Journal of Food Composition and Analysis 66 (2018) 46–54

Contents lists available at ScienceDirect

Journal of Food Composition and Analysis

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

Original research article

Accurate evaluation of sugar contents in stingless bee (Heterotrigona itama) T

honey using a swift scheme
⁎ ⁎
Kuan Wei Sea, Raja Kamarulzaman Raja Ibrahimb, , Roswanira Abdul Wahabc, ,
Sib Krishna Ghoshald
a
School of Graduate Studies, SPS, Universiti Teknologi Malaysia, 81310 UTM, Johor Bahru, Johor, Malaysia
b
Laser Centre, Department of Physics, Faculty of Science, Universiti Teknologi Malaysia, 81310 UTM, Johor Bahru, Johor, Malaysia
c
Department of Chemistry, Faculty of Science, Universiti Teknologi Malaysia, 81310 UTM, Johor Bahru, Johor, Malaysia
d
Department of Physics, AMORG, Faculty of Science, Universiti Teknologi Malaysia, 81310 UTM, Johor Bahru, Johor, Malaysia

A R T I C L E I N F O A B S T R A C T

Chemical compounds studied in this article: We propose a rapid scheme for the precise determination of the sugars content in 62 stingless bee (Heterotrigona
Fructose (PubChem CID: 5984) itama) honeys harvested at diverse locations in Malaysia. This scheme combined Fourier transform infrared and
Glucose (PubChem CID: 79025) attenuated total reflectance (FTIR-ATR) techniques with chemometric partial least square (PLS) regression
Sucrose (PubChem CID: 5988) analysis, wherein high performance liquid chromatography (HPLC) is used as reference technique. Results re-
Maltose (PubChem CID: 23615261)
vealed that such honeys contained (per 100 g) high amount of maltose (9.74–54.3 g) and low quantity of fructose
Acetonitrile (PubChem CID: 6342)
(5.37–19.9 g) as well as glucose (3.91–27.2 g). The first derivative data transformation in the wavenumber
Keywords: within 1500 to 750 cm−1 produced the optimum PLS prediction performance with corresponding R2 value over
Food analysis
0.976 and very low standard error of prediction (per 100 g) for fructose (0.562 g), glucose (0.855 g) and maltose
Food composition
(1.55 g). The results proved that the proposed scheme is potentially useful for fast and accurate quantification of
Stingless bees (Heterotrigona itama) honey
Sugar contents sugars content in H. itama honeys.
FTIR-ATR
HPLC
Chemometrics
PLS regression analysis

1. Introduction noteworthy to mention that the market price of H. itama honey is about
USD 100 per kg (Shadan et al., 2017; Razak et al., 2016). Such high
Honey is a type of natural sweetener produced by honey bees, col- price is due to elevated contents of flavonoids and polyphenols present
lected from the nectar of flowers. Invertase, an enzyme produced in the in stingless bee honeys as compared to that produced by the Apis spp.
hypopharyngeal gland of honey bees hydrolyzes the disaccharides (Biluca et al., 2016; Rodriguez-Malaver, 2013; Rodriguez-Malaver
present in the nectar into monosaccharides (Almeida-Muradian, 2013), et al., 2009).
which subsequently form the dominant sugar composition in honey. The Malaysian bee farmers prefer to foster stingless bee (H. itama)
Apart from sugar and the enzyme, honey also contains minute amount species because such populations are less vulnerable to seasonal changes
of organic acids, proteins and minerals. This composition tends to vary and capable to survive in harsh environments (Kelly et al., 2014). These
according to the bee species, floral source and geographical region are the possible reasons for the H. itama being the most domesticated
(Karabagias et al., 2014). Honey produced by stingless bees called species of stingless bee in Malaysia (Razak et al., 2016). Current esti-
Heterotrigona itama (H. itama) is highly preferred by consumers due to mation showed that the stingless bee honey sold in the Malaysian open
its perceived therapeutic and nutritional benefits. Such honey is be- market is largely produced by H. itama (Kelly et al., 2014). Despite the
lieved to have curative properties for throat inflammation, gastritis, popularity of H. itama honey among consumers, studies relating the
cataract as well as assisting in post-birth recovery (Vit et al., 2004). The sugar composition of this premium food remained relatively limited.
plethora of health benefits coupled with limited production of this food Hence, we aim to develop a rapid and accurate technique to identify and
commodity have inflated the economic value of H. itama honey as quantify the individual sugar components in H. itama honey collected
compared to honey produced by honey bees called Apis mellifera. It is from different geographic locations in Malaysia.

⁎
Corresponding authors.
E-mail addresses: [email protected] (R.K.R. Ibrahim), [email protected], [email protected] (R.A. Wahab).

https://doi.org/10.1016/j.jfca.2017.12.002
Received 2 July 2017; Received in revised form 17 November 2017; Accepted 4 December 2017
Available online 08 December 2017
0889-1575/ © 2017 Elsevier Inc. All rights reserved.
K.W. Se et al. Journal of Food Composition and Analysis 66 (2018) 46–54

Fig. 1. Conceptual view of the proposed scheme for rapid

evaluation of sugar content in H. itama honey.
FTIR-ATR (“Fourier transform infrared together with atte-
nuated total reflectance spectroscopy”); HPLC-RID (“high
performance liquid chromatography combined with re-
fractive index detector”); PLS (“partial least square”); Xcal
(“X-variables in the calibration dataset”); Xval (“X-variables in
the validation dataset”); Ypredicted (predicted Y-variables);
Ymeasured (“measured Y-variables”), R2 (“coefficient of de-
termination”); SEC (“standard error of calibration”), SECV
(“standard error of cross validation”); RPD (“residual pre-
dictive deviation”); SEP (“standard error of prediction”).

