NMR-directed design of pre-TCRβ and pMHC molecules implies a distinct geometry for pre-TCR relative to αβTCR recognition of pMHC

The pre-T cell receptor (pre-TCR) guides early thymocytes through maturation processes within the thymus via interaction with self-ligands displayed on thymic epithelial cells. The pre-TCR is a disulfide-linked heterodimer composed of an invariant pre-TCR α (pTα) subunit and a variable β subunit, the latter of which is incorporated into the mature TCR in subsequent developmental progression. This interaction of pre-TCR with peptide-major histocompatibility complex (pMHC) molecules has recently been shown to drive robust pre-TCR signaling and thymocyte maturation. Although the native sequences of β are properly folded and suitable for NMR studies in isolation, a tendency to self-associate rendered binding studies with physiological ligands difficult to interpret. Consequently, to structurally define this critical interaction, we have re-engineered the extracellular regions of β, designated as β-c1, for prokaryotic production to be used in NMR spectroscopy. Given the large size of the full extracellular domain of class I MHC molecules such as H-Kb, we produced a truncated form termed Kb-t harboring properties favorable for NMR measurements. This system has enabled robust measurement of a pre-TCR–pMHC interaction directly analogous to that of TCRαβ–pMHC. Binding surface analysis identified a contact surface comparable in size to that of the TCRαβ–pMHC but potentially with a rather distinct binding orientation. A tilting of the pre-TCRβ when bound to the pMHC ligand recognition surface versus the upright orientation of TCRαβ would alter the direction of force application between pre-TCR and TCR mechanosensors, impacting signal initiation.

␣␤T cell-mediated immunity serves to combat cellularbased pathologies that afflict the vertebrate host, including those caused by viruses and cancers. At the core of T cell recognition and activation function is the surface T cell receptor (␣␤TCR), 2 consisting of a clone-specific TCR␣␤ heterodimer and invariant dimeric CD3 signaling subunits (CD3⑀␥, CD3⑀␦, and CD3) that are non-covalently associated to form an eight-transmembrane-component complex. TCR␣␤ recognizes "foreign" peptides derived from proteins unique to a pathogenic process such as a viral protein or mutated host gene product within a transformed cell. A key feature of this recognition is that a peptide is presented in the context of peptidemajor histocompatibility complex (pMHC) molecules on the surface of the abnormal cell or cross-presented on dedicated antigen-presenting cells that acquire the protein and surfacearray the peptide. The pathogenically derived foreign pMHCs must be recognized in the context of tens of thousands of self-pMHCs generated as a result of the breakdown of endogenous proteins that are displayed on the same host cell. This discrimination is tuned through the harnessing of external cellular forces generated as a T cell scans its environment for foreign pMHC during immune surveillance as well as through the actuation of intracellular cytoskeletal forces that dynamically tune immune recognition (1)(2)(3)(4)(5). Indeed, it appears that the T cell utilizes piconewton-level forces to initiate signaling-linked structural transitions that extend TCR-pMHC bond lifetime (1,4,6,7), although the atomistic details of this are yet to be fully resolved.
The requirement for exquisite ligand discrimination, to selectively sense foreign pMHC over self-pMHC, drives selection processes in thymocyte development to eliminate ␣␤TCRs with excess self-reactivity without removing those harboring the potential to bind foreign peptides. This is no simple task as a protective ␣␤TCR should recognize self-MHC molecules while distinguishing foreign peptides from self-peptides in the groove of the self-MHC molecule. This is more difficult consid-This work was supported by National Institutes of Health Grants P01GM04746 and R01AI37581 and National Institute of Biomedical Imaging and Bioengineering Grant P41-EB002026 (to G. W.) and R01AI19807 (to E. L. R.) and R01AI100643 (to E. L. R. and M. J. L.). The authors declare that they have no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This article contains Fig. S1. NMR resonance assignments for VSV8/K b and VSV8/K b -t were deposited in the Biological Magnetic Resonance Data Bank (BMRB) with accession numbers 27285 and 27286, respectively. 1 To whom correspondence should be addressed. E-mail: gerhard_ wagner@hms.harvard.edu.
ering that at the TCR-pMHC interaction surface the TCR makes many more atomic contacts with the MHC than with the peptide. The pre-TCR pT␣-␤ heterodimer is an obligate precursor to TCR␣␤ in the development of ␣␤ T cells (8). Pre-TCRs express the somatically rearranged TCR␤ gene product, the ␤ subunit, paired with an invariant pT␣ subunit, itself comprising a single C␣-like constant domain and lacking a variable domain and associated complementarity-determining regions (CDRs), which drive ligand recognition within TCR␣␤. Nevertheless, the pre-TCR is itself a fully functional receptor that includes the same CD3 subunits to initiate signaling (6,9). Both the CDR loops of V␤ and a hydrophobic patch on V␤, exposed as a consequence of being in an unpaired V module topology, mediate ligand binding to pMHC (9). The self-pMHC interaction specificity of the pre-TCR drives expansion of the CD4 Ϫ CD8 Ϫ double-negative thymocytes (6,9) and appears to foster preferential expansion of those thymocytes that recognize peptides in the context of self-MHC. Furthermore, the relaxed peptide specificity of the pre-TCR relative to the ␣␤TCR that appears at the CD4 ϩ CD8 ϩ double-positive thymocyte stage has also been observed and is consistent with the notion of pre-TCR introducing a self-MHC bias (6,10,11). Following pre-TCR-driven ␤ chain selection, the ␣␤TCR replaces the pre-TCR, and each TCR clonotype rigorously selects for a precise peptide specificity using its paired V␣V␤ module. Desirable TCRs undergo positive selection (survival), whereas harmful (strongly autoreactive) TCRs undergo negative selection (deletion) at the CD4 ϩ CD8 ϩ double-positive stage and/or further developmental stages (8,(12)(13)(14).
Because the pre-TCR plays a major role in this two-step T cell repertoire refinement, we sought to characterize the binding event between ␤ and pMHC to decipher the parameters of this recognition relative to those of the TCR␣␤-pMHC interaction (9). More importantly, we would like to delineate differences between the pre-TCR and the ␣␤TCR recognition events to better understand the biological function of the pre-TCR while providing new insights into TCR function (3,4,6). Due to inherent self-association of component molecules (15) and the large size of the complex, attempts to fully define the interactions have been difficult (9,15). We therefore designed a model system to explore the pre-TCR-pMHC interaction in atomic detail. Experiments presented here establish a working model of the pre-TCR-pMHC interaction through the design of a modified ␤ chain (9) and a truncated mutant of pMHC (16) to describe the pre-TCR-pMHC interaction surfaces.

