Review: Plant protein flavor chemistry fundamentals and techniques to mitigate undesirable flavors

Serap Vatansever, Bingcan Chen, Clifford Hall

PDF version https://aocs.onlinelibrary.wiley.com/doi/epdf/10.1002/sfp2.1025

Abstract

Many plant protein ingredients have unpleasant flavor (e.g., green, beany, bitterness) characteristics. Isolation of the protein fraction will reduce the intensity of the undesirable flavors but in most cases the flavor issue is not eliminated. The objective of this review is to provide information about undesirable flavors associated with plant proteins and approaches to mitigate these flavors. The focus of the review will be on flavor associated with plant proteins from pulses, soybean, and oilseeds (e.g., hempseed, flaxseed). While other plant proteins exist, they will not be covered extensively in this limited review. The undesirable flavors are defined as volatile organic compounds (VOCs) and non-volatile organic compounds (non-VOCs). Example VOCs include aldehydes, ketones, and isobutyl pyrazines whereas saponins, bitter peptides, and phenolics are among the most common non-VOCs. The raw material used as the source for the protein dictates the undesirable flavors remaining in the protein ingredient. The intensity of the binding of the VOCs and non-VOCs to protein serves as the basis for the detection and difficulty in removing these compounds from the protein. A discussion about flavor binding and processing methods to mitigate unwanted flavors in protein ingredients will be provided. Methods presented will focus on extraction and thermal treatment. Although flavor masking is an approach to minimize undesirable, this topic will not be covered. Understanding the compounds responsible for the flavor issues and methods to minimize their impacts on sensory characteristic of protein ingredients can provide manufacturers with solutions to addressing flavor issues inherent to plant proteins.

INTRODUCTION

There continues to be significant discussion around food and protein needs by 2050 (Sijpestijn et al., 2022). The expectation is that plant proteins will contribute to the increased protein demand. Plant-based sources of proteins have a disadvantage over animal proteins in that the compounds that co-extract with isolated plant proteins are complex and have definable negative sensory characteristics. The undesirable flavors and approaches to mitigate compounds that contribute flavors will be the focus of this review.

Flavors can be defined as volatile organic compounds (VOCs) and non-volatile organic compounds (non-VOCs). VOCs are compounds that elicit a response through a nasal or oral nasal mechanism. These compounds are associated with the aroma aspect of flavors and bind to receptors as the odorant passes the cilia (location of the receptors) in the nasal cavity. In contrast, non-VOCs elicit a response through binding taste receptors on the tongue or interacting with nerves in the oral cavity. These compounds are responsible for the sense of gustation or taste (e.g., bitter) and trigeminal (e.g., spicy) responses. Regardless of the compound, the intensity of the response will depend on the release of the compound of interest from the matrix. In the context of this review, the interaction of the chemical compound (VOCs or non-VOCs) with protein will dictate the perception of the flavor in terms of intensity and specific attribute (e.g., aroma, bitterness) recognized.

During the protein isolation, some VOCs and non-VOCs will co-extract with proteins. The level of compounds remaining, and the intensity of the interaction will dictate how easily the chemical compounds are removed from the protein ingredient. Likewise, the intensity of the perception of the off flavor can be impacted. Thus, understanding the flavor characteristics of the raw material is necessary to minimize potential off flavors in the protein ingredients. An effective approach to mitigate the off flavors can be developed once the mechanism(s) driving the interaction between the protein and flavor is understood. The remaining part of this review will focus on the type and source of off-flavors, compounds that contribute the undesirable flavor, protein-flavor interaction mechanisms and methods to mitigate the off-flavors.

OFF FLAVORS IN RAW MATERIALS

The interest in plant-based protein has significantly diversified the raw materials used for protein isolation. While soybean is still the major plant base protein source, significant interest in pulses (e.g., pea and lentil), canola, flaxseed, hempseed, and sunflower has led to commercial protein products. The raw materials generally are in the form of a meal that is obtained from a de-fatting or similar process. Generally, the number of volatiles or the intensity of the volatiles decreases with protein purification. However, understanding the raw material and associated flavor compounds is still important because methods to mitigate flavor issues can target those compounds that are associated with undesirable sensory characteristics.

For legumes such as pea and soybeans, VOCs include alcohols, aldehydes, ketones, furans, and alkyl methoxypyrazines (Roland et al., 2017). These VOCs are responsible for the distinct pea, green, grassy, and beany flavor characteristics of legumes (Azarni, Boye, Warkentin, Malcolmson, Sabik, & Bellido, 2011; Murat et al., 2013; Wang et al., 2021). However, enzymatic (hydrolytic and oxidative process) and non-enzymatic (autoxidative) reactions also cause the formation of off-aroma compounds (Azarni, Boye, Warkentin, Malcolmson, Sabik, & Bellido, 2011; Murray et al., 1976; Sessa & Rackis, 1977). Based on available VOC data of different pulses, the same VOCs would be expected regardless of the legume used as a raw material. Guo et al. (2019) reported 73 volatiles were present in raw sunflower with the main volatile compounds being α-pinene, hexanal, furfural, octane, γ-butyrolactone. In flaxseed, the common reported odor active compounds include benzaldehyde, hexanal, hexan-1-ol, 2-methylbutanol, nonanal, 3,5-octadien-2-one, pentane-2,3-dione, pentan-1-ol, 1-penten-3-ol (Hou et al., 2021; Yang et al., 2021). Song et al. (2022) observed 110 volatiles from hempseed; however, only 35 volatiles were considered as odor active. The main odor active compounds included dimethyl trisulfide, eucalyptol, ethyl hexanoate, 3-methylbutanal, and trans-2-nonenal. Unlike other oilseeds, the volatiles in whole canola seeds have not been extensively evaluated. Rusinek et al. (2019) reported 24 volatiles in whole canola seeds. These authors found that 4-H-1-benzopyran-4-one was predominant in a stored sample and determined that the volatile was of microbial origin. In contrast, nonanal was the predominant compound among the 23 VOCs identified in rapeseed meal (Hao et al., 2020).

While the information above did not cover volatiles associated with raw materials extensively, there is amble evidence that many of the volatiles in raw materials are present in protein ingredients (Benavides-Paz et al., 2022a, 2022b; Murat et al., 2012, 2013; Xu et al., 2020; Zhogoleva et al., 2023). Furthermore, assessing flavor compounds in the protein ingredient is essential since newly created VOCs, such as (Z)-2-penten-1-ol, 3-octanol, 2-ethyl-1-hexanol, undecanal and heptadecanal-(E)-2-octenal, can occur during the isolation process (Murat et al., 2013; Xu et al., 2020). Liu et al. (2023) recently reported the diverse nature of the aroma compounds of 24 different commercial pea protein products. Variability in the volatiles and sensory properties among protein products (Liu et al., 2023) likely contributes to the failure of some protein products in product development applications. To a lesser extent, non-VOCs such as saponins and phenolic compounds are lower in protein isolates compared to the raw material source.

Like the VOCs, diversity exists in types and amounts of the non-VOCS in the raw materials. Furthermore, the non-VOCs tend to be present in hulls of seeds and thus any raw material, such as a whole seed or meal, that contains hull material will likely be the source of non-VOCS in protein ingredients. For example, Frolov et al. (2013) identified 66 bound phenolics from canola hull cell walls while Reim and Rohn (2015) reported that on a per weight basis, pea hulls had approximately 1.5–2 times more saponins than the cotyledons. Lin et al. (2006) reported that 41% of the seed saponins were retained in soy protein isolate; thus, removal of the hull prior to protein isolation likely will result in lower saponin and other non-VOCs in the protein isolate.

Glucosinolates have been documented as bitter compounds in plants from the Brassicacea family, specifically Brassica species consumed as vegetables (Drewnowski & Gomez-Carneros, 2000; Groenbaek et al., 2019; Nor et al., 2020). Therefore, the glucosinolates identified in cold pressed rapeseed protein isolate (Miklavčič Višnjevec et al., 2021) likely contribute to the bitterness of rapeseed and canola protein isolate. The identified compounds confirmed that glucosinolates coextracted with protein, as supported by their presence in whole seed (Millán et al., 2009; Sosulski & Dabrowski, 1984; Velasco et al., 2008). However, these compounds tend to degrade or are removed during processing, resulting in lower concentrations in the protein isolates compared to rapeseed and canola meals (Miklavčič Višnjevec et al., 2021).

