Food Chemistry 272 (2019) 723–731

Contents lists available at ScienceDirect

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

Comparative evaluation of the volatile profiles and taste properties of T

roasted coffee beans as affected by drying method and detected by electronic
nose, electronic tongue, and HS-SPME-GC-MS
⁎ ⁎
Wenjiang Donga, Rongsuo Hua, Yuzhou Longa, , Hehe Lib, , Yanjun Zhanga, Kexue Zhua,
Zhong Chua
a
Spice and Beverage Research Institute, Chinese Academy of Tropical Agricultural Sciences (CATAS), Wanning, Hainan 571533, China
b
Beijing Laboratory for Food Quality and Safety, Beijing Technology and Business University, Beijing 100048, China

A R T I C LE I N FO A B S T R A C T

Keywords: In this study, room-temperature drying, solar drying, heat pump drying (HPD), hot-air drying, and freeze drying
Drying techniques were applied to investigate the volatile profiles and taste properties of roasted coffee beans by using electronic
Roasted coffee beans nose, electronic tongue, and headspace solid-phase microextraction gas chromatography-mass spectrometry (HS-
Electronic nose SPME-GC-MS). Results indicated that the drying process markedly affected pH, total titratable acidity, total
Electronic tongue
solids, and total soluble solids. Significant differences existed among all samples based on drying method; and
HS-SPME-GC-MS
the HPD method was superior for preserving ketones, phenols, and esters. Principal component analysis (PCA)
combined with E-nose and E-tongue radar charts as well as the fingerprint of HS-SPME-GC-MS could clearly
discriminate samples from different drying methods, with results obtained from hierarchical cluster analysis (the
Euclidean distance is 0.75) being in agreement with those of PCA. These findings may provide a theoretical basis
for the dehydration of coffee beans and other similar thermo-sensitive agricultural products.

1. Introduction and deterioration, including browning, softening, and rotting owing to

In recent decades, with the development of the economy, social and
cultural awareness, and an increase in standards of living, consumers
have exhibited increased interest in healthy and well-flavoured foods
and have begun to search for diets with health-promoting properties as
well as those reducing potential disease risk. Beverages comprise an
important component of the daily diet, with coffee being a very popular
brewed beverage that is consumed worldwide; moreover, global coffee
consumption has also increased at an average annual rate of 1.9% over
the past 50 years. Coffee, which is characterised by a pleasant taste
sensation comprising a balanced combination of bitterness, sourness,
nuttiness, and astringency, is derived from a tropical evergreen shrub of
the genus Coffea, which includes the two most common cultivated
species, Coffea canephora and Coffea arabica (Cheng, Furtado, Smyth, &
Henry, 2016). In China, coffee is mainly distributed in Hainan and
Yunnan provinces; robusta coffee, recognised by its appearance and
earthy flavour, is cultivated in Hainan province and Arabica coffee,
having the characteristics of sweet-caramel aroma and meta-acid, is
cultivated in Yunnan province (Dong, Tan, Zhao, Hu, & Lu, 2015).
However, fresh coffee fruits are susceptible to postharvest pathogens
inherent enzymes, if processing is not performed in a timely manner,
which severely influences the composition of coffee beans and reduces
their commercial value. Drying represents one of the most effective
procedures to address this issue and has promoted the development of a
highly efficient coffee industry. Therefore, it is necessary to explore the
effect of drying techniques on the sensory quality of roasted coffee
beans, as the biochemical composition of green coffee beans is decisive
for flavour quality through roasting-induced chemical reactions.
Drying technology, as an effective method to remove water from
foods and to reduce water activity, plays an important role in the fields
of food manufacturing and food processing worldwide. Nevertheless,
drying can affect the major quality parameters associated with dried
food products, such as colour, flavour, microbial load, texture, taste,
aroma, and nutritional properties (Zhang et al., 2017). As volatile
compounds are associated with coffee flavour and consequently impact
its acceptance, the aroma component of flavour for coffee is regarded as
especially important and contributes to flavour diversity, whereas
drying could lead to the change of volatile profile and the formation of
novel characteristic flavours. Studies conducted to investigate the vo-
latile compounds in roasted coffee beans (Toledo, Pezza, Pezza, & Toci,

⁎
Corresponding authors.
E-mail addresses: longyuzhou6090@126.com (Y. Long), xyzhehe@126.com (H. Li).

https://doi.org/10.1016/j.foodchem.2018.08.068
Received 7 June 2018; Received in revised form 16 August 2018; Accepted 16 August 2018
Available online 17 August 2018
0308-8146/ © 2018 Elsevier Ltd. All rights reserved.
W. Dong et al.
2016) have identified more than 800 volatile compounds, although
previous studies have revealed that only about 20–30 volatile com-
pounds, including sulfur compounds, aldehydes, ketones, furans, pyr-
roles, and pyrazines, are predominantly responsible for the coffee
aroma. Notably, although some volatiles have low odour thresholds and
concentration, they could lead to a greater olfactory contribution to
coffee flavour (Yang et al., 2016).