Interaction of pMHC with the ␤ subunit of the pre-TCR is hampered by self-association of ␤ through the constant domain
In an attempt to characterize an interaction between a ␤ chain and a pMHC, we utilized two distinct ␤ chains with only 34% sequence identity within their V␤ regions, N30␤ and N15␤, for which the ␣␤TCR of each recognizes the same pMHC, a vesicular stomatitis virus nucleoprotein octapeptide bound to H-2K b (VSV8/K b ) (9,17,18). Each ␤ subunit pro-motes ␤ selection in thymic culture systems (9) and could potentially provide complementary information on the pre-TCR-pMHC complex. We first added unlabeled VSV8/K b to a purified 15 N-labeled sample of N30␤ and acquired 1 H-15 N HSQC spectra with transverse relaxation optimized spectroscopy (TROSY) selection of the mixture and compared it with a spectrum of N30␤. Utilizing the backbone assignments of N30␤ (18), we were able to determine the interaction site by mapping the combined 1 H-15 N chemical shift changes (CCSCs) on a previously determined X-ray crystallographic model (15) (Fig. 1, A and B). The largest chemical shift changes were not localized to any single site on the surface of N30␤ but did cluster in the C domain at the conserved C␣-or pT␣-C␤ interface (19,20). Additional changes were seen in the C domain AB and EF helices, which are located proximal to the C terminus of the TCR. There were some scattered changes within the V domain, but these were not clustered as tightly as those identified in the C domain. This result suggested that any V domain interaction was weaker than the nonspecific interactions within the C domain, which is not known to interact with pMHC in the TCR␣␤, making it unlikely that we were observing physiologically relevant interactions. To understand whether the C domain-mediated nonspecific interactions were a general phenomenon, we next studied the interaction of pMHC with N15␤. Because N15␤ is known to self-associate (15) via V␤, we used cross-saturation transfer (CST) NMR experiments, which highlight only regions of heteromolecular association, in contrast to chemical shift perturbation analyses, which also highlight changes in surfaces due to self-association or allosteric changes. In CST experiments, the observed protein is 15 N-labeled and perdeuterated, which implies that it will have no resonances in the aliphatic region of the 1 H NMR spectrum. The binding partner, in this case pMHC, will be protonated and will harbor resonances in the aliphatic region, but because this protein is not 15 N-labeled it will be invisible in a 15 N HSQC spectrum. In CST experiments, we selectively magnetize the unlabeled protein and monitor transfer of magnetization to the labeled protein, which will reveal the direct interaction interface. In the case of N15␤, we found a significant CST effect in the V␤ as well as C␤ (Fig. 1, C and D). Because the pre-TCR contains a pT␣ subunit that lacks a V␣-like subunit ( Fig. 2A), we posit that the changes seen in the V domain were more physiologically relevant than those seen in the C domain (20). Fig. 2A shows the overall topology of the pre-TCR pT␣-␤ heterodimer (20), illustrating that the interaction surfaces highlighted in the N15␤-VSV8/K b interaction are occluded by pT␣ in the C domain but are not occluded in the V domain ( Fig. 1, C  and D). When examining the crystallographic model of the pre-TCR, we noted a hydrophobic cluster, including residues Phe-131, Val-147, and Leu-149 that are occluded in the pre-TCR (Fig. 2B). We hypothesized that mutating the equivalent residues in N15␤ or N30␤ would abrogate the interaction within this non-physiologically relevant binding surface. It was thus clear that to accurately judge the binding capacity of TCR␤, measuring only interaction potential of elements exposed on cell surfaces, we needed to remove the non-physiologic inter-