COMPOUNDS CONTRIBUTING OFF FLAVORS TO PROTEIN INGREDIENT

Volatile organic compounds

The VOCs play a crucial role for overall acceptability of food products because they contribute both acceptable and unacceptable flavors. Analytical tools such as gas chromatography–mass spectrometry (GC–MS) coupled with olfactory detection (GC–MS-O) have advanced our understanding of VOCs in protein ingredients. The GC–MS-O approach allows descriptors (Table 1) to be assigned to volatiles that separate during the GC run along with mass identification. The technique further allows for determining relevancy of the separated VOCs because not all the isolated volatiles have an odor or are at too low a concentration that the panelists cannot identify the compound. For example, 5-methyl-3-hexanol is detectable via MS but is not detected by olfaction (Fang, Chang, et al., 2023; Fang, Gu, et al., 2023) because this compound is not aroma active, that is, does not produce a recognizable odor. In contrast, a compound such as pyrrole produces a nutty aroma but is in sufficiently low concentration that it may not be detectable in pea protein by olfaction (Xu et al., 2020). Furthermore, assigning the odor detected as pleasant or unpleasant can help to develop mitigation processes, where the unpleasant odor compounds can be targeted for removal. Lan et al. (2020) identified 25 volatiles in flaxseed protein and reported that nine were considered unpleasant. Examples included skunky and rancid that was associated with 1-heptanol, beany and earthy ((E)-3-octen-2-one) and sweaty (3-methyl-butanoic acid). Regardless of the raw material, the presence of individual VOCs give rise to similar descriptors (Table 1) when olfactory detection is used; thus, determining the VOCs through GC–MS can be an effective approach to identifying the aroma active compounds (Liu et al., 2023).

TABLE 1. Descriptors for volatiles identified in flaxseed, hempseed, lentil, pea, and soybeans.

• Source: Jakobsen et al. (1998), Wu et al. (2011), Murat et al. (2012, 2013), Tripathi et al. (2014), Lasekan and Yap (2018), Xu et al. (2019, 2020), Lan et al. (2020), Shen, Gao, Xu, Ohm, et al. (2020), Shen, Gao, Xu, Rao, et al. (2020), Shen et al. (2021), Fang, Chang, et al. (2023), Fang, Gu, et al. (2023), Guldiken et al. (2021), Liu et al. (2023); http://www.thegoodscentscompany.com/.

While GC–MS-O is an effective tool to identify the aroma active volatiles (Viana & English, 2021), an individual volatile compound typically does not define the aroma characteristic of a material. For example, pea off flavor compounds include alcohols, aldehydes, ketones, furans, and alkyl methoxypyrazines (Heng, 2005; Roland et al., 2017). These VOCs are responsible for a distinct pea aroma, including green, grassy, beany, mushroom, earthy, and other aromas (Azarni, Boye, Warkentin, Malcolmson, Sabik, & Bellido, 2011; Murat et al., 2013). The volatile compounds of interest are inherent (i.e., biosynthesized by the plant) to the raw material or develop during handling, processing, and storage (Azarni, Boye, Warkentin, Malcolmson, Sabik, & Bellido, 2011; Murray et al., 1976; Sessa & Rackis, 1977). Many of the VOCs are the result of enzymatic and non-enzymatic oxidation processes. A significant number of VOCs found among protein ingredients from different plant materials likely have origins grounded in oxidation (Table 2).

TABLE 2. Volatile organic compounds identified in protein ingredients of flaxseed, hempseed, lentils, pea, and soy.

• Source: F = Flaxseed (Lan et al., 2020); H = Hempseed (Fang, Chang, et al., 2023; Fang, Gu, et al., 2023; Shen, Gao, Xu, Ohm, et al., 2020; Shen, Gao, Xu, Rao, et al., et al., 2020; Shen et al., 2021); L = Lentil (Guldiken et al., 2021); P = Pea (Liu et al., 2023; Murat et al., 2012, 2013; Xu et al., 2019, 2020); S = Soybean (Chen, 2015; Wu et al., 2011).

Pea (i.e., green and earthy) flavor is attributed to a family of compounds called alkyl methoxypyrazines. These compounds are believed to be produced from amino acids in the plant. Alkyl methoxypyrazines attribute an intensive unpleasant green pea perception at very low odor threshold in dry pea (Heng, 2005; Jakobsen et al., 1998; Murray et al., 1976; Vatansever & Hall, 2020). Jakobsen et al. (1998) identified three predominant alkyl methoxypyrazines, which were 2-isopropyl-3-methoxypyrazine (pea aroma), 2-sec-butyl-3-methoxypyranize (bell pepper aroma), and 2-isobutyl-3-methoxypyrazine (green, peapod aroma) in blanched pea.

Lipoxygenase (LOX)-promoted oxidation of lipids contributes some of the VOCs in protein ingredients. Forster et al. (1999) reported a lower concentration of hydroperoxides were formed in a pea seed free of the enzyme LOX-2. Thus, the contribution of LOX in the generation of VOCs starts with hydroperoxide formation. Mechanistically, LOX is a group of non-heme metal-containing dioxygenases that catalyzes the insertion of molecular oxygen into cis, cis,-1,4-pentadiene units of polyunsaturated fatty acids (Hayward et al., 2017; Wu et al., 1995). This gives rise to conjugated unsaturated fatty acid hydroperoxides which ultimately decompose to VOCs. Hexanal, 1-hexanol and 3-hexenal are well documented VOCs arising from LOX promoted oxidation in many plant materials (Table 2). These compounds contribute the cut grass, hay-like and beany odor characteristic in pulses (Guldiken et al., 2021; Heng, 2005; Liu et al., 2023; Murat et al., 2012, 2013; Wang et al., 2020; Xu et al., 2019, 2020), soybean (Chen, 2015; Wu et al., 2011; Zhang et al., 2022), and oilseed crops such as flaxseed and hempseed (Fang, Chang, et al., 2023; Fang, Gu, et al., 2023; Lan et al., 2020; Shen, Gao, Xu, Ohm, et al., 2020; Shen, Gao, Xu, Rao, et al., et al., 2020; Shen et al., 2021). Among the VOCs, alcohols aside from 1-hexanol generated by fatty acid breakdown include 1-pentenol (pungent), 1-octen-3-ol (mushroom), 1-octanol (green and mushroom), and 1-nonanol (green and citrus) among different protein isolates (Tables 1 and 2). Furthermore, the compound (E,E)-3,5-octadien-2-one also was identified as a hay-like odorant in a pea protein isolate (Murat et al., 2013).

Autoxidation is non-enzymatic lipid oxidation pathway and is likely the predominant mechanism by which VOCs form during the storage of protein ingredients. The amount of lipid remaining in the protein after isolation, availability of oxygen and storage temperature are factors that contribute to the formation of primary oxidation products (i.e., hydroperoxides) in protein ingredients. Subsequent decomposition of the hydroperoxides promoted by transition metals or high storage temperature (McClements & Decker, 2018) led to the formation of VOCs similar to LOX-promoted oxidation.

Non-volatile organic compounds

Bitterness and astringency characteristics of protein ingredients are generally attributed to non-VOCs (Roland et al., 2017). While there is a significant level of non-VOCs in raw materials (Karolkowski et al., 2023), not all are present in protein ingredients. The most common non-VOCs include saponins, polyphenols, bitter peptides, and hydroxy fatty acids (Figure 1). These compounds are inherent to the raw material, and during protein extraction processes, tend to bind to protein (Lin et al., 2006; Noguera et al., 2022; Potter et al., 1993; Shao et al., 2009). As a result, the bitter compounds present in the protein ingredient will be dependent on the raw material used for protein isolation.

