Evaluation of Sweetener Synergy in Humans by Isobole Analyses

Abstract

The chemical senses and pharmaceuticals fundamentally depend on similar biological processes, but novel molecule discovery has classically been approached from vastly different vantage points. From the perspective of ingredient and flavor companies, there are countless ingredients that act via largely unknown mechanisms, whereas the pharmaceutical industry has numerous mechanisms in search of novel compounds. Mixtures of agonists can result in synergistic (superadditive) responses, which can be quantified via isobole analysis, a well-proven clinical approach in pharmacology. For the food and beverage industries, bulk (caloric) sweeteners like sugars are a key ingredient in sweetened foods and beverages, but consumers also desire products with fewer calories, which has led to the development of sweet enhancers and sweetener blends intended to achieve synergy or superadditivity. Synergistic mixtures are highly attractive targets commercially as they enable lower usage levels and enhanced efficacy. Although the psychophysical literature contains numerous prior reports of sweetener synergy, others have also noted that classical additive models fail to account for nonlinear dose-response functions. To address this shortcoming, here we systematically apply the isobole method from pharmacology to quantify the presence or absence of psychophysical synergy for binary pairs of sweeteners in a series of 15 separate experiments, each with ~100 adult volunteers (total n = 1576). Generally, these data support the hypothesis that structurally similar sweeteners acting as agonists will not synergize, whereas structurally dissimilar sweeteners binding to overlapping or distal sites can act as allosteric agonists or agonist-antagonists, respectively.

Introduction

The US Centers for Disease Control and Prevention estimates that, as of 2017, over 39% of United States adults have obesity; globally, the number of adults with obesity may exceed 650 million (Hales et al. 2018). Obesity increases the risk of heart disease, Type II diabetes, stroke, and other preventable diseases, costing an estimated $2 Trillion per year globally (Tremmel et al. 2017). The 2015–2020 Dietary Guidelines for Americans recommend that consumption of added sugars should not exceed 10% of daily calories (US Department of Health and Human Services and US Department of Agriculture 2015). Assuming daily intake of 2000 calories, this is approximately 50 grams of added sugars per day, yet the typical American consumes almost twice this amount of sugar, ~94 grams per day. Excess intake of added sugars contributes to obesity, Type II diabetes, as well as dental carries (Hu 2013; Khan and Sievenpiper 2016). Accordingly, dietary sugar reduction is a target of numerous public health agencies and food producers around the globe. One strategy to reduce sugar is replacement with nonnutritive sweeteners (NNSs). Although many NNSs are currently approved for commercial products, NNS containing foods and beverages are not universally palatable (Reed and McDaniel 2006; Simons et al. 2008). Moreover, negative consumer beliefs about artificial sweeteners have led to commercial demand for natural NNS systems, such as the rebaudiosides, but these also suffer from taste issues (Kinghorn and Soejarto 1989; Allen et al. 2013). Blends of sweeteners and positive allosteric modulators (PAMs) offer additional opportunities to lower the levels of sugar found in products while preserving the sweetness consumers desire (Servant et al. 2010).

Sweet taste is primarily transduced by the interaction of specific ligands with the G-protein-coupled receptor (GPCR) heterodimer T1R2/T1R3 expressed in taste buds; the T1R2/T1R3 proteins are encoded by the TAS1R2 and TAS1R3 genes (Nelson et al. 2001; Li et al. 2002; Zhao et al. 2003). Some evidence suggests that other mechanisms may also be involved (e.g., Yee et al. 2011; Sukumaran et al. 2016). Peripheral signals from the taste buds are relayed to the brain, where the intensity, time course, and appetitive/aversive responses are represented and processed into perception (Wang et al. 2018). T1R2 and T1R3 are thought to have similar structure to the C-class GPCR metabotropic glutamate receptors, calcium sensing receptor, and GABA_B (Bettler et al. 2004; Muto 2007; Geng et al. 2012). The sweet taste receptor is highly promiscuous and contains multiple ligand-binding sites for a wide range of molecules that humans describe as sweet; these sites also bind various modulators (DuBois 2004; Cheron et al. 2017). The T1R2/T1R3 heterodimer shows an impressive breadth of ligand affinity and concentration responses (e.g., 0.44 M for sucrose vs. ~200 nM for thaumatin; see DuBois et al. 1991).

The X-ray crystal structures of the metabotropic glutamate receptors (mGluRs) have provided a wealth of information on the structure, ligand binding, and mechanisms of C-class GPCRs (Doumazane et al. 2013; Vafabakhsh et al. 2015). Figure 1 shows a structural model of T1R2 and T1R3, based on previously reported crystal structure of mGluR2 (Monn et al. 2015). The homology model was built with the Venus flytrap domain (VFD) of mGluR2, the cysteine-rich domain (CRD) of mGluR4 and the transmembrane domain (TMD) of rhodopsin (Protein Data Bank [PDB] I: 5CNJ, 5CNK, and 1F88, respectively). The othrosteric VFD is a bilobate (clam shell) structure composed of 2 lobes, named ligand-binding Domain 1 and 2 (LB1 and LB2) (Cheron et al. 2017). Furthermore, electron microscopy and molecular modeling of medaka fish T1Rs and chimeras and mutational analysis in rodent and human receptors have shown that the LB1 and LB2 of each VFD can adopt an open (inactive) and closed (active) sweetener bound confirmations (Masuda et al. 2012; Nango et al. 2016; Nuemket et al. 2017). The CRD is an ~80 amino acid region with 4 highly conserved disulfide bonds and a single free sulfhydryl (Wu et al. 2014). The CRD tethers the extracellular VFD to the cell membrane TMD. The TMD has 7 transmembrane spanning alpha-helixes that transmit the ligand-bound active state to G proteins, such as gustducin, modulating downstream signal transduction cascades (Wong et al. 1996). Alternative mechanisms of sweet taste and synergy may include T1R-independent modulation via glucose transporters (Glut 4 and 5), the sodium/glucose cotransporter 2 (SGLT2), potassium inwardly rectifying channel (Kir 6.1), and the sulfonylurea receptor 1 (SUR1) (Toyono et al. 2011; Yee et al. 2011). Glucose and fructose transporters on taste cells could have modulatory properties as these sugars will increase intracellular metabolic processes. Additionally, enzymes in saliva and on taste cells can metabolize sugars and certain starches (Sukumaran et al. 2016).

