Molecular mechanism of the allosteric enhancement of the umami taste sensation


H. Khandelia, MEMPHYS – Center for Biomembrane Physics, University of Southern Denmark, Odense, Denmark
Fax: +45 65504048
Tel: +45 65503510


The fifth taste quality, umami, arises from binding of glutamate to the umami receptor T1R1/T1R3. The umami taste is enhanced several-fold upon addition of free nucleotides such as guanosine-5′-monophosphate (GMP) to glutamate-containing food. GMP may operate via binding to the ligand-binding domain of the T1R1 part of the umami receptor at an allosteric site. Using molecular dynamics simulations, we show that GMP can stabilize the closed (active) state of T1R1 by binding to the outer vestibule of the so-called Venus flytrap domain of the receptor. The transition between the closed and open conformations was accessed in the simulations. Using principal component analysis, we show that the dynamics of the Venus flytrap domain along the hinge-bending motion that activates signaling is dampened significantly upon binding of glutamate, and further slows down upon binding of GMP at an allosteric site, thus suggesting a molecular mechanism of cooperativity between GMP and glutamate.


G-protein-coupled receptor


molecular dynamics


metabotropic glutamate receptor


principal component analysis


Venus flytrap domain


Umami (the essence of deliciousness) was first suggested as the fifth basic taste quality, in addition to salt, sour, sweet and bitter, by Ikeda in 1909 [1]. However, it is only in the last decade, after the discovery of specific umami receptors located in the taste-cell membranes, that umami has been fully recognized as a basic taste quality in a physiological sense [2,3]. In 2000, the first umami receptor, taste-mGluR4, was discovered [4]. Taste-mGluR4 is a metabotropic glutamate receptor, a special dimeric G-protein-coupled receptor (GPCR) [5] located in the membranes of the taste cells in the taste buds. Taste-mGluR4 is a truncated version of the well-known glutamate receptor mGluR4 in the brain, and is selectively sensitive to l-glutamate. Two other umami receptors were subsequently found, T1R1/T1R3 [3,6] and a special mGlu receptor [7] that is related to the brain glutamate receptor mGluR1. It remains unclear whether the various umami receptors use different signaling pathways [8,9].

Humans recognize the umami taste as an indicator of accessible protein in food, and the taste sensation is triggered by free amino acids, in particular by binding of the l-glutamate ion to umami taste receptors in the tongue. Glutamate (Glu) is the most abundant amino acid in human breast milk [10], suggesting that the new-born child is primed early for umami sensing. The unique aspect of the umami taste is that it can be enhanced several-fold by the presence of free nucleotides such as inosine-5′-monophosphate (IMP), guanosine-5′-monophosphate (GMP) and adenosine-5′-monophosphate (AMP). Both free glutamate and free nucleotides are present in a variety of foods [11].

The synergy between Glu and nucleotides in enhancing the umami taste sensation is fascinating from the molecular point of view. As T1R1/T1R3 is the only umami receptor that is sensitive to both Glu and free nucleotides, it is the focus of this paper. Similar to the receptors for bitter and sweet, T1R1/T1R3 is a GPCR [12]. The sweet and umami taste receptors belong to the class C GPCR family of proteins, which have seven transmembrane segments [12]. Most class C GPCRs are homodimers, but the sweet and umami receptors are heterodimers, specifically T1R1/T1R3 [6] and T1R2/T1R3 [3,13]. As the T1R3 monomer is common to the sweet and umami taste receptors, the T1R1 monomer is likely to be critical for sensation of the umami taste. Like the highly homologous metabotropic glutamate receptors (mGluRs), the monomer of T1R1 consists of an extracellular Venus flytrap domain (VFTD), a membrane-inserted helical segment and an extracellular cysteine-rich domain connecting the two. The VFTD is identified as the ligand-binding domain in homologous proteins such as mGluRs [5]. In mGluRs, the VFTD consists of two lobes, which can remain open or close together in the open (inactive) or closed (active) conformations of the protein [5,14–16]. Glu stabilizes the closed conformation, which activates the downstream signaling pathway. Closure of the VFTD is therefore the key event that sensitizes the umami taste receptor.