Several prior studies have examined the effect of drying procedures
utilising solar drying, thin layer drying, hot air drying (HAD), freeze
drying (FD), and fluidised drying on the quality of coffee beans and
related products. Burmester and Eggers (2010) reported that the heat
and mass transfer of the coffee drying process, along with optimisation
of the industrial application transport coefficients and coffee properties,
thermal conductivities, and effective diffusion coefficients were a
function of the moisture content as well as the volume and densities of
the coffee beans. Nilnont et al. (2012) demonstrated that a mathema-
tical model could accurately simulate the moisture contents in different
components of parchment coffee and effectively explain transport
processes. Suzihaque and Driscoll (2016) studied the effects of solar
radiation, buoyancy of air flow, and optimisation of coffee drying in a
heat recovery dryer, from which a sensitivity analysis of the profit
equation showed that the tray area, coffee bean density, and the price of
labour had the most effect on profit. Additionally, Burmester, Fehr, and
Eggers (2011) examined the thermophysical material properties of an
innovative coffee drying process. In our previous study, different drying
techniques; i.e. room temperature drying (RTD), solar drying (SD), heat
pump drying (HPD), HAD, and FD were applied to investigate the effect
on bioactive components, fatty acid composition, and volatile profile of
green coffee beans (Dong, Hu, Chu, Zhao, & Tan, 2017). Considering
the drying efficiency, good quality retention, and production cost, HPD
exhibited the broadest market prospect for potential application (Dong,
Hu et al., 2017). In particular, drying was considered to constitute a key
factor contributing to a high-quality coffee. However, to the best of our
knowledge, no study has reported on the comparative evaluation of the
profiles of volatile compounds of roasted coffee beans originated from
different drying techniques.
Over the past few years, gas chromatography-mass spectroscopy
(GC–MS), electronic nose (E-nose), and electronic tongue (E-tongue)
have been increasingly applied in the food industry for quality control
(Peris & Escuder-Gilabert, 2016; Huang, Guo, Yuan, Luo, & Yue, 2015).
In particular, solid-phase microextraction (SPME) represents an espe-
cially promising sample preparation technique in food analysis for the
pre-processing procedures of GC–MS, which could provide an accurate
approach for the analysis of type and concentration of volatile com-
ponents, as well as serving as a useful sensorial tool for food flavour
analysis. E-nose and E-tongue are analytical devices consisting of an
array of sensors utilised to detect odorants and tastants. Recently, HS-
SPME-GC-MS, E-nose, and E-tongue were applied to analyse the volatile
and taste compounds of food matrices for, e.g., quality control, fresh
grading, storage conditions, and authentication and differentiation, of
foods, such as vegetables, fruits, tea and coffee, milk, wine, fish, and
meat (Rosa, Leone, Cheli, & Chiofaloa, 2017). Moreover, in our pre-
vious work we found that in the application of E-nose and E-tongue
sensors combined with chemometric multivariate analysis to dis-
criminate seven robusta coffee cultivars, an acceptable performance of
a calibration model could be obtained (Dong, Zhao, Hu, Dong, & Tan,
2017). Nevertheless, these technologies have yet to be applied to assess
the differences between roasted coffee bean samples prepared using
different drying processes.
Here, we hypothesise that different traditional drying processes
differentially impact the volatile profile and taste properties of roasted
coffee beans, and that these differences can be effectively detected
through the use of statistical modelling of the data obtained from the
newly-developed measurement technologies. The objective of the pre-
sent study was therefore to evaluate and compare the odour com-
pounds, aromatic profile, and taste properties of roasted coffee beans
724
Food Chemistry 272 (2019) 723–731
prepared from different drying processing methods (RTD, SD, HPD,
HAD, and FD). HS-SPME-GC-MS, E-nose, and E-tongue techniques were
then utilised to analyse the flavour of coffee samples, and principal
component analysis (PCA) and hierarchical cluster analysis (HCA) were
applied on the data matrix obtained by HS-SPME-GC-MS, E-nose, and E-
tongue, as well as the three-dimensional fusion matrix, to explore fla-
vour differences among various roasted coffee samples.
2. Materials and methods
2.1. Chemicals and reagents
All reagents or solvents were analytical, high-performance liquid
chromatography (HPLC), or GC–MS grade. HPLC grade methanol and
formic acid were supplied by Merck (Darmstadt, Germany). Sodium
hydroxide and hydrochloric acid were purchased from Damao Chemical
Reagent Factory (Tianjin, China). An SPME manual holder and 75 m
Carboxen/polydimethylsiloxane (CAR/PDMS) were obtained from
Supelco (Bellefonte, PA). The standard of C7-C40 saturated alkanes was
provided by Sigma-Aldrich (St. Louis, MO). Milli-Q water was obtained
from a Milli-Q Plus water system (Millipore, Hetai, Shanghai, China).
2.2. Samples and sample preparation
Fresh coffee fruits (C. canephora, Xinglong robusta coffee, ‘Reyan1’)
were harvested from the experimental station in January 2015, at the
Spice and Beverage Research Institute, Chinese Academy of Tropical
Agricultural Sciences, in Hainan, which is one of the famous coffee
growing regions of China. A total of fifteen samples were picked in the
experimental station under the same climate and planting conditions,
and five random samples were merged into one sample to make the
samples more representative; thus three merged samples were used for
the following experiment. All fruits were sampled at a mature stage
with uniform size and colour (red). In this study, all fresh coffee fruits
were processed by a dry process method, being dried until the moisture
content of samples was 11.0 ± 1.0 g/100 g dry weight (DW) when
subjected to different drying techniques. Then, a huller was used to
remove the shell of fruits to obtain green coffee beans for the sub-
sequent experiments.
