Food and Bioprocess Technology

, Volume 4, Issue 3, pp 364–386

Fluorescence Spectroscopy Measurement for Quality Assessment of Food Systems—a Review


    • Department of Food Technology, Gembloux Agro-Bio TechUniversity of Liège
  • Christophe Blecker
    • Department of Food Technology, Gembloux Agro-Bio TechUniversity of Liège
Review Paper

DOI: 10.1007/s11947-010-0370-0

Cite this article as:
Karoui, R. & Blecker, C. Food Bioprocess Technol (2011) 4: 364. doi:10.1007/s11947-010-0370-0


The present review gives an overview of the use of fluorescence spectroscopy (i.e., conventional, excitation–emission matrix, and synchronous fluorescence) for determining changes in food products and their quality during technological process and storage. From the present review, it was shown that fluorescence spectroscopy is able to determine several properties (functional, composition, nutritional) without the use of chemical reagents. This is due to the use of chemometric tools (descriptive and predictive methods). The review focuses on the use of fluorescence spectroscopy for the determination of the quality of animal (i.e., dairy, meat, fish, and egg) and vegetable (oils, cereal, sugar, fruit, and vegetable) products as well as the identification of bacteria of agro-alimentary interest.


Fluorescence spectroscopyFood systemsQualityChemometrics


Public interest in food quality and production has increased in recent decades, probably related to changes in eating habits, consumer behavior, and the development and increased industrialization of the food supplying chains (Christensen et al. 2006). The demand for high quality and safety in food production obviously calls for high standards for quality and process control, which in turn requires appropriate analytical tools to investigate food. Fluorescence spectroscopy is an analytical technique whose theory and methodology have been extensively exploited for studies of molecular structure and function in the discipline of chemistry and biochemistry (Strasburg and Ludescher 1995). Even though fluorescence is one of the oldest analytical methods used (Valeur and Bochon 2001), it has just, recently, become quite popular as a tool in biological science related to food technology. An indication for this popularity is the increasing number of research publications about fluorescence as well as the introduction of new commercially available instruments for fluorescence analysis, particularly, front-face fluorescence spectroscopy (FFFS). Traditional right angle fluorescence spectroscopic technique cannot be applied to thick substances due to large absorbance and scattering of light. Indeed, when the absorbance of the sample exceeds 0.1, emission and excitation spectra are both decreased and excitation spectra are distorted (Karoui et al. 2003). To avoid these problems, a dilution of samples (when it is possible) was performed so that their total absorbance would be less than 0.1. However, the results obtained on diluted solution of food samples cannot be extrapolated to native concentrated samples since the organization of the food matrix is lost. To reply with this request, FFFS could be utilized. The use of only excitation and emission wavelengths could limit the ability of fluorescence spectroscopy to determine the quality of food systems. To comply with this requirement, the variation in the excitation and emission wavelengths allows simultaneous determination of compounds in several foodstuffs. Therefore, it would be interesting to use for each food product different excitation wavelengths simultaneously. This could be realized by using synchronous fluorescence spectroscopy (SFS) which presents two interesting advantages from our point of view: it (1) allows the consideration of the whole fluorescence landscape, i.e., spectra recorded at different offsets and (2) retains information related to several fluorophores compared to a classical emission spectrum, which is mainly specific to a sole fluorophore. Synchronous fluorescence spectra are obtained by simultaneously scanning both the excitation and emission monochromators keeping a fixed wavelength interval, named offset between them. It gives a narrower and simpler spectrum. For the SFS technique, the selection of a wavelength interval is one of the most important experimental parameter and the parameter should be optimized, which is carried out by measuring the spectra at various offsets.

This review will provide the reader with the basic principles of fluorescence including the use of this technique, especially the use of the most common FFFS and synchronous fluorescence for the assessment of the quality of several food systems that will be discussed in detail (Table 1).
Table 1

Overview of the literature survey on fluorescence of different food systems


Dairy products




Edible oils



Fruit and vegetable


Amino acids and nucleic acids

Boubellouta and Dufour (2008), Dufour and Riaublanc (1997), Dufour et al. (2000), Hammami et al. (2010),Herbert et al. (1999, 2000), Karoui and Dufour (2003, 2006), Karoui et al. 2004a, b, 2005a, b, c, 2006b, e, 2007c, Kulmyrzaev et al. (2005), Liu and Metzger (2007), Mazerolles et al. (2001), Rouissi et al. (2008), Schamberger and Labuza (2006), Zaïdi et al. (2008)

Allais et al. (2004), Dufour and Frencia (2001), Frencia et al. (2003), Lebecque et al. (2003), Møller et al. (2003), Sahar et al. (2009a)

Dufour et al. (2003), Karoui et al. (2006f)

Karoui et al. (2006c, d2007d, e)


Karoui et al. (2006a), Zandomeneghi (1999)

Baunsgaard et al. (2000a, b), Bro (1999), Karoui et al. (2007a), Munck et al. (1998), Nørgaard (1995), Ruoff et al. (2006)

Noh and Lu (2007), Seiden et al. (1996)

Ammor et al. (2004), Leblanc and Dufour (2002, 2004), Leriche et al. (2004), Tourkya et al. (2009)



Egelandsdal et al. (1996, 2002, 2005), Skjervold et al. (2003), Swatland (1987), Swatland and Findlay (1997), Swatland et al. (1995a, b)




Engelsen (1997), Guimet et al. (2004, 2005), Kyriakidis and Skarkalis (2000), Poulli et al. (2005, 2007), Zandomeneghi et al. (2005)


De Ell et al. (1996), Hagen et al. (2006), Lötze et al. (2006), Moshou et al. (2005)


Ferulic acid


Symons and Dexter (1991, 1992, 1993, 1994)


Maillard products

Liu and Metzger (2007), Schamberger and Labuza (2006)


Karoui et al. (2006c, d)


Baunsgaard et al. (2000a, b)



Kulmyrzaev et al. (2005)


Dufour et al. (2003), Karoui et al. (2006f)


Ghosh et al. (2005)


Ammor et al. (2004), Leblanc and Dufour (2002, 2004), Tourkya et al. (2009)


Kulmyrzaev et al. (2005)


Oxidation products

Karoui et al. (2007b), Wold et al. (2002, 2005)

Gatellier et al. (2007), Møller et al. (2003), Olsen et al. (2005), Sahar et al. (2009b), Veberg (2006), Veberg et al. (2006)

Hasegawa et al. (1992), Olsen et al. (2006)


Engelsen (1997), Guimet et al. (2004, 2005), Sikorska et al. (2004, 2005)




Sikorska et al. (2004, 2005), Zandomeneghi et al. (2005)


Baunsgaard et al. (2000a, b), Karoui et al. (2007a), Ruoff et al. (2006)


Retinol (vitamin A)

Boubellouta and Dufour (2008), Dufour and Riaublanc (1997), Dufour et al. (2000), Hammami et al. (2010), Herbert et al. (1999, 2000), Karoui and Dufour (2003, 2006), Karoui et al. (2004a, b, 2005a, b, c, 2006b, e, 2007c), Kulmyrzaev et al. (2005), Liu and Metzger (2007), Mazerolles et al. (2001), Rouissi et al. (2008), Zaïdi et al. (2008)


Karoui et al. (2006c, d, 2007d, e)


Riboflavin (vitamin B2)

Boubellouta and Dufour (2008), Hammami et al. (2010), Karoui and Dufour (2003, 2006), Karoui et al. (2005c, 2006b, e, 2007c), Liu and Metzger (2007), Rouissi et al. (2008), Zaïdi et al. (2008)


Zandomeneghi et al. (2003)


Tocopherols (vitamin E)


Guimet et al. (2004, 2005), Kyriakidis and Skarkalis (2000), Poulli et al. (2005, 2007), Sikorska et al. (2004, 2005), Zandomeneghi et al. (2005)


Fluorescence Spectroscopy


Fluorescence is the emission of light subsequent to absorption of ultraviolet or visible light of a fluorescent molecule or substructure, called a fluorophore. Thus, the fluorophore absorbs energy in the form of light at a specific wavelength and liberate energy in the form of emission of light at a higher wavelength. The general principles can be illustrated by a Jablonski diagram (Papageorgiou and Govindjee 2004; Rost 1995; Zude 2008), as shown in Fig. 1.
Fig. 1

Jablonski diagram showing the basic principle in fluorescence spectroscopy

The first step (1) is the excitation, where light is absorbed by the molecule, which is transferred to an electronically excited state, meaning that an electron goes from the ground singlet states, S0, to an excited singlet state, S1. This is followed by a vibrational relaxation or internal conversion (2), where the molecule undergoes a transition from an upper electronically excited state to a lower one, S1, without any radiation. Finally, the emission occurs (3), typically 10−8 s after the excitation, when the electron returns to its more stable ground state, S0, emitting light at a wavelength according to the difference in energy between the two electronic states. This explanation is somewhat simplified. In molecules, each electronical state has several associated vibrational states. In the ground state, almost all molecules occupy the lowest vibrational level. By excitation with ultraviolet or visible light, it is possible to promote the molecule of interest to one of several vibrational levels for the given electronically excited level. This implies that absorption and fluorescence emission does not only occur at one single wavelength, but rather over a distribution of wavelengths corresponding to several vibrational transitions as components of a single electronic transition. This is why excitation and emission spectra are obtained to describe the detailed fluorescence characteristics of molecules. In fact, fluorescence is characterized by two wavelength parameters that significantly improve the specificity of the method, compared to spectroscopic techniques based only on absorption.

