Peptide–MHC (pMHC) binding to a human antiviral T cell receptor induces long-range allosteric communication between pMHC- and CD3-binding sites

T cells generate adaptive immune responses mediated by the T cell receptor (TCR)–CD3 complex comprising an αβ TCR heterodimer noncovalently associated with three CD3 dimers. In early T cell activation, αβ TCR engagement by peptide–major histocompatibility complex (pMHC) is first communicated to the CD3 signaling apparatus of the TCR–CD3 complex, but the underlying mechanism is incompletely understood. It is possible that pMHC binding induces allosteric changes in TCR conformation or dynamics that are then relayed to CD3. Here, we carried out NMR analysis and molecular dynamics (MD) simulations of both the α and β chains of a human antiviral TCR (A6) that recognizes the Tax antigen from human T cell lymphotropic virus-1 bound to the MHC class I molecule HLA-A2. We observed pMHC-induced NMR signal perturbations in the TCR variable (V) domains that propagated to three distinct sites in the constant (C) domains: 1) the Cβ FG loop projecting from the Vβ/Cβ interface; 2) a cluster of Cβ residues near the Cβ αA helix, a region involved in interactions with CD3; and 3) the Cα AB loop at the membrane-proximal base of the TCR. A biological role for each of these allosteric sites is supported by previous mutational and functional studies of TCR signaling. Moreover, the pattern of long-range, ligand-induced changes in TCR A6 revealed by NMR was broadly similar to that predicted by the MD simulations. We propose that the unique structure of the TCR β chain enables allosteric communication between the TCR-binding sites for pMHC and CD3.

T cells play a major role in generating adaptive immune responses to microbes and cancers. This process is mediated by the T cell receptor (TCR) 4 -CD3 complex, which is composed of a genetically diverse ␣␤ (or ␥␦) TCR heterodimer in noncovalent association with three invariant CD3 dimers: CD3⑀␥, CD3⑀␦, and CD3 (1,2). The TCR mediates recognition of peptide fragments bound to major histocompatibility complex (MHC) molecules on antigen-presenting cells (APCs). These peptides are generated by proteolytic degradation of foreign or self-proteins within cells expressing MHC class I or II molecules (3). Following TCR binding to peptide-MHC (pMHC) ligands, the CD3 molecules transmit activation signals to the T cell. X-ray crystallographic studies of TCR-pMHC complexes have revealed the molecular basis for TCR recognition of foreign and self-antigens (4 -8). In addition, much is known about the downstream T cell-signaling cascade, following phosphorylation of immunoreceptor tyrosine-based activation motifs (ITAMs) of CD3 subunits by the Src kinase Lck associated with CD4 or CD8 (9). These ITAMs, which are located at the cytoplasmic face of the TCR-CD3 complex, undergo dissociation from the lipid bilayer upon TCR engagement of pMHC, rendering them accessible to Lck (10 -12). However, the mechanism(s) by which TCR ligation is first communicated to the CD3 signaling apparatus, a process termed early T cell activation, remains largely a mystery (13-15).
A variety of models have been proposed to explain early T cell activation. However, all these models invoke one or more of the following four basic mechanisms: aggregation, segregation, mechanotransduction, and/or conformational change (13,15). Aggregation of TCR-CD3 complexes following TCR engagement could lead to enhanced phosphorylation of CD3 ITAMs by increasing the proximity of associated Lck (13). Bindinginduced segregation of TCR-CD3 complexes away from the inhibitory phosphatase CD45 has also been proposed to explain TCR triggering (16). According to the mechanotransduction model, the TCR converts mechanical energy derived from T cell scanning of APCs into a biochemical signal upon specific pMHC ligation (17).
Other models invoke ligand-induced conformational changes at sites distant from the pMHC-binding site (allostery) as a mechanism for TCR triggering (12,14). A special attraction of allosteric models is that they can account for triggering at low densities of pMHC agonists on APCs. In particular, findings that T cell signaling begins before formation of the immunological synapse (14,18) and that a single pMHC suffices for early T cell activation (19 -21) suggest that triggering must be intrinsic to the interaction of TCR-CD3 with pMHC. However, X-ray crystallographic studies of TCRs in free and pMHC-bound forms have so far failed to identify consistent conformational changes in the TCR C␣ or C␤ domains that could be unambiguously attributed to antigen binding (8). One possibility is that such ligand-induced, long-range conformational changes simply do not occur and that TCR triggering is mediated by other mechanisms. However, another possibility is that the relevant changes may be in protein dynamics, a parameter that cannot be accessed via the static snapshots provided by crystallography. Indeed, recent studies in several systems have demonstrated that ligand binding can alter protein flexibility at distant sites, resulting in long-range transmission of biological signals, even in the absence of crystallographically observed structural changes (22)(23)(24)(25)(26)(27). This process, known as dynamic allostery, provides a unifying mechanism for the general phenomenon of allostery (23,24,27).
Initial evidence for dynamically driven TCR signaling was obtained by hydrogen/deuterium (H/D) exchange, which showed that pMHC binding globally rigidified the TCR (28). More recently, NMR characterization of the ␤ chain of a mouse TCR (B4.2.3) specific for an HIV-1 gp120-derived peptide bound to a mouse MHC class I molecule (H2-D d ) revealed amide resonance changes in C␤ upon pMHC ligation (29). Some of these changes were near constant ␤ domain (C␤) residues previously implicated by NMR to interact with CD3 (30,31). Here, we report NMR characterization and molecular dynamics (MD) simulations of both the ␣ and ␤ chains of a human antiviral TCR (A6) that recognizes the Tax antigen (LLFGYPVYV) from human T cell lymphotropic virus-1 (HTLV-1) bound to the human MHC class I molecule HLA-A2 (32).

