Introduction

Unconventional T cells covers a variety of cell subsets that span both the innate and adaptive immunity, which can recognize a wide range of non-polymorphic ligands, and plays an important role in tissue homeostasis, disease and therapy1,2,3,4,5. As a subset of unconventional T cells, mucosal-associated invariant T cells (MAIT cells) are abundant in humans6 and are activated by cells infected with various strains of bacteria and yeast whose activation required cognate interaction between an mucosal-associated invariant T cell receptor (TCR) and a major histocompatibility complex (MHC) class I-like protein known as MHC class I-related protein 1 (MR1), which can present bacteria-derived ligands to TCRs as antigens2,7,8. Interestingly, MR1 is encoded by a non-polymorphic gene and exhibits strong conservation among mammals9 and is ubiquitously expressed by almost all nucleated human cells10. humans, vitamin B2 (or riboflavin, RBF) metabolites are the ligands of MR1, and the ligand-MR1 complex are recognized by the MAIT cell’s TCR11. The prototypical antigen of MR1 is the small molecule 5-(2-oxopropylideneamino)−6-D-ribitylaminouracil (5-OP-RU), which forms a covalent bond to MR1 via Schiff base linkage12, whereas non-covalent bound antigens, including 7-hydroxy-6-methyl-8-D-ribityllumazine (RL-6-Me-7-OH; hereafter referred to as 1VY), form hydrogen bond(s) and hydrophobic interactions with MR1 (Fig. 1)13. The ribityl moiety tail of these ligands (5-OP-RU and 1VY) plays a significant role in MR1’s binding to the TCR, which interacts with the TCR via hydrogen bonds. Non-covalently MR1-binding natural ligands with the ribityl group such as riboflavin-derived photolumazine I (PLI) and III (PLIII)14 and RBF itself are included in a broad ligandome15. Experiments have shown that only PLI and PLIII activate MAIT cells as antigens, whereas RBF does not result in MAIT activation15.

Fig. 1
figure 1

MR1 and its natural ligands. (a) Crystal structure of MR1 bound to 1VY (PDB ID: 4L4V). Shown are MR1 in a white cartoon representation with the sidechains as a thin stick representation, and the non-covalently bound ligand as a stick model (yellow), both colored by the CPK scheme. Also shown are the location of the A’ and F’ sites, with the residues of Tyr7, Arg9, Lys43, Tyr62 in A’ and those of Glu76, Tyr112 and Asn146 in F’, as well as Arg94 in the boundary between the two sites are labeled. The structure images were produced using Molmil16, a web-based molecular viewer developed by Protein Data Bank Japan17,18,19. (b) Comparison between the ligand 1VY with a known experimental structure, to the ligands PLI, PLIII and RBF, which are used in this study. Also indicated for the ligands are the atoms used for the PCA. For each ligand, the atom-pairs are indicated with letters, a–d, while atoms that are part of the protein-ligand pairs are indicated with a * symbol. (c) Initial structure of MR1 and PLI before the start of the McMD simulations, with the ligand in the unbound state, and the cylinder shown. Also shown is the axis of the cylinder \(\overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\rightharpoonup}$}}{\lambda }\) used to restrain the ligand (see Section S2).

Full size image

Structurally, MR1 consists of three domains, α1-α3, and a non-covalently bound protein, β−2 microglobulin, like is found in the MHC class I proteins11. The α1 and α2 domains form the antigen binding groove, with two α-helices sitting atop a 7-strand antiparallel β-sheet (Fig. 1). While peptides with a length between 8 and 11 residues bind in an extended conformation to fill the groove in the MHC class I proteins20, only a small part of the groove is used for ligand recognition in MR1. The ligand binding site of MR1, called the A’ site21, comprises of hydrophobic and basic residues such as a lysine residue (Lys43), where aromatic residues (Tyr7 and Tyr62) and basic residues at the edge of the A’ site (Arg9) form hydrophobic contacts and hydrogen bonds with the ligand, respectively. The other site, the F’ site whose amino acid residues are more exposed than the A’ site has been shown to not interact with its antigen or the TCR21. In addition, Arg94 is located at the boundary between A’ and F’ sites. While Lys43 in the A’ site forms a Schiff-base link with the ligand of 5-OP-RU12, it does not form any interaction in the non-covalently bound 1VY ligand (Fig. 1)13. Only a limited part of the 1VY ligand, in particular, the ribityl moiety tail is exposed to the solvent, which is presented to the TCR and forms hydrogen bonds to the TCR molecule13. Although the tertiary structure of the Schiff base-forming ligands such as 5-OP-RU is well-studied, structure information of non-covalent ligands such as the above mentioned PLI, PLIII and RBF ligands, is limited to 1VY.

