Food Chemistry 194 (2016) 873880

Contents lists available at ScienceDirect

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

Analytical Methods

NIR detection of honey adulteration reveals differences in water spectral

pattern
Gyrgy Bzr a,b,, Rbert Romvri a, Andrs Szab a, Tams Somogyi a, Viktria les a,
Roumiana Tsenkova b,
a
Institute of Food and Agricultural Product Qualification, Faculty of Agricultural and Environmental Sciences, Kaposvr University, 40 Guba Sndor Str, Kaposvr 7400, Hungary
b
Biomeasurement Technology Laboratory, Graduate School of Agricultural Science, Kobe University, 1-1 Rokkodai, Nada-ku, Kobe 657-8501, Japan

a r t i c l e i n f o a b s t r a c t

Article history: High fructose corn syrup (HFCS) was mixed with four artisanal Robinia honeys at various ratios (040%)
Received 24 July 2014 and near infrared (NIR) spectra were recorded with a fiber optic immersion probe. Levels of HFCS adul-
Received in revised form 17 August 2015 teration could be detected accurately using leave-one-honey-out cross-validation (RMSECV = 1.48;
Accepted 22 August 2015
R2CV = 0.987), partial least squares regression and the 13001800 nm spectral interval containing absorption
Available online 28 August 2015
bands related to both water and carbohydrates. Aquaphotomics-based evaluations showed that unifloral
honeys contained more highly organized water than the industrial sugar syrup, supposedly because of
Keywords:
the greater variety of molecules dissolved in the multi-component honeys. Adulteration with HFCS caused
Near infrared spectroscopy
Aquaphotomics
a gradual reduction of water molecular structures, especially water trimers, which facilitate interaction with
Honey other molecules. Quick, non-destructive NIR spectroscopy combined with aquaphotomics could be used to
Fructose describe water molecular structures in honey and to detect a rather common form of adulteration.
Water 2015 Elsevier Ltd. All rights reserved.
Aliment
Adulteration
Fraud

1. Introduction
Honey is an important commodity in the food market and is
acknowledged as a healthy natural sweetener, which is consumed
alone or as an ingredient in a variety of sweet products. The worlds
average annual honey production was 1.5 billion kg between 2005
and 2010 (FAOSTAT, 2013). Since consumption of honey in Hun-
gary is very low (0.24 kg/capita/year), the country has become
the second biggest honey exporter in the European Union, deliver-
ing 19 million kg per year to the market, covering more than 18% of
the total EU export of honey. China and many European countries,
particularly in the East, produce considerable amounts of False
Acacia (Robinia pseudoacacia L.) unifloral honey, which is one of
the most valuable honey products on the European market. How-
ever, the characteristics of Robinia honey (i.e. liquid due to the high
fructose content glucose:fructose = 1:1.41.7, very light colored
and flavored) can be easily counterfeited (Persano Oddo & Piro,
2004).
The European Commission has defined honey (European
Commission, 2002) and stipulated that nothing should be added
Corresponding authors at: Institute of Food and Agricultural Product Qualifi-
cation, Faculty of Agricultural and Environmental Sciences, Kaposvr University, 40
Guba Sndor Str, Kaposvr 7400, Hungary (G. Bzr).
E-mail addresses: bazar@agrilab.hu (G. Bzr), rtsen@kobe-u.ac.jp (R. Tsenkova).
to it. However, since there is a great demand for high quality honey
and consumers intend to pay high prices for these products, valu-
able honeys have become a target of adulteration with cheap sweet
foreign components worldwide. Adulterated honeys are often
labeled as natural and priced the same as pure honey, which is
both fraudulent and unfair to consumers (Chen et al., 2011). In gen-
eral, the adulteration of honey does not pose a health risk. How-
ever, it does influence market growth negatively by damaging
consumer confidence (Mehryar, 2011). Consequently, identifica-
tion and authentication of unadulterated honey is important for
processors, retailers and consumers, as well as regulatory authori-
ties (Chen et al., 2011).
Large-scale honey adulteration appeared on the world market
in the 1970s with the introduction of high-fructose corn syrup
(HFCS) by the industry (Mehryar, 2011). Based on the pioneering
investigations of Doner and White (1977), HFCS content in natural
honey samples was described using mass spectroscopy (Doner and
White, 1978), gasliquid chromatography (Doner, White, &
Phillips, 1979) and high pressure liquid chromatography (White,
1980). Although these methods and their further developed appli-
cations can accurately detect honey adulteration, they have a num-
ber of disadvantages that discourage practical application in
screening systems, i.e., the need for highly skilled labor,
time-consuming, destructive and ultrapure preparation and large

