Food Chemistry 226 (2017) 202–211

Contents lists available at ScienceDirect

Food Chemistry

journal homepage: www.elsevier .com/locate / foodchem

Rapid prediction of phenolic compounds and antioxidant activity ofSudanese honey using Raman and Fourier transform infrared (FT-IR)spectroscopy

http://dx.doi.org/10.1016/j.foodchem.2017.01.0240308-8146/� 2017 Elsevier Ltd. All rights reserved.

⇑ Corresponding author.E-mail address: zou_xiaobo@ujs.edu.cn (Z. Xiaobo).

Haroon Elrasheid Tahir a, Zou Xiaobo a,⇑, Li Zhihua a, Shi Jiyong a, Xiaodong Zhai a, Sheng Wang a,Abdalbasit Adam Mariod b,c

a School of Food and Biological Engineering, Jiangsu University, 301 Xuefu Rd., 212013 Zhenjiang, Jiangsu, ChinabCollege of Sciences and Arts-Alkamil, University of Jeddah, P.O. Box 110, Alkamil 2193, Saudi ArabiacDepartment of Food Science & Technology, College of Agricultural Studies, Sudan University of Science & Technology, P.O. Box 71, Khartoum North, Sudan

a r t i c l e i n f o

Article history:Received 18 June 2016Received in revised form 28 December 2016Accepted 6 January 2017Available online 7 January 2017

Keywords:HoneyAntioxidant activityPhenolic compoundsSpectroscopyHPLC-DADPartial least square regression (PLSR)

a b s t r a c t

Fourier transform infrared with attenuated total reflectance (FTIR-ATR) and Raman spectroscopy com-bined with partial least square regression (PLSR) were applied for the prediction of phenolic compoundsand antioxidant activity in honey. Standards of catechin, syringic, vanillic, and chlorogenic acids wereused for the identification and quantification of the individual phenolic compounds in six honey varietiesusing HPLC–DAD. Total antioxidant activity (TAC) and ferrous chelating capacity were measuredspectrophotometrically. For the establishment of PLSR model, Raman spectra with Savitzky-Golaysmoothing in wavenumber region 1500–400 cm�1 was used while for FTIR–ATR the wavenumber regionsof 1800–700 and 3000–2800 cm�1 with multiplicative scattering correction (MSC) and Savitzky–Golaysmoothing were used. The determination coefficients (R2) were ranged from 0.9272 to 0.9992 forRaman while from 0.9461 to 0.9988 for FTIT-ART. The FTIR–ATR and Raman demonstrated to be simple,rapid and nondestructive methods to quantify phenolic compounds and antioxidant activities in honey.

� 2017 Elsevier Ltd. All rights reserved.

1. Introduction

Phenolic compounds are among the most chemically heteroge-neous groups of compounds produced in plants. Phenolic acid andflavonoid compounds are the utmost important groups ofphenolic compounds, with approximately 1000 compoundsalready reported (da Silva, Gauche, Gonzaga, Costa, & Fett, 2016).The concentration of these compounds depicts the quality of honeyand accounts not only for antioxidant activity but also for its sen-sory features such as color (Lachman, Orsák, Hejtmánková, & Kovárová, 2010; Tahir, Xiaobo, Zhihua, & Yaodi, 2015). Previous studiesrevealed the strong relationship between consumption of honeyand low risk of cardiovascular diseases and cancer (Erejuwa,Sulaiman, & Ab Wahab, 2014; Tomasin, de Andrade, & Gomes-Marcondes, 2015).

Nowadays, a great attention has been directed to the use of nat-ural antioxidants from honeybees and honey products. Antioxidantactivity and phenolic compounds have become the most importantfactors to evaluate the quality and functional properties of honey

(Lianda, Sant’Ana, Echevarria, & Castro, 2012). High-performanceliquid chromatography combined with photo diode array detectoror diode array detector (HPLC-PDA/DAD) is the conventionalmethod used in the determination of phenolic compounds inhoney. Despite its precision, however, conventional method istime-consuming, expensive and require a large amount of solvents(Kralj Cigic & Prosen, 2009). Therefore, simple, rapid and inexpen-sive technologies are required. Spectroscopic technologies arewidespread in the analysis of main food components. They havealso become popular in the analysis of honeys quality due to theirbenefits (including rapidity, directness, cost-effectiveness and usu-ally no need for sample preparation). Near infrared (NIR) spec-troscopy was successfully used to determine the total contents ofphenolic, flavonoid and antioxidant capacity in honey as alterna-tive to spectrophotometer (Escuredo, Carmen Seijo, Salvador, &Inmaculada González-Martín, 2013; Tahir, Zou, Shen, Shi, &Mariod, 2016). Fourier transform infrared (FT-IR) and Raman spec-troscopy have feasible application in the assessment honey quality,which are used mainly to determine sugar, moisture and acidity(Anjos, Campos, Ruiz, & Antunes, 2015; Zhu et al., 2010; Özbalci,Boyaci, Topcu, Kadılar, & Tamer, 2013), and to discriminate honeybotanical origin (Corvucci, Nobili, Melucci, & Grillenzoni, 2015;

