This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/authorsrights
 Author's personal copy

Food Research International 61 (2014) 112119

Contents lists available at ScienceDirect

Food Research International

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

Detection of ground roasted coffee adulteration with roasted soybean

and wheat
Elis Daiane Pauli, Franciele Barbieri, Patrcia Salomo Garcia, Tiago Bervelieri Madeira,
Vinicius Ricardo Acquaro Junior, Ieda Spacino Scarminio,
Carlos Alberto Paulinetti da Camara, Suzana Lucy Nixdorf
Universidade Estadual de Londrina, Centro de Cincias Exatas, Departamento de Qumica, Laboratrio DIA, Campus Universitrio, P.O. Box: 10001, Londrina,Paran 86057-970, Brazil

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

Article history: Coffee as high-price commodity is quite vulnerable to adulteration by cheaper roasted grains. Frauds are conven-
Received 29 November 2013 tionally detected by microscopy; however, this technique is limited semi-quantitative assays that require train-
Received in revised form 5 February 2014 ing and skilled analysts. In this regard, carbohydrates as major macronutrients of grains can be used as chemical
Accepted 23 February 2014
markers to qualitatively and quantitatively assess coffee authenticity. Although, some tamper's studies have
Available online 3 March 2014
already been reported, this paper approaches on new analytical resources for detection of ground roasted coffee
Keywords:
adulteration, applying roasted soybean and wheat as sources of fraud. The characterization of the pure roasted
Carbohydrates coffee beans and of adulterations proles was taken by total carbohydrates validated method based on high-
HPLC performance anion-exchange chromatography with pulsed amperometric detection. The inuence of each
Chemometrics matrix was evaluated employing the simplex-centroid design for experiments with mixtures thus relating the
Frauds mixing ratio with each monosaccharide by response surface. The proposed models were effective in recognition
Coffee authenticity and prediction of different mixture proportions, thereby allowing the distinction of genuine coffee by principal
Predictive models component analysis and linear discriminant analysis. Predominantly, pure roasted coffee presented higher levels
of galactose and mannose. Glucose can be considered as a marker for wheat adulteration and fructose for
soybean, respectively. These results correspond to polysaccharides of pure raw grains, conrming this approach
as a feasible analytical tool for detection of adulterants in ground roasted coffee.
2014 Elsevier Ltd. All rights reserved.

1. Introduction adulteration detection in roasted ground coffee (Associao Brasileira

The growing number of studies applying analytical tools aimed to
detect and quantify adulterants in food shows that this is a common
practice (Ellis et al., 2012). Coffee as one of the most popular drinks con-
sumed for its refreshing, stimulating taste (Arya & Rao, 2007) and health
benets (Marcucci, Benassi, Almeida, & Nixdorf, 2013), is the second
commodity worldwide (Ebrahimi-Najafabadi et al., 2012; Marcucci
et al., 2013), whence of the 132.5 million bags traded during 2011/12,
the highest part was of Coffea arabica (61.8%) (ICO [ICO], 2013).There-
fore, coffee adulteration represents a matter of economic order (Patil,
2011) taking husks, sticks, corn, cocoa seed, barley, wheat middling,
chicory, brown sugar, maltodextrins, glucose syrups, soybean, triticale
and acai seeds as some of the described llers (Assad, Sano, Cunha,
Correa, & Rodrigues, 2002; Domingues et al., 2014; Prodolliet et al.,
1995; Sano, Assad, Cunha, Correa, & Rodrigues, 2003).
Once simple visual inspection is not capable of differentiating genuine
coffee of fraudulent samples; microscopy is the most used technique for
Corresponding author. Tel.: +55 43 3371 4836.
E-mail address: snixdorf@uel.br (S.L. Nixdorf).
da Indstria de Caf [ABIC], 2013; Brasil. Agncia Nacional de Vigilncia
Sanitria, 2003). However, coffee color, particle size and oily texture
encourage tamper (Assad et al., 2002) making easy the admixture and in-
corporation of various roasted and grinded cheapest llers (Reis, Franca,
& Oliveira, 2013a). So, it is a great challenge to nd a standard technique
non-subjective, selective for distinct markers and quantitative reproduc-
ible for industrial quality control. In Brazil, the lack of an appropriate tech-
nique for adulteration detection, limited to 1% of foreign material
in roasted ground coffee (ABIC, 2013), has led to revocation by the
Normative Instruction 007/2013 of the IN 16/2010 of MAPA (Ministry
of Agriculture, Livestock and Food Supply of Brazil) (Brasil. Ministrio
da Agricultura, Pecuria e Abastecimento, 2013). Aimed at detecting
adulterants in coffee, spectroscopy (micro Raman, UVVis, Mid and
Near infrared), chromatography (HPLC and GCMS), digital image pro-
cessing and electrophoresis have been applied (Ebrahimi-Najafabadi
et al., 2012). Reis et al. (2013a) and Reis, Franca, and Oliveira (2013b)
by infrared spectroscopy using principal component analysis with predic-
tion models, allowed the successful discrimination between pure roasted
coffee and tampered samples with some adulterants, despite not provid-
ing reference values for comparison by other techniques. However,