NMR assignment of TCR A6[␣-2 H/ 13 C/ 15 N]␤ and A6␣[␤-2 H/ 13 C/ 15 N]
We utilized the antiviral TCR A6 as a representative example of a human MHC class I-restricted TCR. Crystal structures for both the unbound and pMHC-bound states of the A6 ectodomain are known (32)(33)(34), and therefore the binding epitope is well-established and can be reliably compared with results from solution NMR experiments. Moreover, A6 binds Tax-HLA-A2 relatively tightly with a dissociation constant (K D ) of ϳ1 M (33). The TCR A6 ectodomain is an ϳ50-kDa heterodimer consisting of a 204-residue ␣ chain and a 245-residue ␤ chain. We prepared samples with the following labeling patterns, A6[␣-2 H/ 13 C/ 15 N]␤ and A6␣[␤-2 H/ 13 C/ 15 N], and we used transverse relaxation optimized (TROSY)-based triple resonance experiments to assign H N , N, C ␣ , C ␤ , and CO NMR signals in each chain. Backbone amide assignments were obtained for 78% of the nonproline residues in the ␤ chain, accounting for 91% (178/196) of detectable amide peaks (Fig. 1, A and B). For the ␣ chain, backbone amide assignments were determined for 63% of the nonproline residues, corresponding to 74% (122/165) of the detected amide signals (Fig. 2, A and B). In both the ␣ and ␤ chains, unassigned amides were mostly due to either missing amide signals in the 2D 1 H-15 N TROSY-HSQC spectrum or present amide signals but with weak or nonexistent carbon correlations in triple resonance data. In a small number of cases, amide peaks had isolated carbon correlations that could not be linked to assigned regions unambiguously. The TCR A6 assignments for the ␣ and ␤ chains have been deposited in the Biological Magnetic Resonance Bank (BMRB accession codes 27440 and 27441, respectively).

TCR A6 conformation and dynamics in solution
Assigned backbone carbon chemical shifts (C ␣ , C ␤ , and CO) were used to determine the secondary structure elements in solution for TCR A6 ␣ and ␤ chains (Figs. 1C and 2C, upper panel) (35), and these were compared with the structural features obtained from X-ray crystallography (34). In the A6 ␤ chain, 16 of the 18 ␤-strands observed in the crystal structure and the two ␣ helices in C␤, ␣A and ␣B, are identified by chemical shift analysis. Two ␤-strands in the crystal structure, strand CЈ in the variable ␤ domain (V␤) and strand D in C␤, are not recognized as ␤-strands by NMR. This is most likely due to the absence of sufficient unambiguous assignments, although strand D is also not identified by NMR in another human TCR (MS2-3C8) where more assignments were made in this region (30). Furthermore, the two short ␣A and ␣B helical regions in C␤ are not recognized as helices by NMR. For the A6 ␣ chain, 14 of the 16 ␤-strands in the crystal structure are also recognized as ␤-strands by NMR chemical shift analysis. Two secondary structural elements in particular, strand E and helix ␣A near the C terminus of the C␣ domain, were not identified by NMR due to the relatively large number of unassigned residues in this region. Interestingly, this part of the C␣ domain is poorly resolved with missing electron density in a number of TCR crystal structures suggesting increased dynamics (8). Nevertheless, most of the NMR-derived secondary structure elements in TCR A6 corresponded well with the X-ray structure in both the ␣ and ␤ chains, indicating that the conformations in the crystal and solution forms are similar.
In addition to giving important structural information, backbone chemical shifts also provide useful insights into the flexibility of polypeptide chains in the form of order parameters, S 2 (36). For the ␤ chain of A6, secondary structure elements correspond well with high order parameters (S 2 ϳ0.9), consistent with more rigid parts of the TCR structure (Fig. 1C, center panel). In contrast, loops linking secondary structures generally have lower order parameters (S 2 ϳ0.6 -0.8), indicating increased flexibility. Notably, all three complementarity-determining region (CDR) loops (CDR1␤, CDR2␤, and CDR3␤) have relatively low-order parameters, signifying that they are flexible in solution. The exception is the fourth hypervariable loop in V␤, designated HV4␤, which is relatively rigid based on chemical shift derived order parameters. Similar patterns are observed for the ␣ chain. High S 2 values correlate well with secondary structure, and lower S 2 values align with loop regions. As for the ␤ chain, the HV4␣ loop is the least flexible of the hypervariable loops based on chemical shift data. The C-terminal region of the C␣ domain (residues 175-204) appears to be quite dynamic with order parameters in the 0.2-0.7 range, although this section also has assignment gaps as A, extent of backbone assignments for the TCR A6 ␤ chain mapped onto the A6 X-ray structure (PDB accession code 3QH3) (34). Assigned regions of the ␤ chain are shown in wheat; unassigned regions including prolines are shown in gray; and the ␣ chain is in green ribbon format. B, two-dimensional 1 H-15 N TROSY-HSQC spectrum of A6␣[␤-2 H, 13 C, 15 N] with backbone amide assignments. C, TCR A6 ␤ chain secondary structure and dynamics from chemical shift data. Hypervariable regions are highlighted (light red). Top panel, secondary structure elements from chemical shift analysis. Confidence levels from TALOS-N (35) are plotted versus residue number for ␤-strands (black) and ␣-helices (gray). Center panel, chemical shift derived order parameters (S 2 ) from TALOS-N. Bottom panel, crystallographic temperature (B) factors for the C ␣ atoms in the TCR A6 ␤ chain, obtained using PDB 3QH3.