H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211 203

Jandric et al., 2015; Pierna, Abbas, Dardenne, & Baeten, 2011).Recently, FT-IR and Raman have been applied effectively to esti-mate the antioxidant capacity of various foods (Ferreira et al.,2013; Orzel, Stanimirova, Czarnik-Matusewicz, & Daszykowski,2015; Wu et al., 2016). FT-IR principally presents information fromfrequencies of fundamental molecular vibrations and mostlyreveals sharp peaks and different spectral characteristics(Schönbichler et al., 2014). Compared to FTIR, Raman depends onscattering, which is termed as an inelastic scattering of the incidentlight from a sample and the frequency shift by the energy of itscharacteristic molecular vibrations (Wu et al., 2016). The differentenergy during scattering is assigned to the interactions betweenchemical bonds present in the samples and the photon. Thus, thecombination of these two different techniques can provide com-prehensive information about the chemical compositions of honey.To the best of our knowledge, FT-IR and Raman have not beenapplied in the prediction of antioxidant activity and phenolic com-pounds in honey. Therefore, the objective of this work was toassess the feasibility of Raman and Fourier transform infrared withattenuated total reflectance (FTIR-ATR) techniques combined withpartial least-squares (PLSR) regression for the prediction of pheno-lic acids, flavonoid and antioxidant activity of Sudanese honeys.

2. Materials and methods

2.1. Honey samples

Honeys were collected from various geographical and floral ori-gin. The samples include six kinds as follow: Acacia nilotica(n = 10), Acacia seyal (n = 20), Ziziphus spina-christi (n = 20),Amaranthus graecizan (n = 10), Eucalyptus spp. (n = 10) and multi-floral (n = 30). The detailed information about the floral and geo-graphical origins of samples was described in our previous work(Tahir et al., 2015).

2.2. Chemicals reagents

All solvents, phenolic compounds used for HPLC analysis wereof HPLC grade and the rests of chemicals were of analytical grade.Catechin and phenolic acids (Chlorogenic, syringic, vanillic,p-hydroxybenzoic, p-coumaric, caffeic, ferulic, gallic and cinnamicacids) and 1,1-diphenyl-2-picrylhydrazy (DPPH) were supplied by(Sigma chemical CO. USA). Ferrozine (2,4,6-Tris (2-pyridyl)-s-triazine) was supplied by (Sigma chemical Co. Switzerland).Ferrous chloride, quercetin and methanol were supplied by (Sino-pharm chemical reagent Co., Ltd China). Milli-Q water purificationsystem (Millipore Corp., Billerica, MA, USA) was used for waterpurification.

2.3. Reference chemical analysis

2.3.1. Total antioxidant activityThe total antioxidant activity (TAC) was quantified using our

previous method (Tahir et al., 2015, 2016). Measurements wereconducted in triplicate and the mean value was stated as mg ofquercetin equivalent antioxidant content (QEAC) per 100 g ofhoney.

2.3.2. Ferrous ion chelating activityFerrous ion chelating capacity was quantified following the

assay reported recently (Xiong et al., 2013). In this experiment,0.1 mL of ferrous chloride (2.0 mmol/L) was mixed 3.6 ml of honeydissolve in water 1:10 respectively. Thereafter, the reaction wasstarted by 0.2 mL Ferrozine (5.0 mmol/L). The reaction mixtures

were kept for 10 min at 25 �C, the absorbance of the mixture wasmeasured at 562 nm against a blank. The capacity of ferrouschelating was computed using the following formula:

Chelatin capacity % ¼ A1� A2A3

� �� 100

where A1 is the absorbance of the sample, A2 is the absorbance ofthe blank and A3 is the absorbance of the control. The control onlycontains FeCl and ferrozine (Kiliç & Yes�iloglu, 2013).

2.3.3. Extraction of phenolic compoundsExtraction of phenolic compounds was conducted following the

reported method (Gašic et al., 2014) with minor modifications.Honey samples (10 g) were dissolved in 40 mL of purified water,adjusted to pH 2 with HCl (0.1%) and sonicated for 15 min at25 �C. Subsequently, the honey solutions were filtered through cot-ton wool to remove the solid particles. An SPE with C18 cartridges(Applied Separations Inc., Allentown, PA, USA) was conditioned bypassing 10 mL of methanol and 10 mL purified water. The filteredhoney solutions were loaded on the SEP cartridges and washedwith 40 mL with acidified purified water (pH 2) to eliminate allsugars and other polar components of honey. The absorbed pheno-lic compounds were eluted with methanol (2 mL), and then theextracts were filtered through a 0.45-lm membrane filter priorto being quantified by HPLC-DAD.

2.3.4. HPLC-DAD analysisA 1000 lg/mL standard solution of phenolic compounds were

prepared in methanol. The working standard solutions wereprepared by diluting the stock solution at 0.1, 2.5, 5, 50, 100 and1000 lg/mL. All the standard solutions were stored in the dark bot-tles at 4 �C. The Shimadzu LC 20A system (Tokyo, Japan) equippedwith degasser (DGU-20A3), a binary pump (LC-20AD), autosampler(SIL-20 AC), autoinjector (SIL-20AC), column oven (CTO-20AC),communication bus module CBM 20A, and DAD detector SPD-M20A was used. The reversed-phase Zorbax SB-C18 column(250 mm 149 � 4.6 mm, 5 lm particle size, Agilent Technologies)was used for separation of phenolic compounds. The mobile phasecontained A (0.1% formic acid) and B (100%methanol). The gradientprogram was as follows: 0.0–5.0 min, 0–10% B; 5.0– 20.0 min, 40%B; 20.0–32.0 min, 45% B; 32.0–45.0 min, 50% B; 45.0–70.0 min, 80B%; 70.0–75, 5%B. The injection volume for all samples was 5 lL, andthe flow rate was 0.2 mL/min. The results were monitored from210 to 500 nm. The concentrations of phenolic compounds in thesamples were quantified from the obtained standard curves. Thephenolic compounds in the samples were identified by comparingtheir retention times with authentic phenolic compound stan-dards. Limits of detection (LOD) and limits of quantification(LOQ) were computed using standard deviations of the responses(SD) and the slopes of the calibration curves (S) based on thefollowing formulas: LOD = 3⁄3 (SD/S) and LOQ = 10 (SD/S). The val-ues of standard deviations and slopes were obtained from thecalibration curves. Table S1 shows the retention time for differentphenolic compounds, their detection limit, limit of quantificationand linear regression equations.