Saponins are glycosides of sapogenins (aglycone form). Saponins associated with protein ingredients are typically triterpenoid glycosides. Much of our understanding of saponins in protein ingredients is based on soy protein research (Fang et al., 2004; Hu et al., 2002; Lin et al., 2006; Rickert et al., 2004; Yoshiki et al., 1998). The amphiphilic (i.e., having both hydrophilic and hydrophobic moieties) nature of saponins allows them to be extracted with protein. For example, Lin et al. (2006) observed that 41% of the saponins in soybean were retained in soy protein isolate. While limited information exists regarding the saponin content of protein sources other than soy, saponin content has been reported among different types of seeds and plant materials (Table 3). Thus, the presence of saponins in the raw material will mean that some form of saponins will be present in the protein ingredient.

TABLE 3. Saponins present in protein ingredients or seeds.

• * Found in hypocotyl of soybean seed and not in protein ingredient.

Heng, Vincken, Hoppe, van Koningsveld, Legger, et al. (2006) and Heng, Vincken, Hoppe, van Koningsveld, Decroos, et al. (2006) reported pea saponins as two groups, saponin Bb (also referred to as Soyasaponin I) and saponin ßg, also known as DDMP (2,3-dihydro-2,5-dihydroxy-6-methyl-4H-pyran-4-one) saponin or soyasaponin βg, in 16 pea cultivars. These authors considered the DDMP saponin (ßg) as the natural form present in the pea while saponin Bb was a decomposition product. Daveby et al. (1997) observed only saponin Bb to be present in peas but acknowledge that the preparation of the extracts may have contributed to the decomposition of the ßg saponin. Other researchers have supported changes in the saponins due to sample preparation or through processing. For example, the DDMP saponins αg, βg, and βa were converted to soyasaponin Ba, Bb, and Bb', respectively, in soy protein isolate (Rickert et al., 2004) while DDMP saponin βg was converted to soyasaponin Bb during the isolation of pea protein (Noguera et al., 2022) and in chickpeas and lentils subjected to various processing methods (Ruiz et al., 1996, 1997). Hu et al. (2002) observed the presence of DDMP saponins βg, and βa and soyasaponin Bb and Bc in texturized soy protein while soy flour only contained the DDMP saponins. While DDMP- and B-type saponins predominate in pulses such as pea, lentil and chickpea, E-type (saponin Be) has been reported in chickpea and soy proteins (Fang et al., 2004; Kerem et al., 2005; Serventi et al., 2013). Among saponin types, DDMP saponin has higher bitterness intensity than saponin B (Heng, Vincken, Hoppe, van Koningsveld, Legger, et al., 2006; Heng, Vincken, Hoppe, van Koningsveld, Decroos, et al., 2006). However, both saponins still exhibited bitterness and thus their presence in protein ingredients can be problematic. Besides bitterness, saponins may be perceived as astringent and having a metallic flavor (Roland et al., 2017). Variability in saponin concentration will dictate the bitterness intensity. While saponins may be present in protein ingredients, the concentration will dictate the impact on taste perception. Gläser et al. (2020) reported that soyasaponin I concentration (1 mmol/kg) did not exceed the taste threshold (i.e., dose-over threshold (DoT)-factor) to produce bitterness in a pea protein isolate. However, they did find that the DoT for astringency was exceeded and thus at the concentration measured (1 mmol/kg) was sufficient to impart astringency but not bitterness. Ongkowijoyo et al. (2023) also reported soyasaponin I concentration in a 10% solution of a commercial pea protein isolate was below the bitterness threshold and suggested that the bitterness observed was not due to this saponin. However, the conversion of DDMP saponins to soyasaponin I likely occurred during the production of the commercial pea protein isolates and resulted in a less bitter product due to the lower bitterness intensity of soyasaponin I compared to DDMP saponins (Heng, Vincken, Hoppe, van Koningsveld, Legger, et al., 2006; Heng, Vincken, Hoppe, van Koningsveld, Decroos, et al., 2006).

Phenolic compounds have long been associated with bitterness of protein ingredients (Sosulski, 1979). While many researchers have reported the presence of phenolic compounds in flours, less information is available on the presence of phenolic compounds in protein ingredients (Table 4). Phenolic compounds isolated from soy protein isolate include phenolic acids and isoflavones (Alu'datt et al., 2013; Fang et al., 2004; Shao et al., 2009). Phenolic acids and condensed tannins were identified in canola/rapeseed protein isolates (Thiyam et al., 2009; Xu & Diosady, 2002). Thiyam et al. (2009) noted the loss of syringaldehyde and canolol in highly purified protein isolates of rapeseed; furthermore, sinapine was still retained in the protein isolate. In contrast, kaempferol-3-O-(2′″-O-sinapoyl-β-sophoroside) was recently identified as the bitter compound in rapeseed/canola protein isolate (Hald et al., 2019; Siebert et al., 2022). Chlorogenic acid and cryptochlorogenic acid were found in sunflower protein isolate (Saricaoglu et al., 2023) while nine phenolic acids were identified in flaxseed protein isolate (Alu'datt et al., 2013). Cosson, Meudec, et al. (2022) identified six phenolic acids and 10 flavonoids in pea protein isolate. Nine additional phenolics acids were also tentatively characterized. Kaempferol glucosides accounted for most of the flavonoids in the pea protein isolate. Correlation analysis of bitterness and astringency scores with the individual compounds indicated that caffeic acid and quercetin-3-O-glucoside were positively correlated to bitterness and astringency, respectively (Cosson, Meudec, et al., 2022). While oat protein is not readily available as a protein isolate, several researchers have identified phenolic acids as the source of the astringency in oat flour (Günther-Jordanland et al., 2016, 2020), which might be co-extract with proteins and thus contribute undesirable flavor to the protein ingredients or products like oat milk. Other compounds such as bitter peptides, cyclolinopeptides, hydroxy and oxidized fatty acids, and glucosinolates have been identified or theorized as compounds contributing astringency and bitterness of protein isolates. However, limited research on these compounds in protein ingredients exist.

TABLE 4. Phenolic compounds that are present in protein ingredients or seeds.

While numerous studies have been completed on bitterness of enzyme hydrolyzed proteins, limited research has been conducted to characterize bitter peptides in protein isolates. Taylor et al. (2004) identified a PA1b peptide variant in isolated pea protein that had insecticidal activity. More recently, Ongkowijoyo et al. (2023) supported the presences of the bitter peptide PA1b in a commercial pea protein isolate. This bitter peptide was 37 amino acids in length. Cosson, Oliveira, et al. (2022) used mass spectrometry coupled with the BIOPEP database to determine that 14 of the 106 peptides, in pea protein, associated with bitterness had eight amino acid residues. Cyclolinopeptides are peptides of eight to nine amino acids in length that exist is a circular arrangement (Brühl et al., 2007; Lang et al., 2022). These peptides are primarily associated with linseed (i.e., flaxseed) and were identified as contributors to bitterness of flaxseed oil. While these peptides are predominantly in the oil fraction, protein meals can be bitter partly due to the cyclolinopeptides (Aladedunye et al., 2013). Furthermore, the bitterness of these compounds increases upon oxidation of methionine-to-methionine sulfone (Aladedunye et al., 2013; Brühl et al., 2007; Jadhav et al., 2013; Morita et al., 1999). In general, there are limited studies that have focused on bitter peptides in protein isolates. Thus, additional research in this area is needed to determine how widespread these compounds are in plant protein ingredients.