Allosteric changes in T cell receptor induced by peptide-MHC
mentioned above. Chemical shift-based order parameters were further compared with the crystallographic temperature (B) factors (Figs. 1C and 2C, lower panel). Generally, lower S 2 values coincide with higher B-factors for both the ␣ and ␤ chains, although differences between secondary structure and loop regions are more evident in the NMR data. The elevated B-factors in the C-terminal part of the C␣ domain further support the notion of increased chain flexibility in this region. Overall, the chemical shift assignments provided a set of highly self-consistent data on the solution conformation and dynamics of both chains in the TCR A6 heterodimer.

Allosteric changes in T cell receptor induced by peptide-MHC
bound states of TCR A6, monitoring chemical shift perturbations (CSPs) and peak intensity changes upon binding. Spectra of the pMHC complexes with A6[␣-2 H/ 15 N]␤ and A6␣[␤-2 H/ 15 N] indicated that the unbound and bound states were in slow exchange, consistent with the relatively tight binding (K D ϳ1 M) between TCR A6 and Tax-HLA-A2 (33). Addition of unlabeled Tax-HLA-A2 to A6␣[␤-2 H/ 15 N] produced large CSPs for certain residues in the TCR A6 ␤ chain (Fig. 3, A and B), indicating perturbations in local structure. Moreover, whereas all TCR A6 ␤ chain backbone amide resonances broadened upon complex formation with Tax-HLA-A2, some resonances had a disproportionate loss of peak intensity (Fig. 3, A and C), suggesting changes in their main-chain dynamics on the NMR time scale (microseconds to milliseconds).
Spectra of labeled TCR A6/unlabeled Tax-HLA-A2 were recorded over a range of TCR A6 concentrations from 60 to 200 M using 1.2-2.0 eq of pMHC. In all cases, the CSPs and peak intensity changes observed had a similar pattern, showing no indication of a secondary binding site. Consistent with our observations, a previous study on the TCR A6 interaction with Tax-HLA-A2 using analytical ultracentrifugation, dynamic light scattering, and surface plasmon resonance demonstrated that only 1:1 TCR-pMHC complexes were detectable in solution (37). Experimentally significant changes in chemical shifts were ascribed to values of ⌬␦ total greater than or equal to the mean plus 1 S.D., corresponding to values of ϳ0.06 ppm for the ␤ chain and ϳ0.02 ppm for the ␣ chain. By comparison, in control experiments where TROSY-HSQC spectra were recorded on unbound TCR A6 at different concentrations and with different samples, the ⌬␦ total for any given amide signal was less than 0.005 ppm for nearly all residues and in the range of 0.005-0.01 ppm for a few others. This is similar to what was observed for the TCR MS2-3C8 ␤ chain in a previous study (30).
In summary, the NMR binding results indicate that many residues in both the ␣ and ␤ chains of TCR A6 are perturbed upon binding to Tax-HLA-A2. This includes not only amino acids at the crystallographic binding interface but also numerous long-range effects, many of which are located at the interfacial regions between domains. Outside the binding inter-face with Tax-HLA-A2, a significant number of TCR A6 residues with perturbations in one chain are therefore in close contact with affected residues in the other. Thus, the independent Tax-HLA-A2 binding experiments on labeled ␣ and ␤ chains provide complementary information about changes at the A6 heterodimer interface that are self-consistent. For example, perturbed residues Ser-33, Gly-42, Gly-107, and Gln-104 in the A6 ␤ chain are in proximity to perturbed residues Phe-33, Tyr-35, Ser-42, Leu-87, Ala-89, and Phe-100 from the ␣ chain at the V␣/V␤ interface (Fig. 7A). Similar observations were made at the C␣/C␤ interface: residues Phe-129, Ser-132, Glu-135, Thr-143, Val-145, Ser-192, and Arg-196 in the ␤ chain undergo changes upon pMHC binding and are near perturbed residues Lys-130, Leu-134, Thr-136, and Asp-137 in the ␣ chain (Fig.  7B). Connecting these V␣/V␤ and C␣/C␤ regions, a number of residues at the interface between the V␤ and C␤ domains of A6 show changes upon engaging Tax-HLA-A2 (Fig. 7C). Together, these chemical shift and peak intensity perturbations indicate that binding of pMHC to A6 causes a series of structural and possibly dynamic changes in the TCR ectodomain. These changes in conformation and dynamics are likely small amplitude but are nonetheless transmitted from the Tax-HLA-A2-binding site in the V␣ and V␤ domains at one end of the molecule to the C␣ and C␤ domains at the other end. Residues in the ␣A and ␣B helices of C␤ have been previously shown to form at least part of the docking site for CD3 signaling molecules (30,31). Perturbations in NMR signals detected in this study propagate all the way to the CD3-docking site and neighboring residues, indicating a linkage between the pMHC and CD3 sites and suggesting a possible allosteric mechanism for the initiation of T cell signaling. Further support for this hypothesis was obtained from MD simulations.