Figure 1.

A scale model of sweeteners and the T1R2/3 heterodimer. Three major domains are illustrated. The first is the VFD, each composed of 2 lobes, LB1 and LB2 forming a pocket that closes upon ligand binding, thus providing the conformational change to activate downstream pathways. LB1 and LB2 for T1R2 are shown as blue and magenta. Fructose, sucrose, sucralose, aspartame, and RebA are putative T1R2-binding sweeteners. LB1 and LB2 for T1R3 are shown as red and orange. AceK, and Na-saccharin are T1R3 agonists. The second domain is the CRD that tethers the VFD to the TMD. T1R2 and T1R3 both have CRDs shown in yellow. The T1Rs are in the family of GPCRs with classic 7- transmembrane alpha-helices that form the TMD. The TMD spans the lipid bilayer and converts VFD conformational changes into intracellular signal transduction cascades. NHDC is thought to bind to the TMD, while thaumatin is believed to interact with the CRD and LB2 of T1R3. The binding site for RebA is currently not known; however, recent molecular modeling studies have predicted that RebA and sucralose and sucrose can bind to both T1R2 and T1R3. Low molecular weight sweeteners were rendered in classic sphere representation of hydrogens in white, carbons in black, sulfur in yellow, oxygen in red, and nitrogen in blue thaumatin, a 22.4 kDa protein (PDB ID: 1RQW), is shown as a ribbon diagram.

T1Rs binding sites have been investigated by in vitro cell-based assays, in vivo animal 2-bottle preference, brief access, nerve recordings, in silico homology modeling, as well as with a variety of gene knockouts and mutational analyses (Nelson et al. 2001; Zhao et al. 2003; Morini et al. 2005; Delay et al. 2006; Winnig et al. 2007; Ohta et al. 2011; Maillet et al. 2015). Collectively, these results suggest acesulfame-potassium (AceK), saccharin, aspartame, fructose, sucrose, and sucralose interact with the VFD of T1R2, while sucrose and sucralose also activate T1R3 (Zhao et al. 2003; Nie et al. 2005; Masuda et al. 2012; Maillet et al. 2015; Assadi-Porter et al. 2018). Furthermore, the LB2 and CRD on T1R3 contacts the sweet-tasting protein thaumatin; and the TMD on T1R3 binds neohesperidin dihydrochalcone (NHDC) and cyclamate (Jiang et al. 2005; Winnig et al. 2007). Figure 1 illustrates sweeteners that are positioned near their putative binding domains, with the caveat that some ligands, specifically rebaudioside A (RebA), sucrose, and sucralose, may possibly bind to more than one site.

Synergistic mixtures of sweeteners are potentially attractive to ingredient suppliers and food manufacturers, as blends can reduce off tastes, lower cost, or provide ingredient functionality that is lacking for individual NNSs. Here, we evaluated psychophysical results for 15 binary mixtures of bulk and high-potency sweeteners using the isobole method from pharmacology. We compare our results to prior literature and discuss these data in terms of contemporary models of sweet taste receptor activation. This work also illustrates the application of the isobole method to predict synergy for novel sweetener combinations.

Quantitation of mixture effects was first introduced by Loewe in 1928 as the general isobole equation, and similar approaches have been used widely to evaluate drug combinations for synergistic action (see (Tallarida 2016) for a comprehensive review). Here, we use the specific model and definitions proposed by Suhnel in 1993 (Sühnel 1993). Specifically, synergy occurs when the observed effect of the mixture of 2 sweeteners is greater than what would be expected from each agonist in isolation; similarly, suppression (i.e., antagonism) is defined as when the observed effect for the mixture of 2 sweeteners is less than what would be expected from each sweetener (agonist) in isolation. Zero interaction (or simple additivity) occurs when the observed effect is no different from what one would expect from the individual effects of each agonist by themselves (for additional detail, see Sühnel 1993; Fleming et al. 2016). In the isobole method, a zero-interaction response surface can be determined from the individual dose-response functions for each ligand (Tallarida 2016). The zero-interaction response surface is composed of many hypothetical pairs of ligands, called “additive isoboles,” which are mixtures of Ligands A and B in specific concentrations that elicit equivalent responses. Thus, if the 3D dose-response surface is visualized as a 2D contour plot from above, the additive isoboles form parallel straight lines on the diagonal. Observed responses for specific ligand pairs (here, sweetener blends) can then be compared with the zero-interaction surface, either directly (in a 3D plot) or as deviations from a straight (additive) isobole in a 2D top down contour plot (e.g., see Supplementary Figure 1 or Wolf et al. 2010). Pairs of sweeteners (i.e., mixtures) that evoke observed responses that fall on the zero-interaction surface (or on the straight isobole) show simple additivity, whereas any substantial deviation from the predicted response surface indicates synergy or suppression. In a 3D plot, these responses fall above or below the zero-interaction surface, whereas, in a 2D plot, they are shown as curved lines that bow toward or away from the origin (see Supplementary Figure 1).