Pre-TCR-pMHC interaction
action within the C␤ interface normally contacting C␣ or pT␣ (19,20). Due to the placement of the hydrophobic residues Phe-131, Val-147, and Leu-149 in ␤-sheet secondary structures, we chose replacements with known ␤-strand-forming propensities (21), F131R, V147Q, and L149Q, each of which places a charged or hydrophilic residue at this interface.

Mutagenesis of C␤
The 15 N HSQC spectrum of the N30␤ mutant F129R/ V145Q/L147Q, denoted N30␤-c1, is quite similar to that of the WT (Fig. 2C) as is the case with N15␤ F128R/V144Q/L146Q (N15␤-c1) compared with WT ( Fig. 2D), indicating an overall similar fold of the molecule. When examining residues in the C␤ domain of N30␤-c1, these resonances are shifted compared with those in N30␤, whereas those in the V domain remain unchanged (Fig. 2E), indicating that any changes due to the mutation are local and do not affect V domain-binding regions. We conclude that the replacement of the Phe/Val/Leu hydrophobic cluster resulted in a ␤ chain with a fold similar to that of the WT (9, 18).

Rotational correlation time ( c ) measurements show diminished self-association in ␤-c1
Using NMR, one can measure relaxation rates from which molecular correlation times such as c can be extracted. The c may report on the oligomeric states of the observed protein, in solution, at the given concentration and in equilibrium where the interacting partners are not distinguished. Utilizing transverse relaxation correlation time (TRACT) experiments (22), we were able to measure c values, which are positively correlated to the apparent size of a molecule, inclusive of oligomeric state, for N30␤ and N15␤ (Fig. 2F). c increased with increasing  x and y to highlight the contact surface. Side chains of residues contacting (i.e. within 4 Å of) pT␣ are shown in stick representation. Three hydrophobic residues targeted for mutagenesis are colored yellow. Numbering is according to that found in the crystal structure 3OF6 and is offset from those of N15 and N30␤ by three or two residues, respectively. C and D, overlay of 1 H-15 N TROSY-HSQC of N30␤ (18) and N30␤-c1 (9) (C) or N15␤ (18) and N15␤-c1 (9) (D) to illustrate similarity of spectra. E, region of overlaid spectra shown in C with labeled peaks indicating lack of chemical shift changes in the V domain (Thr-6, Thr-32, Asp-25, Leu-111, and Leu-117), whereas those in the C domain (Val-125, Tyr-184, Arg-189, Gln-209, and Phe-210) show significant changes. F, rotational correlation times of WT and mutant TCR␤s suggest abrogation of self-association within C␤. The plot shows c versus concentration for N15 (blue diamonds) and N30␤ (red triangles). Lines show exponential fits of points to guide the eye. WT is shown with filled symbols and solid lines, and c1 mutants are shown with open symbols and dotted lines.

