Food Chemistry 346 (2021) 128871

Contents lists available at ScienceDirect

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

Simultaneous quantification of ethylene glycol and diethylene glycol in

beer by gas chromatography coupled to mass spectrometry
Leila Rodrigues Caldeira a, b, Fernando Diniz Madureira b, Thalita De Faria Maia b, Carlos
Vitor Muller c, Christian Fernandes a, *
a
Faculdade de Farmácia, Universidade Federal de Minas Gerais, Belo Horizonte, MG 31270-901, Brazil
b
Laboratório Federal de Defesa Agropecuária, Avenida Raja Gabaglia, 245, Setor H, Bairro Cidade Jardim, Belo Horizonte, MG 30380-103, Brazil
c
Ministério da Agricultura, Pecuária e Abastecimento, Esplanada dos Ministérios, Bloco D, Brasília, Distrito Federal 70043-900, Brazil

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

Keywords: An analytical method for the simultaneous determination of ethylene glycol and diethylene glycol in beer was
Glycols developed and validated according to current legislation. This method includes the application of sample dilution
Beverage with ethanol followed by quantification using gas chromatography coupled to mass spectrometry. All figures of
Food contaminants
merit were within the limits established by regulation. The recoveries of the analytes, expressed as mean re­
Gas chromatography
covery, were between 91.9% and 108.9%. Precision, in terms of repeatability and intermediate precision, was
Mass spectrometry
established (relative standard deviations were lower than or equal to 10%). The limits of detection (10.0 and 5.0
Chemical compounds studied in this article:
Ethylene glycol (PubChem CID: 174)
mg.L− 1 for ethylene glycol and diethylene glycol, respectively) and quantification (15.0 mg.L− 1 for ethylene
Diethylene glycol (PubChem CID: 8117) glycol and diethylene glycol) obtained were appropriate. Finally, the present method was applied for determi­
nation of ethylene glycol and diethylene glycol in 701 beer samples (from 67 different brands and 128 different
labels), proving to be reliable.

1. Introduction (GC–MS) (FDA, 2007; Fu, Hao, Parker, & Knapp, 2011), gas chroma­
tography with flame ionization detection (GC-FID) (Holloway, Math­
Ethylene glycol (EG) and diethylene glycol (DEG) (Fig. 1) are clear, eswaran, Leeks, Bradby, & Wahab, 2010), ultra-high performance liquid
hygroscopic, odorless and sweet tasting liquids. These glycols are chromatography coupled to high resolution mass spectrometry
miscible with water and polar organic solvents. They are mainly used as (UHPLC–HRMS) (Hernández, Ibáñez, & Sancho, 2008), attenuated total
humectant, solvent, antifreeze and coolants agents in the production of reflection Fourier transform infrared (ATR-FTIR) spectroscopy (López-
automotive antifreeze and coolant formulations, aircraft anti-icing and Sánchez, Domínguez-Vidal, Ayora-Cañada, & Molina-Díaz, 2008); direct
deicing materials (Fowles, Banton, Klapacz, & Shen, 2017). analysis in real time-mass spectrometry (DART-MS) (Self, 2013) and
EG and DEG have physicochemical properties similar to non-toxic excipients by GC-FID (Tokunaga, Osako, & Sato, 2012) have been
and more expensive glycols, such as glycerol (GRO) and propylene reported.
glycol (PPG) (Fig. 1). As a result, they are often used as adulterants in As previously described, several analytical techniques are described
products based on GRO or PPG, mainly in pharmaceutical products such in the literature for the analysis of toxic glycols in different matrices,
as toothpaste, injectable drugs, antipyretic medication and couch syrup and, due to the physical-chemical characteristics of this group of com­
(Marraffa, Holland, Stork, Hoy, & Hodgman, 2007). pounds, GC–MS is a highly recommended technique for its analysis.
Unlike PPG and GRO, which are both generally recognized as safe in Published analytical methods for analysis of DEG in food are more
food, cosmetics and drugs, there are no approved use for EG and DEG in limited. There are few studies from the 1980s regarding the analysis of
these products because of their high toxicities. In fact, there are several DEG in wine (Caccamo, Di Coarcia, & Samperi, 1986; Kaiser & Rieder,
reports of incidents involving human poisoning, mainly caused by DEG. 1985; Lau & Weber, 1987; Lawrence, Chadha, Lau, & Weber, 1986)
The determination of DEG or both EG and DEG in pharmaceutical because, in that time, some events related to the addition of DEG in this
products by gas chromatography coupled to mass spectrometry beverage occurred. This practice was done to improve its sweet taste.