Although the chemosensory literature contains many reports of sweetener synergy (with or without isobole analysis; see Hayes 2008), other researchers have also noted that most commonly used models of additivity implicitly assume linear dose-response functions (Lawless 1998). Other studies have tried to overcome this issue by studying the behavior of self-mixtures (i.e., “mixing” a compound with itself to achieve various concentrations) to account for nonlinear psychophysical functions when assessing mixture interaction (Frank et al. 1989). However, self-mixture models can potentially lead to erroneous conclusions: that is, what appears to be synergy or superadditivity instead reflects a straightforward, albeit nonlinear, psychophysical function for that specific stimulus. The isobole method does not make the same implicit assumptions about the shape of the dose-response function, thereby avoiding issues found in prior reports of sweetener synergy within the chemosensory literature.

Materials and methods

Overview

In a series of 15 separate experiments, convenience samples of adult volunteers were given solutions of sweeteners to taste. They were asked to rate sweetness and bitterness on a horizontal general labeled magnitude scale (gLMS) after being oriented to the scale. Within any 1 experiment, participants received 15 different solutions that consisted of either a single sweetener or a binary combination. Roughly 100 adults participated in each experiment, with a total n of 1576 across all experiments. All data were collected in a custom-build state-of-the-art sensory testing facility (the Sensory Evaluation Center at Penn State) between October 2014 and November 2015 at a rate of 1–2 experiments per month. Numerous other studies on other stimuli were run in this facility during the same time period. A total of 9 different sweeteners were tested across the 15 experiments, representing an incomplete set of all possible combinations. The full set (i.e., 36 separate experiments, with 3600+ participants) could not be tested due to logistical and resource constraints; the binary pairs tested here were based on putative binding sites for ligands on the sweet heterodimer at the time of testing.

Participants

Volunteers were recruited from an existing database of individuals who have previously opted in to be contacted about “taste tests” in our facility. This pool consists of age-diverse individuals in central Pennsylvania who are affiliated with Penn State and/or live in nearby communities. Participants were prescreened via an online questionnaire to meet broad inclusion criteria: aged 18–64, nonsmokers, no cheek, lip, or tongue piercings, and no known food allergies or defects of smell or taste, no pregnant or breastfeeding women, and no history of choking or difficulty swallowing. Detailed demographics were not collected here, and these free-living adults were presumed to be nominally healthy, although no attempts were made to screen for specific neurological or metabolic conditions. Generally, most of our participants self-identify as white (consistent with the demographics of the local region), and we typically have a gender split of ~65% women and ~35% men. Study procedures were exempted from institutional review board review by the Penn State Office of Research Protections under the wholesome foods exemption in 45 CFR 46.101(b)(6). All participants provided informed consent, which was documented via a yes/no click-through question presented at the beginning of each test session. This work complied with the Declaration of Helsinki for medical research involving human subjects.

Stimuli

Within a single experiment, 15 stimuli were presented in 10-mL aliquots to participants at room temperature (~20 °C) in plastic medicine cups labeled with random 3-digit blinding codes on a single tray. A complete block design was used (i.e., all participants received all samples), and presentation order was counterbalanced across participants in a Williams design to minimize order, position, and carryover effects. Binary mixtures and single-compound solutions were prepared in reverse osmosis (RO) water at levels roughly isosweet with 4%, 6%, and 8% sucrose. These concentrations, commonly called sucrose equivalents (SEs), were estimated from earlier work (DuBois et al. 1991). At each sucrose equivalent level, the blends of sweeteners varied: 100% A, 75% A/25% B, 50% A/50% B, 25% A/75% B, and 100% B.

The sweeteners and highest concentrations tested were here: sucrose (233 mmol L-1; Domino), fructose (361 mmol L-1; Alfa Aesar), thaumatin (7.65e-4 mmol L-1; Alfa Aesar), AceK (5.9 mmol L-1; donated by PepsiCo), aspartame (0.76 mmol L-1; Spectrum), sucralose (0.4 mmol L-1; Tate & Lyle), saccharin (3.1 mmol L-1; Spectrum), neohesperidin dihydrochalcone (0.94 mmol L-1; Tokyo Chemical Industry Co.), and RebA (1.0 mmol L-1; Tokyo Chemical Industry Co.). A detailed list of concentrations is provided in Supplementary Tables S1 and S2 in the Supplementary Materials. All stimuli were prepared at least 24 h prior to testing.

Protocol

All psychophysical data were collected using Compusense five v5.2. Prior to tasting samples, participants were oriented to a horizontal gLMS by rating imagined oral and nonoral sensations (Hayes et al. 2013). The gLMS is a line scale with semantic labels placed at empirically derived intervals (barely detectable, 1.4; weak, 6; moderate, 17; strong, 35; very strong, 51; strongest imaginable sensation of any kind, 100); critically, it provides ratio level data similar to magnitude estimation (Bartoshuk et al. 2004). Results from the gLMS orientation were used to verify that participants used the scale as instructed. After orientation, a tray containing all 15 test samples in clear plastic medicine cups was passed to participants via a serving hatch. Stimuli were evaluated in semi-isolated sensory testing booths under white light. To minimize errors from tasting the wrong sample, stimuli were placed on a paper tray template with arrows that indicated the order of tasting; participants were also asked to double check that the 3-digit code on the medicine cup matched the 3-digit code shown on the computer screen. Participants were told to take the entire sample into their mouth and swirl it around the mouth for 5 s before spitting it out. A fixed 5-s interval was encouraged via software using a countdown timer visible to participants. After spitting, participants rated the intensity of sweetness and bitterness on separate gLMS scales for each quality (Allen et al. 2013; Nolden et al. 2016). The participants then rinsed thoroughly ad libitum with room temperature RO water before proceeding to the next stimulus. A minimum interstimulus interval of 45 s was enforced via software.

