Professional Documents
Culture Documents
Food Chemistry
journal homepage: www.elsevier.com/locate/foodchem
School of Food and Biological Engineering, Jiangsu university, 301 Xuefu Rd., 212013 Zhenjiang, Jiangsu, China
College of Sciences and Arts-Alkamil, University of Jeddah, P.O. Box 110, Alkamil 21931, Saudi Arabia
c
Department of Food Science & Technology, College of Agricultural Studies, Sudan University of Science & Technology, P.O. Box 71, Khartoum North, Sudan
b
a r t i c l e
i n f o
Article history:
Received 13 September 2015
Received in revised form 5 March 2016
Accepted 10 March 2016
Available online 11 March 2016
Keywords:
Honey
Colorimetric sensor array
SPME-GC/MS
Multivariate analysis
Sensory analysis
PLSR
a b s t r a c t
Aroma profiles of six honey varieties of different botanical origins were investigated using colorimetric
sensor array, gas chromatographymass spectrometry (GCMS) and descriptive sensory analysis.
Fifty-eight aroma compounds were identified, including 2 norisoprenoids, 5 hydrocarbons, 4 terpenes,
6 phenols, 7 ketones, 9 acids, 12 aldehydes and 13 alcohols. Twenty abundant or active compounds were
chosen as key compounds to characterize honey aroma. Discrimination of the honeys was subsequently
implemented using multivariate analysis, including hierarchical clustering analysis (HCA) and principal
component analysis (PCA). Honeys of the same botanical origin were grouped together in the PCA score
plot and HCA dendrogram. SPME-GC/MS and colorimetric sensor array were able to discriminate the
honeys effectively with the advantages of being rapid, simple and low-cost. Moreover, partial least
squares regression (PLSR) was applied to indicate the relationship between sensory descriptors and
aroma compounds.
2016 Elsevier Ltd. All rights reserved.
1. Introduction
More than 600 volatile compounds have been identified in
honey samples. These compounds belong to different chemical
families, including aldehyde, ketone, acid, alcohol, hydrocarbon,
norisoprenoids, terpenes, benzene compounds, esters, furan and
pyran derivatives (Karabagias, Badeka, Kontakos, Karabournioti, &
Kontominas, 2014a; Plutowska, Chmiel, Dymerski, & Wardencki,
2011).
Traditionally, the aroma of honey is determined by sensory
analysis, where 10 or 12 assessors recognise and score welldefined descriptors. However, this method is expensive and
time-consuming (Vaclavik & Christian, 2014). As a substitute,
instrumental techniques, such as solid-phase microextraction couple to gas chromatographymass spectrometry (SPME-GCMS),
can be used to recognise and measure singular aroma components.
Although SPME-GCMS shows good reproducibility, sensitivity
and provides qualitative and quantitative data for these compounds (Cuevas-Glory, Pino, Santiago & Sauri-Duch, 2007), it is relatively high-cost and time-consuming (pnik, Pazitn, ika, &
Szolcsnyi, 2014). Investigation of volatile compounds of honey
up to date has given slight importance to the relationship between
Corresponding author.
E-mail address: zou_xiaobo@ujs.edu.cn (Z. Xiaobo).
http://dx.doi.org/10.1016/j.foodchem.2016.03.032
0308-8146/ 2016 Elsevier Ltd. All rights reserved.
38
DR jRa Rb j
DG jGa Gb j
DB jBa Bb j
DR DG DB
Rb Gb Bb
39
(Mathworks, Natick, MA) in Windows 7. PLSR mapping was performed using Unscrambler version 10. (Camo Software, Oslo,
Norway).
3. Results and discussions
3.1. Colorimetric sensor array analysis
Principal component analysis (PCA) and hieratical cluster analysis (HCA) were applied in the present work. PCA-like methods can
be selected principally for the determination of the correlation
amid data (Gasper, Mijatovic, Bnard, Derenne, Kiss, &
Goormaghtigh, 2010). On the other hand, if we want to study the
clustering of similar data collected from various samples, cluster
analysis must be used (Wang & Mizaikoff, 2008). Related samples
tend to be grouped in the same cluster and the degree of difference
between the groups and the degree of difference between the clusters is stated (Ward Jr, 1963). PLSR was used to find the relationship between sensory analysis results and SPME-GCMS data.