2.3. Drying and roasting procedures
The fresh coffee fruits were split into five equal portions for the
subsequent drying processing; the final water content of green coffee
beans was 11.0 ± 1.0 g/100 g DW. (a) RTD: The first portion of fresh
coffee fruit was dried using natural air flow at room temperature of
approximately 25 °C for seven days. (b) SD: The second portion of fresh
coffee fruit was dried in the SD chamber for five days. (c) HPD: The
third portion of fresh coffee fruit was dehydrated using a dehumidified
heat pump (RF-25II; Guangzhou Quanneng Energy Technology Co.,
Ltd., Guangzhou, China) at 50 °C for 72 h and 70% relative humidity.
(d) HAD: The fourth portion of fresh coffee fruit was dehydrated using
an electro-thermal blowing dry box (DHG-9030A, Shanghai Yiheng
Laboratory Instrument Co., Ltd., Shanghai, China) at 50 °C for 108 h. (e)
FD: The fifth portion of fresh coffee fruit was dried in a freeze dryer
(JDG-0.2, Lanzhou Kejin Freeze Drying Instruments Co., Ltd., Lanzhou,
China) under 40 Pa vacuum pressure for 25 h. A detailed description of
drying procedures is provided in Dong, Hu et al. (2017). The dried
green coffee beans were then used for the roasting step. Samples of
100 g were taken from each portion and roasted using a PRE 1 Z drum
coffee roaster (Emmerich am Rhein, Germany). The initial roasting
temperature was 180 °C, after which the roasting temperature varied
from 180 °C to 200 °C to medium degree for 10 min. Coffee roasting
conditions such as time and temperature were defined by the colour
index of L*. All samples were ground using the ‘fine’ option of an
electronic coffee grinder (Kenia, Mahlkönig, Germany) and the ground
W. Dong et al.
Fig. 1. Total soluble solids content (a), total solids content (b), pH value (c), and total titratable acidity (d) of roasted coffee beans prepared by different drying
methods.
coffee was sieved using a standard sieve to exclude large particles.
2.4. Physicochemical analysis
The crude extract for physicochemical analysis was prepared by
mixing 8.25 g coffee bean powder with 150 mL distilled water in a hot
water bath (95 °C, 5 min) with stirring; after extraction, the extract was
filtered through Whatman filter paper and the filtrate was stored at
4 °C. The physicochemical parameters, total soluble solids (TSS), total
solids (TS), pH, and total titratable acidity (TA) were determined ac-
cording to the methods of Gloess et al. (2013) with minor modifica-
tions. The TSS content of the coffee samples was measured in °Brix
using an analogue refractometer (PAL-A; Atago, Tokyo, Japan). For TS,
expressed as g/100 g DW, 4.00 g sample were accurately weighed and
dried in a well-ventilated oven at 105 ± 2 °C until reaching a constant
weight. The pH value was quantified using a pH meter (FE-20; Mettler-
Toledo, Greifensee, Switzerland). The TA, as milliequivalents of citric
acid per 100 g dry matter, was determined by titration with 0.1 M
NaOH solution up to pH 8.1 until the pink end-point using phe-
nolphthalein indicator.
2.5. E-nose analysis
The E-nose analyses were performed using a commercial Fox 3000
electronic nose (Alpha M.O.S., Toulouse, France). The structure of E-
nose consists of pattern recognition software, HS-100 autosampler, and
a sensor array unit. The sensor array is composed of six metal oxide
semiconductor sensors, T30/1, T70/2, PA/2, LY2/AA, and LY2/gCT to
725
Food Chemistry 272 (2019) 723–731
detect sample gas. Coffee powder sample (1 g) was put into a 10-mL vial
and incubated for 20 min at 60 °C (300 rpm). Processed pure air was
used as carrier gas to clean the sensor array to cause the signal response
to return to zero; then the sample was measured using an injection
volume of 300 µL. Each sample was measured in triplicate, and the
mean value was applied for further analysis.
2.6. E-tongue analysis
Coffee taste attributes were analysed using an Astree II potentio-
metric electronic tongue (Alpha M. O. S., Toulouse, France). This E-
tongue system was comprised of an LS48 automatic sampler, sensor
array, reference electrode (Ag/AgCl), and a Chemometric software
package. The E-tongue contained seven chemical sensors: ZZ, JE, BB,
CA, GA, HA, and JB, which are potentiometric sensors with an organic
membrane coating with specific sensitivity and selectivity for each
sensor. Prior to analysis, 0.01 M hydrochloric acid, sodium chloride,
and sodium-L-glutamate were used to calibrate and diagnose the sen-
sors and system. The instrument parameters were as follows:
delay = 0 s, acquisition time = 120 s, interval = 1 s, stirring rate = 1,
and acquisition period = 1. Between each measurement, 80 mL deio-
nised water were used to clean the sensors, and the sensor signals were
acquired using Astree software v.3.0.1.
2.7. Volatile compounds analyses
The volatile compounds of roasted coffee samples were determined
using HS-SPME-GC-MS. An accurately weighed 1.0 g sample was filled
W. Dong et al. Food Chemistry 272 (2019) 723–731

Fig. 2. E-nose responses for representative roasted coffee beans (a), and mean values of seven sensors (b) for coffee samples. PCA of E-nose data for coffee samples
with different drying methods (c: score plot, d: loading plot). (RTD ( ), SD ( ), HPD ( ), HAD ( ), FD ( )).