Modeling synergy

The 3D response surfaces in Figure 2 were graphed using Grapher for OSX. Deviations from the zero-interaction surface were quantified mathematically via Suhnel’s (Sühnel 1993) proposed index of interaction, defined as:

Figure 2.

Data summarizing responses for 3 aspartame stimuli, 3 AceK stimuli, and 9 aspartame/AceK blends. Panels A and B are 3D plots with the 2 sweeteners on the x and y axes, with perceived sweetness on the z axis, shown from 2 different angles. The spheres indicate observed means for these sweeteners in isolation or blends. The zero-interaction surface for aspartame and AceK was determined from their respective dose-response functions and is shown as the gray surface in the 3D plots. Deviations above the zero-interaction surface correspond to I-values below 1 (indicating synergy), whereas deviations below the surface correspond to I-values above 1 (indicating suppression); values on or near the surface indicate zero interaction. Panel C shows the group means and standard errors for rated sweetness for the same 15 stimuli.

$${\displaystyle \begin{array}{l}{\displaystyle \left(\mathrm{c}\mathrm{A}/\mathrm{C}\mathrm{A}\right)+\left(\mathrm{c}\mathrm{B}/\mathrm{C}\mathrm{B}\right)=I}\end{array}}$$ (1)

where cA and cB are the concentrations or doses of stimuli A and B used in the blended pair, and CA and CB are the concentrations of A and B that would individually produce the same perceived intensity in isolation. An index of interaction, I, less than 1 indicates synergy, more than 1 indicates antagonism, and around 1 indicates zero interaction (Also, we acknowledge that the use of upper and lower case to distinguish between terms in equation (1) is potentially confusing but do so nonetheless to stay consistent with Suhnel’s prior work).

For each sweetener pair, 3 separate I-values were calculated at each level of sucrose equivalents: 1 for the 75% A/25% B blend, 1 for the 50% A/50% B blend, and 1 for the 25% A/75% B blend. The mean of these 3 I-value estimates was calculated and treated as the relative I-value for that specific level of sweetness (i.e., 4%, 6%, or 8%, in sucrose equivalents). These means are shown in the bar graphs for Figures 3–7. Inferential statistics were not performed on the I-values, as it would require estimating dose-response functions for each sweetener at the individual level. Instead, variation in I-value was accounted for by considering the range 0.9 < I < 1.1 to be the zero-interaction criterion as done previously (Fleming et al. 2016).

Figure 3.

Mean I-values for AceK studies at SE4, SE6, and SE8. AceK (T1R3 VFD) was tested with fructose, aspartame (T1R2 VFD), and sucralose (T1R2/3), saccharin (T1R3 VFD), NHDC (TMD), thaumatin (T1R3 CRD–TMD), and RebA (T1R2/3 VFD). The columns represent means and bars are standard errors. The light gray horizontal bar in the middle represents the I-value range that indicates zero interaction.

Figure 7.

Mean I-values for RebA studies at SE4, SE6, and SE 8. RebA was tested with AceK (T1R3 VFD), fructose, and sucralose (T1R2 VFD), NHDC (TMD), and thaumatin (T1R3 CRD). The columns represent means and bars are standard errors. The light gray horizontal bar in the middle represents the I-value range that indicates zero interaction.

Molecular modeling of T1R2–T1R3

A homology model of human T1R2–T1R3 VFD was generated from mGluR2 (PDB ID: 5CNI); the CDR of mGluR3 (PDB ID: 5CNK) and the TMD of rhodopsin (PDB ID: 1F88) in Swiss Model (Palczewski et al. 2000; Monn et al. 2015; Waterhouse et al. 2018) were utilized to illustrate the respective domains. PDBs were imported into PyMol V2.0 and ray-traced rendered for publication (Schrodinger 2015). Sizes of sweeteners are in scale with receptors, demonstrating size considerations. Thaumatin structure was imported from (PDB ID: 1RQW) and sweet ligands SMILES were used to generate the PDB files.

Results

To quantify synergy, we determined interaction values (I-values) for every binary mixture using the concentrations tested (in mmol/L) and concentrations of single-compound solutions that would elicit the same perceived intensity. To calculate the single-compound concentrations, we determined dose-response functions for the individual sweeteners as power functions. From the dose-response functions of the individual sweeteners, we determined the zero-interaction surface (shown as the curved gray surface in Figure 2). Deviations above the zero-interaction surface correspond to I-values less than 1, indicating synergy.