Multicanonical molecular dynamics (McMD) simulation, which is one of enhanced sampling methods, enables us to study phenomena such as the binding between molecules at atomistic resolution, not offered by either experimental methods or conventional MD simulations. We have developed a ligand-receptor molecular docking implementation using McMD, see Section S1 for an explanation of the McMD theory22,23. This simulation protocol has been successfully applied to a wide range of cases, such as small-molecule ligands and medium-sized ligands including high-affinity peptides of around 10 residues24,25,26,27,28,29,30,31. Finally, we have also employed this method for the docking of RBF to an RNA riboswitch32, RNA force field evaluation33, and for the sampling of enzyme substrate interactions34. In McMD, the bias as a function of the potential energy (or temperature) enables McMD simulations to adaptively modulate the bias given the density of states as in the generalized ensemble methods. Thus, the potential energy surface functions as a reaction coordinate, which does not depend on any prior knowledge (e.g., the native receptor-ligand complex). The canonical ensemble at any given temperature, which is one of the physico-chemically acceptable ensembles, can be constructed from the multicanonical ensemble by using a reweighting procedure. The free energy landscape (FEL), which governs the thermodynamic properties of a system, can then be obtained by mapping the reweighted structural ensemble onto a reaction coordinate such as a binding path or onto one or more principal components obtained from Principal Component Analysis (PCA). Analysis of the FEL then uncovers the stable bound complexes as sampled by the McMD simulations. We built a framework of tools to systematically analyze the canonical ensemble, producing representative structures ranked by their free energy23. Finally, by analyzing the representative complex structures, we can also postulate whether the complex allows known binders to form trimer structures. Alternatively, the representative structures could prevent binding, as we showed in our recent work analyzing how the peptide inhibitor OS1 impedes the protein-protein interaction between GPIbα and the von Willebrand factor30.

We previously reported the McMD-based dynamic docking analysis between MHC class I and 9- or 10-residue peptides, where an exhaustive search of the peptides’ conformation and configuration were executed in the region including its binding groove and the surrounding bulk28,31. Here, we target a similar protein and one of the ligands used in one of our other works32 and apply the McMD protocol to explore docking configurations between MR1 and each of four natural ligands (1VY, PLI, PLIII and RBF) with a flexible MR1 molecule in explicit solvent. We show that while two compounds (PLI and PLIII) bind with a similar configuration to 1VY in the experimental structure, RBF binds in a nonspecific manner. Our analyses have revealed the binding mechanism, where binding mainly occurs at the A’ site, but that the F’ site might play a role in trapping the RBF at this site, thereby preventing an immune response. Finally, we discuss the effects of the ligands in activating MAIT cells.

Results and discussion

Independent McMD-based dynamic docking simulations (Section S2) were performed for each of the four ligands (1VY, PLI, PLIII and RBF) starting from the unbound state, with the production run producing 12 µs of data for each system, with snapshots saved at 10 ps intervals (Table S10). The ligand 1VY, whose complex structure with MR1 has been experimentally determined, was included as a control. Characteristics of the convergence of both the McMD pre-run and production run are described on Tables S1-S8, where all simulations demonstrated satisfactory convergence. Subsequently, we applied our dynamic docking analysis protocol to study and compare the structural ensembles (Section S3). Table 1 lists the representative rk structures on the FEL (Fig. S1) with a cluster free energy (CFE) of less than 2 kcal/mol for each of the structural ensembles at 300 K and their properties. The reference structure for PLI, PLIII, and RBF is derived from the experimental structure of MR1 in complex with 1VY and a MAIT cell-TCR (PDB ID: 4L4V), where the ligand is replaced with each of the target ligands. For the analysis between MR1 and 1VY, the experimental structure is used as the reference structure to calculate the R(ref) and root-mean-square-deviation (RMSD) values. Here, the ligand binding R(ref)-value, which ranges from 0 to 1, serves as a measure of the conservation of intermolecular interactions between the ligand and the MR1 with respect to the reference structure, with values above 0.8 indicating high conservation (see Section S3 for an explanation of the R-value). Comparing the results, it shows that the configuration of the MR1-1VY complex at rank 3 (r31VY) has the same binding configuration as the experimental structure, with the configuration at rank 1 (r11VY) partially matching the experimental structure. The configuration of the MR1-PLI complex at rank 2 (r2PLI) has a similar binding configuration to the reference structure, while for the MR1-PLIII complex, this is the case for rank 1 (r1PLIII). On the other hand, for the MR1-RBF complex, no similar binding configurations that were similar to the reference structure were found. This makes sense, as PLI and PLIII were found to activate MAIT cells, while RBF was found to not result in MAIT cell activation.