Pre-TCR-pMHC interaction
concentration for each protein, ranging from 18 to 23 ns for N30␤ at 50 -500 M, whereas N15␤ varied from 23 to 32 ns over similar concentrations. At the lowest concentrations measured here, the c was 18.6 and 22.7 ns. These values were greater than the predicted value of 16.3 ns for these 27-kDa proteins as calculated using the software HYDRONMR (23). Although we cannot rule out self-association via V␤, we posit that at least part of this self-association is the result of intermolecular interactions of C␤ (15). N15␤ self-associates more than N30␤ (p ϭ 0.028), supporting the idea that N15␤ utilizes the V domain more in self-association. c for N30␤-c1 was found to be 15.2 and 15.8 ns for 100 and 200 M samples, respectively (Fig. 2F). This is consistent with essentially monomeric TCR␤ as determined from hydrodynamic radius calculations using HYDRONMR (23) wherein a value of 16.3 ns is predicted for the 27-kDa monomer. This supports the idea that N30␤-c1 exists as a monomer at these concentrations and self-associates significantly less than WT (p ϭ 0.035). In contrast, N15␤-c1 exhibits an increase in c at concentrations from 100 to 500 M ( Fig. 2F) with values comparable with those seen for the WT N30␤ but significantly less self-association than WT N15␤ (p ϭ 0.032). It is expected that N15␤-c1 continues to self-associate more than N30␤-c1 (p ϭ 0.042) because X-ray crystallographic analyses have revealed V␤-V␤ association in N15␤-c1, whereas N30␤-c1 contained contacts mainly utilizing C␤ (15). This analysis also agrees with observations that N15␤ will utilize the V␤ hydrophobic patch more readily in the context of pre-TCR recognition of pMHC (9).

Backbone resonance assignments of full-length VSV8/K b
In our previous study, we were able to demonstrate an interaction between N15␤-c1 and labeled K b within the VSV8/K b complex (9). To characterize the binding site on pMHC, we required the backbone resonance assignment of VSV8/K b . Utilizing non-uniform sampling (24), we collected high resolution TROSY versions of HNCA/HNCOCA/HNCO/HNCACO/ HNCACB spectra on perdeuterated 2 H/ 13 C/ 15 N-labeled heavy chain with perdeuterated but otherwise unlabeled ␤ 2 -microglobulin (␤ 2 m), the smaller protein subunit of the MHC heterotrimer, and unlabeled VSV8. As shown in Fig. 3A, signals of cross-peaks were quite strong in the HNCA experiment with signals for both the HN correlated and its preceding (i and i Ϫ 1, respectively) residue in each spin system. Conversely, for the HNCACB (Fig. 3B), most spin systems lack cross-peaks for Cb i Ϫ 1 , whereas ϳ30% lack cross-peaks for Cb i . Similarly, the HNCO experiment provided robust Co i Ϫ 1 cross-peaks, whereas the HNCACO lacked cross-peaks for most Co i resonances. Because of the sparse coverage in sequential connectivities, we pursued orthogonal means of backbone assignment using the program RASP (25) to iteratively utilize chemical shift predictions to find potential resonance assignments. These assignments were cross-checked manually against a high-resolution 15 N dispersed TROSY-HSQC-NOESY experiment for confirmatory cross-peaks indicative of through-space contacts. This procedure was critical in assigning the ␤-sheet that forms the base of the peptide-binding groove and those that comprise the ␣3 domain and allowed the completion of 73% of the nonproline backbone amide resonance assignments (Fig. 4A). Depicted in Fig. 4, B and C, is the completeness of assignment overlaid on the X-ray structure of VSV8/K b (26). Notably, the ␣2 helix, proximal to the peptide-binding loop, with a likely role in pre-TCR binding, remained largely unassigned. Whether this is related to the specific peptide in the MHC groove or a more common feature of K b per se requires further study but implies that mobility in this region might have important consequences for dynamics of ␣␤TCR-pMHC as well as pre-TCR-pMHC interaction.