Quantum Yield (Efficiency)

Each molecule presents a specific property, which is described by number, named quantum yield or quantum efficiency (ϕ).
$$ \phi = \frac{\text{number of quanta emitted }}{{\hbox{number of quanta absorbed }}} = {\hbox{quantum}}\;{\hbox{yield}} $$

As illustrated in Eq. 1, the higher the value of ϕ, the greater the fluorescence of a compound (e.g., chlorophyll). A practically non-fluorescent molecule (e.g., carotenoids) is one whose quantum efficiency is zero or so close to zero that the fluorescence is not measurable. All energy absorbed by such a molecule is rapidly lost by collision deactivation.

Excitation and Emission Spectra

Excitation Spectrum

The excitation spectrum is defined as the relative efficiency of different wavelengths of exciting radiation in causing fluorescence. The shape of the excitation spectrum should be identical to that of the absorption spectrum of the molecule and independent of the wavelengths at which fluorescence is measured. However, this is seldom the case because the sensitivity and the bandwidth of the spectrophotometer (absorbance spectrum) and the spectrofluorimeter (excitation spectrum) are different. In addition, for many food samples, scattering properties and energy transfer between neighboring molecules could contribute to this difference. A general rule of thumb is that the strongest (generally the longest) wavelength peak in the excitation spectrum is chosen for excitation of the sample. This minimizes possible decomposition caused by the shorter wavelength, higher energy radiation.

Emission Spectrum

The emission spectrum of a compound results from the radiation absorbed by the molecule. The emission spectrum is the relative intensity of radiation emitted at various wavelengths. In theory, the quantum efficiency and the shape of the emission spectrum are independent of the wavelength of the excitation radiation. In practice, this is not the case. Indeed, it has been shown that fluorescence of chlorophyll from a green leaf has a lower short wavelength emission maximum when excited with green light than when excited with blue light (Buschmann 2007). Green light penetrates more deeply into the leaf since it is less absorbed than blue light and the green light-excited fluorescence from more inside the leaf is more readily re-absorbed by the chlorophylls on its way to the sample surface. The re-absorption of fluorescence is particularly high in the short wavelength fluorescence where it overlaps with the absorption spectrum of chlorophyll. If the exciting radiation is at wavelength that differs from the wavelength of the absorption peak, less radiant energy will be absorbed and hence less will be emitted.

Stokes Shift

According to the Jablonski diagram (Fig. 1), the energy of emission is lower than that of excitation. This implies that the fluorescence emission occurs at higher wavelengths than the absorption (excitation). The difference between the excitation and emission wavelengths is known as Stokes shift (Valeur and Bochon 2001), as indicated with the arrow in Fig. 2, marking the difference between the excitation and emission spectrum of tryptophan fluorescence spectra scanned on milk submitted to ultra-high temperature.
$$ {\hbox{Stokes}}\;{\hbox{shift}}\;\left( {{\hbox{c}}{{\hbox{m}}^{ - 1}}} \right) = {10^7}\left( {\frac{1}{{{\lambda_{\rm{ex}}}}} - \frac{1}{{{\lambda_{\rm{em}}}}}} \right), $$
where λex and λem are the maximum wavelengths (nanometer) for excitation and emission, respectively.
Fig. 2

Excitation (full line) and emission (dotted line) tryptophan fluorescence spectra recorded on UHT milk

Normally, the emission spectrum for a given fluorophore is a mirror image of the excitation spectrum, as seen to some extent in Fig. 2 for tryptophan. The general symmetric nature is a result of the same transitions being involved in both absorption and emission and the similarities of the vibrational levels of S0 and S1. However, there are several exceptions, since several absorption bands can be observed in the excitation spectrum but only the last peak is observed in the emission spectrum, representing the transition from S1 to S0. The fluorescence of vitamin A, as seen in Fig. 3, is an example of this, with three absorption peaks and only one emission peak. Normally, only emission or excitation spectra (i.e., one excitation or emission wavelength) are recorded when investigating the fluorescence of a sample. However, it can be beneficial and informative to obtain the entire fluorescence landscape (also known as two-dimensional fluorescence spectroscopy) in order to find the exact excitation and emission maxima, as well as the correct structure of the peaks. Furthermore, it facilitates more appropriate analysis of fluorescence data from complex samples with more fluorophores present (Christensen et al. 2006).
Fig. 3

Excitation (full line) and emission (dotted line) vitamin A fluorescence spectra recorded on UHT milk

Factors Affecting Fluorescence Intensity

Several factors related to the nature and the concentration of fluorophores of food samples influence the fluorescence intensity.


Fluorescence quenching represents any process leading to a decrease in fluorescence intensity of the sample (Lakowicz 1983). It is related to the deactivation of the excited molecule by either intra- or intermolecular interactions. There are two types of quenching: statistic and dynamic. The first occurs when the formation of the excited state is inhibited due to a ground state complex formation in which the fluorophore forms non-fluorescent complexes with a quencher molecule. Dynamic or collisional quenching refers to the process when quenchers deactivate the behavior of the excited state after its formation. The excited molecule will be deactivated by either intramolecular interaction (collision) or intermolecular activity (interaction with other molecules). One of the best-known quenchers is oxygen. A higher temperature also results in larger amount of collisional quenching due to the increased velocities of molecules. Resonance energy transfer can also be considered as dynamic quenching, since an interaction between the donor and acceptor molecules could take place inducing a full or partial deactivation of the excited fluorophore (donor). The energy transfer does not involve emission of light, but a dipole–dipole interaction between the donor and acceptor molecule.

Concentration and Inner Filter Effect

The equation defining the relationship of fluorescence intensity to concentration is:
$$ {I_{\rm{f}}} = \phi {I_0}\left( {1 - {{10}^{ - \varepsilon lc}}} \right), $$
where If is the fluorescence intensity, ϕ the quantum yield, I0 the intensity of the incident light, ε the molar absorptivity, l the optical depth of the sample, and c the molar concentration of the fluorophore.
According to Lakowicz (1983), for low absorbance (<0.05), the equation can be written as:
$$ {I_{\rm{f}}} = 2.3\phi \;{I_0}\;\varepsilon \;l\;c, $$
where If is the fluorescence intensity, ϕ the quantum yield, I0 the intensity of the incident light, ε the molar absorptivity, l the optical depth of the sample, and c the molar concentration of the fluorophore. The decrease is in a part caused by an attenuation of the excitation beam in the areas of the solution in front of the detection system and by the absorption of the emitted fluorescence within the solution. This is defined as the inner cell or inner filter effect. The equation expressing the fluorescence intensity indicates that there are three major factors other than concentration that could affect the fluorescence intensity:
  1. 1.

    The quantum efficiency ϕ; the greater the ϕ, the greater the fluorescence intensity.

  2. 2.

    The intensity of incident light I0; a more intense source will yield greater fluorescence. In actual practice, a very intense source can cause photodecomposition of the sample. Hence, one compromise is a source of moderate intensity (such as a mercury or xenon lamp).

  3. 3.

    The molar absorptivity of the compound, ε. In order to emit radiation, a molecule must first absorb radiation. Hence, the higher the ε, the better the fluorescence intensity of the compound will be.


It should be remembered that the overall fluorescence intensity of a given sample is expressed as the sum of the fluorescence contribution from each of the inherent fluorophores present in the sample. However, due to the complex systems of food products, the fluorescence intensity may not be additive because the quenching phenomenon and interactions with the molecular environment of the fluorophores may take place.

Molecular Environment

The local environment of a fluorophore has an important effect on the shape of the fluorescence spectra. In more polar environments, the fluorophore in excited state will relax to a lower energy state of S1. This means that the emission of polar fluorophores will be shifted towards longer wavelengths (lower energy) in more polar solvents.

The structure of macromolecules and the location in macromolecules can also have a large effect on the fluorescence emission and quantum yield of a fluorophore. Temperature, pH, and color strongly affect the fluorescence signal. Increased temperature leads to increased movement of the molecules, and thereby more collisions, thus inducing a reduced fluorescence signal. It is therefore important that all samples in an experiment present the same temperature. The pH value affects the fluorescence, and most hydroxyl aromatic compounds fluoresce better at high pH (Guilbaut 1989). The color of the sample can affect both the shape and the intensity of the spectra. Dark samples will reabsorb more of the fluorescence than bright samples.


Scattering of the incident light affects the fluorescence signal. As mentioned in the previous section, the absorbance of the sample measured plays an important role in fluorescence measurements. Especially in turbid solutions and solid opaque samples (like most foods), the amount of scattered and reflected light affects the measurements considerably, with respect to both the sampling (i.e., the optical depth of the sampling) and the obtained (fluorescence) signal. Scattered light can be divided into Rayleigh and Raman scatter.