http://dx.doi.org/10.1016/j.foodchem.2015.08.092
0308-8146/ 2015 Elsevier Ltd. All rights reserved.
874 G. Bzr et al. / Food Chemistry 194 (2016) 873880
instrumental analysis. Thus, there is an obvious demand for devel-
oping easy to use, rapid, non-destructive, and low-cost analytical
methods to detect and quantify HFCS adulteration in commercial
honey lots.
Near infrared (NIR) spectroscopy is a rapid, highly accurate, ver-
satile, multi-analytical technique based on the electromagnetic
absorption of organic compounds in the short wavelength infrared
range (7802500 nm). It can be applied to qualitative or quantita-
tive analyses of multiple components of a sample by a single mea-
surement. Moreover, no reagents are required and no hazardous
waste is produced. The NIR technique has been applied in a wide
range of fields in the last decades and several of these applications
are currently in use as routine analyses, even in on-line monitoring
systems, both in agriculture and the food industry (Downey, 1996;
Huang, Yu, Xu, & Ying, 2008; Prieto, Roeho, Lavn, Batten, & Andrs,
2009).
The use of the NIR technique for food authentication was
reviewed by Reid, ODonnell, and Downey (2006). In regards to
honey, Cho and Hong (1998) and Ha, Koo, and Ok (1998) developed
NIR prediction method for conventional quality parameters of
Korean honeys using transflectance cuvettes with gold and
aluminum reflector back. Garca-Alvarez, Huidobro, Hermida, and
Rodrguez-Otero (2000) analyzed fructose, glucose and moisture
content of honeys, and reported proper results in calibration and
independent validation for all three components when the
calibration was generated using transflectance spectra. As an
authentication tool, NIR spectroscopy has been applied success-
fully for rapid determination of the floral origin of honeys (Chen
et al., 2012; Escuredo, Gonzlez-Martn, Rodrguez-Flores, & Seijo,
2015). Transflectance NIR spectra of Irish artisanal honeys
adulterated with fructose and glucose were acquired by Downey,
Fouratier, and Kelly (2003) using a gold back reflector with
0.1 mm sample thickness, and accurate classification models were
generated to reveal adulteration. Kelly, Petisco, and Downey (2006)
investigated the detection of honey adulteration with sugar-beet
invert syrup and HFCS using both qualitative and quantitative
NIR models. Applying a fiber optic probe, successful classification
was performed by Chen et al. (2011) using FT-NIR spectra of
unadulterated and adulterated Chinese honeys collected from
apiaries and purchased in local groceries.
Water is one of the most studied compounds in physics and
chemistry, the properties of which are explained by several differ-
ent structural models based on the number, nature and strength of
hydrogen bonds (Segtnan, Sasic, Isaksson, & Ozaki, 2001).
Aquaphotomics, as a new approach in NIR spectroscopy, considers
water as a multi-element system that can be described by its
multi-dimensional NIR spectra (Tsenkova, 2009). Since water-
related H-bonds are present in most natural samples, this analyti-
cal approach, using perturbed water in different environments as a
mirror for the rest of the molecules in the sample, can be applied
effectively in various fields (Kinoshita et al., 2012; Tsenkova,
Meilina, Kuroki, & Burns, 2009).
The objective of the present study was to develop an applicable
NIR model for screening unifloral Robinia honey using a fiber-optic
probe. Recent findings of aquaphotomics were implemented to
interpret the chemometric calibration models for measuring the
level of HFCS adulteration, and to extract important information
on functionality of the honey related to its water structures.
2. Materials and methods
2.1. Honey and HFCS samples
Pure Robinia honey samples were collected from four
geographic regions of Hungary, during periods of False Acacia
flowering in 2012. Liquid, cleaned, sterile, ion exchanged and
filtered isoglucose syrup (HFCS) was produced in a high-
temperature closed process, contained 40% fructose and 33%
glucose at 18.7% of moisture. All samples were stored in dark glass
vials at room temperature (25 C) for approximately 5 months,
until NIR analysis. Because of the lack of refrigeration slight chem-
ical changes may have occurred in the samples. However, from a
practical point of view, this represents the real-life situation as
regards commercial honeys (Kelly et al., 2006). One day prior to
NIR scanning, individual honey samples were diluted randomly
with HFCS (n = 40, range of honey concentration = 10060%;
mean SD = 80.79 12.89). Samples (about 50 g) were produced
for each mixture by measuring the mass of the constituents (one
of the Robinia honeys plus HFCS) into glass vials on a laboratory
scale and blending with a glass rod. HoneyHFCS mixtures were
covered and stored at room temperature for 24 h in order to
remove any bubbles produced during blending, since these could
interfere with NIR scanning through light scattering. Four pure
honeys, the HFCS sample, and randomly selected (n = 19) honey
HFCS mixtures were tested for dry matter content after NIR scan-
ning, by means of conventional drying at 105 C for 4 h, in three
replicates, and mean values were applied for NIR calibration. The
standard error of laboratory (SEL) was determined for the reference
data according to Workman and Mark (2006).
2.2. Near infrared spectra
Pure and artificially adulterated honey samples in glass holders
were scanned at room temperature (25 C) using a NIRSystems
6500 spectrometer (FOSS NIRSystems, Inc., Laurel, MD, USA)
equipped with an OptiProbe fiber optic immersion sampling unit
with 2 mm layer thickness. Transflectance spectra (1100
2500 nm) were recorded with a spectral step of 2 nm as an average
of 32 scans. The acquisition of logR 1 spectra was performed with
VISION 2.51 software (FOSS NIRSystems, Inc., Laurel, MD, USA). The
noise level of model 6500 monochromator was certified for
50 lAbs unit. Slightly increased noise level was experienced during
performance tests because of the application of the fiber optic
probe (noise level of the 13001800 nm region <80 lAbs unit).
The sample set was scanned in two runs, on two consecutive days.
The samples were presented for scanning in random order in both
repeated series. Six consecutive spectra were recorded for each