Large size of VSV8/K b -N15␤-c1 complex impacts measurement of binding interface
The backbone assignments were then used to interpret the previously reported 1 H-15 N TROSY-HSQC measurement of

Pre-TCR-pMHC interaction
unlabeled N15␤ addition to VSV8/K b wherein the heavy chain of K b was 15 N-labeled (9). As shown in Fig. 5A, chemical shift perturbation data analysis suggests a binding interaction involving the canonical peptide-binding groove previously reported for all TCR␣␤-pMHC interactions. However, the data are not unequivocal, with a number of residues distal to the canonical surface and within the ␣3 subunit revealing changes similar to those seen at the putative interfacial region. The regions associated with the largest peak intensity losses (Fig.  5B) conform to the region surrounding the peptide, clustering more closely to the canonical binding interface. Whether due to the large size of the complex or the weak nature of the binding event, a more precise interaction map could not be obtained. This, combined with the incomplete assignment of the region surrounding the peptide-binding pocket, led us to attempt to engineer improvements in the biophysical behavior of the pMHC complex.

Engineering a truncated pMHC model
The inability to obtain the complete backbone resonance assignments of the receptor-binding portion of the MHC may have been due to the large size of the pMHC complex or the conformationalheterogeneitywithinthecomplex.TRACTmeasurements of VSV8/K b at 100 M indicate a c of 32 ns, corresponding to a predicted molecular mass of 85 kDa for a spherical molecule, nearly double that of the 45-kDa monomeric pMHC. This long correlation time may be caused by the oblong shape of the molecule but regardless explains quite well the poor quality of the triple-resonance backbone experiments. To obtain more complete backbone assignments while improving the quality of spectra when measuring the ␤-pMHC complex, we designed a truncated K b molecule (termed K b -t) lacking the ␣3 domain and ␤ 2 m (16). TRACT measurements at 67 M indicate that VSV8/K b -t is monomeric with a c of 15 ns, corresponding to an expected molecular mass of 28 kDa, which is in better agreement with its molecular mass of 22 kDa.

Backbone resonance assignment of VSV8/K b -t
We proceeded to assign backbone resonances for 2 H/ 13 C/ 15 N-labeled VSV8/K b -t, acquiring triple-resonance experiments as for the full-length VSV8/K b . The HNCA shows a complete cross-peak complement for each HN-correlated spin system (Fig. 6A) as does the HNCACB (Fig. 6B). This allowed a more facile assignment of the majority of HN resonances present within the TROSY-HSQC spectrum, assigning 139 of 178 non-proline residues (78%; Fig. 7A). Surprisingly, similar to the full-length VSV8/K b (Fig. 4), the ␣2 helix flanking the peptidebinding groove remained recalcitrant to assignment with 19 residues within the ␣2 helix as well as nine more directly contacting this region remaining unassigned. We suspect, given the high quality of the spectra as well as the known mobility of the ␣2 helix (27), that this inability to assign this region of the protein reflects an intermediate (ϳms-s)-timescale motion (28) on the part of the peptide and ␣2 helix, broadening these resonances. That it is unassigned in both full-length and truncated constructs argues that this is an intrinsic property of the molecule and not the result of experimental manipulation.

Pre-TCR-pMHC interaction N15␤-c1 recognizes VSV8/K b -t via CDR and V␤ patch region
As the truncated K b construct appeared folded in a manner consistent with the fold of the full-length heterotrimeric mole-cule, we tested pre-TCR recognition of VSV8/K b -t by incubation with 15 N-labeled N15␤-c1. Spectral changes (Fig. 8A) are consistent with the binding event reported previously (9) with

Pre-TCR-pMHC interaction
chemical shift and intensity changes consistent with binding kinetics in the fast-to-intermediate exchange regime (28). The changes are almost exclusively located in V␤ (Fig. 8, B and C) with significant overlap between regions with chemical shift perturbation and intensity loss. The largest chemical shift and intensity changes are located in the CDR1 and -3 loops as well as V␤ patch region. Note that the CDR2 loop is not assigned likely due to conformational exchange. A region with smaller chemical shift changes is found distal to the interaction surface but still within V␤, clustered within a region encompassing residues 15-25 (the AB loop and B strand) and residues 74 -85 (E strand and EF loop). These changes may occur as a result of secondary rearrangements within the binding interface and are consistent with a dynamic view of pre-TCR or TCR interactions with pMHC (4, 6,9,18,29). The magnitude of chemical shift changes are similar when 200 M N15␤-c1 is added to fulllength VSV8/K b compared with VSV8/K b -t with the 95th percentile of CCSC equal to 0.064 (9) versus 0.060, respectively. Peak intensity losses, as expected, were lower overall due to the smaller size of the ␤-pMHC complex when truncated pMHC was used with median intensity loss (thus measuring residues not directly interacting with TCR␤) of 27 versus 47% (9) when the full-length MHC is utilized. Less overall peak broadening allowed more precision in measuring peak positions for chemical shift perturbation analysis and allowed a greater dynamic range for distinguishing exchange broadening leading to peak intensity losses (Fig. 8, B and C).

