Structure of a fully assembled tumor-specific T cell receptor ligated by pMHC

1To provide host protection, the immune system must deal with all comers. For T cells, which orchestrate the adaptive response, the solution is to render down pathogens and tumors to short (up to 24-residue) peptides for presentation to T cell receptors (TCRs) as complexes with major histocompatibility complex (MHC) proteins (peptide-loaded MHC [pMHC]). Despite decades of research, how the TCR initiates T cell signaling after binding peptide-loaded MHC class I or II molecules remains one of the great enigmas of immunology. The TCR belongs to the immunoglobulin superfamily of cell surface proteins and comprises complexes of clonotypic antigen-binding αβ heterodimers, each consisting of variable (V) and constant (C) domains, associated in a 1:1:1:1 ratio with invariant CD3-εδ, CD3-εγ, and CD3-ζζ dimers essential for signal transduction. Although many receptors for soluble ligands, e.g., growth factor receptors, are triggered by autophosphorylation, the TCR lacks associated kinase activity. Instead, after pMHC binding, the TCR is phosphorylated on tyrosine residues in its intrinsically unstructured cytosolic region by extrinsic Src-family kinases. Despite being among the most-studied membrane receptors, it is not yet understood how ligand-binding results in TCR triggering.

2TCRs diffuse on the cell surface as individual proteins poised to initiate signaling upon encountering single pMHC complexes. Given this constraint, it has been widely anticipated that TCR signaling would be initiated by spontaneous allosteric rearrangements. Using environment-sensitive fluorescent probes and structural analyses, a candidate conformational change in the AB loop of Cα triggered by pMHC engagement has been observed. Nuclear magnetic resonance (NMR) studies and molecular dynamics (MD) simulations identified putative small-amplitude structural rearrangements also involving the Cα AB loop, along with two additional sites: the Cβ FG loop projecting from the Vβ/Cβ interface and several Cβ residues at or close to the Cβ αA helix forming part of the CD3 docking site. Other NMR analyses suggested that conformational remodeling of the Cβ H3 helix region at the membrane-proximal face of the TCR accompanies pMHC binding. Crystal structural studies of soluble unbound and liganded TCR-αβ ectodomains, on the other hand, offer little support for allosteric effects since, typically, only minor adjustments in the ligand-binding site have been observed.

3Other explanations for receptor signaling allow rigid-body interactions of TCR-αβ with pMHC and for structural changes to occur elsewhere, either spontaneously or under force. It is suggested, for example, that CD3-ζζ loses cohesion with TCR-αβ or that the membrane-proximal cytosolic regions of the CD3-ζ subunits form a closer association, allowing signaling. Evidence that the TCR might function as a mechanosensor has come from experiments in which tangential forces applied to the receptor were observed to initiate calcium signaling. It was suggested that anisotropic forces created by the directional scanning of antigen-presenting cells (APCs) by T cells displaces TCR-αβ relative to the CD3 subunits, leading to signaling. Finally, a mechanism has been proposed wherein signaling does not rely on ligand-induced structural rearrangements in the assembled complex, of any type. According to this concept, referred to as the “kinetic-segregation” model, pMHC binding prevents the diffusion of the receptor out of regions of close contact between T cells and APCs that favor signaling. These regions are proposed to favor signaling because they would be depleted of phosphatases such as CD45 that are larger than pMHC/TCR complexes and counteract the activities of kinases. An understanding of the structural similarities or differences between fully assembled unliganded and pMHC-bound TCRs would dramatically reduce the number of possible explanations for how signaling is initiated by the TCR.

