To apply load, we used harmonic positional restraints on the terminal C_α atoms at given extensions (Figure 2A; see Laddered extension with added strands in Computational methods). This was based on the realistic situation of immune surveillance. When a T-cell interacts with an APC, other molecules such as CD2 and CD58 maintain the separation near the ∼120 Å span of the TCRαβ-pMHC complex (Reinherz et al., 2023). The force applied to the complex fluctuates via thermal motion and through cellular activities such as coupling to the actomyosin machinery within the T-cell and APC (Liu et al., 2016; Feng et al., 2017; Reinherz et al., 2023). Restraining terminal C_α atoms in simulation mimics the membrane anchoring of these molecules with a relatively constant spacing and fluctuating force. The measured instantaneous force varies in magnitude and direction across coordinate frames. Averaged over time, it becomes mainly longitudinal (Table 1). Despite the temporal variation in the applied force, the results below show that the response of the complex depends on the average force and the ligand.

In simulation, soft harmonic positional restraint can be used for conformational sampling (Pitera and Chodera, 2012). Rather than positional sampling, our goal is applying force at a given extension, for which we used a stiff 1-kcal/[mol.Ų] potential (Laddered extension with added strands) that yields thermal fluctuations of amplitude ∼0.8 Å (Hwang, 2007). While it is also possible to apply a constant force (Gomez et al., 2021; Stirnemann, 2022), it would not reflect the membrane-anchored state. In addition, the extension of the complex will fluctuate under a constant force, and the actual load that the TCRαβ-pMHC interface experiences will likewise fluctuate. How load propagates through the complex and distributes across the interface is a subject of a future study.

We assessed the effect of load on WT first by counting high-occupancy contacts with pMHC (Figure 3A; see Contact analysis in Computational methods). WT^low had the least number of contacts, followed by WT⁰ and WT^high, indicating low and high loads may have opposite effect on the interfacial stability. Vαβ-pMHC without the C-module formed the most contacts. This indicates that without a proper load, the C-module is detrimental to the stability of the interface with pMHC, as we found previously for the JM22 TCR (Hwang et al., 2020).

Time-dependent changes in the interfacial contacts were monitored using the Hamming distance H (Hamming, 1950; Hwang et al., 2020). H is the number of the initial high-occupancy contacts (those with greater than 80% average occupancy during the first 50 ns) that are subsequently lost during the simulation. A low H means that such contacts persist while a high H means the corresponding number of initial high-occupancy contacts are missing. Consistent with the contact count, H remained low for WT^high and Vαβ-pMHC (Figure 3B). In WT⁰ and WT^low, H increased after about 500 ns and 900 ns, respectively. Thus, the relatively high number of interfacial contacts for WT⁰ (Figure 3A) is due to the formation of new contacts rather than by maintaining the initial contacts.

Occupancy heat maps provide the time dependence of individual contacts. For WT^high and Vαβ-pMHC, high-occupancy contacts persist throughout the simulation (blue regions in Figure 3C and Figure 3—figure supplement 1) while WT⁰ or WT^low exhibited breakage of contacts, especially when H increased (dashed arrows in Figure 3D and E). Differences in the interfacial contacts also manifest in their location. We displayed the backbone C_α atoms of Vα and Vβ residues that form contacts with pMHC with greater than 80% average occupancy (averaging was done after the initial 500 ns; Figure 3F). In WT⁰, contacts are spread apart, and in WT^low they lie mostly along the length of the peptide. These layouts potentially make interfacial contacts more prone to break via easier access by water molecules. In WT^high and Vαβ-pMHC, high-occupancy contacts form more compact clusters. Exposure to water of the TCRαβ residues involved in those contacts was measured by their buried surface area (BSA), which follows the same trend as the number of high occupancy contacts (Figure 3A vs. Figure 3G) Furthermore, this trend also applied to the root-mean square fluctuation (RMSF) of Tax peptide backbone C_α atoms, even though RMSF values were small (Figure 3H).

Experimentally, the WT complex has a relatively strong affinity as a TCRαβ (about 1 μM) (Ding et al., 1999), which may be why the interface with pMHC involved more contacts in WT⁰ compared to WT^low. At low load, the short distance between restraints on the ends of the complex (Figure 2A) allows wider transverse motion that in turn generates a shear stress or a bending moment at the interface. Transverse stress will be less for WT⁰ where the end moves freely, and for WT^high where lateral motion is suppressed. The extent of transverse motion can be seen by the RMSF of the center of mass of the C_α atoms of the peptide in the transverse direction, which was 16.3 Å for WT^low and 12.7 Å for WT^high. The high stability of Vαβ-pMHC and WT^high agree well with the results for JM22 (Hwang et al., 2020).