* Corresponding author at: Universidade Federal de Minas Gerais, Av. Pres. Antônio Carlos, 6627, Pampulha, 31270-901 Belo Horizonte, MG, Brazil.
E-mail address: cfernandes@farmacia.ufmg.br (C. Fernandes).

https://doi.org/10.1016/j.foodchem.2020.128871
Received 19 August 2020; Received in revised form 4 December 2020; Accepted 9 December 2020
Available online 15 December 2020
0308-8146/© 2020 Elsevier Ltd. This article is made available under the Elsevier license (http://www.elsevier.com/open-access/userlicense/1.0/).
L.R. Caldeira et al. Food Chemistry 346 (2021) 128871

were acquired and used to prepare stock solutions. Ethanol (Merck,

Darmstadt, Germany) was used to dilute the beer samples and to prepare
the stock solutions of all analytes.
Individual standard solutions of EG and DEG in ethanol were pre­
pared to be used in pool stock standard solutions, which contained both
EG and DEG in ethanol (10,000 mg.L− 1 or 1000 mg.L− 1). The pool stock
standard solutions were stored at room temperature in silanized ambar
glass flasks. Pool working standard solutions were prepared by diluting
appropriate amounts of the pool stock solution in ethanol.
Stock solution of the internal standard (IS) 1,4-butanediol (1,4-
BuOH) was prepared at the concentration of 10,000 mg.L− 1, which was
submitted to a 100-fold dilution to prepare the working solution of IS.

2.2. Instruments and apparatus

A tube shaker (IKA, Germany) and a centrifuge (Thermo Scientific,

Fig. 1. Molecular structures of ethylene glycol, diethylene glycol, propylene
Waltham, USA) were used in sample preparation. Micropipettes with
glycol and glycerol.
capacities of 50–5000 μL (Gilson, Middleton, USA) were employed in
several steps of the experimental procedure. Quantification was per­
Only one of these studies applied GC–MS (Lau & Weber, 1987).
formed using a gas chromatograph (GC system Shimadzu, model GC
More recent, another study in this field was published, which is an
2010 Plus, Japan) equipped with an auto sampler (Shimadzu, model,
analytical method by GC-FID to assay DEG, diethylene glycol monoethyl
AOC-5000 Plus, Japan) and coupled to a mass spectrometer (Shimadzu,
ether, coumarin and caffeine in some industrial food products
model GCMS-QP2010 Ultra, Japan). For the chromatographic runs, a
comprising soft drink and juice, infant formula and infant food, cereal,
polyethylene glycol capillary column (Restek Rtx-Wax, State College,
flour and snacks (Rahim et al., 2011). The scope of the method does not
USA) with dimensions of 30 m × 0.25 mm × 0.25 μm was used. The
include EG. The reported quantification limit values refer to the con­
software GCMS Solution, version 4.41 (Shimadzu, Japan), was used for
centration of analytes in the injected samples, not necessarily the con­
data acquisition and processing.
centration of analytes in the original samples, since it does not consider
the sample dilution that occurs in products with higher water content
2.3. Samples
during sample preparation. Solvent-based calibration curves were used,
but the possibility of matrix effect in different matrices in which the
Three blank beer samples were used in the validation experiments.
method was applied was not discussed.
For the occurrence study, 701 beer samples (from 67 different brands
To the best of our knowledge, no study has been reported in the peer-
and 128 different labels), provided by the Brazilian Ministry of Agri­
review literature dealing with the simultaneous determination of EG and
culture, Livestock, and Food Supply, were analyzed using the developed
DEG in food. The simultaneous quantification of both analytes was re­
analytical method. Most of the samples (87%) corresponded to one
ported in four studies in non-food matrices and only one of them applied
investigated brand (n = 612), and the others comprised different brands,
GC–MS (FDA, 2007; Holloway et al., 2010; Tokunaga, Osako, & Sato,
domestic or imported (n = 89). The samples were conditioned in a
2012; Self, 2013). A disadvantage of using GC-FID (Holloway et al.,
refrigerator (8 ± 2 ◦ C) until the time of analysis.
2010; Tokunaga, Osako, & Sato, 2012) is its lower selectivity compared
to GC–MS, which can be a problem in the analysis of more complex
2.4. Sample preparation
matrices. Also, the ranges of concentration tested in the analysis of toxic
glycols in non-food matrices are usually relatively high. Unlike phar­
For sample preparation, 1 mL of beer was transferred to a propylene
maceutical products and excipients, there is no maximum permitted
centrifuge tube and 4 mL of ethanol were added and well mixed. Sub­
limit for EG and DEG in food, so it is necessary low limits of detection
sequently, the mixture was centrifuged at 3739g (4000 rpm) for 5 min.
and quantification for an analytical method to be applied in food.
Then, 1 mL of the supernatant was transferred to a vial containing 0.1
Additionally, excipients are a cleaner matrix and do not present the same
mL of the internal standard solution (1,4-BuOH) at 100 mg.L− 1. The vial
analytical challenge as a food matrix.
was homogenized in a tube shaker. In the case of samples with high
Glycols have been sometimes used as refrigerant agents in different
content of the analytes, a second dilution step was employed, also using
stages of beer production and must be contained in closed systems,
ethanol as diluent, in order to keep the concentration of the analytes in
without any contact with the product. Therefore, the determination of
the sample extract inside the range of the calibration curve.
glycols in food is necessary in some circumstances. However, no studies
were found in the peer-review literature dealing with determination of
2.5. Mass spectrometry and chromatography optimization
EG or DEG in beer. In the present study, an analytical method for the
determination of these two contaminants in beer was developed and
Helium was used as carrier gas in the flow rate of 1 mL.min− 1. The
validated to be applied in analysis of 701 beer samples marketed in
temperature of the injector and detector was at 250 ◦ C and 230 ◦ C,
Brazil. The complete method involves the simple dilution of beer sam­
respectively. The oven temperature was set at 100 ◦ C for 1 min,
ples and subsequent quantification of the analytes by GC–MS. The hy­
increased to 250 ◦ C at 10 ◦ C.min− 1 and, then, held for 4 min in this
pothesis tested in this study is that the developed analytical method is
temperature. The samples were injected on split mode at a ratio of 1:25
reliable for detection and quantification of EG and DEG in beers.
and the volume of injection was 0.50 µL.
Some parameters of the gas chromatography and the mass spec­
2. Materials and methods
trometry (column temperature program, carrier gas flow rate, split ratio
and injection volume) were optimized by injection of standard solutions
2.1. Reagents, chemicals and stock solutions
of the analytes and the IS.
The MS detector was tuned with the standard spectrum autotune and
Standards of ethylene glycol, diethylene glycol (Sigma Aldrich,
the MS data were acquired in the Selected Ion Monitoring (SIM) mode
Overijse, Belgium) and 1,4-butanediol (Sigma Aldrich, Saint Louis, USA)
using electron ionization with an electron energy of 70 eV. The

