Interactions of Milk Proteins and Volatile Flavor Compounds: Implications in the Development of Protein Foods



ABSTRACT:  This review encompasses the binding of volatile flavor compounds by milk proteins in aqueous solutions. The presence of proteins in a food matrix can result in a decrease in aroma perception and in an unpleasant aroma profile, because of binding of the desirable flavor compounds to proteins. Hence, various analytical methods used to measure the extent and the type of binding, and the determination of the binding parameters, are evaluated in this review. The binding of various flavor compounds by individual milk proteins is discussed and compared in terms of their binding affinity for flavor compounds. Furthermore, the influence of temperature and ultra-high pressures on the interactions between proteins and flavors is considered in detail. The implications of protein–flavor binding in the development of protein foods are discussed.


One of the most important criteria for consumer acceptance of foods is flavor. Food matrix components, such as proteins (Gremli 1974; Damodaran and Kinsella 1980a, 1980b, 1981a, 1981b; Farès and others 1998; Lübke and others 2002; Gianelli and others 2005), carbohydrates (Yven and others 1998; Heinemann and others 2001; Philippe and others 2003; Jouquand and others 2004), and lipids (Ebeler and others 1988; van Ruth and others 2002; Meynier and others 2003), are known to interact with flavor compounds. Proteins are added to foods primarily because of their functional properties, such as emulsifying and stabilizing capacities, and because of their nutritional value. However, interactions between proteins and flavors are known to influence the perceived flavor of a food product (Overbosch and others 1991; Land 1996). Protein ingredients not only reduce the perceived impact of desirable flavors but also may transmit undesirable off-flavors to foods, especially whey protein and soy protein products (Mills and Solms 1984; Semenova and others 2002a). In addition, proteins may change the texture of a food that is gelling, and thus decrease the flavor perception due to inhibition of mass transfer (Jaime and others 1993; Carr and others 1996; Wilson and Brown 1997).

Two different types of interaction can occur between proteins and flavor compounds: (1) reversible (physicochemical) binding, including hydrogen bonds, hydrophobic interactions, and ionic bonds, and (2) irreversible (chemical) binding via covalent linkages, that is, amide and ester formation, and the condensation of aldehydes with amino (NH2) and sulfhydryl (SH) groups (Solms and others 1973; Overbosch and others 1991; Mottram and others 1996; Meynier and others 2003, 2004). The type of binding between a protein and a flavor compound depends on the characteristics of both the protein and the flavor, and thus there is no universal mechanism for flavor binding in foods (Solms and others 1973). However, most flavors show hydrophobic, reversible binding to proteins (Gremli 1974; Damodaran and Kinsella 1980b, 1981a, 1981b, 1983; O'Neill and Kinsella 1987; Pelletier and others 1998). Adsorption and absorption also exist but are specific for low-moisture food systems (Maier 1972, 1973, 1975; Le Thanh and others 1992; Landy and others 1997; Mironov and others 2003), and are not discussed in this review.

In the area of protein–flavor interactions, studies have been conducted mainly with milk proteins (Andriot and others 2000; Lübke and others 2002) and soy proteins (Damodaran and Kinsella 1981a; Li and others 2000), but also with a range of other food proteins, such as fababean protein (Ng and others 1989a, 1989b; Semenova and others 2002a), pea protein (Dumont and Land 1986), ovalbumin (Maier 1970; Ebeler and others 1988), fish actomyosin (Damodaran and Kinsella 1983), and myoglobin (Gianelli and others 2005). The studies on milk proteins are of great importance because milk proteins are utilized in numerous food products, including dairy products, bakery and confectionary products, and meat products (Mulvihill 1992).

The demand for healthier low-fat foods is increasing. However, in the absence of fat, altered flavor–matrix interactions result in a dramatic change in flavor profiles (Shamil and Kilcast 1992; Plug and Haring 1993; Hatchwell 1996). In fat-reduced or “light” foods the dominant components are carbohydrates or proteins, which interact differently with aroma compounds compared with fat, and thus change the perceived flavor. In addition, most of the fat replacers used in these foods are composed of proteins or carbohydrates or both. A better understanding of the science behind protein–flavor interactions is required for the development of improved food flavor, particularly that of low-fat foods, and for minimizing the presence of off-flavors in protein-containing foods.

The aim of this review is to discuss the flavor-binding ability of individual milk proteins and milk protein products, and the influence of processing, particularly temperature and high pressure, on protein–flavor binding. Methodologies used to investigate protein–flavor interactions in different systems are evaluated. Implications of protein–flavor interactions in the development of protein foods, and future research, are also considered.

Analysis of Flavor Binding by Proteins

Two approaches may be used to study the interactions between flavor molecules and macromolecules, such as proteins: instrumental techniques and sensory analysis. The systems considered are often very simple, consisting of 1 flavor compound and 1 protein, usually in an aqueous solution.

Flavor binding has been examined predominantly under equilibrium conditions. A common technique is equilibrium dialysis, which is still commonly used (Damodaran and Kinsella 1981a, 1981b, 1983; Druaux and others 1995; Farès and others 1998; Muresan and others 2001; Burova and others 2003; Jung and Ebeler 2003a; Guth and Fritzler 2004). However, this technique is time consuming, and flavor losses during the experiment may occur. Another frequently used equilibrium method is static headspace analysis (O'Keefe and others 1991a; Charles and others 1996; Andriot and others 2000; van Ruth and Villeneuve 2002; Heng and others 2004; Liu and others 2005a). A drawback of the static headspace method is poor sensitivity for compounds with little volatility. For some samples this problem can be overcome by split less or on-column techniques, or by increasing the temperature. However, thermal reactions may occur. Aroma compounds that are not detectable by flame ionization detection (FID) or mass spectrometry (MS) can often be detected using the headspace gas chromatography–olfactometry (GC–O) technique (Widder and Fischer 1996).

Recently, static headspace solid-phase microextraction (SPME) has been found to be very useful for the examination of protein–flavor binding (Roberts and Pollien 2000; Adams and others 2001; Fabre and others 2002; Zhu 2003; Jung and Ebeler 2003a; Gianelli and others 2005). The SPME technique has been developed by Pawliszyn and co-workers (Arthur and Pawliszyn 1990; Zhang and Pawliszyn 1993; Zhang and others 1994). In headspace SPME, a fused silica fiber coated with a thin layer of a selective coating extracts the flavor compounds from the headspace above the sample (Figure 1). The analytes are concentrated in the coating and are then transferred to the analytical instrument for desorption and analysis (Zhang and Pawliszyn 1993). It is a sensitive, rapid, inexpensive, selective, and solvent-free sample preparation technique, and is compatible with a wide range of separation methods such as GC, GC–MS, and high-performance liquid chromatography (HPLC) (Zhang and others 1994). Headspace SPME has been found to be superior to both static and dynamic headspace analysis for the measurement of milk protein–flavor interactions (Fabre and others 2002). The technique can also be used to quantify the flavor concentrations after equilibrium dialysis (Zhou and others 2002).