Rayleigh scatter refers to the scattering of light by particles and molecules smaller than the wavelength of the light. Rayleigh is so-called elastic scatter, meaning that no energy loss is involved, so the wavelength of the scattered light is the same as that of the incident light. Rayleigh scatter can be observed as a diagonal line in fluorescence landscapes for excitation wavelengths equaling the emission wavelength. The signal from fluorophores with little Stokes shift will be situated close to the scattering line, and therefore be most affected by Rayleigh scatter. Due to the construction of grating monochromators used for excitation in most spectrofluorometers, some light at the double wavelength of the chosen excitation will also pass through to the sample. For this reason, an extra band of Rayleigh scatter, so-called second-order Rayleigh, will typically appear in fluorescence measurement for emission wavelengths at twice the given excitation wavelength. Rayleigh scatter can be disregarded by measuring and considering the fluorescence signal only between the first- and second-order Rayleigh scatter.

Raman scatter is inelastic scatter due to absorption and re-emission of light coupled with vibrational states. A constant energy loss will appear for Raman scatter, meaning that the scattered light will have a higher wavelength than that of the excitation light, with a constant difference in wavelength. In liquid samples, the solvent is decisive in the amount and nature of Raman scatter, while for solid samples it will typically be an expression of the bulk substances. Raman scatter can in most cases be neglected because of its weak contribution to the fluorescence signal.


The basic setup for an instrument for measuring steady-state fluorescence is shown in Fig. 4. The spectrofluorimeter consists of a light source (generally xenon or mercury lamp); a monochromator and/or filter(s) for selecting the excitation wavelengths; a sample compartment; a monochromator and/or filter(s) for selecting the emission wavelengths; a detector, which converts the emitted light to an electric signal; and a unit for data acquisition and analysis. The sampling geometry can have a substantial effect on the obtained fluorescence signal. If absorbance is less than 0.1, the intensity of the emitted light is proportional to the fluorophore concentration and excitation and emission spectra are accurately recorded by a classical right-angle fluorescence device. In this case, the excitation light travels into the sample from one side, and the detector is positioned at right angles to the center of the sample. When the absorbance of the sample exceeds 0.1, the intensity of emission and excitation spectra decreases and excitation spectra are distorted. To avoid these problems, dilution of samples (when it is possible, i.e., liquid samples) is currently performed so that their total absorbance will be less than 0.1. However, the results obtained on diluted solutions of food samples cannot be extrapolated to native concentrated samples since the organization of the food matrix is lost. In addition, the dilution may change the concentration of other relevant fluorescent species below or close to the detection limit of fluorescence. Moreover, for solid samples, the dilution cannot be realized (e.g., meat, cheese). To avoid these problems, FFFS can be used (Fig. 4). In this manner, it is possible to measure more turbid or opaque samples, since the signal becomes more independent of the penetration of the light through the sample. However, when front-face sampling is used, the amount of scattered light detected will increase due to the higher level of reflection from the surface topology of the sample and sample holder. To minimize these effects, it is recommended that the sample is not placed with its surface oriented at an angle of 45° to the incident beam, but rather at 30°/60° to the light source and the detector (Lakowicz 1983).
Fig. 4

Basic setup of a spectrofluorimeter

Data Analysis

Data analysis of fluorescence spectra has been well established by Smilde et al. (2004). Fluorescence is inherently multidimensional. Indeed, multidimensional fluorescence signals recorded from a sample can conveniently be presented as a matrix of fluorescence intensities as a function of excitation and/or emission wavelengths. Due to the neighboring wavelengths, highly correlated data present in emission and excitation spectra have been pointed out (Smilde et al. 2004). In this case, principal component analysis (PCA), common component and specific weights analysis (CCSWA), partial least squares regression (PLS), factorial discriminant analysis (FDA), parallel factor analysis (PARAFAC), etc. have proven to be powerful methods for the extraction of valuable information (Boubellouta and Dufour 2010; Christensen et al. 2006; Hammami et al. 2010; Kulmyrzaev and Dufour 2010).

Applications of Fluorescence in Foods and Drinks

Recently, the application of fluorescence spectroscopy in combination with multidimensional statistical techniques for the evaluation of food quality has increased. In most of the research papers, the obtained fluorescence signal was assigned to specific fluorophores after fixing the excitation or the emission wavelength.

Dairy Products

Dairy products contain several intrinsic fluorophores, which represent the most important area of fluorescence spectroscopy. They include the aromatic amino acids and nucleic acids (AAA+NA) tryptophan, tyrosine, and phenylalanine in proteins; vitamins A and B2; nicotinamide adenine dinucleotide (NADH) and chlorophyll; and numerous other compounds that can be found at a low or very low concentration in food products.

Dufour and Riaublanc (1997) investigated the potential of FFFS to discriminate between raw, heated (70 °C for 20 min), homogenized, and homogenized and heated milks. The authors applied PCA to the tryptophan and vitamin A fluorescence spectra, and good differentiation between milk samples as a function of homogenization and heat treatment applied to milk samples was observed. They concluded that the treatments applied to milk induced specific modifications in the shape of the fluorescence spectra. Recently, Kulmyrzaev et al. (2005) confirmed these earlier findings. In their research, the emission and excitation spectra of different intrinsic probes (i.e., AAA+NA, NADH, and FADH) were used to evaluate changes in milk following thermal treatments in the range of 57–72 °C for 0.5–30 min. The PCA applied to the normalized spectra allowed good discrimination of milk samples subjected to different temperatures and times; the obtained results were confirmed recently by Boubellouta and Dufour (2008) who discriminated milk samples according to heating in 4–50 °C and acidification (pH ranging from 6.8 to 5.1). Boubellouta et al. (2009) determined the effect of the adding calcium, phosphate, and citrate at different levels (i.e., 3, 6, and 9 mM) on the molecular structure of skimmed milks; the addition of phosphate induced molecular change (observed by mid-infrared and synchronous fluorescence) that was different than that of calcium and citrate; the origin of this difference was not depicted by the authors. In these three latter research studies, milk samples were heated at only relatively low temperatures allowing the non-monitoring of the development of Maillard browning reaction; this was realized after by Schamberger and Labuza (2006); the fluorescence spectra of milks were processed for 5, 15, 20, 25, and 30 s in 5 °C increments from 110 to 140 °C and found to be well correlated with hydroxylmethylfurfural. Indeed, the R2 values of 0.95 were found continuously throughout the emission wavelength range of 394 to 447 nm. In addition, the fluorescence levels increased with higher time–temperature combinations. One of the main conclusions of this study was that FFFS could be considered a promising method for measuring Maillard browning in milk; the authors encouraged the use of this technique for on-line monitoring of thermal processing of milk. Later, Liu and Metzger (2007) confirmed the aforementioned results following the use of FFFS for monitoring changes in non-fat dry milk (n = 9) collected from three different manufacturers and stored at four different temperatures (4, 22, 35, and 50 °C) for 8 weeks. Different intrinsic probes (fluorescent Maillard reaction products (FMRP), riboflavin, tryptophan, and vitamin A) were used, and each of the considered spectral data sets allowed good discrimination of milk samples kept at 50 °C from the others. In addition, good discrimination of milk samples as a function of the storage time was observed. In a similar approach, Feinberg et al. (2006) used fluorescence spectroscopy to identify five types of heat treatments (pasteurization, high pasteurization, direct ultra-high temperature (UHT), indirect UHT, and sterilization) of 200 commercial milk samples stored at 25 and 35 °C for 90 days. By applying FDA, Feinberg et al. (2006) found that tryptophan fluorescence spectra could be considered well-adapted to discriminate sterilized milks and probably pasteurized milks from the other milk samples. However, this intrinsic probe failed to discriminate the other types of milk. An explanation could be that fluorescence spectra were recorded in the pH 4.6 soluble fraction of the milk sample, inducing a loss of information contained in milk samples.

Regarding milk coagulation, Herbert (1999) and Herbert et al. (1999) used FFFS to monitor milk coagulation at the molecular level. Three different coagulation processes have been studied: the glucono-δ-lactone (GDL), the rennet-induced coagulation system, and a mixed GDL and rennet-induced coagulation system. Emission fluorescence spectra of the tryptophan were recorded for each system during the milk coagulation kinetics. By applying the PCA to normalized fluorescence spectra data sets of the three systems, detection of structural changes in casein micelles during coagulation and discrimination of different dynamics of the three coagulation systems was achieved. Herbert et al. (1999) concluded that FFFS allows the investigation of network structure and molecular interactions during milk coagulation.

Most of the aforementioned studies regarding discrimination of milk were performed at a laboratory scale on extreme and controlled samples. Milk products from mountain areas are reputed to have specific organoleptic and nutritional qualities (Bosset et al. 1999; Coulon and Priolo 2002; Renou et al. 2004), and the tracing of milk production sites is therefore important in order to avoid fraud. In this context, the potential of FFFS to discriminate between milks according to their geographical origin was explored. Forty milk samples—8 produced in lowland areas (430–480 m), 16 produced in mid-level areas (720–860 m), and 16 produced in mountain (1,070–1,150 m) areas—from the Haute-Loire department in France at key periods of animals feeding were analyzed (Karoui et al. 2005c). Tryptophan fluorescence spectra, AAA+NA spectra, and riboflavin spectra were recorded directly on milks, with excitation wavelengths set at 290, 250, and 380 nm, respectively. The excitation spectra of vitamin A were also recorded, with the emission wavelength set at 410 nm. By applying FDA to the spectral collection, a trend to a good separation between milks as a function of their origins was observed. The best results were obtained with AAA+NA fluorescence spectra, since 81.5% and 76.9% of the calibration and validation spectra, respectively, were correctly classified. However, some misclassification occurred between milks produced in mid-level areas and the other milk samples. In the same context, FFFS has demonstrated its potential for monitoring ewe’s milk samples according to the feeding system, i.e., scotch bean versus soybean meals (Hammami et al. 2010; Rouissi et al. 2008) and genotype (Zaïdi et al. 2008).