2.4. Spectroscopic analysis

Honey samples were heated at 40 �C and agitated to dissolvethe crystals in order to obtain homogenous samples prior spectro-scopic analysis (Tahir et al., 2016). For FT-IR and Raman Standardapproaches were employed following the methods reported inthe literature (Corvucci et al., 2015; Jandric et al., 2015).

204 H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211

2.4.1. FTIR-ART spectroscopyFor measurement of honey spectra, approximately 300 lL of

each honey was investigated on a Thermo Scientific Nicolet iS50FT-IR Spectrometer (Thermo Fisher, USA) using a diamond singlereflection attenuated total reflectance (ATR) accessory equippedby a zinc selenide crystal. The software OMNIC version 8 was usedfor spectral acquisition. Infrared spectra were obtained in therange of 4000–600 cm�1 with a spectral resolution of 4 cm�1. The

(a)

(b)

(c)

(d)

Fig. 1. Spectra of different honey varieties. FT-IR spectra with transmittance value 500–correction (MSC) and Raman spectra with Raman shift value 3500–50 cm�1; (c), raw sp

averaged spectra were obtained using 32 scans includingsubtraction of a background scan of the clean diamond crystal.The diamond was cleaned between samples using alcohol.

2.5. Raman spectroscopy

Spectra of the honey samples were acquired on a DXR LaserRaman Spectrometer (Thermo Fisher, USA) coupled with excitation

4500 cm�1; (a), raw spectra; (b), processed spectra using multiplicative scatteringectra; (d), processed spectra using Savitzky–Golay smoothing.

H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211 205

laser: 532 nm; the spectrograph allows a resolution of 5 cm�1

using 900 lines/mm. Each honey was investigated, using a laserpower at the sample of 10 mW. Approximately 500 lL of eachhoney was positioned on a glass slide; the visual center was cor-rectly identified through the microscope, and then the spectrumwas collected. A spectrum from each honey sample was collectedfor 5 min, using the continuous extended scan from 50 to3500 cm�1.

The full spectra of honey samples indicates the presence ofsome regions that incorporate noises to the calibration modelscaused by interferences mainly water. Thus, only the fingerprintsregions were investigated: 1500–400 cm�1 for Raman while forFT-IR were 1800–700 and 3000–2800 cm�1. The number of vari-ables for Raman and FT-IR was 1142 and 2699, respectively.

2.6. Multivariate data analysis

The obtained spectra of FT-IR and Raman contain noises in addi-tion to the honey information. Thus, the mathematical preprocess-ing is essential to reduce the interferences and undesirableinformation as well as improve the contribution of chemicalconstituents. In this work, the data was first preprocessed by theSavitzky–Golay algorithm (15 data points, and a second-orderpolynomial function). Afterward, three pretreatments techniquesincluding, standard normal variate (SNV), multiplicative scatteringcorrection (MSC), de-trending (DT) were applied comparatively.The optimum pretreatment technique was obtained based on thelowest root mean square error of cross validation (RMSECV) andhighest correlation coefficients (R) in the calibration set (Rc) andprediction set (Rp) of PLSR. The optimum pretreatment for Ramanwas Savitzky–Golay smoothing (SG) while for FT-IR was MSC withSavitzky–Golay smoothing. Fig. 1 shows the raw and pretreatedspectra of honey samples.

The classification of samples into calibration and prediction setswas conducted by the following our previous method (Tahir et al.,2016). All 100 samples were randomly allocated into two subsets.One of the subsets called calibration set was used to establish amodel, and the second one called the prediction set was used toevaluate the robustness of the model. The classification of the sam-ples into sets was as follows: all samples were divided according totheir corresponding y-value (viz. the reference chemical data).With regard to partition the calibration/prediction spectra, seven-teen spectra of every fifty kinds of honey were selected for theprediction set. Thus, the calibration and prediction sets consist of64 and 36 spectra, respectively. PLSR is commonly applied as sta-tistical method for building linear regression models based onthe variable matrix X (FT-IR or Raman data) and variable matrix

Table 1Descriptive statistics for phenolic acids, flavonoid, ferrous chelating activity and TAC in Su

Mean SD

Calibration set (n = 64)Catechin lg/g 17.53 12.46Chlorogenic acid lg/g 7.20 5.96Syringic acid lg/g 30.11 12.44Vanillic acid lg/g 4.11 2.26Ferrous chelating activity % 71.89 11.69TAC (QEAC mg/100 g) 16.27 2.49Prediction set (n = 36)Catechin lg/g 19.37 14.14Chlorogenic acid lg/g 6.46 5.26Syringic acid lg/g 27.74 12.10Vanillic acid lg/g 3.78 1.64Ferrous chelating activity % 68.59 9.55TAC (QEAC mg/100 g) 15.42 2.75

Sample number, n; quercetin equivalent antioxidant content (QEAC);Standard dCV = [{SD/Mean} � 100].