2
L.R. Caldeira et al. Food Chemistry 346 (2021) 128871

monitored fragments were at m/z 61 (*), 31 and 43 for EG; at m/z 75 (*)
(Nx1 − 1)s2a1 + (Nx2 − 1)s2a2
and 45 for DEG; and at m/z 44 (*), 57 and 71 for 1,4-BuOH. The s2a1,a2 = (3)
instrumental response was the ratio of the peak chromatogram area of (Nx1 + Nx2 − 2)
quantifier’s fragment of the analyte and the peak chromatogram area of
(Nx1 − 1)s2b1 + (Nx2 − 1)s2b2
the fragment m/z 44 of the IS. The validated quantifier’s fragment ions s2b1,b2 = (4)
(Nx1 + Nx2 − 2)
were those marked with the symbol (*) and the other monitored frag­
ments were the qualifiers.
in which s2a1,a2 is the pooled variance of the intercepts, s2b1,b2 is the pooled
2.6. Validation variance of the slopes, Nx1 and Nx2 are the number of independent
concentration levels of the calibration curves.
Figures of merit (selectivity, linearity, matrix effect, accuracy, pre­ After the evaluation of the variances of the two compared curves, the
cision, limit of detection and limit of quantification) were estimated slopes and the intercepts of them were compared applying t-test at a
according to the validation protocol specified in guidelines of the Bra­ probability level of 95%.
zilian Ministry of Agriculture, Livestock and Food Supply (Ministry of In cases where the variances of solvent-based and matrix-matched
Agriculture, Livestock and Food Supply of Brazil, 2011, 2015) and calibration curves were equivalent, the t values were calculated as
Brazilian Health Regulatory Agency (Brazilian Health Surveillance described in Eqs. (5) and (6). In this case the critical t value was obtained
Agency, 2012, 2017). by a t Student distribution applying a probability level of 95% and de­
grees of freedom (ν) calculated in Eq. (7).
2.6.1. Selectivity |a1 − a2 |
Injection of standard solutions of EG, DEG and 1,4-BuOH into the ta,calc = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) (5)
chromatographic system were performed to verify the purity of these s2a1,a2 1
Nx1
+ N1x2
standards. Then, blank beer samples, fortified with these standards,
were also injected into the system, in order to determine the retention
time of the analytes. To perform selectivity test, three blank beer sam­ |b1 − b2 |
tb,calc = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) (6)
ples were injected to verify the presence of interfering compounds in the 1 1
s2b1,b2 +
same retention time of the analytes. Nx1 Nx2

2.6.1.1. Matrix effect. Solvent-based and matrix-matched calibration ν = Nx1 + Nx2 − 2 (7)
curves were built and compared to verify the presence of matrix effect.
To build the solvent-based calibration curves, from the 1000 mg.L− 1 in which a1 and a2 are the values of the intercepts of solvent-based and
pool standard stock solution, pool working standard solutions were matrix-matched calibration curves, and b1 and b2 are the values of the
prepared to give rise to the six different levels (3, 5, 10, 20, 30 and 50 slopes of solvent-based and matrix-matched calibration curves.
mg.L− 1). In the case where the variances were different, the calculated t value
To build the matrix-matched calibration curve, from the 10,000 mg. was obtained using the Eq. (8). In this case, the critical t value is ob­
L− 1 pool standard stock solution, pool working standard solutions were tained by a t Student distribution at a probability level of 95% and de­
prepared to fortify beer blank samples. The blank extracts were prepared grees of freedom calculated by the Eq. (9).
by applying the optimized extraction procedure, which employs a 5-fold |b1 − b2 |
dilution, in ethanol, of the blank beer sample followed by centrifugation tb,calc = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅ (8)
s2b1 s2
as previously described. Hence, 100 μL of each pool working standard Nx1
+ Nb2x2
solution were transferred to a 10 mL volumetric flask and its volume was
( )2
completed with the beer blank extracts. Each pool of working standard s2b1 /Nx1 + s2b2 /Nx2
solution had a different concentration to give rise to the six different ν= − 2 (9)
(s2b1 /Nx1 ) (s2b2 /Nx2 )
levels (3, 5, 10, 20, 30 and 50 mg.L− 1) of the matrix-matched calibration N +1
+ N +1
x1 x2