http://dx.doi.org/10.1016/j.foodres.2014.02.032
0963-9969/ 2014 Elsevier Ltd. All rights reserved.
 Author's personal copy
E.D. Pauli et al. / Food Research International 61 (2014) 112119
carbohydrates as major constituents in grains including green coffee bean
(48% in dry weight) (Arya & Rao, 2007; Redgwell, Trovato, Curti, &
Fischer, 2002) were used as tracers for instant coffee adulterations
(Bernal, Del Nozal, Toribio, & Del Alamo, 1996; Blanc, Davis, Parchet, &
Viani, 1989; Pauli, Cristiano, & Nixdorf, 2011) used by the norm ISO
11292 high performance liquid chromatography with anion-exchange
column and pulsed amperometric detection (HPLCHPAECPAD) with-
out need of derivatization (Garcia et al., 2009). Studies using carbohy-
drates as markers are still scarce in the literature, having described
monosaccharides separately only to few kinds of adulterants mentioned,
like spent coffee ground (Mussatto, Carneiro, Silva, Roberto, & Teixeira,
2011), added to instant coffee (Pauli et al., 2011) and to ground roasted
coffee (Domingues et al., 2014; Garcia et al., 2009; Reis et al., 2013a,
2013b). Despite the relative success achieved by some approaches for
determining coffee authenticity (Domingues et al., 2014; Ebrahimi-
Najafabadi et al., 2012; Reis et al., 2013a, 2013b), studies applying multi-
variate analysis (Box & Draper, 1987; Montgomery, 2001) to verify the
effect of interactions between different proportions for simultaneous
articial mixtures of roasted soybean and wheat into ground roasted
coffee have not yet been reported elsewhere. So, this approach proposes
combining chemometric tools with the use of the HPLCHPAECPAD
method, based on carbohydrate prole and their monosaccharide's
contents, to characterize the genuine coffee, distinguishing adulterated
samples with different proportions of roasted soybean and wheat.
2. Materials and methods
2.1. Mobile phases and standard solutions
Mobile phases 1.4 and 300.0 mmol L1 of sodium hydroxide were
prepared using carbonate-free NaOH 48.9% solution (Isosol, Brazil) by
degassing ultrapure water (Milli-Q, Millipore, Milford, USA) for
20 min prior, with nitrogen gas of 99.99973% purity (LINDE, Brazil).
Carbohydrate standards, provided by Merck (Darmstadt, Germany),
were stored in a glass desiccator under vacuum over phosphorus pent-
oxide (Merck) for 1 week. The standard solution was of: mannitol:
0.0165; arabinose: 0.200; galactose: 0.666; glucose: 0.250; xylose:
0.0788; mannose: 0.500; and fructose: 0.250 mmol L1.
2.2. Sample preparation of arabica coffee and adulterants (soybean
and wheat)
Arabica coffee green beans, classied as hard with less than 80
defects, were supplied by a coffee industryCacique de Caf Solvel.
In order to obtain a representative sample of coffee, one blend was
made of 200.0 g of the most commonly varieties cultivated in the
north of Paran (Iapar 59, Obat, Catua, Novo Mundo and Icatu). 2.4. Statistical analysis
The chosen representative adulterants were: commercial soybean
(cultivar D 206), the second most cultivated by Brazilian producers in
5 macro-regions; and wheat (cultivar Quartzo) grown on 3 micro-
regions of Paran, and in the 2 micro-regions from So Paulo, Santa
Catarina and Rio Grande do Sul, provided by a rural cooperative
COROL (Londrina, Paran, Brazil).
From each of the three matrices described, 1000.0 g was roasted
(Rod bell, Brazil) ground and sieved (35 mesh500 mm) and submit-
ted to color and moisture evaluation. Color measurements were per-
formed using a colorimeter (CR-400, Konica Minolta, Japan) with
standard illumination D65 and colorimetric normal observer angle of
0, with reading area of 8 mm. Measurements were based on the CIE
L*a*b* three dimensional Cartesian (xyz) color space represented by:
Luminosity (L*), ranging from 0 (black) to 100 (white)z axis; param-
eter a*, representing the greenred color componentx axis; and
parameter b*, representing the blueyellow componenty axis.
The chromatic hue was calculated by the arctangent of the ratio b*/a*,
(H* = arc tan b*/a*). Moisture content was determined using infrared
thermo-balance (Bel Engineering, Italy) at 105 C for 5 min.
113
Component proportions for experiments with mixtures are
constrained to a constant sum (1 or 100%) being commonly summarized
by triangular graphs. So, the simplex-centroid design for experiments
with mixtures in this study was composed by 10 experiments done in du-
plicate, with samples of pure roasted-arabica coffee, soybean and wheat
arranged at vertices of an equilateral triangle (1:0:0corresponding to a
100% for a particular component and 0 for all other components); binary
blends (1:1:0) involving a mixture of 2 components (50:50 % w/w) dis-
posed on middle of the edges; ternary mixtures arranged at axial points
(in ratio of 66:17:17%, 17:66:17%, and 17:17:66 % w/w/w), being the cen-
tral point (roasted coffee: soybean: wheat in ratio of 34:33:33 % w/w/w)
made in triplicate.
The hydrolysis was done in accordance with ISO 11292 (1995). So,
the different proportions of roasted sample according to the simplex-
centroid design with mixtures, always reaching 0.3000 g on a dry
weight basis, were transferred to a screw-cap erlenmeyer (500 mL,
Schott, Brazil) by adding 50 mL of 1.00 mol L 1 hydrochloric acid
(F. MAIA, Brazil) and placed in a thermostated water-bath (Fisatom
550, Brazil) at 85 C for 180 min. After cooled and ltered through
white paper in a 100 mL volumetric ask, solution was lled up
with ultrapure water. An aliquot of 10.0 mL was passed through
preconditioned cartridge (Sep Pak C18, Waters, USA) coupled with
membrane (0.22 m, nylon, Millipore, USA) being ready for chromato-
graphic analysis.
2.3. Chromatographic system and analytical conditions
The chromatographic system consisted of: inert pump of high-
pressure LC-10Ai (Shimadzu, Kyoto, Japan); 3-way low-pressure
NResearch-1367-72 solenoid valve; autosampler SIL-Prominence 20A
(Shimadzu); guard column and column CarboPac PA1 (4.0 x 250 mm,
10 m; Dionex, Sunnyvale, USA); column oven (Waters) coupled with
650 CHX controller (Pickering); source of 12 V RS570 (Stanford);
electrochemical cell ED-50 (Dionex); potentiostat Autolab PGSTAT 30
and interface (Eco-Chemie, Utrecht, Netherlands); software GPES for
data acquisition and processing, and software INTEGRA for areas
integrations.
The HPLC system runs at isocratic mode with NaOH 1.4 mmol L1
(as eluent: 045 min and for re-equilibrate: 57.672.6 min) and
NaOH 300.0 mmol L 1 (for regeneration: 45.157.5 min). Flow
rate: 1.0 mL min 1; injection vol.: 20.0 L; pre-column and column:
CarboPac PA-1 at 28 C; amperometric pulse waveform in ED-50-Au:
+ 0.20 V (400 ms); + 0.65 V (200 ms) and 0.20 V (400 ms).
The variation of the total carbohydrate related to different blends
was studied by the described simplex-centroid design for experi-
ments with mixtures. The effect of the mixtures was evaluated and
displayed by the response surface generated, expressed in percent-
age of each monosaccharide in dry basis (% db). The low degree poly-
nomials, given by linear or quadratic polynomial equations, were
chosen as models to represent the response as a function of the var-
iable mixture. Analysis of variance (ANOVA) allowed creating
models at 5% levels of signicance. The model has set a correlation
between experimental versus predicted values. The distinction be-
tween the samples was performed by applying principal component
analysis (PCA), and the separation of the groups taken from the hier-
archical analysis by the largest Euclidean distance in the dendro-
gram. Besides the multivariate analysis used a linear discriminant
analysis (LDA) was performed on the carbohydrate data obtained
from the different blends. All the statistical analyses, including re-
sponse surfaces, ANOVA, models, PCA and LDA, were carried out
using Statistica 8.0 software (statsoft Inc., Tulsa, USA).
 Author's personal copy