We analyzed the motion between Vα and Vβ (Vα-Vβ motion) to find its effect on the TCRαβ-pMHC interface. Compared to the unliganded systems (Vαβ and Tαβ; Table 2), the number of high-occupancy Vα-Vβ contacts increased slightly in Vαβ-pMHC (‘V’ in Figure 4A), while it decreased in full TCRαβ-pMHC complexes (‘0’, ’Low’, and ‘High’ in Figure 4A). This shows that the Vα-Vβ interface is difficult to organize with the restrictions imposed by the bound pMHC, except in the absence of the constant domains. The number of Vα-Vβ contacts in the liganded systems is the smallest for WT^low, similar to the case for the number of contacts with pMHC (Figure 3A), again reflecting a destabilizing effect with low load.

To measure the Vα-Vβ motion, coordinate trajectories were first aligned to the stably folded β-sheet cores of the two domains. Triads were then assigned to the cores and principal component analysis (PCA) was performed on the triad trajectories (Figure 4B; Variable domain triads and PCA; Hwang et al., 2020). Since triads were assigned in the same way, the relative Vα-Vβ motion in different simulations can be studied by comparing their PCA.

The amplitude of PC1 is lower when the number of Vα-Vβ contacts is higher (Figure 4A vs. Figure 4—figure supplement 1A). Directions of PCs differed to varying extents (arrows in Figure 4—figure supplement 1B). Similarity of the directions was measured by the absolute value of the dot product between PCs as 18-dimensional unit vectors (for the six arms from two triads) in different systems. A value of 1 corresponds to the same direction, and 0 means an orthogonal direction (Figure 4—figure supplement 1C). For PC1, a high degree of similarity was observed between Tαβ and Vαβ, which is consistent with their similarity in the number of Vα-Vβ contacts (Figure 4A) and PC amplitudes (Figure 4—figure supplement 1A). Among triad systems with bound pMHC, WT^low differed significantly in the PC1 direction compared to others (Figure 4—figure supplement 1C, darker colors). The dot products varied more for PC2 and PC3, which capture finer motions with smaller amplitudes (Figure 4—figure supplement 1C).

To determine how the Vα-Vβ motion influences the interface with pMHC, we measured the distance between CDR3 loops (CDR3 distance), which play a central role in peptide discrimination (Figure 2B and C and Figure 4—figure supplement 2A–C). Unliganded Tαβ and Vαβ had greater fluctuation in the CDR3 distance (larger std), as they are unrestrained by pMHC. Among the pMHC-bound systems, WT^high and Vαβ-pMHC had small CDR3 distance (averages of 10.3 and 10.5 Å, respectively; Figure 4—figure supplement 2B C). CDR3 distance was larger for WT⁰ (11.3 Å), which reflects an altered interface with pMHC. For WT^low, the CDR3 distance varied more widely, with more than a 2-fold increase in standard deviation. The increase in CDR3 distance of WT^low happens after the increase in H (800 ns; Figure 3B and D and Figure 4—figure supplement 2C), suggesting a loss of contacts at the interface is related to the Vα-Vβ motion.

PCA decomposes the Vα-Vβ motion into mutually orthogonal directions. We made 2-dimensional histograms of each of these projections versus the corresponding CDR3 distance (Figure 4—figure supplement 1D). If any of the PC modes is strongly correlated with the CDR3 distance, the corresponding histogram would exhibit a slanted profile. However, no clear correlation could be seen (Figure 4—figure supplement 1D), suggesting that the changes in CDR3 distance may depend on combinations of PCs. We addressed this possibility by considering angles between matching arms of the two triads that do not rely on PCA (Figure 4B). The e1-e1 angle, herein called ∠e1 (and similarly define ∠e2 and ∠e3; Figure 4B), can change either by the e1 arms turning up and down (‘flap’) or in and out of the page (‘twist’) in Figure 4B. Angles ∠e2 and ∠e3 depend primarily on rotation indicated by dashed arrows in Figure 4B (‘scissor’) (Hwang et al., 2020).