curve. Another approach used to evaluate matrix effect was the comparison
These curves were prepared in triplicate (three distinct days) and of the concentrations of fortified beer samples, obtained using solvent-
injected into the chromatographic system in triplicate. Matrix effect was based and matrix-matched calibration curves. Blank beer samples
evaluated by the comparison of the slopes and the intercepts of the were fortified in four concentration levels, in a minimum of 12 samples
solvent-based calibration curve with those of the matrix-matched cali­ per level. They were submitted to sample preparation and injected into
bration curve as described in Ministry of Agriculture, Livestock and Food the chromatographic system. Solvent-based and matrix-matched cali­
Supply of Brazil (2011). bration curves were also prepared and analyzed. Concentrations of the
First, the variances of the slopes and intercepts of the two types of samples estimated by the two different calibration curves were
curves were compared by F-test using the following equations (Eqs. (1) compared applying t-test, since the variances of the groups were not
and (2)): statistically different, as verified by F-test.
s2a1
F= (1) 2.6.2. Linearity
s2a2
The calibration curve covered the range of 3–50 mg.L− 1. Regression
s2b1 was performed by the weighted least squares method using dispersion as
F= (2) a weighting factor. This procedure was adopted due to the hetero­
s2b2
scedastic behavior of the instrumental response. The linear adjustment
in which s2a1 and s2a2 are the variances of the intercepts; s2b1 and s2b2 are the quality was evaluated by applying the t-test (Ministry of Agriculture,
variances of the slopes, being the bigger variance in the numerator. Livestock and Food Supply of Brazil, 2011). In this test the null hy­
In cases where the variances of solvent-based and matrix-matched pothesis (H0) is that in which the correlation coefficient is statistically
calibration curves were equivalent, the pooled variance was calculated equal to zero; and the alternative hypothesis (H1) is that in which this
using the Eqs. (3) and (4). coefficient is statistically different from zero. The experimental t value
was calculated as in Eq. (10).

3
L.R. Caldeira et al. Food Chemistry 346 (2021) 128871

L− 1.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅
|rw | Nx − 2
trw = √̅̅̅̅̅̅̅̅̅̅̅̅̅ (10) ( )
1 − r2w IR − a VF1 VF2
C= × × × CF Rec + CIP (12)
b VI1 VI2
in which rw is the correlation coefficient and Nx is the number of
experimental values for the calibration curve. where IR = instrumental response (peak area ratio between analyte and
If the calculated trw is higher or equal to the critical t value obtained IS); a = intercept of the calibration curve; b = slope of the calibration
by the two-tailed t distribution with 95% of confidence level and (Nx− 2) curve; VF1 = final volume of the first sample dilution with ethanol (mL);
degrees of level, the range of the calibration curve is considered linear. VI1 = volume measured of beer sample (mL); VF2 = final volume of the
second sample dilution with ethanol (mL); VI2 = initial volume of the
2.6.3. Accuracy and precision second sample dilution with ethanol (mL); CFRec = correction factor due
Accuracy was estimated via the attainment of recovery rates. Preci­ to systematic error of recovery; CIP = correction of intermediate
sion was assessed by the repeatability and intermediate precision re­ precision.
sults. For the recovery and accuracy evaluation, six blank samples were From Eq. (3), the coefficients of sensitivity for each of the identified
fortified and analyzed at four concentration levels and on three distinct uncertainty source were calculated according to Eqs. (13)–(20).
days. For the fortification experiments, beer blank samples were forti­
fied at concentrations 15, 25, 250 and 2000 mg.L− 1 of EG and DEG. All ∂C
= −
VF1 × VF2 × CF Rec
(13)
the samples were submitted to sample preparation protocol applying 5- ∂a b × VI1 × VI2
fold dilution with ethanol followed by centrifugation. The sample for­ ( )
tified at 2000 mg.L− 1 was diluted in a second step using 20-fold dilution
∂C VF1 × VF2 × CF Rec a − IR
= × (14)
in ethanol, since its initial concentration was above the upper limit of ∂b VI1 × VI2 b2
the validated calibration curve. This concentration was used in order to
∂C VF1 × VF2 × CFRec
include, in the validation procedure, the case of beer samples with high = (15)
∂IR b × VI1 × VI2
concentration of EG and/or DEG that would extrapolate the validated
linear range if only the standardized 5-fold dilution procedure was ∂C VF1 × VF2 × CF Rec × (a − IR)
performed in sample preparation. So, the concentration of the sample = (16)
∂VI1 b × VI1 2 × VI2
extracts injected into the chromatography system were 3, 5, 50 and 20
mg.L− 1, arising from the beer samples fortified at 15, 25, 250 and 2000 ∂C VF1 × VF2 × CF Rec × (a − IR)
mg.L− 1, respectively. = (17)
∂VI2 b × VI1 × VI2 2
Since there is no maximum tolerated level established for EG and
DEG in beer, generic acceptance criteria for the results of accuracy and ∂C VF2 × CF Rec × (IR − a)
= (18)
precision was used, as stated in the validation guide of Brazilian Ministry ∂VF1 b × VI1 × VI2
of Agriculture, Livestock and Food Supply (Ministry of Agriculture,
Livestock and Food Supply of Brazil, 2015). The mean recovery values ∂C VF1 × CF Rec × (IR − a)
= (19)
were considered acceptable only if the differences between the mean of ∂VF2 b × VI1 × VI2
the experimental values and the theoretical values were within ±20%.
For precision, the values calculated by the Horwitz equation (Eq. (11)) ∂C
=
(IR − a) VF1 VF2
× × (20)
were used as the limit for each fortification level. ∂CF Rec b VI1 VI2