Figure 1—.

Illustration of the partition process in headspace SPME for investigating interactions of proteins (P) and flavor compounds ( •) (not to scale)

However, there are some limitations of headspace SPME, such as the difficulty in using external standards for more complex sample matrices (Yang and Peppard 1994). Instead, standard addition and isotopic dilution can be used for quantitative analysis of such samples by SPME (Hawthorne and others 1992). In addition, competition of flavor compounds in the fiber coating can cause biases in the quantitative analysis (Coleman 1996; Grote and Pawliszyn 1997; Roberts and others 2000). Hence, the method is more suitable for simple systems rather than complex systems. In addition, it has to be ensured that the amount of flavor extracted from the headspace is small to avoid re-equilibration of the system. The principle behind headspace SPME is the equilibrium partition process of analytes between the sample, the headspace above the sample, and the fiber coating (Figure 1). The overall equilibrium of the system can be described by the following equation:


where [F ]f is the concentration of the flavor compound in the fiber coating, [F ]h in the headspace, [F ]w in the aqueous phase, and [F ]b bound by the protein. The partitioning of the flavor compound in the system depends on the fiber–headspace partition coefficient (Kfh), the headspace–water partition coefficient (Khw), and the binding constant between the protein and the flavor (K) (Figure 1). Too long extraction times and too high values of Kfh would result in flavor being released from the sample solution, and, as a consequence, protein-bound flavor could be released into the aqueous phase and headspace, and the measured binding constant K would be underestimated. Hence, sampling parameters have to be chosen with care.

Other equilibrium techniques include liquid–liquid partitioning (Spector and others 1969; Spector 1975; Damodaran and Kinsella 1980b) and radioactivity counting (King and Solms 1979). For a few hardly volatile and water-soluble flavor compounds, such as vanillin or benzaldehyde, HPLC methods have been developed to determine the free flavor after equilibration with the protein (Ng and others 1989a; McNeill and Schmidt 1993; Andriot and others 1999; Li and others 2000; Chobpattana and others 2002). However, several disadvantages are associated with equilibrium methods, such as long equilibration times and thus the possible degradation of the volatiles during this period.

The conformational stability of protein–flavor complexes upon heat denaturation can be examined using differential scanning calorimetry (DSC) methods (Burova and others 1999; Semenova and others 2002b, 2002c). DSC also allows the calculation of binding constants using bovine serum albumin (BSA) as an internal molecular detector since its denaturation temperature and thus conformational stability depends on the amount of flavor bound (Burova and others 2003). However, the amount of bound flavor determined this way is not precise.

Some dynamic methods, that is, dynamic headspace analysis (exponential dilution), are used currently (Jouenne and Crouzet 1996; Farès and others 1998; Jouenne and Crouzet 2000b). The flavor compounds in the headspace are of greater importance than those in the food as they can travel to the nose during eating and stimulate the olfactory receptors in the nasal cavity (Linforth and Taylor 1993). In dynamic headspace analysis, the exponential decrease of the flavor concentration in the headspace above the sample is measured, while the sample is exhausted of aroma volatiles by passing an inert gas through it or sweeping it over its surface. The gas leaving the system is sampled at regular intervals and analyzed using GC. Purge-and-trap analysis is a commonly used method of dynamic headspace analysis and enables fast determination of the activity coefficient of the flavor compound in a model food system (Sorrentino and others 1986; Jouenne and Crouzet 1996, 2000b). Disadvantages are the purging of water along with the analytes, decreasing the adsorption capacity of the trap, and diffusion or “breakthrough” losses of analytes not retained by the trap (Stevenson and others 1996).

Other dynamic methods involve liquid chromatography, such as size exclusion chromatography (Hummel and Dreyer 1962; Dumont 1987; Pelletier and others 1998), dynamic coupled column liquid chromatography (DCCLC) (Langourieux and Crouzet 1995; Jouenne and Crouzet 2000a), or affinity chromatography on protein-bonded stationary phases (Pelletier and others 1998; Sostmann and Guichard 1998; Reiners and others 2000). The methods based on liquid chromatography need small amounts of product and are rapid, thereby reducing the degradation of analytes. In spite of this, affinity chromatography only provides global affinities. The number of binding sites has to be determined through exclusion size chromatography or other methods. Another drawback of affinity chromatography is that possible conformational changes in proteins due to their immobilization have to be considered, and some protein-binding sites may be hindered, as observed for isoamyl acetate (Pelletier and others 1998).

The above-mentioned methods can be used to demonstrate the existence and extent of molecular interactions between aroma volatiles and proteins or other food constituents. However, they do not provide information about the nature of interactions. Spectroscopic techniques have been used successfully to gain insight into the nature of interactions. Binding of a flavor compound to a protein molecule can generate conformational modifications of the protein (Damodaran and Kinsella 1980b; Lübke and others 2000). These conformational changes, and thus the binding of ligands can be measured by following the change in protein tryptophan fluorescence, which can be increased or quenched depending on the ligand present (Muresan and others 2001). The technique also indicates if a ligand binds in the vicinity of a tryptophan residue. Assuming that the change in fluorescence depends on the amount of protein–ligand complex, fluorescence spectroscopy can be used to determine binding constants (Dufour and Haertlé 1990; Frapin and others 1993; Liu and others 2005a). However, the protein investigated needs to contain at least 1 tryptophan residue. In addition, for ligands that quench the tryptophan fluorescence, the binding is overestimated and should always be verified using a second method (Muresan and others 2001).

Another technique monitoring conformational changes of proteins upon flavor binding is infrared (IR) spectroscopy (Lübke and others 2000). The technique is a very useful tool to study protein secondary structures. Generally, the amide I bands region between 1600 and 1700 per cm reveals the most of the information as it is highly conformation-sensitive. When using either fluorescence or IR spectroscopy it has to be considered that flavor binding without conformational changes in the protein cannot be investigated using these techniques.