Over the years, different analytical techniques such as high perfor- honey, the established HPLC method is used as the reference technique
mance liquid chromatography (HPLC) and high performance anion- to explicitly profile the main sugar components and validate the data
exchange chromatography integrated with pulsed amperometric de- obtained using the swift scheme. Previously, reference methods such as
tection (HPAEC-PAD) have been used to qualitatively and quantita- HPLC and HPAEC-PAD have been used to authenticate the sugars
tively evaluate various compound in honey including sugars, fluor- content predicted using PLS-integrated FTIR spectroscopy. Many stu-
oquinolones antibiotic, etc. (Kek et al., 2017; Tayeb-Cherif et al., 2016; dies used the PLS regression models constructed from raw FTIR spectra
Cherif et al., 2015; Anjos et al., 2015). Despite the proven effectiveness (Ruoff et al., 2006; Tewari and Irudayaraj, 2004), first derivative (Anjos
and accuracy of HPLC and HPAEC-PAD to analyze sugar content in et al., 2015) and second derivative (Huang et al., 2016; Wang et al.,
honey, the overall protocol remains laborious and time-consuming, as 2010) as the main data preprocessing method for sugar contents eva-
well as requires costly reagents and cumbersome disposal practices. To luation. However, a comparative study between the spectral data pre-
overcome such limitations, a fast, portable, inexpensive, and precise processing (raw, first or second derivative) and the predicted accuracy
method for assessing the sugars content in honey is demanded. Herein, of the PLS model is not yet performed. Therefore, the objective of this
we propose a swift scheme which combines FTIR-ATR spectrometry study is to employ PLS integrated FTIR to quantitatively assess the
with chemometric PLS regression analysis to identify and quantify the sugar content in H. itama honey. Performance of the corresponding PLS
individual sugar in H. itama honey. The term ‘swift’ used in this study regression models for predicting the main sugar components is com-
signifies the rapid and facile nature of the proposed scheme to analyze pared and evaluated in terms of the raw data, pre-treated first and
the sugar composition in a honey sample. Typically, each test requires second derivative data as well.
less than 5 min for spectral acquisition and sugar prediction. Con-
ceptual view of the proposed scheme is illustrated in Fig. 1. The well 2. Material and methods
celebrated chemometrics regression analysis (PLS regression) is used to
correlate the FTIR-ATR spectral data for precise determination of var- 2.1. Honey samples
ious types of sugars content in H. itama honey. The applicability of this
method has been clearly demonstrated for predicting diverse minerals A total of 62 samples of pure H. itama honey are collected from
(Escuredo et al., 2015), methylglyoxal and antibacterial activity reputable bee farms located in five different states of Malaysia (43 from
(Sultanbawa et al., 2015) as well as several physicochemical parameters Johor, 14 from Terengganu, 3 from Kedah, 1 from Pahang and 1 from
such as pH, free acidity, electric conductivity, ash, moisture and hy- Kelantan). School of Food Science and Technology (Universiti Malaysia
droxymethylfurfural (HMF) content (Anguebes et al., 2016) present in Terengganu, UMT) and Johor Entrepreneur of Stingless Bee Society are
honeys. the main provider of these samples. For reliable outcome with im-
To minimize the uncertainties within the proposed scheme and to proved statistics, random sampling is performed wherein several
simultaneously detect as well as quantify sugar content in H. itama honeys are collected from different bee hives at the same farm