into 20-mL screw vials fitted with polytetrafluorethylene-silicone septa;
after incubation at 60 °C for 20 min, the volatiles were then sampled
with a 75-µm CAR/PDMS fibre supplied by Supelco (Bellefonte, PA) for
40 min. After extraction, the fibre was immediately inserted into the
injection port of the GC–MS for the desorption step at 250 °C for 3 min.
GC–MS analyses were carried out on an Agilent 7890A GC 5975C
system coupled with a quadrupole mass filter for mass spectrometric
detection (Agilent Technologies, Santa Clara, CA). The chromato-
graphic separation was achieved on a DB-WAX capillary column
(30 m × 0.25 mm, df = 0.25 m). Analytical conditions were as follows:
the injector and transfer line temperatures were 230 and 250 °C, re-
spectively; the oven program began with a 40 °C hold for 2 min, was
raised to 130 °C at 1.5 °C/min, then raised to 220 °C at 4 °C/min and
held for 5 min. Helium was used as the carrier gas, with the flow of He
at 1 mL min−1, and detection was performed in full scan mode over the
mass range m/z 35–300 (Dong et al., 2015). The qualitative identifi-
cation of compounds was achieved by comparison of their mass spectra
and retention times to those of pure standards, as well as calculating the
retention indices (RI) and matching against the NIST MS Search pro-
gram, version 08. The quantification of compounds was achieved by
integrating the selected specific ions to obtain relative peak areas.
2.8. Data analysis and chemometric procedures
Statistical analysis was performed through one-way analysis of
variance (ANOVA), followed by Duncan’s multiple test to identify sig-
nificant differences among roasted coffee samples from different drying
techniques using SPSS version 20.0 software (SPSS Inc., Chicago, IL); all
statistical tests were two-tailed and p < 0.05 or less was considered as
significant. Chemometrics methods, PCA, and HCA were operated
under the Matlab software platform (v. R2010a Mathworks Inc., Natick,
MA) environment. All data are expressed as the means ± SD (standard
deviation) of triple measurements.
3. Results and discussion
3.1. Effect of drying methods on the TSS and TS content of roasted coffee
beans
The TSS and TS content of the roasted coffee samples from different
drying methods is shown in Fig. 1a–b. The TSS content of all samples
ranged between 1.93 and 2.50 °Brix, RTD samples had the lowest TSS
content (1.93 °Brix), followed by FD (2.01 °Brix), with the HPD and
HAD methods resulting in almost the same TSS content (2.31 °Brix and
2.30 °Brix, respectively), whereas SD samples had a significantly
(p < 0.05) higher TSS content than that of other samples, which was in
accordance with our previous study (Dong, Zhao et al., 2017). In
comparison, Gloess et al. (2013) investigated the effects of nine
common extraction methods on the TSS content of coffee, which ranged
from about 0.40 to 6.40 °Brix. The broad range observed was likely due
to the different brewing mechanisms, although most were similar to our
results. The TS content was mainly related to the body; i.e. mouthfeel or
texture, which constitutes an additional important sensory descriptor
for coffee. In the present study, the levels of TS content in samples
ranged from 1.92 to 2.73 g/100 g DW, with a statistical significant
difference (p < 0.05) being observed in the five investigated types of
roasted coffee samples, except between HAP and HAD, and RTD and FD
samples. The SD method was considered to allow better retention of TS

726
W. Dong et al.
Fig. 3. E-tongue responses for representative roasted coffee beans (a), and mean values of seven sensors (b) for coffee samples. PCA of E-tongue data for roasted
coffee samples with different drying methods (c: score plot, d: loading plot). (RTD ( ), SD ( ), HPD ( ), HAD ( ), FD ( )).
for coffee samples.
3.2. Effect of drying method on the pH and TA of roasted coffee beans
The pH value and TA of the roasted coffee samples from different
drying techniques are presented in Fig. 1c–d. Perceived acidity re-
presents one of the key perceptions within the coffee organoleptic
profile, as well as being considered as a particularly important criterion
for a good cup of coffee. In the current study, the pH value of the
samples ranged from 5.54 to 5.70, among which, the HAD samples had
the highest pH value (5.70), whereas SD samples had the lowest (5.54).
The pH value showed statistically significant differences (p < 0.05)
among all different dried roasted coffee samples. The TA of coffee ex-
tract was in the range of 1.95–2.15 mL of 0.1 M NaOH, whereas the TA
of SD, RTD, and FD samples was higher than that of HPD and HAD
samples, which may be related to the formation of acidic compounds
during heat treatment in the HPD and HAD methods. Caporaso,
Genovese, Canela, Civitella, and Sacchi (2014) reported that the pH
values of Arabica coffee brews by Neapolitan, Espresso, American, and
Moka techniques varied from 5.52 (Neapolitan coffee) to 6.01 (Amer-
ican coffee), which is in accordance with the results of our study. Gloess
et al. (2013) reported similar results for the TA after nine common
extraction methods, which indicated Nespresso samples as being the
most acidic brew (pH = 5.41); although French press (pH = 5.92) was
somewhat higher, the different roasting degree appeared to have no
substantial influence on pH value.