Baur et al. (1977) reported hydroxy fatty acids 9,12,13-trihydroxyoctadec-10- and 9,10,13-trihydroxyoctadec-11-enoic acids, which resulted from linoleic acids treatment with soy protein, had significant bitterness. Similar hydroxy fatty acids, as well as monoglycerides, were identified as bitter components of oat lipid (Biermann & Grosch, 1979). Fourteen lipids and lipid oxidation products have been identified as contributing to the bitterness in pea protein isolates (Gläser et al., 2020, 2021). These authors reported a series of hydroxy fatty acids that contributed to the bitterness. However, the 9,10,11-trihydroxyoctadec-12-enoic and 11,12,13-trihydroxyoctadec-9-enoic acids were determined to be the hydroxy fatty acids that contributed most to the bitterness. Ongkowijoyo et al. (2023) also reported the presence of hydroxy fatty acids in a commercial pea protein isolate but found that the concentrations were lower than the threshold values for bitterness. Furthermore, 1-linoleoyl glycerol had significant bitterness (Gläser et al., 2020, 2021). These reports support the bitterness of hydroxy fatty acids. Like bitter peptides, few researchers have attempted to characterize the hydroxy fatty acids and thus additional research on this topic is warranted.

MECHANISMS OF FLAVOR/OFF-FLAVOR BINDING

In general, the bonding between protein and flavor compounds can be categorized into physical trappings and molecular interactions, both of which are dependent on several factors. However, molecular interactions such as hydrogen boning, hydrophobic interactions, Van der Waals forces, ionic interactions, and covalent bond formation are the key to understanding potential flavor issues in protein ingredients (Figures 2 and 3). Furthermore, the interactions driving flavor binding to protein can be impacted by environmental moisture (Zhou & Cadwallader, 2006).

Regardless of the flavor compound, interactions with carboxyl (COOH), sulfhydryl (SH), amine (NH₂), and hydroxyl (OH) groups of protein are reversible (Reineccius, 2005). In contrast, disulfide (SS), SH, and NH₂ of proteins can undergo irreversible binding through covalent bond formation with the flavor compound (Reineccius, 2005; Wang & Arntfield, 2014, 2015a, 2015b, 2016a; Zhang et al., 2022). Pulse proteins for example have high lysine (NH₂ source) content and thus have the potential to bind flavors through both reversible and irreversible mechanisms. However, any protein can bind unwanted flavor compounds (i.e., volatile or non-volatile) leading to proteins with undesirable flavor.

The presence of beany, grassy off-flavors for example relates to the binding of the volatile compounds to protein (Wang & Arntfield, 2016a; Zhang et al., 2021; Zhang et al., 2023). Aldehydes tend to bind pea protein over a broader pH range than ketones (Heng et al., 2004; Nguyen et al., 2014). Regardless of pH, vicilin protein tends to bind both aldehydes and ketones while legumin only binds aldehydes under neutral pH (Heng et al., 2004). In general, any protein isolate will likely bind aldehydes through hydrogen bonding, hydrophobic interactions, or covalent bond formation if high concentrations of lysine and arginine (NH₂ source) and sulfur (SH source) amino acids are present (Guo et al., 2019, 2020, 2023; Semenova et al., 2002; Wang & Arntfield, 2015a, 2015b, 2016a, 2017). In contrast, alcohols and ketones tend to bind proteins through weaker hydrogen bonding via the hydroxyl or ketone group or hydrophobic interactions via their aliphatic side chain (Bi et al., 2022; Wang & Arntfield, 2015a). Zhang et al. (2023) reported that ketones (i.e., carbonyl group) undergo stable hydrogen bond formation with lysine residue 412 and asparagine residue 484 in soy protein isolate. Other VOCs such as pyrazine, the undesirable compound in dry pea, undergoes hydrogen bonding between the nitrogen in the heterocyclic ring and the available amine or hydroxyl groups of the protein (Juhás & Zitko, 2020). The change (less α-helix and more β-sheets) in secondary structures facilitated the interaction between 2-methylfuran and soy protein (Zhang et al., 2023). These authors proposed that 2-methylfuran binding to soy protein occurred through hydrogen bonds with aspartic acid 137 and tyrosine 135, electrostatic interactions with aspartic acid 137, and hydrophobic interactions with isoleucine 30 and leucine 22. The multiple interactions between soy protein and 2-methylfuran likely contributed to the higher binding percentage of 2-methylfuran compared to whey protein isolate, which bound 2-methylfuran via hydrophobic interactions (Zhang et al., 2023). Like VOCs, non-VOCs also have an affinity for proteins via multiple interactions.

The interactions between proteins and polyphenolics are driven by hydrogen bonding, ionic interactions, covalent bonding, and hydrophobic interactions (Kroll et al., 2003; Ruan et al., 2022). Chlorogenic acid is a phenolic compound that binds protein resulting in a bitter and green-discolored sunflower protein isolate (Pedrosa et al., 2000). Jia et al. (2022) reported that chlorogenic acid can bind via covalent and noncovalent interactions depending on pH during protein isolation. For example, the di-hydroxy moiety on benzene ring can undergo oxidation leading to the formation of Michael's and Shiff base adducts with proteins (Figure 3). Xu and Diosady (2000) reported that ionic interactions were the predominant binding mechanism between canola protein and phenolics. However, the chemical interactions involved are dependent on the phenolic compound. Flavonoids make up the largest and most diverse groups among the polyphenolics. Flavonoids tend to contribute greater bitterness than non-flavonoid (e.g., phenolic acids) compounds due to the structural features such as ortho-hydroxyl groups and neighboring substitutions at the bonding sites. Although bitterness was not evaluated, Parolia et al. (2022) reported greater binding of quercetin to lentil protein isolate than ellagic acid and rutin. Rutin is a glycoside (i.e., rutinose) of quercetin substituted at the 3 position on ring C. The reduction in binding between rutin and the lentil protein isolate was proposed to be caused by the rutinose moiety interfering (through steric hindrance) with the noncovalent interactions between the B ring ortho-hydroxyl group of rutin and the lentil protein (Parolia et al., 2022). However, the rutinose moiety could also inhibit covalent bonding between rutin and protein (Figure 3).

Like flavonoids, tannins are bitter due to their multiple ortho-hydroxyl groups (Soares et al., 2018). Conflicting reports regarding polyphenolic size and bitterness perception exist in literature. The perceived bitterness tends to decrease as the degree of polymerization (i.e., molecular weight) of the polyphenol, within the same class of flavonoid, increases (Peleg et al., 1999). Sun et al. (2007) also reported that lower molecular weight phenolic compounds tended to have more bitterness and less astringency compared to high-molecular weight phenolic compounds. However, Hufnagel and Hofmann (2008) found the exact opposite where increasing bitterness corresponded to increasing polymerization (i.e., increased molecular weight). Inconsistencies in results likely are the result of differences in the testing methodologies. Viana and English (2021) reviewed chromatographic approaches used in flavor analysis of pulses and pulse by-products. These authors recommended that advanced methods involving mass spectrometry coupled with olfactory detections be expanded to facilitate our understanding of aroma active compounds in pulses. This approach is effective in assessing VOC but not non-VOCs. Instead, use of HPLC-based approaches coupled with mass spectrometry and sensory analysis (Cosson, Meudec, et al., 2022; Ongkowijoyo et al., 2023) are better suited for assessing non-VOC compounds. The main drawback to these methods is the potential variability of the sensory panelists. While training of panelists is essential for these methods, humans still impart subjectivity (i.e., bias) into the measure (Meilgaard et al., 2016). To overcome subjectivity in sensorial tests and determine how a compound elicits a perceived flavor sensation, the use of taste receptor cell assays have become common analytical tools. For example, taste receptor cell assays have been used to assess bitterness for almost two decades.