47
K.W. Se et al. Journal of Food Composition and Analysis 66 (2018) 46–54

considering their color variability. Supplementary Table S1 enlists the named factor, and the concentration of sugars present in the honey (Y-
details of collected honeys in terms of their geographical location, label, variables) is predicted.
and year of harvesting. The honey samples are stored in darkness at Generally, PLS regression model is constructed via three steps such
room temperature (26 ± 2 °C) prior to the following analytical mea- as building of the calibration model, optimizing the model using the
surements. cross-validation and authenticating the test dataset. The entire datasets
are separated into calibration dataset and validation dataset (test da-
2.2. High performance liquid chromatography (HPLC) taset). The calibration dataset is used for PLS calibration and the vali-
dation dataset is utilized for the prediction which in turn checked the
Sugars content in all honey samples are quantified using HPLC validity of the calibration model (Gallardo-Velazquez et al., 2009). The
equipped with refractive index detector (RID) (Agilent 1100, Santa performance of PLS calibration is evaluated by calculating the standard
Clara, California, United States) following the AOAC Official Method error of calibration (SEC) using the following equation:
977.20. A 5% (w/v) solution of honey dissolved in ultrapure water
N
(Milli Q system, Merck Millipore, Burlington, Massachusetts, United ⎛∑ (Ai − Ai* )2 ⎞
i=1
States) is filtered through 0.45 μm PTFE filter and injected to the HPLC SEC = ⎜ ⎟
⎜ N−f−1 ⎟
system. Zorbax Carbohydrate Column (Agilent, Santa Clara, California, ⎝ ⎠ (1)
United States) having 5 μm particle size, 250 mm in length and 4.6 mm
where N is the sample size, f is the number of factor in the calibration,
inner diameter is used as the stationary phase. An isocratic elution
Ai and Ai* are the known and calculated specimen concentrations, re-
mobile phase consisting of HPLC grade (purity ∼ 99.9%) acetonitrile
spectively.
(Qrec, Selangor, Malaysia) and water (75:25 v/v) is used. Sample vo-
The leave-one-out cross validation (LOOCV) is used to identify the
lume of 40 μL is injected with a flow rate of 1.4 mL/min with the
number of optimized latent variables (factor) based on the minimum
column temperature maintained at 30 °C. Run time per sample is set at
value of predicted residual error sum of squares (PRESS). This step is
25 min synchronized with standard analytical grade (purity > 99.0%)
necessary as to avoid over fitting of a calibration model. In the process
glucose (QRec) and sucrose (Qrec) as well as HPLC grade
of LOOCV, the first spectrum from the calibration dataset is removed for
(purity ∼ 99.9%) fructose (Systerm, Selangor, Malaysia) and maltose
prediction and the remaining spectra are used to construct the cali-
(Sigma-Aldrich, St. Louis, Missouri, United States). The sugars content
bration model. Consequently, the second spectrum is omitted for pre-
in the honey samples are identified by comparing them with the re-
diction and the calibration is built using the remaining spectra. Similar
tention time of standard sugars (fructose, glucose, sucrose and maltose).
procedure is carried out until all the sample spectra have been con-
Quantitative analyses are performed by preparing standard solutions of
sidered as unknown for prediction. Standard error of cross validation
fructose, glucose, sucrose and maltose at different concentrations.
(SECV) is a type of mean-square error which defines the variability of
Subsequently, the calibration curves are constructed using the re-
the bias (accuracy) on the PLS calibration model. The calibration per-
spective peak area of analytes. Each measurement is repeated three
formance of PLS regression is also statistically evaluated through re-
times to calculate the mean and standard deviation. Supplementary Fig.
sidual predictive deviation (RPD). The RPD value signified the ratio of
S1 shows the HPLC calibration curves for fructose, glucose, sucrose and
standard deviation (SD) of population in the calibration dataset to the
maltose.
SECV. RPD value is calculated using the relation:
2.3. Fourier transform infrared with attenuated total reflectance (FTIR- SD
RPD =
ATR) spectroscopy SECV (2)

RPD estimates the quality of the developed calibration model to

Room temperature FTIR spectra in the wavenumber range of
predict the chemical data (Naes et al., 2002; Williams, 2001). A high
4000–650 cm−1 are recorded using NIR/MIR Frontier spectrometer
RPD value indicates a greater probability to accurately predict the
(Perkin Elmer, Waltham, Massachusetts, United States) which is
analyte of interest outside the calibration dataset (Cozzolino et al.,
equipped with deuterated triglycine sulphate (DTGS) detector and in-
2014; Naes et al., 2002; Williams, 2001). A value of RPD above 3 is
terfaced to computer software (Spectrum, v10.4.0, Perkin Elmer).
considered as moderate and suitable for screening purposes. Con-
Attenuated total reflectance (ATR) sampling accessory used a diamond
versely, RPD value over 5 infers the adequacy of a model to be used for
and zinc selenide (ZnSe) crystal composite for spectral acquisition.
accurate prediction (Huang et al., 2016; Cozzolino et al., 2014; Naes
Triplicate spectra for each sample are acquired with a total accumula-
et al., 2002; Williams, 2001).
tion of 16 scans at spectral resolution of 2 cm−1. Each spectrum is
Using the validation dataset (not included in the calibration da-
standardized against the background spectrum of air and corrected for
taset), the prediction capability of the calibration model is evaluated
baseline prior to the further analysis. The ATR crystal is gently washed
based on the standard error of prediction (SEP) via:
by distilled water and dried with acetone wash after every measure-
ment. Often, the background spectrum is inspected to ensure the N
⎛∑ (Ai − Ai* − Bias )2 ⎞
complete absence of any residue from the previous analysis. SEP = ⎜
i=1
⎟
⎜ M−1 ⎟
⎝ ⎠ (3)
2.4. Chemometric analysis of FTIR-ATR spectra
where Ai* is the sample concentration computed by the calibration
The Unscrambler X software (v10.3, CAMO, Oslo, Norway) is used equation and M is the sample size in the validation dataset. SEP signifies
for the subsequent chemometric data analysis. Partial least square (PLS) the magnitude of error when independent samples (validation dataset)
regression is constructed based on FTIR-ATR spectral data for predic- are predicted using the constructed PLS calibration model (Santos et al.,
tion of sugars content in the studied honeys. The PLS regression algo- 2013). It is important to note that the equation for computing SECV is
rithm is based on the mathematical correlation between 2 independent same as that of SEP (Mark and Workman, 2003). The only difference is
data matrices (Li et al., 2017). PLS regression model correlates the the sample size that is used for cross validation and prediction. For this
FTIR-ATR spectral data (called X variables) to the concentration of purpose, twenty five randomly selected samples are considered as va-
sugars in honey acquired from the reference (HPLC) method (specified lidation dataset and the remaining 37 samples are treated as calibration
as Y-variables). The X-variables represent data matrix originated from dataset. Following simple data preprocessing method (Savitzky-Golay
IR spectral data with rows for samples and columns for wavenumber. method with second polynomial order), the first derivative (dy/dx) and
The X-variables are projected into small number of latent variables second derivative (d2y/dx2) of the data is utilized. In this method, five

48
K.W. Se et al. Journal of Food Composition and Analysis 66 (2018) 46–54

Fig. 2. Bar diagram showing various sugar contents in the studied honeys obtained from the Malaysian state of (a) Johor and (b) others. Dotted line represents the total fructose and
glucose content established by Codex Alimentarius Standard for Honey (2001).