Molecular dynamics simulations of TCR A6 in unbound and pMHC-bound states
Separate MD simulations were carried out for 1200 ns using unbound and pMHC-bound structures of TCR A6 as starting points. Comparison between unbound and Tax-HLA-A2bound states provided an in silico picture of how pMHC binding affects A6 main-chain flexibility, as measured by the root Figure 4. Effect of Tax-HLA-A2 binding on TCR A6 ␤ chain residues. A, TCR A6 ␤ chain residues with experimentally significant changes (Ն mean plus 1 S.D.) upon addition of Tax-HLA-A2 are highlighted. The A6 color scheme is as follows: ␤ chain (wheat); unassigned ␤ chain residues, including prolines (gray); significant CSPs (cyan); significant peak intensity changes (red); and ␣ chain (green). The portion of the Tax-HLA-A2 at the interface with TCR A6 is shown as a gray ribbon for the MHC and pink for the peptide (PDB code 1QRN) (33). B, surface representation of the TCR A6 ␤ chain in two orientations with the same color scheme as in A.

Allosteric changes in T cell receptor induced by peptide-MHC
mean square deviation (RMSD) and the root mean square fluctuation (RMSF) for the C␣ atom of each residue from the 200-ns time point to the end of the simulations. As seen from plots of RMSD trajectories (Fig. S1), the simulation stabilized around 150 ns and TCR conformations sampled around the crystal structures during the rest of the simulation. Accord-  A, TCR A6 ␣ chain residues with experimentally significant changes (Ն mean plus 1 S.D.) upon addition of Tax-HLA-A2 are highlighted. The color scheme is as follows: ␣ chain (green); unassigned ␣ chain residues including prolines (gray); significant CSPs (purple); significant peak intensity changes (orange); and ␤ chain (wheat). The portion of Tax-HLA-A2 at the interface with TCR A6 is shown as a gray ribbon for the MHC and pink for the peptide (PDB code 1QRN) (33). B, surface representation of the A6 ␣ chain in two orientations with the same color scheme as in A.

Allosteric changes in T cell receptor induced by peptide-MHC
ingly, we selected the conformations after 200 ns for RMSF analysis. The four domains of TCR A6 behaved differently, with fewer RMSF changes in V domains than C domains upon pMHC binding. Unbound C␣ and C␤ domains display similar fluctuations. However, upon pMHC binding, the C␤ domain is stabilized, whereas the C␣ domain is destabilized. Binding of pMHC to TCR A6 decreased the flexibility of most residues in CDR1␤, CDR3␤, CDR1␣, CDR3␣, and most C␤ domain loops. However, CDR2␤ and the C␣ domain became more flexible.
The rigidification effects therefore tend to be larger for CDR loops in the ␤ chain than in the ␣ chain (Fig. 8, A and B). Moreover, the general decrease in flexibility propagates well into the TCR A6 ␤ chain structure, extending through the V␤ domain to the adjacent FG, BC, and DE loops in C␤, and into the C␤ domain itself, including perturbations to ␣A and ␣B at the CD3-docking site (Fig. 8C) (30,31). In contrast, the MD simulations predict that pMHC binding causes a combination of both increased and decreased flexibility in the TCR A6 ␣ chain. Thus, although there is rigidification of CDR1␣ and CDR3␣ and other parts of the V␣ and C␣ domains upon pMHC ligation, there is also increased flexibility in CDR2␣ and the V␣-C␣ linker. Overall, the most notable feature from the MD simulations is that pMHC binding induces extensive long-range effects on TCR A6 flexibility in both chains but particularly the ␤ chain. Indeed, the largest and most extensive RMSF differences observed (1-2 Å) correspond to increased rigidity and are located in the FG loop of the C␤ domain, far from the pMHCbinding site (Fig. 8C). In a similar case, a recent MD simulation of single-chain Fv antibodies found that the linker between the V H and V L domains exhibited the dominant dynamical response by being coupled to nearly the entire protein (38). Our MD simulations are consistent with previous H/D exchange experiments showing that pMHC binding induced a global reduction in TCR A6 flexibility that included the C␤ FG loop (28).
To compare chemical shift changes from MD simulation and NMR, we used ShiftX2 (39) to predict chemical shifts of unbound (5000 conformations) and pMHC-bound TCR A6 (5000 conformations) using the conformations obtained from the last 500 ns of the 1.2-s simulations. There are excellent correlations between calculated and experimentally assigned chemical shifts for unbound TCR: R 2 ϭ 0.89 and 0.93 for ␣-carbon atoms in the ␣ and ␤ chains, respectively; the corresponding correlations for carbonyl carbon atoms are R 2 ϭ 0.43 and 0.45. We compared the chemical shift change of amide nitrogen atoms, which are only assigned for pMHC-bound TCR. Although there were no quantitative correlations between simulations and NMR, we could reproduce the largest chemical shift perturbation of Gly-40 in the TCR ␤ chain (Fig. S2).
Comparison between the MD and NMR results indicates that the pattern of long-range effects on TCR A6 is broadly similar, despite differences in the time scales over which observations were made. Thus, changes in RMSFs out to 1.2 s in the MD simulations exhibit comparable effects propagating into the C domain to those detected by NMR on the longer microseconds-milliseconds time scale. It should be noted that the motions seen on the MD time scale employed here will not lead to line broadening. In analogy with the MD simulations, NMR perturbations in the A6 TCR ␤ chain extend from V␤ to C␤ and include changes in residues from the FG loop and adjacent regions (Fig. 4A). NMR shift perturbation effects are smaller for CDR loops in the ␣ chain (Fig. 5B) than the ␤ chain (Fig. 3B), in parallel with the reduced ⌬RMSF values seen for the ␣ chain CDR loops in MD simulations, suggesting that the TCR ␤ chain is the major source of propagating long-range effects (see "Discussion").
In summary, pMHC binding leads to long-range changes in TCR A6 conformation and/or flexibility. The patterns of perturbations indicated experimentally by NMR and predicted computationally by MD simulations overlap to a significant degree. Changes induced by pMHC ligation extend from the binding site at one end of the TCR molecule to a region near the transmembrane helices and the docking site for CD3 molecules (30,31), suggesting a plausible pathway for allosteric control of T cell signaling. We propose that signals are relayed from the TCR V domains to three key allosteric sites in the C domains: 1) the unique 16-residue C␤ FG loop that projects from the V␤/C␤ interface; 2) a cluster of C␤ residues in or near the C␤ ␣A helix, a region that interacts with CD3 (30, 31); and 3) the C␣ AB loop located at the membrane-proximal base of the