sample at each time. Thus, one sample was represented by 12 spec-
tra in total, from two repeated measurement days. The total num-
ber of scanned samples (pure honeys, HFCS and mixtures) was
n = 41, and the total number of stored spectra was n = 492. During
scanning, the OptiProbe was immersed into a sample and glass
holders were rotated between repeated spectral acquisitions in
order to flush the measuring slit with fresh sample. The OptiProbe
was washed with hot water and dried after each sample.
2.3. Spectral analysis
Recorded spectra were exported from VISION in NSAS file for-
mat and The Unscrambler 9.7 (CAMO Software AS, Oslo, Norway)
was applied for data processing and analysis. Low energy light at
longer wavelengths could not pass through the samples, meaning
absorbance spectra above 1850 nm showed stray light. In the
shorter wavelength region (11001200 nm) undefined noise was
detected during basic spectral evaluation. Considering these
effects, the 13001800 nm range was selected for evaluation of
NIR data. Moving average smoothing was applied with 15 spectral
points before standard normal variate (SNV) transformation
(Barnes, Dhanoa, & Lister, 1989) in order to correct scatter effect.
Derivative spectra were calculated using the gap-segment method
(Norris, 2001) with various gap and segment sizes. Principal
 G. Bzr et al. / Food Chemistry 194 (2016) 873880
component analysis (PCA) was used to describe the basic multidi-
mensional characteristics of the NIR data matrix (Cowe & McNicol,
1985). Calibration for HFCS or dry matter content was conducted
using principal component regression (PCR) and partial least
squares regression (PLSR), and models were tested by cross-
validation (Ns, Isaksson, Fearn, & Davies, 2002). In case of PCR,
the principal components are generated solely from the spectral
data and are optimized to describe the highest proportion of vari-
ance of the spectral variables. In contrast, PLSR applies also the
Y-variables of the predictable constituents and optimizes the latent
variables of the model to describe as much amount of the variance
of X-(spectral data) and Y-variables (reference data) as possible, at
the same time. Leave-one-honey-out cross-validation was applied,
where spectra for one of the four honeys and all of its mixtures
were omitted from the calibration, and the model was tested on
the remaining samples. The procedure was repeated iteratively
until all groups were used for testing. The precision and accuracy
of the chemometric models were evaluated by the determination
coefficient (R2, R2CV) and the root mean square error (RMSE, RMSECV)
of calibration and cross-validation, respectively (Ns et al., 2002).
Regression vectors of the calibration models were analyzed in
order to assign the characteristic water bands, water matrix coor-
dinates (WAMACs), showing considerable changes in honey mix-
tures according to HFCS content. The variation of WAMACs
describes the water spectral pattern (WASP), which can be visual-
ized in aquagrams (Kinoshita et al., 2012). The star-chart applied
for the aquagram displays average normalized absorbance values
from different sample groups at several water bands on the axes
originating from the center of the graph. Absorbance values at
the WAMACs were placed on the respective radial axes. Aquagrams
were prepared in Microsoft Office Excel 2010 (Microsoft Co., Red-
mond, WA, USA).
3. Results and discussion
3.1. Dry matter based evaluation
The dry matter content of pure honey samples and HFCS was
87.2 0.65 (mean SD) and 81.3%, respectively, while that of the
honeyHFCS mixtures (n = 19) was 86.4 0.71 (mean SD). Stan-
dard error of laboratory (SEL) for the reference measurement was
Fig. 1. (a) Raw and (b) 2nd derivative spectra of four honeys, HFCS, and honeyHFCS mixtures (n = 492; blue spectra: HFCS, reddish spectra: pure and adulterated honeys).
875
0.11. Based on the almost 5% difference in moisture of the test
materials, a strong negative correlation (r = 0.99) was found
between dry matter and HFCS content of the mixtures of the four
individual honeys. The correlation was weaker (r = 0.81) when
it was calculated for all the tested honeyHFCS mixtures together.
3.2. Basic spectral differences
Line plots in Fig. 1 show the NIR spectra of the samples investi-
gated. The 2nd derivative spectra were calculated with 5-point gap
and 5-point segment. The negative peaks of the 2nd derivative
graphs expose the underlying peaks appearing overlapped in the
raw spectrum.
The average spectrum of the four pure honeys and that of the
HFCS sample was calculated. Raw average spectra and their 2nd
derivatives are shown in Fig. 2. The mean spectrum of HFCS was
subtracted from the honeys and the difference is plotted to empha-
size spectral regions showing the biggest variation in the two test
materials. In this case, the averaging of spectra decreased the over-
all noise level and, thus, a more sensitive setting for derivative cal-
culation was applied: 5-point smoothing, then 2nd derivatives
with 5-point gap and 1-point segment. This setting could find
and visualize narrow bands within the spectra.
As for band assignment it must be noted that the spectral step
and bandpass of the applied spectrometer was 2 nm and 10 nm,
respectively. Because of the application of the fiber optic probe,
increased noise level (<80 lAbs) was experienced and, thus, it
was necessary to involve smoothing procedures with several
points in the chemometric models. Based on all these effects, small
shifts of the bands can be expected during assignment and com-
parison with literature data.
The peak of the derivative graphs in Figs. 1 and 2 are consistent,
despite being calculated differently, and accentuate the following
bands: 1356, 14341436, 1476, 1502, and 1580 nm, representing
water with less or more Hbonds (Rambla, Garrigues, & de la
Guardia, 1997; Segtnan et al., 2001; Xantheas, 1995) as well as
1690, 17281730, and 1780 nm, which are CH stretching in sug-
ars (Golic, Walsh, & Lawson, 2003; Kelly et al., 2006; Workman,
2001). The signal at 1690 nm is characteristic for fructose
(Williams and Norris, 2001) and indicates the marked difference
in glucose:fructose ratio of the investigated HFCS and honeys (glu-
cose:fructose = 1:1.21.3 and 1:1.41.7 in HFCS or Robinia honey,
876 G. Bzr et al. / Food Chemistry 194 (2016) 873880