Nuclear magnetic resonance (NMR) spectroscopy techniques are a very valuable tool to explore mechanisms of protein–flavor interactions. Two-dimensional (2D) NMR spectroscopy reveals conformational changes of the protein upon flavor binding, and those amino acid residues involved in the binding so that information about the location and number of binding sites on a protein can be obtained (Lübke and others 2002). Diffusion-based NMR techniques are fast and easy but do not offer information about the nature of interactions. The pulsed field gradient NMR (PFG-NMR) method could provide binding constants K and number of binding sites n on the proteins. However, using this technique it is assumed that there are no interactions between the proteins or the flavor ligands themselves but this assumption is not true. A powerful tool for rapid screening of flavors that have an affinity for a protein is the diffusion-based nuclear Overhauser effect (NOE) pumping technique (Jung and others 2002; Jung and Ebeler 2003b). This technique should only be used as a screening method as it lacks sensitivity.

Sensory analysis has also been applied to examine the effect of proteins on flavor perception (Ng and others 1989b; Hansen and Heinis 1991, 1992; McNeill and Schmidt 1993; Reiners and others 2000). Rating of flavor intensities in the presence and absence of protein gives insight into the effect of protein–flavor binding on flavor perception and complements instrumental techniques. However, an intensive training of the taste panel is necessary to obtain precise results, and thus is time consuming and costly.

Determination of Binding Parameters

To characterize the binding of a specific flavor to a specific protein, binding parameters, such as an intrinsic binding constant K, the number of binding sites n, the Hill coefficient h, the Gibbs' free energy of binding ΔG, the enthalpy change ΔH, and the entropy change ΔS, can be determined experimentally. Different plots have been used to determine these parameters, including the Scatchard plot (Scatchard 1949), the Klotz plot (Klotz and others 1946; Klotz and Urquhart 1948), and the Hill plot (Hill 1910; Yven and others 1998). The experimental determination of K as a function of temperature allows the determination of the thermodynamic parameters ΔG, ΔH, and ΔS (Steinhardt and Reynolds 1969).

The frequently used Scatchard and Klotz models assume equal and independent binding sites on a protein. However, proteins can possess nonequivalent binding sites, for example, 1 or more high-affinity primary binding sites and a group of lower affinity secondary binding sites. This results in nonlinear Scatchard and Klotz plots and complicates the evaluation of binding data because it is often difficult to extrapolate the binding plots to obtain the exact number of equivalent binding sites. Furthermore, binding sites can be dependent. The initial binding of aroma compounds can cause a protein to undergo conformational changes (Damodaran and Kinsella 1980b), revealing new binding sites. If the initial binding of a ligand to a protein produces a tendency to bind more ligand, positive cooperativity exists. If binding is restricted after the initial binding of a ligand, the protein shows negative cooperativity. If the Scatchard and Klotz plots are nonlinear, the Hill plot (Hill 1910) should be used to determine if cooperativity between binding sites exists. Therefore, a reason for variations in binding parameters between studies may be the choice of binding model.

In addition, a complete binding analysis from the initial signs of binding to saturation is often not realizable in flavor binding experiments, for example, because of the low water solubility of most flavor compounds. The range of flavor concentration varies from one study to another, making comparison of binding data impossible because the amount of flavor molecules present determines the occupation of binding sites. If flavor concentrations are low only the high-affinity binding sites are occupied, whereas high flavor concentrations may also result in the occupation of secondary binding sites.

Flavor Binding by Milk Proteins

β-Lactoglobulin (β-lg)

Of all the different food proteins, β-lg has been used extensively as a model protein for studying protein–flavor interactions, because of its well-defined structure and properties (McKenzie 1971; Kinsella and Whitehead 1989; Batt and others 1994).

The 3-dimensional structure of bovine β-lg has been determined by high-resolution crystallographic studies (Sawyer and others 1985; Papiz and others 1986; Monaco and others 1987) and has been shown to be similar to that of serum retinol-binding protein (RBP) (Papiz and others 1986; North 1989). β-lg is built up of 2 β-sheets, formed from 9 strands converging at 1 end to form a hydrophobic calyx or pocket, and a flanking 3-turn α-helix (Sawyer and others 1985; Papiz and others 1986) (see Figure 1 in Wu and others [1999]). This pocket serves as a binding locus for apolar molecules such as retinol (K= 5 × 107 M−1) (Fugate and Song 1980) and long-chain fatty acids (K= 105–107 M−1) (Spector and Fletcher 1970; Díaz de Villegas and others 1987; Pérez and others 1989; Frapin and others 1993; Wu and others 1999; Ragona and others 2000).

Fourier transform IR spectroscopy has been used to determine the conformational changes in β-lg upon the addition of ligands. The technique confirmed the binding of β-ionone, retinol, and fatty acids into the central cavity (Lübke and others 2000). For other ligands (p-cresol, eugenol, 2-nonanone, and γ-decalactone), no conformational changes of the protein were observed and the authors suggested either a binding to the protein surface or a binding into the central cavity without inducing a conformational change. However, an NMR study did confirm conformational changes of β-lg upon binding of flavor molecules (γ-decalactone and β-ionone) (Lübke and others 2002). It appeared that the side chains of several amino acids close to the central hydrophobic cavity (Leu46, Ile56, Met107, and Gln120) were affected by the binding of γ-decalactone, whereas the binding of β-ionone affected amino acids located in a groove near the outer surface of the protein (Leu104, Tyr120, and Asp129), a groove that is close to a region that has been described by Monaco and others (1987) and Lübke and others (2002). This study confirms the existence of 2 different binding sites on β-lg for aroma compounds, but it is still not entirely clear which flavors bind preferentially to which site.

Therefore, the most probable binding site for flavor compounds is the hydrophobic pocket of β-lg. In addition to this primary binding site, β-lg is thought to contain weaker secondary binding sites that are capable of undergoing interactions with hydrophobic molecules, such as flavor compounds (Spector and Fletcher 1970; Robillard and Wishnia 1972; Monaco and others 1987; O'Neill and Kinsella 1988; Dufour and Haertlé 1990).

β-lg is known to interact with several flavor compounds, such as alkanes (Wishnia and Pinder 1966; Mohammadzadeh-K. and others 1967, 1969a, 1969b), ketones (O'Neill and Kinsella 1987; Andriot and others 2000; Guichard and Langourieux 2000; Jouenne and Crouzet 2000b), aldehydes (van Ruth and Villeneuve 2002), ionones (Dufour and Haertlé 1990; Jouenne and Crouzet 2000a; Lübke and others 2002; Jung and Ebeler 2003b), lactones (Sostmann and Guichard 1998; Lübke and others 2002; Guth and Fritzler 2004), and esters (Pelletier and others 1998; Guichard and Langourieux 2000; Jouenne and Crouzet 2000b; Reiners and others 2000). As a result of different experimental methodologies and conditions, various binding parameters have been derived in different studies, making comparison of binding data difficult.