Y (reference chemical data) by assuring that, all latent variablesare arranged on the basis of their relevance for predicting Y(Wold, Sjöström, & Eriksson, 2001). The latent variables (LVs) ofthe PLSR models were measured by leave-one-out cross-validation by mapping the number of factors against the RMSECV.The optimum number of LVs was determined by the lowest valueof RMSECV. The performance of the PLSR models was assessedusing Rc, Rp, and RMSECV. Generally, an excellent model musthave high correlation coefficients, as well as low RMSECV and rootmean standard error of prediction (RMSEP) (He, Sun, & Wu, 2014;Kamruzzaman, ElMasry, Sun, & Allen, 2013). The spectralpretreatments and PLSR were executed using Matlab Version7.1(Mathworks Inc., Natick, Mass., USA).

3. Results and discussions

3.1. Phenolic compounds and antioxidant activity in honey

Several phenolic compounds have been monitored in Sudanesehoney samples (Table S2). Vanillic acid, chlorogenic acid and syrin-gic acid and catechin were detected in all honey samples analyzed,while Gallic acid was not detected in any samples. Therefore, thepredominant phenolic compounds were selected for developingprediction models. The reference chemical data are stated as themean, standard deviation, and coefficient of variation. The descrip-tive statistics depicted in Table 1, showed a great variation in phe-nolic compounds and antioxidant activities in the calibration andexternal validation sets. This variation was due to the differencesof the geographical and floral sources of honey samples (Tahiret al., 2016). From Table 1 it can be concluded that the Sudanesehoney samples contain a considerable amount of bioactivecompounds with varied concentrations, revealing a significantantioxidant capacity. The range of six different measurements inthe calibration set covered the range in the prediction set and itis clear that their standard deviations were not significantly differ-ent. Thus, the distribution of the chemical data in the calibrationand prediction sets was appropriate for developing models todetermine these parameters in honey.

3.2. Spectra interpretations

FTIR and Raman spectra of different kinds of honey analyzed areshown in Fig. 1. The spectra procured in this work are comparablewith those described in the literatures (Anjos et al., 2015; Piernaet al., 2011; Özbalci et al., 2013). Regarding the TFTIR spectra,as could be seen in Fig. a and b, the peaks in the region

danese honeys.

Min Max CV

1.99 45.35 71.110.98 20.55 82.7914.54 50.90 41.311.88 9.96 54.9149.55 88.78 16.2711.42 20.95 15.60

1.88 45.55 73.010.96 19.78 81.3914.54 50.26 43.641.88 9.35 43.3852.40 87.88 13.9311.38 20.96 17.83

eviation, SD; Maximum, Max; Minimum, Min; Coefficient of variation, CV;

206 H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211

2800–3000 cm�1 have been assigned to the CAH, OAH, and NH3;these stretching are mainly due to carbohydrates, carboxylic acids,free amino acids and phenolics (Anjos et al., 2015; Movasaghi,Rehman, & Rehman, 2008). The bands at 1700–1600 cm�1 wereoriginated to stretching band of carbonyl groups C@O and C@Cstretching, this region was found to be related to phenolic mole-cules (Preserova, Ranc, Milde, Kubistova, & Stavek, 2015). Accord-ing to Gok, Severcan, Goormaghtigh, Kandemir, and Severcan(2015), the vibration in the region 1540–1175 cm�1 is the resultof deformations of OAH, CAO, CAH and C@C. As well, the samespectral region corresponds to flavanol and phenol (Masek,Chrzescijanska, Kosmalska, & Zaborski, 2014; Nickless, Holroyd,Stephens, Gordon, & Wargent, 2014). The vibrations in the regions940–1175 cm�1 are due to C–OH group as well as the stretchesCAC and CAO in the carbohydrate structure and CAO in the phenol(Masek et al., 2014).

In Fig. 1.c and d, the Raman spectra were characterized by theintense bands around 2941 and 2903 cm�1 was assigned to CAHstretching, and the vibrational observed in the low wavenumberregions 200–500 cm�1 has been assigned to CACAC, CACAO,CAO and CAC (Özbalci et al., 2013). The band at 518 cm�1 wasoriginated from strong deformation of CACAC and CACAO. Thestrong band at 630 cm�1 was assigned to ring deformation. Thebands around 705 cm�1 originated to stretching CAO, CACAOand OACAO bending and the bands in region 822–866 cm�1

were assigned to the CAH, CAOAH, and CH2 bending. The vibra-tion around 95 cm�1 was related to CAH and CAOAH (Zhu et al.,2010). The peak at 1075 cm�1 was assigned mainly to bendingvibration of CAH and CAOAH in carbohydrates and slight contri-bution by the vibration of CAN bond in proteins and amino acid.The peak at 1127 cm�1 could be due to the combination ofstretching vibration of CAO bond and vibration of CAN bond ofcarbohydrate and amino acids, respectively. A strong peak at1266 cm�1 was found to be due to the deformation vibration ofCACAH, OACAH, and CAOAH. The absorbance band at1460 cm�1 was originated from the combination of bendingvibration of CH2 and the vibration of COO– group. Besides, thesame spectral region has been attributed to the flavanol andorganic acids (Nickless et al., 2014; Paradkar & Irudayaraj,2002). The small band observed at 1365 cm�1 might be attribu-ted to the symmetric deformation in the plane of CH2 (Piernaet al., 2011).