114 E.D. Pauli et al. / Food Research International 61 (2014) 112119

Average content of carbohydrates (in g 100 g1) determined by HPAECPAD following the simplex-centroid design of mixture experiments for roasted samples of arabica coffee, soybean and wheat (n = 2, except for central point with n = 3).
3. Results and discussion

coffee 17% soybean 66%

17co:66so:17wh
Pure samples of coffee, soybean and wheat (cheaper adulter-
ants) were characterized by color, moisture and carbohydrate con-

0.02

0.03
0.01
0.02
0.04
0.03
wheat 17%
tents. It was impossible to compare roasted soybeans and roasted

wheat carbohydrates amounts obtained by this study with the lit-

2.26
3.65

0.68
7.61
0.98
1.69
10
erature, since data were not found. Due to the lack of information
regarding carbohydrate composition, polysaccharides of raw beans

coffee 17% soybean

were used as reference, since considering hydrolysis should pro-

17co:17so:66wh
duce the corresponding monosaccharides. The effect of each carbo-

17% wheat 66%

0.02

0.05
0.10
1.18
0.21
0.05
hydrate was evaluated by a simplex-centroid design for experiments

with mixtures, shown by the response surfaces. ANOVA determined

2.37

0.24
2.01
30.22
2.70
1.63
the quality of models, and principal component analysis and linear

9
discriminant analysis were able to discriminate different samples,
distinguishing pure coffee of adulterated ones, based on their carbohy-

coffee 66% wheat 17%

drate prole.

66co:17wh:17so
soybean 17%
3.1. Characterization of samples of coffee, soybean and wheat

0.08

0.02
0.16
0.32
0.04
0.03

Samples were roasted until reaching similar color to arabica

2.19
4.85
7.82

0.19
0.78
5.34
roasted coffee, so that the incorporation of the grains could not be vi-

8
sually noticed. Color parameters (L* and H*) were measured in dupli-

coffee 34% wheat 33%

cate, since they can inuence on carbohydrates during roasting
process. Color reectance (L*) was 18.0 1.12 for coffee, 37.4

34co:33wh:33so
3.23 for soybean, and 47.4 0.30 for wheat, with chromaticity

0.09

0.07
0.09
0.32
0.04
0.14
soybean 33%
(H*) of 51.4 0.2; 69.2 0.0 and 64.0 0.1, respectively. Hence,

coffee was classied as medium roasted according to Campanha

0.41
2.27
3.58
11.79
1.58
3.02
(2008), while adulterant roasting was considered light (Dias, 2005).

7
Moisture (n = 2) was determined for dry bases being 3.88 0.02%
w/w for coffee, 1.99 0.01% w/w for soybean and 1.78 0.01% w/w

0.09

0.07
0.06
0.80
0.16
0.02
soybean 50%

50wh:50so
for wheat. wheat 50%

2.50
2.05

0.58
20.09
2.24
0.30
6

0.09

0.01
0.01
0.06
0.02
0.21
50co:50wh
wheat 50%
coffee 50%

19.55

0.08
2.20
3.22

2.11
4.14
5

soybean 50%

0.01
0.14

0.02
0.21

0.02
0.27
coffee 50%

50co:50so

0.89
2.18

0.50
4.73

0.32
4.11
4

40.75 2.93
2.63 0.27
0.36 0.04

4.18 0.42
0.07 0.00
wheat 100%

100wh

ND: not detected. SD: standard deviation. n: number of observations.