The TASTE 2 Receptor (TAS2R), also referred to as human TAS2R (hTAS2R), have been used to assess bitterness of phenolic compounds. There are 25 subtypes of this receptor; however, not all receptors will interact with all bitter compounds (Adler et al., 2000; Behrens & Meyerhof, 2006). The receptors hTAS2R14 and hTAS2R39 were found to be activated by 68 and 70 flavonoids, respectively, with 58 flavonoids activating both receptors (Roland et al., 2013). The glucoside forms tended only to activate hTAS2R39 (Roland et al., 2011, 2013). These authors observed that genistein had the lowest threshold (4–8 μM) among isoflavones while genistin (isoflavone glucoside) had threshold value of 500 μM (Roland et al., 2011). Soares et al. (2013) found that (−)-epicatechin had the highest (>3000 μM) threshold values compared to procyanidin trimer (35.6 μM) and a penta-O-galloyl-d-glucopyranose (PPG) compound (6.6–8.5 μM). The increasing number of OH substitutions favored interactions with taste receptors, resulting in lower thresholds. However, (−)-epicatechin was able to activate hTAS2R4, hTAS2R5, and hTAS2R39, while PPG activated hTAS2R5 and hTAS2R39 and procyanidin trimer only the hTAS2R5. An interesting finding in this study was that malvidin-3-glucoside activated hTASR7 at a concentration of 12.6 μM (Soares et al., 2013), which is significantly lower concentration than typical activation by flavonoid glycosides. Phenolic acid esters activated hTAS2R4 at concentrations higher than hydrolyzed and condensed tannins (Soares et al., 2018). The hydrolysable tannins activated hTASR7 while condensed tannins activated hTASR5, hTASR7, hTASR39, and hTASR43 depending on the tannin. The data regarding taste receptors clearly supports the bitterness of phenolic substances.

Our understanding of polyphenolic contributions to the undesirable astringency flavor characteristic has been documented for over 80 years. For example, astringency of unripe bananas was theorized to be associated with the formation of tannin complexes with protein (Barnell & Barnell, 1945). Astringency (i.e., drying sensation) tends to increase with increasing polymer size due to hydrogen bond formation between the hydroxyl groups of the polyphenol and carboxyl groups of the protein (Peleg et al., 1999). Furthermore, astringent polyphenols sufficiently alter the protein structure causing precipitation (Bate-Smith, 1973; McManus et al., 1981). More recently, the interaction between salivary proline-rich proteins and polyphenolics has been extensively characterized (García-Estevez et al., 2017; Ployon et al., 2018; Soares et al., 2011, 2016). In addition, a trigeminal response via an activation of a G protein by phenolic compounds has been identified as contributing to astringency (Schöbel et al., 2014). Therefore, the presence of phenolic compounds in protein ingredients will contribute to flavor concerns such as bitterness and astringency. Knowing the compounds that contribute to off flavors of protein ingredients is important because effective strategies can be developed to mitigate these compounds.

APPROACHED TO MINIMIZE FLAVOR OF PROTEIN INGREDIENTS

Introduction

Proteins can readily bind to both VOCs and non-VOCs (Roland et al., 2017), making removal of these compounds difficult. The type of interactions between proteins and off flavor compounds can determine if the off flavor will be problematic. If a compound is not released during consumption, the impact on product quality will be minimized (Frank et al., 2012). However, the protein digestibility might be impacted. Therefore, strategies to minimize off-flavors must be based on the underling chemistry of the protein and flavor complex. Methods to mitigate flavor concerns include extraction, thermal and hydrothermal treatment, protein modification, fermentation and germination, and plant breeding.

Extraction methods

Extraction is the most common approach to mitigating flavors from protein. Solvents such as ethanol and other alcohols, acetone, and water are frequently used to remove targeted compounds. Recently, supercritical carbon dioxide has been used to remove off flavor compounds in peas. The discussion below represents just a handful of studies that have targeted off flavor removal from plant sources as well as protein ingredients.

Volatiles organic compounds

Solvent extraction has yielded encouraging results in improving the sensory quality of pulse protein ingredients by removing off flavor compounds (Chang et al., 2019; Hillen, 2016; Wang et al., 2020). Diluted organic solvents (acetone, ethanol, isopropanol at concentrations of 35%–95% vol/vol) were employed to eliminate undesirable volatiles from lentil protein isolate (Chang et al., 2019). The results support that ethanol and isoproponal were effective solvents for deflavoring the protein sample at concentrations of 35%, 55%, and 75% vol/vol, in particular efficacy observed at 75% vol/vol. In contrast, the use of acetone led to an increase in the intensity of these volatiles. Similarly, Wang et al. (2020) reported alcohol washing with ethanol and isoproponal at concentrations of 50% and 80% vol/vol was effective in reducing the unpleasant flavor of pea protein concentrate. In another study, solvent extraction assisted with high pressure was employed to enhance the sensory quality of pea flour used different food applications (Hillen, 2016). Data showed that cake and cookie made with extracted pea flour received better sensory acceptance scores than those prepared with raw pea flour. However, the intensities of off-flavor compounds in the extracted pea flour were significantly higher than in the raw pea flour (Hillen, 2016). Gohl (2019) found that aqueous ethanol (47.5%) using a 63 min extraction effectively removed most VOCs and resulted in higher cookie sensory score compared to the cookies made with raw flour.

Supercritical carbon dioxide (SC-CO₂) extraction is widely recognized as a green technology for separating natural substances, including flavors and fragrances, lipids, food colorants, and volatile compounds from biological matrices (Shao et al., 2014; Vatansever et al., 2021). Kumari et al. (2016) reported the successful use of SC-CO₂ extraction to separate residual phospholipids, a major contributor to off-flavor generation in soymeal, compared to conventional defatting methods. The addition of ethanol to SC-CO₂ extraction has been shown to enhance the separation of polar compounds, such as alcohols, phenolic compounds, terpenoids and carotenoids, from plant sources (Xu et al., 2015). Cocero and Calvo (1996) reported the use of SC-CO₂ modified with ethanol for extracting sunflower oil from seeds. The results of this study revealed that elevating the ethanol concentration improved solubility during the extraction process, leading to an increased extraction of phospholipids from the seeds. The addition of 5% ethanol as a modifier to SC-CO₂ removed medium chain aldehydes, ketones, and alcohols from soy protein isolate (Maheshwari et al., 1995). In a recent study, SC-CO₂ + ethanol extraction was successfully employed as a flavor modification method to improve the sensory quality of pea flour by removing off-flavor compounds (Vatansever & Hall, 2020). The undesirable flavor compounds were effectively reduced under optimum conditions (22% ethanol, 86°C, and 42.71 MPa) of the extraction system based on instrumental analyses and olfactory evaluation. Sensory analysis of both deflavored and non-deflavored pea flours indicated significantly lower bitterness and pea flavor intensities in the deflavored pea flour compared to non-deflavored pea flours of varying particle sizes (Vatansever et al., 2021).

Adsorbent and ion exchange resin have been used during protein extraction. β-cyclodextrin removed 94% of the 2-nonanone bound to soy protein (Arora & Damodaran, 2010). Sapabeads SP 207 (polystyrene adsorbent) removed approximately 30%–41% hexanal from soy protein isolate (Inouye et al., 2002) and 32%–41% of the aldehydes from lentil protein isolate (Guldiken et al., 2021). In addition, these authors found other adsorbents effectively removed ketones and alcohols.

Non-volatile organic compounds

Much of the research on non-VOCs extraction has focused on analytical methods for quantifying these compounds while fewer have targeted specifically removal of the non-VOCs from protein ingredients. In contrast, process development has been one area where significant research has been done to create low phenolic content protein extracts (González-Pérez et al., 2002; Malik et al., 2016; Malik & Saini, 2017). Alteration of the protein functionality is a limitation of removing non-VOCs by solvent extraction (Malik et al., 2016; Roland et al., 2017; Subaşi et al., 2020). However, not all properties of proteins are negatively affected by solvent extraction.

Several researchers have modified traditional isoelectric precipitation method for protein isolation using alcohol washes (González-Pérez et al., 2002; Jia et al., 2021; Malik et al., 2016; Peng et al., 2021). The presence of aqueous ethanol (60% or less) during protein isolation were effective in producing protein ingredients from soybean (Peng et al., 2021) and sunflower (Jia et al., 2021) with less phenolic compounds compared to ethanol-free extraction of proteins. Malik and Saini (2017) reported a 90% reduction in phenolics for sunflower protein isolated prepared by first extracting the kernels with 60% methanol followed by protein isolation. Nandasiri et al. (2021) reported that 70% methanol tended to remove kaempferol derivative slightly better than 70% ethanol using accelerated solvent extractions.