To validate our approach, we first examined the well-known synergy for aspartame/AceK mixtures and simple additivity of saccharin/AceK mixtures as positive and negative controls, respectively. I-values calculated for the various aspartame/AceK pairs tested here ranged from 0.49 to 0.60, confirming previous reports of synergy (DuBois 2004). This can be seen in Figure 2 as most observed responses for aspartame/AceK mixtures fell above the zero-interaction surface. Conversely, I-values of 1 for the saccharin/AceK pairs for the high and medium concentrations (i.e., levels chosen to be roughly equivalent in sweetness to 8% and 6% sucrose (e.g., SE8 and SE6) indicate zero interaction. For the lowest concentration (SE4) saccharin/AceK pairs, the I-values were greater than 1.1 (1.49 to 1.63), indicating suppression. Finally, Figure 4 also shows that sucralose/sucrose mixtures had I-values near 1 for SE4–8 (I-values 0.96 to 1.04), indicating the absence of any interaction between the 2 sweeteners.

Figure 4.

Mean I-values for sucralose studies at SE4, SE6, and SE8. Sucralose (T1R2 VFD) was tested with AceK (T1R3 VFD), NHDC (TMD), sucrose (T1R2 VFD), thaumatin (T1R3 CRD), and RebA. The columns represent means and bars are standard errors. The light gray horizontal bar in the middle represents the I-value range that indicates zero interaction.

Next, we systematically tested AceK with a variety of other sweeteners that reportedly bind to different sites on the T1R2/T1R3 receptor. As shown in Figure 3, the fructose/AceK pairs consistently elicited responses greater than would be predicted from the individual dose-response functions. This was true for low, medium, and high concentrations (i.e., SE4, SE6, and SE8). The presence of synergy for fructose/AceK pairs is indicated by I-values below 0.8, ranging from 0.62 to 0.75. As shown in Figure 3, we also found strong evidence that AceK synergizes with both NHDC and RebA, with I-values ranging from 0.35 to 0.55 and 0.38 to 0.54, respectively. Sucralose synergizes weakly with AceK, with I-values ranging from 0.7 to 0.84. Thaumatin/AceK blends either failed to interact or showed evidence of suppression with I-values of ranging from 0.9 to 1.3.

Subsequently, we tested sucralose with various other sweeteners (Figure 4). To facilitate comparisons, AceK data are included in Figures 4–7. Our data indicate that sucralose strongly synergizes with NHDC and RebA, with I-values ranging from 0.38 to 0.66 for sucralose/NHDC and sucralose/RebA pairs, respectively. Conversely, sucralose also appears to synergize weakly with thaumatin at low concentrations (SE4 I-values = 0.8); however, SE6 and SE8 of sucralose/thaumatin blends do not appear to interact, with I-values of 0.9 and 1.1, respectively.

We then tested thaumatin with 3 additional sweeteners, as shown in Figure 5. Thaumatin/NHDC combinations consistently showed evidence of synergy. The I-values ranged from 0.3 to 0.64, with greater synergy at high concentrations. Finally, RebA did not synergize with thaumatin. The I-values for the binary combinations ranged from 0.70 in the 8% SE concentration to >5 for the blends at 4% SE. Notably, the 4% SE concentration level showed very high I-values for the combinations of RebA and thaumatin, ranging from 2.54 for the 50:50 blend to >5 for the 25:75 RebA/thaumatin blend. These I-values were the highest found in any of our studies and indicate suppression.

Figure 5.

Mean I-values for thaumatin studies at SE4, SE6, and SE8. Thaumatin (T1R3 CRD) was tested with AceK (T1R3 VFD), NHDC (TMD), sucralose (T1R2 VFD), and RebA. The columns represent means and bars are standard errors. The light gray horizontal bar in the middle represents the I-value range that indicates zero interaction.

Next, we tested NHDC pairwise with sucralose, thaumatin, and RebA. Notably, NHDC appears to synergize well with all 4 of the sweeteners shown in Figure 6. The different NHDC combinations resulted in consistently low I-values: 0.35 to 0.55 for NHDC/AceK blends, 0.39 to 0.66 for NHDC/sucralose blends, 0.31 to 0.67 for NHDC/thaumatin blends, and 0.21 to 0.59 for NHDC/RebA blends. These I-values were among the lowest found across our studies.

Figure 6.

Mean I-values for NHDC studies at SE4, SE6, and SE8. NHDC (TMD) was tested with AceK (T1R3 VFD), sucralose (T1R2 VFD), thaumatin (T1R3 CRD), and RebA. The columns represent means and bars are standard errors. The light gray horizontal bar in the middle represents the I-value range that indicates zero interaction.

In our last series of experiments, shown in Figure 7, we tested RebA pairwise with a series of other sweet ligands. We did so for 2 reasons: first, there is a strong commercial demand for natural NNSs and, second, the binding domains for RebA have yet to be experimentally determined (Nuemket et al. 2017). As noted above, RebA synergized with both AceK and sucralose, with I-values ranging from 0.38 to 0.54 and 0.38 to 0.72, respectively. Notably, RebA also synergized with fructose (see Figure 7), with mean I-values ranging from 0.66 in the highest concentration to 0.84 in the lowest concentration. For fructose/RebA pairs, the I-values were lower in the 8% SE level, suggesting there may have been more synergy at higher concentrations. This contrasts with the NHDC/sucralose data showing greater synergy at lower concentrations. Meanwhile, RebA synergized strongly with NHDC at all concentration levels, with I-values ranging from 0.21 to 0.58. Again, as with the fructose/RebA pairs, greater synergy was found in the higher concentrations for RebA/NHDC mixtures. Finally, RebA did not synergize with thaumatin. As mentioned previously, the RebA/thaumatin blends in the low and medium concentrations were among the highest I-values found in our study.