727
Food Chemistry 272 (2019) 723–731
3.3. E-nose
3.3.1. E-nose response to roasted coffee samples
The electronic nose was a better method for analysing aroma, as it
could acquire complete information related to volatile compounds in
samples rather than qualitative and quantitative results (Tiggemann
et al., 2017). The typical E-nose responses to the roasted coffee samples
are shown in Fig. 2a, where G/Go is used as the response value, i.e., the
ratio between sample gas and pure air resistivity. Each curve represents
the variation trend of a corresponding sensor with increasing time until
a stable equilibrium is finally reached. It can be observed that the re-
sponse values of sensors P30/2, PA/2, T70/2, and T30/1 increased; in
contrast, the response values of sensor LY2/AA decreased, whereas the
sensor LY2/gCT remained almost unchanged in the first 0 to 15 s.
Subsequently, the response value of sensor P30/2 increased drastically
and reached a stable equilibrium, and the response value of sensors PA/
2, T70/2, T30/1 decreased gradually until reaching a steady value.
However, the response value of sensor LY2/AA increased gradually in
the interval of 20–90 s.
The typical responses of the six sensors of the volatile compounds in
roasted coffee samples from different drying techniques are presented
in Fig. 2b. As can be observed, the variation trends of signals from
different samples were similar; the responses for sensors P30/2 and PA/
2 were significantly higher than those of the other sensors, and the
responses of P30/2 for HAD, HPD, and FD samples were significantly
different from those of the other dried samples. Moreover, the relative
values of T30/1, T70/2, and LY2/AA were in the intermediate value,
and significant differences were also observed between HPD, HAD, and
FD samples, RTD, and SD samples. Notably, the responses of sensors in
W. Dong et al.
Fig. 4. Total ion chromatogram of volatiles from roasted coffee samples with different drying methods (a). Relative content of volatile compounds among different
classes in coffee samples (b).
our study were in accordance with the report of Severini, Ricci, Marone,
Derossi, and De Pilli (2015), in which E-nose was applied to analyse and
discriminate the changes in chemical attributes and aromatic profile of
espresso coffee as a result of the extraction time and grinding level. The
differences may have resulted from the formation of novel volatile
compounds and other similar volatile compounds among the various
drying conditions (Yang et al., 2016). However, it is difficult to dis-
criminate different dried samples only from observed sensor response
values; thus, the signal information of the six sensors needs to be further
processed and its internal relationships and differences explored.
3.3.2. PCA of volatile compounds in roasted coffee samples
PCA constitutes a statistical tool used to explain differentiation
between samples and to extract information from the variables that
mainly affects the sample spatial distribution. This procedure can re-
duce the dimensionality of the data matrix while retaining most of the
information of the original data, as well as explain the relationship of
728
Food Chemistry 272 (2019) 723–731
objects and the correlation structure of the variables (Kaya, Yidiz, &
Ünlütürk, 2015)). PCA was performed on the dataset of the response
values of E-nose to evaluate the influence of drying processing on the
grouping of the roasted coffee samples. The two-dimensional bi-plots of
score and loading of the dried samples are presented in Fig. 2c–d. The
principal components (PCs) PC1 and PC2 represented 48.5% and 37.1%
of the total variance, respectively, with the cumulative contribution
rate of the first two PCs accounting for 85.6%, which indicates they are
sufficient to explain the total variance in the dataset (Wold, Esbensen, &
Geladi, 1987). It can be observed that grouping of the samples was
clearly apparent in the bi-plot of PCA, with the RTD and HAD samples
being located to the right of the X-axis, whereas the SD, HPD, and FD
samples were located to the left of the X-axis, among which a slight
overlap was observed between SD and HPD groups. Considering the
loading plot demonstrating that the relationship between the E-nose
variables P30/2, LY2/AA, and LY2/gCT had a higher influence on the
differentiation ability, the sensor P30/2 likely contributed more to the
W. Dong et al.
Fig. 5. PCA bi-plot for coffee samples using the volatiles matrix as input vari-
ables (a: score plot, b: loading plot). (RTD ( ), SD ( ), HPD ( ), HAD ( ), FD
( )).
HAD samples and the sensor LY2/AA contributed more to the FD
samples, whereas the response values of LY2/gCT, T70/2, and PA/2
were more correlated with the SD and HPD samples.
3.4. E-tongue
3.4.1. E-tongue response to roasted coffee samples
The mechanism of E-tongue was similar to that of E-nose, being able
to detect taste profiles by electronic sensors to mimic taste perception in
humans. A representative sample of the E-tongue sensor response to a
roasted coffee sample is presented in Fig. 2a. Each curve represents the
corresponding potentiometric difference value of a sensor against time.
At first, all sensors (HA, ZZ, JB, GA, JE, BB, and CA) retained their
initial values, then the responses of sensors GA and JE changed sig-
nificantly in the first 10 s, decreasing and increasing, respectively,
whereas the other sensors, HA, ZZ, JB, BB, and CA, changed relative
slowly. From 10 to 120 s, the response values of all sensors varied
slowly and reached a dynamic equilibrium. The signal values of the
120th s of each sensor were used for further statistical analysis. In ad-
dition, ANOVA was utilised to analyse the significant difference of
sensors from various types of roasted coffee samples. Fig. 2b shows the
characteristic signal response of the tested samples, the result indicates
a high stability of each sensor, with a significant difference being
729
Food Chemistry 272 (2019) 723–731
observed between different dried samples. Overall, all samples showed
a similar changing trend for the seven sensors albeit with different re-
sponse intensity. E-nose has also been used in the quality control and
taste characterisation of food chemistry, such as by Lopetcharat et al.