RSDHorwitz = 2(1− 0.5logC)

(11) For each sample, the uncertainty contributions were combined,
generating the combined standard uncertainty (uc). It was decided to
where RSDHorwitz = Horwitz relative standard deviation; C = concen­ express the result with the expanded uncertainty (U), multiplying the
tration in the mass fraction. combined standard uncertainty by a coverage factor, calculated from the
effective degree of freedom and adopting a 95% confidence level.
2.6.4. Limit of detection (LOD) and limit of quantification (LOQ)
To establish the LOD, six beer blank samples were fortified with 3. Results and discussion
crescent levels of EG and DEG. The LOD was defined as the lowest
concentration in which all the six samples presented chromatographic 3.1. Method optimization
peaks with signal to noise ratio of 3 for all monitored fragments (m/z) of
the analytes. The developed procedure for sample preparation was simple and
The LOQ was determined experimentally as the lowest concentration fast. Two solvents were tested to dilute the beer samples – ethanol and
of the analytes that presents acceptable results of precision and accu­ methanol. Both solvents gave similar results in terms of analytes peak
racy. It was assumed that the expanded uncertainty in the LOQ must not area and peak shape (data not shown). The choice of ethanol was based
exceed 50% of the concentration result. on its characteristics: environmentally-friendly, less toxic and cheaper
than methanol.
For the instrumental step, 250 ◦ C was chosen for injector tempera­
2.7. Measurement uncertainty (uc) ture based on the high boiling point (244–245 ◦ C) of DEG. The influence
of the speed of temperature rise on the chromatography separation was
The uncertainty calculation was performed according to the Guide evaluated at three distinct values: 15 ◦ C.min− 1, 12 ◦ C.min− 1 and 10 ◦ C.
for the Expression of Measurement Uncertainty (Hall, 2008). min− 1. Although the speed of 10 ◦ C.min− 1 increases analysis time, it was
The measurement uncertainty was calculated using the bottom-up chosen because it was better for chromatography separation of the
methodology. The main sources of uncertainty concerning the vali­ analytes from other matrix compounds.
dated analytical method were those arising from the estimate of the It is well known that carryover is commonly associated with glycol
regression of the calibration curve, variation of instrumental response, analysis (Fu, Hao, Parker, & Knapp, 2011). Some procedures were
sample volume, procedure of dilution, repeatability, as well as correc­ applied in the analytical method to reduce the extent of this phenome­
tion factors for recovery and intermediate precision. Hence, the mea­ non. First, the importance of holding the temperature of 250 ◦ C for 4 min
surement uncertainty was calculated using the measurement function in the end of the GC oven program to proper elution of DEG was
(Eq. (12)), in which the analyte concentration (C) was expressed in mg.

4
L.R. Caldeira et al. Food Chemistry 346 (2021) 128871

Fig. 2. Extracted ion chromatograms (EIC) for each analyte and internal standard (IS) obtained under the optimized conditions of analysis.

observed. It avoids residues of non-evaporated DEG to stay in the in the retention times of the analytes.
capillary column. Second, the presence of analytes in the syringe of the
injector may be a cause of carryover. If the syringe is not properly 3.2.1. Matrix effect
cleaned between analyzes, carryover will cause inconsistent results. The results of the comparison between the slope and intercept values
Some rinsing procedures of the syringe were tested and a rinsing pro­ of solvent-based and matrix-matched calibration curves are presented in
tocol in which the syringe was rinsed with ethanol 3 times before in­ Supplementary Material (Table S1). As the analysis of these parameters
jection and 15 times after the injection was established. Another fact demonstrate, the equations of the curves were statistically different,
that favors carryover is the nature of beer, which is essentially an showing matrix effect.
aqueous matrix. Since the injected samples can contain up to 20% of The results obtained in the comparison of the fortified beer samples
water in their composition, a liquid with high expansion volume, it is concentration, estimated using these two different calibration curves,
common the occurrence of the backflash phenomenon. This occurs when also corroborates the occurrence of matrix effect (Table S2). The mean
the sample expansion volume exceeds the capacity of the liner, which values of EG and DEG concentrations, estimated by the solvent-based or
drives the vapor cloud out of it, resulting in poor sample transfer onto matrix-matched calibration curves, in all levels (except for DEG 50 mg.
the column (Fu, Hao, Parker, & Knapp, 2011). This can also cause L− 1), presented significant difference (p value <0.05).
carryover. In order to avoid the backflash phenomenon, the split mode
of injection was chosen, which reduces the amount of water injected into 3.3. Linearity
the chromatographic system. The split ratio was also optimized, being
defined as 1:25 to avoid backflash phenomenon without losing detect­ Linearity was performed on solvent-based and matrix-matched
ability. Based on the importance of reducing the amount of water calibration curves. Since significant matrix effect was demonstrated,
injected into the chromatographic system, it was not advisable to reduce the results presented here are only those obtained by matrix-matched
the dilution factor of the beer samples under 5-fold. The dilution of the calibration curve.
samples also contributes to minimize matrix effects and, also, to pre­ First, the presence of outliers was assessed upon the application of
serve the mass spectrometer clean for a longer period of time. the t-test defined in Ministry of Agriculture, Livestock and Food Supply
The optimized conditions (previously exposed) yielded chromato­ of Brazil (2011) guideline, using the Eq. (21):
grams of beer samples as those presented in Fig. 2. As can be observed,
the fragment ions monitored in the first 7 min of the chromatographic ty =
|ey |
(21)
run were those with m/z 45.0; 43.0, 61.0 and 31.0. From 7 min to the sres
final of the chromatographic run the monitored fragment ions were
those with m/z 44.0; 45.0, 57.0, 71.0 and 75.0. where |ey | is the instrumental residue of a specific replicate of the cali­
bration curve and sres is the sample standard deviation of the instru­
mental responses in the same concentration level.
3.2. Selectivity Since all calculated values of t were lower than the critical t value of
the two-tailed t distribution with 95% of confidence level and 5 degrees
Analysis of blank samples confirmed the method selectivity, since no of freedom (DG), no data was considered outlier.
interfering compounds were detected at a signal-to-noise ratio of up 3:1 A heteroscedastic behavior of the instrumental response variances