smoothing data points are applied to the FTIR-ATR spectral data to mentioned standard (60 g/100 g). Recently, Kek et al. (2017) reported
compare their PLS prediction performances. that the average value of fructose and glucose in three Malaysian
stingless bees’ honey (H. itama) is as low as 15.8 g/100 g and 9.22 g/
3. Results and discussion 100 g, respectively. Another study on eleven different stingless bees
(Tetragonula laeviceps-pagdeni) honey species from Thailand also re-
3.1. HPLC analysis of sugar content vealed the presence of low level of fructose (17.0 g/100 g) and glucose
(14.0 g/100 g) (Chuttong et al., 2016). Moreover, the fructose and
As aforementioned, HPLC is used as the reference method for qua- glucose contents in Brazilian stingless bee honey meet the Codex
litative and quantitative determination of sugar content in H. itama Standard (de Sousa et al., 2016; Biluca et al., 2016). Later, Kek et al.
honey samples. Supplementary Table S2 enlists the concentrations of (2017) demonstrated that the sucrose concentrations in Malaysian H.
fructose, glucose, sucrose and maltose in the 62 studied honeys. itama honey are quite high (32.3 g/100 g). On the contrary, out of our
According to the results, the H. itama honey showed an average fructose 62 honeys, only five of them displayed the presence of sucrose with an
concentration of 13.4 ± 3.83 g/100 g, with concentration ranging average of 3.46 ± 2.49 g/100 g (ranging from 1.01 ± 0.35 to
from 5.37 ± 0.01 to 19.9 ± 0.35 g/100 g. Glucose concentration is 6.50 ± 0.14 g/100 g), which is well below the permitted limit set out
slightly higher with an average of 16.4 ± 5.86 g/100 g, corresponding by the Codex Alimentarius (2001).
to concentration between 3.91 ± 0.09 to 27.2 ± 0.19 g/100 g. While Fig. 2(a–b) shows the measured sugars content in all 62 studied
there is yet a standard available for stingless bee honey, the Codex stingless bee honeys. Cumulative sugar (fructose + glucose + su-
Alimentarius (2001) states that the sum of fructose and glucose content crose + maltose) content in nearly all samples exceeded the limit set by
should be higher than 60 g/100 g for conventional sting bee honey. The Codex Alimentarius (2001). It is asserted that this variation may set a
cumulative fructose and glucose content in the studied honey are new benchmark for monitoring the quality of stingless bee honeys. This
clearly well below (29.8 g/100 g) the level set forth by the above- departure is because all the earlier studies on sugars profiling in

49
K.W. Se et al. Journal of Food Composition and Analysis 66 (2018) 46–54

Fig. 3. FTIR-ATR spectra of 62 studied stingless bee honeys.

stingless bee honey overlooked the maltose content (Kek et al., 2017; of anomeric carbon (950–750 cm−1) which is often preferred for car-
Biluca et al., 2016; Suntiparapop et al., 2012; Guerrini et al., 2009). bohydrates analysis (Gok et al., 2015). Therefore, this carbohydrate
Interestingly, detected amount of maltose in all the studied H. itama dominated region (1500–750 cm−1) is further utilized to construct the
honey is widely varied and ranged from 9.74 ± 0.46 to model for sugars concentration prediction using partial least square
54.3 ± 2.66 g/100 g with an average of 33.8 ± 10.6 g/100 g. Con- (PLS) regression.
versely, some studies on the sugars content in Brazilian stingless bee
(Melipona subnitida Duke and Melipona scutellaris Latrelle) honeys 3.3. Sugars prediction via chemometric-integrated FTIR spectroscopy
showed the complete absence of maltose (de Sousa et al., 2016). The
observation on distinctively high level of maltose in the studied honey Before constructing the PLS calibration model, correlation among
is comparable to the one found in Thai stingless bee honey species various samples are established using principal component analysis
(Lepidotrigona terminata, L. flavibasis, and Tetragonula laeviceps-pagdeni (PCA) based on the FTIR-ATR spectra in the region of 1500–750 cm−1.
complex) (Chuttong et al., 2016). This wide variation among honey This is prerequisite because all the 62 honeys are collected from diverse
samples from different species of stingless bee is typically influenced by hives and localities of Malaysia. Fig. 4(a–c) depicts the PCA scores plot
the processes carried out by complex salivary secretions, the nectar of all studied H. itama honey from raw, first and second derivative
collected from flowers and regional climatic conditions, as described by processed FTIR-ATR data in the wavenumber within 1500–750 cm−1.
Jalil et al. (2017) and Anjos et al. (2015). Principal component 1 (PC1) and component 2 (PC2) revealed an ac-
cumulation with 97%, 88% as well as 73% data variability for raw, first
3.2. FTIR-ATR spectral analysis and second derivative data, respectively. The PCA scores plots depicted
the absence of any data separation or clustering among the various
Fig. 3 illustrates the FTIR-ATR spectra of the studied honeys in the samples. This confirmed the samples’ solitary class and thus chosen to
wavenumber region of 4000–650 cm−1. Two dominant absorption build the PLS calibration model. The model is constructed based on the
bands are observed at 3270 cm−1 and 1642 cm−1, which are allocated FTIR-ATR spectra with reference to HPLC outcome. The PLS regression
to the OeH stretching and OeH deformation of water, respectively model predicted the contents of fructose, glucose and maltose only. The
(Anjos et al., 2015). The emergence of a band around 2938 cm−1 is inadequacy of detected sucrose (present only in 5 out of 62 of honey
assigned to the CeH stretching of carboxylic acid and the NH3 samples) made it unsuitable for chemometric data analysis.
stretching of free amino acids (Gallardo-Velazquez et al., 2009). The Fig. 5(a–c) displays the plot of PLS regression coefficient against the
observed absorption bands originated from the presence of acid (at low spectral region of interest (1500–750 cm−1) for fructose, glucose and
concentrations) in the honey samples (Tewari and Irudayaraj, 2004; maltose. The high positive and negative values of PLS regression coef-
Sivakesava and Irudayaraj, 2001) is excluded from the chemometric ficient signify a strong correlation with the actual concentration of the
model construction. The weak absorption band around 1740 cm−1 is investigated sugar. Conversely, PLS regression coefficient equivalent to
approved to the C]O stretching that aroused mainly from the presence zero indicates the lowest correlation. Fig. 5 reveals the difference in the
of carbohydrates and tiny amount of protein (NeH bending of amide I) contribution levels of wavenumbers in the range 1500–750 cm−1 cor-
in the honeys (Gok et al., 2015). respond to actual concentration of fructose, glucose and maltose pre-
The spectrum for carbohydrates typically shows multiple absorption sent in the studied samples. This finding again verified that the wave-
bands within the region of 1500–750 cm−1 (Wang et al., 2010; number within 1500–750 cm−1 is comprised of the most important
Gallardo-Velazquez et al., 2009; Tewari and Irudayaraj, 2004) in which vibrational mode of sugars and thereby affirmed the usability of this
spectroscopic variability is reportedly the highest in honey (Fig. 3). This spectral region for sugars prediction.
region encapsulated the deformation of eCH2 and angular deformation Table 1 summarizes the PLS calibration parameters such as optimal
of CeCeH and HeCeO linkages at band 1500–1200 cm−1 (Hineno, number of factor, coefficient of determination (R2), SEC, SECV, RPD,
1977). The significant stretching vibrational modes of carbohydrates and prediction parameters including SEP based on FTIR-ATR data (raw
(CeO and CeC) are appeared around 1200–950 cm−1 (Pataca et al., as well as processed first derivative and second derivative) in the region
2007). This region also allows the assessment of the fingerprints region of 1500–750 cm−1. The R2 value denotes the correlation between the