Table I provides a summary of sweetener combinations, sorted in groups showing full synergy (I-values <0.6), partial synergy (I-values of 0.9–0.6), zero interaction (I-values 0.9–1.1), and suppression (I-values >1.0) based on calculations from psychophysical ratings in the full set of different experiments shown in Figures 3–7.

Discussion

Here, we systematically tested 15 binary pairs of sweeteners in a large cohort of volunteers (n = 1576) using contemporary psychophysical methods, and analyzed these data using the isobole method to quantify potential synergy. Below, these results are compared with prior psychophysical literature and current models of sweet taste receptor putative binding sites. All compounds examined here are intrinsically sweet, implying that sweet taste, by nature, has multiple mechanisms of activation.

Previous studies of sweetener synergy have assumed a linear dose response or attempted to address this concern utilizing self-mixtures (Frank et al. 1989). One hallmark of receptor activity is the dose-response function, which generally forms a rectangular hyperbola. Interpreted from a receptor occupancy model, low to high concentration of a compound will have a linear response function, a transition phase, and a maximal signal as the receptor sites are saturated. Prior data suggests that sweeteners have dose-response functions that can be saturated at high concentrations (Ayya and Lawless 1992; Antenucci and Hayes 2015).

As noted in the introduction, putative T1Rs binding sites have been investigated using a wide range of methods across multiple species. Figure 1 illustrates sweeteners that are positioned near their putative binding domains, although it should also be cautioned that some sweet agonists appear to be dual binding—that is, they may bind to more than 1 region. In general, while a host of possible receptor mechanisms have been described, agonists bind to orthosteric sites, activate receptors, and are competitive in mixtures. In contrast, positive or negative allosteric modulators have no intrinsic activity by themselves as they require an agonist to be present; in mixtures with an agonist, they can enhance or diminish receptor responses to said agonist. So-called allosteric agonists have both intrinsic activity but can also have positive or negative effects on receptor activation (Monn et al. 2015). Lastly, agonist–antagonists have intrinsic activity alone but can inhibit another agonist (Monn et al. 2015).

Positive and negative controls

The I-values obtained here for AceK/aspartame mixtures show highly synergistic responses (Figure 3; Table 1), in agreement with numerous prior studies. These studies have consistently found synergy for this pair of sweeteners, irrespective of the methodology or food matrix used: 21-point category scale (Frank et al. 1989), 15-cm line scale (Schiffman et al. 1995), time intensity, magnitude estimation (Schifferstein 1995), 9-cm structured line scale with in a peach nectar juice drink (Melo et al. 2013), and 50-point composite scorecard in a lassi drink (George et al. 2010). Likewise, the sustained commercial success of global soft drink brands containing AceK/Aspartame blends further support this conclusion. As a negative control, we also tested AceK and saccharin (Figure 3; Table 1). Our analysis of these data failed to show any evidence of synergy, in agreement with prior work (Frank et al. 1989; Ayya and Lawless 1992; Schiffman et al. 1995). Indeed, the suppression observed here for the least sweet AceK/saccharin blend is very similar to the finding from Ayya and Lawless (1992), who also observed subadditivity (i.e., suppression) for their least sweet AceK/xaccharin mixture. Elsewhere, Schiffman et al. (1995) reported suppression AceK/saccharin blends (albeit for all 3 levels they tested). Collectively, these controls suggest the utility of the isobole method as applied here.

Table 1.

Summary tables of high and partial synergy, no interaction, and suppression of the sweetener pairs

Full synergy (I-value = <0.6)
Sweetener 1	Sweetener 2
AceK	Aspartame
NHDC	AceK
NHDC	Sucralose
NHDC	Thaumatin
NHDC	Reb A
Reb A	Sucralose
Reb A	AceK
Partial synergy (I-value = 0.9–0.6)
Sweetener 1	Sweetener 2
AceK	Sucralose
AceK	Fructose
Reb A	Fructose
Zero interaction (I-value = 0.9–1.1)
Sweetener 1	Sweetener 2
Sucralose	Thaumatin
Sucralose	Sucrose
Suppression (I-value >1.0)
Sweetener 1	Sweetener 2
AceK	Saccharin
AceK	Thaumatin
Reb A	Thaumatin

Several in vitro cell-based studies have been dedicated to mapping the binding sites of T1R2/3; however, to our knowledge, binary mixtures of sweeteners are generally underrepresented (Morini et al. 2005; Delay et al. 2006; Winnig et al. 2007; Ohta et al. 2011; Maillet et al. 2015; Assadi-Porter et al. 2018). We sought to compare current models of receptor activation to our binary mixture results. Although the crystal structures of mammalian T1Rs have not been published, several C-class GPCRs have been solved to with and without bound ligands (Wu et al. 2014). The VDF of T1R2/3 and mGLuR2 have sequence identity of 27% and 29%, respectively (Cheron et al. 2017). Homology models constructed in silico from these structures can be useful in generating hypotheses (Figure 1). Masuda et al. and Maillet et al. showed that point mutations in 11 amino acids residing near the N-terminus of the T1R2 VFD abolished aspartame activity and mutation of 3 amino acids in T1R2 VFD abolished responses to AceK and saccharin (Masuda et al. 2012; Maillet et al. 2015). T1R2 amino acid residue D142 overlaps with AceK, saccharin, and aspartame and is thought to reside deep within the VFD pocket. Residues E382 and R383, important for AceK and saccharin, would be located near the opening face of the VFD (Masuda et al. 2012). Based on these reports, one hypothesis would be that that the psychophysical synergy observed for AceK and aspartame would predict dual binding to T1R2 VFD or allosteric agonism. Conversely, AceK and saccharin would presumably be agonists competing for the same single-binding site, in agreement with the zero interaction seen here. Notably, novel dual-binding ingredients of T1R2 VFD could be a useful commercial strategy to achieve further sugar reduction.