Table 1
Results for linearity, LOD, LOQ and uncertainty arising from the validation of the analytical method developed for the analysis of EG and DEG in beer samples.
Analyte Intercept Slope r r2 t-valuea LOD (mg.L− 1) LOQ (mg.L− 1) uc (max) (mg.L− 1) U (max) (%)

EG (m/z 61) 0.0022 0.0039 0.990 0.979 27.57 10.0 15.0 2.4 33
DEG (m/z 75) 0.0245 0.0625 0.993 0.987 34.86 5.0 15.0 2.3 28
a
tcritical = 2.120 (16 degrees of freedom; 95% of confidence level); LOD = limit of detection; LOQ = limit of quantification; U = maximum expanded measurement
uncertainty in LOQ level.

5
L.R. Caldeira et al. Food Chemistry 346 (2021) 128871

Table 2
Results for precision and accuracy arising from the validation of the analytical method applied to the analysis of EG and DEG in beer samples.
Figure of merit Analyte EG DEG
a a a a
Levels of concentration 1st level 2nd level 3rd level 4th level 1st levela 2nd levela 3rd levela 4th levela

Precision RSD (%) of Repeatability 1st day 6.4 8.6 3.9 2.1 5.2 7.9 5.1 4.0
2nd day 7.4 6.9 1.9 2.6 6.0 2.5 3.2 4.4
3rd day 6.4 7.6 3.6 2.2 6.3 6.0 3.2 3.7
RSD (%) of Intermediate precision 8.0 10.1 5.1 3.6 8.4 7.5 4.3 8.9
Acceptance criteria of Precision – Repeatability and 13.6 12.6 10.2 8.9 13.6 12.6 10.2 8.9
Intermediate precision: RSDHorwitz (%)

Accuracy Mean recovery (%) 91.9 94.8 85.9 94.0 94.1 100.8 100.5 108.9
Difference between experimental and theoretical 8.1 5.2 14.1 6.0 5.9 0.8 0.5 8.9
value (%)
Acceptance criteria to the difference between 20.0 20.0
experimental and theoretical value (%)b
a
Spiked concentration: 1st level = 3 mg.L− 1 in the vial or 15 mg.L− 1 in the beer, 2nd level = 5 mg.L− 1
in the vial or 25 mg.L− 1
in the beer, 3rd level = 20 mg.L− 1
in
the vial or 2000 mg.L− 1 in the beer, 4th level = 50 mg.L− 1 in the vial or 250 mg.L− 1 in the beer.
b
Acceptance criteria based on Brazil (2015).

was observed, so the weighted least squares method regression was quantifying ions, the qualifying ions were also detected at the LOD level.
applied on the experimental data. The quality of the linear fit was The LOQ values allowed the quantification of glycols in beer when
confirmed by the excellent correlation coefficients (r) obtained present in low concentrations, ensuring precision and accuracy of the
(>0.990), as displayed in Table 1. Even more important than deter­ result.
mining the correlation and determination coefficients, is the statistical
analysis of the significance of the regression. Moreover, the t-test applied 3.6. Measurement uncertainty
on each set of data confirmed the adequacy of the linear model since the
tvalue ≫ tcritical (2.120 for 16 degrees of freedom with 95% confidence The measurement uncertainty (uc) for each analyte was calculated
level) for each analyte (Table 1). using the bottom-up approach. Sensitivity coefficients were obtained by
applying Eqs. (13)–(21). Table 1 shows the expanded measurement
3.4. Accuracy and precision uncertainty (U) for each compound, calculated at the LOQ level. It
should be noted that the results obtained are in conformity with the
For accuracy, all recovery rates obtained herein fall in the acceptance maximum expanded uncertainty assumed in percentage terms. The
criteria (Table 2). To confirm the analytical performance, Horwitz uncertainty estimation model showed that, regardless of the concen­
relative standard deviation (RSDHorwitz) values were used as the limit for tration of EG and/or DEG in beer samples, the main uncertainty
the RSD in the precision estimate. It can be observed (Table 2) that contribution was the uncertainty due to the non-correction of the result
repeatability and intermediate precision were within the limits recom­ by the recovery, followed by the intermediate precision uncertainty and,
mended at the four levels for each compound. to a lesser extent, repeatability. These three sources of uncertainty
together contributed between 90 and 100% of the total uncertainty.
3.5. LOD and LOQ
3.7. Occurrence study
The experimental LOD and LOQ values, shown in Table 1, demon­
strate the high detectability of the analytical method. The LOD values For the occurrence study, 612 beer samples of one investigated brand
enabled the detection of glycols in beers when present in low concen­ were analyzed using the analytical method developed and 51 of them
trations, while avoiding false positive results, since in addition to the presented at least one glycol, which represent about 8% of these