Table 1 Properties of metastable binding configurations between MR1 and ligands.
Full size table

For the dynamic docking simulations between 1VY and MR1, the top-ranking configuration, r11VY, only partially matches the experimental structure. As shown in Fig. 2a, two of the representative structures bind to the A’ site, while one binds to the F’ site. r11VY occupies the A’ site but adopts an inverted tail orientation of the ribityl moiety compared to the experimental structure, with the tail interacting with Arg61 (Fig. 2b). In contrast, configuration r21VY binds to the F’ site, forming interactions with Arg9, Glu149, and Asn146, and stacking against Trp69 (Fig. 2c). Configuration r31VY, however, binds in a manner closely resembling the experimental structure, with the tail engaging Arg94 (Fig. 2d). Comparing r11VY and r31VY, MR1 maintains a largely similar overall conformation, with only minor shifts in the positions of Arg9, His58, Arg61, Tyr62, and Gln153. Yet, r11VY is preferred over r31VY. One key factor is the strong hydrogen bond network formed between Tyr62 and Arg9, which stabilizes the stacking interaction between Tyr62 and the ligand in r11VY. In r31VY, Glu76 moves closer to Arg9, displacing it from the ligand and Tyr62, thereby breaking the stabilizing hydrogen bond network. This loss of interaction destabilizes the complex, despite r31VY’s closer match to the experimental structure. Although Arg94 partially compensates for this loss in r31VY, it is not positioned to form a comparable hydrogen bond network with Tyr62. This suggests that the conformational stability driven by the Tyr62-Arg9 interaction is a critical determinant in favor of r11VY. To verify these results, we performed additional MD simulations starting from these structures at high temperature (400 K), where we subsequently performed intermolecular stability analysis on the binding configurations. The 1VY structures show that r31VY remains the most stable across both 300 K and 400 K simulations. With a stability R-value of 0.999 at 300 K and 0.908 at 400 K, r31VY exhibits excellent structural fidelity, closely matching the experimental structure (Table 1, see the table caption for an explanation of the stability R-value). In contrast, r11VY and r21VY show lower stability and increasing deviation at 400 K. Collectively, r31VY remains the most stable even at 400 K, and is identified by our McMD-based docking protocol as the top candidate in the blind prediction, indicating that our simulations have successfully reproduced the experimental structure.

Fig. 2
figure 2

Dynamic docking results between MR1 and 1VY. (a) Representative structures rk1VY, which are metastable complexes between MR1 and 1VY ranked based on their population density (Table 1). The protein structures are rendered as a cartoon model, and the ligand as a stick model. Here, r11VY, r21VY and r31VY are colored cyan, magenta, and yellow, respectively. (b) Closeup of r11VY, with hydrogen bonds made by the ligand shown and Arg9, Lys43 and Arg61 have been labeled. (c) Closeup of r21VY, with hydrogen bonds made by the ligand shown and Arg9, Asn146 and Glu149 have been labeled. (d) Closeup of r31VY, with hydrogen bonds made by the ligand shown and Lys43 and Arg94 have been labeled.