ND
3
soybean 100%

0.03

0.01
0.05

0.03
0.06

0.02

100so

1.49

1.10
2.36
3.44

0.56
0.37
2

0.04
0.29

0.01
0.16

0.28
coffee 100%

100co

0.26
2.02

0.07
5.76

7.06
ND
1

Sample composition

Arabinose SD
Galactose SD

Mannose SD
Fructose SD
Glucose SD
Design points

Xylose SD
Sample code

Fig. 1. Chromatograms of total carbohydrates: (a) standard of monosaccharides,

Table 1

(b) roasted arabica coffee 100%, (c) roasted soybean 100%, (d) roasted wheat 100%
(diluted 1:3). Peaks (pk): (1) mannitol, (2) arabinose, (3) galactose, (4) glucose, (5) xylose,
(6) mannose, and (7) fructose, determined by HPAECPAD.
 Author's personal copy

E.D. Pauli et al. / Food Research International 61 (2014) 112119 115

3.2. Carbohydrates in pure matrices

A standard solution with seven monosaccharides was used for quan-

Fig. 2. Chromatograms for articial mixtures of soybean and wheat in pure coffee. Peaks:
(1) mannitol, (2) arabinose, (3) galactose, (4) glucose, (5) xylose, (6) mannose,
(7) fructose.
tifying carbohydrates through the external standardization method
(Fig. 1a). As carbohydrate content depends directly on the analytical
conditions (Domingues et al., 2014), chromatographic proles using
HPAECPAD for the samples: coffee and adulterant, soybean and
wheat are represented in Fig. 1b, c, and d, respectively. The method
was appropriate for quantication due to its precision (RSD b 5.00%)
and linearity (r N 0.99).
According to Fig. 1, the prole of coffee carbohydrates presents
quantitative differences in relation to the adulterants, but the monosac-
charides obtained by hydrolysis are in accordance to the compositional
description of polysaccharides, regarding the studied samples. Ground
roasted coffee presented considerable amount of arabinose (2.02
0.29 g 100 g1), else the higher amounts of galactose (5.76 0.16 g
100 g1) and mannose (7.06 0.28 g 100 g1) (Fig. 1b and Table 1).
This can be justied by the composition of arabica coffee beans, consisting

Table 2
Predictive mathematical models with respective adjusted coefcients of determination (R2 adj) for carbohydrate applying ANOVA for the matrices.

Carbohydrate Equation Model Type R2 adj

Arabinose 1 1.99co + 2.34so + 2.56wh linear 0,6154

Galactose 2 5.77co + 3.34so + 0.34wh + 1.04coso + 1.06cowh + 1.00sowh quadratic 0,9924
Glucose 3 Model with lack of t
Xylose 4 0.07co + 0.51so + 4.09wh linear 0,9876
Mannose 5 7.09co + 0.35so + 0.03wh + 1.69coso + 2.53cowh + 0.45sowh quadratic 0,9965
Fructose 6 0.003co + 1.08so + 0.05wh linear 0,9806

co: coffee; so: soybean; wh: wheat. R2 adj: adjusted coefcient of determination. In bold the signicant coefcients.

Table 3
ANOVA for models adopted following the simplex-centroid design for experiments with mixtures of roasted samples of arabica coffee, soybean and wheat (n = 2, except for central point
with n = 3).

Variation source Sum of squares Degrees of freedom Mean square (MQ) F-value Probability
(SQ)

Arabinose
Regression (R) 0.4910 2 0.2455 17.00 0.00
Residuals (r) 0.2599 18 0.0144
Lack of t (lf) 0.0469 7 0.0067 0.35 0.91
Pure error (pe) 0.2129 11 0.0193
Total (T) 0.7509 20 0.0375

Galactose
Regression (R) 44.9736 6 7.4956 609.89 0.00
Residuals (r) 0.1720 14 0.0123
Lack of t (lf) 0.0423 3 0.0141 1.19 0.36
Pure error (pe) 0.1298 11 0.0118
Total (T) 45.1457 20 2.2573

Xylose
Regression (R) 29.1761 5 5.8352 298.73 0.00
Residuals (r) 0.2930 15 0.0195
Lack of t (lf) 0.0414 4 0.0103 0.45 0.77
Pure error (pe) 0.2516 11 0.0229
Total (T) 29.4691 20 1.4734

Mannose
Regression (R) 101.1501 6 16.8583 900.88 0.00
Residuals (r) 0.2620 14 0.01871
Lack of t (lf) 0.0169 3 0.0056 0.25 0.86
Pure error (pe) 0.2451 11 0.0223
Total (T) 101.4121 20 5.0706

Fructose
Regression (R) 2.2356 5 0.4471 230.27 0.00
Residuals (r) 0.0019 15 0.0019
Lack of t (lf) 0.0018 4 0.0018 0.90 0.50
Pure error (pe) 0.0020 11 0.0020
Total (T) 2.2647 20 0.1132
 Author's personal copy