Regarding the use of FFFS for monitoring the quality of cheeses during ripening, Dufour et al. (2000) and Mazerolles et al. (2001) used FFFS to monitor 16 semi-hard cheeses produced and ripened under a controlled scale. By applying PCA to the normalized tryptophan fluorescence spectra, good discrimination of cheeses presenting a ripening time of 21, 51, and 81 days was observed, while an overlap was observed between cheeses aged 1 day and those aged 21 days. The spectral pattern of tryptophan indicated a red shift of aged cheeses suggesting that the environment of ripened cheeses was more hydrophilic than that of young (1-day-old) cheeses. This phenomenon was assigned to partial proteolysis of casein as well as to the salting phenomenon, which may induce some changes in the tertiary and quaternary structures of casein micelles. Regarding the fluorescence spectra of vitamin A, two shoulders located at 295 and 305 nm and a maximum located at 322 nm were observed (Dufour et al. 2000). In addition, the shape of the spectra changed with ripening time. By applying PCA to the normalized vitamin A spectra, better discrimination of cheeses aged 21, 51, and 81 days from those aged 1 day was achieved. The authors determined the link between the data recorded by mid-infrared (MIR) and fluorescence by using canonical correlation analysis (CCA). Correlation coefficients of 0.58 or more were obtained suggesting that fluorescence and MIR spectra might provide a common description of the investigated cheese samples during ripening.

Karoui et al. (2006b) continued this work by recording tryptophan, vitamin A, and riboflavin spectra of 12 semi-hard cheeses (Raclette) of 4 different brands, which were produced during summer period at the industrial level. By applying CCSWA to the spectral data sets and physicochemical data, good discrimination of the four brands was observed. The same research group (Karoui and Dufour 2006) evaluated the potential of FFFS to predict the rheological parameters of 20 semi-hard cheeses at the end of their ripening stage (60 days) from fluorescence spectra recorded at a young stage (2 days old). By using tryptophan fluorescence spectra scanned on cheeses aged 2 days and at 20 °C, the storage modulus (G′), loss modulus (G″), strain, tan (δ), and complex viscosity (η*) were predicted by using PLS regression with leverage correction with R higher than 0.97. The obtained results were confirmed recently by Boubellouta and Dufour (2010), reporting that synchronous fluorescence spectroscopy could be used for the determination of fat melting and cheese melting of two cheese varieties (i.e., Comté and Raclette).

In another context, FFFS has been used for the authentication of different varieties of soft, semi-hard, and hard cheeses during ripening at the retail stage (Herbert 1999; Herbert et al. 2000; Karoui and Dufour 2003; Karoui et al. 2003, 2004a, b, 2005a, b). Herbert et al. (2000) explored the potential of FFFS to discriminate between different soft cheese varieties. Tryptophan and vitamin A spectra were acquired on the cheese samples. The environment of the tryptophan residues was found to be relatively more hydrophilic for the ripened cheeses than for those at the young stage. This phenomenon was attributed to partial proteolysis of caseins during ripening, resulting in an increase of tryptophan exposure to the solvent. To test the accuracy of FFFS in differentiating between the eight soft cheeses, the authors applied FDA to the most relevant PCs, and good discrimination of cheeses was observed, with better results obtained with vitamin A spectra (96% and 93% for the calibration and validation samples, respectively) than with tryptophan spectra (95% and 92% for the calibration and validation samples, respectively). However, in their investigations, samples were studied at the center of the cheese, which could induce some misclassification between the investigated cheeses. In the case of soft cheeses, protein breakdown, lipolysis, pH, etc. differ significantly between the surface and the center. To reply with this request, the matrix structure of three retailed soft cheeses, produced each with different manufacturing process, was studied from the surface to the center of the cheese using FFFS, among other techniques (Karoui and Dufour 2003). Cheese slices, 5-mm thick, were cut from the surface to the center of the samples. PCA applied to the tryptophan fluorescence spectra recorded for each cheese variety showed good discrimination of cheese samples as a function of their location. The environment of tryptophan residues was found to be more heterogeneous in the surface samples than in the center samples; this was attributed to the changes in the extent and type of protein–protein interactions in the protein network depending on the sampling zone. One of the limiting points of this study was the low number of cheeses. In addition, only tryptophan and vitamin A fluorescence spectra were studied in this research. To reply with this request, later, Karoui et al. (2006e, 2007c) used FFFS to investigate changes at the molecular level of both the external (E) and central (C) zones of 15 ripened soft cheeses produced according to traditional and stabilized cheese-making procedures. The CCSWA was applied to the tryptophan, vitamin A, and riboflavin spectral data sets, and the plane defined by the common components 1 (q1) and 3 (q3) showed clear discrimination between the cheese varieties and sampling zones. From the obtained results, it was concluded that CCSWA allowed very efficient management of all the spectroscopic information collected on the investigated soft cheeses. Each of the fluorophore provides information, which can be used for recognizing the cheese variety and sampling zone. The CCSWA method sums up this information using two common components (q1 and q3), taking into account the relation between the different fluorescence data sets. The result obtained from CCSWA did not observed with the PCA performed separately on each of fluorescence spectra, illustrating that CCSWA methodology allowed efficiently the use of all the spectroscopic information provided by the three intrinsic probes. In another study, Karoui et al. (2005a) attempted to discriminate 25 Gruyère and L’Etivaz Protected Designation of Origin cheeses. Emission spectra were scanned following excitation at 250 and 290 nm, and excitation spectra following emission at 410 nm. By applying FDA, 100% correct classification was obtained from the emission and excitation spectra, suggesting the use of FFFS as an accurate technique for the determination of the geographic origin of cheeses. These findings were fully supported on Emmental cheeses originating from different European countries and manufactured during both winter and summer seasons (Karoui et al. 2004a, b, 2005b). One of the strong conclusions of these studies was that FFFS allows a good discrimination between Emmental cheeses produced from raw milk and those made with thermized milk. This was supported by the accuracy of FFFS to predict chemical parameters collected from the E and C zones of 15 soft cheeses being assessed (Karoui et al. 2006e). The PLS regression applied to the normalized vitamin A fluorescence spectra provided the highest values of R2, since values of 0.88, 0.86, 0.86, and 0.84 were found for fat, dry matter, fat in dry matter, and water-soluble nitrogen, respectively.

Finally, FFFS was used for monitoring the oxidation of dairy products. Good estimation of the sensory analyses from FFF spectra of both sour cream (Wold et al. 2002) and cheese (Wold et al. 2005) has been achieved. Wold et al. (2005) showed that naturally occurring porphyrin and chlorophylls play an important role as photosensitizers in dairy products. The degradation of these components showed higher correlation with sensory measured lipid oxidation. Recently, the feasibility of the use of FFFS as a non-destructive technique for monitoring oxidation at the molecular level of semi-hard cheeses, made with cow’s milk and collected during both the grazing period (summer) and the stabling period (autumn), was examined at the surface (20 mm from the rind) and the inner (40 mm from the rind) layers throughout the ripening stage—i.e., 2, 30, and 60 days old by Karoui et al. (2007b). By applying FDA to the 400–640-nm emission fluorescence spectra recorded at the surface layer, correct classification was observed for 100% and 91.7% for the calibration and the validation spectra, respectively. With regard to the samples cut from the inner layers, the authors stated that the 400–640-nm emission fluorescence spectra failed to discriminate cheeses that were either 2 or 30 days old. The main conclusion of this study was that throughout ripening the riboflavin component was affected primarily by oxygen and light (Marsh et al. 1994), while the physicochemical modification that takes place during ripening seemed to present lesser effect than did light and oxygen.

Meat and Meat Products

Research regarding the application of FFFS for the evaluation of meat products has focused on the measurements of fluorescence from collagen, adipose tissues, and protein (Newman 1984; Jensen et al. 1989; Frencia et al. 2003).

The collagen in connective tissue is known to be an important parameter of meat quality, as it is related to the tenderness and texture of the meat. Collagen exists in several different genetic forms, four of which have been found to be present in muscle: types I, III, IV, and V. Types I, III, and IV presented similar fluorescent characteristics when they were excited in the 330–380-nm spectral region (Hildrum et al. 2006). Elastin, another important fluorophore in meat, presents quite similar fluorescence properties to those of collagen types I, III, and IV (Egelandsdal et al. 2005).