Full size image

Taking a closer look at the representative structures for PLI (Fig. 3a), we can see that the two configurations bind as mirror images of each other, with the tail and carboxyl group switching places. In the case of configuration r2PLI with a similar binding configuration to the reference structure (RMSD = 2.00 Å in Table 1), the carboxyl group interacts with Lys43 and Tyr7, while the tail section forms a complex hydrogen bond network with Arg94 and Tyr152 (Fig. 3c). On the other hand, the carboxyl group of configuration r1PLI interacts with Tyr62 and Arg9 (Fig. 3b), while the tail does not form hydrogen bonds with MR1, only relying on basic packing. The binding of r11VY and r1PLI are quite similar. Although r11VY has a direct hydrogen bond network formed between Arg9-Tyr62 and Arg9-1VY, with Arg9 as the center residue, in r1PLI the ligand is the center residue using its carboxyl group. The validation MD analyses reveal that still, r1PLI exhibits a higher stability in maintaining MR1-PLI interactions with a stability R-value of 0.920, versus a stability R-value of 0.838 for r2PLI (Table 1). This suggests that packing is energetically more important for PLI than hydrogen bonding, which we can also see in the relative accessible surface area (RASA) value being much lower for r1PLI. We also qualitatively analyzed the effectiveness of MAIT cell-TCR binding based on the structure of 4L4V. Here, although r1PLI does not have any apparent clashes with the MR1 molecule (Fig. S2a), it fails to form favorable interactions with the TCR molecule, suggesting it cannot support productive TCR engagement. On the other hand, the tail section of r2PLI forms a hydrogen bond network between MR1 and TCRαTyr95, effectively acting as a molecular bridge that facilitates TCR engagement (Fig. S2b). Furthermore, with the TCR binding, the surface exposure of r2PLI is also reduced, offsetting the energetic penalty. This data suggests that a configurational shift occurs upon binding to the TCR molecule, where if PLI is simply bound to MR1, is r1PLI preferred, while the TCR molecule would only bind r2PLI, driving the configurational ensemble to r2PLI.

Fig. 3
figure 3

Dynamic docking results between MR1 and PLI. (a) Representative structures rkPLI, which are metastable complexes between MR1 and PLI ranked based on their population density (Table 1). The protein structures are rendered as a cartoon model, and the ligand as a stick model. Here, r1PLI and r2PLI are colored cyan and magenta, respectively. (b) Closeup of r1PLI, with hydrogen bonds made by the ligand shown and Arg9 and Tyr62 have been labeled. (c) Closeup of r2PLI, with hydrogen bonds made by the ligand shown and Tyr7, Lys43, Arg94 and Tyr152 have been labeled.

Full size image

For PLIII, there are three metastable structures (Fig. 4a). Here, configuration r1PLIII binds in a comparable manner to MR1 as configuration r2PLI (RMSD = 1.86 Å in Table 1), with the double-ring scaffold forming hydrogen bonds with Arg9, Ser24 and Tyr62, while being sandwiched between Trp69 and Tyr7 (Fig. 4b). The tail forms hydrogen bonds with Arg94 and faces outwards. The indole group is surrounded in a hydrophobic pocket by Tyr7, Leu54, His58, Trp59, Tyr62, Trp164 and Phe168. On the other hand, in r2PLIII, the scaffold forms only a weak hydrogen bond with Trp156 while sandwiched between Tyr7 and Tyr62, the tail forms hydrogen bonds with His58 and Glu160, while the indole group is in the back of the MR1 molecule (Fig. 4c). In configuration r3PLIII, the scaffold forms more interactions, with it forming hydrogen bonds with Arg9 and Lys43 while sandwiched between Tyr7 and Tyr63, the tail faces outwards without interacting with anything and the indole group is exposed while forming a weak interaction with Gln153 (Fig. 4d). Here, r1PLIII clearly makes the most favorable interactions, and the validation simulations confirm this, with r1PLIII (stability R-value = 0.855 in Table 1) scoring higher than r2PLIII (stability R-value = 0.802) and r3PLIII (stability R-value = 0.835), however this is lower than the stability of r1PLI. For PLIII we also analyzed the efficacy of TCR binding, which indicated that r1PLIII is the most likely configuration to bind to the TCR, with some induced fit of the tail most likely occurring as the TCR complex binds (Fig. S3a). On the other hand, both r2PLIII (Fig. S3b) and r3PLIII (Fig. S3c) have either clashes and/or weak interactions, limiting their ability to bind the TCR.

Fig. 4
figure 4

Dynamic docking results between MR1 and PLIII. (a) Representative structures rkPLIII, which are metastable complexes between MR1 and PLIII ranked based on their population density (Table 1). The protein structures are rendered as a cartoon model, and the ligand as a stick model. Here, r1PLIII, r2PLIII and r3PLIII are colored cyan, magenta, and yellow, respectively. (b) Closeup of r1PLIII, with hydrogen bonds made by the ligand shown and Arg9, Ser24, Tyr62 and Arg94 have been labeled. (c) Closeup of r2PLIII, with hydrogen bonds made by the ligand shown and His58, Trp156 and Glu160 have been labeled. (d) Closeup of r3PLIII, with hydrogen bonds made by the ligand shown and Arg9, Lys43 and Gln153 have been labeled.