Fig. 2. (a) Raw and (b) 2nd derivative average spectra of pure honeys and HFCS, and their subtracted spectra.

respectively). Application of this single point of the spectrum, how-

ever, is problematic since the fructose ratio of artisanal honeys
may vary greatly. The dominant peak of the subtracted spectrum
at 1418 nm, in Fig. 2, is related to the ascending part of the water
first overtone absorption peak (1st overtone of OH stretching)
(Osborne, Fearn, & Hindle, 1993). The peak position of the sub-
tracted spectrum is not exactly in the position of the characteristic
peak but on its side, showing that differences in this region origi-
nate from the fact that HFCS provides a water peak shifted to the
shorter wavelength regions, while this absorption band in honey
appears at longer wavelengths, without great change in the absor-
bance level. This reflects that spectra contain information mainly
on the water structure. Subtracted peaks of Fig. 2 show spectral
variation between HFCS and honey at the 1st overtone of CH
bonds of carbohydrates around 1690, 1728, and 1780 nm (Golic
et al., 2003; Kelly et al., 2006; Workman, 2001).
PCA score plot generated on all recorded spectra and the line
plot of the first principal component (PC) are shown in Fig. 3. As
the score plot depicts, the basic spectral variation of the samples
resembles the adulteration level. HFCS samples form a separate
group, while honeyHFCS mixtures are plotted in the order of
the mixing ratio, between pure honeys and HFCS. The first PC
describing 97% of the total variance of the NIR data is dominated
by the difference in the spectral properties of the test materials.
The subtracted spectrum of non-derivative spectra shown in
Fig. 2 is nearly identical with the first PC (Fig. 3b).

