Food Analytical Methods (2022) 15:728–738

https://doi.org/10.1007/s12161-021-02152-8

Spectrofluorimetric Determination of Phenylalanine in Honey

by the Combination of Standard Addition Method and Second‑Order
Advantage
Daphne Chiara Antônio1 · Bruno Gonçalves Botelho1 · Marcelo Martins Sena1,2

Received: 14 April 2021 / Accepted: 26 October 2021 / Published online: 4 November 2021
© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2021

Abstract
The development of rapid, reliable, and environmentally friendly analytical methods for food analysis is increasingly required.
In this sense, the combination of spectrofluorimetry and second-order calibration is a promising alternative to the determina-
tion of naturally fluorescent analytes in food matrices, as compared to the laborious and time-consuming chromatographic
methods. This work combined the chemometric second-order method parallel factor analysis (PARAFAC) with spectrofluor-
imetry aiming to develop a new analytical method to quantify phenylalanine in honey samples. Honey samples of different
botanical and geographical origins were diluted with deionized water, and their excitation-emission matrices were recorded
between 280–350 nm (excitation) and 360–490 nm (emission). Due to the presence of matrix effect, second-order standard
addition was employed. This methodology utilizes the second-order advantage, which allows to quantify an analyte in the
presence of uncalibrated/unmodeled interferences. Phenylalanine concentrations were estimated in the range from 5.7 to
32.5 mg ­kg−1. Method validation was performed by verifying the agreement between the results of the developed method
and from an independent methodology based on liquid chromatography. In addition, proper figures of merit were estimated
for the proposed method, such as sensitivity, analytical sensitivity, limits of detection, and quantification.

Keywords Honey · Phenylalanine · Amino acid · Molecular fluorescence · Three-way analysis · Chemometrics

Introduction Honey presents several pharmacological properties,

such as antioxidant, antimicrobial, antidiabetic, and anti-
Honey is a complementary food, beneficial to human health, inflammatory activity, and wound healing (Leyva-Jimenez
and its composition varies according to botanical origin, et al. 2019). Pharmacological properties are frequently
agricultural practices, and climatic and environmental con- attributed to honey minor compounds, which are usu-
ditions (Kavanagh et al. 2019; Noviyanto and Abdulla 2020). ally present at concentration levels below 0.001 g L ­ −1.
It is composed of carbohydrates, water, and other minor Therefore, sensitive techniques able to detect and quan-
compounds, such as organic acids, amino acids, proteins, tify these compounds are required. Analytical techniques
minerals, vitamins, and lipids. The main amino acids in most applied to the determination of organic compounds
honey are phenylalanine, proline, and glutamic acid (Pereira in honey are gas and liquid chromatographies (Leyva-
et al. 2017; Roldán et al. 2011; Rebane and Herodes 2010; Jimenez et al. 2019; Shen et al. 2019; Allenspach et al.
Iglesias et al. 2006). 2018; Chen et al. 2016; Otter 2012; Kelly et al. 2010).
Despite their high sensitivity and selectivity, chromato-
graphic techniques are expensive, consume solvents, and
* Marcelo Martins Sena require laborious sample preparation and well-trained pro-
[email protected] fessionals, resulting in low throughput analysis. On the
1 other hand, spectroscopic techniques have been exploited
Departamento de Química, Instituto de Ciências Exatas
(ICEx), Universidade Federal de Minas Gerais (UFMG), as simple, rapid, and environmentally friendly analytical
Belo Horizonte, MG 31270‑901, Brazil alternatives, since they allow simultaneous multicompo-
2
Instituto Nacional de Ciência E Tecnologia Em Bioanalítica, nent analysis with minimum sample preparation. Specifi-
Campinas, SP 13083‑970, Brazil cally, molecular fluorescence spectroscopy is a simple,

13
Food Analytical Methods (2022) 15:728–738 729

fast, and sensitive technique that can be applied to quali- trilinear factors, usually based on the iterative alternating
tative and quantitative analysis of fluorophores, molecules least squares (ALS) algorithm (Bro 1997).
that show fluorescence, through the relationship between The determination of amino acid profile using spectro-
absorbed and emitted energy (Lakowicz 2006). fluorimetry has been explored for the study of the botanical
Over the last decades, fluorescence spectroscopy has and geographical origin of food and for the detection of food
become popular for the analysis of food, environmental and fraud (Strelec et al. 2018; Siddiqui et al. 2017; Lenhardt
biological samples, mainly due to its wide linear response et al. 2015; Sergiel et al. 2014; Ruoff et al. 2005). Honey
range, high sensitivity and selectivity, and low cost in com- contains several intrinsic fluorophores, such as proteins,
parison to other alternatives (Christensen et al. 2006; Lako- peptides, and free amino acids, which can be used as mark-
wicz 2006). Its potential has been improved when combined ers for characterization studies (Lenhardt et al. 2015; Ruoff
with chemometric methods. The combination of spectro- et al. 2005). Several articles have determined botanical ori-
fluorimetry and chemometrics has been applied to direct gin and checked authenticity of honey samples using dif-
determinations in honey and other food matrices (Lenhardt ferent analytical techniques (Sharin et al. 2021; Shen et al.
et al. 2015; Lenhardt et al. 2017; Sádecká and Tóthová 2007; 2019; Kavanagh et al. 2019; Lenhardt et al. 2015); however,
Ruoff et al. 2005). Fluorescence spectra can be recorded at most of them focused on discriminant analysis by compar-
a single fixed excitation wavelength (emission spectra) or ing honey profiles instead of quantifying specific analytes.
at a single fixed emission wavelength (excitation spectra). Among the main amino acids present in honey, phenylala-
When both wavelength ranges are simultaneously varied, an nine is the only intrinsic fluorophore. Thus, it can be directly
excitation-emission matrix (EEM) is obtained for each sam- determined by fluorescence spectroscopy, which represents
ple. Then, these EEM can be arranged in a three-way array a promissory alternative for honey characterization. Pheny-
and processed by multiway models. Multiway or second- lalanine is an essential aromatic amino acid; that is, it must
order models take advantage of the whole chemical infor- be ingested by diet and cannot be synthesized by the human
mation contained in EEM and in many situations provide body. It can be found in two isomer forms, d-phenylalanine
more predictive classification and calibration models than and l-phenylalanine (Fig. 1), the latter being the most com-
first-order models built with only emission (or excitation) mon in food, such as meat, fish, egg, honey, peanut, and
spectra. Thus, this type of chemometric model is a suitable milk derivatives (Perkowski and Warpeha 2019). In addition,
approach in food analysis, especially for matrices of great phenylalanine has been determined as a discriminant marker
complexity (Christensen et al. 2006). to trace the floral origin of rosemary, eucalyptus, lavender,
Multiway calibration methods can be applied to pro- thyme, and orange blossom honeys, among others (Biluca
cess higher order spectrofluorimetric data arrays, allow- et al. 2019; Chen et al. 2016; Rebane and Herodes 2008;
ing to determine the number of fluorescent components, to Hermosín et al. 2003).
estimate their pure spectra, and to determine their relative Thereby, the objective of this article was to develop a
concentrations in each sample. PARAFAC (parallel factor new direct spectrofluorimetric method for the quantification
analysis) has been the most used method for this task (Bro of phenylalanine in polyfloral honey using PARAFAC. The
1997). Other methods, such as multivariate curve resolution- content of this natural component is an important parameter
alternating least squares (MCR-ALS) (de Juan and Tauler that contributes to the characterization of honeys in rela-
2021), and N-way partial least squares (N-PLS) (Bro 1996), tion to their pharmacological properties, their botanical and
have also been employed. PARAFAC has been the most used geographical authentication, and the possible detection of
chemometric method for the multiway processing of spectro- frauds. Due to the presence of a matrix effect that changes
fluorimetric data in general, and particularly in food analy- the fluorescence signal of phenylalanine depending on the
sis (Christensen et al. 2006; Ríos-Reina et al. 2017; Rubio honey composition, a strategy of second-order standard
et al. 2019; Lobo-Prieto et al. 2020). Briefly, PARAFAC is addition was adopted. Considering the variability in honey
a method that decomposes three-dimensional data into three composition, an important advantage of this strategy is the