4The recent cryogenic electron microscopy (cryo-EM)-based structure of a non-clonotypic, unliganded TCR disclosed the overall arrangement of the TCR-αβ/CD3 subunits. Here, we took advantage of a high-affinity, tumor-reactive TCR to obtain the structure of a fully assembled ligand-bound receptor. Our 3.08 Å cryo-EM structure establishes the structural basis of CD3-εδ and CD3-εγ heterodimer stabilization and an unexpected role for lipids in TCR assembly, and reveals that antigen recognition by the receptor relies on the cooperative effects of small adhesive proteins. Most importantly, despite some minor conformational rearrangements, the structure reveals that the TCR is largely unchanged by ligand binding unlike most, if not all other classes of signaling receptors.

5To isolate a ligand-bound complex, we utilized an affinity-matured TCR-αβ (GPa3b17) being developed for immunotherapy, i.e., ImmTACs, which binds with high affinity (KD = 13 pM) to the melanoma-derived gp100 antigenic peptide (YLEPGPVTV) in complex with the most prevalent human leukocyte antigen (HLA), HLA-A∗02:01 (HLA-A2). Expressed at both wild-type (WT) and very low levels in leukemic (Jurkat) T cells, the TCR induced calcium signals comparable with those triggered by a non-affinity-matured TCR upon stimulation with gp100/HLA-A2-presenting supported lipid bilayers (SLBs; Figure S1A). Moreover, CD69 expression was upregulated on the GPa3b17-expressing cells following incubations with gp100 peptide-pulsed HLA-A2+ THP-1 cells (Figure S1B). This confirmed that the signaling capacity of the GPa3b17 receptor is comparable with that of more typical TCRs. The signaling observed in the absence of the gp100 peptide (Figure S1B) likely reflects the cross-reactivity promoting effects of the affinity-enhancing mutations in the HLA-A2-binding regions of the GPa3b17 TCR. cDNAs encoding the six subunits of the TCR, separated by viral 2A ribosome-skipping sites and distributed across three lentiviral vectors (Figure S1C), were expressed in Chinese hamster ovary (CHO) cells because they assembled more homogeneous complexes than HEK293 cells. The cytosolic domains of the CD3 subunits were truncated to limit heterogeneity and the CD3-δ chain tagged with GFP2 to monitor receptor expression and purification. Although partially assembled complexes could reach the surface of CHO cells, gp100/HLA-A2 complexes bound only to cells transduced by all three lentiviruses.

6For structural analysis, surface-expressed TCRs were tagged with soluble gp100/HLA-A2 monomers bearing a C-terminal affinity epitope of the Ca2+-dependent anti-Protein C antibody HPC-4. After cell lysis, the assembled TCR complexes were solubilized with glyco-diosgenin (GDN) and isolated using an HPC-4 immunoaffinity matrix and size-exclusion chromatography (SEC). All TCR subunits were identified using SDS-PAGE (Figure S1D). The gp100/HLA-A2/TCR formed a stable complex of ∼250 kDa detectable by native PAGE, which was super-shifted by binding two CD3-ε-specific UCHT1 Fab fragments (Figure S1E). Monodispersity was confirmed by SEC, insofar as the TCR eluted as a single peak that was quantitatively shifted after incubation with two equivalents of UCHT1 Fab (Figures S1F and S1G).

7The structure of the gp100/HLA-A2/TCR complex was determined by cryo-EM using all-gold supports covered with a hydrophilized graphene monolayer to preserve the fully assembled multiprotein complex (Figures S2E and S2F). UCHT1 Fab was used to increase particle stability and boost resolution (see online methods STAR Methods). Unexpectedly, almost no density was observed for UCHT1 Fab in the final high-resolution cryo-EM reconstruction, although reference-free 2D classification of the gp100/HLA-A2/TCR particles revealed weak density for the bound Fab and for the GFP2-tag fused to the CD3-δ chain (Figure S2G). An overall resolution of 3.08 Å was obtained for the gp100/HLA-A2/TCR complex, which comprised 11 individual proteins and peptides. The TCR-αβ heterodimer, bound to the trimeric pMHC I, was positioned at the center of the assembly, resting in a “half cup” formed by the heterodimeric CD3-εδ and CD3-εγ ectodomain pairs and supported by the packing of eight transmembrane (TM) helices including the CD3-ζζ TM regions (Figures 1A–1C). The pMHC I-engaged TCR was tilted 59° relative to the plane of the membrane, resulting in an offset of ∼60 Å in the membrane-anchoring points of the TCR and pMHC I (Figures 1C and 1D).