3.3. Calibration models on dry matter and purity

PCR and PLSR were applied for calibration on dry matter and
purity (pure honey content against HFCS) of pure and adulterated
honeys. Spectra of pure HFCS were not applied for any of the mod- Fig. 3. (a) PCA score plot and (b) loadings of the 1st PC calculated on the entire
sample set (n = 492) after 15-point smoothing and SNV transformation of the NIR
els. Analyses were run using different spectral intervals, i.e. the
spectra (13001800 nm).
13001800 nm region was divided in two parts: 13001600 nm
representing the first overtone region of water stretching vibra-
tions (OH bands), and 16001800 nm, predominantly represent-
ing the first overtone region of carbohydrates (CH bands). formed before or after 2nd derivation. There were no differences
Second derivatives as spectral pretreatment were applied on in the appearance of the spectra when SNV was applied prior or
SNV transformed spectra. Since the order of the applied spectral post derivation, and only slight differences were observed in the
treatments may affect the results (Fearn, 2008), SNV was per- results, while regression vectors showed the same bands in both
 G. Bzr et al. / Food Chemistry 194 (2016) 873880
Fig. 4. (a, c) Regression vectors of PLSR models with (b, d) the results of calibrations and cross-validations on (a, b) dry matter content and (c, d) purity (honey percentage
against HFCS) of adulterated honey samples using the 2nd derivative of the smoothed (15 points) and SNV transformed 13001800 nm spectral interval (nr. of applied
factors = 5).
cases. Thus, only results achieved with SNV made prior to the
calculation of derivatives are shown here.
Both PCR and PLSR models were developed using the aforemen-
tioned three spectral ranges, applying three different spectral
treatments: (a) smoothing with 15 points and SNV, (b) 2nd deriva-
tive with 5 point gap and 15 point segment, (c) smoothing with 15
points, SNV, and 2nd derivative with 5 point gap and 15 point seg-
ment. Only results of PLSR are summarized here, but the PCR mod-
els and their regression coefficient vectors were also evaluated in
order to assign the important spectral bands.
Fig. 4 shows the calibration and cross-validation results of the
PLSR models achieved with the 2nd derivatives of the smoothed
and SNV transformed spectra in the 13001800 nm interval. The
regression vectors shown in Fig. 4 accentuate the bands having
dominance in the calibrations. Results from the PLSR calibration
models with different spectral intervals are summarized in Table 1.
RMSECV results on dry matter content are comparable with the
SEL value of the reference method, which shows the strength of the
applied NIR models. Calibration models provided results at high
regression and low error level on purity. Results achieved with
smoothed and SNV transformed non-derivative spectra were the
most accurate when the 13001800 nm spectral interval was
applied: error of cross-validation (RMSECV) was less than 1.5%.
Using these spectral treatments, characteristic region of carbohy-
drates (16001800 nm) led to weaker results on purity than that
of water (13001600 nm). Applying only the 1st overtone region
of water, the error was 4% (R2CV = 0.91), still less than 1/3 of the stan-
dard deviation of the reference values. The application of derivatives
877
revealed signals from the carbohydrate region (16001800 nm) and
the error (RMSECV) decreased to 1.5%. It is remarkable, however, that
the addition of water regions and application of the whole range
(13001800 nm), generally, increased the accuracy and precision of
the models. This shows that water bands hold additional information
on the constituents of the mixtures.
Kelly et al. (2006) investigated the detection of honey adulter-
ation with HFCS using transflectance spectra of 0.1 mm thick layer
of samples recorded with a camlock cell and gold reflector. The
applied sample set comprised 83 authentic honeys, of which 10
were adulterated with HFCS at the inclusion level of 10%, 30%,
50% and 70%. The total number of samples in HFCS calibration
was 123. PLS regression was used to predict the adulteration level,
and the best coefficient of determination was 0.72 for HFCS con-
tent, with an RMSECV of 11.9, achieved with the 2nd derivative
spectra of 11002498 nm region. Compared to the results of
Kelly et al. (2006), the precision and accuracy of the models
reported in the present study are more robust, although spectra
were recorded with a fiber optic probe, but there was a great differ-
ence in sample number and the range and variance of the adulter-
ation level and, thus, results are not strictly comparable.
Averaging repeated scans is a useful method to reduce errors
originating from sampling, environmental or instrumental noise
(Norris, 2002). In this case, averaging was applied to determine
the influence of noise on the results. The six consecutive spectra
of each sample were averaged in pairs: average of 1st and 6th,
2nd and 5th, 3rd and 4th was calculated. Applying this procedure,
the possible distorting effect of consecutive scanning was also
878 G. Bzr et al. / Food Chemistry 194 (2016) 873880