(2016), who provided a new approach to correlate between overall
difference decision and E-tongue results as well as to discriminate civet
coffee from different origins, obtaining a satisfactory performance. In
addition, E-nose was also used to characterise espresso coffees brewed
with different thermal profiles, thus providing useful information for
the optimisation of processing conditions (Buratti, Benedetti, &
Giovanelli, 2017).
3.4.2. PCA of taste compounds in roasted coffee samples
PCA was also performed on the E-nose dataset to distinguish roasted
coffee samples from different drying techniques, taking the response
values of sensors as input variables. Fig. 3c–d shows the score and
loading plots of the first and the second components, PC1 and PC2, in
the two-dimensional PC space, with the first two PCs explaining 79.0%
of the total variance, at 58.0% and 21.0% for PC1 and PC2, respec-
tively. It can be observed that the samples exhibit a tendency to form
three major groups in the score bi-plot of PCA. A group comprised of SD
and HPD samples was located close together in the first quadrant except
for one HPD sample that was in the Y-axis; the RTD processed samples
were all in the fourth quadrant with positive and negative score values
at PC1 and PC2, whereas the HAD and FD samples were all almost
located in the third quadrant of the graph except for one FD sample.
Fig. 3d represents loading plots of different kinds of roasted coffee
samples, the analysis of which can explain which sensor combinations
were responsible for the distribution and grouping of the samples in the
score plot. The HAD and FD samples were differentiated by CA with
negative score values at PC1 and PC2; the GA and JE were mainly re-
sponsible for the sample discrimination in the direction of PC1, whereas
BB, HA, and ZZ contributed more to the separation along with the di-
rection of PC2. In comparison, in a previous study regarding E-tongue
combined with PCA, we reported that when Chinese robusta coffee
samples from different cultivars were discriminated using E-tongue
data with the aid of chemometrics, a satisfactory performance was
achieved (Dong, Zhao et al., 2017).
3.5. HS-SPME-GC-MS
3.5.1. HS-SPME-GC-MS analysis of volatile components in roasted coffee
samples
The volatile profile of coffee constitutes a very important organo-
lepic characteristic for the acceptance and preference of high quality
coffee brew. In order to evaluate the effect of drying methods on the
flavour of roasted coffee beans, HS-SPME-GC-MS was used to analyse
the volatile compounds of each sample afforded by different drying
methods. The total ion current chromatogram of all samples with dif-
ferent drying methods and the relative content of volatile compounds
from different classes in coffee samples are shown in Fig. 4a, b, and the
detailed volatile composition of samples is listed in Table 1 in the
Supplementary material. A total number of 133 volatile compounds
were detected in roasted samples. The volatile compounds were clas-
sified into 11 groups based on the chemical properties, including 11
furans, 19 pyrazines, 31 ketones, 9 aldehydes, 9 phenols, 7 pyrroles, 7
pyridines, 5 esters, 8 acids, 7 alcohols, and 18 others (Table 2,
Supplementary material). Sunarharum, Williams, and Smyth (2014)
also reviewed that coffee volatile compounds comprise several classes,
which was in accordance with our present study. Although there was no
significant difference in the qualitative profile of roasted coffee samples
prepared from different drying methods, an ANOVA analysis performed
to compare the peak area of volatile compounds of different classes
indicated that a significant difference could be observed among the
majority of these compounds. (See Fig. 5).
The difference in degree of change of volatile compounds between
W. Dong et al.
Fig. 6. Dendrogram obtained by HCA of roasted coffee samples with a dataset comprising combined E-nose, E-tongue, and HS-SMPE-GC-MS data.
the different drying methods possibly arose from the different drying
mechanism and dehydration behaviour, as the concentrations of furans
and pyrazines were the highest when compared with those of other
classes. Approximately 111, 109, 102, 93, and 90 kinds of volatile
compounds were identified in samples from the RTD, SD, HPD, HAD,
and FD methods, respectively. For RTD and SD samples, a total 11 kinds
of furans were detected, with the concentration of furans in particular
being highest (25.46%) in the RTD sample, with the 2,2′-methylenebis-
furan being only found in this sample. 2-Furanmethanol and furfural
contribute substantially to the furans of coffee samples; furthermore, 2-
furanmethanol accounted for about 19.0% of total volatile compounds.
A total of 19 kinds of pyrazines were detected, which contribute to the
toasty and earthy flavour, with methylpyrazine and dimethylpyrazine
accounting for 5.5% and 10.0%, respectively. In HPD and HAD drying,
the number of volatiles decreased when compared to that of RTD and
SD drying; this was attributed to long exposure to high temperature air
resulting in the loss of flavour precursors, although we found that the
HPD method was superior in preserving ketones, phenols, esters, and
other volatiles.