Table 2Calibration statistics for phenolic compounds content and antioxidant activities determin

Spectroscopic technique Parameters LVs

Raman Catechin lg/g 6Chlorogenic acid lg/g 6Syringic acid lg/g 7Vanillic acid lg/g 6Ferrous chelating activity % 5TAC (QEAC mg/100 g) 5

FT-IR Catechin lg/g 8Chlorogenic acid lg/g 8Syringic acid lg/g 8Vanillic acid lg/g 7Ferrous chelating activity % 8TAC (QEAC mg/100 g) 7

Raman-FT-IR Catechin lg/g 8Chlorogenic acid lg/g 8Syringic acid lg/g 8Vanillic acid lg/g 8Ferrous chelating activity % 8TAC (QEAC mg/100 g) 8

N: number of samples; LVs: latent variables, SD: standard deviation; RMSECV: Standard Ecorrelation coefficients in the calibration set (R2c) and prediction set (R2p); QEAC: quer

3.3. Quantitative analysis

Table 2 displays the results of PLSR analysis using both FT-IRand Raman data for quantitative determinations of catechin,chlorogenic, syringic, vanillic acids, TAC and ferrous chelatingcapacity in honey.

3.3.1. Results of PLSR modelsPLS is of special interest because, it can analyze data with

strongly collinear correlated, noisy, and various X-variables, andalso simultaneously model several response variables, Y, i.e., pro-files of performance (Wold et al., 2001). PLS and large X-variablesshowed good quantitative results in many studies (DurakliVelioglu et al., 2017; Friedel, Patz, & Dietrich, 2013; Georgouli,Martinez Del Rincon, & Koidis, 2017; Tahir et al., 2016; Wu et al.,2016; Özbalci et al., 2013). PLSR models were constructed usingthe spectra set both in full cross-validation and prediction set. PLSRmodels were performed using selected regions for FT-IR andRaman. The calibration models were developed using Raman andFT-IR spectra set of the randomly chosen 64 honey samples asshown in Table 2. The robust PLSR model must have a high corre-lation coefficient, low RMSECV, RMSEP and a small number of LVs,preferably less than ten (Liang, Lu, Hu, & Kitts, 2016). The numberof LVs obtained in different models presented in Table 2 was smal-ler than nine, hence demonstrating that the developed modelswere not over fitted (Abdi, 2010). Good calibration models wereachieved both in FT-IR and Raman regions. In fact, R2c values incross-validation were higher than 0.95 and RMSECV values for allparameters analyzed were relatively low and ranged from 0.40 to1.14 and 0.33–0.86 for Raman and FT-IR models, respectively.The performance of the models of phenolic compounds and antiox-idant activities in honey was confirmed using the remaining 36honey samples. Both techniques revealed excellent capability inpredicting the concentrations of phenolic compounds and antioxi-dant activities in different honey samples, as shown high R2pvalues (>0.92). Although the RMSEP of TAC in both FT-IR andRaman seem to be slightly higher but still remains good.

These findings revealed that the PLSR models successfullycaptured essential FT-IR and Raman spectra features for phenoliccompounds existent in different honey samples. Figs. 2 and 3represent the regression lines of PLSR models for catechin, chloro-genic, syringic, vanillic acids, TAC and ferrous chelating activity

ed using PLSR for Raman and FT-IR Spectroscopy.

Calibration set (N = 64) Prediction set (N = 36)

R2c RMSECV R2

p RMSEP

0.9988 0.44 0.9992 0.410.9950 0.42 0.9962 0.330.9990 0.41 0.9930 1.000.9677 0.40 0.9272 0.460.9904 1.14 0.9922 0.990.9561 0.51 0.9763 0.430.9988 0.46 0.9988 0.400.9944 0.44 0.9940 0.430.9992 0.33 0.9924 1.080.9688 0.40 0.9461 0.450.9934 0.86 0.9791 1.630.9681 0.47 0.9549 0.530.9990 0.44 0.9986 0.450.9954 0.40 0.9942 0.410.9990 0.40 0.9990 0.380.9508 0.40 0.9728 0.430.9910 1.00 0.9920 1.050.9647 0.50 0.9610 0.49

rror of Cross-Validation; RMSEP: root mean square error of prediction; R2: multiplecetin equivalent antioxidant content.

Pred

icte

d va

lue

for c

hlor

ogen

ic a

cid

(µg/

g)

Reference value for catechin (µg/g)

Pred

icte

d va

lue

for c

atec

hin

(µg/

g

Reference value for vanillic acid (µg/g)

Pred

icte

d va

lue

for v

anill

ic a

cid

(g/

g)

Reference value for chlorogenic acid (µg/g)

Pred

icte

d va

lue

for f

erro

us c

hela

ting

activ

ity %

Reference value for ferrous chelating activity %

Pred

icte

d va

lue

for s

yrin

gic

acid

(µg/

g)

Reference value for syringic acid (µg/g)

Reference value for TAC (QEAC /100g)

Pred

icte

d va

lue

forT

AC

( QEA

C /1

00g)

Fig. 2. Correlation plots for the prediction of catechin, chlorogenic acid syringic acid ferrous chelating activity and quercetin equivalent antioxidant content (QEAC) using PLSbased on the FT-IR spectra.

H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211 207

obtained by FT-IR and Raman, respectively. All constructed calibra-tion models in the FT-IR and Raman regions presented excellentresults for the six parameters measured. Although the results offerrous chelating capacity in FT-IR and Raman were moderatelygood, the predicted results of TAC were not good as compared tothose obtained for the other parameters. We expected these resultsfor antioxidant activities since the FT-IR and Raman spectra repre-sent functional group characteristics of specific chemical con-stituents in honey; but, FT-IR and Raman do not reflect possible

interactions and synergistic effects (competitive inhibitions) amidhoney antioxidants. These features, perhaps are related to the freeradical reaction mechanisms and could describe why the results ofthe most phenolic compounds in honey exhibited greater R2p val-ues compared to those R2p of antioxidant activity.