Fig. 1  Structural formula of

phenylalanine optical isomers.
a d-Phenylalanine. b l-Pheny-
lalanine

13
730 Food Analytical Methods (2022) 15:728–738

possibility to quantify the analyte in the presence of uncali- Creek, CA, USA) equipped with two Czerny-Turner mono-
brated interferences. This chemometric strategy, which has chromators, and a Xenon discharge lamp pulsed at 80 Hz
been applied to clinical analysis (Sena et al. 2006; Bernardes with a half peak height of 2 ls (peak power equivalent to
et al. 2010; Privitera and Lozano 2017), is applied here for 75 kW). A high-performance R298 photomultiplier tube
the first time to food analysis. The developed method is detector was used for the collection of the fluorescence
a clean, low-cost, and rapid alternative to the traditional spectra. The temperature was hold at 25.00 ± 0.05 °C using
protocols. a Peltier thermostat. The spectrometer was interfaced with
a computer with Cary Eclipse 1.1 (132) scan software for
spectral acquisition and exportation. After an initial opti-
Materials and Methods mization step, EEM were selected by varying the excita-
tion wavelength (λex) ranging between 280 and 350 nm (step
Materials 10 nm) and recording the emission spectra (λem) from 360
to 490 (step 1 nm). Excitation and emission slits were both
All the utilized solvents were from Sigma-Aldrich HPLC set at 5 nm, and the scan rate was fixed to 600 nm ­min−1.
grade. All chemicals were analytical-reagent grade and Rayleigh scatter was removed from all PARAFAC models
utilized without further purification. The l-phenylalanine through the selection of a proper working spectral region in
standard (99%) was obtained from Synth (São Paulo, Brazil). EEM. The characteristic region of elastic scattering provides
Ultrapure deionized water was used throughout (Milli-Q, a non-trilinear analytical signal unrelated to the sample com-
Merck, Darmstadt, Germany). position. Therefore, this region should be removed before
building multilinear PARAFAC models (Bro 1997).
Sample Preparation
Phenylalanine Quantification
Honey samples were purchased from local markets and
stored at room temperature (25 °C) in the original pack-
Quantification of phenylalanine was carried out by the stand-
ages until analysis. Nine samples from different origins
ard addition method, in a situation of variable total volume
and brands were analyzed. Table 1 presents botanical and
with continuous variation of standard (Bader 1980). Four
geographical origins of the honey samples certified by the
constant volumes of the standard solution were directly
respective producers. Aliquots of 2.00 g of each sample were
added in the cuvette, thus simplifying the experimental pro-
placed into a 50-mL polypropylene tubes with 40.00 mL of
cedure, and eliminating the need to prepare one solution for
deionized water. Sample solutions were homogenized and
each addition. In this case, the standard addition curve is
stored at 4 ± 2 °C until analysis by molecular fluorescence
based on Eq. (1) (Bader 1980):
spectroscopy. A standard solution of l-phenylalanine (Synth,
São Paulo, Brazil) 20 mg L ­ −1 was also prepared in deion-
[ ]
Vx Cx NVs Cs
ized water. R=k + (1)
Vx + NVs Vx + NVs

Fluorescence Analysis where R is the instrumental response, here the sample scores
of phenylalanine obtained by PARAFAC, Vx and Cx are the
Fluorescence measurements were performed using a Varian initial volume and concentration of the sample, Cs is the
fluorescence spectrophotometer model Cary Eclipse (Walnut concentration of the standard solution, Vs is the volume

Table 1  Botanical and Samples Botanical origin Geographical origin

geographical origins of the
honey samples analyzed in this S1 Polyfloral honey National Beekeeping Cooperative, Minas Gerais, Brazil
study, certified by the producers
S2 Polyfloral honey Southeast of Minas Gerais, Minas Gerais, Brazil
S3 Polyfloral honey National Beekeeping Cooperative, Minas Gerais, Brazil
S4 Polyfloral honey National Beekeeping Cooperative, Minas Gerais, Brazil
S5 Wildflower honey—honey produced National Beekeeping Cooperative, Minas Gerais, Brazil
with nectar from different flowers
S6 Wildflower honey Midwest of Minas Gerais, Brazil
S7 Polyfloral honey National Beekeeping Cooperative, Minas Gerais, Brazil
S8 Polyfloral honey National Beekeeping Cooperative, Minas Gerais, Brazil
S9 Predominantly wild flowering honey Santa Barbara, Minas Gerais, Brazil