It was recently suggested that 5′-ribonucleotides such as IMP and GMP significantly enhance the sensitivity of the umami receptor to glutamate by binding to the outer cleft of the VFTD [17]. The evidence for this proposed mechanism was based on the following observations. First, using chimeric T1R receptors, it was established that the T1R1 and the T1R2 receptors are responsible for ligand binding via the VFTDs for the umami and sweet receptors, respectively, and the same domain was responsible for the enhancing effects of IMP and GMP. Binding free energy calculations from Molecular dynamics (MD) simulations of homology models of T1R1 revealed that glutamate binding was very favorable at T1R1 (ΔGbinding approximately −19 kT), and occurred only transiently on T1R3 (ΔGbinding approximately −kT), where kT is thermal energy. [18]. Second, mutagenesis of residues that are conserved between mGluR1 and T1R1 and that lie close to the outer cleft of the VFTD resulted in a loss of the umami enhancement effect of GMP. Mutagenesis of glutamate-binding residues did not alter the enhancement effect [17].

Here, we take a detailed look at the dynamics on the atomic scale for the VFTD of the T1R1 monomer in the presence and absence of the two ligands Glu and GMP, in order to unravel the molecular mechanism behind the synergistic effect of the two ligands on the dynamics of the receptor. The goal of the simulations is to obtain molecular insights into the sensation of the umami taste. Sweet taste receptors may also work by a similar allosteric synergistic mechanism [19], and a deeper insight into the allosteric mechanism at the receptor level may lead to new therapies against diabetes.

Although ionotropic glutamate receptors have been investigated previously by MD simulations [20], only a short 2 ns simulation of the mGluR ligand-binding domain has been reported previously [21]. The only previous simulation study of the umami receptor [18] focused on protein–glutamate binding interactions and energies. The current study focuses on the allosteric stabilization of the active state of the protein by GMP, and hence the very core of the well-known synergy in the umami taste.

We found that the VFTD of T1R1 is a highly dynamic moiety, and both the closed and open conformational states are captured in the simulations. GMP is found to stabilize the active (closed) state, and the mechanism of stabilization of the closed conformation by GMP is explained.


Dynamics of the T1R1 VFTD

Two simulations each for the glutamate-free form (NOGlu1 and NOGlu2) and the glutamate-bound form (Glu1 and Glu2) were implemented, and four simulations for the glutamate-bound protein with the nucleotide GMP (GMPGlu1–GMPGlu4) were implemented. Interactive visual inspection indicated that the VFTD of T1R1 is a highly dynamic moiety, in particular when no ligand was bound (NOGlu simulations). Closure of the VFTD was measured by the angle formed between lines joining the Glu-binding site and the pincer residues H308 and H71. H308 lies on the upper lobe, close to the purported GMP-binding site, and H71 lies on the lower lobe at the same site. The angle is 64° in the open conformation and 35° in the closed conformation. As shown in Fig. 1, the angle approaches 35° during the course of the NOGlu1 simulation, and the VFTD closes in the absence of a ligand. In NOGlu2, the protein remains in the open conformation, and accesses conformations in which the two lobes are further apart than in the crystal structures. Addition of glutamate drives the protein towards a closed conformation in both Glu simulations. Addition of the second ligand GMP further drove the angle to values below 35° in two of the four GMPGlu simulations.

Figure 1.

 (A) Angle between the two lobes of the VFTD. A five-point average is shown. The horizontal lines represent values for the homology models for the closed (35°) and open (64°) conformations of the VFTD. For clarity, data for only one each of the Glu and GMPGlu simulations are shown. In one of the two NOGlu simulations (NOGlu2), only the open conformation is accessed, while in the other, the VFTD closes spontaneously. Glu1 closes, and the closed conformation is stabilized further in GMPGlu1. (B) Ensemble-averaged values for the angles reported in (A). Note that the closed state is further stabilized in two of four GMPGlu simulations.

The radius of the binding vestibule of the VFTD leading to the glutamate-binding site was measured using the program HOLE [22]. Fig. 2 shows snapshots from the Glu and GMPGlu simulations, showing that the aperture is further closed when GMP is present, suggesting stabilization of the closed (active) conformation upon binding of GMP.

Figure 2.