116 E.D. Pauli et al. / Food Research International 61 (2014) 112119

with 5060% of polysaccharides arabinogalactans and galactomannans,
described by Redgwell and Fischer (2006) and Arya and Rao (2007).
The content of arabinose was lower, probably due to the degradation
into low molecular weight carbohydrates (Lima, Loh, & Buckeridge,
2004) during the roasting, and follows a susceptibility order of: arab-
inose N galactose N mannose N glucose (Arya & Rao, 2007). The small
amount of xyloglucan in the cell wall composition of green coffee
beans, according to Redgwell et al. (2002); Fischer, Reimann, Trovato,
and Redgwell (2001); and Redgwell and Fischer (2006), can justify
the low concentration found for glucose (Fig. 1bpk 4 with 0.26
0.04 g 100 g1Table 1) and xylose (pk 50.07 0.01 g 100 g1) in
ground roasted coffee. These results conrm the same trend shown by
Oosterveld, Voragen, and Schols (2003), Garcia et al. (2009) and
Domingues et al. (2014).
For crops like soybean, the hydrolysis of the sucrose can partly justify
production of glucose and fructose in addition to the presence of these
carbohydrates already associated with this matrix, as described in the
literature. According to Liu (1997) and Rivas (2006) soybean is free
from starch and constituted of 5.0% of sucrose, 3.6% of arabinoxylans
and 2.3% of galactans. Its cell wall contains about 30% pectin, 50%
hemicelluloses and 20% cellulose. The glucose amount was 1.49
0.03 g 100 g1 (Table 1, Fig. 1cpk 4) while fructose, monosaccharide
detected only in the soybean was 1.10 0.01 g 100 g1 (Fig. 1cpk
7). Therefore, the arising of 3.44 0.06 g 100 g1 (Table 1, Fig. 1c
pk 3) of galactose, as well as 2.36 0.05 g 100 g1 (Table 1, Fig. 1c
pk 2) of arabinose and 0.56 0.03 g 100 g1 (Table 1, Fig. 1cpk 5)
of xylose, can thus be explained.
The high level of glucose (Fig. 1dpk 440.75 2.93 g 100 g1
Table 1) measured in wheat can be justied by the hydrolysis of part
of the starch, which is the basic composition of the endosperm of this
grain, corresponding to approximately 80%, equivalent to 82% of the
dry weight (Orth & Shellenberger, 1988; Pomeranz, 1988). It was so

Fig. 3. Response surface contour curves for mixture of coffee, soybean and wheat: (a) arabinose; (b) galactose; (c) glucose; (d) xylose; (e) mannose; (f) fructose. (o) Experimental points of
simplex-centroid design with mixtures clockwise: vertex 100%coffee (100co), soybean (100so) and wheat (100wh); binary blends in edges50% (coffee: soy-50co:50so) (soy: wheat-
50so:50wh) (coffee:wheat-50co:50wh).Ternary mixtures-center point 33% (coffee:soy:wheat-34co:33so:33wh); axial points 66:17:17% (coffee:soy:wheat-66co:17so:17wh), (coffee:
wheat:soybean-17co:66so:17wh), (coffee: soy: wheat-17co:17so:66wh).
 Author's personal copy

E.D. Pauli et al. / Food Research International 61 (2014) 112119 117

high, that it required a dilution of three times for its quantication. Also,
arabinose and xylose were derived by the hydrolysis of arabinoxylans
found in cell wall of wheat, containing 88% of arabinoxylans and 4% of
cellulose, according to Lineback and Rasper (1988). They appear in
higher proportion for wheat (Fig. 1d and Table 1) with 2.63 0.27 g
100 g1 (pk 2) of arabinose and 4.18 0.42 g 100 g1 of xylose (pk 5).
Mannitol was not detected in any of the analyzed matrices (Fig. 1b, c
and dpk 1). This was expected, since their presence was not reported
for the raw material of coffee, soybean and wheat in the literature. How-
ever, the fact that mannitol was below its limit of detection may be a
relevant information, considering the creation of a more complete
model for coffee adulteration. Thus, if mannitol is found, it can represent
fraud by a matrix containing this type of carbohydrate in their composi-
tion, like coffee husks, just distinguished by high levels of this sugar
alcohol (Garcia et al., 2009), but by exclusion, can also help indicate
that tampering was not due to soybeans and wheat.
The mannose (Fig. 1pk 6) had opposite behavior compared to
xylose. Although mannose amount (Table 1) in coffee was high, it was
low for soybean (0.37 0.02 g 100 g 1), and even lower for wheat
(0.07 0.00 g 100 g1).