Allosteric changes in T cell receptor induced by peptide-MHC
TCR. The biological relevance of each of these allosteric sites to signal transduction is supported by previous mutagenesis and functional studies using transfected or transgenic T cells (29, 40 -42), as discussed below.

Analysis of domain interface residue conservation and energetics
To identify conserved interdomain contacts that may mediate allosteric communication between TCR V and C domains, we collected and analyzed sets of nonredundant ␣ and ␤ chain structures from the Protein Data Bank (PDB) (Tables S1 and S2) (43). To avoid over-representation of V domain residues due to biased V␣ and V␤ gene usage in X-ray structures, these sets are nonredundant by TRAV or TRBV gene used. These were analyzed for residues frequently observed at V␣/C␣ or V␤/C␤ domain interfaces (present in Ͼ90% of structures), and computational alanine scanning was used to assess relative impact of side chains on interdomain assembly (Table S3). In V␣/C␣ interfaces (Fig. 9A), a glutamate residue (Glu-14 in TCR A6) is present in 93 and 94% of human and mouse germline genes, respectively (Fig. S3), and is proximal to a lysine residue at position 168 in C␣. Lys-168 is present in human and mouse TCRs, suggesting the possibility of a conserved salt bridge interaction. However, the average energetic impact of these residues on domain interactions is not predicted to be high in the Rosetta scoring function used (44). In contrast with V␣/C␣ interfaces, a larger set of residues was observed in V␤/C␤ interfaces (Table  S3), several of which have major calculated effects on interdomain interface energy. These include a cluster of five co-located hydrophobic residues (Fig. 9B), of which two are near the N terminus of the V␤ domain. Although these V␤ residues, located at positions 10 and 12, are not fully conserved among TRBV germline genes (Fig. S2), they are predominantly (position 10) or always (position 12) hydrophobic. This cluster of hydrophobic residues in the ␤ chain overlaps with positions associated with pMHC binding perturbation in NMR, specifically Leu-118 and Tyr-216 (Fig. 7C), and is proximal to residues Thr-115 and Tyr-152, which likewise showed NMR perturba-tions upon pMHC binding and exhibited some calculated impact on domain interface energetics (Table S3).
An analysis of V␣/V␤ interface residues based on all functional germline TRAV, TRAJ, TRBV, and TRBJ gene sequences (human and mouse) downloaded from the IMGT database (45) revealed additional highly conserved positions that potentially contribute to interdomain communication. V␣ residues in the V␣/V␤ interface that undergo signal perturbations include the following: Tyr-35 (94% identical in human TRAV sequences; 80% identical in mouse TRAV sequences); Ala-89 (80 and 92%); Phe-100 (96% identical in human TRAJ sequences; 95% identical in mouse TRAJ sequences); and Gly-103 (100 and 100%). Perturbed V␤ residues in the interface with V␣ include: Gly-40 (76% identical in human TRBV sequences; 44% identical in mouse TRBV sequences); Gly-42 (76 and 39%); Gln-104 (50% identical in human TRBJ sequences; 33% identical in mouse TRBJ sequences); and Gly-107 (100 and 100%). Additionally conserved V␣/V␤ interface residues include Gln-37 on the ␣ chain (91% conserved in human and 92% conserved in mouse), which interacts with Gln-37 on the ␤ chain (98 and 96%), in   (Table S1). B, structures shown are 27 ␤ chain X-ray structures (Table S2). The structures are nonredundant by TRAV or TRBV gene usage and are superposed by V␣ or V␤ domains. The orientation between V␣ and C␣ domains displays greater variability than that between V␤ and C␤ domains. Selected frequently observed V/C interface residues are shown as sticks and labeled by residue number (based on A6 TCR numbering). Highly conserved residues are labeled by predominant or sole amino acid(s) observed at that position in structurally characterized TCRs and human and mouse germline genes.