Table 1 in Robinia honey. These bands are the WAMACs for this complex
Results of PLSR calibrations and cross-validations on dry matter content and purity of system, and the variation in absorbance values at these bands
adulterated honey mixtures applying different spectral regions and various spectral
treatments.
describe the WASP of the sample sets. In our application, the aqua-
gram shows a relative fingerprint of a sample, as it contains nor-
Range Y var.a Nr.b RSQc RMSEd RSQCVe RMSECVf malized absorbance values at specific water bands, and the graph
(a) Smoothing (15 points), SNV of a group depends on the dominance and characteristics of the
13001800 nm DMg 5 0.981 0.10 0.981 0.12 other samples. The similarity of honeys and their unlikeness from
Purityh 5 0.997 0.72 0.988 1.49
13001600 nm DM 2 0.937 0.19 0.924 0.23
HFCS can be seen very clearly using this visualization method.
Purity 4 0.969 2.25 0.910 4.01 Aquagram in Fig. 5a shows the pure honey samples against the
16001800 nm DM 4 0.979 0.11 0.958 0.17 HFCS sample, while Fig. 5b demonstrates the aquagram of adulter-
Purity 2 0.946 2.98 0.892 4.49 ated mixtures of one honey (H1), plotted together with the mean
(b) 2nd derivative (gap = 5 points, segment = 15 points) graph of pure honeys and HFCS.
13001800 nm DM 2 0.975 0.12 0.945 0.20 The bands accentuated by the regression vectors of the devel-
Purity 6 0.996 0.81 0.985 1.62
oped models, and applied for aquagrams, are consistent with the
13001600 nm DM 3 0.971 0.13 0.958 0.17
Purity 4 0.934 3.31 0.889 4.49 assigned peaks of the derivative spectra (Figs. 1 and 2), although
16001800 nm DM 5 0.978 0.11 0.955 0.18 the calculation method of derivatives was not the same during
Purity 4 0.996 0.86 0.989 1.52 the different procedures this demonstrates the stability of the
(c) Smoothing (15 points), SNV, 2nd derivative (gap = 5 points, segment = 15 applied methodology. Basically, the 13221326 nm and 1358
points) 1366 nm regions represent non or less H-bonded water, where sig-
13001800 nm DM 5 0.981 0.10 0.984 0.11 nals of free OH vibrations dominate (Xantheas, 1995), but this
Purity 5 0.995 0.90 0.987 1.48
13001600 nm DM 2 0.949 0.17 0.943 0.20
region is also related to highly organized water in water solvation
Purity 4 0.917 3.72 0.816 5.74 shell as it corresponds to the 1st overtone of those bands described
16001800 nm DM 2 0.976 0.12 0.960 0.17 in relation with aqueous protons (Headrick et al., 2005), and free
Purity 4 0.993 1.00 0.987 1.54 OH in water clusters (Mizuse & Fujii, 2012). Temperature studies of
a
Reference variable of calibration model. liquid water demonstrated that the signals of less H-bonded water
b
Number of latent variables applied in the calibration model. appear around 1412 nm, and water having more H-bonds provides
c
Determination coefficient of calibration. signals at 1462 and 1490 nm (Collins, 1925; Segtnan et al., 2001).
d
Root mean square error of calibration.
e
Segtnan et al. (2001) also assigned an intermediate stage of water
Determination coefficient of cross-validation.
f
Root mean square error of cross-validation.
at 1438 nm. Bands above 1500 nm are related to the 1st overtone
g
Dry matter, in percentage (mean SD = 86.57 0.75; SEL = 0.11; nr. of of ice-like clusters of water, having highly organized structure of
samples = 23; nr. of spectra = 276). molecules (Headrick et al., 2005; Mizuse and Fujii, 2012), and such
h
Pure honey content of sample against HFCS, in percentage structure of water is expected around the hydrated macro-
(mean SD = 80.79 12.89; nr. of samples = 40; nr. of spectra = 480).
molecules. Rambla et al. (1997) assigned 15801590 nm region
for aqueous solutions of fructose (1583 nm), sucrose (1584 nm)
reduced. The PCR and PLSR models described above were per- and glucose (1587 nm), which are the main sugars of honey.
formed on the averaged data set. Results were slightly weaker than Different honeys in Fig. 5a show relatively large variation at
those obtained with the original spectra, and regression vector 14641472 nm and 15001504 nm bands. It can be assumed that
peaks appeared at the same positions as without averaging. This honeys functionality is highly related to these bands representing
showed that the calculation of models was not greatly influenced altering number of H-bonds in dynamic water structure that is the
by random noise or sample-light interactions, and the assigned essential form of water in biological systems (Chandler, 2002). It is
bands contained reliable information. also notable how these bands changed with thinning (Fig. 5b).
The peak positions in the regression coefficient vectors of the Especially the 14641472 nm band of the active water trimers
related models were collected to summarize the bands having (Tsenkova, 2009) decreased when honeys were adulterated with
the biggest weights in calibration on dry matter or HFCS content. industrial fructose syrup and, thus, structures which characterize
The accentuated regions were very similar in case of PLSR and functional water facilitating interactions with other molecules
PCR models. Bands having the highest importance in the calibra- reduced gradually during adulteration with HFCS.
tion on the adulteration level were not analogous with those of The aquagrams of the present study demonstrate evidence on
the calibration model on dry matter content (Fig. 4). This difference the differences in the water structure of the investigated unifloral
was observed in both PLSR and PCR models, and may reflect that Robinia honeys and the applied HFCS. The orientation of the graphs
although dry matter content (or moisture) can be predicted per- suggest that honeys contain larger amount of highly organized