50
K.W. Se et al. Journal of Food Composition and Analysis 66 (2018) 46–54

Fig. 4. Principal component analysis scores plot of

62 studied honeys from (a) raw data, (b) 1st deri-
vative and (c) 2nd derivative pre-processed FTIR-
ATR data in the wavenumber range of
1500–750 cm−1.

actual and predicted values with R2 ≈ 1 implies perfect correlation. validation dataset, the constructed PLS calibration model is first opti-
The PLS calibration model produced high R2 (Table 1) for fructose mized using LOOCV. Furthermore, the calibration performance is as-
(0.969–0.979), glucose (0.923–0.976), and maltose (0.958–0.980) sessed using the magnitude of SECV and RPD. The calculated RPD va-
corresponding to raw, first derivative and second derivative data, re- lues are discerned to be all above 3, which indicated the good quality of
spectively. The computed SEC of all PLS calibration model are in the the constructed calibration model for predicting the sugars content in
range of 0.557–2.19 which is with the acceptable level of 3 (Gallardo- stingless bee honeys. Meanwhile, the first derivative of FTIR-ATR data
Velazquez et al., 2009). Before predicting the sugars content using provided the highest RPD values for fructose (5.48), glucose (4.73) and

51
K.W. Se et al. Journal of Food Composition and Analysis 66 (2018) 46–54

(1.55 g/100 g) obtained from the first derivative data. This observation
reaffirmed that the first derivative transformation of FTIR-ATR spectral
data could yield the best PLS prediction performance. Conversely, it is
revealed that the second derivative transformation of FTIR-ATR data is
inappropriate towards predicting the maltose content (corresponding
SEP value above 3) in terms of calibration and assessments.
Table 2 depicts a comparative evaluation of the present scheme with
existing art of the techniques integrated with PLS regression analysis.
The prediction accuracy of PLS is compared in terms of the calibration
parameter, RPD values and prediction parameters including R2 and SEP
(or root mean square error of prediction, RMSEP) which are extracted
from the previous findings. The RPD values for fructose (5.48), glucose
(4.73) and maltose (4.74) those are acquired from our calibration
models are found to be higher than the one reported by Anjos et al.
(2015). Following similar scheme (Ruoff et al., 2006) the values of R2
(SEP) for fructose, glucose and maltose are obtained to be 0.841
(1.20 g/100 g), 0.943 (0.900 g/100 g) and 0.250 (0.600 g/100 g), re-
spectively. Furthermore, utilization of near infrared (NIR) spectroscopy
with PLS regression (Ruoff et al., 2007) yielded lower R2 and higher
SEP values compared to our prediction. Based on these findings, Ruoff
concluded that the NIR and MIR spectra for maltose (R2 ≈ 0.25) is
uncorrelated, which indicated the incapability of such model towards
predicting the maltose content in honey.
Apart from combining infrared spectroscopy with PLS, some studies
mingled Raman spectroscopy with PLS regression analysis (Mignani
et al., 2016; Ozbalci et al., 2013). Moreover, the values of R2 achieved
by us are superior than the earlier report based on Raman spectroscopy
(Table 2), which signified better prediction accuracy of the proposed
models. In short, present study demonstrated that first derivative of
FTIR-ATR data permits the construction of a reliable and accurate PLS
model to predict the concentration of fructose, glucose, and maltose
present in H. itama honeys. This rapid, facile and yet accurate scheme
can be adopted in practice to monitor the quality of Malaysian stingless
bee honey.
The proposed swift scheme integrated the FTIR-ATR spectroscopy
with PLS regression. It is more convenient due to easy handling and
possibly incurring lower operating costs without requiring any ex-
tensive sample preparation steps. The present scheme typically takes
around 5 min to complete sugars prediction in H. itama honey.
Meanwhile, current standard method to determine sugar concentration
in honey (AOAC Official Method 977.20) employs HPLC which is not
suitable for large scale sample analysis because it not only requires
toxic reagents and skillful operator but also involves tedious sample
preparation procedure. As a conclusion, the proposed swift scheme
offers facile and rapid evaluation of sugars content in honey, which is
suitable in real field application and large scale investigation.