O'Neill and Kinsella (1987), using equilibrium dialysis, showed that β-lg has a high binding affinity and one main binding site for methyl ketones. The slopes of the Klotz plots for 2-heptanone, 2-octanone, and 2-nonanone indicated binding constants (K ) of 152, 481, and 2439 M−1, respectively. As the binding constant K for β-lg and homologous series of esters, aldehydes, ketones, and alcohols increased with increasing chain length (hydrophobicity) within the same chemical class, hydrophobic interactions between β-lg and flavor compounds were suggested (O'Neill and Kinsella 1987; Pelletier and others 1998; Guichard and Langourieux 2000). However, earlier work by Jasinski and Kilara (1985), using the same method, reported a much lower value of K= 122 M−1 for 2-nonanone and about 14 binding sites were evident. The authors suggested that unfolding of the protein may explain the high number of binding sites determined, but it is unclear how this unfolding may have occurred. In our opinion the flavor protein ratios used by Jasinski and Kilara (1985) are comparatively high so that weaker, secondary binding sites on the protein might be occupied as well, which would also decrease the overall binding constant. Besides, they added sodium azide as an antibacterial agent, which has later been shown to influence protein–flavor interactions (O'Keefe and others 1991b; Farès and others 1998; Reiners and others 2000).

The presence or absence of certain functional groups and steric factors is also known to have a marked influence on the degree of flavor binding by proteins. The interactions of an alkyl chain with β-lg can be reduced by a polar group; that is, the introduction of a hydroxyl function (OH) is responsible for weaker interactions (Reiners and others 2000). The introduction of an aldehyde group led to a slight increase in binding. In general, the binding capacities of proteins increase from alcohols to ketones and aldehydes.

A recent fluorometric study on the binding of retinol and γ-undecalactone to β-lg showed that there is competition between the ligands (Muresan and others 2001). The effect of the chain length on the free energy of interaction between β-lg and different lactones indicated that the interaction is mainly hydrophobic, which led the authors to conclude that the preferential binding site for the lactones is likely to be the central cavity. This hypothesis was confirmed by competition experiments between β-ionone and other flavor compounds (Sostmann and Guichard 1998). The authors concluded that (1) lactones have some affinity for the central cavity as well, and (2) α-ionone, β-damascenone, methyl benzoate, and unsaturated aliphatic aldehydes and ketones bind nonspecifically to the protein. In contradiction, Guth and Fritzler (2004) suggested a binding position of γ-decalactone, which is different from that of the central cavity since both retinol and palmitate only slightly inhibited the binding of the lactone. These contradictory results demonstrate that the location of binding sites for flavor molecules on proteins needs to be investigated further. Even flavor binding locations on the frequently studied β-lg are still not fully understood.

Dufour and Haertlé (1990), using fluorescence spectroscopy, suggested that the central hydrophobic pocket has a narrow specificity to the structure formed by the conjugated double bonds of the β-ionone ring and isoprenoid chain, present in both β-ionone and retinol. The authors could not demonstrate binding between β-lg and α-ionone, since it did not quench the tryptophan fluorescence. The ionone isomers differ only in the position of the cyclohexene double bond. On the other hand, α-ionone might be bound by β-lg, but without interacting with tryptophan. In contrast, Lübke and others (2000), using IR spectroscopy, showed that the binding of retinol and β-ionone into the hydrophobic cavity of β-lg induced no significant conformational change in the protein, whereas the binding of α-ionone did, probably due to the “wrong” position of the double bond (Lübke and others 2000). Jung and Ebeler (2003b), using a diffusion-based NOE pumping technique, demonstrated that the binding of β-ionone by β-lg was significantly higher than the affinity of α-ionone. In addition, α-ionone was bound only at pH 9, whereas β-ionone was bound at pH 3 to 11, with the greatest binding affinity at pH 9, and the lowest at pH 11, due to alkaline denaturation and aggregation. The reason why these authors, contrary to Lübke and others (2000), did not observe any binding of α-ionone at neutral pH may be the low sensitivity of the NOE pumping method. The complementary results of the above studies show the importance of using a combination of techniques to obtain more reliable results.

α-Lactalbumin (α-la)

α-la is a compact globular protein, stabilized by 4 intrachain disulfide bonds (Kinsella and Whitehead 1989), and it plays an important role in the synthesis of lactose (Wong 1988). However, only a few studies have investigated the binding of flavor compounds to α-la, as its affinity for flavor compounds is believed to be lower than that of β-lg. Using headspace analysis, α-la was found to bind various amounts of aldehydes and methyl ketones (Franzen and Kinsella 1974), but binding constants were not given. The study of Jasinski and Kilara (1985), which considered the binding of 2-nonanone and nonanal to α-la, showed very weak binding of both flavors to the protein, as determined by equilibrium dialysis. Because Jasinski and Kilara (1985) underestimated the binding of 2-nonanone to β-lg, they possibly underestimated the binding to α-la as well. Further examination of flavor binding by α-la should be performed.

Bovine serum albumin (BSA)

BSA binds a large variety of compounds, including retinol (Futterman and Heller 1972), long-chain fatty acids (Morrisett and others 1975; Spector 1975; Pérez and others 1989), alkanes (Wishnia and Pinder 1964; Mohammadzadeh-K. and others 1967, 1969a, 1969b), and aldehydes and ketones (Beyeler and Solms 1974; Franzen and Kinsella 1974; Damodaran and Kinsella 1980b; Jung and Ebeler 2003a). The protein is composed of a single polypeptide chain, which is folded so that 3 or 4 spherical units are formed. The binding sites for fatty acids are probably located in crevices between the spherical units (Spector 1975).

Using static headspace analysis, native BSA was found to decrease the vapor pressure of diacetyl over its aqueous solution. As little as 0.5% protein caused a 25% reduction in volatility (Land and Reynolds 1981), indicating a very high affinity of BSA to bind diacetyl. Beyeler and Solms (1974) used equilibrium dialysis to study the interactions of a large number of flavor compounds with BSA. The binding constants ranged from 0 to 10 × 103 M−1, and decreased in the sequence aldehydes > ketones > alcohols. The authors suggested that the binding of aroma compounds was due to both hydrophobic and electrostatic forces. King and Solms (1979) found the interaction between labeled (14C) benzyl alcohol and denatured BSA to be reversible. The interaction was independent of pH and ionic strength, confirming a dominance of hydrophobic interactions.