Histograms of the three angles (Figure 4C) show a clearer difference than individual PCs among the systems tested, and the CDR3 distance varies with the angles (Figure 4D). The wider distributions for angles of WT^low and WT^high reflect their higher PC amplitudes (Figure 4—figure supplement 1A). The symmetric distributions of ∠e2 and ∠e3 can be seen from the two peaks for WT^low, which is due to the reciprocal behavior of the scissoring motion involving the two angles (Figure 4B, side view, and Figure 4C, open diamonds). The two peaks are also related to the changes in the CDR3 distance (Figure 4D), which reflects an agitating effect of the mild load on the scissor motion. Given the definitions of the angles, the CDR3 distance will increase (dashed arrows in Figure 4B) with larger ∠e1 or ∠e3, or with smaller ∠e2 (Figure 4D). WT⁰, despite the apparent stability of the interface, had a larger CDR3 distance than WT^high and Vαβ-pMHC, again indicating a disrupted interface. These results show that the CDR3 motion is coupled to the Vα-Vβ motion, especially the scissoring motion.

We next analyzed the motion of the V-module relative to the C-module (V-C motion). The number of high-occupancy Cα-Cβ contacts did not vary significantly (in the 31–34 range) and they were more than the number of Vα-Vβ contacts, similar to the case for the JM22 TCR (Hwang et al., 2020). The C-module thereby influences the V-module as a single unit. The V-C motion was analyzed by performing PCA on the bead-on-chain (BOC) model constructed based on the β-sheet core of each domain, and hinges between V- and C-domains denoted as Hα and Hβ (Figure 5A; V-C BOC and PCA). Coordinate trajectories were aligned to the C-module so that the motion of the V-module relative to the C-module in different simulations can be compared. Across different systems, amplitudes of PCs were similar (Figure 5—figure supplement 1). PC1 (V-C bend; Figure 5A) was similar among systems, as seen by the values of dot products being close to 1.0 (Figure 5B). Directions of higher PCs varied more, similarly as higher PCs for the Vα-Vβ motion.

We noticed that Vα bends more compared to Vβ, as can be gleaned from the longer PC arrows for Vα (Figure 5A). We quantified this asymmetry by subtracting the amplitudes of motion for domains in the β chain from those for the matching domains in the α chains, where positive or negative values respectively indicate greater or less motion of the α compared to the β chain (Figure 5C). Compared to Tαβ, binding of pMHC increases the α-chain motion, which is the greatest in WT^low (Figure 5C, PC1 in top row). The greater degree of Vα-Cα motion is consistent with the smaller number of Vα-Cα contacts compared to Vβ-Cβ (Figure 5—figure supplement 1).

The asymmetry was further analyzed by measuring hinge angles ∠TCRα and ∠TCRβ independently of PCA (Figure 5A). Distributions of ∠TCRα varied more compared to ∠TCRβ (Figure 5D). A wide distribution of ∠TCRα for WT^low is related to the increase in the CDR3 distance and concomitant changes in the triad arm angles later during the simulation (Figure 4—figure supplement 2C). For WT^low and WT^high, CDR3 distance decreases with increasing hinge angles, especially with ∠TCRα (Figure 5E), which suggests that unbending of the V-module under load helps with bringing the CDR3 loops closer together. In WT^low, this state is not maintained and ∠TCRα decreases (more bending) as the CDR3 distance increases (Figure 5E) which happens after the increase in H (Figure 3B). These results suggest a mechanism by which the asymmetric response of the whole TCRαβ to load affects the binding with pMHC by controlling the relative positioning between the CDR3 loops via the Vα-Vβ motion. For this, the Cβ FG-loop plays a critical role, as simulations of the bound complex without the Cβ FG-loop resulted in a smaller ∠TCRβ and an over-extended ∠TCRα (see Appendix 1).

In the WT crystal structure, the side chain of Y5 in the Tax peptide is located between the CDR3 loops of Vα and Vβ while V7 mainly contacts CDR3β (Figure 2C). P6 makes one contact with CDR3β. The side chain of Y8 is located between CDR3β and the α1 helix of MHC. Crystal structures of point mutants of these four residues are very similar in terms of interfacial contacts, docking angle, and CDR loop conformations, with the only structurally observable difference located at CDR3β (Figure 2B and C; Ding et al., 1999; Scott et al., 2011). However, point mutations profoundly affect dynamics of the complex, as explained below.