3.3. Monosaccharides in mixtures

The overlay of chromatograms in Fig. 2 shows variations in concen-

tration of each carbohydrate by the articial mixture of adulterants. A
reduction in mannose (pk 6) and galactose (pk 3) can be evidenced
by a decrease in the proportion of coffee from 100% to 66%, with a
marked increase in glucose (pk 4) and xylose (pk 5) with increasing
soybean and wheat, producing the appearance of fructose (pk 7).
Arabinose (pk 2) remained virtually unchanged.
Table 1 shows the amount of carbohydrates from different blends of
adulterants, according to the simplex-centroid design for experiments
with mixtures.
From carbohydrate contents (Table 1), predictive mathematical
models were created (Eq. 16, Table 2) and by ANOVA (Table 3) their
suitability was tested, considering the F values that do not suffer from
lack of t being signicant. To allow a better view on how the content
of a specic carbohydrate varies as a function of the three matrices Fig. 4. Plots of experimental values measured versus predicted values by linear model of
(coffee, soybean and wheat) response surfaces were generated (Fig. 3). regression for carbohydrate glucose (a) with their residuals values (b).
The linear model was the best t to the variation of carbohydrates
arabinose, xylose and fructose for different matrices, with adjusted
coefcients of determination (R2 adj) of 0.62, 0.99 and 0.98, respec- Although the model has not been adjusted by the analysis of variance,
tively (Eq. 1, 4 and 6). The other carbohydrates, mannose and galactose there was a good agreement between the experimental and predicted
were best described by quadratic models (Table 2, Eq. 2 and 6) with values for glucose by the model with a correlation coefcient of 0.98
settings to the experimental data (R2 adj) above 0.99. The response (Fig. 4a), explaining 98.04% of data variance. It also appears that the
surfaces generated from the models are shown in Fig. 3. residues were random (Fig. 4b), without evidence of trends, with values
From the mathematical model coefcients, it can be observed that distributed above and below zero, following a normal distribution.
the carbohydrate galactose (Table 2, Eq. 2, Fig. 3b) and mannose
(Table 2, Eq. 5, Fig. 3e) is present in higher concentrations in the coffee 3.4. Ability to discriminate pure from adulterated coffee by soybean
matrix. Fructose has a higher amount in the soybean matrix (Table 2, and wheat
Eq. 6, Fig. 3f) represented by more intense staining at vertex of pure soy-
bean (100so). Coffee, soybean and wheat contribute almost equally to Projection of the variables (monosaccharides) (Fig. 5a) and disper-
the amount of arabinose (Table 2, Eq. 1, Fig. 3a). Xylose had larger sion of samples (Fig. 5b) by principal component analysis were used
amount in wheat (Table 2, Eq. 4, Fig. 3d) and even higher for glucose to discriminate pure ground roasted arabica coffee from adulterated
(Fig. 3C), although a model had not been adjusted for this carbohydrate. articial mixture of roasted soybean and wheat. The dendrogram
It is noteworthy that although the coefcient of determination (R2) (not shown) based on hierarchical analysis by Euclidean distance sepa-
was 0.98 for the linear model of glucose, it showed lack of t. And so rated the samples into 4 groups presenting also the shape of triangle
at principle it should not proceed with testing signicance in regression. (Fig. 5b), which almost match fully the experimental points used in
But in this case it appears that the lack of signicant adjustment for the F the simplex-centroid design for experiments with mixtures.
test is due to the small value of the quadratic sum of the pure error Together, the principal components (PC) 1 and 2 in Fig. 5a explain
(ANOVA), a consequence of the very small difference between the du- 94.81% of data variability. The samples with positive score values
plicates (Table 2), which justies the lack of t of the model (Box & distributed to the right of PC1 are more inuenced by the responses of
Draper, 1987; Breitkreitz, 2007; Montgomery, 2001). However, no sig- variables mannose and galactose that contribute to this component
nicant graphical differences on observed versus predicted values (Fig. 5a), represented by pure arabica coffee (100co) located at the ex-
(Fig. 4a) and on predicted versus raw residuals (Fig. 4b) from the models treme right of the x axis of Fig. 5b. At PC2 (y axis), the carbohydrates
of diagnosis were observed. that inuence the values of the pure wheat samples (100whFig. 5b)
 Author's personal copy
118 E.D. Pauli et al. / Food Research International 61 (2014) 112119
Fig. 5. (a) Principal component analysis for the variables arabinose, galactose, glucose,
xylose, mannose and fructose determined by data of the simplex-centroid design for ex-
periments with mixtures of coffee, soybean and wheat, made in duplicate. (b) Pure arabica
coffee encoded by 100co starting in the upper right corner clockwise, pure soybean
(100so) and pure wheat(100wh); binary mixtures of 50% (50co:50so), (50so:50wh),
(50co:50wh); and ternary mixtures in center of 34% of coffee, 33% of soybean and
33% of wheat (34co:33so:33wh); and in axial points 66% of coffee,17% of soybean
and 17% of wheat (66co:17so:17wh); 17% of coffee, 66% of soybean and 17% of
wheat (17co:66so:17wh); and 17% of coffee,17% of soybean and 66% of wheat (17
co:17so:66wh).
with positive score are glucose and xylose (Fig. 5a), and of the pure soy-
