, only thespectra between 1000 and 2200 nm were used in the subsequentanalysis.

3. Results and discussion

3.1. Scattering effect of different particle sizes

NIR absorbance spectra of coffee powder with different particlesizes are shown in Fig. 1. All these different particle sizes of coffeepowder were obtained from one green coffee sample. Differentabsorbance spectra were observed for the different particle sizes ofthe same coffee sample. From Fig. 1, it can be seen that absorbanceintensity increased as particle size increased. The reason for this isthat the space between particles increases as particle size increases.As a consequence, the light absorbed in the increased space be-tween particles contributes to a higher absorbance. For sampleswith the same particle size, absorbance is the same. Therefore, in anattempt to reduce the effect of different particle sizes, and keep CGAas the only variable, we choose to use coffee powder of a certainparticle size in this study. Coffee samples with larger particle sizeshave a lower absorbance, and visa versa. The particle size distri-bution might differ among samples because they have differentmoisture content. But as I did not measure the moisture contentand the particle distribution, I chose 600 mm to eliminate sampleparticle size effect.

3.2. CGA concentration analysis

The total concentration of CGA in the unroasted green coffeebeans was around 10%, and the concentration decreased to around1% during roasting, which has been reported to be due to thethermal breakage of carbonecarbon covalent bonds (Perrone et al.,2010). The CGA concentration range of the coffee samples is shownin Table 1.

3.3. Observations on the NIR spectra

The original absorbance spectra of the different roasted coffeebean samples are shown in Fig. 2. The overall shape of the roastedcoffee bean spectrawasqualitativelyexamined inorder to extract theinfluence of roasting on the spectral profiles. Differences in absor-bance spectra intensity were related to roasting degrees (differentroasting temperature and time). The most significant spectra

differences appeared around 1200e1350 nm, 1450e1850 nm, and1900e2050nm. This is thought to be due to structural changes in thechemical composition during roasting (Ribeiro et al., 2011).

Most absorption bands in the near-infrared region are overtonesor combination bands of the fundamental absorption bonds, whichinclude the CeH bond, NeH bond and OeH bond. Decreases inspectral intensity can be observed in Fig. 2 as roasting temperatureincreased. During roasting, complex chemical reactions occur, suchas the Maillard reaction and pyrolysis, resulting in the release ofwater and volatiles. Thus, water loss and chemical change, such asthe formation of flavor compounds, could explain the observedchanges in absorbance intensity. As we expected, the water bondintensity around 1450 nm (first overtone of OeH stretching), and at1940 nm (a combination band of OeH stretching and OeH defor-mation) decreased gradually with darker roasting of the coffeebeans (higher temperature or longer time). However, the molecularovertone and combination bands (CeH, OeH, NeH) in the NIR re-gion are very broad and complex, and it is difficult to assign specificfeatures to specific chemical components. Thus, multivariate cali-bration techniques are generally used with NIR spectra.

3.4. Regression model of CGA concentration

Outliers in spectral data were identified by the Hotelling's T-square statistic method using the score plot of the PLS model.Hotelling's T-square statistic is a measure of the variation of asample within a PLS model and a high T-square level for a certainsample indicates a high influence of the sample on the model. Theconfidence interval was defined by Hotelling's T-square ellipse witha 95% confidence level. Three samples were identified as outliers inthis experiment.

The capability of the prediction model was evaluated by usingseveral parameters, including the root mean square error of crossvalidation (RMSECV), determination coefficient of cross validation(RCV2 ), and ratio prediction to deviation (RPD). The results obtainedwhen using the PLSR to model CGA concentration on the basis ofNIR spectra in combination with different preprocessing methods,

Table 2Results of the cross validationmodels developed by PLSR based on NIR preprocessedspectra.

Processing LVs Cross Validation

RCV2 RMSECV (%)

MSC 5 0.72 1.161st Derivate 2 0.59 1.432ndDerivate 4 0.45 1.65Normalize 5 0.72 1.17SNV 7 0.76 1.10De-trending 7 0.71 1.20

SNV is standard normal variate.

J. Shan et al. / Engineering in Agriculture, Environment and Food 7 (2014) 139e142142

were used to select the most suitable latent variables (LVs) of eachmodel (see Table 2). Multiple scatter correction (MSC) was firstapplied and then other preprocessing methods were applied on thespectra.