13
Food Analytical Methods (2022) 15:728–738 731

of each standard addition, N is the number of the addition Data Analysis and Software
(from 0 to 4), and k is a constant of proportionality. For
practical applications, the above equation is rearranged as Data were processed using Matlab software, version 7.10.0
(Vx + NVs)R = kVxCx + NkVsCs, and the entire left hand of (R2010a) (MathWorks, Natick, USA), along with the PLS_
this expression is plotted versus N, resulting in a straight line Toolbox package, version 6.71 (Eigenvector Technologies,
with intercept kVxCx and slope kVsCs. Manson, USA). The graphical interface Multivariate Cali-
Each honey sample was analyzed in triplicate. For bration 2 (MVC2), written by Olivieri et al. (2009), was
each measurement, 2.00 mL of a diluted honey sample employed for estimating analytical figures of merit and con-
was placed into a 10.00-mm quartz cuvette containing centration standard errors. This freely available toolbox for
a micro magnetic bar, and the spectrum was recorded. MATLAB allows data processing using several second-order
In the sequence, four successive additions of 10 µL of multivariate calibration methods, such as PARAFAC, MCR-
a 20 mg L ­ −1 l -phenylalanine solution were performed. ALS, and N-PLS.
After each addition, samples were homogenized under
magnetic stirring, and the respective spectra recorded. Theoretical Framework: PARAFAC and Second‑Order
Figure 2 shows a scheme summarizing the experimental Standard Addition Method
procedure adopted for the developed method.
The structural basis of the PARAFAC model is represented
by Eq. (2):
HPLC Analysis ∑F
𝐗ijk = a b c
f =1 if jf kf
+ eijk (2)
Chromatographic measurements were carried out on a
liquid chromatograph aiming at amino acid quantification where each element of X (ijk) will represent the fluorescence
based on the procedure published by Allenspach et al. intensity of ith sample at the excitation wavelength j and the
(2018). A liquid chromatograph Alliance HT Waters emission wavelength k. This three-way array is decomposed
model e2795 (Milford, MA, USA) in-line degasser cou- into a set of three vectors corresponding to each of the three
pled to a fluorescence detector Waters 2475 was utilized. dimensions, respectively. The elements aif are collected in
A reversed‐phase Nova-Pak C18 column with an average the vector a(I,F), containing the I sample scores relative to
particle size of 4 µm and dimensions of 150 mm × 4 mm each fth fluorophore/factor (F is the total number of factors).
and a fluorescence detector set at λexc 260 nm and λ em Similarly, b(jf) and c(kf) vectors contain excitation and emis-
300 nm were used. A 200 μL aliquot of each sample was sion loadings representing the respective spectral profiles for
dissolved with 0.1% (v/v) TFA (Sigma-Aldrich, MO, each fluorophore. Finally, data variance not captured by the
USA) in HPLC‐grade water: 0.1% (v/v) TFA in ace- model is collected in the elements of the residual vector e(ijk).
tonitrile (30:70 v/v). The gradient program was an iso- One of the most advantageous features of PARAFAC
cratic elution with 85% of 0.1% (v/v) TFA in water–15% for spectral curve resolution is the concept of uniqueness.
0.1% (v/v) TFA in acetonitrile for 25 min with a flow This mathematical property permits the model to provide
rate of 0.3 mL ­min−1. The injection volume was 4 μL at unique solution for data decomposition, which is free of
30 ± 5 °C. rotation ambiguity. Because of the uniqueness, it is possi-
ble to determine the number of fluorophores and obtain their
pure spectra provided that the data is actually trilinear, the

Fig. 2  Scheme summarizing the

experimental procedure for the
developed spectrofluorimetric
method

13
732 Food Analytical Methods (2022) 15:728–738

correct number of factors is chosen, and the signal-to-noise order to calibrate the variance of all the possible interfer-
ratio is appropriate. A more detailed discussion about the ences. The other main alternative, MCR-ALS, is also able
uniqueness of PARAFAC models can be found elsewhere to provide second-order advantage, but has the disadvantage
(Bro 1997, 1998). of applying bilinear data decomposition to trilinear data.
The use of first-order calibration methods, such as in tra- Consequently, PARAFAC has the advantage of providing
ditional multivariate calibration with PLS, allows the quan- unique solutions (uniqueness), while MCR-ALS decompo-
tification of a given analyte without the need of physical sition suffers from a certain degree of ambiguity (de Juan
separation, provided that the interferences must be present and Tauler 2021).
in the calibration set. The use of some second-order cali-
bration methods, such as PARAFAC, provides the second-
order advantage. Second-order advantage is a key aspect Results and Discussion
in the chemometric methodology used in this study, since
it allows the mathematical separation of the analyte signal Preliminary Exploratory Analysis
and its determination in the presence of unmodeled sample
components that are not included in the calibration set. This Aiming to assess the most appropriate excitation-emission
ability of determining analytes in the presence of unknown wavelength range for this analysis, a preliminary scanning
interferences is attractive from the point of view of green was performed based on the parameters found in the lit-
chemistry (Olivieri and Escandar 2019). erature (Stanković et al. 2019; Bong et al. 2016; Lenhardt
The standard addition method is traditionally used in et al., 2015; Chen et al. 2014; Sádecká and Tóthová 2007).
univariate calibration aiming to cope with matrix effects Figure 3 shows the EEM of sample S2 as a contour map
that change the analyte signal from sample to sample. The and as emission spectra at different excitation wavelengths
combination of standard addition and multiway methods (excitation range of 250–350 nm, and emission range of
that provide second-order advantage resulted in the so- 300–800 nm). It can be noted that the honey fluorescence
called second-order standard addition method (SOSAM). band is mainly contained in the emission range between 300
The original version of SOSAM was proposed in 1995 by and 550 nm. Thus, the emission wavelength range between
applying direct trilinear decomposition (DTD) to kinetic- 360 and 490 nm was chosen for honey analysis aiming to
spectrophotometric data (Booksh et al. 1995). In the last provide a rapid procedure, and to avoid the scattering region
years, PARAFAC has been shown the best multiway method
to be combined with the standard addition method in appli-
cations such as the determination of drugs in complex bio-
logical matrices (Sena et al. 2006; Bernardes et al. 2010).
This combination has also been applied to food matrices,
such as in the determinations of sulphaguanidine residues in
honey (Muñoz de la Peña et al. 2007) and carminic acid in
fruit juices (Samari et al. 2010), though employing a more
laborious standard addition strategy, a constant total volume
procedure that required individual solutions for each concen-
tration level of the analytical curve.
In the developed method for the quantification of phe-
nylalanine in honey samples of different origins, standard
addition was required to cope with the observed matrix
effect (as discussed in the subsection “PARAFAC Models”
of “Results and Discussion”), while second-order advantage
was demanded to cope with the presence of different uncali-
brated interferences without the need of physical separation.
PARAFAC was chosen because of its advantages over the
main multiway alternatives. N-PLS has a faster algorithm
and incorporates the dependent variables (phenylalanine
concentration) in the independent/spectral variables. How-
ever, N-PLS does not provide second-order advantage and Fig. 3  Excitation-emission matrix (EEM) of sample S2, obtained in a
preliminary exploratory analysis, as a a contour map and b emission
therefore cannot be used in our methodology (Bro 1996,
spectra at different excitation wavelengths. Excitation and emission
1998). For this same type of application, robust N-PLS wavelength ranges between 250–350 nm and 280–800 nm, respec-
models would require dozens or hundreds of samples in tively