Fig. 3. Number of beer samples, contaminated with at least one of the glycols, classified in levels of concentrations of EG and DEG.

6
L.R. Caldeira et al. Food Chemistry 346 (2021) 128871

Fig. 4. Ascending concentration of EG and DEG in each beer samples contaminated with at least one of glycol.

samples. Among these, six samples were contaminated only with DEG at glycols. It is important to highlight that the collection of samples was not
low level (five of them under LOQ) and three samples were contami­ representative of the Brazilian beer consumption market, since the
nated only with EG at low level. The other 42 contaminated samples performance of the analysis had the purpose of inspection, in such a way
showed both EG and DEG at different levels of concentration. Concen­ that the majority of the analyzed beers belonged to one investigated
trations of EG and DEG in beer samples contaminated with at least one brand. Positive results for the presence of EG and/or DEG were found in
glycol were stratified by level and the number of samples present in each only this brand, related to a specific failure in the production of beers.
class is shown in Fig. 3. This was the first case of DEG and EG contamination in foods consumed
Beer samples that presented at least one glycol above the LOQ are in Brazil. The developed analytical method was essential for the inves­
also shown in Fig. 4, in ascending order of the sum of the concentration tigation of the case in question, providing subsidies for the decision of
of EG and DEG, showing the level of each glycol and it expanded mea­ the inspection organizations, aiming to guarantee food safety.
surement uncertainty. It is noticed that most contaminated samples
presented concentration of total glycols between 15 and 500 mg.L− 1, but 4. Conclusions
higher concentrations were found, with the maximum value found being
9677 mg.L− 1 for the sum of EG and DEG. A method that employs a simple sample preparation procedure
Despite these findings, one can assume that beer consumption in combined with GC–MS was developed for the simultaneous determina­
Brazil is safe, since 93% of samples analyzed – including 92% (n = 561) tion of ethylene glycol and diethylene glycol in beer. The entire pro­
of beer samples of the investigated brand and all the beer samples of the cedure was validated according to the directive from Brazilian Ministry
other 66 different brands (n = 89) – did not present any of these toxic of Agriculture, Livestock and Food Supply, since this is the organization