 The vestibule leading to the GMP-binding site and thereafter the glutamate-binding site is color-coded based on to the radius of the largest-fitting spherical probe (blue, wide; green, medium; red narrow), calculated using the program HOLE [22]. The simulation snapshots are taken from the final conformations of the Glu1 and GMPGlu1 simulations. The radius of the vestibule is significantly reduced (red) in GMPGlu1 close to the GMP-binding site.

Principal component analysis (PCA) identified the essential degrees of freedom of the protein. Analysis of the eigenvalues and eigenvectors of the covariance matrices of the NOGlu, Glu and GMPGlu simulations revealed that the largest eigenvalue corresponded to the motion that opens and closes the VFTD. The displacement along the 1st eigenvector, corresponding to the square root of the first eigenvalue, was highest in NOGlu1, was reduced upon addition of Glu in Glu1, and was further reduced upon addition of GMP in GMPGlu1 (Fig. 3). Thus, Glu and GMP clearly restrict the motion of the protein along the protein’s most essential degree of freedom: opening and closing of the VFTD. Furthermore, Glu and GMP are synergistic in their impact on the conformational equilibrium of the VFTD. Fig. 4 shows snapshots of the NOGlu1 simulation projected upon the first eigenvector, and it is clear that the protein spans the entire range of conformations ranging from an open VFTD to a completely closed domain, which corresponds to the active protein conformation. An animation of the protein motion along the first eigenvector is shown in Movies S1, S2 and S3 for the NOGlu1, Glu1 and GMPGlu1 simulations, respectively. See Figs S2-S4 for the starting snapshots for the movies. The projected trajectories also access conformations of the VFTD that have a much wider entrance to the Glu-binding site than the open conformation in the crystal structures [5,15,16].

Figure 3.

 Eigenvalues for the covariance matrix of backbone atoms of the protein from simulation snapshots between = 50 ns and = 100 ns. The displacement along eigenvalue 1, corresponding to the hinge-bending motion, which is largest without any ligand, is reduced upon binding of glutamate, and further reduced upon binding of GMP in two GMPGlu simulations.

Figure 4.

 Interpolation snapshots between the two extreme projections of the backbone simulation trajectory along the first eigenvector in the NOGlu1 simulation. Six interpolated snapshots are shown. The figure represents the extremes of the conformations that were accessible in the simulations. The red and blue colors correspond to the protein regions comprising the upper and lower VFTD lobes, respectively. Residues H71 and H308 (shown as red balls) move closer together as the lobes close. Residue Y220 (blue and cyan balls) represents the region near the glutamate-binding site.

Interactions of Glu with T1R1

In the Glu simulations, the main residues coordinating the glutamate residue are A170, R329, S172 and T149. There were also interactions with R277 in some simulations. Most of the protein–glutamate interactions were similar to those reported previously [18]. However, the γ-carboxylate of glutamate interacted strongly with R329 in all simulations reported here.

Interactions of GMP with T1R1

The initial placement of GMP in our simulations was based on mutagenesis data from Zhang et al. [17]. After the initial 10 ns restrained dynamics, the GMP molecule remained stably bound near the opening of the VFTD vestibule between the two lobes. In the four GMPGlu simulations, the GMP molecules adopted a variety of conformations. In three instances, the GMP remained oriented in the vestibule with its guanosine base pointing into the glutamate-binding site and its phosphate pointing out towards the vestibule entry. However, in the fourth instance, the phosphate was closer to the glutamate, while the base pointed towards the opening of the vestibule. The orientation of GMP was contingent upon its initial placement in the vestibule. As a result of its variable position in the vestibule, GMP interacted with a large number of residues. Fig. 5 shows radial distribution functions between GMP and protein residues. In most cases, the phosphate group was stabilized by the side chains of K328 and H308 and the backbone of A330. H308 also interacted with the GMP base by means of a stacking interaction. Surprisingly, we did not find any significant interactions between GMP and H71, which was postulated to be one of the pincer residues that bind GMP [17]. S306 and R307 also bind to the phosphate group, strongly in some instances. However, the phosphate group of GMP remains highly mobile, although bound to the protein. In a frequently occurring motif, GMP binds to the helix between residues 300 and 310 on the upper lobe of the VFTD: the residues A302 (via the base), S306 (via the phosphate group) and H308 (via stacking with the base and interactions with the phosphate group) interact with GMP. On the lower lobe, the loop between residues 378 to 386 sandwiches the GMP in the vestibule. S382 and S385 often participate in donating hydrogen bonds to the GMP base.