4. Conclusion

For the first time, this paper proposed an efficient scheme to profile
accurately the sugars content in diverse Malaysian stingless bee
Fig. 5. PLS regression coefficient of (a) fructose, (b) glucose and (c) maltose in the wa- (Heterotrigona itama) honeys. It is demonstrated that H. itama honeys
venumber range of 1500–750 cm−1.
contained an elevated level of maltose (33.8 g/100 g), low level of
fructose (13.4 g/100 g) and glucose (16.4 g/100 g) as well as insignif-
maltose (4.74). Yet again, the emergence of high RPD values (close to 5) icant amount of sucrose (3.46 g/100 g). It is established that FTIR-ATR
strongly indicated the good predictive capability (very accurate) of the spectrometry integrated with chemometric analysis may constitute a
calibration models for quantifying the sugars concentration in H. itama basis for routine analysis of sugars content in H. itama honey. The
honeys. constructed PLS regression models are optimized by cross validation
Validation dataset is projected to the pre-constructed PLS calibra- and validated using independent test dataset. The developed models
tion model to further validate the model’s quality. The similar magni- revealed best predictive capability to estimate the fructose, glucose and
tude of calculated SEPs to that of SECV clearly indicated the reliability maltose contents wherein the first derivative treatment of the FTIR-ATR
and accurate prediction performance of the constructed PLS regression spectral data is exploited. The excellent attributes of the results sug-
model. High R2 values are achieved for the predicted concentrations of gested that the simplistic combination of FTIR-ATR spectrometry with
fructose (0.942–0.981), glucose (0.934–0.981) and maltose PLS regression analysis is advantageous for the fast and precise pre-
(0.898–0.977). These high R2 values corresponded to the low SEP va- diction of sugars content in honey. This study is an effort to develop a
lues for fructose (0.562 g/100 g), glucose (0.855 g/100 g) and maltose standard for stingless bee honey’s sugar content.

52
K.W. Se et al. Journal of Food Composition and Analysis 66 (2018) 46–54

Table 1
PLS regression statistics for calibration and prediction of detected sugars in honey specimens using FTIR−ATR and HPLC data correlation.

Pre-processed FTIR-ATR data Property of interest Factors Calibration Prediction

R2a SECb (g/100 g) SDc SECVd (g/100 g) RPDe R2 SEPf (g/100 g)

Raw data Fructose 6 0.969 0.685 3.87 0.888 4.36 0.942 0.795
Glucose 7 0.976 0.870 5.63 1.31 4.29 0.965 1.18
Maltose 8 0.978 1.61 10.8 2.49 4.31 0.965 2.01
dy Fructose 4 0.976 0.600 3.87 0.71 5.48 0.976 0.562
dx Glucose 6 0.976 0.864 5.63 1.19 4.73 0.981 0.855
Maltose 6 0.980 1.51 10.8 2.27 4.74 0.977 1.55

d2y Fructose 4 0.979 0.557 3.87 0.83 4.68 0.981 0.518

dx 2 Glucose 3 0.923 1.57 5.63 1.89 2.98 0.934 1.75
Maltose 4 0.958 2.19 10.8 3.58 3.00 0.898 3.25

37 samples used in the calibration dataset; 25 samples used in the validation dataset.
a
R2, coefficient of determination.
b
SEC, standard error of calibration.
c
SD, standard deviation of population from calibration dataset.
d
SECV, standard error of cross validation.
e
RPD, residual predictive deviation.
f
SEP, standard error of prediction.

Table 2
Comparative evaluation of the proposed scheme with existing art-of-the techniques integrated with PLS regression analysis.

Technique Reference Calibration parameter Prediction parameter

RPDc R2d SEPe or RMSEPf (g/100 g)

Fructose Glucose Maltose Fructose Glucose Maltose Fructose Glucose Maltose

a
FTIR-ATR (MIR spectroscopy) Anjos et al. (2015) 2.55 2.67 2.03 0.842 0.880 0.605 – – –
Ruoff et al. (2006) – – – 0.841 0.943 0.250 1.20 0.90 0.60

NIRb spectroscopy Ruoff et al. (2007) – – – 0.810 0.884 0.233 1.60 1.30 0.900

Raman spectroscopy Mignani et al. (2016) – – – 0.820 0.920 0.660 0.760 1.10 0.530
Ozbalci et al. (2013) – – – 0.965 0.964 0.949 – – –

Present scheme 5.48 4.73 4.74 0.976 0.981 0.977 0.562 0.855 1.55

Missing value means not reported in the previous works.

a
MIR, mid-infrared.
b
NIR, near-infrared.
c
RPD, residual predictive deviation.
d
R2, coefficient of determination.
e
SEP, standard error of prediction.
f
RMSEP, root mean square error of prediction [RMSEP = Bias2 + SEP2].