Full size image

In the case of RBF, there are six metastable configurations (Fig. 5a), of which two bind at the F’ site. The scaffold of configuration r1RBF forms hydrogen bonds with Arg94 and Gln153, and on one side interacts with Trp156 (Fig. 5b). The tail faces outwards and interacts with Tyr152. The scaffold of configuration r2RBF does not form any hydrogen bonds, but is sandwiched between Tyr7 and Tyr62, while the tail forms hydrogen bonds with Tyr7 and the backbone of His58 (Fig. 5c). On other hand, configuration r3RBF binds at the F’ site and forms multiple hydrogen bonds using its scaffold with Arg94, Tyr112, Asn146 and Gln153, while on one side interacting with Tyr143 (Fig. 5d). Its tail face inwards, interacting with Glu76. Similarly, configuration r4RBF also binds at the F’ site, but does not form any hydrogen bonds using its scaffold nor does it interact strongly with any aromatic residues, although Tyr92, Tyr112, Phe119 and Trp143 surround the pocket (Fig. 5e). Its tail faces sideways towards the A’ site and forms hydrogen bonds with Glu76 and Glu149. The scaffold of configuration r5RBF only forms a hydrogen bond with Tyr7 and is sandwiched between Tyr62 and Trp164 (Fig. 5f). Its tail interacts with Glu160 and Arg61 and faces outwards. Finally, configuration r6RBF, although it has several potential partners, since the interactions are quite exposed on the surface, they appear to be quite weak. Stability analysis shows that configurations r1RBF, r2RBF and r5RBF exhibit medium-strong stability, whereas the remaining configurations are medium-weak. The reduced stability at 400 K for some of the configurations reflects increased thermal motion disrupting key ligand-receptor contacts, highlighting the thermodynamic fragility of certain binding configurations. Investigating the efficacy of TCR binding shows that r1RBF, r5RBF and r6RBF prevent TCR binding, due to clashes formed, while r2RBF, r3RBF and r4RBF form no interactions (Fig. S4). These diverse and unstable binding configurations, some of which occupy the F’ site, do not support the structural features required for productive TCR binding, as shown by our analysis of the interface geometry and hydrogen bonding patterns.

Fig. 5
figure 5

Dynamic docking results between MR1 and RBF. (a) Representative structures rkRBF, which are metastable complexes between MR1 and RBF ranked based on their population density (Table 1). The protein structures are rendered as a cartoon model, and the ligand as a stick model. Here, r1RBF, r2RBF, r3RBF, r4RBF, r5RBF and r6RBF are colored cyan, magenta, yellow, pink, lime, and navy blue, respectively. (b) Closeup of r1RBF, with hydrogen bonds made by the ligand shown and Arg94, Tyr152, Gln153 and Trp156 have been labeled. (c) Closeup of r2RBF, with hydrogen bonds made by the ligand shown and Tyr7 and Tyr62 have been labeled. (d) Closeup of r3RBF, with hydrogen bonds made by the ligand shown and Glu76, Arg94, Tyr112, Asn146 and Gln153 have been labeled. (e) Closeup of r4RBF, with hydrogen bonds made by the ligand shown and Glu76 and Glu149 have been labeled. (f) Closeup of r5RBF, with hydrogen bonds made by the ligand shown and Tyr7, Arg61, Tyr62, Glu160 and Trp164 have been labeled.

Full size image

Compared to PLI and PLIII, RBF binds in more various configurations, with some configurations occupying the F’ site (a previously identified ligand-binding pocket). The multiplicity (number of configurations required to reach a cumulative population of 90%) of the RBF-MR1 complex is five distinct configurations, while the multiplicity of PLI and PLIII are both two configurations. Considering the population density of the clusters and the stability (high stability R-values over 0.800 at 400 K) of the corresponding representative structures (Table 1), we can postulate that PLI binds the strongest to MR1, followed by PLIII and finally RBF. Finally, inspection of how the TCR molecule might bind to the MR1-ligand complexes (Table 2; Fig. 6) showed that both the MR1-PLI and MR1-PLIII complexes can successfully bind and should lead to MAIT activation. Although PLI has two stable configurations, only r2PLI supports productive TCR binding. Therefore, despite its high stability, only one configuration (r2PLI) leads to effective TCR engagement. In contrast, PLIII has only one configuration (r2PLIII) with minor clashes, which might be rescued by induced fit, and no other configuration that supports TCR binding. On the other hand, none of the MR1-RBF complexes would appear to lead to successful TCR binding and therefore no MAIT activation, with some configurations even preventing TCR binding.