Damodaran and Kinsella (1980a, 1980b, 1981c) extensively studied the interactions between flavor compounds, in particular 2-nonanone and BSA. Using liquid–liquid partitioning, they determined a binding constant for 2-nonanone and BSA of K= 1800 M−1 with 6 primary binding sites on the protein molecule (Damodaran and Kinsella 1980a, 1980b). More recently, PFG-NMR spectroscopy revealed quite different binding parameters for the BSA/2-nonanone system with K= (833 ± 15) M−1 and n= 7 (Jung and others 2002). Differences in binding parameters for systems containing BSA are often due to the type of BSA used since BSA products contain varying amounts of fatty acids, which are tightly bound by the protein and thus reduce the binding of flavors. The above-mentioned values agree with the number of binding sites in BSA for long-chain n-alcohols (Steinhardt and Reynolds 1969) and free fatty acids (Spector and others 1969; Spector 1975). In addition, a large number of weaker, secondary binding sites are probably present (Spector and others 1969). This was confirmed by Guth and Fritzler (2004) who suggested 1 or 2 high-affinity binding sites, and a large number of lower affinity sites for γ- and δ-lactones on BSA.

A recent study using DSC suggested 2 and 3 binding sites and binding constants of 600 and 300 M−1 for vanillin (4-hydroxy-3-methoxybenzaldehyde) and 2-octanone for BSA, respectively (Burova and others 2003). We consider that the low binding parameters compared to other flavor ligands could be explained by the slightly acidic conditions (pH = 6.4) used by these authors, since lowering the pH by 1.8 units reduced the binding of γ-decalactone on BSA by 40% (Druaux and others 1995), probably due to conformational changes in the protein.

Using predominantly chromatographic methods, Dhont (1987) provided some evidence that albumin bound vanillin irreversibly to a substantial extent, but Dumont (1987) showed that the binding was reversible. It may be possible that the binding sites for vanillin on BSA are not equivalent; binding could occur both reversibly via noncovalent interactions, and irreversibly via the aldehyde function. Alaiz and Girón (1994) observed irreversible binding of 2-octenal to BSA. The authors suggested a covalent binding of 2-octenal, through its double bond, with the imidazole ring of histidine in BSA.

Jasinski and Kilara (1985), using equilibrium dialysis, compared the binding between BSA and 2 flavors, 2-nonanone and nonanal. For both flavors, they observed strong binding to the protein, with nonanal having a higher affinity than 2-nonanone. The authors attributed the stronger binding of nonanal to the position of the functional group, or the higher reactivity of aldehydes compared to ketones. A reduction in the available ɛ-amino groups of BSA on analysis of the nonanal–protein complex could not be found by Damodaran and Kinsella (1980b), indicating that the 1-position of the keto group caused the higher binding affinity. The 2-position of the keto group in 2-nonanone would give rise to more steric hindrance to hydrophobic interactions.

Whey protein products

A few studies have been reported on the binding of flavor compounds by whey protein products. Jasinski and Kilara (1985) investigated the binding of 2-nonanone and nonanal to whey protein concentrate (WPC). They found a large number of binding sites with strong binding affinity. Using HPLC, a weak interaction between whey protein isolate (WPI) and vanillin in a sweetened drink was demonstrated by McNeill and Schmidt (1993). Unfortunately, no binding parameters were determined. Li and others (2000) extended their research and found that the interaction between vanillin and WPI was strong, with an average binding constant of 1713 M−1 (12 °C) and 0.67 binding sites on WPI. Recently, a high binding affinity between 2-nonanone and WPI was confirmed using headspace SPME (Zhu 2003). A binding constant of K= 2059 M−1 (25 °C) and on average 1 binding site per protein molecule were found. The strong interactions between whey protein products and flavor compounds indicate that the addition of these proteins to food products, even at very low concentrations, could influence the flavor profile of the food.


A few studies have dealt with the behavior of aroma compounds in the presence of sodium caseinate or casein. Bovine sodium caseinate is a useful model for investigating the interactions between aroma and protein because of its well-known functional properties and its wide use in dairy, as well as nondairy, food products (Mulvihill 1992).

In the presence of sodium caseinate (10%), Reineccius and Coulter (1969) noted a decrease in the diacetyl headspace concentration of nearly 50%. In an aqueous solution containing only 1% casein, lower volatilities of acetone and acetaldehyde were observed (Maier 1970). Even the addition of only 0.1% sodium caseinate induced a decrease in the volatility of flavor compounds in aqueous solution in the following order: β-ionone > n-hexanol > ethyl hexanoate, isoamyl acetate (Voilley and others 1991). The intensity of the odor due to aldehydes, stored in a mixture with casein, decreased with increasing time of storage (Pokorný and others 1976). However, Le Thanh and others (1992) did not find a decrease in the headspace concentrations of acetone and ethyl acetate in the presence of 10% sodium caseinate. The authors attributed this to the residual NaCl present from the preparation of caseinate, because salts increase the concentration of volatile compounds in the headspace (Nawar 1966; Land and Reynolds 1981).

In a model dairy protein drink, sodium caseinate (6%) was shown to interact with vanillin. A significant decrease in free vanillin in the drink was shown by HPLC (McNeill and Schmidt 1993). Using similar methodology, moderate binding of vanillin to sodium caseinate was shown by Li and others (2000). Landy and others (1995) observed a major influence of sodium caseinate on the headspace–liquid partition coefficients and on the relative volatility of diacetyl, ethyl butanoate, and ethyl hexanoate, but not of ethyl acetate, in solutions containing 0.5% and 5% protein, respectively. The presence of strong interactions between diacetyl and sodium caseinate was suggested by the retention of diacetyl in the dialysis sac even after exhaustive dialysis against water (Farès and others 1998). Conversely, an exhaustive dialysis in the presence of benzaldehyde and sodium caseinate resulted in no volatile compound retained by the protein, revealing weak bonds between benzaldehyde and sodium caseinate (Farès and others 1998). In contrast, the aliphatic aldehyde hexanal has been found to bind covalently to sodium caseinate (Meynier and others 2004).

Dubois and others (1996) showed that, in a model cheese system, the volatility of diacetyl decreased slightly with increasing calcium caseinate content whereas a change in the fat content of up to 30% did not affect the volatility. The hydrophilicity of diacetyl was used as an explanation by these authors.