The advantage of solvent extraction to remove flavors is based on the availability and cost of equipment. For example, a mixing tank is commonly used to extract protein; thus, the reuse of this tank would be applicable to solvent extraction. Furthermore, decanters or centrifuges that are currently in place in protein isolation facility could be used to remove solvent from the isolated protein prior to drying operations. Therefore, of the de-flavoring approaches that could be used, solvent extraction is the most adaptable since no or limited addition capital would be needed beyond the unit operations required for protein isolation. Furthermore, safety concerns regarding utilization of solvent extraction are minimal. Ethanol for example is widely utilized in the food industry with few restrictions. In contrast, the regulations (Code of Federal Regulations, CFR, 2023a, 2023b, 2023c) around acetone (21 CFR 173.210), isopropyl alcohol (21 CFR 173.240), and methanol (21 CFR 173.250) limit the usage and residual limits to 30, 50, and 50 ppm in spice oleoresins, respectively. Furthermore, isopropyl alcohol cannot exceed 6 ppm in lemon oil or 2% in hop extracts (provided a cooking step follows extract addition). While the application of acetone, methanol, and isopropyl alcohol for deflavoring is not listed under 21 CFR 173, it is likely that protein isolates would need to meet the same residue level found for spice oleoresins if these solvents were used for deflavoring.

Supercritical fluid extraction (SFE) using carbon dioxide and alcohol modifiers (SC-CO₂) and subcritical water extraction (SWE) are additional extraction techniques for removing non-VOCs. However, unlike solvent extraction, specialized equipment that can be pressurized and, when necessary, reach high temperatures (up to 250°C) if required for super- and sub-critical extraction techniques. Supercritical carbon dioxide with alcohol modifiers was evaluated as an extraction technique in soy flour (Rostagno et al., 2002) and soy meal (Zuo et al., 2008). The highest total concentration (31 μg/g flour) of isoflavones removed from the soy flour was from the SFE process using carbon dioxide with methanol modifier (10 mol%; SC-CO₂ + MeOH) at 50°C and 36 MPa (Rostagno et al., 2002). However, conventional method using ultrasonic-assisted extraction and Soxhlet extraction recovered 312 and 213 μg/g flour, respectively. In contrast, more daidzein 30.9 μg/g flour was recovered using SC-CO₂ + MeOH at 50°C and 36 MPa. Zuo et al. (2008) reported SC-CO₂ + MeOH (7.8 mass%) extraction conditions that achieved the highest (87.3%) isoflavone recovery included 40°C, 50 MPa and a carbon dioxide rate of 9.8 kg/h. The modifier in this study was aqueous methanol (20%:80% vol/vol).

Unlike SFE, SWE has a significant potential to extract or promote degradation of carbohydrates, fats, proteins, and polyphenolics compounds. The application of SWE as a means to remove phenolic compounds from canola meal showed that extraction at 160°C and 6.9 MPa for 30 min produced the highest extraction yield and total phenolic content per gram meal basis compared to the SWE completed at 110°C and 6.9 MPa for 30 min (Hassas-Roudsari et al., 2009). Conjugated isoflavones (i.e., β-glycoside, malonyl glycosides and acetyl glycoside) undergo conversion to aglycone forms with prolonged processing at temperatures over 100°C (An et al., 2023; Nkurunziza, Pendleton, and Chun, 2019; Nkurunziza, Pendleton, Sivagnanam, et al., 2019). The malonyl glycosides were the most susceptible to SWE conditions. An et al. (2023) reported that the optimal SWE conditions, based on total isoflavone yield from soy flour, were 150°C for 25 min when pressure was maintained at 10 MPa. Using response surface methodology, Nkurunziza, Pendleton, and Chun (2019) reported the optimal SWE conditions were 146.23°C, 3.98 MPa, and solid to liquid ratio of 20 mg/mL for removing isoflavones from dried Okara.

Regardless of the SWE protocol, solvent extraction tends to remove greater amounts of total isoflavones (An et al., 2023; Nkurunziza, Pendleton, and Chun, 2019; Nkurunziza, Pendleton, Sivagnanam, et al., 2019). However, under select SWE conditions, the conversion from glycoside to the aglycone form was observed. For example, An et al. (2023) reported significant increases in genistein at processing conditions of 180°C and 10 MPa for 30 min. Furthermore, genistein concentration did not surpass the concentration of an 80% methanol extract obtain from soxhlet until 185 minutes of processing at 140 and 3.75 MPa (Nkurunziza, Pendleton, & Chun, 2019). These authors reported the presence of daidzein (another aglycone) in the extract obtained by SWE but not soxhlet (Nkurunziza, Pendleton, and Chun, 2019; Nkurunziza, Pendleton, Sivagnanam, et al., 2019). The authors assessing SWE did not evaluate the sensory characteristics of the extracted soy flour or okara; however, the removal of the aglycones from the soy flour and okara are significant because they tend to have greater bitterness compared to the glycosidic forms (Roland et al., 2011). Thus, the soy flour and okara would be expected to have less bitterness. The limitations of SWE include the co-extraction of other material such as proteins and that individual polyphenolic removal has optimal conditions (Cheigh et al., 2015; Hassas-Roudsari et al., 2009; Ko et al., 2014). Within the flavonol aglycone group for example, quercetin was optimally extracted at 170°C and 10 MPa for 10 min while Kaempferol optimal extraction was 190°C for 15 min at 10 MPa (Ko et al., 2014). Thus, the removal of polyphenolic compounds from protein ingredients will require knowledge of the specific phenolic compounds requiring extraction.

The use of adsorbent and ion exchange technologies to produce low phenolic-containing protein isolates has been demonstrated in canola (Xu & Diosady, 2002) and sunflower (Pickardt et al., 2015; Weisz et al., 2010). In these procedures, the protein is generally isolated through isoelectric precipitation followed by mixing an adsorbent such as polyvinylpyrrolidone (PVP), Dowex™, or Amberlite™, with or without subsequent filtration steps. Xu and Diosady (2002) reported a 50% reduction in phenolic acids in soy protein isolate subjected to PVP treatment. In sunflower protein isolation, Weisz et al. (2010) reported that ion exchange resins reduced phenolic compounds by up to 96% while approximately 85% reduction in phenolic compounds were observed using other adsorbents. Other researchers found that a combination of adsorbent and anion exchange resin reduced the phenolics by 99% in a pilot scale preparation of sunflower protein isolate (Pickardt et al., 2015).

Modification of protein

Physical denaturation

Modification through denaturing has been shown to impact binding of off flavors to protein. Denaturation of protein using non-chaotropic salts or extreme pH led to a reduction in ketone bind to pea protein isolates (Wang & Arntfield 2015a, 2015b). Sastry and Rao (1990) reported similar trends but for the binding of chlorogenic acid with polyphenol-free sunflower protein. However, modification through denaturation is dependent on the volatiles present in the protein extraction system and method of denaturation, as denaturing can enhance aldehyde binding (Wang & Arntfield, 2014). Blocking reactive sites on the protein can be an alternative to denaturation.

High pressure processing (HPP) has been widely evaluated as a method for modifying protein functional properties. However, the impact of HPP on flavor of protein has not been widely reported for plant proteins. Flavor research associated with HPP processing of diary protein has been reported (Kühn et al., 2006). The few research studies on flavor relative to HPP show the tendency of weaking of interactions between plant protein and flavor compounds (Bi et al., 2022; Houde et al., 2018). The pea milk processed with higher pressure (550 MPa) retained more hexanal than the pea milk processed at 200 MPa (Bi et al., 2022). Mixed results were obtained for the removal of isoflavones from soybean by HPP. No reductions in isoflavones were observed for soymilk or smoothies made with soy after HPP processing (Andrés et al., 2016; Jung et al., 2008). However, an interconversion of the malonyl form to β-glucosides was observed (Jung et al., 2008). In contrast, Ueno et al. (2019) reported the reduction of isoflavone in soybean processed at pressure over 200 mPa. The bitterness and astringency of the HPP process soybean were reduced and theorized that the reduction in undesirable characteristics were due to the isoflavone reduction; however, other bitter and astringent compounds such as saponins were not evaluated. Furthermore, conversion of the malonyl form to β-glucosides might contribute to the reduction in astringency (Aldin et al., 2006). The few research studies that have been reported involving HPP and flavor characterization suggest a benefit removal of undesirable compounds; thus, additional research in his area is warranted.