Table 2 Bindability of ligand in representative structures to MAIT-TCR.
Full size table
Fig. 6
figure 6

Recognition of PLI and PLIII by MAIT-TCR. Global overview (a) and zoomed-in version (b) of TCR bound to r1PLI. Global overview (c) and zoomed-in version (d) of TCR bound to r1PLIII. Here, MR1 is shown as a white cartoon, PLI & PLIII in yellow CPK, TCRα in cyan cartoon and TCRβ in magenta cartoon. In the zoomed in figures (b and d), the sidechains within 5Å of the ligand are also shown.

Full size image

Table S9 lists the RMSD values of key MR1 binding-site residues across the various ligand-bound configurations, highlighting those with deviations exceeding 3.0 Å from the experimental structure (PDB ID: 4L4V). Among these, Lys43 and Tyr62 consistently exhibit low RMSD values across multiple representative configurations, particularly in the r11VY and r3PLI configurations. This observation indicates that these residues maintain stable orientations in native conformations resembling the experimentally observed structure. Their low RMSD values suggest strong structural and energetic coupling to the native-like binding pose, underscoring a key role in stabilizing the receptor-ligand interface in a biologically relevant conformation. In contrast, other residues such as His58, Arg61, and Trp143 exhibit higher RMSD values, especially in metastable configurations (e.g., r21VY, r2PLIII, and r6RBF), where deviations exceed 3.0 Å. These larger RMSDs reflect alternative, energetically accessible conformations but deviate significantly from the experimental reference and are induced by ligand binding. Collectively, the relative stability of Lys43 and Tyr62 thus serves as a structural signature for identifying binding configurations that closely resemble the experimentally observed native state, highlighting their critical role in maintaining a biologically relevant receptor-ligand interface.

Harriff et al. previously showed experimentally that RBF does not result in MAIT cell activation15, but no structural explanation was given at the time. Here, we have performed extensive dynamic docking simulations that elucidate how the compound could bind to MR1. Our results indicate that PLI and PLIII can both facilitate the binding of the TCR, but RBF binding would prevent this, in line with Harriff et al.’s experiments. RBF also binds in a more diverse manner, with even some configurations binding to the F’ site. Therefore, MR1 does not specifically bind RBF, unlike PLI and PLIII. Furthermore, our analysis of potential TCR complex structures shows that none of the RBF-bound configurations support productive TCR binding. For TCR binding to occur, the ligand tail must project outward to form a compatible interface. In RBF, however, to position the tail in a way that allows TCR docking, the hydrophilic portion would need to penetrate the deep pocket (interacting with Arg9 and Lys43), while the hydrophobic side would be exposed to solvent, a configuration that is energetically unfavorable due to the associated desolvation penalties. This structural incompatibility directly translates into a failure to support productive TCR engagement, consistent with experimental observations that RBF does not activate MAIT cells. Since RBF is also used by eukaryotes and has been found to be an important nutrient, where deficiency could lead to oxidative stress, mitochondrial dysfunction, neurological disorders and various cancers35, RBF is an important compound that is found throughout our bodies, therefore having RBF lead to MAIT cell activation would lead to catastrophe. Combined, this suggests that MR1 has evolved to avoid recognizing RBF in a way that would enable TCR engagement, likely due to the risk of inappropriate immune activation against a vital nutrient, resulting in RBF binding in configurations that do not support TCR binding and thus fail to trigger MAIT cell activation. On the other hand, one would expect that metabolites derived from microbial sources or RBF precursors of RBF would be recognized by MR1 and MAIT, thereby using those as signals for infection, while not responding to the presence of the useful RBF compound. These structural and energetic constraints suggest that MR1 has evolved a mechanism to distinguish between microbial metabolites and essential nutrients, thereby preventing inappropriate immune activation against a vital vitamin.