Adipose tissue contains fluorescent molecules that are specific for fat. Indeed, it has been shown by several authors (see, for example, Ramanujam 2000; Skjervold et al. 2003) that the fat-soluble vitamins A, D, and K exhibit fluorescence in the 387–480-nm spectral region after excitation at 308–340 nm. Swatland (Swatland 1987; Swatland et al. 1995a, b; Swatland and Findlay 1997) could be considered the pioneer in the field of meat, with a series of papers on different aspects of the use of FFFS, starting in 1987. His work focused on measuring collagen and elastin fluorescence from the connective tissues in meat. The author reported that after excitation at 365 nm, the fluorescence emission spectra of adipose tissue exhibited a maximum at 510 nm with a secondary plateau varying from 430 to 450 nm (Swatland 1987). The obtained fluorescence spectra of various meats were found to be correlated with biochemical and sensory analyses such as chewiness (Swatland et al. 1995a), palatability (Swatland et al. 1995b), and toughness (Swatland and Findlay 1997). Most of the studies realized by Swatland and co-workers were analyzed using univariate data analytical approach, by the comparison of single wavelength or extracted fluorescence peak features. Although interesting results were achieved, the use of multivariate statistical analyses is needed for curve resolution and useful quantitative measurements due to the complexity of fluorescence spectra. Egelandsdal et al. (1996) applied PCA and PLS regression to the fluorescence spectra scanned on meat products after excitation at between 300 and 400 nm for studying isolated perimysial sheets from a type I muscle and found a high correlation between perimysial breaking strength and fluorescence emission spectra recorded after excitation at 335 nm. Wold et al. (1999a, b) later confirmed these previous investigations and suggested that it would be possible to measure the amount of connective tissue in ground meat by using an excitation wavelength of 380 nm. They have also showed that both connective tissue and intramuscular fat content could be measured using an excitation wavelength of 332 nm. Egelandsdal et al. (2002) studied six different batches of beef longissimus dorsi samples originating from 151 animals by using FFFS and Warner–Bratzler (WB) peak values. By applying PLS regression, poor to good (R = 0.45–0.84) correlations between WB peak values and the emission spectra were obtained. In addition, minor difference in predictability was observed using excitation wavelengths at 332 or 380 nm. The emission wavelengths containing the most relevant information about WB peak values were found in the 360–500-nm spectral range. Emission wavelengths around 375 nm, following excitation at 332 nm, were found to be related to a component in the perimysial tissue, most likely present in collagen I or III. In another study, FFFS was explored to predict the age of Parma hams produced from two muscles (semi-membranosus and biceps femoris) and aged 3 months (young), 11 and 12 months (matured), and 15 and 18 months (aged) by Møller et al. (2003). Using PLS regression, prediction of the age was considered as good, with a relative error of prediction of approximately 1 month; also, a good correlation between fluorescence and chemical, sensory, and physical parameters was found.

Egelandsdal et al. (2005) tried to throw more light on the phenomena affecting the fluorescence signal and the ability of the FFFS technique to quantify collagen contents. Beef masseter and latissimus dorsi and pork glutens medius muscles, among others, were chosen for their wide differences in color and connective tissue quality and content. PLS regression was applied to their fluorescence spectra in order to predict collagen (measured as hydroxyproline), and good results were obtained (root mean square error of prediction (RMSEP) of 0.55%). Similar prediction results were obtained with complex sausage batters consisting of different kinds of muscles and presenting a large span in myoglobin and realistic ranges in collagen and fat. One of the interesting results obtained in this study was that FFFS gave lower prediction errors for collagen content than did near-infrared (NIR) reflectance when applied to the same batters. Recently, Sahar et al. (2009a) tried to predict some parameters in meat with excitation set at 290 nm (emission 305–400 nm) and the obtained results were not successful since only 53% and 55% of correct classification for protein and cooking loss, respectively, in the validation data sets were obtained.

Differences in the level of collagen within a muscle or between different muscles led to a huge difference in tenderness (Light et al. 1985). Using excitation wavelengths set between 332 and 380 nm, FFFS found a promising technique for estimating tenderness in such muscles (Hildrum et al. 2006). Recently, tryptophan fluorescence spectra scanned on two beef muscles (longissimus thoracis and infraspinatus) 2 and 14 days post mortem showed a maximum located around 336 nm (Dufour and Frencia 2001). In addition, the maximum emission of aged muscles showed a shift to higher wavelengths (red shift). The PCA performed on spectra showed good discrimination of samples according to the muscle type and aging. However, with only a limited number of meat samples, the models suffered from over-fitting and consequently were not very robust against the inclusion or exclusion of samples. Further analyses with more samples are necessary to substantiate these models. This would allow more variability of the chemical properties and thus development of general mathematical models for better accuracy of the FFFS technique. Frencia et al. (2003) therefore assessed the potential of FFFS to discriminate between five muscle types presenting different levels of collagen contents at two points in time (2 and 14 days post mortem). Applying FDA to the tryptophan fluorescence spectra, correct classification rate of 82% was obtained. The authors concluded that FFFS is a powerful technique that allows a relatively good identification of muscle types according to maturation. Preliminary results have shown that results obtained at the molecular level by FFFS are related to the macroscopic levels (sensory and rheology data sets). Indeed, by applying CCA to the fluorescence spectra (spectrofluorimeter with a front-face device or coupled to a fiber-optic) and mechanical properties (texturometer) or sensory attributes, a strong correlation between the different methods was found. The authors concluded that a common description of the samples was possible from both the fluorescence and the rheology or sensory data, which was later confirmed by the investigation of Lebecque et al. (2003) on 25 longissimus thoracis samples from animals presenting different ages (between 2.5 and 8 years) and sexes. The authors depicted that the phenomena observed at the molecular and macroscopic levels were related to the changes in the texture of meat during aging. The obtained results were confirmed by Allais et al. (2004) on meat emulsion and frankfurter reporting that high correlation was obtained from fluorescence spectra and rheology methods.

Lipid oxidation is one of the factors limiting the quality and acceptability of meat and meat products (Veberg 2006). As explained above, lipid oxidation can be determined by several methods, such as 2-thiobarbituric acid, sensory analysis, and dynamic headspace gas chromatography combined with mass spectrometry. Veberg et al. (2006) used FFFS and other destructive methods to determine the level of lipid oxidation and explored the usefulness of this technique for detection of low levels of lipid oxidation in turkey meat stored during 9 months at −10 and −20 °C. By applying PLS regression between fluorescence spectra and the values obtained by the other traditional techniques, good correlations between fluorescence spectra and thiobarbituric acid reactive substances (TBARS), hexanal, 1-penten-3-ol, total components and sensory measured rancidity and intensity were found. One of the interesting conclusions of this study was that FFFS could be considered as a sensitive and non-destructive method for early lipid oxidation determination in turkey meat. Sahar et al. (2009b) confirmed the obtained results recently on chicken meat since FFFS was found to be able to determine heterocyclic aromatic amines in grilled meat. The obtained results confirmed the previous findings of Olsen et al. (2005) who monitored poultry meat freeze-stored in air at −20 °C for 26 weeks and stated that FFFS and gas chromatography coupled with mass spectrometry could detect oxidative changes in pork back fat earlier than the sensory panel and the electronic nose could. However, the correlation of fluorescence spectra with sensory analyses was found to be poorer and less than that observed between sensory analysis and gas chromatography with mass spectrometry. In a similar approach, Gatellier et al. (2007) monitored lipid oxidation of chicken meat by using FFFS and TBARS over 9 days. In their research study, three chicken genotypes representative of French production were compared (i.e., Standard, Certified, and Label). The samples were stored in darkness at 4 °C, and a good correlation between emission spectra recorded after excitation at 380 nm and TBARS was found. In addition, the genotype meat was found to present an effect on the shape of fluorescence emission spectra, since the fluorescence intensity of the emission spectra of Certified and Label animals after 7 days of refrigerated storage was significantly higher than that of Standard chicken meat samples. The obtained results confirmed previous findings of Munck (2001) reporting that fluorescence could be used in slaughtering and cutting by robots.


The lipid fraction of marine fish has been shown to contain a high level of polyunsaturated fatty acids. During storage and/or processing, the degradation of polyunsaturated fatty acids can lead to the development of primary and secondary products, resulting in the formation of fluorescent compounds and the loss of essential nutriments. Indeed, Gardner (1979) has shown that relatively higher fluorescence intensity was observed at longer wavelength maxima as the quality of fish decreased.

The 493/463- and 327/415-nm ratios have proved to be a more effective index of fish quality than other common assessment methods. Hasegawa et al. (1992) used FFFS for the quantitative assessment of oxidative deterioration in freeze-dried fish. Two excitation wavelengths, i.e., 370 and 450 nm exhibiting maximum emissions at 460 and 500 nm, respectively, were used. For fish samples stored at 25 °C in the dark, an increase in the fluorescence intensity at 500 nm was noted, while that at 460 nm remained unchanged. The fluorophores observed at 460 nm have been attributed to the reaction between reducing sugars (e.g., glucose and ribose) and amino acid compounds inducing activation of the Maillard reaction, while that observed at 500 nm has been attributed to the lipid oxidation products. In a similar approach, Olsen et al. (2006) used an excitation wavelength of 382 nm to record spectra in the 450–750-nm region on four different batches of salmon pâté stored at 4 °C for 4, 8, and 13 weeks. Citric acid or calcium disodium ethylene-diamine tetraacetate was added as metal chelators to two batches, whereas no chelator was added to the third batch. The three investigated batches contained oil, while a fourth one was made with the same amounts of ingredients but without any oil. The obtained results showed an increase in the fluorescence intensity with increased storage time for all the batches. In addition, the shape of the spectra changed largely between samples containing oil and those without oil. Indeed, samples containing oil exhibited the highest intensity in the range of 470–475 nm, while those with no added oil presented maxima between 440 and 450 nm. By applying PCA to the collection of spectral data, a clear difference between samples according to the storage time was observed; the largest variation in the data sets was attributed to whether the sample contained oil or not; the storage time was the second most important factor that led to this discrimination. The authors applied PLS regression to estimate the age of salmon pâté. The correlation coefficients for the sensory attributes, dynamic head-space data, fluorescence spectra, and electronic nose sensor responses were 0.64, 0.94, 0.93, and 0.70, respectively. The corresponding RMSEP were 3.8, 1.7, 1.8, and 3.5, respectively, illustrating that FFFS could be a suitable technique to measure lipid oxidation. This observation is confirmed by the highest level of correlation found between fluorescence spectra and sensory attributes, among the other analytical techniques. Recently, FFFS has been used to monitor fish freshness for four different species—cod, mackerel, salmon, and whiting fillets—at 1, 5, 8, and 13 days of storage (Dufour et al. 2003). Emission spectra of AAA+NA, tryptophan, and NADH were scanned after excitation at 250, 290, and 336 nm, respectively. The spectra of the first two excitation wavelengths showed maxima located at 338 and 336 nm, respectively, while the last excitation wavelength showed two maxima located at 414 and 438 nm. For all three excitation wavelengths, the shape of the spectra illustrated some differences according to storage time, suggesting that a fluorescence spectrum may be considered as a fingerprint. Applying FDA to the tryptophan fluorescence spectra allowed 56% of samples to be correctly classified. Better classification was obtained from AAA+NA and NADH, since 92% and 74% correct classification was observed, respectively. The authors concluded that AAA+NA fluorescence spectra could be considered as fingerprints that may allow discrimination between fresh and aged fish fillets.