Structural modification

Various structural modifications, including enzymatic and chemical modification have been employed to improve protein functionality by altering protein structure. These methods have also been tested to modify the flavor characteristics of proteins. Arteaga et al. (2020) investigated the effects of enzymatic hydrolysis on the sensory perception of pea protein isolates. In this study, among sensory attributes, only bitterness changed significantly after 15- and 120-min of hydrolysis, which was attributed to the formation of bitter peptides. The authors also noted a strong correlation between bitterness and the degree of hydrolysis due to the formation of small peptides containing hydrophobic amino acid residues. Wang and Arntfield (2016b) investigated effects of enzymatic and chemical modifications on the binding properties of pea protein to selected aroma compounds. Enzymatic hydrolysis resulted in an increased binding to octanal and dibutyl disulfide, while it reduced the binding for 2-octanone and hexyl acetate.

Deamidation of soy protein using protein-glutaminase reduced binding affinity of soy protein (Suppavorasatit et al., 2013). These authors reported deamidation the reduction in binding affinity of soy protein with vanillin and maltol was caused by a change in interactions (i.e., hydrophobic and covalent) to hydrogen bonding and van der Waals forces, which are weaker interactions. Application of the deamidation process to soymilk showed that the vanilla and cotton candy intensity was higher in the deamidated-treated soymilk compared to a soymilk control (Suppavorasatit et al., 2013). Fang et al. (2020) also found that deamidation of pea protein improved sensory characteristics, that is, less beany and bitter, of the resulting pea protein. The increase in Umami intensity of deamidated wheat gluten was proposed as one reason for decrease in bitterness observed by a sensory panel (Liu et al., 2017). Overall, the deamidation of protein supports improvement in flavor characteristics of protein ingredients; however, additional research on deamidation is needed.

The modification of pea protein through acetylation and succinylation led to the release of octanal and hexyl acetate, while 2-octanone and dibutyl disulfide binding increased at a low ratio of reactants to protein mass (Wang & Arntfield, 2016b). Succinylation was also effective in preventing chlorogenic acid binding to sunflower protein (Sastry & Rao, 1991). These authors also observed that N-ethylmaleimide was ineffective in preventing chlorogenic acid from binding sunflower protein. Additional observation showed that succinylation blocked some of the binding sites but not binding affinity. Therefore, for the modification of a protein to be effective, the binding site must be modified to prevent the targeted VOCs or non-VOCs from binding.

The conjugation of plant proteins with carbohydrates presents an attractive option for the removal of undesirable odor-active compounds. The conjugation of pea proteins with gum Arabic through a Maillard reaction reduced the intensity of selected off-flavor compounds, which were attributed to the beany flavor (Zha et al., 2019). Additionally, cyclodextrins are known for their ability to entrap and mask volatiles using their hydrophobic cavities. The combination of cyclodextrin with pea protein through solid-based spray drying significantly mitigated off-flavor compounds (Cui et al., 2020a, 2020b). Cyclodextrins have also been reported to bind hydrophobic amino acids associated with bitterness (Kryachko et al., 2023).

Germination and fermentation

Germination is a process for improving the nutritional value and reducing anti-nutritional factors of pulse seeds. Basically, germination starts from seed hydration. Water is absorbed by dry seeds with a concentration gradient and diffusion. Recently, researchers found that controlled germination could potentially improve the flavor profile of pulses. For instance, Xu et al. (2019) studied the dynamic flavor change of chickpea, lentil, and yellow pea during pulse germination. Totally, 124 of volatiles have been recorded during pulse germination. They noted an increase in the intensity of off flavor compounds, including increased hexanal, 1-hexanol, 2-pentylfuran, and 2-methoxy-3-isopropyl pyrazine, in germinated lentil and yellow pea seeds. However, germination reduced the off-flavor intensity for chickpea seeds. Most of the volatiles increased during a long time of germination. However, with 1 or 2 days of germination, the volatiles in the pulse seeds did not develop extensively. Interestingly, some volatiles decreased in the first days of chickpea germination. For example, 2,4-nonadienol and hexanal, two of the major contributors of beany flavor, decreased 72% and 70%, respectively, after 2 days of chickpea germination. An increased intensity of key volatiles, such as hexanal, 1-hexanol, and alkyl methoxypyrazines, which are strongly associated with undesirable flavors, and formation of new volatiles, including dimethyl sulfide and 2-methylbutanal, causing meaty and sulfur aromas in germinated lupin and soybean seeds were reported (Kaczmarska et al., 2018).

Troszyńska et al. (2011) investigated the impact of germination on the flavor profile of lentils. After 7 days of germination, a reduction in the intensities of bean and green flavors was observed, while the intensities of the off odor, bitterness, and astringency were elevated, indicating an overall reduction in the sensory quality of germinated lentils. The increase of bitterness and astringency was attributed to the formation of tannins and catechin during germination (Vidal-Valverde et al., 1994). In another study, bitterness was reduced during this process by degrading tannins. Furthermore, germination has been found to decrease antinutritional factors, including the undesirable flavors caused by lipid oxidation (Simons, 2011). As an added benefit, germination has the potential to increase the levels of phytonutrients such as vitamins, phytosterols, saponins, and phenolics (Simons, 2011). A drawback of germination is that as inherent off flavors are reduced, new undesirable flavors may arise. While the practice of germination provides some benefits, this process appears limited as a deflavoring method for legume ingredients.

Fermentation, another bio-processing strategy, has traditionally been employed to enhance flavor and nutrient content in foods (Kaczmarska et al., 2018; Schindler et al., 2012). Lactic acid fermentation has garnered attention for its potential to improve flavor characteristics of plant protein ingredients by reducing off flavors and generating new volatiles through transaminase and lyase pathways, and non-enzymatic conversions (Shi et al., 2021). Schindler et al. (2012) examined the impact of lactic acid fermentation on the aroma profile of pea protein extracts using Lactobacillus plantarum and Pediococcus pentosaceus. This fermentation process exhibited notable potential to improve the flavor characteristics of pea protein extracts by producing pleasant volatiles that masked off-flavors or reduced key volatiles, like hexanal. Similarly, Shi et al. (2021) reported that lactic acid fermentation of pea protein using L. plantarum noticeably decreased the intensity of the off-flavor compounds, including aldehydes and ketones, while slightly increasing the alcohol content. Sensory evaluation of the samples indicated an improvement in the aroma and taste attributes of pea protein through this fermentation. However, lactic acid fermentation applied to lupin flour using yoghurt culture resulted in an increased intensity of off-flavor compounds, contributing to mushroom, green, soil, and nutty aromas, as revealed by instrumental analysis and olfactory evaluation (Kaczmarska et al., 2018). Ben-Harb et al. (2019) also suggested that fermentation could produce pleasant flavors to mask the off-flavors. They screened 55 strains from different microbial species for improving the aroma of pea. Two yeast strains, Candida catenulate and Geotrichum candidum, were able to colonize pea-based products and positively influence the release of volatile compounds by generating roasted/grilled, fruity, and lactic aroma for pea.

Based on the findings of fermentation and germination, there is an inconsistency in the flavor profile of legumes. Variations in flavor characteristics of germinated and fermented legumes are likely attributed to factors, such as the source of protein ingredients and the specific experimental processes. It appears that the effectiveness of these methods may be crop-specific, necessitating the optimization of the processing parameters.