Allosteric changes in T cell receptor induced by peptide-MHC
addition to residues Tyr-35 and Phe-106 on the ␤ chain, which are both fully conserved in human and mouse. Of note, the interaction between residues at position 37 was previously noted by others as a key mediator of V␣/V␤ positioning (46).

Discussion
The NMR and MD results presented here indicate that pMHC binding has long-range effects on TCR conformation and dynamics. Similar long-range stabilization effects on TCR upon pMHC ligation have been observed using H/D exchange measurements (28). Taken together, the NMR, MD, and H/D exchange results illustrate that comparable changes in TCR chain flexibility have been predicted or detected experimentally over a range of time scales from hundreds of nanoseconds to seconds or longer. Importantly, similar observations have been made in other protein systems displaying dynamic allostery (e.g. catabolite activator protein, 20S core particle proteasome, and PDZ proteins), in which long-range transmission of biological signals was shown to occur via alterations in protein flexibility at distant sites upon ligand binding (22)(23)(24)(25)(26)(27). Dynamic allostery could also explain the exceptional ability of a single pMHC molecule binding to an exclusively monomeric TCR to activate T cells, without any requirement for ligand-induced receptor dimerization or oligomerization (21).
We propose that the unique structure of the TCR ␤ chain enables allosteric communication between the TCR-binding sites for pMHC and CD3. In sharp contrast to V␣ and C␣ domains (and to antibody V L and C L or V H and C H 1 domains), V␤ and C␤ domains are in intimate contact in all TCR structures determined to date (8). The average total buried surface area between V␤ and C␤ is 1007 Ϯ 105 Å 2 for 27 nonredundant ␤ chains analyzed here (1123 Å 2 for A6), compared with 539 Ϯ 71 Å 2 for V␣ and C␣ for 28 nonredundant ␣ chains (522 Å 2 for A6). Corresponding buried surfaces between antibody V H and C H 1 domains, or between V L and C L domains, vary from about 200 to 450 Å 2 , depending on the elbow angle. Extensive and highly conserved interactions between V␤ and C␤ (see above) impose on the TCR ␣␤ heterodimer an overall rigid structure that is largely devoid of flexibility in the region homologous to the elbow of an antibody Fab fragment. This may have important implications for signal transduction, because a rigid conformation could facilitate transmission of any allosteric changes that may occur in the TCR upon binding pMHC to associated CD3 molecules in the TCR-CD3 complex (47).
In support of this idea, we observed experimentally significant CSPs or losses of peak intensity for numerous residues in the tightly packed V␤/C␤ interface, including Thr-115, Leu-118, Lys-119, Phe-122, Phe-151, Tyr-152, Tyr-216, and Glu-220, which would provide an obvious path for allosteric signal transmission from V␤ to C␤. By contrast, considerably fewer NMR signal perturbations were detected for residues in the much smaller and loosely packed V␣/C␣ interface, making allosteric signaling across this interface less likely. Accordingly, we propose that spectroscopic changes in V␣ upon Tax-HLA-A2 binding are relayed to C␣ primarily through the ␤ chain rather than across the V␣/C␣ interface.
In addition to the close juxtaposition of V␤ and C␤, another distinctive structural feature of TCR ␤ chains is a unique 16-residue FG loop in C␤ (residues 216 -231 in TCR A6) that protrudes between the V␤ and C␤ domains (Fig. 4A). This elongated FG loop, which is not present in C␣ or in antibody C L or C H 1 domains, has been implicated in TCR signaling (48). In particular, deletion of the C␤ FG loop in TCR transgenic mice, while maintaining TCR-CD3 expression and copy number on the surface of T cells, dramatically reduced their sensitivity to activation by cognate peptide (40,41). MD simulations of TCR A6 in unbound versus pMHC-bound states revealed that the C␤ FG loop is sensitive to ligand binding (Fig. 8A). Consistent with MD results, NMR analysis revealed significant signal perturbations for C␤ FG loop residues Tyr-216, Glu-220, Gln-226, and Asp-227 upon TCR engagement of Tax-HLA-A2 (Fig. 7C).
Unlike the TCR, which comprises variable ␣ and ␤ chains, the pre-TCR consists of a variable TCR ␤ chain paired with an invariant pre-TCR ␣ chain (pre-T␣). Signaling through the pre-TCR during early T cell development terminates TCR␤ rearrangements and triggers TCR␣ rearrangements, leading to assembly of ␣␤ TCR heterodimers (49). Ligands for the pre-TCR are believed to be self-antigens on thymic stroma, possibly pMHC (50). The pre-T␣ chain consists of a single immunoglobulin-like domain that interacts exclusively with the C␤ domain, leaving the V␤ domain unpaired (51). Despite the absence of V␣, the overall conformation of the TCR ␤ chain in the pre-T␣/TCR␤ heterodimer is essentially identical to that in the TCR. Most notably, the intimate association between V␤ and C␤ described above, including specific contacts across the V␤/C␤ interface, is maintained, suggesting that allosteric signals may traverse similar pathways in TCR and pre-TCR ␤ chains.
In a previous study, site-directed fluorescence labeling of a human TCR (LC13) specific for an Epstein-Barr virus peptide presented by HLA-B8 identified a discrete conformational change in the AB loop of the C␣ domain (residues 123-130 in TCR A6) upon pMHC ligation (42). Mutagenesis of residues within the C␣ AB loop, which is located at the base of the TCR proximal to the membrane (Fig. 6A), impaired antigen-specific triggering of LC13, suggesting a functional role for this loop in early T cell activation. Unlike fluorescence-based methods, NMR does not require the introduction of bulky fluoroprobes that may perturb protein conformation or dynamics. It is therefore noteworthy that NMR analysis of TCR A6 revealed signal perturbations for C␣ AB loop residues Ser-128, Asp-129, and Lys-130 upon engagement of Tax-HLA-A2 (Fig. 6A), in agreement with fluorescence-based results for LC13 (42). Moreover, the fact that pMHC binding induces a conformational and/or dynamic change at the same C␣ site in LC13 and A6 implies a conserved allosteric mechanism, given that these TCRs recognize different ligands using unrelated V␣/V␤ gene combinations.