Aubourg et al. (1998) used fluorescence spectroscopy to monitor changes that occurs in sardines stored at −18 and −10 °C. Sardines stored at −18 °C were sampled after 0.5, 2, 4, 8, 12, and 24 months, and those stored at −10 °C were sampled at 3, 10, 25, 60, and 120 days. Fluorescence was measured at two excitation–emission maxima, 327/463 and 393/463 nm. The authors used the fluorescence ratio, defined as the fluorescence intensity at 327/463 nm over the fluorescence intensity at 393/463 nm, determined in aqueous and organic solutions (phases resulted from lipid extraction). This ratio was found to increase throughout the whole storage time at the two temperatures when it was determined in the aqueous solution. One of the most limiting points of this study was that the use of only maxima of emission and excitation wavelengths could induce some loss of information contained in the fluorescence spectra. Recently, Karoui et al. (2006f) explored the potentiality of FFFS to differentiate between fresh and frozen-thawed fish. A total of 24 fish (12 fresh and 12 frozen-thawed) were analyzed by using excitation wavelengths set at 290 (tryptophan) and 340 nm (NADH). The emission spectra of NADH of fresh fish showed a maximum at 455 nm and a shoulder at 403 nm, while frozen-thawed fish was characterized by a maximum located at 379 nm and a shoulder at 455 nm. By applying PCA to NADH spectra, good discrimination between fresh and frozen fish samples was observed, which was confirmed by applying FDA to the first five PCs of the PCA performed on the NADH spectra. Indeed, 100% correct classification was obtained for the calibration and validation data sets, respectively. One of the interesting conclusions of this research was that NADH fluorescence spectra may be considered a promising tool for differentiating between fresh and frozen-thawed fish samples.

Eggs and Egg Products

In order to ensure high and consistent egg quality, an attractive and alternative strategy for determining the state of egg freshness can be achieved by sensor technologies. These techniques (such as NIR, MIR, fluorescence spectroscopies, etc.) appear to be promising tools for non-destructively determining egg freshness. Such methods cannot eliminate the need for more detailed physicochemical analyses, but they may help to screen for samples that require further examination. Freshness makes a major contribution to the quality of eggs and egg products. One of the main concerns of the egg industry is the systematic determination of egg freshness because consumers may perceive variability in freshness as lack of quality.

The changes that occur in eggs during storage are many and complex and affect the functional properties of egg yolk and egg albumen. These changes include thinning of albumen, increase of pH, weakening and stretching of the vitelline membrane, and increase in the water content of the yolk. Freshness can be explained to some extent by objective sensory, (bio)chemical, microbial, and physical parameters and can therefore be defined as an objective attribute. Knowledge of the various descriptors of properties that are encountered in eggs immediately after laying must be known, as well as the changes in properties that take place over time. This information can be gained by performing controlled storage experiments that extend from the time after laying; loss in freshness and spoilage can thus be monitored. Posudin (1998) assessed the potential of FFFS to determine egg freshness by using ultraviolet radiation for the quality evaluation of eggs with differing levels of pigmentation. The emission spectra of different eggs showed two maxima located at 635 and 672 nm (ascribed to the pigments of porphyrin and porphyrin derivatives of florin and oxoflorin) after excitation at 405, 510, 540, and 557 nm. The intensity at 672 nm depends on the egg freshness. The autofluorescence of fresh egg is stronger than that of old one, since the intensity of autofluorescence depends on the amount of porphyrin on the shell surface. From these preliminary results, the author concluded that fluorescence spectroscopy could be a promising approach for quantitative estimation of porphyrin in eggs and thus determine egg freshness. Recently, FFFS was used to monitor egg freshness during storage (Karoui et al. 2006c, d). The authors found that FMRP (excitation 360 nm; emission 380–580 nm) recorded on thick and thin albumens and vitamin A scanned on egg yolk (emission 410 nm; excitation 270–350 nm) could be considered as powerful tools for the evaluation of egg freshness stored at room temperature, while tryptophan fluorescence spectra recorded on thick and thin albumens and egg yolk failed to discriminate between fresh and aged eggs. Using excitation at 360 nm, the emission spectra recorded on fresh thick egg albumen exhibited two maxima located at 410 and 440 nm, respectively. Similar results were obtained on thin albumen of fresh eggs. The very characteristic fluorescence spectra of thick and thin albumen of eggs stored for a long time (i.e., 18 days or more) at room temperature showed a shoulder located at 414 nm and a maximum at approximately 438 nm. In addition, as the spectra showed large differences between fresh thin/thick egg albumens and those stored for a long time (29 days), the authors considered the spectra as fingerprints for freshness identification. Indeed, thick and thin albumens of fresh eggs within 2–3 days of laying had the highest intensity at 410 nm, while aged eggs had the lowest one. The authors concluded that the shape of FMRP is correlated with storage time: thick albumen of fresh eggs had the lowest ratio of fluorescence intensity FI440 nm/FI410 nm (i.e., 1.0), while that of eggs stored for 29 days had the highest (i.e., 1.30). The changes in the FI440 nm/FI410 nm ratio has been ascribed to the change in the viscosity of both thick and thin egg albumens and the formation of furosine during storage (Birlouez-Aragon et al. 1998; Kulmyrzaev and Dufour 2002). Indeed, it has been reported that during egg storage, a decrease in the viscosity of thick albumen was observed (Lucisano et al. 1996). This phenomenon has been attributed to the separation of the β-fraction of ovomucin, rich in carbohydrate, and from the ovomucin–lysozyme complex. However, in the study by Karoui et al. (2006c, d), eggs were stored at room temperature, and little attention has been given to the influence of temperature and relative humidity variations on fluorescence measurements, although Stadelman et al. (1954) observed a linear decrease of −1.15 Haugh Units per 10 °C increase in testing temperature. Therefore, Karoui and co-worker have continued this work by investigating changes at the molecular level of 126 eggs stored at 12.2 °C and 87% RH for 1, 6, 8, 12, 15, 20, 22, 26, 29, 33, 40, 47, and 55 days (Karoui et al. 2007d). Of the intrinsic fluorophores tested, only PCA applied to the vitamin A fluorescence spectra allowed good identification of eggs as a function of their storage time. By applying FDA to the AAA+NA spectra, correct classification rates of 69.4% and 63.9% were observed for the calibration and validation sets, respectively. Quite similar results were obtained with AAA+NA scanned on egg yolks. The best results were obtained with vitamin A fluorescence spectra, where correct classification rates of 97.7% and 85.7% in the calibration and validation sets were obtained, respectively. The authors concluded that vitamin A fluorescence spectra provide useful fingerprints allowing the identification of eggs during storage at low temperature and could be considered as a powerful intrinsic probe for the evaluation of egg freshness. Karoui and co-workers have continued this work by testing the ability of vitamin A fluorescence spectra to monitor changes at the molecular level of 225 eggs stored at 12.2 °C and 87% RH in an atmosphere containing 2% (n = 108) and 4.6% (n = 99) of CO2 for 55 days (Karoui et al. 2007e). Again, vitamin A fluorescence spectra allowed good discrimination of eggs according to both storage time and conditions, while more overlapping between egg samples was observed when the other intrinsic probes were investigated: eggs aged 22 days or less were separated from those aged 26 days or more. In addition, eggs were found to be well separated for each storage time, except for those samples aged 20 and 22 days and those aged 26 and 29 days, where some overlapping was observed.