Conclusion

We performed four independent dynamic docking simulations to investigate how four ligands, 1VY, PLI, PLIII, and RBF, bind to MR1 and whether these interactions enable TCR engagement and MAIT cell activation, with 1VY (PDB ID: 4L4V) included as a control. Our simulations successfully reproduced the 4L4V structure, and subsequent analyses focused on the other ligands. Both PLI and PLIII bind MR1 in a comparable manner as 1VY does and with high specificity. On the other hand, RBF binding is dissimilar to 1VY and binding is less specific. Furthermore, while PLI and PLIII bind in a manner that exposes the ribityl moiety tail to allow TCR binding and ultimately leading to MAIT cell activation, the same was not true for RBF, which could only bind in configurations where the tail was not exposed. Given that RBF is a vital nutrient, MR1 likely evolved to avoid recognizing intact RBF in a way that would trigger immune activation. Instead, MR1 may recognize microbial-derived metabolites or precursors of RBF as danger signals, enabling infection detection without inappropriate responses to essential nutrients. These findings reveal that MR1 selectively enables MAIT cell activation through specific, tail-exposing binding configurations for microbial metabolites like PLI and PLIII, while actively excluding the essential nutrient RBF via structurally incompatible and non-productive binding, thereby preventing inappropriate immune responses to vital dietary components.

Methods

Since no crystal structure of the ligands (PLI, PLIII and RBF) with MR1 was available, we used the structure of MR1 in complex with 1VY (PDB ID: 4L4V) as a template13, modifying the ligand to match PLI and PLIII to produce reference structures. For the RBF ligand, we used the parameters from our previous work where we performed dynamic docking simulations between RBF and a riboswitch molecule32. For RBF, we roughly positioned the molecule to match that of 1VY in 4L4V to produce the reference structure that we use for comparison during the analysis stage. The PLI and PLIII ligands were prepared in the same way as the RBF ligand was previously, where Gaussian36 at the HF/6-31G* level was used to optimize the geometries and calculate the electron density, followed by RESP37,38 to finally obtain the partial atomic charges by fitting the density.

The system setup, simulation parameters, and computational protocol are analogous to those employed in our prior work on MHC molecules28,31. MR1 was truncated to retain only the ligand-binding domain, removing the α3 and β2m-like domains not involved in ligand or TCR recognition, to improve computational efficiency and an NME cap was added to the C-terminal of the MR1 molecule. To parameterize the system, we used the Amber ff99SB-ILDN force field39 for the MR1 molecule, where a disulfide bond was added between Cys98 and Cys161 of the MR1 molecule. The first principal axis of inertia of the MR1 molecule was aligned with the Z-axis, while the second one was with the Y-axis. Then, the ligand was translated 30 Å along the X-axis further into the bulk region (to generate the unbound structure). Gromacs 2023.1 was used to perform the preparations40, while a modified version of it was used to perform the dynamic docking simulations22,23. A triclinic box of was placed around each of the systems and filled with TIP3P waters41, and NaCl was added to neutralize the system and bring its concentration to 0.1 M, using the monovalent ion parameters by Joung et al.42 To further prepare the system, two iterations of energy minimizations were performed, with the first iteration having position restraints on the heavy MHC and ligand atoms until a maximum force of 1000 kJ/mol/nm2 was achieved, followed by a second energy minimization without position restraints until to a maximum force of 100 kJ/mol/nm2 was achieved. Next, 100 ps of NVT simulation at 300 K using the Bussi thermostat43 with position restraints on the heavy protein atoms, with velocities initialized at 300 K following a Maxwell distribution was performed. Finally, 100 ps NPT simulation with position restraints using the Parrinello-Rahman barostat was used to equilibrate the pressure44. The long-range electrostatics were calculated using the zero-dipole summation method, which is a cutoff-based approach utilizing a well-defined pairwise function45, with the damping factor α set to 0 Å−1 and the atom-based cutoff length set to 12 Å. A time-step of 2 fs was used, with LINCS46 to constrain the bond lengths and SETTLE47 to constrain the water geometries. The characteristics of the systems and the simulation parameters are summarized in Table S10. Similarly, the McMD simulations were executed in a similar manner as our previous work (see also Section S2)28, with a description of the analyses performed on the multicanonical ensemble in Section S3.

Rendering of the images was performed using Molmil, a web-based platform. Versioning is not applicable due to the web-based nature of the tool. The software is accessible at https://pdbj.org/molmil2/.