Allosteric changes in T cell receptor induced by peptide-MHC
residues within this cluster reduced antigen-dependent activation of B.4.2.3 without affecting cell-surface expression, pMHC affinity, or thermal stability of the TCR, suggesting a functional role for this allosteric site in TCR signaling (29). Importantly, the C␤ ␣A helix has been identified by NMR as a docking site for CD3 subunits (30,31). On the C␣ side of the C␣/C␤ interface, Leu-134, Phe-135, Thr-136, and Asp-137, which are in proximity to perturbed C␤ residues near the C␤ ␣A helix, also undergo amide resonance changes in the bound state (Fig. 7B).
Although allosteric propagation of biological signals over long distances has been demonstrated for many proteins, including cell-surface receptors such as the EphA2 receptor (52,53), the enormous diversity of TCRs poses a major conceptual challenge for allosteric models of T cell activation. How can conserved structural or dynamic changes in the TCR C␣ and C␤ domains, such as those in the C␣ AB loop and C␤ ␣A helix discussed above, be generated, given the diversity of TCR-pMHC binding interfaces and of V␣ and V␤ sequences? At least part of the answer appears to be that perturbations in the highly variable CDR loops caused by pMHC ligation are relayed to the C domains through the V␣/V␤ and V␤/C␤ interfaces, whose structures are relatively conserved across different TCRs (8), as described above. Indeed, this pathway is supported by NMR results for both TCRs A6 and B4.2.3 (29), in which numerous residues in the V␣/V␤ and V␤/C␤ interfaces experience CSPs and peak intensity changes. Additionally, an analysis of TCR binding conformational changes through comparison of 20 unbound and pMHC-bound TCR crystallographic structure pairs found notable conformational changes centered at ␤ chain residue Gly-40 (54), a V␣/V␤ interface residue with significant NMR signal perturbation in this study.
Although it may be intellectually appealing to search for a single molecular mechanism to explain early T cell activation, the long evolution of the TCR-CD3 complex may have led to the emergence of several distinct, yet interdependent, mechanisms (15). For example, our complementary NMR and MD analyses of TCR A6 have demonstrated that the C␤ FG loop is sensitive to allosteric perturbation upon pMHC binding. In the mechanotransduction model of TCR triggering, mechanical force arising from movement of the T cell relative to the APC is thought to be transferred to CD3⑀␥ via the C␤ FG loop (55). Conformational selection related to the dynamics of this loop could potentiate interactions with CD3⑀␥, which NMR has shown to interact only weakly with unbound TCR (30,31), thereby linking dynamic allostery with mechanotransduction. Indeed, several cooperative TCR triggering mechanisms may be required to activate different T lymphocyte subsets (CD4 ϩ and CD8 ϩ T cells, NKT cells, and regulatory T cells) at different anatomical sites (thymus and secondary lymphoid organs) and different stages of differentiation (naive, mature, and memory).

NMR spectroscopy
NMR spectra were acquired on Bruker AVANCE III 600 and 900 MHz spectrometers using Z-gradient 1 (57), and analyzed with SPARKY (58). Analysis of secondary structured regions using chemical shift data were obtained from TALOS-N (35). Protein structures were displayed and analyzed using PyMOL (Schrodinger).
Binding experiments were performed at 30°C utilizing 80 M TCR A6 samples, labeled at either the ␣ chain or ␤ chain, and unlabeled Tax-HLA-A2. Prior to mixing, a two-dimensional 1 H-15 N TROSY-HSQC spectrum was acquired for each A6-labeled sample as a control. Tax-HLA-A2 was exchanged into buffer conditions matching those of the A6 samples. Unlabeled Tax-HLA-A2 (1.2-2.0 molar eq) was added to 200 l of 80 M labeled A6 and concentrated back to 200 l using membrane filtration. A 2D 1 H-15 N TROSY-HSQC spectrum was then recorded for pMHC-bound A6␣[␤-2 H/ 13 C/ 15 N] and pMHCbound A6[␣-2 H/ 13 C/ 15 N]␤. Bound state peaks were assigned conservatively by inspection utilizing the assignments for the unbound state. Chemical shift perturbations between free and pMHC-bound states of TCR A6 were determined from ⌬␦ total ϭ ((W H ⌬␦ H ) 2 ϩ (W N ⌬␦ N ) 2 ) 1/2 , where ⌬␦ H and ⌬␦ N represent 1 H and 15 N chemical shift changes, respectively, between free and pMHC-bound states of A6. Weighting factors of W H ϭ 1 and W N ϭ 0.2 were used. Values of ⌬␦ total greater than or equal to the mean plus 1 S.D. were considered to be experimentally significant. This is a conservative estimate based on errors in TROSY-HSQC chemical shift measurements in multiple identical A6 samples and also in previous TCR samples (30). The loss of peak intensity for TCR A6 resonances upon Tax-HLA-A2 binding was measured using SPARKY (58). Experimentally significant changes were given by values greater than or equal to the mean percent loss in intensity value plus 1 S.D.