FD drying decreased the varieties of total volatile compounds sig-
nificantly. Although previous research reported that FD methods could
retain the volatiles to the greatest extent, our studies demonstrated that
the long vacuum treatment of FD obviously affected the variety and
quantity of roasted coffee samples. In general, the postharvest dehy-
dration is one of the most important factors in regulating the formation
and release of volatile compounds. Specifically, there are four pathways
to generate volatiles during the drying process: Maillard reaction, lipid
oxidation and degradation, interaction between oxidised lipids and
amino acids, and long-chain compound degradation (Yang et al., 2016;
Deng, Luo, Wang, & Zhao, 2015). The drying mechanisms and path-
ways that led to the various kinds and quantity of the volatiles from
roasted coffees identified in the present study should be further in-
vestigated in future studies.
3.5.2. PCA of HS-SPME-GC-MS analysis
PCA was performed on the data matrix (15 samples × 11 volatile
classes) to observe possible sample distribution according to different
drying methods. The PCA plot presented in Fig. 2a, b explains 70.9% of
the total cumulative variance contribution. PC1 and PC2 account for
49.8% and 21.1% of the total variance, respectively, as shown in the
score plot of PCA (Fig. 2a), with roasted samples clustering into five
730
Food Chemistry 272 (2019) 723–731
distinct groups in the dimensional graph based on the PC1 × PC2 score.
The HAD and FD processed samples were all located in the fourth
quadrant of the graph, whereas samples of HPD were in the upper
quadrant and most RTD and SD samples localised in the third quadrant
of the graph except for one sample from SD treatment. Fig. 2b shows the
relationship of dried samples with volatile compounds. To identify
which compounds contribute to the class separation of the different
groups, it can be observed that the esters, aldehydes, and acids are
highly and positively related to PC1, and mainly distinguish HAD and
FD samples from other dried samples. In addition, phenols have a high
score in the positive direction of PC2, which contributes greatly to the
separation of HPD samples. Finally, RTD and SD samples are char-
acterised by the highest content of pyrazines and alcohols, which were
highly and negatively correlated with PC1. Overall, these results in-
dicate that the three analytical instruments, E-nose, E-tongue, and HS-
SPME-GC-MS could be used for the quality evaluation of roasted coffee
beans from five drying methods, and that the results were consistent
across platforms.
3.6. HCA of the fusion dataset of E-nose, E-tongue, and HS-SPME-GC-MS
In order to confirm the results obtained from PCA, HCA (Granato,
Santos, Escher, Ferreira, & Maggio, 2018) was applied to the fused
dataset comprised of E-nose, E-tongue, and HS-SPME-GC-MS data,
employing the Ward’s linkage method and Euclidean distances to cal-
culate the interpoint distances between samples dried by different
methods. The dendrogram obtained, shown in Fig. 6, indicates the re-
lationships and distribution patterns among the roasted coffee samples
with different drying methods. The dendrograms clearly showed that
the 15 tested samples were clustered into two major groups. Group A
consisted of six samples (HAD and FD), group B consisted of nine
samples (RTD, SD, and HAD), and samples with the same drying modes
were closely clustered but fully discriminated from each other in the
dendrogram. These classification groups were consistent with the PCA
results, in which all roasted coffee samples were fully distinguished
according to their drying methods.
4. Conclusion
In this study, the effect of drying processes (RTD, SD, HPD, HAD,
and FD) on the aromatic profile and taste properties of dried coffee
W. Dong et al.
beans was studied. Fingerprint analysis, E-nose, E-tongue, and HS-
SPME-GC-MS, coupled with chemometrics methods, including PCA,
HCA, and ANOVA, were applied to the single or fused dataset to eval-
uate the quality of roasted coffee beans. Consistent with our hypothesis,
we found that the drying process had a marked effect on pH, TA, TS,
and TSS, and significant differences were found among samples with
different drying methods. In addition, the results indicated that the HPD
method was superior in preserving ketones, phenols, and esters.
Furthermore, PCA combined with E-nose, E-tongue, and HS-SPME-GC-
MS could effectively discriminate roasted coffee samples derived from
different drying processes. HCA coupled to the fused data matrix
achieved similar results when compared with PCA. In summary, our
study supports and encourages the application of this method for
roasted coffee sample discrimination and the establishment of similar
methods for other heat-sensitive agricultural products.
Acknowledgments
This work was supported by the National Natural Science
Foundation of China (31501404, 31872888), the Fund of the Beijing
Laboratory for Food Quality and Safety, Beijing Technology and
Business University (FQS-201806), and the Fund of Central Public-in-
terest Scientific Institution Basal Research of CATAS (1630142017005).
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in the
online version, at https://doi.org/10.1016/j.foodchem.2018.08.068.
References
Buratti, S., Benedetti, S., & Giovanelli, G. (2017). Application of electronic senses to
characterize espresso coffees brewed with different thermal profiles. European Food
Research and Technology, 243, 511–520.
Burmester, K., & Eggers, R. (2010). Heat and mass transfer during the coffee drying
process. Journal of Food Engineering, 99, 430–436.
Burmester, K., Fehr, H., & Eggers, R. (2011). A comprehensive study on the themophy-
sical material properties for an innovative coffee drying process. Drying Technology,
29, 1562–1570.
Caporaso, N., Genovese, A., Canela, M. D., Civitella, A., & Sacchi, R. (2014). Neapolitan
coffee brew chemical analysis in comparison to expresso, moka and American brews.
Food Research International, 61, 152–160.
Cheng, B., Furtado, A., Smyth, H. E., & Henry, R. J. (2016). Influence of genotype and
environment on coffee quality. Trends in Food Science and Technology, 57, 20–30.