bean samples (100soFig. 5b) are inuenced by fructose in the negative
scores (Fig. 5a). The mixtures follow the proportion of the pure matri-
ces. So, it can be seen that this approach was able to separate pure coffee
(group I), of samples adulterated by soybean and wheat (group IV) from
pure soybean (group II) and from pure wheat (group III).
Applying linear discriminant analysis (LDA) (Fisher, 1936) con-
sidering carbohydrate contents of the 10 points of the simplex-
centroid as classication, it has not been possible to obtain a regres-
sion, because of the small amount of data (duplicate) (Table 1). How-
ever, using the point of pure coffee (1Table 1) designated as
coffee, adding to this classication 4 more coffee samples previous-
ly analyzed, as this analysis requires at least 5 data, and the other
points of the simplex-centroid (210Table 1) those containing
any amount of soybean and/or wheat designated as adulterated
the function: + (0.226506*arabinose) + (0.104662 galactose) +
(0.005468*glucose) + (0.112465*xylose) + (0.027360*mannose) +
(0.435309*fructose) were obtained. This function was able to discrimi-
nate different samples according to their carbohydrate composition,
separating into 2 groups on Node 1(split constant 0.788963), thus
allowing correct classication of 100%, i.e., 5 samples were classied
as coffee and 19 as adulterated samples.
4. Conclusion
The study based on the simplex-centroid design for experiments
with mixtures allowed creating predictive models for different percent-
age proportions of cheaper roasted grains such as soybean and wheat
added to the roasted ground coffee. This type of fraud is imperceptible
to the naked eye, and there are no studies describing in the literature
so far this simultaneous adulteration.
The determination of carbohydrates carried out by the validated
method made it possible to establish the prole and reference amounts
for genuine coffee, which can be distinguished by the principal compo-
nent analysis and linear discriminant analysis of faked samples. The re-
sults obtained herein conrmed that HPLCHPAECPAD associated
with chemometrics presents a potential to be used as a routine
approach for the adulteration detection and authenticity checking for
ground roasted coffee. Further studies will be conducted for the detec-
tion and discrimination of other common adulterants such as barley
and impurities like husks and sticks used in tapered coffee to expand
to a complete model.
Acknowledgments
The authors acknowledge nancial support from the following
Brazilian Government Agencies: CNPq, CAPES, Fundao Araucria and
FINEP.
References
ABIC (2013). Norma de Qualidade Recomendvel e Boas Prticas de Fabricao de Cafs
Torrados em Gro e Cafs Torrados e Modos. Retrieved November 18, 2013, from:
http://www.abic.com.br/publique/media/NMQ_LEGISLAcaO_NORMAQUALIDADE.pdf.
Arya, M., & Rao, L. J. M. (2007). An impression of coffee carbohydrates. Critical Reviews in
Food Science and Nutrition, 47, 5167.
Assad, E. D., Sano, E. E., Cunha, S. A.R., Correa, T. B.S., & Rodrigues, H. R. (2002).
Identicao de impurezas e misturas em p cbede caf por meio de comportamento
espectral e anlise de imagens digitais. Pesquisa Agropecuria Brasileira, 37, 211216
(Retrieved November 18, 2013, from: http://ainfo.cnptia.embrapa.br/digital/bitstream/
AI-SEDE/21836/1/pab20_360.pdf).
Bernal, J. L., Del Nozal, M. J., Toribio, L., & Del Alamo, M. (1996). HPLC analysis of in wines
and instant coffees using anion exchange chromatography coupled to pulsed amper-
ometric detection. Agricultural and Food Chemistry, 44, 507511.
Blanc, M. B., Davis, G. E., Parchet, J. M., & Viani, R. (1989). Chromatographic prole of
carbohydrates in commercial soluble coffees. Agricultural and Food Chemistry, 37(4),
926930.
Box, G. E. P., & Draper, N. R. (1987). Empirical model-building and response surfaces. NY:
John Wiley & Sons.
Brasil. Agncia Nacional de Vigilncia Sanitria (2003). Regulamento Tcnico de Avaliao
de Matrias Macroscpicas e Microscpicas Prejudiciais Sade Humana em
Alimentos Embalados (contaminantes e fragmentos de insetos) (RDC n 175/2003).
Retrieved November 16, 2013, from: http://portal.anvisa.gov.br/wps/wcm/connect/
efabae00474580fc8d11dd3fbc4c6735/RDC_175_2003.pdf?MOD=AJPERES
Brasil. Ministrio da Agricultura, Pecuria e Abastecimento (2013). Regulamento Tcnico
para o Caf Torrado em Gro e Torrado e Modo (IN 007/2013). Retrieved November
16, 2013, from: http://sistemasweb.agricultura.gov.br/sislegis/action/detalhaAto.do?
method=consultarLegislacaoFederal.
Breitkreitz, M. C. (2007). Otimizao simultnea de variveis de processo e mistura em
cromatograa lquida de alta ecincia. (master's thesis). SP: Universidade Estadual
de Campinas (Retrieved November 21, 2013, from: http://cutter.unicamp.br/
document/?code=000408465).
Campanha, F. G. (2008). Discriminao de espcies de caf (Coffea arabica e Coffea
canephora) pela composio de diterpenos. (master's thesis). Londrina, PR:
Universidade Estadual de Londrina (Retrieved November 21, 2013, from: http://
www.bibliotecadigital.uel.br/document/?code=vtls000127750).
Dias, R. C. (2005). Discriminao de espcies de caf (Coffea arabica e Coffea canephora) em
diferentes graus de torra. (master's thesis). Londrina, PR: Universidade Estadual de
Londrina (Retrieved November 21, 2013, from: http://www.bibliotecadigital.uel.br/
document/?code=vtls000108937).
Domingues, D. S., Pauli, E. D., Abreu, J. E. M., Massura, F. W., Cristiano, V., Santos, M. J., et al.