Molecular dynamics simulations and analysis
The experimental X-ray crystal structures of the complex between Tax-HLA-A2 and TCR A6 were employed as starting points for the MD simulations. The D/E chains of A6 from the A6 X-ray structure in complex with Tax-HLA-A2 (PDB code 1QRN) (33) was used for the simulation of unbound TCR. The complex between a high-affinity version of A6 (A6c134) and Tax-HLA-A2 (PDB code 4FTV) was used as a homology template to prepare the TCR-pMHC structure by mutating in silico the corresponding ␤ chain residues back to TCR A6 (A6c134/A6: Ala/Met-98, Gly/Ser-99, Gly/Ala-100, Arg/Gln-101, Ala/Glu-133, Glu/Ala-133, and Asn/Asp-204). MD simulations verified that TCRs A6 and A6c134 have almost identical dynamics (data not shown). The starting structures were then solvated by TIP3P water molecules, and sodium and chloride ions were added to neutralize the system and to achieve a total concentration of ϳ150 mM. Box sizes for the TCR-pMHC complex were 156 ϫ 156 ϫ 152 Å 3 with 307,057 atoms and 80 ϫ 93 ϫ 84 Å 3 with 59,085 atoms. The resulting solvated systems were energy-minimized for 5000 conjugate gradient steps, with the protein fixed, whereas water molecules and counterions were allowed to move. This was followed by an additional 5000 conjugate gradient steps, where all atoms were allowed to move. In the equilibration stage, each system was gradually relaxed by performing a series of dynamic cycles. In the production stage, all simulations were performed using the NPT ensemble at 300 K for 300 ns. All MD simulations were performed using the NAMD program (60) with CHARMM27 force field with CMAP correction (61,62). This force field is known to substantially overstabilize helical structures, and the effect on these simulations is unknown (63). The short-range van der Waals interactions were calculated using the switching function, with a twin range cutoff of 10.0 and 12.0 Å. The long-range electrostatic interactions were calculated with the Particle Mesh Ewald method with a cutoff of 12.0 Å. To minimize the effects caused by crystal packing, the initial 200-ns simulations were not used in RMSF calculations. Thus, the conformations generated from the last 1000 ns of 1200-ns MD simulation trajectories (10000 frames for each simulation) were used for RMSF analysis. To project out the effect of interdomain motion on RMSF values, each domain (V␣, V␤, C␣, and C␤) was superimposed separately. The averaged structures from the last 1000 ns were used as reference. The conformers generated from 500-ns simulations (5000 frames) were used for NMR chemical shift calculation using the ShiftX2 algorithm (39).

Collection and analysis of TCR structures
TCR ␣ and ␤ X-ray structures were obtained from the RCSB PDB (43). Structures were retained based on X-ray structure resolution (Ͻ3.0 Å), the presence of both V and C domains, and the lack of missing residues near V/C domain interfaces. Human and mouse TRAV and TRBV germline amino acid reference sequences were obtained from the IMGT database (45) and were used to assign TRAV and TRBV genes to each structure and remove redundancies. Furthermore, chimeric human-mouse TCR chains were removed. Structures were renumbered based on the TCR A6 structure (PDB code 3QH3) (34), and interdomain interface residues were identified based on a 5 Å distance cutoff to any atom in the other domain. For this analysis, the V␣ C terminus was defined as residue 111, and the V␤ C terminus was residue 117. To account for V␣/C␣ and V␤/C␤ interface residues, only positions Ͼ2 residues apart in sequence were used for interface residue determination. Computational alanine scanning was performed using the "interface" protocol in Rosetta version 2.3 (44), which was used to model alanine point mutations in each structure and calculate the interdomain binding energy changes (⌬⌬G).
Prior to computational alanine scanning, all structures were subjected to constrained minimization using the FastRelax protocol (64) in Rosetta version 3 (weekly release 2017.29), to alleviate any minor structural defects from the X-ray structures that would potentially bias energetic calculations. Example command line flags used for FastRelax are as follows: This performs computational alanine scanning on an input TCR structure, 3QH3.pdb, with the set of mutations specified in the file "3QH3.muts.txt". The format of that file is shown in the following example, which specifies three alanine point mutations in the D chain (positions 14, 79, 80):

TCR germline sequence analysis
TRAV, TRBV, TRAJ, and TRBJ germline amino acid reference sequences were obtained from the IMGT database (45). Only functional TRAV and TRBV genes based on IMGT annotation were used, and only the first allele for each gene was included, to avoid over-representation of genes with multiple alleles. This resulted in 45 human and 108 mouse TRAV sequences, 48 human and 22 mouse TRBV sequences, 50 human and 39 mouse TRAJ sequences, and 13 human and 11 mouse TRBJ sequences. TRAV and TRBV sequences were aligned by IMGT, and TRAJ and TRBJ sequences were aligned using CLUSTAL (65). Graphical representation of TRAV and TRBV sequence propensities was performed using the WebLogo tool (59).