Conflict of interest References

All authors have read and understood the Journal of Food A.O.A.C. (Association of Official Analytical Chemist), 2006. In: Horwitiz, W. (Ed.),
Composition and Analysis’s policy and declare that we have no conflict Official Methods of Analysis of the AOAC, 18th ed. Association of Official Analytical
Chemists, Washington D.C., USA.
of interests. Almeida-Muradian, L.B., 2013. Tetragonisca angustula pot-honey compared to Apis melli-
fera honey from Brazil. In: Vit, P., Pedro, S.R.M., Roubik, D.W. (Eds.), Pot-Honey: A
Legacy of Stingless Bees. Springer, New York, pp. 375–382.
Anguebes, F., Pat, L., Ali, B., Guerrero, A., Cordova, A.V., Abatal, M., Garduza, J.P., 2016.
Acknowledgements Application of multivariable analysis and FTIR-ATR spectroscopy to the prediction of
properties in Campeche honey. J. Anal. Methods Chem. 2016, 1–14.
Authors are thankful to the Malaysian Ministry of Higher Education Anjos, O., Campos, M.G., Ruiz, P.C., Antunes, P., 2015. Application of FTIR-ATR spec-
troscopy to the quantification of sugar in honey. Food Chem. 169, 218–223.
and UTM for financial support through grants (FRGS:
Biluca, F.C., Braghini, F., Gonzaga, L.V., Costa, A.C.O., Fett, R., 2016. Physicochemical
R.J130000.7826.4F869) and (RU/GUP: Q.J130000.2526.13H67 and profiles, minerals and bioactive compounds of stingless bee honey (Meliponinae). J.
17H19). We are also grateful to the President of Johor Entrepreneur of Food Compos. Anal. 50, 61–69.
Cherif, K.T., Peris-Vicente, J., Carda-Broch, S., Esteve-Romero, J., 2015. Analysis of da-
Stingless Bee Society and School of Food Science and Technology, UMT
nofloxacin, difloxacin: ciprofloxacin and sarafloxacin in honey using micellar liquid
for supplying the honeys. chromatography and validation according to the 2002/657/EC decision. Anal.
Methods 7, 6165–6172.
Chuttong, B., Chanbang, Y., Sringarm, K., Burgett, M., 2016. Physicochemical profiles of
stingless bee (Apidae: Meliponini) honey from South East Asia (Thailand). Food Chem.
Appendix A. Supplementary data 192, 149–155.
Codex Alimentarius, 2001. Revised codex standard for honey. 24th Session of the Codex
Supplementary data associated with this article can be found, in the Alimentarius.
Cozzolino, D., Roumeliotis, S., Eglinton, J., 2014. Feasibility study on the use of atte-
online version, at https://doi.org/10.1016/j.jfca.2017.12.002. nuated total reflectance MIR spectroscopy to measure fructan content in barley. Anal.

53
K.W. Se et al. Journal of Food Composition and Analysis 66 (2018) 46–54

Methods 6, 7710–7715. Fourier transform infrared spectrometry. Talanta 71, 1926–1931.