HCA and PCA were performed using Origin Pro version 9.0 (Origin
Lab. Corp., Northampton, MA) and Matlab Version 7.10.0
a
b
H1
H2
H6
H7
H3
H8
H4
H5
H9
H10
d
d
Before exposure
After exposure
Color difference
Fig. 1. Colorimetric maps of six honey varieties (a), plot of the first two principal components by PCA with data acquired from the colorimetric sensor (b), image of the
colorimetric sensor array before exposure, after exposure to honey samples, and the difference image (c) and HCA dendrogram of six kinds of honeys (d).
n.d.
n.d.
0.55 0.00
2.82 0.10
n.d.
0.85 0.10
0.39 0.00
2.08 0.20
2.06 0.20
0.46 0.00
n.d.
n.d.
1.24 0.20
n.d.
0.55 0.00
0.62 0.10
n.d.n.d.
n.d.
3.56 0.0
0.40 0.01
n.d.
n.d.
0.54 0.00
3.21 0.00
n.d.
0.20 0.00
n.d.
3.02 0.20
2.35 0.00
n.d.
1.53 0.20
n.d.
n.d.
n.d.
0.32 0.00
n.d.
0.28 0.00
0.37 0.00
5.53 0.2
1.59 0.10
KI, Kovats index; n.d., not detected.
H9
H8
n.d.
n.d.
0.34 0.00
0.47 0.00
n.d.
0.61 0.00
n.d.
3.97 0.30
1.72 0.00
1.24 0.00
n.d.
n.d.
n.d.
n.d.
n.d.
n.d.
n.d.
n.d.
2.54 0.3
n.d.
n.d.
n.d.
1.33 0.20
4.17 0.00
n.d.
n.d.
n.d.
12.34 1.20
3.47 0.10
1.21 0.00
n.d.
0.40 0.00
n.d.
1.09 0.00
0.43 0.00
n.d.
0.61 0.00
n.d.
1.56 0.1
1.42 0.30
H7
H6
n.d.
0.60 0.00
1.18 0.00
0.97 0.10
n.d.
n.d.
n.d.
16.75 1.30
2.14 0.00
5.22 0.30
1.24 0.10
0.36 0.00
n.d.
0.50 0.10
0.37 0.00
0.48 0.10
2.04 0.10
n.d.
n.d.
1.23 0.00
n.d.
0.98 0.00
1.16 0.10
2.96 0.50
0.88 0.20
0.89 0.10
n.d.
6.54 0.20
5.59 0.20
1.02 0.10
0.63 0.00
1.00 0.10
14.45 1.00
1.86 0.20
0.69 0.10
n.d.
0.52 0.00
1.83 0.30
0.90 0.2
1.67 0.10
H5
H4
n.d.
39.61 1.20
n.d.
0.78 0.00
2.06 0.20
3.17 0.20
6.79 1.10
0.81 0.10
1.92 0.30
n.d.
n.d.
n.d.
n.d.
0.79 0.10
n.d.
n.d.
n.d.
n.d.
4.18 0.4
n.d.
2.33 0.10
n.d.
1.15 0.20
1.91 0.20
3.31 0.00
2.73 0.10
n.d.
14.30 0.50
4.83 0.00
2.91 0.50
n.d.
0.62 0.0
n.d.
0.22 0.00
0.24 0.01
0.10 0.00
1.88 0.10
n.d.
n.d.
1.29 0.04
H3
H2
1.16 0.10
n.d.
0.25 0.00
0.53 0.00
2.24 0.00
3.58 0.00
n.d.
17.73 1.10
2.99 0.00
3.40 0.00
4.00 0.40
n.d.
n.d.
0.61 0.01
0.12 0.00
0.18 0.00
1.74 0.00
n.d.
n.d.
5.54 0.30
1.61 0.10
n.d.
n.d.
n.d.
3.48 0.10
3.55 0.20
n.d.
20.39 1.50
5.82 0.10
4.81 0.00
n.d.
0.57 0.10
n.d.
n.d.
n.d.
0.55 0.01
1.39 0.10
n.d.
n.d.