13
Food Analytical Methods (2022) 15:728–738 733

between 300 and 350 nm and the second-order diffraction 2019; Lenhardt et al. 2015). Another spectral band present
above 490 nm. The excitation wavelength range was set in all the samples is in the emission range between 360 and
between 280 and 350 nm with a step of 10 nm, aiming to 400 nm, at excitation wavelengths from 280 to 300 nm.
avoid as much as possible the scattering region and provid- This analytical signal may be assigned to aromatic amino
ing a short time of analysis (Lakowicz 2006; Christensen acids (Stanković et al. 2019; Lenhardt et al. 2015). In the
et al. 2006; Ruoff et al. 2005). sequence, PARAFAC models were built to determine the
number of fluorescent patterns and to estimate their resolved
Fluorescence Fingerprints spectra.

Figure 4 shows the EEM recorded for honey samples as PARAFAC Models
contour maps after removing Rayleigh scattering. As can
be observed, the emission patterns and intensities may vary Three individual PARAFAC models were built for each
among samples due to their different botanical origins as sample, one for each replicate. For each model, three-way
previously reported in the literature (Lenhardt et al. 2015; data arrays of dimensions 5 (measurements) × 130 (emission
Chen et al. 2014; Sádecká and Tóthová 2007). wavelengths) × 8 (excitation wavelengths) were analyzed.
A spectral band is present in all the honey samples in In these arrays, the first mode stands for number of meas-
the excitation range between 300 and 350 nm and in the urements, corresponding to the original sample plus four
emission range between 400 and 470 nm. This region may standard additions, and the second and third modes represent
be assigned to phenolic compounds that have often been the number of emission and excitation wavelengths, respec-
associated to the botanical origin of honeys (Stanković et al. tively. Non-negativity constraint was applied to all the three

Fig. 4  Fluorescence contour maps obtained for honey samples after removing Rayleigh scattering. a S1. b S2. c S3. d S4. e S5. f S6. g S7. h S8.
i S9

13
734 Food Analytical Methods (2022) 15:728–738

modes. The number of factors for each model was deter- in shifts of the spectral bands to neighboring wavelengths.
mined based on several criteria, such as the relation between This results in a matrix effect that prevents the use of exter-
the number of factors and the core consistency diagnostic nal calibration curves to quantify phenylalanine in those
(CORCONDIA) values (Bro and Kiers 2003), observation matrices.
of estimated loadings/resolved spectra, split-half analysis, The second factor (red solid lines in Fig. 5b, c) shows
and model residuals (Bro 1997). excitation maximum around 350 nm, and emission maxi-
Since each sample can show specific spectral features mum at 443 nm. This factor may be assigned to phenolic
(Fig. 4) and due to the presence of matrix effect, the num- compounds that constitute the fluorescence background of
ber of factors may change from sample to sample. For all honey samples. A variety of phenolic compounds are present
the samples, the best PARAFAC models were selected with in honey, such as hydroxycinnamic acids, chlorogenic acid,
two or three factors, with the corresponding CORCONDIA caffeic acid, coumarins, and stilbenes. Phenolic compounds
values varying from 62 to 99%, and with mean of 73%. All have been considered honey markers and their concentra-
these models accounted for at least 98% of the spectral data tions vary widely according to the floral origins (Kavanagh
variance. Figure 5 shows excitation and emission loadings et al. 2019; Stanković et al. 2019; Bong et al. 2016; Ser-
estimated by PARAFAC for one replicate of one of the ana- giel et al. 2014). Thus, spectral profiles can vary depend-
lyzed honey samples (S4), as well as its fluorescence surface. ing on the honey samples. For most samples, a third factor
Excitation and emission loadings estimated for the other (blue dotted lines in Fig. 5b, c) was also estimated, with
eight honey samples show similar profiles. excitation maximum at 310 nm, and emission maximum
Two factors are common in all these models. The first around 420 nm. This factor may be related to Maillard reac-
factor (black solid lines in Fig. 5b, c) showed two local tion products, such as furosine and hydroxymethylfurfural
excitation maxima at 280 nm and 350 nm, and an emis- (HMF) (Christensen et al. 2006).
sion maximum around 380 nm. This factor was assigned
to phenylalanine based on spectral comparison with the lit- Quantification of Phenylalanine by Second‑Order
erature (Stanković et al. 2019; Chen et al. 2016; Lenhardt Standard Addition
et al. 2015; Pérez et al. 2007; Hermosín et al. 2003). This
factor was also the only one associated with a significant A previous attempt to use an external calibration model
increase in sample loadings (scores) as a function of stand- with different honey samples demonstrated its infeasibility
ard additions. Excitation maxima at 258 nm (Christensen since the presence of matrix effect was observed. Then,
et al. 2006), 260 and 282 nm (Lakowicz 2006) and emission the strategy of standard addition was employed. PARA-
maxima at 284 nm (Christensen et al. 2006) and 280 nm FAC measurement loadings (first mode) of the first fac-
(Lakowicz 2006) for phenylalanine in water have also been tor were used to construct univariate linear regressions/
reported. Interactions of phenylalanine with other compo- standard addition curves by applying Eq. 1, thus estimat-
nents of the matrix such as honey and honeydew can result ing phenylalanine concentration in each sample. Good