Fischer and Widder (1997) developed a method based on headspace GC–O to measure the retention of esters and heptanal in aqueous solutions with a casein content varying from 0 to 12%. Generally, the aroma retention increased with increasing protein content. A recent study by Zhu (2003) demonstrated weak binding of 2-nonanone by sodium caseinate, using headspace SPME. An average affinity constant of K= 1858 M−1 and on average 0.3 binding sites per protein molecule were found. The above studies clearly show that caseins and sodium caseinate are capable of binding several different flavor compounds. Therefore, when adding caseins to a food product, it has to be considered that flavor may be bound by the proteins and made unavailable for perception.

Hardly any information is available on the flavor binding behavior of the individual caseins αs1-, αs2-, β-, and κ-casein. Of all the caseins, β-casein is the most hydrophobic (Swaisgood 1992), and thus could have higher affinity constants for lipophilic flavor compounds. In addition, casein molecules have a tendency to undergo self-association or association with each other, depending on environmental conditions, such as protein concentration, pH, and ionic strength. It is unknown how this association behavior influences flavor binding. Recently, Burova and others (2003) estimated binding constants around 100 M−1 for 2-octanone or vanillin on β-casein. Systematic studies on the binding of selected flavor compounds by individual caseins and their mixtures under different conditions are required to fully understand this complex system.

Comparison of flavor binding capacities of milk proteins

The binding parameters of the most studied flavor compound, 2-nonanone, for milk proteins obtained by different authors vary significantly, as shown in Table 1. Researchers have used different experimental approaches to investigate protein–flavor binding, which may be the reason for some of the variation in the results (O'Keefe and others 1991b; Stevenson and others 1996). Nevertheless, there are obvious trends, such as decreasing affinity constants in the order BSA > β-lg > α-la.

Table 1—.  Binding data for the interactions between 2-nonanone and milk proteins (25 °C): n, number of binding sites per monomer; K, intrinsic binding constant
WPC611920000Equilibrium dialysisJasinski and Kilara (1985)
0.253000000Fluorescence spectroscopyLiu and others (2005b)
WPI12059Headspace SPMEZhu (2003)
Sodium caseinate0.31858Headspace SPMEZhu (2003)
β-Lg12439Equilibrium dialysisO'Neill and Kinsella (1987)
0.26250 (≤ 40 ppm)Static headspace analysisCharles and others (1996)
0.51667 (≥ 45 ppm) 
14122Equilibrium dialysisJasinski and Kilara (1985)
α-La3311Equilibrium dialysisJasinski and Kilara (1985)
BSA5–61800Liquid–liquid partitioningDamodaran and Kinsella (1980b)
1514100Equilibrium dialysisJasinski and Kilara (1985)
7833PFG-NMR spectroscopyJung and others (2002)

An early study that compared the binding of diacetyl by sodium caseinate and whey protein showed similar flavor binding for both proteins, as determined by static headspace analysis (Reineccius and Coulter 1969). However, this observation was disputed by several later studies, which showed that whey protein generally has a stronger flavor-binding capacity than casein (Hansen and Heinis 1991; Hansen and Booker 1996; Li and others 2000; Zhu 2003) (Figure 2). Under identical experimental conditions, WPI was found to have higher affinity for vanillin than sodium caseinate (Li and others 2000). These results agree with the findings of Hansen and Booker (1996), who examined the binding of vanillin, benzaldehyde, citral, and d-limonene to sodium caseinate and WPC. They reported that whey protein exhibited greater degrees of binding than casein of these flavor compounds. Recently, Zhu (2003), using headspace SPME, determined a higher average binding constant for 2-nonanone and WPI than for 2-nonanone and sodium caseinate (Table 1). In contrast, McNeill and Schmidt (1993) reported that sodium caseinate interacted significantly more than WPI with vanillin in sweetened drinks. The authors did not provide any explanation for this surprising result. Their finding highlights that a complex matrix, containing sucrose and emulsifier as well as protein, can influence flavor binding to proteins differently than a simple matrix. Therefore, it is crucial to explore basic systems first to be able to interpret results in multicomponent systems.

Figure 2—.

Vanillin flavor intensity relative to the reference in the presence of sodium caseinate (CAS) and whey protein concentrate (WPC). The reference vanillin concentration was 3.38 × 10−6 mM in a 2.5% sucrose solution. For each protein type, bars with dissimilar letter codes indicate significant differences between means. (Reproduced with permission from Hansen and Heinis (1991). Copyright 1991 American Dairy Science Assn.)

BSA has been shown to interact strongly with vanillin, whereas sodium caseinate and WPI showed similar and significantly lower binding of vanillin. Hydrogen bonding appeared to be a major force for the interaction of vanillin and sodium caseinate. However, hydrophobic interaction seemed to be more important than hydrogen bonding in the vanillin–BSA system (Chobpattana and others 2002). Binding of γ- and δ-lactones has been shown to be much stronger to BSA in comparison to β-lg (Guth and Fritzler 2004). From all of these studies, it becomes obvious that BSA is the milk protein that is most capable of flavor binding, followed by β-lg.

Influence of Heat Treatment on Protein–Flavor Binding

Flavor binding by proteins is very dependent on the conformational state of the proteins, and all the factors that alter the conformation, that is, temperature (Li and others 2000; Chobpattana and others 2002), pH (Druaux and others 1995; Jouenne and Crouzet 1996; Andriot and others 1999; Jouenne and Crouzet 2000a, 2000b), and ionic strength (Mohammadzadeh-K. and others 1969b; Damodaran and Kinsella 1981c). It has been demonstrated that conformational changes affect both the binding affinity and the number of binding sites on proteins for aroma compounds.

In practice, heat-denatured proteins are of greater importance than native proteins, because heat treatment is an important step during the processing and preparation of many foodstuffs containing protein (de Wit 1981). Heat treatments usually cause denaturation of proteins, which involves unfolding and subsequent aggregation of unfolded protein molecules. The process of unfolding may reveal binding sites that were previously buried and thus may result in an increase in binding. In contrast, the subsequent formation of protein aggregates may release flavor molecules from the binding sites again.

Below the denaturation temperature

The effects of temperature on protein–flavor interactions appear to be dependent on the type of protein and the type of flavor compound. The affinity of γ-decalactone for BSA in a model wine system was found to be higher at 10 °C than at 20 °C and 30 °C, possibly due to structural changes of the protein at low temperatures, whereas the number of binding sites (n= 6–7) was not modified at the 3 temperatures (Druaux and others 1995). However, an increase in temperature from 10 to 35 °C had little effect on the binding of 2-nonanone to BSA (Damodaran and Kinsella 1980b). This discrepancy could be explained by the presence of ethanol (10% w/w) and salts in the model wine system used by Druaux and others (1995), which may have had an effect on BSA conformation. In addition, γ-decalactone and 2-nonanone might bind on different sites on BSA due to their structural differences. There is not much information available about flavor binding sites on BSA. A large number of binding sites have been suggested by several authors (Spector and others 1969; Damodaran and Kinsella 1980a, 1980b; Jung and others 2002; Guth and Fritzler 2004), which makes the interpretation of results even more difficult. Some of the sites might be susceptible to temperature changes, whereas others might not be.