Finally, the findings revealed that the FTIR and Raman spec-troscopy could be used as a simple and direct method to enablethe apiculture and pharmaceutical industries to efficiently assessthe antioxidant activity and health benefits compounds.

Reference value for TAC (QEAC /100g)Reference value for ferrous chelating activity %

Pred

icte

d va

lue

for

ferr

ous c

hela

ting

activ

ity %

Pred

icte

d va

lue

for

TAC

(QEA

C /1

00g)

Reference value for catechin (µg/g)

Pred

icte

d va

lue

for

cate

chin

(g/

g)

Reference value for chlorogenic acid (µg/g)

Pred

icte

d va

lue

for

chlo

roge

nic

acid

(g/

g)

Reference value for vanillic acid (µg/g)

Pred

icte

d va

lue

for

vani

llic

acid

(g/

g)

Pred

icte

d va

lue

for

syri

ngic

aci

d (

g/g)

Reference value for syringic acid (µg/g)

Fig. 3. Correlation plots for the prediction of catechin, chlorogenic acid syringic acid ferrous chelating activity and quercetin equivalent antioxidant content (QEAC) using PLSbased on the Raman spectra.

208 H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211

3.3.2. Results of PLSR models based on data fusionData fusion process seeks to integrate data from the same set of

samples collected from multiple blocks into a single model whichcan lead to enhancement of the accuracy of the predictions morethan single block achieved. Although both Raman and FT-IRshowed good results, however, the results of vanillic acid andTAC were not good enough compared to other parameters mea-sured. Thus, data fusion method has been applied to investigatewhether the combination of spectra can improve the predictioncapacity of these parameters. In this work, PLSR model based on

data fusion was developed. For integrating FT-IR and Raman, thespectra were scaled using simple mean-centering and variancescaling prior to combining spectra to give alike weights to the dataobtained from techniques of different features. Afterward, thespectra of two techniques fused in a single matrix had the numberof columns equal to the number of honey samples and the numberof rows equal to the number of a total number of data (selectedregions) from FT-IR and Raman. Fig. 4 represents the correlationplots of chemical data against predictions for the six parametersobtained from combined spectra. As depicted in Table 2, the

Pred

icte

d va

lue

for

cate

chin

(g/

g)

Reference value for catechin (µg/g) Reference value for chlorogenic acid (µg/g)

Pred

icte

d va

lue

for

chlo

roge

nic

acid

(g/

g)

Pred

icte

d va

lue

for

syri

ngic

aci

d (

g/g)

Reference value for syringic acid (µg/g)

Pred

icte

d va

lue

for

vani

llic

acid

(g/

g)

Reference value for vanillic acid (µg/g)

Pred

icte

d va

lue

for

ferr

ous c

hela

ting

activ

ity %

Reference value for ferrous chelating activity % Reference value for TAC (QEAC /100g)

Pred

icte

d va

lue

for

TAC

( QEA

C /1

00g)

Fig. 4. Correlation plots for the prediction of catechin, chlorogenic acid syringic acid ferrous chelating activity and quercetin equivalent antioxidant content (QEAC) using PLSbased on the combined spectra.

H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211 209

determinations performed using the Raman regions gave slightlybetter results than those using the FT-IR regions, which was inagreement with results reported in the literature (Wu et al.,2016). This might be attributed to the differences of the vibrationalfeatures of functional groups in the two technologies. Comparisonof the Raman results with the fused spectra showed that the R2pvalues were nearly similar, demonstrating that the added FT-IRdata is relatively not used. Among the six parameters analyzedthe fusion of the selected regions slightly improved the predictionof syringic and vanillic acids. Several authors reported that the datafusion did not always enhance individual results (Apetrei et al.,

2012; Cosio, Ballabio, Benedetti, & Gigliotti, 2007; Masnan et al.,2012) and in some cases was influenced negatively (Roussel,Bellon-Maurel, Roger, & Grenier, 2003). These findings revealedthat the data fusion of FT-IR and Raman does not significantlyenhance prediction results and in some parameters had negativeeffect (Table 2).

4. Conclusion

This work represents the first study to examine the capability ofFTIR and Raman spectroscopy combined with chemometrics for

210 H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211

the prediction of TAC, ferrous chelating capacity catechin, syringic,vanillic, and chlorogenic acids in honey. For constructing PLSRmodel, the wavenumber range 1500–400 cm�1 was selected forRaman, while for FT-IR, 1800–700 and 3000–2800 cm�1 were uti-lized. It was noted that the performances of models based on theRaman spectra were slightly better than those based on the FTIRspectra. The prediction results of vanillic acid and TAC for bothtechniques were not good compared to those of other parametersmeasured. Therefore, the combination of the two spectra was usedto examine whether it can improve the prediction results. Wenoted that a combination of two complementary data for the samesample, i.e. FT-IR and Raman spectroscopy did not improve predic-tion models. Regardless of the results of combined data, bothRaman and FT-IR showed satisfactory prediction capacity for allparameters investigated. The results demonstrated that the devel-oped PLS models based on Raman and FTIR were superior to thoseestablished using NIR spectra (Escuredo et al., 2013; Tahir et al.,2016). Since this work is a feasibility study, further study with alarge number of honey varieties from different countries arerequired prior such an approach is accepted by the apicultureand pharmaceutical industries with confidence.

Acknowledgements

We are grateful for the support from the departmentof the National Science and Technology support program(2015BAD17B04, 2015BAD19B03), the National Natural ScienceFoundation of China (Grant No. 61301239), the Natural ScienceFoundation of Jiangsu Province (BK20130505) and China Postdoc-toral Science Foundation (2013M540422, 2014T70483).