de Sousa, J.M.B., de Souza, E.L., Marques, G., Benassi, M.D.T., Gullon, B., Pintado, M.M., Razak, S.B.A., Aziz, A.A., Ali, N.A., Ali, M.F., Visser, F., 2016. The sustainable integration
Magnani, M., 2016. Sugar profile: physicochemical and sensory aspects of monofloral of meliponiculture as an additional income stream for rubber smallholders in
honeys produced by different stingless bee species in Brazilian semi-arid region. LWT Malaysia. International Rubber Conference 143–156.
Food Sci. Technol. 65, 645–651. Rodriguez-Malaver, A.J., Rasmussen, C., Gutierrez, M.G., Gil, F., Nieves, B., Vit, P., 2009.
Escuredo, O., Gonzalez-Martin, M.I., Rodriguez-Flores, M.S., Seijo, M.C., 2015. Near in- Properties of honey from ten species of Peruvian stingless bees. Nat. Prod. Commun.
frared spectroscopy applied to the rapid prediction of the floral origin and mineral 4, 1221–1226.
content of honeys. Food Chem. 170, 47–54. Rodriguez-Malaver, 2013. Antioxidant activity of pot-honey. In: Vit, P., Pedro, S.R.M.,
Gallardo-Velazquez, T., Osorio-Revilla, G., de Loa, M.Z., Rivera-Espinoza, Y., 2009. Roubik, D.W. (Eds.), Pot-Honey: A Legacy of Stingless Bees. Springer, New York, pp.
Application of FTIR-HATR spectroscopy and multivariate analysis to the quantifica- 475–480.
tion of adulterants in Mexican honeys. Food Res. Int. 42, 313–318. Ruoff, K., Iglesias, M.T., Luginbuhl, W., Bosset, J., Bodganov, S., Amado, R., 2006.
Gok, S., Severcan, M., Goormaghtigh, E., Kandemir, I., Severcan, F., 2015. Differentiation Quantitative analysis of physical and chemical measureands in honey by mid-infrared
of Anatolian honey samples from different botanical origins by ATR-FTIR spectro- spectrometry. Eur. Food Res. Technol. 223, 22–29.
scopy using multivariate analysis. Food Chem. 170, 234–240. Ruoff, K., Luginbuhl, W., Bogdanov, S., Bosset, J., Estermann, B., Ziolko, T.,
Guerrini, A., Bruni, R., Maietti, S., Poli, F., Rossi, D., Paganetto, G., Muzzoli, M., Kheradmandan, S., Amado, R., 2007. Quantitative determination of physical and
Scalvenzi, L., Sacchetti, G., 2009. Ecuadorian stingless bee (Meliponinae) honey: a chemical measurands in honey by near-infrared spectrometry. Eur. Food Res.
chemical and functional profile of an ancient health product. Food Chem. 114, Technol. 225, 415–423.
1413–1420. Santos, P.M., Pereira-Filho, E.R., Rodriguez-Saona, L.E., 2013. Rapid detection and
Hineno, M., 1977. Infrared spectra and normal vibrations of (D-glucopyranose. quantification of milk adulteration using infrared microscpectroscopy and chemo-
Carbohydr. Res. 56, 219–227. metrics analysis. Food Chem. 138, 19–24.
Huang, Y., Carragher, J., Cozzolino, D., 2016. Measurement of fructose, glucose: maltose Shadan, A.F., Mahat, N.A., Wan Ibrahim, W.A., Ariffin, Z., Ismail, D., 2017. Provenance
and sucrose in barley malt using attenuated total reflectance mid-infrared spectro- establishment of stingless bee honey using multi-element analysis in combination
scopy. Food Anal. Methods 9, 1079–1085. with chemometrics techniques. J. Forensic Sci. 1–6.
Jalil, M.A.A., Kasmuri, A.R., Hadi, H., 2017. Stingless bee honey: the natural wound Sivakesava, S., Irudayaraj, J., 2001. A rapid spectroscopic technique for determining
healer: a review. Skin Pharmacol. Physiol. 30, 66–75. honey adulteration with corn syrup. J. Food Sci. 66, 787–791.
Karabagias, I.K., Badeka, A.V., Kontakos, S., Karabournioti, S., Kontominas, M.G., 2014. Sultanbawa, Y., Cozzolino, D., Fuller, S., Cusack, A., Currie, M., Smyth, H., 2015. Infrared
Botanical discrimination of Greek unifloral honeys with physico-chemical and che- spectroscopy as a rapid tool to detect methylglyoxal and antibacterial activity in
mometric analyses. Food Chem. 165, 181–190. Australian honeys. Food Chem. 172, 207–212.
Kek, S.P., Chin, N.L., Tan, S.W., Yusof, Y.A., Chua, L.S., 2017. Classification of honey from Suntiparapop, K., Prapaipong, P., Chantawannakul, P., 2012. Chemical and biological
its bee origin via chemical profiles and mineral content. Food Anal. Methods 10, properties of honey from Thai stingless bee (Tetragonula leaviceps). J. Apic. Res. 51,
19–30. 45–52.
Kelly, N., Farisya, M.S.N., Kumara, T.K., Marcela, P., 2014. Species diversity and external Tayeb-Cherif, K., Peris-Vicente, J., Carda-Broch, S., Esteve-Romero, J., 2016. Use of mi-
nest characteristics of stingless bees in Meliponiculture. Pertanika J. Trop. Agric. Sci. cellar liquid chromatography to analyze oxolinic acid, flumequine: marbofloxacin
37, 293–298. and enrofloxacin in honey and validation according to the 2002/657/EC decision.
Li, S., Zhang, X., Shan, Y., Su, D., Ma, Q., Wen, R., Li, J., 2017. Qualitative and quanti- Food Chem. 202, 316–323.
tative detection of honey adulterated with high-fructose corn syrup and maltose Tewari, J., Irudayaraj, J., 2004. Quantification of saccharides in multiple floral honeys
syrup by using near-infrared spectroscopy. Food Chem. 218, 231–236. using Fourier transform infrared microattenuated total reflectance spectroscopy. J.
Mark, H., Workman, J., 2003. Statistics in Spectroscopy. Academic Press, San Diego, Agric. Food Chem. 52, 3237–3243.
California, USA. Vit, P., Medina, M., Enriquez, M.E., 2004. Quality standards for medicinal uses of
Mignani, A.G., Ciaccheri, L., Mencaglia, A.A., Sanzo, R.D., Carabetta, S., Russo, M., 2016. Meliponinae honey in Guatemala, Mexico and Venezuela. Bee World 85, 2–5.
Dispersive Raman spectroscopy for the nondestructive and rapid assessment of the Wang, J., Kliks, M.M., Jun, S., Jackson, M., Li, Q.X., 2010. Rapid analysis of glucose,
quality of southern Italian honey types. J. Lightwave Technol. 34, 4479–4485. fructose: sucrose and maltose in honeys from different geographic regions using
Naes, T., Isaksson, T., Fearn, T., Davies, T., 2002. A User-Friendly Guide to Multivariate Fourier transform infrared spectroscopy and multivariate analysis. J. Food Sci. 75,
Calibration and Classification. NIR Publication, Chichester, UK. C208–C214.
Ozbalci, B., Boyaci, I.H., Topcu, A., Kadilar, C., Tamer, U., 2013. Rapid analysis of sugars Williams, P.C., 2001. Implementation of near-infrared technology. In: Williams, P.C.,
in honey by processing Raman spectrum using chemometric methods and artificial Norries, K.H. (Eds.), Near Infrared Technology in the Agricultural and Food
neural networks. Food Chem. 136, 1444–1452. Industries. American Association of Cereal Chemists Inc., St. Paul, Minnesota, USA,
Pataca, L.C.M., Neto, W.B., Marcucci, M.C., Poppi, R.J., 2007. Determination of apparent pp. 145–169.
reducing sugars: moisture and acidity in honey by attenuated total reflectance-

54