Figure 5.

 (A) Radial distribution functions between protein residues and various parts of GMP in the GMPGlu1 simulation. Only heavy atoms of GMP were used. A five-point average is shown. (B) Simulation snapshot showing the interaction of GMP with the upper and lower lobes of the VFTD. The glutamate-binding site is to the right. Helix 300–310 (green) lies on the upper lobe, and residues 378–386 (purple) lie on the lower lobe. Inset: interaction of GMP with the protein in GMPGlu2. A frequently occurring motif is the stacking interaction between H308 and the GMP base.


The synergy between nucleotides (GMP in the present study) and the ligand glutamate in controlling the dynamics of the umami receptor T1R1/T1R3 has been investigated using MD simulations. Binding of GMP to the opening of the vestibule of ligand-binding VFTD on the T1R1 unit further stabilized the closed (active) conformation of the protein in at least 50% of the simulations implemented with GMP. Analysis of the principal dynamic modes of the protein confirmed that the displacement (proportional to the eigenvalues) along the opening and closure of the VFTD is reduced upon binding of GMP to the VFTD. It is further interesting to note that, in some simulations with GMP, the first principal mode of protein motion is not closure and opening of the VFTD, but a twisting motion around an axis different from the VFTD hinge axis, suggesting that the movement of the VFTD that switches the protein between active and inactive conformations is significantly dampened by addition of GMP. The PCA thus confirms that binding of GMP drives the conformational equilibrium of the VFTD towards the active (closed) conformation.

In simulations without any bound ligand, the protein accesses conformations where the two lobes of the VFTD are separated even further than observed in crystal structures of mGluR1s [5]. We cannot exclude the possibility that the excess conformational freedom may be a consequence of the simulation set-up, in which neither the cysteine-rich region nor the other monomer T1R3 was simulated. It is possible that these structural features may exert restraints on the amplitude of VFTD motion. On the other hand, it is possible, as suggested by the present simulations, that the VFTD of the full umami receptor is very flexible, perhaps more so than the mGluR1 family of proteins. The two NoGlu simulations diverged from each other. In NoGlu1, the VFTD closes, but in NoGlu2 it does not. The difference underlines the conformational flexibility of the domain in the absence of glutamate. If simulated long enough, both the simulations will access both the open and closed conformations. Both NoGlu1 and NoGlu2 are ‘correct’ simulations in the sense that they access expected conformations of the protein. The differences between the two simulations stress the importance of running multiple copies of simulations, particularly when the time scale of the phenomenon of interest (in this case opening and closing of the VFTD) occurs on unknown time scales or longer time scales. Even without glutamate, NoGlu1 can adopt conformations similar to Glu1 (Fig. 1), suggesting that even without glutamate, the free protein accesses the glutamate-bound conformations. Glutamate stabilizes the bound state of the protein, and dampens the hinge-bending motion, as seen in the eigenvalue calculations. A similar conclusion was drawn from a detailed conformational analysis of ubiquitin, in which long time-scale simulations of free ubiquitin accessed the conformational heterogeneity of 46 different crystal structures of the protein, in many of which the protein was bound to other proteins or ligands [23].

Comparison with experimental data and testable predictions

The single most important phenomenon explained by the simulations is that the presence of nucleotides enhances the sensation of the umami taste. The simulations show that the presence of GMP further stabilizes the closed (active) state of the protein once glutamate is bound. An inherent assumption in this analysis is that GMP binds after glutamate.

As GMP binds at the entrance of the binding cleft, it is very likely that binding of GMP will be enhanced if glutamate is already bound, because the upper and lower lobes of the VFTD will be closer, and will be optimally positioned for GMP to bind. Furthermore, if GMP bound first, it would sterically block the entrance to the VFTD vestibule, and prevent access of glutamate to the glutamate-binding site. Thus, it may be that, when both ligands are available, GMP binding after glutamate is more likely than the other way around. GMP still enhances the response of mutants where the glutamate-binding site has been compromised, showing that, in itself, GMP is able to activate the receptor [17]. Further mutational analysis comparing the rate of activation of the receptor by GMP in the presence and absence of glutamate-binding should be able to confirm the sequential binding hypothesis, as well as the prediction that GMP binds at the entrance of the vestibule.