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.foodchem.2017.01.024.

References

Abdi, H. (2010). Partial least squares regression and projection on latent structureregression (PLS Regression). Wiley Interdisciplinary Reviews: ComputationalStatistics, 2(1), 97–106.

Anjos, O., Campos, M. G., Ruiz, P. C., & Antunes, P. (2015). Application of FTIR-ATRspectroscopy to the quantification of sugar in honey. Food Chemistry, 169,218–223.

Apetrei, I. M., Rodríguez-Méndez, M. L., Apetrei, C., Nevares, I., del Alamo, M., & deSaja, J. A. (2012). Monitoring of evolution during red wine aging in oak barrelsand alternative method by means of an electronic panel test. Food ResearchInternational, 45(1), 244–249.

Corvucci, F., Nobili, L., Melucci, D., & Grillenzoni, F.-V. (2015). The discrimination ofhoney origin using melissopalynology and Raman spectroscopy techniquescoupled with multivariate analysis. Food Chemistry, 169, 297–304.

Cosio, M. S., Ballabio, D., Benedetti, S., & Gigliotti, C. (2007). Evaluation of differentstorage conditions of extra virgin olive oils with an innovative recognition toolbuilt by means of electronic nose and electronic tongue. Food Chemistry, 101(2),485–491.

da Silva, P. M., Gauche, C., Gonzaga, L. V., Costa, A. C. O., & Fett, R. (2016). Honey:Chemical composition, stability and authenticity. Food Chemistry, 196, 309–323.

Durakli Velioglu, S., Temiz, H. T., Ercioglu, E., Velioglu, H. M., Topcu, A., & Boyaci, I. H.(2017). Use of Raman spectroscopy for determining erucic acid content incanola oil. Food Chemistry, 221, 87–90.

Erejuwa, O. O., Sulaiman, S. A., & Ab Wahab, M. S. (2014). Effects of honey and itsmechanisms of action on the development and progression of cancer.Molecules,19(2), 2497–2522.

Escuredo, O., Carmen Seijo, M., Salvador, J., & Inmaculada González-Martín, M.(2013). Near infrared spectroscopy for prediction of antioxidant compounds inthe honey. Food Chemistry, 141(4), 3409–3414.

Ferreira, B. S., de Almeida, C. G., Le Hyaric, M., de Oliveira, V. E., Edwards, H. G. M., &de Oliveira, L. F. C. (2013). Raman spectroscopic investigation of carotenoids inoils from amazonian products. Spectroscopy Letters, 46(2), 122–127.

Friedel, M., Patz, C.-D., & Dietrich, H. (2013). Comparison of different measurementtechniques and variable selection methods for FT-MIR in wine analysis. FoodChemistry, 141(4), 4200–4207.

Gašic, U., Keckeš, S., Dabic, D., Trifkovic, J., Milojkovic-Opsenica, D., Natic, M., et al.(2014). Phenolic profile and antioxidant activity of Serbian polyfloral honeys.Food Chemistry, 145, 599–607.

Georgouli, K., Martinez Del Rincon, J., & Koidis, A. (2017). Continuous statisticalmodelling for rapid detection of adulteration of extra virgin olive oil using midinfrared and Raman spectroscopic data. Food Chemistry, 217, 735–742.

Gok, S., Severcan, M., Goormaghtigh, E., Kandemir, I., & Severcan, F. (2015).Differentiation of Anatolian honey samples from different botanical origins byATR-FTIR spectroscopy using multivariate analysis. Food Chemistry, 170,234–240.

He, H.-J., Sun, D.-W., & Wu, D. (2014). Rapid and real-time prediction of lactic acidbacteria (LAB) in farmed salmon flesh using near-infrared (NIR) hyperspectralimaging combined with chemometric analysis. Food Research International, 62,476–483.

Jandric, Z., Haughey, S. A., Frew, R. D., McComb, K., Galvin-King, P., Elliott, C. T., et al.(2015). Discrimination of honey of different floral origins by a combination ofvarious chemical parameters. Food Chemistry, 189, 52–59.

Kamruzzaman, M., ElMasry, G., Sun, D.-W., & Allen, P. (2013). Non-destructiveassessment of instrumental and sensory tenderness of lamb meat using NIRhyperspectral imaging. Food Chemistry, 141(1), 389–396.

Kiliç, I., & Yes�iloglu, Y. (2013). Spectroscopic studies on the antioxidant activity of p-coumaric acid. Spectrochimica Acta Part A: Molecular and BiomolecularSpectroscopy, 115, 719–724.

Kralj Cigic, I., & Prosen, H. (2009). An overview of conventional and emerginganalytical methods for the determination of mycotoxins. International Journal ofMolecular Sciences, 10(1), 62–115.

Lachman, J., Orsák, M., Hejtmánková, A., & Kovárová, E. (2010). Evaluation ofantioxidant activity and total phenolics of selected Czech honeys. LWT – FoodScience and Technology, 43(1), 52–58.

Lianda, R. L., Sant’Ana, L. D. O., Echevarria, A., & Castro, R. N. (2012). Antioxidantactivity and phenolic composition of Brazilian honeys and their extracts. Journalof the Brazilian Chemical Society, 23(4), 618–627.

Liang, N., Lu, X., Hu, Y., & Kitts, D. D. (2016). Application of Attenuated TotalReflectance-Fourier Transformed Infrared (ATR-FTIR) spectroscopy todetermine the chlorogenic acid isomer profile and antioxidant capacity ofcoffee beans. Journal of Agricultural and Food Chemistry, 64(3), 681–689.