Pre-TCR recognizes a surface similar to TCR␣␤ on pMHC
Because we were able to assign a significant portion of the putative interface, we proceeded to measure 1 H-15 N TROSY-HSQC spectra of 1 H/ 15 N-labeled K b -t bound to unlabeled VSV8. With titration of unlabeled N15␤-c1, we measured chemical shift changes and intensity losses within K b -t (Fig. 9, A  and B). We observed significant chemical shift changes localized to the ␣1 helix, which comprises residues 57-85, and assigned portions of the ␣2 helix, which comprises residues 138 -176, as well as residues forming the "floor" of the peptidebinding groove (Fig. 9, C and D). Despite the lack of assignments within the ␣2 helix and the peptide VSV8 itself, these chemical shift perturbations define a contiguous binding interface. The chemical shift perturbation map compares favorably with that measured using the full-length K b molecule (Fig. 5), demonstrating the validity of improvements made by shortening the construct. Indeed, if one overlays the chemical shift perturbation map obtained with the truncated pMHC onto the structure of VSV8/K b bound to N15␣␤ (30), the extent of the pre-TCR-pMHC interaction matches with the TCR␣␤-pMHC interface determined by X-ray crystallography (Fig. 10). This strongly suggests that the pre-TCR recognizes a similar surface on pMHC as does TCR␣␤. To highlight the distinctness  Completeness of assignment is 78% non-proline residues. B and C, VSV8/K b structural model (Protein Data Bank code 1KPU) with assigned residues within K b colored blue, unassigned residues in white, prolines in cyan, and VSV8 in purple. C is rotated by 90°about the x axis relative to B.

Pre-TCR-pMHC interaction
of geometries that the respective interfaces place upon the molecular complex, we utilized HADDOCK2.2 web server (31) to generate a preliminary model of the pre-TCR-pMHC interaction interface (Fig. 10D). Given the lack of orientational restraints, it is not surprising that the low-energy conforma-tions clustered into two distinct docking orientations offset by ϳ180°as typified by the three lowest-energy conformations shown here. Each of the models utilizes the CDR2 and -3 loops as well as the surface that normally interacts with V␣ within the ␣␤ heterodimer, i.e. the V␤ patch and surrounding regions. It is

Pre-TCR-pMHC interaction
likely that interaction of CDR3 and the V␤ patch also requires neighboring CDR2 to participate in the modeled complex despite the absence of NMR-generated constraints. The models also provide conformations that do not result in obvious clashes with known pre-TCR (15, 20, 32) or pMHC glycosylation sites (33) (Fig. S1). The apposition of these elements results in a more  15 N-labeled VSV8/K b -t. Spectra are overlaid for VSV8/K b -t alone (red) and with addition of 200 (pink), 300 (lavender), or 500 (blue) M N15␤-c1. B, CCSCs for spectra outlined in A for 200 (dark blue), 300 (blue), or 500 (cyan) M N15␤-c1. Residues that differ at the p Ͻ 0.01 (**) or p Ͻ 0.05 (*) level between the CCSC for that residue and the median CCSC for each experiment are indicated. C and D, schematic representation of VSV8/K b -t with assigned residues colored blue, unassigned residues in white, prolines in cyan, and VSV8 in purple. Significantly perturbed residues by CCSC at p Ͻ 0.01 are highlighted as red spheres, and those that differ at p Ͻ 0.05 are shown as orange spheres. Residue identities are indicated. D, view rotated by 90°about the x and y axes relative to C with the ␣2 helix proximal. Figure 10. Modeling pre-TCR-pMHC complex. A-C, location of residues participating in interaction with N15␤ in the N15␤-VSV8/K b X-ray structure (30). A, interacting residues indicated by red spheres were defined by CCSC differing at the p Ͻ 0.01 level between the CCSC for that residue and the median CCSC for each experiment. The VSV8/K b -t structure is depicted with assigned residues colored blue, unassigned residues as well as ␣3 and ␤ 2 m subunits not present in this construct colored white, and VSV8 in purple. B and C are zoomed in to highlight the interaction site with ␣2 proximal in B and rotated 180°about the y axis and ␣1 proximal in C. D, three representative conformers of N15␤-c1 interacting with VSV8/K b as determined by molecular modeling using CST data for the N15␤-c1 interaction surface (9) and VSV8/K b -t CCSC data (Fig. 9) as ambiguous restraints input for HADDOCK2.2 (31). The conformers are representative of the three lowest-energy clusters of binding conformations and are color-coded with one shown as van der Waals surface representation and two shown in ribbon. The C␤ domain membrane-proximal end for each conformer is indicated with a color-coded "C." The models are aligned by overlay of the pMHC, shown in gray schematic with ␣1 and ␣2 labeled. E, side view with only conformer 1 of ␤ shown and differing by a rotation of 90°about the x axis relative to D. VSV8/K b -t is in the same approximate orientation as VSV8/K b shown in A with ␣2 facing front.