3.30 0.10
1216
1363
1894
1932
915
918
1086
1472
1536
1587
1664
1573
1677
1857
1189
1272
1516
1668
1618
1717
3-Methyl-1-butanol
1-Hexanol
Benzyl alcohol
Phenyl ethyl alcohol
2-Methylbutanal
3-Methylbutanal
Hexanal
Furfural
Benzaldehyde
5-Methylfurfural
Safranal
2-Methylpropanoic acid
2-Methylbutanoic acid
Hexanoic acid
2-Heptanone
2-Methyldihydro-3(2H)-furanone
2-Acetylfuran
Acetophenone
3,7-Dimethyl-1,5,7-octatrien-3-ol,
4-Oxoisophorone
A1
A2
A3
A4
B1
B2
B3
B4
B5
B6
B7
C1
C2
C3
D1
D2
D3
D4
E1
E2
H1
Compound
Code
of data using the first two PCs is shown in Fig. 1b. The contribution
of the first two PCs was 56.21% of the total variance (PC1 = 39.80%,
PC2 = 16.41%). The data are clearly grouped into ten groups based
on the amount of the volatiles in the honey samples. As we can
see from this figure, the data points of H5 and H6 were slightly
more distributed compared to other honey sample groups. This
could be attributed to the potent volatility of the volatile compounds identified in these samples. It can be seen from Fig.1b that
H8 was positioned between acacia honeys H3 and H9, which
belong to different geographical sources. Besides, overlapping
was found between H3 and H8, which may be due to the similarity
of dyevolatile interactions. Additionally, jujube honeys H1 and H6
which also belong to different geographical areas were successfully
discriminated from other samples. However, the clustering of H1
and H6 in the different clusters may be due to the differences of
dyevolatile interactions. Therefore, the composition and sensory
characteristics of honey differ considerably depending on botanical
and geographical origins. Moreover, multifloral honeys (H4, H5 and
H7) from different locations were clearly discriminated. The results
of the colorimetric sensor array indicated that the volatile compositions of honeys analysed varied considerably.
HCA study average between groups linkage method was used
to explore the data (R, G, and B value of each dot and Euclidean distance) obtained from colorimetric reactions. The HCA reveals the
level of similarity of the array reactions between the different compounds by the datas full dimensionality. Additionally, HCA was
able to identify any misclassification between distinct measurements of components. From Fig. 1d, it can be seen that the samples
were obviously separated into two groups: H5 with high carboxylic
acids in group (A) and others honey samples in group (B). According to Janzen et al. (2006), carboxylic acids exhibited strong
response, while alcohol showed weak response just as the mammalian olfactory system. Examination of Fig.1b and d showed that
reasonable clustering was procured in both PCA and HCA, yet, in
both cases, honeys with the same botanical origin did not cluster
closely with each other. Besides, correct clustering of multifloral
honeys (H4, H5 and H7) from different geographical regions by
HCA further proved the sensor to have excellent performance in
discriminating different honeys. This may be due to the fact that
honey properties and composition depend not only on the
nectar-providing plant species, but also on other factors such as
bee species, mode of storage, and even harvest technology and
conditions. These findings are partly in agreement with results of
GCMS data and sensory evaluation analysis.
H10
Table 1
Percentage composition (%) of volatile components used for discrimination of Sudanese honey samples.
40
PC1
PC2
PC3
A1
A2
A3
A4
B1
B2
B3
B4
B5
B6
B7
C1
C2
C3
D1
D2
D3
D4
E1
E2
0.5253
0.8357
0.1019
0.0796
0.8056
0.7699
0.8333
0.6798
0.1606
0.6375
0.2121
0.0893
0.1561
0.1865
0.0790
0.5987
0.6101
0.1241
0.9110
0.6287
0.5330
0.3849
0.6644
0.8756
0.3930
0.5196
0.3947
0.5895
0.7642
0.0266
0.6658
0.4993
0.5266
0.3325
0.5934
0.4883
0.0895
0.6861
0.0259
0.3574
0.1802
0.0136
0.1022
0.2431
0.2768
0.0747
0.0286
0.1927
0.5696
0.5277
0.3098
0.7923
0.6459
0.7311
0.6579
0.0599
0.3317
0.4362
0.1130
0.0614
and environmental conditions (Anklam, 1998). The results demonstrated critical differences in volatile profiles among honey samples of the same floral source, but from different geographical
areas. For instance, the number of compounds of Jujube honeys
H1 and H6 were found to be different.