Masek, A., Chrzescijanska, E., Kosmalska, A., & Zaborski, M. (2014). Characteristics ofcompounds in hops using cyclic voltammetry, UV–VIS, FTIR and GC–MSanalysis. Food Chemistry, 156, 353–361.

Masnan, M. J., Mahat, N. I., Zakaria, A., Shakaff, A. Y. M., Adom, A. H., & Sa’ad, F. S. A.(2012). Enhancing classification performance of multisensory data throughextraction and selection of features. Procedia Chemistry, 6, 132–140.

Movasaghi, Z., Rehman, S., & Rehman, I. U. (2008). Fourier transform infrared (FTIR)spectroscopy of biological tissues. Applied Spectroscopy Reviews, 43(2),134–179.

Nickless, E. M., Holroyd, S. E., Stephens, J. M., Gordon, K. C., & Wargent, J. J. (2014).Analytical FT-Raman spectroscopy to chemotype Leptospermum scoparium andgenerate predictive models for screening for dihydroxyacetone levels in floralnectar. Journal of Raman Spectroscopy, 45(10), 890–894.

Orzel, J., Stanimirova, I., Czarnik-Matusewicz, B., & Daszykowski, M. (2015).Prediction of the hydrophilic antioxidant capacity of tomato pastes from theIR and fluorescence excitation–emission spectra of extracts and intact samples.Talanta, 138, 64–70.

Özbalci, B., Boyaci, _I. H., Topcu, A., Kadılar, C., & Tamer, U. (2013). Rapid analysis ofsugars in honey by processing Raman spectrum using chemometric methodsand artificial neural networks. Food Chemistry, 136(3–4), 1444–1452.

Paradkar, M., & Irudayaraj, J. (2002). Discrimination and classification of beet andcane inverts in honey by FT-Raman spectroscopy. Food Chemistry, 76(2),231–239.

Pierna, J. A. F., Abbas, O., Dardenne, P., & Baeten, V. (2011). Discrimination ofCorsican honey by FT-Raman spectroscopy and chemometrics. BiotechnologieAgronomie Societe Et Environnement, 15(1), 75–84.

Preserova, J., Ranc, V., Milde, D., Kubistova, V., & Stavek, J. (2015). Study of phenolicprofile and antioxidant activity in selected Moravian wines during winemakingprocess by FT-IR spectroscopy. Journal of Food Science and Technology, 1–10.

Roussel, S., Bellon-Maurel, V., Roger, J.-M., & Grenier, P. (2003). Authenticatingwhite grape must variety with classification models based on aroma sensors,FT-IR and UV spectrometry. Journal of Food Engineering, 60(4), 407–419.

Schönbichler, S., Falser, G., Hussain, S., Bittner, L., Abel, G., Popp, M., et al. (2014).Comparison of NIR and ATR-IR spectroscopy for the determination of theantioxidant capacity of Primulae flos cum calycibus. Analytical Methods, 6(16),6343–6351.

Tahir, H. E., Xiaobo, Z., Zhihua, L., & Yaodi, Z. (2015). Comprehensive evaluation ofantioxidant properties and volatile compounds of Sudanese honeys. Journal ofFood Biochemistry, 39(4), 349–359.

Tahir, H. E., Zou, X., Shen, T., Shi, J., & Mariod, A. A. (2016). Near-Infrared (NIR)spectroscopy for rapid measurement of antioxidant properties anddiscrimination of Sudanese honeys from different botanical origin. FoodAnalytical Methods, 9(9), 2631–2641.

Tomasin, R., de Andrade, R. S., & Gomes-Marcondes, M. C. C. (2015). OralAdministration of Aloe vera (L.) Burm. f. (Xanthorrhoeaceae) and Honeyimproves the host body composition and modulates proteolysis through

H.E. Tahir et al. / Food Chemistry 226 (2017) 202–211 211

reduction of tumor progression and oxidative stress in rats. Journal of MedicinalFood.

Wold, S., Sjöström, M., & Eriksson, L. (2001). PLS-regression: A basic tool ofchemometrics. Chemometrics and Intelligent Laboratory Systems, 58(2), 109–130.

Wu, Z., Xu, E., Long, J., Pan, X., Xu, X., Jin, Z., & Jiao, A. (2016). Comparison betweenATR-IR, Raman, concatenated ATR-IR and Raman spectroscopy for thedetermination of total antioxidant capacity and total phenolic content ofChinese rice wine. Food Chemistry, 194, 671–679.

Xiong, X., Li, M., Xie, J., Jin, Q., Xue, B., & Sun, T. (2013). Antioxidant activity ofxanthan oligosaccharides prepared by different degradation methods.Carbohydrate Polymers, 92(2), 1166–1171.

Zhu, X., Li, S., Shan, Y., Zhang, Z., Li, G., Su, D., et al. (2010). Detection of adulterantssuch as sweeteners materials in honey using near-infrared spectroscopy andchemometrics. Journal of Food Engineering, 101(1), 92–97.

本文献由“学霸图书馆-文献云下载”收集自网络，仅供学习交流使用。

学霸图书馆（www.xuebalib.com）是一个“整合众多图书馆数据库资源，

提供一站式文献检索和下载服务”的24 小时在线不限IP

图书馆。

图书馆致力于便利、促进学习与科研，提供最强文献下载服务。

图书馆导航：

图书馆首页 文献云下载 图书馆入口 外文数据库大全 疑难文献辅助工具