Fig. 5  a Fluorescence surface, and b excitation and c emission loadings estimated by PARAFAC for honey sample S4. In b and c, black solid
lines, red solid lines, and blue dotted lines represent the loadings for the first, second, and third PARAFAC factors, respectively

13
Food Analytical Methods (2022) 15:728–738 735

correlations (correlation coefficients between 0.96 and Table 2  Phenylalanine concentration determined for each honey sam-
0.99) were observed between the instrumental response ple by spectrofluorimetric data and second-order standard addition,
and checked by HPLC–UV (n = 3) (mean ± SD)
and the addition of phenylalanine standard solution for all
the samples. Figure 6 shows the standard addition curve Samples [Phenylalanine] (mg ­kg−1)
built with PARAFAC loadings for one replicate of honey PARAFAC HPLC–UV
sample S4. In this figure, the increase of phenylalanine
fluorescence intensity with the phenylalanine standard S1 6.1 ± 0.4 6.3 ± 0.4
solution addition (20 mg ­L−1) through the spectral con- S2 9.3 ± 1.0 9.6 ± 0.7
tour maps presented below the curve can also be observed. S3 5.7 ± 0.2 5.8 ± 0.5
Estimated phenylalanine concentrations for each sample S4 8.2 ± 0.6 8.9 ± 0.2
(n = 3, mean ± standard deviations, SD) obtained from S5 13.9 ± 1.7 14.5 ± 0.1
PARAFAC models are presented in Table 2. S6 32.5 ± 3.4 32.1 ± 0.6
Phenylalanine concentration varied significantly among S7 30.6 ± 1.0 31.8 ± 1.1
samples, from 5.7 to 32.5 mg ­kg−1. This variation may be S8 8.5 ± 0.2 8.2 ± 0.3
explained due to the different botanical and floral origins S9 9.9 ± 0.9 9.5 ± 0.7
of the analyzed honeys (in addition to other factors, such Mean 13.9 14.1
as bee feeding, climatic conditions, processing methods). Median 9.3 9.5
This analytical range agrees with the literature for the Min 5.7 5.8
phenylalanine concentration in other types of honey, such Max 32.5 32.1
as polyfloral (7.92–140.65 mg ­kg−1, Rebane and Herodes
(2010); 28.27 ± 13.37 mg ­kg−1, Iglesias et al. (2006)), hon-
eydew (13.13–31.39 mg ­kg−1, Rebane and Herodes 2010), Figures of Merit
phacelia (20.41 ± 11.99 mg ­kg−1, Kuś et al. 2018), heather
(11.01–66.26 mg ­kg−1, Rebane and Herodes 2010), and Aiming to evaluate the performance of the developed spec-
dandelion (15.34–61.93 mg ­kg −1, Rebane and Herodes trofluorimetric method, suitable figures of merit (FOM)
2010). were calculated (Olivieri 2014). Table 3 shows the estimated
The results of the developed spectrofluorimetric method values for precision, sensitivity (SEN), selectivity (SEL),
were verified by HPLC–UV. Reference values provided by the inverse of analytical sensitivity (γ−1), limit of detection
HPLC for each honey sample are also presented in Table 2. (LOD), and limit of quantification (LOQ).
A paired t test at 95% significance level and with 8 degrees These FOM were specifically calculated for each sam-
of freedom corroborated that there are no significant dif- ple. This is typical for second-order calibration models,
ferences between the results of the two methods (estimated since the composition of interferences can vary from
t of 1.20 versus tabulated t of 2.306). sample to sample. Thus, mean values of FOM are not

Fig. 6  Standard addition curve

built with PARAFAC loadings
of the first factor from the first
mode for honey sample S4.
Insets show excitation-emission
contour maps relative to each
point of the curve

13
736 Food Analytical Methods (2022) 15:728–738

Table 3  Figures of merit for the Figures of merit

quantification of phenylalanine
in honey by the developed Samples
spectrofluorimetric method
S1 S2 S3 S4 S5 S6 S7 S8 S9

Precision (%)a 6.6 10.7 3.5 7.3 12.2 10.5 3.3 2.4 9.1
Sensitivity 1333 1967 1600 2333 607 430 527 1900 1239
Selectivity 0.76 0.83 0.88 0.83 0.83 0.94 0.82 0.82 0.63
γ−1 (mg ­kg−1)b 0.02 0.02 0.02 0.02 0.06 0.07 0.07 0.01 0.02
LOD (mg ­kg−1) 0.3 0.1 0.1 0.2 0.3 0.5 0.4 0.2 0.01
LOQ (mg ­kg−1) 0.9 0.4 0.3 0.6 0.9 1.4 1.1 0.7 0.04
a
Relative standard deviation of triplicates. bInverse of analytical sensitivity