Pre-TCR-pMHC interaction
horizontal orientation of the ␤ subunit relative to the MHC peptide-binding groove, suggesting an acute angle of contact between pre-TCR and pMHC. This is readily observed by comparing the long axis of the ␤ subunit shown in Fig. 10A with that in Fig. 10E.

Discussion
This study highlights the rational design of recombinant prokaryotically expressed proteins, which in the wild-type form are not ideal for NMR studies. Specifically, domains that self-associate or interact within non-physiologically relevant regions ( Figs. 1 and 2) in comparison with their eukaryotically expressed transmembrane protein counterparts were rendered less reactive, whereas proteins that are simply too large for NMR to measure with sufficient precision (Figs. [3][4][5] have been pared down to their essential components. Structure-based design considerations allowed these hurdles to be overcome, thereby acquiring more accurate measurement of their binding capacity, thus leading to biologically important observations ( Fig. 10 and Refs. 6 and 9). TRACT measurements clearly report the predicted decrease in c and thus apparent molecular size at any given concentration with mutation of C␤ residues. Additionally, the increase of c with increasing concentration is lesser for the c1 mutant, making readily apparent the difference in molecular behavior conferred by these mutations. Similarly, building on prior design of pMHC for molecular studies (16), we were able to improve an existing model system for NMR study through modification of pMHC (Figs. 6 -9).
The chemical shift perturbation analysis of N15␤ binding to VSV8/K b -t implies a binding mode in which the pre-TCR utilizes a similar binding interface on pMHC as seen for TCR␣␤. As shown by the overlay of chemical shift perturbation data for the pre-TCR model with crystallographic data for the TCR system (Fig. 10, A-C), the interaction surface is similar between the systems. Once high-resolution structures are available, it will be valuable to compare the determinants of binding for the pre-TCR system versus the TCR, particularly focusing on the complementarity of the interacting surfaces as the weaker binding as measured by biomembrane force probe (9) or optical tweezers (6) for the N15␤-c1 for VSV8/K b as compared with that of N15␣␤ (9) implies suboptimal contacts in the former despite apparently similar contact area with pMHC ligands. The fact that the only other pre-TCR tested, N30␤, exhibits even more attenuated binding than N15␤ (9) suggests that weaker interactions are the norm for pre-TCR. In contrast, we have noted that the pre-TCR-N15␤ exhibits a force-bond lifetime profile similar to that of the OT1 ␣␤TCR-Ova/K b system (9). It should be noted, additionally, that there are likely an array of self-pMHCs that are the physiologic ligands for any given pre-TCR (6,9) and that some may possess as high or even higher affinity than that found for the cognate ligand of the respective TCR␣␤. Because of the convoluted and transient procession from thymic progenitor to T cell (34), it may be difficult to define the "true" ligand for any pre-TCR. Although there is good evidence for self-pMHC promotion of pre-TCRmediated proliferation and progression (6,9), the existence of non-MHC ligands for pre-TCR (35,36) has not yet been ruled out.
Meanwhile, as implied by single-molecule studies, the V␤ patch region actively participates in binding by replacing V␣ (6). In support of this hypothesis, the molecular modeling using ambiguous NMR restraints (Fig. 10D) shows conformations that uniformly predict a more horizontal apposition of pre-TCR to pMHC. In the pre-TCR, the long axis of the ␤ subunit is roughly parallel to the MHC groove, whereas in the TCR it is perpendicular. If this is the case, then this implies an initial contact orientation prior to the application of mechanical force that is rather tilted relative to the more upright TCR when bound to pMHC. This tilted interface would affect the loadsustaining ability and transmission of force through the pre-TCR with implications for the mechanisms of catch-bond formation and mechanical triggering. The sensitivity of pre-TCRbearing thymocytes to triggering may be potentiated by such a mechanism. It has been seen that the pT␣-␤ is more compliant than the TCR␣␤, undergoing a catch-bond requisite reversible transition more readily (6). The unpaired V␤ domain structure in the case of the pre-TCR relative to the paired V␣V␤ module in the TCR␣␤ can explain this compliance difference. As we have previously suggested (15), the location of glycans on murine pT␣ rules out some homodimerization models (20) of pre-TCR function. Although glycan sites do not appear to directly conflict with ligand binding measured here, the proximity of the ␤ chain (Figs. 10, D and E, and S1) or the pT␣ subunit (Fig. S1) in these models to the conserved glycan sites on pMHC means that some guiding role for glycan modulation of early thymocyte progression directly through the impact of these interactions or indirectly via recruitment of other cell surface proteins (37) cannot be excluded. The initial pre-TCR structural binding model presented here in conjunction with single-molecule biophysical analysis performed previously (6) suggests that the biology of the pre-TCR and ␣␤TCR will manifest important distinctions built around a common ␤ chain module.