fectly by NIR models, and it highly correlates with the adulteration water than the industrial fructose syrup, supposedly because of
level, NIR spectra contain more information on the level of adulter- the larger variety of molecules (proteins, fibers, vitamins, minerals)
ation than solely the moisture content of the samples. This dissolved in the multicomponent system of honey, while HFCS has
extended information is represented in the chemometric models relatively large amount of free OH in unstructured water with less
on HFCS concentration. Presumably, the additional information H-bonds. This difference explains the increased liquidity of HFCS
may lay in the structure of water molecular system related to compared to Robinia honeys, even at a lower fructose ratio (glu-
moisture and purity, as it can be seen in the peak shifts mentioned cose:fructose = 1:1.21.3 or 1:1.41.7 in HFCS or Robinia honey,
at Fig. 2. respectively). The shift of the dominant water peak described at
Figs. 2 and 3 also supports that the investigated HFCS contains less
H-bonded water than the Robinia honeys.
3.4. Aquaphotomics based evaluation of water structural changes in It must be noted that fructose and glucose show considerable
adulterated honeys differences in the applied 13001600 nm NIR range, i.e. both
mono-saccharides provide signals because of their OH bonds,
Based on the consistency of assigned bands of the subtracted but dominant regions of fructose appear at 14201440 nm, while
spectra and the regression coefficient vectors, ten characteristic those of glucose appear at 15001540 nm (Williams and Norris,
ranges can be defined which are responsible for HFCS detection 2001). These characteristics are not represented in the aquagrams
 G. Bzr et al. / Food Chemistry 194 (2016) 873880
Fig. 5. (a) Aquagram of the investigated pure Robinia honeys and HFCS. (b) Aquagram of the adulterated mixtures of one honey (H1-purity %) plotted with the mean graph of
the pure honeys and graph of HFCS.
since the changes of the aquagrams are not appearing according to
the ratio and spectral differences of the fructose and glucose in the
composites (i.e. the ratio of fructose is higher in honey than in
HFCS, still, HFCS is shifted to the 14201440 nm region). We
assume that information of water had the most impact in the PLSR
and PCR calibrations on HFCS adulteration of Robinia honeys, and
the aquagrams are presenting this information effectively.
4. Conclusions
Although strong negative correlation was found between the
dry matter content and the adulteration level of Robinia honeys
mixed with HFCS, the conventional measurement of the dry matter
content can hardly be used as a quick adulteration monitoring
technique since this measure is allowed to vary in a relatively wide
range even in artisanal Robinia honeys. Based on the NIR spectral
information of honeys recorded with fiber optic probe, quick ana-
lytical tests can be developed to detect aliment adulteration. The
most accurate NIR models for predicting adulteration level with
the lowest cross-validation error (RMSECV = 1.48%) were achieved
within the whole spectral range of 13001800 nm, containing the
879
absorption bands of both water and carbohydrates. Using NIR tech-
nique and aquaphotomics, it has been demonstrated that there is a
well describable difference in the water molecular structure in
Robinia honey and HFCS. The aquagrams presented in this study
emphasize that the investigated unifloral Robinia honeys contain
larger amount of highly organized water than the added industrial
fructose syrup. The difference may be caused by the larger variety
of molecules dissolved in the multicomponent system of honeys,
while the simpler matrix of HFCS may have relatively large amount
of less H-bonded, unstructured water. Adulteration of Robinia hon-
eys with HFCS caused gradual reduction of water molecular struc-
tures facilitating the interactions with other molecules. This
reasoning elucidates the understanding of honeys functionality
and, therefore, needs further and more detailed investigations.
Acknowledgements
This work is dedicated to the memory of Kroly Kaffka. The
financial support of JSPS Postdoctoral Fellowship for Foreign
Researchers (P12409) and TMOP 4.2.2.-B research Grant is highly
acknowledged. Authors would like to thank the assistance of the
880 G. Bzr et al. / Food Chemistry 194 (2016) 873880
Institute of Apiculture and Bee Biology at Research Centre for Farm
Animal Gene Conservation, and Hungrana Starch and Isosugar
Manufacturing and Trading Co. Ltd.
References
Barnes, R. J., Dhanoa, M. S., & Lister, S. J. (1989). Standard normal variate
transformation and de-trending of near-infrared diffuse reflectance spectra.
Applied Spectroscopy, 43, 772777.
Chandler, D. (2002). Hydrophobicity: Two faces of water. Nature, 417, 491.
Chen, L., Wang, J., Ye, Z., Zhao, J., Xue, X., Heyden, Y. V., & Sun, Q. (2012).
Classification of Chinese honeys according to their floral origin by near infrared
spectroscopy. Food Chemistry, 135, 338342.
Chen, L., Xue, X., Ye, Z., Zhou, J., Chen, F., & Zhao, J. (2011). Determination of Chinese
honey adulterated with high fructose corn syrup by near infrared spectroscopy.
Food Chemistry, 128, 11101114.
Cho, H. J., & Hong, S. H. (1998). Acacia honey quality measurement by near infrared
spectroscopy. Journal of Near Infrared Spectroscopy, 6, A329A331.
Collins, J. R. (1925). Change in the infra-red absorption spectrum of water with
temperature. Physical Review, 26, 771779.
Cowe, I. A., & McNicol, J. W. (1985). The use of principal components in the analysis
of near-infrared spectra. Applied Spectroscopy, 39, 257266.
Doner, L. W., & White, J. W. (1977). Carbon-13/carbon-12 ratio is relatively uniform
among honeys. Science, 197, 891892.
Doner, L. W., White, J. W., & Phillips, J. G. (1979). Gasliquid chromatographic test