representative of the model. Precision was estimated as Conclusion

relative the standard deviation (RSD) of triplicates for
each sample. RSD varied between 2.4 and 12.2%. The Determination of phenylalanine in honey is important
following FOM are calculated based on the concept of net because this analyte is one of the predominant free amino
analyte signal (NAS) and a deeper discussion about them acids in this food matrix and has been considered a bio-
including the respective equations can be found elsewhere marker of its botanical and geographical origin. This study
(Bernardes et al. 2010; Olivieri 2014). SEN is the variation developed a direct and environmentally friendly spectro-
of the analytical signal caused by a change in the analyte fluorimetric method to determine phenylalanine in honey
concentration. The large range of SEN values observed based on multiway calibration using PARAFAC. Honey
for honey samples (430–2333 mg ­kg−1) corroborated the samples from different origins were analyzed, with the phe-
presence of matrix effect and the need to employ the stand- nylalanine concentration ranging from 5.7 to 32.5 mg ­kg−1.
ard addition method. Samples presenting lower SEN, such These results were verified by HPLC and the proposed
as S5–7 (Table 3), may be associated with the presence method was validated by estimating proper figures of merit,
of a more intense fluorescence quenching. SEL estimates such as precision, sensitivity, analytical sensitivity, and
the fraction of the sample analytical signal used to pre- limits of detection and quantification. Thus, the developed
dicted analyte concentration. These values were estimated method provided robust and reliable predictions for one of
between 63 and 94%. In truth, this is a not so useful FOM, the most important compositional parameters of honeys,
since there are no practical reasons to establish limits for which can be useful for several purposes, such as quality
SEL. Analytical sensitivity (γ) is the relation between SEN control, authentication testing, detection of frauds, and eval-
and the instrumental noise, and allows comparing different uation of pharmacological properties.
methodologies, since it is independent on the food matrix Due to the presence of a strong matrix effect, this determi-
and the analytical technique employed. The inverse of γ, nation cannot be performed using external validation. Thus,
γ−1, is an estimate of the minimum concentration differ- an analytical strategy employing second-order standard addi-
ence that the method can discriminate in the absence of tion allowed to quantify an analyte in the presence of uncali-
experimental error. This inverse value also defines the brated interferences that may be present in different honeys.
number of decimal places that should be used to express The developed method presented several advantages over
the prediction results. Although the estimated inverses of the methods currently used for this determination, which
analytical sensitivities varied at the second decimal place are mainly based on chromatographic techniques. Among
(0.01–0.07 mg ­kg −1), we decided to express the results these advantages, rapidity, relatively low cost, no need of
with only one decimal place as a more realistic representa- sample pretreatment, no need of solvents or reagents, and
tion. Finally, LOD and LOQ are calculated as 3.3 and 10 no production of chemical waste can be cited. The analyti-
times the inverse of the analytical sensitivity (γ−1), respec- cal and chemometric strategy employed in this study can be
tively, and their estimates varied to a great extent (0.1–0. suggested to be applied to quantify phenylalanine or other
5 mg ­kg−1/0.3–1.4 mg ­kg−1). This large variation of LOD fluorescent analytes in different food matrices that present
and LOQ also reflected the presence of matrix effect, and matrix effect.
the range of these values was considered satisfactory for
practical purposes in routine honey analysis. Acknowledgements The authors are grateful to the Brazilian govern-
ment agencies CNPq, FAPEMIG, and CAPES, for financial support,