Modified agonists Y5F and V7R had about the same number of contacts with pMHC as the WT complexes, but high load resulted in fewer contacts, indicating a potential slip bond behavior (Figure 6A), although loss of contacts in Y5F^high might have been due to a higher load experienced compared to other complexes at the same extension (23.7 pN; Table 1). Antagonists P6A and Y8A had overall fewer contacts with pMHC without a consistent load dependence (Figure 6A). This trend was also seen in BSA profiles of residues forming high-occupancy contacts with pMHC (Figure 6B). For modified agonists, higher load also resulted in greater increase of H, whereas the trend was opposite for antagonists (Figure 6—figure supplement 1). The large number of contacts with pMHC for modified agonists (Figure 6A) despite an increase in H suggests an altered binding rather than maintaining the initial contacts.

In contact heat maps, the Y5 residue of the WT peptide forms a hydrogen bond with αS31 and nonpolar contacts with a few residues in both Vα and Vβ (Figure 3C–E, Figure 6C). In Y5F, the hydrogen bond with αS31 cannot form, and many of the nonpolar contacts with F5 break under load later during the simulation (Figure 6—figure supplement 2A). The breakage coincides with the increase in H (Figure 6—figure supplement 1A). In addition, contacts involving Y8 and V7 also break in Y5F^high (Figure 6—figure supplement 2A). Thus, the Y5-αS31 hydrogen bond may stabilize the interface with pMHC by arranging other nearby residues to form nonpolar contacts in favorable positions; its absence would make the nonpolar contacts more prone to break under load. The relative stability of Y5F⁰ can also be seen by the similarity in the locations of high-occupancy contact residues between WT and Y5F⁰ (Figure 3F vs. Figure 6—figure supplement 3A). Experimentally, Y5F has kinetic and cytotoxicity profiles similar to WT (Hausmann et al., 1999; Scott et al., 2011). Its dependence on load needs further experimental analysis. On the other hand, V7 of the WT peptide forms nonpolar contacts with residues in CDR3β (Figure 3C–E, Figure 6C). In V7R, nonpolar contacts with CDR3β form with reduced occupancy, and contacts involving Y8 are also disrupted (Figure 6C vs. Figure 6—figure supplement 2B).

For antagonists, more contacts broke, which again involve non-mutated residues such as Y5 and V7 (Figure 6D–F, Figure 6—figure supplement 2C). The greater number of contacts in Y8A^high compared to Y8A⁰ and Y8A^low (Figure 6A) despite smaller number of contacts involving key peptide residues Y5–A8 (Figure 6—figure supplement 2C) suggests formation of additional contacts with other parts of MHC as a result of an altered interface. Experimentally, binding of the A6 TCR to pMHC containing the P6A or Y8A peptide was not detected in vitro (Ding et al., 1999). Thus, Y8A in principle could exhibit a catch bond, but forming the complex in the loaded state may be kinetically inaccessible.

The modified agonists had more Vα-Vβ contacts than WT^high while the antagonists had fewer, except for Y8A^high (Figure 7—figure supplement 1A). While the amplitude of Vα-Vβ motion was generally in a range similar to the WT systems (Figure 4—figure supplement 1A vs. Figure 7—figure supplement 1B), the CDR3 distance was larger for all mutant systems except for P6A, which had a weak dependence on triad angles (Figure 7A, B). The angles in turn varied among systems and loading conditions (Figure 7—figure supplement 2). These results suggest that point mutations to the peptide cause alterations in the load-dependence of the interface and the Vα-Vβ motion.

The mutants affected the average V-C BOC similarly as that for dFG^high (Figure 7C and D vs. Appendix 1—figure 1C). Among them, Y8A^high had an average BOC approaching that of WT^high, which aligns with the comparable number of contacts with pMHC (Figure 6A). However, the location of its Hα differed (Figure 7D), and the CDR3 distance was larger (Figure 7B). To quantify deformation of the average BOC, we measured displacements of centroids from the corresponding ones in WT^high, which were overall greater for the α chain than the β chain (Figure 7—figure supplement 3A B). Consistent with this, the mutants had fewer Vα-Cα contacts than WT^high and a similar number of Vβ-Cβ contacts (Figure 7—figure supplement 3C D).

Similar to the WT systems, the greater motion of the α chain than the β chain was observed in the mutant systems, as seen from the differences in V-C PC1 amplitudes (Figure 7—figure supplement 4A B). However, dot products of the BOC PC1 between WT and mutants revealed that the direction of motion differed by varying degrees, which was more for V7R^high and Y8A (Figure 7—figure supplement 4C vs. Figure 5B). Thus, point mutations on the WT peptide can affect the conformational motion of the whole TCRαβ, in addition to the average BOC.