A decrease in temperature from 12 to 4 °C increased the number of binding sites and the binding constants for vanillin on casein and whey protein (Li and others 2000). The authors concluded that the changes in the number of binding sites and the binding affinity could be attributed to possible changes in the tertiary and quaternary structures of the protein at 4 °C. The exact changes to protein structures under these conditions need to be investigated.

Mills and Solms (1984), using static headspace analysis, observed an increase in the binding of heptanal to whey protein with an increase in temperature from 25 to 50 °C. The amount of heptanal irreversibly bound markedly increased at 50 °C. These results were attributed to covalent binding of the aldehyde and ɛ-amino groups of lysine residues in the proteins. An increase in temperature seems to enhance the binding of “reactive” flavors, whereas reversibly bound flavors may be released or not affected at all, depending on the nature of flavor compound. Clearly, further work is needed to fully understand the temperature dependence of these interactions, in particular in the case of reversible binding.

Above the denaturation temperature

Studies on the influence of heat-denaturation on protein–flavor binding are summarized in Table 2. A number of studies reported a decreased binding when proteins are denatured. On exposure of β-lg to 75 °C for 10 and 20 min, the binding affinity for 2-nonanone was weaker than that of the native protein and the number of low-affinity nonspecific binding sites increased (O'Neill and Kinsella 1988). Heat-treated WPI (85 °C, 10 min) had significantly higher vanillin flavor intensity than an untreated WPI (McNeill and Schmidt 1993). Free vanillin as determined by HPLC was also higher in a heated BSA system (68 °C for 30 min and 75 °C for 15 min) than in the nonheated system (Chobpattana and others 2002). In agreement with this, Burova and others (2003) reported the complete loss of vanillin binding after thermal denaturation of BSA. This may have been due to heat-induced structural changes and protein aggregation during the heat treatment. These authors also demonstrated an increase in the denaturation temperature of BSA in the presence of either vanillin or 2-octanone, which means that the flavor ligands increased the conformational stability of the protein (Burova and others 1999, 2003). Nonflavor ligands, such as palmitate, have also been found to stabilize the native structure of β-lg against heat-induced unfolding and denaturation. This stabilizing effect appears to be ligand-dependent; ligands that bind strongly into the hydrophobic calyx seem to be most effective (Considine and others 2005a).

Table 2—.  Influence of heat denaturation on the binding between milk proteins and flavor compounds; ↑ binding increases, ↓ binding decreases
Protein–Flavor Systemθ[°C]/t [min]BindingMethodReference
β-Lg/benzaldehyde70/30Ultra filtration-HPLCHansen and Booker (1996)
β-Lg/2-nonanone75/10 and 20Equilibrium dialysisO'Neill and Kinsella (1988)
WPI/vanillin85/10SensoryMcNeill and Schmidt (1993)
BSA/vanillin75/15Ultra filtration-HPLCChobpattana and others (2002)

On the other hand, when heated and unheated sodium caseinate solutions were compared, no differences were found for vanillin flavor intensity (McNeill and Schmidt 1993) and free vanillin concentration (Chobpattana and others 2002). This was expected as casein has a little secondary and tertiary structure and remarkably high heat stability (Fox and Mulvihill 1982).

However, for 1 milk protein–flavor system, it has been reported that the binding capacity of denatured proteins is generally higher than that of native proteins. Hansen and Booker (1996) reported that the amount of benzaldehyde bound by β-lg increased from 38% to 63% as the temperature was raised from room temperature to pasteurization temperature (70 °C, 30 min). The authors attribute the increase in binding to unfolding of whey proteins upon heating. The previously buried hydrophobic residues may become accessible for interaction with nonpolar flavor molecules, resulting in a greater amount of flavor compounds bound. Since these authors added the flavor before the heat treatment, it is likely that covalent binding may play a role in this case; that is, the aldehyde function of benzaldehyde could be susceptible to reaction with ɛ-amino groups in β-lg, particularly at the elevated temperature.

Most of the reported data on the influence of heat treatment on protein–flavor binding are based on single temperature and heating time. No systematic studies have been reported on the development of interactions with increasing heating time or increasing heating temperature. There might be an early stage with increased binding due to unfolding of the proteins, and a later stage characterized by a decrease in binding due to aggregation. Moreover, it has been established that a number of intermediate species are generated during heat-induced denaturation and aggregation of β-lg and whey proteins (Havea and others 2001). It would be interesting to understand the role of these intermediates in flavor binding.

Influence of High-Pressure Treatment on Protein–Flavor Binding

The use of high pressure as an alternative to heat treatment of foods, including milk and dairy products, is becoming of increasing interest. High-pressure treatment not only results in microbial inactivation, but has also been shown to improve rennet or acid coagulation of milk without detrimental effects on important quality characteristics, such as taste, flavor, vitamins, and nutrients (Trujillo 2002). It is known that high-pressure treatment changes the structure of milk proteins, and is thus likely to affect the protein–flavor interactions. In particular, β-lg has been shown to be very sensitive towards pressure-induced denaturation, and the process of pressure denaturation appears to be similar but not identical to that of heat denaturation (Huppertz and others 2004; Considine and others 2005b). α-La is more resistant to pressure than β-lg (Huppertz and others 2004; Liu and others 2005a). Yang and others (2003) observed that ligand binding by high-pressure-treated β-lg can be increased or decreased compared to the native protein depending on the structure of the ligand. High-pressure treatment of β-lg decreased the affinity for capsaicin on specific binding sites, whereas the binding of the other flavor compounds examined, namely α-ionone, β-ionone, cinnamaldehyde, and vanillin, was unspecific and remained unaffected by high-pressure treatment. The authors explain this finding with the incorporation of water into the protein upon high-pressure treatment, as compared to the transfer of nonpolar groups into water upon heat denaturation of a protein (Hummer and others 1998), so that high-pressure-treated β-lg may not exhibit an increase in surface hydrophobicity, and thus not exhibit an increasing affinity for hydrophobic flavor compounds. However, this would mean that heat-treated β-lg would bind more flavor compared to the native protein, but this is not true in most cases (Table 2).