13
Food Analytical Methods (2022) 15:728–738 737

and Dr. Maria José Nunes de Paiva (Faculty of Pharmacy) and Mirna according to free amino acids content by high-performance liquid
Maciel D’Auriol Souza (Chemistry Department) from Universidade chromatography with fluorescence detection with chemometric
Federal de Minas Gerais for helping in the HPLC-UV analyses. DCA approaches. J Sci Food Agric 97:2042–2049. https://​doi.​org/​10.​
acknowledges CNPq for a PhD scholarship. 1002/​jsfa.​8008
Christensen J, Nørgaard L, Bro R, Engelsen SB (2006) Multivariate
autofluorescence of intact food systems. Chem Rev 106:1979–
Declarations 1994. https://​doi.​org/​10.​1021/​cr050​019q
de Juan A, Tauler R (2021) Multivariate Curve Resolution: 50 years
Ethics Approval This article does not contain any studies with human addressing the mixture analysis problem - a review. Anal Chim
participants or animals performed by any of the authors. Acta 1145:59–78. https://​doi.​org/​10.​1016/j.​aca.​2020.​10.​051
Hermosín I, Chicón RM, Cabezudo MD (2003) Free amino acid com-
Consent to Participate Not applicable. position and botanical origin of honey. Food Chem 83:263–268.
https://​doi.​org/​10.​1016/​S0308-​8146(03)​00089-X
Conflict of Interest Daphne Chiara Antônio declares that she has no Iglesias MT, Martín-Álvarez PJ, Polo MC, de Lorenzo C, González
conflict of interest. Bruno Gonçalves Botelho declares that he has no M, Pueyo E (2006) Changes in the free amino acid contents of
conflict of interest. Marcelo Martins Sena declares that he has no con- honeys during storage at ambient temperature. J Agric Food Chem
flict of interest. 54:9099–9104. https://​doi.​org/​10.​1021/​jf061​712x
Kavanagh S, Gunnoo J, Passos TM, Stout JC, White B (2019) Physico-
chemical properties and phenolic content of honey from different
floral origins and from rural versus urban landscapes. Food Chem
272:66–75. https://​doi.​org/​10.​1016/j.​foodc​hem.​2018.​08.​035
References Kelly MT, Blaise A, Larroque M (2010) Rapid automated high per-
formance liquid chromatography method for simultaneous deter-
Allenspach MD, Fuchs JA, Doriot N, Hiss JA, Schneider G, Steuer mination of amino acids and biogenic amines in wine, fruit and
C (2018) Quantification of hydrolyzed peptides and proteins by honey. J Chromatogr A 1217:7385–7392. https://​doi.​org/​10.​
amino acid fluorescence. J Pept Sci 24:e3113. https://​doi.​org/​10.​ 1016/j.​chroma.​2010.​09.​047
1002/​psc.​3113 Kuś PM, Włodarczyk M, Tuberoso CIG (2018) Nitrogen compounds
Bader M (1980) A systematic approach to standard addition methods in Phacelia tanacetifolia Benth. honey: first time report on occur-
in instrumental analysis. J Chem Educ 57:703–706. https://​doi.​ rence of (−)-5-epi-lithospermoside, uridine, adenine and xanthine
org/​10.​1021/​ed057​p703 in honey. Food Chem 255:332–339. https://​doi.​org/​10.​1016/j.​
Bernardes CD, Poppi RJ, Sena MM (2010) Direct determination of foodc​hem.​2018.​02.​093
trans-resveratrol in human plasma by spectrofluorimetry and Lakowicz JR (2006) Principles of fluorescence spectroscopy. Springer,
second-order standard addition. Talanta 82:640–645. https://​doi.​ New York
org/​10.​1016/j.​talan​ta.​2010.​05.​024 Lenhardt L, Bro R, Zeković I, Dramićanin T, Dramićanin MD (2015)
Biluca FC, Bernal J, Valverde S, Ares AM, Gonzaga LV, Costa AC, Fluorescence spectroscopy coupled with PARAFAC and PLS
Fett R (2019) Determination of free amino acids in stingless bee DA for characterization and classification of honey. Food Chem
(Meliponinae) honey. Food Anal Methods 12:902–907. https://​ 175:284–291. https://​doi.​org/​10.​1016/j.​foodc​hem.​2014.​11.​162
doi.​org/​10.​1007/​s12161-​018-​01427-x Lenhardt L, Zeković I, Dramićanin T, Milićević B, Burojević J,
Bong J, Loomes KM, Schlothauer RC, Stephens JM (2016) Fluo- Dramićanin MD (2017) Characterization of cereal flours by
rescence markers in some New Zealand honeys. Food Chem fluorescence spectroscopy coupled with PARAFAC. Food Chem
192:1006–1014. https://​doi.​org/​10.​1016/j.​foodc​hem.​2015.​07.​118 229:165–171. https://​doi.​org/​10.​1016/j.​foodc​hem.​2017.​02.​070
Booksh K, Henshaw JM, Burgess LW, Kowalski BR (1995) A second- Leyva-Jimenez FJ, Lozano-Sanchez J, Borras-Linares I, Cadiz-Gur-
order standard addition method with application to calibration of rea MDLL, Mahmoodi-Khaledi E (2019) Potential antimicrobial
a kinetics–spectroscopic sensor for quantitation of trichloroeth- activity of honey phenolic compounds against Gram positive and
ylene. J Chemom 9:263–282. https://​doi.​org/​10.​1002/​cem.​11800​ Gram negative bacteria. LWT – Food Sci Technol 101:236–245.
90403 https://​doi.​org/​10.​1016/j.​lwt.​2018.​11.​015
Bro R (1997) PARAFAC. Tutorial and applications. Chemom Intell Lobo-Prieto A, Tena N, Aparicio-Ruiz R, García-González DL, Sikor-
Lab Systems 38:149–171. https:// ​ d oi. ​ o rg/ ​ 1 0. ​ 1 016/ ​ S 0169-​ ska E (2020) Monitoring virgin olive oil shelf-life by fluores-
7439(97)​00032-4 cence spectroscopy and sensory characteristics: a multidimen-
Bro R, Kiers HAL (2003) A new efficient method for determining sional study carried out under simulated market conditions. Foods
the number of components in PARAFAC models. J Chemom 9:1846. https://​doi.​org/​10.​3390/​foods​91218​46
17:274–286. https://​doi.​org/​10.​1002/​cem.​801 Muñoz de la Peña A, Mora Diez N, MahederoGarcía MC, Bohoyo Gil
Bro R (1996) Multiway calibration. Multilinear PLS. J Chemom D, Cañada-Cañada F (2007) A chemometric sensor for determin-
10:47–61. https://doi-org.ez27.periodicos.capes.gov.br/10.1002/ ing sulphaguanidine residues in honey samples. Talanta 73:304–
(SICI)1099–128X(199601)10:1<47::AID-CEM400>3.0.CO;2-C 313. https://​doi.​org/​10.​1016/j.​talan​ta.​2007.​03.​047
Bro R (1998) Multi-way analysis in the food industry: models, algo- Noviyanto A, Abdulla WH (2020) Honey botanical origin classifica-
rithms, and applications. https://​static-​curis.​k u.​dk/​portal/​files/​ tion using hyperspectral imaging and machine learning. J Food
130359​ 61/R​ asmus_B ​ ro.p​ df. PhD Thesis, Universiteit van Amster- Eng 265:109684. https://d​ oi.o​ rg/​10.1​ 016/j.j​ foode​ ng.​2019.​109684
dam, Amsterdam. Accessed in Aug 2021 Olivieri AC (2014) Analytical figures of merit: from univariate to mul-
Chen Q, Qi S, Li H, Han X, Ouyang Q, Zhao J (2014) Determina- tiway calibration. Chem Rev 114:5358–5378. https://​doi.​org/​10.​
tion of rice syrup adulterant concentration in honey using three- 1021/​cr400​455s
dimensional fluorescence spectra and multivariate calibrations. Olivieri AC, Escandar GM (2019) Analytical chemistry assisted by
Spectrochim Acta A 131:177–182. https://​doi.​org/​10.​1016/j.​saa.​ multi-way calibration: a contribution to green chemistry. Talanta
2014.​04.​071 204:700–712. https://​doi.​org/​10.​1016/j.​talan​ta.​2019.​06.​022
Chen H, Jin L, Chang Q, Peng T, Hu X, Fan C, Pang G, Lu M, Wang Olivieri AC, Wu H-L, Yu R-Q (2009) MVC2: A MATLAB graphi-
W (2016) Discrimination of botanical origins for Chinese honey cal interface toolbox for second-order multivariate calibration.