Loop motions in proteins are some of the slowest modes of motion, and often loops move on time scales of micro- or milliseconds. For the umami receptor, this has been confirmed by NMR, which suggests that the closed form of the receptor is stabilized over the time scale of milliseconds [24]. In this respect, we have been fortunate that, in the current set of simulations, we were able to capture the rare event of conformational change between open and closed states. The correlation times of the loop motions in the current simulations are of the order of 1000 ps, and therefore the simulations are long enough to capture the rare event, perhaps because the region in phase space corresponding to the conformational change was accessed sooner rather than later. In any event, PCA is useful for rare event simulations because it helps in extraction of the primary modes of motion, even if the rare event is inaccessible.

We did not observe any direct interaction of H71 with GMP, which is apparently in conflict with mutational data suggesting that H71 mutants do not respond well to signal enhancement by GMP [17]. It is possible that the H71 mutants may have longer-range effects that stabilize the closed state of the protein, but this remains to be clarified. On the other hand, H308 interacts strongly with the phosphate group of GMP in the simulations, and such an interaction should be enhanced if H308 is replaced by a negatively charged residue. The H308–GMP interaction also brings the two lobes of the VFTD together, and therefore stabilizes the active state. Indeed, the H308E mutant was shown to be more responsive to GMP than the wild-type [17]. Residue K328 also interacts strongly with the GMP molecule, and we predict that replacement of K328 by a negatively charged residue should result in a protein that is less active than the wild-type. Another important cationic residue is R329. R329 interacts with the side chain carboxylate of glutamate in the Glu1 and Glu2 simulations. However, once GMP is also introduced, these interactions are lost, because R329 is now partially sequestered towards GMP by the GMP sugar and the phosphate group. As in prior simulations [18], glutamate in our simulations is coordinated by S172, another potential target for testing that had not been previously implicated in glutamate-binding in the umami receptor.

The simulations reported in this paper provide quantitative support for the proposed putative mechanism [17] behind the ubiquitous synergy effect in umami sensation via a detailed and well-defined molecular model. The revelation of an allosteric molecular mechanism at the receptor level behind the synergy in the umami taste holds promise for the design of novel compounds for controlling umami flavoring of foodstuffs, and for investigation of similar allosteric mechanisms in other GPCRs such as sweet taste receptors, providing insight into fundamental allostery in medically important proteins.

Experimental procedures

Homology modeling

The structure of the T1R1 extracellular VFTD without the cysteine-rich region was obtained by homology modeling. The cysteine-rich region was excluded from the model, because it is unlikely to have a direct impact on the binding of Glu or GMP and the dynamics of the VFTD. The amino acid sequences of T1R1 (Uniprot ID Q7RTX1) and mGluR1 (Uniprot ID P23385) were aligned using MAFFT [25]. The sequence alignment is shown in Fig. S1. The sequence identity between the two proteins is 26.3%. However, both proteins have the same glutamate-binding domain, which is a highly conserved structure in eukaryotes. Previous simulations, in which a similar homology model was used to accurately predict glutamate-binding free energies and glutamate-binding sites in the umami receptor [18], confirm the reliability of the homology model. The resulting alignment was fed into MODELLER [26] for homology modeling. The glutamate-free open conformation of mGluR1 was used as a template for modeling the coordinates of the corresponding open conformation of T1R1. The structure of the mGluR1 template was obtained from PDB ID 1EWK, which contains coordinates for the complex form of the dimer, with one monomer in the closed conformation and the other in the open conformation. Coordinates of residues missing in the crystal structures were inserted as random loops using MODELLER. Loop motions may occur over very long time scales (> μs). However, in this case, the loops, which control access to the binding sites, access a wide range of conformations within 100s of nanoseconds. Thus, the results should be independent of the modeled loops. The loops constructed where structural template was missing are distant from the binding sites, and from the VFTD vestibule, and are not expected to affect the results. The ligand glutamate was retained in the structural alignment. Cysteine bridges between residues C67/C109 and C432/C439 in mGluR1 have equivalent pairs C66/C106 and C406/C411 in the T1R1 sequence. The cysteine bridges were created after homology modeling. Simulations initiated in the closed conformation of T1R1 were unnecessary, because the closed conformation was spontaneously accessed in simulations initiated in the open conformation.