Protein production
Unlabeled and isotopically labeled N15␤, N15␤-c1, N30␤, and N30␤-c1 were produced as detailed (18). Briefly, ␤ chains were expressed using Escherichia coli as inclusion body (ib) preparations. These ib preparations were washed thoroughly in 50 mM Tris-Cl, pH 8, 150 mM NaCl (TBS); dissolved in 6 M guanidine hydrochloride; refolded by dilution in 5.4 M guanidine hydrochloride, pH 8, 1 M Arg, 1 mM reduced glutathione, 0.1 mM oxidized glutathione; and subsequent dialysis in TBS. VSV8/K b was prepared as detailed (9,38). VSV8 was chemically synthesized (United Biosystems, Inc.), and K b and ␤ 2 m were produced via E. coli as ib preparations. The ib preparations were washed in TBS and dissolved in 8 M urea. VSV8, Kb, and ␤ 2 m were mixed in 20 mM Tris-Cl, pH 8, 8 M urea buffer and dialyzed serially against 2, 1, 0.5, and 0 M urea in 20 mM Tris-Cl, pH 8, for 2 h or overnight with a final dialysis against 0 M urea, 20 mM Tris-Cl, pH 8, overnight. VSV8/K b -t was prepared identically to VSV8/K b with the exception that 1 mM reduced glutathione and 0.1 mM oxidized glutathione were included prior

Pre-TCR-pMHC interaction
to dialysis-mediated refolding. Proteins were purified by successive rounds of size-exclusion chromatography.

NMR
Standard pulse sequences were utilized on Bruker 750-MHz and Varian 600-MHz spectrometers equipped with cryogenically cooled probes or a Bruker 500-MHz spectrometer with a room temperature probe. Concentrations of proteins were in the range of 50 -500 M, and NMR experiments were conducted at 298 K in PBS (50 mM NaPO 4 , 150 mM NaCl, pH 7.0) unless otherwise indicated. CCSC was calculated using the formula CCSC ϭ ͉5 ϫ ⌬␦( 1 H)͉ ϩ ͉⌬␦( 15 N)͉. All spectra were processed with NMRPipe (39) and visualized using CARA (40).

Cross-saturation transfer experiments
Unlabeled 175 M VSV8/K b was added to uniformly 2 H/ 13 C/ 15 N-labeled 350 M N15␤ in PBS (20 mM NaPO 4 , 50 mM NaCl, pH 7.0) in 80% (v/v) 2 H 2 O/H 2 O. The protonated VSV8/K b was saturated by an adiabatic WURST (wideband, uniform rate, and smooth truncation) pulse (25 ms) with an excitation bandwidth of 750 Hz. CST experiments were performed in an interleaved fashion with the on-resonance saturation pulse centered at 1 ppm (sat) and the off-resonance saturation pulse centered at Ϫ19 ppm (ref). The CST effect was calculated as a ratio of resonance peak intensities (sat/ref). Optimal saturation was found to occur at 3-s saturation time. The top 10th or 25th percentile was taken as the affected region.

Statistical analysis
Statistical significance (p) was calculated by linear regression analysis using the R software package (43).