13
738 Food Analytical Methods (2022) 15:728–738

Chemom Intell Lab Systems 96:246–251. https://​doi.​org/​10.​ Sádecká J, Tóthová J (2007) Fluorescence spectroscopy and chemo-
1016/j.​chemo​lab.​2009.​02.​005 metrics in the food classification - a review. Czech J Food Sci
Otter DE (2012) Standardised methods for amino acid analysis of food. 25:159–173. https://​doi.​org/​10.​17221/​687-​CJFS
Br J Nutr 108:S230–S237. https://​doi.​org/​10.​1017/​S0007​11451​ Samari F, Hemmateenejad B, Shamsipur M (2010) Spectrophotometric
20024​86 determination of carminic acid in human plasma and fruit juices
Pereira AP, Oliveira JM, Mendes-Ferreira A, Estevinho LM, Mendes- by second order calibration of the absorbance spectra-pH data
Faia A (2017) Mead and other fermented beverages. In: Pandey matrices coupled with standard addition method. Anal Chim Acta
A, Du G, Sanroman MA, Soccol CR, Dussap CG (eds) Current 667:49–56. https://​doi.​org/​10.​1016/j.​aca.​2010.​04.​007
developments in biotechnology and bioengineering - food and Sena MM, Trevisan MG, Poppi RJ (2006) Combining standard addition
beverages industry. Elsevier, Amsterdam method and second-order advantage for direct determination of
Pérez AR, Iglesias MT, Pueyo E, González M, de Lorenzo C (2007) salicylate in undiluted human plasma by spectrofluorimetry. Tal-
Amino acid composition and antioxidant capacity of Spanish anta 68:1707–1712. https://​doi.​org/1​ 0.​1016/j.​talan​ta.2​ 005.​08.​024
honeys. J Agric Food Chem 55:360–365. https://​doi.​org/​10.​1021/​ Sergiel I, Pohl P, Biesaga M, Mironczyk A (2014) Suitability of three-
jf062​055b dimensional synchronous fluorescence spectroscopy for finger-
Perkowski MC, Warpeha KM (2019) Phenylalanine roles in the seed- print analysis of honey samples with reference to their phenolic
to-seedling stage: not just an amino acid. Plant Sci 289:110223. profiles. Food Chem 145:319–326. https://d​ oi.o​ rg/1​ 0.1​ 016/j.f​ oodc​
https://​doi.​org/​10.​1016/j.​plant​sci.​2019.​110223 hem.​2013.​08.​069
Privitera ML, Lozano VA (2017) Development of a second-order Sharin SN, Sani MSA, Jaafar MA, Yuswan MH, Kassim NK, Manaf
standard addition fluorescence method for the direct determina- YN, Wasoh H, Zaki NNM, Hashim AM (2021) Discrimination of
tion of riboflavin in human urine samples without previous clean Malaysian stingless bee honey from different entomological ori-
up and separation steps. Microchem J 133:60–66. https://​doi.​org/​ gins based on physicochemical properties and volatile compound
10.​1016/j.​microc.​2017.​02.​033 profiles using chemometrics and machine learning. Food Chem
Rebane R, Herodes K (2008) Evaluation of the botanical origin of 346:128654. https://​doi.​org/​10.​1016/j.​foodc​hem.​2020.​128654
Estonian uni- and polyfloral honeys by amino acid content. J Agric Shen S, Wang J, Chen X, Liu T, Zhuo Q, Zhang S-Q (2019) Evalua-
Food Chem 56:10716–10720. https://​doi.​org/​10.​1021/​jf801​8968 tion of cellular antioxidant components of honeys using UPLC-
Rebane R, Herodes K (2010) A sensitive method for free amino acids MS/MS and HPLC-FLD based on the quantitative composition-
analysis by liquid chromatography with ultraviolet and mass spec- activity relationship. Food Chem 293:169–177. https://d​ oi.o​ rg/1​ 0.​
trometric detection using precolumn derivatization with diethyl 1016/j.​foodc​hem.​2019.​04.​105
ethoxymethylenemalonate: application to the honey analysis. Anal Siddiqui AJ, Musharraf SG, Choudhary MI, Rahman A (2017) Appli-
Chim Acta 672:79–84. https://​doi.​org/​10.​1016/j.​aca.​2010.​04.​014 cation of analytical methods in authentication and adulteration of
Ríos-Reina R, Elcoroaristizabal S, Ocaña-González JA, García- honey. Food Chem 217:687–698. https://​doi.​org/​10.​1016/j.​foodc​
González DL, Amigo JM, Callejón RM (2017) Characterization hem.​2016.​09.​001
and authentication of Spanish PDO wine vinegars using multidi- Stanković M, Bartolić D, Šikoparij B, Spasojević D, Mutavdžić D,
mensional fluorescence and chemometrics. Food Chem 230:108– Natić M, Radotić K (2019) Variability estimation of the protein
116. https://​doi.​org/​10.​1016/j.​foodc​hem.​2017.​02.​118 and phenol total content in honey using front face fluorescence
Roldán A, van Muiswinkel GCJ, Lasanta C, Palacios V, Caro I (2011) spectroscopy coupled with MCR–ALS analysis. J Appl Spectrosc
Influence of pollen addition on mead elaboration: physicochemi- 86:256–263. https://​doi.​org/​10.​1007/​s10812-​019-​00809-1
cal and sensory characteristics. Food Chem 126:574–582. https://​ Strelec I, Brodar L, Flanjak I, Kenjerić FČ, Kovač T, Kenjerić DČ,
doi.​org/​10.​1016/j.​foodc​hem.​2010.​11.​045 Primorac L (2018) Characterization of Croatian honeys by right-
Rubio L, Sanllorente S, Sarabia LA, Ortiz MC (2019) Fluorescence angle fluorescence spectroscopy and chemometrics. Food Anal
determination of cochineal in strawberry jam in the presence Methods 11:824–838. https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 12161-0​ 17-1​ 059-z
of carmoisine as a quencher by means of four-way PARAFAC
decomposition. Food Chem 290:178–186. https://​doi.​org/​10.​ Publisher's Note Springer Nature remains neutral with regard to
1016/j.​foodc​hem.​2019.​03.​113 jurisdictional claims in published maps and institutional affiliations.
Ruoff K, Karoui R, Dufour E, Luginbbühl W, Bosset J, Bogdanov S,
Amadò R (2005) Authentication of the botanical origin of honey
by front-face fluorescence spectroscopy. A preliminary study. J
Agric Food Chem 53:1343–1347. https://​doi.​org/​10.​1021/​jf048​
384q

13