Edible Oils

Olive oil is an economically important product of Mediterranean countries. It has a fine aroma and pleasant taste, with excellent health benefits. The quality of olive oil ranges from the high-quality extra-virgin olive oil (EVOO) to the low-quality olive-pomace oil. EVOO is obtained from the fruit of the olive tree by mechanical pressing and without refining processes. Owing to its high quality, it is the most expensive type of olive oil. For this reason, it may be mislabeled or adulterated for economic reasons. Mislabeling often involves false information regarding the geographic origin or oil variety (Aparicio et al. 1997). Adulteration involves the addition of cheaper oils; the most common adulterants found in virgin olive oil are refined olive oil, residue oil, synthetic olive oil–glycerol products, seed oils, and nut oils (Baeten et al. 1996; Downey et al. 2002; Sayago et al. 2004). Owing to the low price of olive-pomace oil, it is sometimes used to adulterate EVOO. For this reason, a rapid method to detect such a practice is important for quality control and labeling purposes. In this context, Zandomeneghi et al. (2005) recorded fluorescence spectra of EVOO using right angle and FFFS. The former method showed considerable artifacts and deformation, while the latter provided spectra that are much less affected by self-absorption. The authors attributed this to the self-absorption phenomena when using right-angle fluorescence, even when the spectra are corrected for inner filter effects. In another study, Sayago et al. (2004) applied fluorescence spectroscopy for detecting hazelnut oil adulteration in virgin olive oils. Virgin olive, virgin hazelnut, and refined hazelnut oil samples and a mixture of them at 5%, 10%, 15%, 20%, 25%, and 30% adulteration were analyzed after excitation at 350 nm. By performing linear discriminant analysis (LDA), 100% correct classification was achieved. In a similar approach, Kyriakidis and Skarkalis (2000) used excitation wavelength of 360 nm to differentiate between common vegetable oils, including olive oil, olive residual oil, refined olive oil, corn oil, soybean oil, sunflower oil, and cotton oil. All the oils studied showed a strong fluorescence band at 430–450 nm, except for virgin olive oil, which exhibited a low intensity at both 440 and 455 nm, a medium band around 681 nm and a strong one at 525 nm. The latter two bands have been ascribed to chlorophyll and vitamin E compounds, respectively. The very low intensity of the peaks at 445 and 475 nm is due to the high content of phenolic antioxidants, which provide more stability against oxidation. All refined oils showed only one intense peak at 445 nm, which is due to fatty acid oxidation products formed as a result of the large percentage of polyunsaturated fatty acids present in these oils.

Oil (and food systems in general) could be considered as a complex system, which presents a set of different properties and contains many fluorescent molecules. The use of only excitation or emission wavelengths could limit the ability of this technique to determine the quality of oil samples. To comply with this requirement, the variation in the excitation and emission wavelengths allows simultaneous determination of compounds present in oils. This could be realized by using synchronous fluorescence spectroscopy. In this context, Sikorska et al. (2004, 2005) used synchronous fluorescence spectroscopy with excitation wavelengths from 250 to 450 nm and emission spectra in the range 290 to 700 nm. The peak located at 320 nm after excitation at 290 nm has been attributed to tocopherols, while the band located at 670 nm in emission and 405 nm in excitation belongs to pigments of the chlorophyll group. In order to compare the set of synchronous fluorescence spectra of different oils, Sikorska et al. (2005) applied the k-nearest neighbors method, and good discrimination between oil samples with a very low classification error ranging between 1% and 2% and a low standard deviation value was obtained. Guimet et al. (2004) explored the use of FFFS to discriminate between virgin and pure olive oils; the ranges studied were at excitation wavelength λex = 300–400 nm and emission wavelength (λem) 400–695 nm and λex = 300–400 nm and λem 400–600 nm. The first range was found to contain chlorophylls, whereas the second range contained only the fluorescence spectra of the remaining compounds (oxidation products and vitamin E). Later, Guimet et al. (2005) applied PARAFAC to detect the adulteration of EVOO with olive-pomace oil at low levels (5%). Discrimination between the two types of oils was achieved by applying both LDA and discriminant multi-way PLS regression; the latter method gave 100% correct classification. The same technique (synchronous fluorescence) was used to analyze 73 samples, including 41 edible and 32 lampante virgin olive oils collected in October and November 2002 (Poulli et al. 2005). PCA and hierarchical cluster analysis applied to the emission spectra in the range 350–720 nm (at excitation wavelengths varying from 320 to 535 nm) showed good separation between the two types of oils. Recently, the same research group assessed the potential of synchronous fluorescence spectra to detect adulteration of virgin olive oil (VOO) with other oils (Poulli et al. 2007). By applying PLS regression to the excitation spectra recorded 250–720 nm with a wavelength interval of 20 nm, the authors stated that FFFS could be useful for the detection of olive-pomace, corn, sunflower, soybean, rapeseed, and walnut oils in VOO at levels of 2.6%, 3.8%, 4.3%, 4.2%, 3.6%, and 13.8%, respectively. Frying oil deterioration has also been measured by using five selected excitation wavelengths varying from 395 to 530 nm (Engelsen 1997). By applying PLS regression, the author showed good correlation between fluorescence spectra and quality parameters describing the deterioration (e.g., anisidine value, iodine value, oligomers, and vitamin E).

Cereals and Cereal Products

The potential of fluorescence spectroscopy for monitoring cereals has increased over the past few years with the propagated application of chemometric tools and with technical and optical developments of the spectrofluorimeter. Zandomeneghi (1999) used FFFS (excitation 275 nm; emission 280–575 nm) to differentiate between different cereal flours (i.e., rice, creso, maize, pandas). The same research group also utilized visible excitation set at 445 nm (emission 460–600 nm) to differentiate between flours of five different wheat varieties and a good discrimination was observed. In another study, excitation wavelengths set at 275, 350, and 450 nm presenting fluorescence emission maxima at 335, 420, and 520 nm, respectively, were utilized to classify botanical tissue components of complex wheat flour and rye flour; the bands were attributed to AAA, ferulic acid, and riboflavin components, respectively. The last fluorescent component was confirmed by Zandomeneghi et al. (2003), who attributed the band observed at 520 nm to riboflavin, while the band between 430 and 530 nm was found to be proportional to the lutein content of the flour (Zandomeneghi et al. 2000).

Ferulic acid and riboflavin spectra have been reported to have good accuracy for monitoring wheat flour refinement and milling efficiency following the use of fluorescence imaging. Successful classification was obtained, suggesting that FFFS may be used to classify wheat cultivars (Symons and Dexter 1991, 1992, 1993, 1994). These results have recently been confirmed by Karoui et al. (2006a) where tryptophan fluorescence spectra of 59 samples (20 complete Kamut®, semi-complete Kamut®, and soft wheat flours, 28 pasta and 11 semolinas manufactured from complete Kamut®, semi-complete Kamut®, and hard wheat flours) were scanned after excitation at 290 nm. PCA performed on the flours’ spectra clearly differentiated complete Kamut® and semi-complete Kamut® samples from those produced from complete and semi-complete soft wheat flours, while good discrimination of pasta samples manufactured from complete Kamut® and complete hard wheat flours from those made with semi-complete Kamut® and semi-complete hard wheat flours was achieved. The best discrimination was obtained with tryptophan spectra recorded on semolinas, since the four groups were well discriminated. Indeed, by applying FDA to the spectral collection, 86.7% and 87.9% correct classification rates were obtained for the calibration and validation samples, respectively. In a similar approach, emission spectra (370–570 nm) were recorded after excitation at 350 nm on red and white wheat kernels and a clear difference was observed between the two group samples (Ram et al. 2004); this difference has been attributed to the morphological variation in the pericarp and nuclear organization of the two varieties of wheat.