7
L.R. Caldeira et al. Food Chemistry 346 (2021) 128871

responsible for quality control and safety of beverages in Brazil. In References

conclusion, the results obtained in the validation procedure showed that
the analytical method is selective, sensitive, reliable and precise to be Brazilian Health Surveillance Agency. (2012). http://bvsms.saude.gov.br/bvs/saudeleg
is/anvisa/2012/rdc0027_17_05_2012.html/ [Accessed 20 January 2020].
applied in real samples. Thus, the hypothesis tested in this article was Brazilian Health Surveillance Agency. (2017). http://portal.anvisa.gov.br/documents/
confirmed to be true. Moreover, the matrix-matched calibration curves 10181/2721567/RDC_166_2017_COMP.pdf/d5fb92b3-6c6b-4130-8670-4e3263763
shown a linear behavior for both analytes in a wide dynamic range. It 401/ [Accessed 20 January 2020].
Caccamo, F., Di Coarcia, A., & Samperi, R. (1986). Rapid purification with Carbopack B
was also found that the main uncertainty sources of the method were the cartridge and quantitation by gas chromatography. Journal of Chromatography, 354,
non-correction of the result by the recovery, followed by the interme­ 478–481.
diate precision uncertainty and, to a lesser extent, repeatability. This FDA. Lab Method for Gas Chromatography - Mass Spectrometry (GC-MS) Screening
Procedure for the Presence of Diethylene Glycol and Ethylene Glycol in Toothpaste,
analytical method proved to be suitable for analyzing real beer samples 2007. https://www.fda.gov/food/laboratory-methods-food/lab-method-ga
marketed in Brazil. Positive results for the presence of EG and/or DEG s-chromatography-mass-spectrometry-gc-ms-screening-procedure-presence-dieth
were found in 8% of beer samples of only one investigated brand, related ylene-glycol/ [Accessed 14 April 2020].
Fowles, J., Banton, M., Klapacz, J., & Shen, H. (2017). A toxicological review of the
to a specific failure in the production of beers. This was the first case of
ethylene glycol series: Commonalities and differences in toxicity and modes of
DEG and EG contamination in foods consumed in Brazil. All beer sam­ action. Toxicology Letters, 278, 66–83. https://doi.org/10.1016/j.
ples tested, which were not related to the suspect brand, provided toxlet.2017.06.009.
negative results, showing that they are safe for consumption. Fu, Y., Hao, Z., Parker, B., & Knapp, M. (2011). A greener way to screen toothpaste for
diethylene glycol. American Journal of Analytical Chemistry, 2, 938–943.
Hall, B. D. (2008). Evaluating methods of calculating measurement uncertainty.
CRediT authorship contribution statement Metrologia, 45, L5–L8. https://doi.org/10.1088/0026-1394/45/2/N01.
Hernández, F., Ibáñez, M., & Sancho, J. V. (2008). Fast determination of toxic diethylene
glycol in toothpaste by ultra-performance liquid chromatography–time of flight mass
Leila Rodrigues Caldeira: Conceptualization, Methodology, Soft­ spectrometry. Analytical and Bioanalytical Chemistry, 391, 1021–1027. https://doi.
ware, Validation, Formal analysis, Investigation, Data curation, Writing org/10.1007/s00216-008-2082-0.
- original draft, Writing - review & editing. Fernando Diniz Madureira: Holloway, G., Maheswaran, R., Leeks, A., Bradby, S., & Wahab, S. (2010). Screening
method for ethylene glycol and diethylene glycol in glycerin-containing products.
Conceptualization, Methodology, Software, Validation, Formal analysis, Journal of Pharmaceutical and Biomedical Analysis, 51, 507–511. https://doi.org/
Investigation, Data curation. Thalita De Faria Maia: Conceptualization, 10.1016/j.jpba.2009.08.025.
Methodology, Software, Validation, Formal analysis, Investigation. Kaiser, R. E., & Rieder, R. I. (1985). Diethylene gycol in wine. Journal of High Resolution
Chromatography & Chromatography Communications, 8, 863–866.
Carlos Vitor Muller: Conceptualization, Resources, Writing - original Lau, B. P. Y., & Weber, D. (1987). Determination of diethylene glycol in wines by high-
draft, Writing - review & editing, Supervision, Project administration, resolution gas chromatography/mass spectrometry. Journal of Agriculture and Food
Funding acquisition. Christian Fernandes: Conceptualization, Meth­ Chemistry, 412–417. https://doi.org/10.1021/jf00075a032.
Lawrence, J. F., Chadha, R. K., Lau, B. P. Y., & Weber, D. F. (1986). Simplified routine
odology, Resources, Writing - original draft, Writing - review & editing,
method for the determination of diethylene glycol in wines by capillary gas
Supervision, Project administration, Funding acquisition. chromatography with flame ionization detection. Journal of Chromatography A,
213–216.
Declaration of Competing Interest López-Sánchez, M., Dominguéz-Vidal, A., Ayora-Cañada, M. J., & Molina-Díaz, A. (2008).
Assesment of dentifrice adulteration with diethylene glycol by means of ATR-FTIR
spectroscopy and chemometrics. Analytica Chimica Acta, 620, 113–119. https://doi.
The authors declare that they have no known competing financial org/10.1016/j.aca.2008.05.032.
interests or personal relationships that could have appeared to influence Marraffa, J., Holland, M., Stork, C., Hoy, C., & Hodgman, M. (2007). Diethylene glycol:
Widely used solvent present serious poisoning potential. Journal of Emergency
the work reported in this paper. Medicine, 35(4), 401–406. https://doi.org/10.1016/j.jemermed.2007.06.025.
Ministry of Agriculture, Livestock and Food Supply of Brazil. (2011). Available in: https
Acknowledgements ://bibliotecaquimicaufmg2010.files.wordpress.com/2012/02/mapa-2011-manual
-de-garantia-da-qualidade-analitica.pdf [Accessed 20 January 2020].
Ministry of Agriculture, Livestock and Food Supply of Brazil. (2015). Available in:
The authors thank the Brazilian Ministry of Agriculture, Livestock http://www.agricultura.gov.br/assuntos/laboratorios/arquivos-publicacoes-labora
and Food Supply for the financial support, for the beer samples collected torio/manual-de-garantia-qualidade-analitica.pdf. [Accessed 20 January 2020].
Rahim, A. A., Saad, B., Osman, H., Hashim, N., Yahya, S., & Talib, K. M. (2011).
and for the authorization to publish the scientific results. The authors Simultaneous determination of diethylene glycol, diethylene glycol monoethyl ether,
also thank the Brazilian agencies FAPEMIG, CAPES and CNPq for the coumarin and caffeine in food items by gas chromatography. Food Chemistry, 126,
financial support. 1412–1416. https://doi.org/10.1016/j.foodchem.2010.11.121.
Self, R. L. (2013). Direct analysis in real time-mass spectrometry (DART-MS) for rapid
qualitative screening of toxic glycols in glycerin-containg products. Journal of
Appendix A. Supplementary data Pharmaceutical and Biomedical Analysis, 80, 155–158. https://doi.org/10.1016/j.
jpba.2013.02.037.
Supplementary data to this article can be found online at https://doi. Tokunaga, H., Osako, T., & Sato, K. (2012). Determination of ethylene glycol and
diethylene glycol as the adulterant in concentrated gycerin and propylene glycol.
org/10.1016/j.foodchem.2020.128871. Journal of Japanese Cosmetic Science, 36, 265–275. https://doi.org/10.11469/
koshohin.36.269.