Deng, Y., Luo, Y. L., Wang, Y. G., & Zhao, Y. Y. (2015). Effect of different drying methods
on the myosin structure, amino acid composition, protein digestibility and volatile
profile of squid fillets. Food Chemistry, 171, 168–176.
Dong, W. J., Hu, R. S., Chu, Z., Zhao, J. P., & Tan, L. H. (2017). Effect of different drying
techniques on bioactive components, fatty acid composition, and volatile profile of
731
Food Chemistry 272 (2019) 723–731
robusta coffee beans. Food Chemistry, 234, 121–130.
Dong, W. J., Tan, L. H., Zhao, J. P., Hu, R. S., & Lu, M. Q. (2015). Characterization of fatty
acid, amino acid and volatile compound compositions and bioactive components of
seven coffee (Coffea robusta) cultivars grown in Hainan province, China. Molecules,
20, 16687–16708.
Dong, W. J., Zhao, J. P., Hu, R. S., Dong, Y. P., & Tan, L. H. (2017). Differentiation of
Chinese robusta coffees according to species, using a combined electronic nose and
tongue, with the aid of chemometrics. Food Chemistry, 229, 743–751.
Gloess, A. N., Schönbächler, B., Klopprogge, B., D'Ambrosio, L., Chatelain, K., Bongartz,
A., et al. (2013). Comparison of nine common coffee extraction methods:
Instrumental and sensory analysis. European Food Research and Technology, 236,
607–627.
Granato, D., Santos, J. S., Escher, G. B., Ferreira, B. L., & Maggio, R. M. (2018). Use of
principal component analysis (PCA) and hierarchical cluster analysis (HCA) for
multivariate association between bioactive compounds and functional properties in
foods: A critical perspective. Trends in Food Science and Technology, 72, 83–90.
Huang, X. C., Guo, C. F., Yuan, Y. H., Luo, X. X., & Yue, T. L. (2015). Detection of
medicinal off-flavor in apple juice with artificial sensing system and comparison with
best panel evaluation and GC-MS. Food Control, 51, 270–277.
Kaya, Z., Yidiz, S., & Ünlütürk, S. (2015). Effect of UV-C irradiation and heat treatment on
the shelf life stability of a lemon-melon juice blend: Multivariate statistical approach.
Innovative Food Science and Emerging Technologies, 29, 230–239.
Lopetcharat, K., Kulapichitr, F., Suppavorasatit, I., Chodjarusawad, T., Phatthara-aneksin,
A., Pratontep, S., et al. (2016). Relationship between overall difference decision and
electronic tongue: Discrimination of civet coffee. Journal of Food Engineering, 180,
60–68.
Nilnont, W., Thepa, S., Janjai, S., Kasayapanand, N., Thamrongmas, C., & Bala, B. K.
(2012). Finite element simulation for coffee (Coffea arabica) drying. Food and
Bioproducts Processing, 90, 341–350.
Peris, M., & Escuder-Gilabert, L. (2016). Electronic noses and tongues to access food
authentication and adulteration. Trends in Food Science and Technology, 58, 40–54.
Rosa, A. R. D., Leone, F., Cheli, F., & Chiofaloa, V. (2017). Fusion of electronic nose,
electronic tongue and computer vision for animal source food authentication and
quality assessment – A review. Journal of Food Engineering, 210, 62–75.
Severini, C., Ricci, I., Marone, M., Derossi, A., & De Pilli, T. (2015). Changes in the
aromatic profile of espresso coffee as a function of the grinding grade and extraction
time: A study by the electronic nose system. Journal of Agricultural and Food
Chemistry, 63, 2321–2327.
Sunarharum, W. B., Williams, D. J., & Smyth, H. E. (2014). Complexity of coffee flavor: A
compositional and sensory perspective. Food Research International, 62, 315–325.
Suzihaque, M. U. H., & Driscoll, R. (2016). Effects of solar radiation, buoyancy of air flow
and optimization study of coffee drying in a heat recovery dryer. Procedia Engineering,
148, 812–822.
Tiggemann, L., Ballen, S. C., Bocalon, C. M., Graboski, A. M., Manzoli, A., Steffen, J., et al.
(2017). Electronic nose system based on polyaniline films sensor array with different
dopants for discrimination of artificial aromas. Innovative Food Science and Emerging
Technologies, 43, 112–116.
Toledo, P. R. A. B., Pezza, L., Pezza, H. R., & Toci, A. T. (2016). Relationship between the
different aspects related to coffee quality and their volatile compounds.
Comprehensive Reviews in Food Science and Food Safety, 15, 705–719.
Wold, S., Esbensen, K., & Geladi, P. (1987). Principal component analysis. Chemometrics
and Intelligent Laboratory Systems, 2, 37–52.
Yang, N., Liu, C. J., Liu, X. K., Degn, T. K., Munchow, M., & Fisk, I. (2016). Determination
of volatile marker compounds of common coffee roast defect. Food Chemistry, 211,
206–214.
Zhang, M., Chen, H. Z., Mujumdar, A. S., Tang, J., Miao, S., & Wang, Y. C. (2017). Recent
developments in high-quality drying of vegetables, fruits, and aquatic products.
Critical Reviews in Food Science and Nutrition, 57, 1239–1255.