(2014). Detection of roasted and ground coffee adulteration by HPLC by amperometric
and by post-column derivatization UVVis detection. Food Chemistry, 146, 353362.
http://dx.doi.org/10.1016/j.foodchem.2013.09.066.
Ebrahimi-Najafabadi, H., Leardi, R., Oliveri, P., Casolino, M. C., Jalali-Heravi, M., & Lanteri, S.
(2012). Detection of addition of barley to coffee using near infrared spectroscopy and
chemometric techniques. Talanta, 99, 175179.
Ellis, I. D., Brewster, V. L., Dunn, W. B., Allwood, J. W., Golovanov, A. P., & Goodacre, R.
(2012). Fingerprinting food: Current technologies for the detection of food adultera-
tion and contamination. Chemical Society Reviews, 41, 55695868. http://dx.doi.org/
10.1039/C2CS35138B.
 Author's personal copy
E.D. Pauli et al. / Food Research International 61 (2014) 112119
Fischer, M., Reimann, S., Trovato, V., & Redgwell, R. J. (2001). Polysaccharides of green
arabica and robusta coffee beans. Carbohydrates Research, 330, 93101.
Fisher, R. A. (1936). The use of multiple measurements in taxonomic problems. Annals of
Eugenics, 7(2), 179186. http://dx.doi.org/10.1111/j.1469-1809.1936.tb02137.x/pdf.
Garcia, L. M. Z., Pauli, E. D., Cristiano, V., Camara, C. A. P., Scarmnio, I. S., & Nixdorf, S. L.
(2009). Chemometric evaluation of adulteration prole in coffee due to corn and
husk by determining carbohydrates using HPAEC-PAD. Journal of Chromatographic
Science, 47, 825832.
ICO (2013). Relatrio mensal sobre o mercado cafeeiro. Notcias mais recentes sobre
produo, exportaes e consumo de caf (Retrieved November 18, 2013, from:
http://dev.ico.org/documents/cy2012-13/cmr-0913-p.pdf).
ISO/11292 (1995). Instant coffee: Determination of free and total carbohydrate contents.
Method using high performance anion-exchange chromatography.
Lima, D. U., Loh, W., & Buckeridge, M. S. (2004). Xyloglucan-cellulose interaction depends
on the side chains and molecular weight of xyloglucan. Plant Physiology and
Biochemistry, 42, 389394.
Lineback, D. R., & Rasper, V. F. (1988). Wheat carbohydrates. In Y. Pomeranz (Ed.), Wheat:
Chemistry and technology (pp. 277372). St. Paul: AACC.
Liu, K. (1997). Chemistry and nutritional value of soybean components. In K. Liu (Ed.),
Soybeans: Chemistry, technology and utilization (pp. 2595). NY: Aspen Publisher.
Marcucci, C. T., Benassi, M. T., Almeida, M. B., & Nixdorf, S. L. (2013). Teores de trigonelina,
cido 5-cafeoilqunico, cafena e melanoidinas em cafs solveis comerciais brasileiros.
Quimica Nova, 36, 544548. http://dx.doi.org/10.1590/S0100-40422013000400011.
Montgomery, C. D. (2001). Design and analysis of experiments (5th ed.). NY: John Wiley &
Sons.
Mussatto, S. I., Carneiro, L. M., Silva, J. P. A., Roberto, I. C., & Teixeira, J. A. (2011). A study on
chemical constituents and sugars extraction from spent coffee grounds. Carbohydrate
Polymers, 83, 368374. http://dx.doi.org/10.1016/j.carbpol.2010.07.063.
Oosterveld, A., Voragen, A. G. J., & Schols, H. A. (2003). Effect of roasting on the carbohy-
drate composition of coffea arabica beans. Carbohydrate Polymers, 54(2), 183192.
http://dx.doi.org/10.1016/S0144-8617(03)00164-4.
Orth, R. A., & Shellenberger, J. A. (1988). Origin, production, and utilization of wheat. In Y.
Pomeranz (Ed.), Wheat: Chemistry and technology (pp. 114). St. Paul: AACC.
119
Patil, A.R. (2011). A study on consumer protection through maintenance of product safety &
standards in India. Asian Law Institute, 23, (Retrieved November 18, 2013, from:
http://law.nus.edu.sg/asli/pdf/WPS023.pdf).
Pauli, E. D., Cristiano, V., & Nixdorf, S. L. (2011). Mtodo para determinao de carboidratos
empregado na triagem de adulteraes de caf. Quimica Nova, 34, 689694.
http://dx.doi.org/10.1590/S0100-40422011000400023.
Pomeranz, Y. (1988). Chemical composition of kernel structures. In Y. Pomeranz (Ed.),
Wheat: Chemistry and technology (pp. 97158). St. Paul: AACC.
Prodolliet, J., Bruelhart, M., Blanc, M. B., Leloup, V., Cherix, G., Donnelly, C. M., et al. (1995).
Adulteration of soluble coffee with coffee husks and parchments. Journal of AOAC
International, 78(3), 761767.
Redgwell, R. J., & Fischer, M. (2006). Coffee carbohydrates. Brazilian Journal of Plant
Physiology, 18, 165174.
Redgwell, R. J., Trovato, V., Curti, D., & Fischer, M. (2002). Effect of roasting on degradation
and structural features of polysaccharides in Arabica coffee beans. Carbohydrate
Research, 337, 421431. http://dx.doi.org/10.1016/S0008-6215(02)00010-1.
Reis, N., Franca, A. S., & Oliveira, L. S. (2013). Discrimination between roasted coffee,
roasted corn and coffee husks by diffuse reectance infrared Fourier transform spec-
troscopy. LWTFood Science and Technology, 50, 715722. http://dx.doi.org/10.1016/j.
lwt.2012.07.016.
Reis, N., Franca, A. S., & Oliveira, L. S. (2013). Performance of diffuse reectance infrared
Fourier transform spectroscopy and chemometrics for detection of multiple adulter-
ants in roasted and ground coffee. LWTFood Science and Technology, 53, 395401.
http://dx.doi.org/10.1016/j.lwt.2013.04.008.
Rivas, M. B. (2006). Compostos da soja. In M. B. Rivas (Ed.), Soja: Qualidade de vida e sade
com prazer e sabor (pp. 3755). Porto Alegre: Age.
Sano, E. E., Assad, E. D., Cunha, S. A.R., Correa, T. B.S., & Rodrigues, H. R. (2003). Quan-
tifying adulteration in roast coffee powders by digital image processing. Journal
of Food Quality, 26, 123134. http://dx.doi.org/10.1111/j.1745-4557.2003.
tb00232.x.