Highly synergistic mixtures

Highly synergistic sweetener combinations are of interest to the food and beverage industries as they reduce cost and facilitate sugar reduction while improving sensory profiles. As shown in Table 1, of 15 binary mixtures, 7 demonstrated full synergy, with I-values below >0.6. This includes AceK/aspartame blends, which were already discussed above. For NHDC, putative binding has been mapped to the T1R3–TMD and it has intrinsic sweetness (i.e., it is not merely a modulator) (Winnig et al. 2007). Here, AceK/NHDC mixtures were found to synergize in agreement with psychophysical data from Schiffman et al. as well as in vitro data from Fujiwara et al. looking at activity in response to singular and binary sweetener solutions (Schiffman et al. 1979; Fujiwara et al. 2012). Also, NHDC was highly synergistic with sucralose, thaumatin, and RebA (Table 1). Interestingly, Fujiwara et al. proposed a mechanism for synergy involving the TMD-binding site (Fujiwara et al. 2012). They noted that the literature shows binary combinations of either NHDC or cyclamate as one of the combinations resulting in “over additive enhancement of the perceived sweetness,” proposing that binding to the TMD site results in positive allosteric modulation of the sweet receptor, thereby reducing the energy of activation required for sweetener binding to the extracellular sites. NHDC binding has been mapped to the TMD of T1R3 (Winnig et al. 2007).

Sucralose/NHDC mixtures also revealed full synergy. Nuclear magnetic resonance and circular dichroism studies on individually expressed VFDs have shown that sucralose and sucrose bind T1R2 and T1R3; however, the observed Kd values are an order of magnitude less than expected from psychophysical results (Nie et al. 2005; Assadi-Porter et al. 2018). This discrepancy maybe due to cross-subunit interactions or removal of the CDR and TMD. Recently, T1R2 and T1R3, possibly as a homodimer, have been suggested to act as a glucose sensor in the pancreas and other tissue (Laffitte et al. 2014). Because glucose is ~5 mM in blood, a concentration that is not perceptibly sweet, the possibility exists that T1R2–T1R3 together reduces sweetener potency. Evolutionarily, it would be advantageous to detect high levels of nutrients in the environment as opposed to highly sensitive detection that would fail to meet nutritional needs.

Thaumatin is one of the largest proteins (22.4 kDa) to demonstrate intrinsic sweetness, and it is believed to interact with the CRD and VFD-LB2 of T1R3 (see Figure 1). Our results show that thaumatin demonstrated synergy with only NHDC (Table 1), which agrees with NHDC having general PAM properties; however, more precisely, allosteric modulators that have intrinsic activity are termed allosteric agonists (Schwartz and Holst 2007). Results here provide a framework where TMD activation by NHDC alone bypasses VFD/CRD activation and suggests that it enhances activity of T1R2/3 sweeteners in mixtures.

RebA is a natural stevia glycoside sweetener with high commercial relevance; however, other negative sensory attributes have limited its use somewhat. Homology models of medaka fish T1R2/3 with rebaudiosides suggest RebA may be a dual-binding sweetener. However, experimental evidence for the mechanism of RebA interaction with T1R2/3 is unclear (Nuemket et al. 2017). Here, RebA showed a high level of synergy with the synthetic sweeteners AceK, NHDC, and sucralose (Table 1). The present data showing RebA/AceK exhibit synergy conflicts with data from Schiffman et al (1995). Also, RebA/NHDC pairs were also highly synergistic here, consistent with NHDC acting as a general sweetness enhancer (Benavente-García et al. 2001). RebA/sucralose were also highly synergetic: this is interesting due to the putative dual binding of sucralose to T1R2/3. This may suggest dual binding to T1R2 and T1R3 but such predictions have not been experimentally validated.

Partially synergistic mixtures

Of 15 mixtures, blends of fructose/AceK, fructose/RebA, and AceK/sucralose were observed to have partial synergy (I-value 0.9–0.6) with no observed dose effect across SE4–8 (Table 1). Notably, whole nerve recordings from the chorda tympani in T1R3 knockout mouse model demonstrated significant residual activity with sugars but not NNSs, suggesting the possible existence of T1R-independent pathways (Damak et al. 2003). Thus, the partial synergy observed may be due to activation of multiple pathways; however, our AceK/sucralose data is inconsistent with this hypothesis. Alternatively, these sweeteners may compete at T1R bind sites or partially activate the T1R2–T1R3 complex.

No interaction mixtures

Of 15 mixtures tested, only 2—sucrose/sucralose and thaumatin/sucralose—were observed to have zero interaction (I-value 0.9–1.1). Our sucralose/sucrose results conflict with the findings of Choi and Chung; however, they studied the sweetener pair in milk and coffee and noted that their results differed from work in water (Choi and Chung 2015). In particular, milk contains lactose and coffee elicits bitterness and/or sourness that may have resulted in mixture suppression with the sweeteners used in their study. Beyond perceptual effects from bitterness (i.e., the potential for mixture suppression), caffeine is also in the methoxanthine class of CNS stimulants, which can modulate adenosine receptors and can be transported across cell membranes to modulate multiple signal transduction pathways (Stephenson 1977), which may further explain the discrepancy with the data of Choi and Chung. Likewise, Schiffman et al. have shown in vivo effects of caffeine on sweetness (Schiffman et al. 1986). Sucralose and sucrose are structurally similar and are thought to interact with the VFD of both T1R2 and T1R3, and may be considered dual binding (Nie et al. 2005; Assadi-Porter et al. 2018). In summary, the sucrose/sucralose blend showed no interaction here, consistent with their being competitive agonists. Thaumatin/sucralose also demonstrated no interaction (see below for discussion).