To further test effects of point mutations, we introduced in silico point mutations P6A and Y8A to the WT complex (WT to antagonists) and conversely introduced A6P and A8Y mutations to the P6A and Y8A complexes, respectively (antagonists to WT). The in silico antagonists did exhibit reduction in contacts with pMHC while the results were mixed for the in silico WT, especially for A8Y where the introduced tyrosine is bulkier than the original alanine. Nevertheless, these tests support the above results based on the original crystal structures (See Appendix 2 for details).

To probe the dynamic relation between the TCRαβ-pMHC (intermolecular) interface and intra-TCRαβ (intramolecular) interfaces formed between subdomains of the complex, we calculated the total occupancy of the high-occupancy contacts in respective cases (Figure 8A, B). For the intramolecular contacts, we excluded the Cα-Cβ interface contacts since they are larger in number compared to other interfaces and did not differ significantly across different systems. Thus, the C-module moves mostly as a single unit (Hwang et al., 2020).

For WT⁰, the intermolecular contact occupancy stayed at around 20 (WT in Figure 8A, horizontal bar on the bottom) and for WT^low, it decreased later in simulation (WT in Figure 8A, darkening of circles without outline). In comparison, the intramolecular contact occupancy remained relatively constant for both WT⁰ and WT^low (WT in Figure 8B, horizontal bar on the bottom and circles without outline). For WT^high, the intermolecular contact occupancy was steady even with wider fluctuation in force (WT in Figure 8A, outlined circles), and the intramolecular occupancy also remained high, indicating the subdomains are held together tightly (WT in Figure 8B, outlined circles). For dFG^low, the intermolecular contact occupancy stayed low and intramolecular occupancy was relatively constant (dFG in Figure 8A and B, circles without outline). In dFG^high, the contact occupancy with pMHC increased (dFG in Figure 8A, outlined circles), but the intramolecular contact occupancy became low (dFG in Figure 8B, outlined circles), which suggests that the complex is not as tightly coupled compared to WT.

For modified agonists, the no load and low load cases had overall higher occupancy, both with pMHC and within TCRαβ, but occupancy fluctuated more as can be seen by the changes in colors in the occupancy trajectories (Y5F and V7R in Figure 8A and B, horizontal bars on the bottom and circles without outlines). Under high load, intermolecular contact occupancy decreased over time (Y5F and V7R in Figure 8A, darkening of outlined circles) while intramolecular contact occupancy either increased (Y5F^high) or decreased (V7R^high) relative to the respective low load cases. For antagonists, both occupancy measures were lower than the WT, and further reduction could be seen over time in some cases (P6A and Y8A in Figure 8A and B, darkening of colors in outlined circles).

The stability of the TCRαβ-pMHC interface also manifested into their relative motion, which was quantified by the angle between the least-square fit line across the backbone C^_α atoms of the antigenic peptide and the unit vector formed between the centroids of Vα and Vβ (Figure 8—figure supplement 1A–D). For WT, the peptide angle fluctuated more for WT^low than WT^high (WT in Figure 8C) where 58.4°±3.6° (avg±std after 500 ns) for WT^high reflects a diagonal binding. This is consistent with the greater transverse RMSF of WT^low mentioned earlier which may assist with destabilizing the interface. For dFG^low, the peptide changed orientation by more than 20°, and for dFG^high, it stabilized, but at a higher value than WT^high, which also was reached in dFG^low later during simulation, suggesting a more orthogonal binding (dFG in Figure 8C). For modified agonists, similar to the behaviors of the total intra- and intermolecular contact occupancy, the peptide angle was affected more under high loads, again becoming more orthogonal compared to WT^high (Y5F and V7R in Figure 8C). For antagonists, the angle overall fluctuated more under no load or settled to different values under high load. Since the antagonists are loosely coupled (low occupancy in Figure 8A and B), settling of the angle does not indicate stabilization of the interface, as evident from the positional shift of the α2 helix of V7R^high or Y8A^high (Figure 8—figure supplement 1G H) compared to WT^high (Figure 8—figure supplement 1E).