The results indicate that the best regression model for thedetermination of CGA concentration in roasted coffee beans wasachieved when seven LVs were taken into account in the calibrationmodel, in combination with MSC and SNV preprocessed spectradata. The calibrationmodel had a RCV2 of 0.76, and RMSECV of 1.10%.SNV is usually applied on NIR spectra to remove scatter effect. Fromthe results, we could conclude that scattering was creating noiseand interfering with the PLSR model's accuracy. Besides, CGA hasbeen considered to have five groups of isomers. After roasting,considerable changes in the CGA compositionwere observed. Someof the CGA isomers decreased sharply, while others decreasedslightly, while yet others increased due to a transition from oneisomer to another. Thus, the different isomer states will interferewith the results. This was considered to be a part of the errorsource. However, the model still exhibits a high predictive ability.

Fig. 3 shows the scatter plot of reference values of CGA concen-tration vs. prediction values of CGA concentration obtained from the

Fig. 3. Plots of reference CGA concentration values vs. prediction CGA concentrationvalues obtained from cross validation model.

PLSR model in combination with MSC and SNV pretreatments. Thiscan also be used to display the goodness of the final regressionmodel. RCV2 was 0.76, RMSECV was 1.10% and RPD was 2.01, whichindicates the calibration model can accurately predict CGA con-centration. The results show that NIR spectroscopy is a suitablemethod for the quantitative determination of CGA concentration incoffee beans which have been roasted to different degrees.

4. Summary and conclusions

In this study, NIRS was used to determine CGA concentration inground coffee powder. In order to remove the scatter effect associatedwith different particle sizes, we used a specific particle size (600 mm)for all samples when obtaining NIR diffuse reflectance spectra ofpowdered coffee bean samples. The developed model developmentused spectra from 1000 to 2200 nm. In the subsequent analysis, MSCwas first applied to remove baseline shift, and then several pre-processing approaches were applied on the spectra. The best cali-bration results, RCV2 ¼ 0.76 and RMSECV¼ 1.10%, were obtained froma PLSR model with MSC and SNV preprocessed NIR spectra. The re-sults from this paper indicate NIRS, combined with chemometricmethods, can be used to accurately determine CGA concentration incoffee samples. It can also be used to optimize the roasting process soas to retain the greatest amount of CGAwith the appropriate roastingtemperature and time. More studies (such as with different coffeevarieties, particle sizes, more sophisticated mathematical methodsand so on) should be investigated for future online application.

Acknowledgments

The authors thank Professor Garry Piller, Kyoto University forfruitful discussion and support.

References

Belay A, Gholap AV. Characterization and determination of chlorogenic acid (CGA)in coffee beans by UV-Vis spectroscopy. Afr J Pure Appl Chem 2004;3:234e40.

Chen QS, Zhao J, Lin H. Study on discrimination of roast green tea according togeographical origin by FT-NIR spectroscopy and supervised pattern recognition.Spectrochim Acta Part A 2009;72:845e50.

Esteban-Diez I, Gonzalez-saiz JM, Pizarro C. Prediction of sensory properties ofespresso from roasted coffee samples by near-infrared spectroscopy. Anal ChimActa 2004;525:171e82.

Hashimoto A, Mori H, Kanou M, Yamanaka A, Kameoka T. Mid-infrared spectro-scopic analysis on brewed coffee characteristics. In: 10th Asia-Pacific Confer-ence of Chemical Engineering; 2004.

Huck CW, Guggenbichler W, Bonn GK. Analysis of caffeine, theobromine andtheophylline in coffee by near infrared spectroscopy (NIRS) compared to high-performance liquid chromatography (HPLC) coupled to mass spectrometry.Anal Chim Acta 2005;538:195e203.

Perrone D, Donangelo R, Donangelo CM, Farah A. Modeling weight loss andchlorogenic acids content in coffee during roasting. Agric Food Chem 2010;58:12238e43.

Ribeiro JS, Ferreira MMC, Salva TJG. Chemometric models for the quantitativedescriptive sensory analysis of Arabica coffee beverage using near infraredspectroscopy. Talanta 2011;83:1352e8.

Shao XG, Zhuang YD. Determination of chlorogenic acid in plant samples by usingnear-infrared spectrum with wavelet transform preprocessing. Anal Sci2004;20:451e4.

Turgo LC, Macrea R. Chlorogenic acid composition of instance coffee. Analyst1984a;109:263e6.

Trugo LC, Macrae R. A study of the effect of roasting on the chlorogenic acidcomposition of coffee using HPLC. Food Chem 1984b;15:219e27.

Tfouni SAV, Serrate CS, Carreiro LB, Camargo MCR, Teles CRA, Cipolli KMVAB, et al.Effect of roasting on chlorogenic acid, caffeine and polycyclic aromatic hydro-carbons levels in two coffea cultivars: coffea Arabica cv. Int J Food Sci Technol2012;47:406e15.