Suppression

Saccharin/AceK, thaumatin/AceK, and thaumatin/RebA mixtures demonstrated a dose response with decreasing suppression (I-value > 1.1) from SE4–8 (Table 1). Previous work suggests synergy may be stronger at lower concentrations (Curtis et al. 1984; Schiet and Frijters 1988; Schiffman et al. 1995), whereas our results showed the opposite effect. These relationships are consistent with a competitive binding model where, at high ligand concentrations, receptors become saturated, overcoming inhibition. As discussed above, saccharin/AceK are thought to bind to the orthosteric site on T1R3 and thaumatin’s potential to block AceK are consistent with a competition model. Thaumatin-RebA pairs showed the highest suppression, suggesting RebA binds to T1R3. These differences may also result from using a more appropriate approach to model synergy, specifically the isobole method (Fleming et al. 2016; Wang et al. 2018). The approach used here incorporates the individual dose-response functions of each sweetener, whereas previous studies may fail to account for the nonlinear dose-response functions commonly seen in NNSs (Frank et al. 1989). Because thaumatin is thought to function by interacting with CRD and LB2 of T1R3, an intriguing hypothesis is that thaumatin wedges LB2 and LB1 of T1R3 into a closed active conformation but, in doing so, blocks AceK and RebA access (Table 1). Alternatively, thaumatin may also affect T1R2 via cross subunit interaction.

Limitations and conclusions

By definition, the psychophysical ratings obtained here are not simple measures of receptor activation in vivo: rather, they represent the sum of receptor activation, central integration with other chemosensory signals, and judgment processes and potential response biases inherent to the rating process (Lawless 2014). Nonetheless, psychophysical ratings are generally presumed to be roughly reflective of differential receptor activation (e.g., Wooding et al. 2010; Hayes et al. 2015) and peripheral neural events (e.g., Borg et al. 1967). Some of the data described here are presumably influenced by central processing of taste signals, like mixture suppression (Lawless 1977, 1979). This phenomenon would seem to be especially relevant as several of the sweeteners tested here can have notable side tastes (e.g., Allen et al. 2013; Antenucci and Hayes 2015). Indeed, the suppression seen here for some saccharin, AceK, RebA, and thaumatin mixtures might not be purely attributable to competition for various T1R2/T12R3 binding sites. For the AceK/saccharin blends, both ligands activate one variant of the TAS2R31 bitter receptor (e.g., Roudnitzky et al. 2011). When mixed, their collective bitterness may reduce sweetness centrally (Lawless 1979) in some individuals but not others and potentially cause some of the suppression seen here, irrespective of any competition for T1R3 binding sites. Likewise, thaumatin and RebA each have a licorice note (Reyes et al. 2017); whether this might cause some suppression of sweetness when the 2 sweeteners are mixed is unknown. Additional research is needed to determine how other taste or cross-modal inputs (e.g., Wang et al. 2019) may modify sweetness synergy. Here, extensive and systematic psychophysical studies with 1576 volunteers were used to test binary combinations of ligands described as sweet. These data generally are consistent with a modified version of the binding site hypothesis: ligands that bind to different sites on the sweet receptor will show synergy, whereas those that act upon a common or overlapping binding site on the same subunit will not. Although sucralose and RebA appear to have dual-binding characteristics, present results are consistent with sucralose preferentially binding T1R3, whereas RebA preferentially binds T1R2. This hypothesis, in conjunction with present psychophysical data, also provides novel mechanistic predictions with regard to binding sites. Additional work is needed to apply the isobole method in more complex food matrices beyond water (e.g., Wang et al. 2018). By studying binary mixtures of sweeteners that bind to different sites on the sweet receptor, we tested a biologically informed model of sweetener synergy. Present data generally support the hypothesis that structurally similar sweeteners (e.g. AceK/saccharin) acting as agonists will not synergize, whereas structurally dissimilar sweeteners binding to overlapping or distal sites can act as allosteric agonists (e.g. NHDC) or agonist/antagonists (e.g., thaumatin), respectively.

Funding

This work was supported in part by a Bunton–Waller scholarship from the Pennsylvania State University to M.M.R. and discretionary funds from the Pennsylvania State University controlled by J.E.H. Additional support was provided by a grant from the National Institutes of Health [R03DC010904] to J.E.H., and the United States Department of Agriculture and the National Institute of Food and Agriculture via Hatch Act Appropriations [Projects PEN04565 and PEN04332]. None of these organizations have had any role in study conception, design or interpretation, or the decision to publish these data.

Conflict of interest

M.M.R. has no conflicts to disclose. J.E.H. has received speaking and consulting fees from nonprofit organizations, federal agencies, and corporate clients in the food industry. Additionally, the Sensory Evaluation Center at Penn State routinely conducts taste tests for the food industry to facilitate experiential learning for students. None of these organizations have had any role in study conception, design or interpretation, or the decision to publish these data. S.A.G. is the sole proprietor of Radio Taste LLC, a for-profit biotechnology company. No compensation was provided for his work on this project.