In combination with multivariate statistical analyses, fluorescence spectroscopy has proved to be a promising screening method for predicting quality parameters in beet sugar samples (Munck et al. 1998). Indeed, it has been shown that commercial sugars exhibit characteristic fluorescence, which can be used to obtain information regarding minor constituents in the sugar. Fluorescence has successfully been applied to the beet sugar manufacturing process with the use of multivariate data analysis (Munck et al. 1998). The same approach with multiple excitation and emission wavelengths used by Carpenter and Wall (1972) has also been employed, and interesting results were obtained. In a study of beet sugar samples, it was possible to classify white sugar samples according to factory and to predict quality parameters such as amino nitrogen, color, and ash from the fluorescence data of these samples (Nørgaard 1995). The fluorescence data of thick juice samples showed more ambiguous results owing to the more complex sample composition. Another study of beet sugar samples utilized the three-dimensional structure of the fluorescence excitation–emission landscapes to resolve spectral excitation and emission profiles of fluorophores present in sugar with a multi-way chemometric model, PARAFAC (Bro 1999). Four fluorescent components were found to capture the variation in the fluorescence data of 268 sugar samples collected from a beet sugar factory in a single campaign, and two of them showed spectra with a close similarity to the pure fluorescence spectra of the amino acids tyrosine and tryptophan. The concentrations of the four components estimated from the sugar samples could be correlated with several quality and process parameters, and they were characterized as potential indicator substances of the chemistry in the sugar process, which has been confirmed by the use of HPLC analysis combined with fluorescence detection on thick juice samples and evaluation by PARAFAC (Baunsgaard et al. 2000a). Seven fluorophores were resolved from thick juice. Apart from tyrosine and tryptophan, four of the fluorophores were identified as high molecular weight compounds, which were related to colorants absorbing at 420 nm. Three of the high molecular weight compounds were found to be possible Maillard reaction polymers. The last of the seven fluorophores indicated a compound with polyphenolic characteristics. In a fluorescence study of 47 raw cane sugars collected from many different locations and campaign years, three individual fluorophores were found; one of them, representing maximum excitation and emission at 275 and 350 nm, respectively, was characterized as an ultraviolet color precursor that participates in color development during storage. The other two (340, 420 nm and 390, 460 nm excitation/emission in the visible wavelength area) are considered to be potential colorants, which shows a link with their fluorescence behavior (Baunsgaard et al. 2000b). Recently, FFFS has been used to monitor adulteration of honey with cane sugar syrup (Ghosh et al. 2005). Using an excitation wavelength of 340 nm, pure honey samples were characterized by two prominent features—a shoulder located at around 440 nm and a maximum located at 510 nm, which has been ascribed to flavins—while cane sugar syrup samples exhibited a maximum located around 430 nm. The peaks located at 440 and 430 nm in pure honey and sugar syrup samples have been attributed to NADH. Synchronous fluorescence was then applied to differentiate between pure honey and sugar syrup samples, and good discrimination between the two groups was observed. The spectra for cane sugar syrup were characterized by a shoulder around 305 nm and a prominent band around 365 nm, while honey samples had a strong peak around 460 nm and a much weaker peak around 365 nm. The authors observed an increase in the intensity at 365 and 425 nm, as well as the ratio of FI365/FI425, with the increase of cane sugar syrup concentration; thus the ratio of FI365/FI425 has been suggested as a potential method to monitor adulteration of honey with cane sugar syrup. In another study, fluorescence spectra were scanned on 62 honey samples belonging to seven floral origins after excitation at 250 nm (emission 280–480 nm), 290 nm (emission 305–500 nm), and 373 nm (emission 380–600 nm) and emission set at 450 nm (excitation 290–440 nm) by Ruoff et al. (2005) and Karoui et al. (2007a). By applying FDA to the four data sets (concatenation), correct classification rates of 100% and 90% were observed for the calibration and validation samples, respectively. In addition, the seven honey types were well discriminated, indicating that the molecular environments, and thus the physicochemical properties, of the investigated honeys were different. One of the main findings of these studies is that FFFS might be a suitable and alternative technique to classify honey samples according to their botanical origins; this was confirmed recently by Ruoff et al. (2006), who studied 371 honey samples originating from Switzerland, Germany, Italy, Spain, France, Slovenia, and Denmark. By using chemometric tools, the error rates of the discriminant models ranged from 0.1% to 7.5%.

Fruit and Vegetables

Chlorophyll fluorescence has been used as an intrinsic probe to determine the physiological status of whole plants and plant organs (Song et al. 1997). This component has been considered an efficient probe for monitoring apples during maturation, ripening, and senescence (Song et al. 1997). Recently, fluorescence spectroscopy has been considered to have the potential for assessing the mealiness of apples (Moshou et al. 2005), since relatively good correlation was obtained between mealiness and fluorescence spectra. Other authors have used chlorophyll fluorescence of apples as a potential predictor of superficial scald development during storage (De Ell et al. 1996) and for the estimation of anthocyanins and total flavonoids in apples (Hagen et al. 2006). In another study, Lötze et al. (2006) used fluorescence imaging as a non-destructive method for the pre-harvest detection of bitter pit in apples; the same technique had already been utilized to determine apple juice quality (Seiden et al. 1996; Noh and Lu 2007). Two excitation wavelengths, set at 265 and 315 nm, were chosen as they yielded the richest spectra of two juice-apple varieties (Jonagold and Elstar). The spectra showed two excitation–emission maxima (315/440 and 265/350 nm) that have been not attributed to any component in apple juice (Seiden et al. 1996). Applying PCA to the two juice-apple varieties, good discrimination was observed—which was not achieved with titratable acidity or soluble solids data. The authors pointed out that an increase in the ripening process of apples involves an increase in the soluble fluorescent compounds. Good correlation between soluble solids and fluorescence spectra was observed independent of the apple varieties, indicating the possibility of modeling the progression in maturity with information obtained from spectra, while fluorescence was found to correlate poorly to the amount of titratable acids in juice.

Identification of Bacteria of Agro-alimentary Interest

The identification of microorganisms of agro-alimentary interest in food and food products by conventional phenotypic procedures based on morphology and biochemical tests involves a large quantity of reagents and, in some cases, is unable to discriminate microorganisms at the strain level. In this context, Leblanc and Dufour (2002) assessed the potential of different intrinsic probes (i.e., tryptophan, AAA+NA, and NADH) to discriminate between 25 strains of bacteria in dilute suspensions. The best results were obtained by using AAA+NA spectra, where correct classification rates of 100% and 81% were observed for the calibration and validation samples, respectively. The authors noted that fluorescence spectroscopy is able to discriminate and identify bacteria at genus, species, and strain levels. This assumption was later demonstrated by the same research group (Leblanc and Dufour 2004). In their studies, the spectra of three bacteria strains (Lactococcus lactis, Staphylococcus carnosus, and Escherichia coli) were recorded at different growth phases. By applying PCA to the spectra scanned on each bacteria, three groups corresponding to three main phases of growth were identified (lag phase, exponential phase, and stationary phase). The authors then gathered the spectra recorded on the three bacteria into one matrix, and this new matrix was analyzed by PCA. The obtained results showed good discrimination of spectra according to bacteria and metabolic profile. Recently, Ammor et al. (2004) utilized the same technique for the identification of lactic acid bacteria isolated from a small-scale facility producing traditional dry sausages. Again, fluorescence spectroscopy demonstrated its ability to discriminate between Lactobacillus sakei subsp. carnosus and Lactobacillus sakei subsp. sakei. In another approach, Leriche et al. (2004) isolated 30 Pseudomonas spp. strains from milk, water, cheese center, and cheese surface belonging to two traditional workshops manufacturing raw milk Saint Nectaire cheese. By applying FDA to the data sets, clear linkages between groups of isolates were noted. In the first workshop, the milk was implicated being as the sole source of cheese contamination, whereas in the second workshop the milk and cheese center isolates were found to be similar, but different from surface cheese isolates. The authors attributed this contamination at the cheese surface to the water used during the ripening process (washing of the cheese surface). From the results obtained, it was stated that it is possible to characterize, differentiate, and trace Pseudomonas spp. strains using the fluorescence technique. These findings were reinforced by the high correlation (using CCA) observed between the data sets obtained from the metabolic profiling and fluorescence spectroscopy. The obtained results were fully supported by the same research group (Tourkya et al. 2009) since good discrimination, even for strains for which ambiguity still remained after PCR and Analytical Profile Index 20 NE identification.

Saito (2009) depicted that Napa cabbage (Brassica rapa L.) laser-induced fluorescence (LIF) spectra of a normal core and a rotten core show greater intensities in the green wavelength range of the LIF spectrum of the rotten core in comparison to that of the normal core. The integrated peak area between 450 and 600 nm for the rotten core was more than twice that for the normal core. Regarding pear (Pyrus communis L.), differences were found in the LIF spectra from the surface of a young pear and a ripe pear. The intensity of the blue region (400–500 nm) of the LIF spectrum in the ripe pear surface was increased in comparison to the pear picked at harvest time in a still unripe stage. In another approach, (Kim et al. 2003) developed a LIF imaging food system (i.e., meat pork and apple). Multispectral fluorescence emission images were recorded after excitation set at 355 nm and the system fluorescence emission images were captured in the blue, green, red, and far-red regions of the spectrum centered at 450, 550, 678, and 730 nm, respectively, from a 30-cm diameter target area in ambient light. Images of apples and of pork meat artificially contaminated with diluted animal feces with excitation at 355 nm and emission at 450, 550, 678, and 730 nm have also demonstrated the versatility of fluorescence imaging techniques for potential applications in food safety inspection. The most interesting conclusion was that regions of contamination that were not readily visible to the human eye were found to be easily identified from the fluorescence images.


As illustrated in the present review, the environment of intrinsic fluorophores recorded on intact food systems contains valuable information regarding the composition and nutritional values of food products. The huge potential for the application of fluorescence spectroscopy combined with multivariate statistical analyses for the evaluation of food quality has also been demonstrated; the great difference between food systems has been related to the differences in the molecular structure of the samples resulting in a variation of the optical pathway of excitation light and fluorescence inside the optically complex natural food systems. The method is suitable as an effective research tool and can be a part of evaluation procedure for food quality.

Calibration stage and development of the calibration equations are the limiting steps for adopting the fluorescence as a technique for the determination of the quality and authenticate food products, as it is time-consuming and costly procedure. However, when the calibration stage is accomplished successfully, the determination of a chemical property or the geographical origin can be carried out very rapidly with a single analysis for minimal cost.

This development could make the FFFS a powerful tool for its use for on-line process control. In this context, FFFS sensors would give valuable information in comparison to the NIR since the information given by the latter are based on molecular overtone vibrations, which are less sensitive and specific. More recent improvement in the fluorescence instrument passes through the development of spectrometers that allows on-line measurements. The main innovation in this area is the use of fiber optics for the connection between the spectrometer and the sensing device. Even though the review focuses on examples from the food industry, the principles are broader and fluorescence could be applied to other fields (pharmaceutical, biotechnology, etc.). It is therefore expected that in the coming years, FFFS combined with chemometric tools would be a reliable tool for understanding the bases of food molecular structure and, as a consequence, for their qualities.

Copyright information

© Springer Science + Business Media, LLC 2010