Structure preparation was done using CHARMM (Brooks et al., 2009). Simulation systems were based on PDB 1AO7 (Garboczi et al., 1996a) 1QSE, 1QRN, and 1QSF (Ding et al., 1999) and 3QFJ (Scott et al., 2011). Residues from the TCR α- and β-chains were renumbered sequentially from the original non-sequential numbering in the PDB. Throughout the paper we use the renumbered index to refer to a residue. Residues differing at a few locations in some of the PDB files were converted so that all systems have identical sequences except for point mutations introduced in the Tax peptide (details are given below). Disulfide bonds between cysteine residues were introduced as noted in the PDB file. Histidine protonation sites were determined based on the 1QSE crystal structure to promote hydrogen bond formation with neighboring residues. Where neighboring residues were unlikely to hydrogen bond, we assigned the water-facing nitrogen of histidine as charged. This led to protonation of the N^δ atom for all histidine residues except for MHC H263 and β2m H84, where the N^ϵ atom was protonated. For truncated structures, crystal waters within 2.8 Å from the protein atoms were kept in the initially built system. For full structures, all crystal waters were kept.

We extended the termini of the TCRαβ-pMHC complex as handles for applying positional restraints (Figure 2, “added strands”) (Hwang et al., 2020). For MHC, we used the sequence from UniProt P01892, where ²⁷⁶LSSQPTIPI²⁸⁴ was added after E275. For TCRαβ, sequences for the added strands were from GenBank ABB89050.1 (TCRα) and AAC08953.1 (TCRβ), which were ²⁰¹PESSCDVKLVEKSFETDT²¹⁸ and ²⁴⁶CGFTSESYQQGVLSA²⁶⁰, respectively. After adding the strands, a series of energy minimization and a short MD simulation in the FACTS implicit solvent environment (Haberthür and Caflisch, 2008) were performed to relax them and bring together the C-terminal ends of the two TCR chains. The first two N-terminal residues of TCRβ were missing in all structures except for 3QFJ, so they were added and briefly energy minimized.

1AO7 (Tax peptide): In the original PDB 1AO7, coordinates for the Cα domain (D116–S204) and parts of Cβ (E130–T143, K179–R188, S219–R228) are missing. The coordinates listed are based on the renumbered indices. These were built using PDB 1QSE. For the Cα domain, we aligned the Vα domain of 1AO7 and 1QSE (K1–P115) based on their backbone C_α atoms and added the missing Cα domain residue coordinates to 1AO7. After this, we performed a brief energy minimization on the added domain while fixing positions of all other atoms of 1AO7. For missing residues in the Cβ domain, we used backbone C_α atoms of two residues each before and after the missing part to align 1QSE to 1AO7 and filled in coordinates, followed by a brief energy minimization of the added part in 1AO7. In this way, the TCRαβ-pMHC interface of the original 1AO7 is preserved. By comparison, previous simulations mutated PDB 1QRN back to WT (Ayres et al., 2016), which corresponds to the A6P in silico WT system (Appendix 2), or converted a high-affinity variant of A6 (PDB 4FTV) by mutating β-chain residues, in particular nearly the entire CDR3 loop (Rangarajan et al., 2018). Compared to our approach, those preparation methods thereby introduce more perturbation to the interface with pMHC.

The β2m residues C67 and C91 were reverted (C67Y, C91K) based on UniProt P61769 referenced in PDB 1AO7. These agree with the β2m sequence in other structures.

1QRN (P6A): Except for the two N-terminal residues of TCRβ, there were no missing coordinates. This also applies to 1QSE and 1QSF. The following conversions were made to match the sequence with other structures: K150S (TCRα), and A133E and E134A (TCRβ).

1QSE (V7R): No residue conversion was made.

1QSF (Y8A): The following conversions were made: A219R (MHC) and A225T (TCRβ).

3QFJ (Y5F): There were no missing residues. We made the D204N conversion in TCRβ.

WT truncated complexes: For truncation, we used the constructed 1AO7 complex.

Each in silico mutation (Appendix 2—table 1) was performed for low and high load extensions of the complex. To use similar extensions as in the original complexes, we used the last frame of the 4-ns laddered extension simulation. After introducing the in silico mutation, we inspected the structure to ensure there was no steric clash with neighboring residues or water molecules. We performed a short energy minimization to relax the modified residue while keeping coordinates of all other residues except for residues immediately before and after the mutated one on the peptide. We then truncated the water box and re-neutralized the system, after which steps from the initial energy minimization up to the final production run followed the same procedure as explained above.

Coordinates were saved every 20 ps (0.02 ns) during production runs, resulting in 50,000 coordinate frames for 1000 ns. We excluded the initial 500 ns when calculating averages and standard deviations in the number of contacts, CDR3 distance, BSA, PCA values, and angle data. Since all systems were simulated for a minimum of 1 µs, this leaves at least 25,000 frames. We report data prior to 500 ns in trajectory plots and contact occupancy heat maps (e.g. Figure 3B–E).