Thermal and hydrothermal

A plethora of methods have been explored to inactivate antinutritional factors in legumes, a well-documented endeavor in the literature. These approaches have also been attempted to improve their flavor attributes of legume ingredients. Typically, thermal treatments have been reported as effective tools to diminish undesirable flavor, as indicated by the decreased intensity of beany flavor in precooked flours. Microionization treatment at 130 and 150°C resulted in a reduction in odor-active compounds in chickpea and with a lesser to lentil flours, and it improved the sensory quality of the end-products fortified with these microionized flours (Shariati-Ievari et al., 2016). In the case of pea protein isolate, it exhibited a lower pea flavor intensity when subjected to spray drying in comparison to freeze drying. Additionally, spray-dried pea protein isolate had increased emulsion and foaming capacities (Hoang, 2012).

Water treatments, including soaking and blanching, can be employed to remove water-soluble off-flavor compounds, such as alcohols and non-volatile compounds (Roland et al., 2017). Additionally, blanching can deactivate LOX in pulse seeds (Jakobsen et al., 1998). However, it results in irreversible alterations to the structure of pulse ingredients, thereby affecting their functionality. Specifically, blanching requires substantial water and energy resources. With a similar approach to blanching, heat-steam treatment exhibited a notable potential to remove off-flavors of pulse ingredients. Heat-steam treatment at 120 and 140°C with 10% steam significantly improved the sensory quality of yellow pea flour and navy bean flour (Bourré et al., 2019). This treatment resulted in reduced pea and beany aromas compared to untreated flours. Furthermore, both heat-steam treated flours had lower bitterness intensity. However, heat treatment alone, without steam, was less effective in improving the flavor characteristics of these flours. Overall, hydrothermal processes in general improve sensory properties of flour; however, like germination these methods are not suitable for deflavoring protein ingredients due to the impact on protein functionality.

Filtration methods

Membrane separation methods, such as ultrafiltration (removing particles in the range of 0.001–0.02 μm), have commonly been employed to produce plant protein isolates with increased purity, recovery, higher functionality, and reduced anti-nutritional components (Mondor et al., 2010; Singhal et al., 2016). Protein isolation from plant materials, such as legumes, is mostly performed by alkaline extraction coupled with ultrafiltration and/or diafiltration (Singhal et al., 2016). The ultrafiltration/diafiltration (UF/DF) sequence is often used to improve protein recovery (Mondor et al., 2010). Microfiltration and UF procedures are well-known methods for whey protein fractionation, such as producing whey protein isolate from whey protein concentrate. These filtration methods have been reported to improve flavor characteristics of whey protein isolates through removing various volatile compounds derived from lipid oxidation, as well as reducing fat content (Mortenson et al., 2008). Ser et al. (2008) reported that ultrafiltration removed approximately 20%–30% of the glucosinolates from canola protein.

In recent studies, Benavides-Paz et al. (2022a), Benavides-Paz et al. (2022b) investigated the impact of process parameters (e.g., solubilization, neutralization, UF/DF, homogenization) on the flavor profile of pea protein isolates produced by two extraction methods, alkaline and salt extractions coupled with the UF/DF sequence (AE-UF/DF and SE-UF/DF). Both studies showed that there was a slight change in number of volatiles in pea samples obtained at different steps, such as pea flour (initial sample), neutralization, UF/DF, homogenization, and pea protein isolate (final product). However, the relative amounts of the selected volatiles used for quantification varied at different steps of both pea protein extraction processes. Most volatiles, such as hexanal, were in high concentrations in the protein sample obtained from neutralization step; however, a significant reduction in the volatiles of the protein sample was observed after the UF/DF step using the SE-UF/DF method (Benavides-Paz et al., 2022b). Likewise, a significant reduction in the relative concentration of selected volatiles was observed in the protein sample after the UF/DF and AE-UF/DF processes (Benavides-Paz et al., 2022a). The use of filtration methods has been widely documented as an effective method for protein isolation. However, much of the research deals with functionality of the protein obtained from the process. Only a few plant protein research studies involving filtration methods and flavor characterization of resulting protein products have been published. Therefore, this area of research should be expanded to cover the lack of information specific to filtration methods for isolating plant proteins and the impact on flavor compounds.

Plant breeding

While plant breeding can be a potential avenue to reducing unwanted components, plant breeding programs are long-term, that is, 5-to-10-year endeavors. Furthermore, the compounds that humans find objectionable, for example, saponins, play a vital role in the plant defense against pests such as insects (Güçlü-Üstündağ & Mazza, 2007; Qasim et al., 2020). Cultivar selection of protein crops with minimal off-flavor profile could be a potential short-term approach to select protein sources with less flavor (Roland et al., 2017). For example, pea cultivars with varying off flavor components have been reported (Arteaga et al., 2021; Azarni, Boye, Warkentin, & Malcolmson, 2011; Cui et al., 2020; Malcolmson et al., 2014). However, ideal cultivar selection can be challenging due to likely differences in the flavor profile of crops impacted by diverse factors, such as year and location (Azarni, Boye, Warkentin, & Malcolmson, 2011). LOX might be controlled through plant breeding to prevent unsaturated fatty acid breakdown and improve sensory properties of pulse crops and their ingredients. In LOX free soybean, hexanal was lower than the LOX containing soybean, but other volatiles were not impacted (Matoba et al., 1996). Thus, food processing technologies to improve flavor is likely a better short-term solution to plant breeding approaches.

CONCLUSIONS

The intent of this review was to focus on the off flavors associated with plant protein ingredients. There are several research and review articles in literature that focus on flavor of animal proteins, primarily dairy. This review focuses only on plant protein, thus literature on dairy proteins and flavor issues related to dairy protein were not included. However, this body of knowledge is supportive of the types of proteins and flavor interaction presented in the current review. While understanding the flavor compounds of the whole material is important, the data cannot necessarily be translated to protein ingredients. The significance of this review relates to the compounds isolated from plant proteins. There were many references not used due to the limited scope of this review. The authors strongly suggest that researchers working in this field consider the body of knowledge on protein isolates, flavor interactions, and mitigation strategies to investigate literature from 1970 to 2000. Many research articles with focus on soy protein highlight flavor binding to the protein, and methods to improve flavor characteristics. In general, a body of knowledge on VOCs and non-VOCs supports that plant proteins from different sources have similar off-flavor compounds. However, minor differences exist in the flavor compounds, but not extensively such that an overall strategy to mitigate off-flavor using similar technologies cannot be applied. Any mitigation strategy must target the fundamental interactions between the flavor compound and the protein. In this review, the interactions were identified as reversible and nonreversible. Reversible interactions include hydrogen bonding, hydrophobic interactions, and electrostatic/ionic as examples. Flavor compounds will interact with carboxyl (COOH), sulfhydryl (SH), amine (NH₂), and hydroxyl (OH) groups of protein. In contrast, disulfide (SS), SH, and NH₂ of proteins can undergo irreversible binding through covalent bond formation with the flavor compound. Pulse proteins for example have a high lysine (NH₂ source) content and thus have the potential to bind flavors through both reversible and irreversible mechanisms. Overall, any covalently bound flavor to protein will not be removed by current technologies without substantial changes to the protein structure. In contrast, reversible interactions can be disrupted using physical and chemical modification (e.g., heat, HPP), alcohol solutions, super critical fluid and subcritical extraction methods, and to a lesser extend filtration. While germination and fermentation have been presented, these methods are not well suited for deflavoring since these processes tend to generate new compounds. Based on a balance between removal of multiple flavor compounds and capital inputs, the most promising approach to remove flavors is the use of solvent extraction, specifically aqueous ethanol. The ethanol has moderate polarity and will remove reversibly bound compounds such as aldehydes and ketone and those bound via hydrophobic interactions. Addition of water to the ethanol (30:70 vol/vol) creates a solvent that can disrupt hydrogen bonds and electrostatic interactions, allowing for polyphenolics for example to be removed.

AUTHOR CONTRIBUTIONS

Serap Vatansever: Created images and wrote text dealing flavor compounds and mitigation strategies. All authors contributed to and approved the final draft of the manuscript. Bingcan Chen: Created images and wrote text dealing with mitigation strategies. Clifford Hall: Conceived and structured the review paper, and overseen the submission and revision of the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

ETHICS STATEMENT

No human or animal subjects were used in creation of this review article.