New insight into the effects of high hydrostatic pressure (HHP) treatment on protein–flavor binding was given by Liu and others (2005a), using fluorescence spectroscopy and static headspace analysis. The number of binding sites and the apparent dissociation constants for benzaldehyde and methyl ketones on WPC were either unaffected or increased upon HHP treatment, depending on the structure of the flavor compound, the flavor concentration, and the HHP treatment times. Using SDS-PAGE, the same authors (Liu and others 2005b) showed that during the time to reach the target pressure (600 MPa), dissociation of protein aggregates present in WPC occurred, which may have exposed more binding sites. The presence of aggregates may have resulted from the ultra filtration and drying procedures used in WPC manufacture. Liu and others (2005a) pointed out the importance of careful selection of flavor concentrations and HHP treatment conditions for desired outcomes in food applications.

Thorough research on the effect of high-pressure treatment on protein–flavor binding is clearly needed. The studies by Yang and others (2003) and Liu and others (2005a) are the only ones dealing with the influence of high-pressure treatment on protein–flavor interactions. Nonflavor ligands, when added prior to high-pressure treatment, have been found to inhibit the formation of intermediate, non-native protein species (Considine and others 2005b). The same effect may exist with flavor ligands but has not been studied yet. This area clearly warrants further investigations.

Implications in the Development of Protein Foods

Proteins do not contribute to flavor directly, but protein–flavor interactions can cause the aroma profile of a food to become unbalanced and unpleasant since proteins bind flavors to differing extents, depending mainly on the nature of the protein and flavor compound. This problem occurs particularly in high protein-based foods and in low-fat foods because fat is the preferred carrier of flavor compounds. Ideally, flavor molecules are preserved in the food during storage and processing, and slowly released during consumption of the food. However, in the presence of protein and absence of fat, some flavor compounds are lost due to the absence of fat, whereas others are tightly bound by the proteins, preventing them to be released and perceived during mastication. In such foods it is extremely difficult to control flavor. If proteins are present, the amount of flavor compounds added usually has to be increased to compensate for the protein-bound flavor.

The processing of foods, and in particular heat treatment, has been shown to have a marked effect on milk protein–flavor interactions. Even slight changes in temperature below the denaturation temperature of the proteins have been shown to influence the binding of flavors (Mills and Solms 1984; Druaux and others 1995). Severe heat treatments caused a decrease in binding for most flavors (O'Neill and Kinsella 1988; McNeill and Schmidt 1993; Chobpattana and others 2002; Burova and others 2003), but an increase in binding has also been reported (Mills and Solms 1984; Hansen and Booker 1996). Thus heat treatment may drastically change the flavor profile of a food product. In the development of protein-based foods, factors changing the conformation of a protein, that is, pH, ionic strength, so on, have to be considered because they potentially alter the binding of flavors to the protein. The presence of flavor compounds during heat treatment has been shown to increase the protein conformational stability (Burova and others 1999, 2003), which in turn may alter the characteristics of the end product.

Concluding Remarks

The presence of proteins in flavored low-fat food products causes a great challenge for flavor scientists because many proteins are able to bind several flavor compounds tightly and influence the perceived aroma profile significantly. BSA is the milk protein most capable of binding volatile flavor compounds, followed by β-lg. A large number of instrumental methods have been used to investigate the type and the extent of flavor binding, which makes it difficult to compare results between studies. Sensory methods are very useful because they complement instrumental techniques and give insight into the effect of protein–flavor binding on flavor perception. The characteristics of the aroma compounds and the proteins determine the extent of binding, which can be influenced by several parameters. Although extensive information on the binding of various flavor compounds to different food proteins is available, there are many apparent contradictions and disagreements among various studies. For example, the effects of heat treatments on protein–flavor interactions are not fully understood. The development of protein–flavor interactions with heating time or rather with increasing temperature should be looked at and compared with the corresponding conformational state of the protein. Research on the influence of high-pressure treatment of milk proteins on the proteins' flavor binding behavior needs to be investigated further as well.

A good knowledge of the physicochemical interactions that occur between aroma compounds and proteins is required to improve food flavoring and to make protein-based foods, for example, “light” dairy products, sensorily more acceptable to the consumer. In particular, the nature and the location of binding sites on proteins for flavors need to be investigated further. It is vital to obtain more consistent results between different instrumental methods. Some of the early methods such as equilibrium dialysis and exponential dilution are still going to be frequently used. We believe some of the newer techniques, such as SPME and NMR, need to be developed further to investigate protein–flavor interactions and be used more frequently in the future. SPME is fast, solvent-free, and very sensitive. The main advantages of NMR techniques are speed and insight into binding mechanisms and binding topology.

Furthermore, the focus should be more on sensory techniques, because instrumental flavor-binding studies do not show if and how bound flavor is perceived during consumption. A growing area of research comprises the mechanisms of in vitro and in vivo flavor release from foods. A detailed discussion of in vivo flavor release from protein containing systems is beyond the scope of this review but some information is given below. Model mouth systems have been frequently used to measure flavor release in vitro (Roberts and Acree 1995; Deibler and others 2001; Rabe and others 2002). Methods for measuring in vivo volatile release from foods have been summarized previously (Taylor and Linforth 2000). There is a variety of objective and subjective in vivo techniques to investigate flavor release in-nose or in-mouth, including concentrating the exhaled volatiles followed by GC-MS (Denker and others 2006), atmospheric pressure chemical ionization (APCI)-MS (MS-Nose) (Taylor and others 2000; Weel and others 2002; Lethuaut and others 2004), and proton-transfer reaction (PTR)-MS (Lindinger and others 1998; Mestres and others 2005; Boland and others 2006) to analyze breath volatiles during eating in real time, and sensory techniques such as the time-intensity (TI) approach (Weel and others 2002; Lethuaut and others 2004; Mestres and others 2005). No systematic studies have been reported on the relationship between extent or strength of flavor binding to proteins and its in vivo release. Interactions of WPI and aldehydes in solution were shown to be less significant in-nose than under static headspace conditions, possibly due to the highly dynamic conditions in vivo (Weel and others 2003). Milk proteins, in particular whey proteins, reduced the in-mouth release of selected flavor compounds from coffee (Denker and others 2006).

To date, most systems that have been investigated consist of 1 protein and 1 aroma compound in an aqueous solution. Food systems are much more complex, consisting of several food matrix components and flavor mixtures. Thus, further research on these complex systems is becoming increasingly important. However, primarily consistent results using simple systems should be obtained before investigating complex systems.