Crystal waters from the template structure were discarded. The simulation time scales were long enough for water to diffuse into the hydrophilic binding vestibule, and hydration of the VFTD was very similar to that observed in the crystal structure (data not shown). All eight water molecules of hydration within 5 Å of the ligand in the crystal structure were also observed in the simulation. It has been previously shown that water molecules, being highly mobile, are able to attain crystallographic positions in simulations, even when the ligand-binding site is screened from bulk water by a hydrophobic barrier such as a protein or membrane [27].

Simulation set-up

Two simulations each of the glutamate-free form (NOGlu1 and NOGlu2) and the glutamate-bound form (Glu1 and Glu2) were implemented, and four simulations of the glutamate-bound protein with the nucleotide GMP (GMPGlu1–GMPGlu4) were implemented. The GMP molecule was placed between residues H308 and H71, which were called ‘pincer’ residues by Zhang et al. [17] because they are part of the putative allosteric site that accommodates GMP in the VFTD. The GMP molecule was randomly rotated around its own axes at its binding site in the four copies of the GMPGlu simulations. Structural information on the nucleotide-binding site on T1R1 is unavailable, and reliable prediction of ligand-binding sites is still not possible by MD simulations. The data obtained by Zhang et al. [17] were therefore used to assign the initial position of GMP. The initial conformations of the protein for the GMP-containing simulations were obtained from 10 ns snapshots of the Glu1 simulations. The protein complexes were solvated in a dodecahedral cell with approximately 26 000 water molecules, such that the distance between periodic images of the protein was at least 2.5 times the cut-off distance for non-bonded interactions. Na+ and Cl ions were used as counter-ions to keep the systems electrostatically neutral. Simulations were run in GROMACS version 4.5.3 [28–31] using the CHARMM27 all-atom force field with CMAP [32]. An energy minimization using the steepest-decent method with all bonds constrained was performed until the maximum force experienced by the system was below 500 kJ·mol−1·nm−1. After successful energy minimization, the systems were subject to MD simulations in the NPT statistical ensemble using the leap-frog integrator [33] with a time step of 2 fs. Temperature coupling was performed separately for the protein and the rest of the system using the Nose–Hoover thermostat [34] with a reference temperature of 310 K and a time constant of 0.4 ps for both subgroups. Isotropic pressure coupling was applied using the Parrinello–Rahman barostat [35] with a coupling constant of 4 ps and a reference pressure of 1.0 bar. A compressibility value for pure water of 4.5 × 10−5 bar−1 was used. Bonds containing hydrogen atoms were constrained using the LINCS algorithm [36,37], and water molecules were constrained using SETTLE [38]. Periodic boundary conditions were applied in all directions. A neighbor list with a 12 Å cut-off was used for non-bonded interactions and was updated every 20 fs. The van der Waals interactions were switched off using a switching function between 10 and 11 Å. The electrostatic interactions were treated using the particle mesh Ewald method [39,40] using a 4th-order spline and Fourier spacing of 1.5 Å. The real-space sum for the particle mesh Ewald method was cut off at 12 Å. The center of mass translation was removed every 20 fs. Simulations were run for at least 100 ns each. Trajectories were sampled every 10 ps. In the GMPGlu simulations, the GMP was kept under harmonic restraints for the first 10 ns, which were then released for the production runs. For calculation of ensemble-averaged properties, the first 50 ns of each simulation were discarded. The analysis was performed using GROMACS or custom-made programs. PCA was performed on each simulation to extract the key orthogonal modes of the protein motion. Visualization and snapshots were rendered using VMD [41].


The MEMPHYS Center for Biomembrane Physics is funded by the Danish National Research Foundation. H.K. is funded by a Lundbeck Junior Group Leader Fellowship. The computations were performed at the Danish Center for Scientific Computing at the University of Southern Denmark.