for honey adulteration by high fructose corn sirup. Journal of Association of
Official Analytical Chemists, 62, 186189.
Downey, G. (1996). Authentication of food and food ingredients by near infrared
spectroscopy. Journal of Near Infrared Spectroscopy, 4, 4761.
Downey, G., Fouratier, V., & Kelly, J. D. (2003). Detection of honey adulteration by
addition of fructose and glucose using near infrared transflectance
spectroscopy. Journal of Near Infrared Spectroscopy, 11, 447456.
Escuredo, O., Gonzlez-Martn, M. I., Rodrguez-Flores, M. S., & Seijo, M. C. (2015).
Near infrared spectroscopy applied to the rapid prediction of the floral origin
and mineral content of honeys. Food Chemistry, 170, 4754.
European Commission (2002). Council Directive 2001/110/EC (20 December 2001)
relating to honey. Official Journal of the European Union, L010, 4752.
FAOSTAT (Statistical Database of the Food and Agriculture Organization of the
United Nations). URL http://faostat.fao.org/ Accessed 11.12.13.
Fearn, T. (2008). The interaction between standard normal variate and derivatives.
NIR News, 19(7), 1617.
Garca-Alvarez, M., Huidobro, J. F., Hermida, M., & Rodrguez-Otero, J. L. (2000).
Major components of honey analysis by near-infrared transflectance
spectroscopy. Journal of Agricultural and Food Chemistry, 48, 51545158.
Golic, M., Walsh, K., & Lawson, P. (2003). Short-wavelength near-infrared spectra of
sucrose, glucose, and fructose with respect to sugar concentration and
temperature. Applied Spectroscopy, 57, 139145.
Ha, J., Koo, M., & Ok, H. (1998). Determination of the constituents of honey by near
infrared spectroscopy. Journal of Near Infrared Spectroscopy, 6, A367A369.
Headrick, J. M., Diken, E. G., Walters, R. S., Hammer, N. I., Christie, R. A., Cui, J., ...
Jordan, K. D. (2005). Spectral signatures of hydrated proton vibrations in water
clusters. Science, 308, 17651769.
Huang, H., Yu, H., Xu, H., & Ying, Y. (2008). Near infrared spectroscopy for on/in-line
monitoring of quality in foods and beverages: A review. Journal of Food
Engineering, 87, 303313.
Kelly, J. D., Petisco, C., & Downey, G. (2006). Potential of near infrared transflectance
spectroscopy to detect adulteration of Irish honey by beet invert syrup and high
fructose corn syrup. Journal of Near Infrared Spectroscopy, 14, 139146.
Kinoshita, K., Miyazaki, M., Morita, H., Vassileva, M., Tang, C., Li, D., ... Tsenkova, R.
(2012). Spectral pattern of urinary water as a biomarker of estrus in the giant
panda. Scientific Reports, 2, 856. http://dx.doi.org/10.1038/srep00856.
Mehryar, L. (2011). Honey and honey adulteration detection: a review. In 11th
International Congress on Engineering and Food. Athens, Greece.
Mizuse, K., & Fujii, A. (2012). Tuning of the internal energy and isomer distribution
in small protonated water clusters H+(H2O)48: an application of the inert gas
messenger technique. The Journal of Physical Chemistry A, 116, 48684877.
Ns, T., Isaksson, T., Fearn, T., & Davies, T. (2002). A user-friendly guide to
multivariate calibration and classification. Chichester: NIR Publications.
Norris, K. H. (2001). Understanding and correcting the factors which affect diffuse
transmittance spectra. NIR News, 12(3), 69.
Norris, K. H. (2002). More on the factors affecting diffuse transmittance and
reflectance spectra. NIR News, 13(3), 811.
Osborne, B. G., Fearn, T., & Hindle, P. H. (1993). Practical NIR spectroscopy with
applications in food and beverage analysis (2nd ed.). Harlow, New York: Longman
Scientific & Technical, Wiley.
Persano Oddo, L., & Piro, R. (2004). Main European unifloral honeys: Descriptive
sheets. Apidologie, 35, S38S81.
Prieto, N., Roeho, R., Lavn, P., Batten, G., & Andrs, S. (2009). Application of near
infrared reflectance spectroscopy to predict meat and meat products quality: A
review. Meat Science, 83, 175186.
Rambla, F. J., Garrigues, S., & de la Guardia, M. (1997). PLS-NIR determination of
total sugar, glucose, fructose and sucrose in aqueous solutions of fruit juices.
Analytica Chimica Acta, 344, 4153.
Reid, L. M., ODonnell, C. P., & Downey, G. (2006). Recent technological advances for
the determination of food authenticity. Trends in Food Science and Technology,
17, 344353.
Segtnan, V. H., Sasic, S., Isaksson, T., & Ozaki, Y. (2001). Studies on the structure of
water using two-dimensional near-infrared correlation spectroscopy and
principal component analysis. Analytical Chemistry, 73, 31533161.
Tsenkova, R. (2009). Aquaphotomics: dynamic spectroscopy of aqueous and
biological systems describes peculiarities of water. Journal of Near Infrared
Spectroscopy, 17, 303314.
Tsenkova, R., Meilina, H., Kuroki, S., & Burns, D. H. (2009). Near infrared
spectroscopy using short wavelengths and leave-one-cow-out corss-
validation for quantification of somatic cells in milk. Journal of Near Infrared
Spectroscopy, 17, 345351.
White, J. W. J. R. (1980). Detection of honey adulteration by carbohydrate analysis.
Journal of the Association of Official Analytical Chemistry, 66, 1118.
White, J. W., & Doner, L. W. (1978). Mass spectroscopic detection of high-fructose
corn sirup in honey by use of 18C/12C ratio: Collaborative study. Journal of the
Association of Official Analytical Chemistry, 61, 746750.
Williams, P., & Norris, K. (2001). Near-infrared technology in the agricultural and food
industries (2nd ed.). Minnesota, USA: American Association of Cereal Chemists,
Inc.
Workman, J. (2001). Handbook of organic compounds. London: Academic Press.
Workman, J., & Mark, H. (2006). Limitations in analytical accuracy, Part I: Horwitzs
trumpet. Spectroscopy, 21, 1824.
Xantheas, S. S. (1995). Theoretical study of hydroxide ionwater clusters. Journal of
the American Chemical Society, 117, 1037310380.