3.2.2. GC/MS multivariate analysis
In order to better elucidate the data obtained from SPME-GC/MS,
the 20 abundant volatiles or the active aroma volatiles were manually assimilated and their relative percentages were used for further
multivariate analysis (see Table 1). The first three principal components (PCs) covered 71.33% of the total variance (PC1 = 29.78,
PC2 = 25.87, PC3 = 15.68). Table 2 presented the loadings of the
original variables in the first three PCs. The results indicated that
PC1 positively correlated with furfural, 5-methylfurfural, 2methyldihydro-3(2H)-furanone, 2-furyl methyl ketone and 2,6,6-tri
methyl-2-cyclohexene-1,4-dione; volatiles negatively correlated to
PC1 included 1-hexanol, 2-methylbutanal, 3-methylbutanal, hexanal and 3,7-dimethyl-1,5,7-octatrien-3-ol. Alcohols (benzyl alcohol, phenylethyl alcohol), aldehydes (benzaldehyde, safranal) and
ketones (acetophenone) which are key compounds contributing
to the aroma of honeys showed a positive correlation to PC2, while
41
Fig. 2. (a) HCA dendrogram of different honeys and (b) plot of the first two principal components showing data obtained from GCMS.
42
Fig. 3. Radar graphic presenting sensory profile of six honey verities analyzed.
Fig. 4. PLS regression map showing the relationship between GC/MS data correlated with sensory data in the honeys studied. Code of the aroma compounds
correspond to those in Table 1.
In the current study, we have applied PLSR to display the correlation between the aroma compounds identified by GCMS and
sensory attributes. To correlate aroma compounds and sensory
attributes, the 20 selected volatile compounds (Table 1) were
assigned as independent variables and the six sensory data were
assigned as dependent variables. The PLSR modelling between
selected volatile and sensory attributes was presented in a twofactor model elucidating 92% and 58% of the difference in X (designated volatiles) and in Y (sensory attributes) respectively. The
inner ellipse revealed 50% of the explained variance and the outer
ellipse is the unit-circle representing 100% of the explained variance. Fig. 4 shows the resultant correlation loading plot of the first
two factors. As can be seen from this figure, the aroma attributes
fruity and flowery were closed located in the positive region
of the two dimensions and were positively correlated to phenylethyl alcohol, benzaldehyde, safranal, 2-heptanone, acetophenone
and hotrienol. These outcomes exhibited that not all the sensory
attributes could be explained by the measured aroma compounds.
Negative relationships were found between flowery and fruity
descriptors and 3-methyl-1-butanol, furfural, 2-methyldihydro-3
(2H)-furanone, 2-furyl methyl ketone and 4-oxoisophorone. The
positive and negative correlations suggest that the perception of
an aromatic note is influenced not only by the existence of a few
compounds whose aromas form the note, but also by the presence
of other odorants that affect negatively the perception of such aromatic note (Aznar et al., 2003). As can be seen from Fig. 4, green/
grassy was positioned on the positive site of the first factor
and negative site of the second factor and positively correlated to
1-hexanol, 2-methylbutanal, 3-methylbutanal and hexanal. High
correlations were observed between 1-butanol, furfural, and
2-methyldihydro-3(2H)-furanone, 2-furyl methyl ketone and
4-oxoisophorone. Buttery descriptor was positively correlated
to 2-methylpropanoic acid, and 2-methylbutanoic acid. As these
volatiles and buttery descriptor were positioned close to the
centre, they are not well described by factor 1 and factor 2. In
general, good relationships between sensory assessment and
GC/MS data were found.
4. Conclusions
To the best of our knowledge, this is the first work revealing the
discrimination of honey based on the aroma compounds assessed
with colorimetric sensor array, sensory analysis and GCMS analysis combined with multivariate analysis. The outcomes demonstrated that both the colorimetric sensor array and GCMS
effectively discriminated the honeys. Furthermore, good relationships between GCMS and sensory evaluation were found. There
was a decent compatibility between colorimetric sensor array, sensory analysis and GCMS data, indicating that the colorimetric sensor array has the effectiveness to discriminate honeys from
different botanical origins.
Acknowledgment
We acknowledge the financial support of National Science and
technology support program (2015BAD17B04), Chinese 863
Program, (Grant No. 2011AA00807), NSFC (Grant No. 6091079),
the Jiangsu province science fund for distinguished young
scholars (BK20130010), New century excellent talents in university (NCET-11-0986), Specially-appointed professors by Jiangsu
province.
43