8Having pMHC I bound to the fully assembled TCR revealed how the TM receptor engages ligands. Recognition of pMHC I by the TCR exclusively involved the TCR-αβ complementarity determining regions (CDRs), consistent with soluble TCR-αβ/pMHC ectodomain-only crystal structures. The tumor-associated gp100 epitope was embedded in the MHC I heavy chain in the canonical mode for 9mer peptides, and the quality of the EM map at a local resolution of 2.6 Å allowed unambiguous placement of all residues of the tumor-associated antigen, re-emphasizing the important role of Val9 in anchoring the heteroclitic YLEPGPVTV peptide in the MHC I α1α2 groove (Figures 2A and 2B; Figure S3; ). The GPa3b17 TCR engaged the pMHC in a diagonal arrangement (rotation: 93°; tilt: 7°) that completely enveloped the peptide and left the pMHC unchanged relative to the unliganded state. The pMHC/TCR interface had a shape complementarity (Sc) of 0.648 and comprised more than twenty contacts, burying an overall surface of 2,154 Å2, with the buried area evenly distributed between both TCR-αβ subunits (TCR-α: 49%, TCR-β: 51%). Similar interfaces were found across diverse, soluble pMHC/TCR-αβ ectodomain pairs studied by X-ray crystallography, suggesting that our findings are generalizable. The GPa3b17 TCR had been mutated in three blocks using a “CDR walking” strategy to maximize its therapeutic activity by increasing its affinity. Except for the D95S mutation, which interacts with Arg65 of the MHC I heavy chain, other residues mutated in CDR3α (L98M and V99Q) did not contact the pMHC and must act indirectly, whereas changes to CDR2β (Q51W, I52A, V53Q, and N54G) and CDR3β (I96W and G98A) would have enhanced Sc in the region of the α1 helix of the MHC molecule and/or peptide (Figures 2C and 2D). The improvements in MHC recognition obtained with these mutations likely account for the cross-reactivity of this TCR in our assays (Figure S1B).

9Our high-resolution reconstruction allowed the molecular contacts underpinning the assembly of the TCR to be visualized at a high level of detail. The structures of the extracellular IgSF domains of the TCR-αβ and CD3 heterodimers are well characterized by NMR and X-ray crystallography. How these parts of each subunit are linked to the membrane by the connecting peptides (CPs) and how the heterodimers interact with one another and with the CD3-ζ homodimer to form the assembled complex has been intensely studied. Although the CP of TCR-α is unusually long (20 residues), the CP of TCR-β is significantly shorter (8 residues); the CPs of the CD3 heterodimers are even shorter and characterized by highly conserved CXXC motifs.

10For the TCR-αβ and CD3 heterodimers, the CPs serve to create rigid links to the membrane, but in two quite different ways. For TCR-αβ, the TCR-α CP reaches around and under TCR-β, where it forms a short helix that fills the space between TCR-Cβ and the tops of the TCR-αβ TM helices. The TCR-α CP helix is held in place by a disulfide bond to the TCR-β CP, bringing the CPs into proximity and reinforcing the linker region (Figure 3A). In the CD3 heterodimers, an elegant and evidently highly conserved arrangement of disulfides plays the major stabilizing role in the linker region. Strikingly, the cysteines comprising the conserved CXXC motifs adjacent to the TM regions of the CD3-δ, -ε, and -γ subunits form two pairs of interacting intramolecular cystines in each heterodimer (Figures 3B and 3C) which, in the assembled complex, are located either side of the “top” of the TCR-αβ TM bundle. The cystines are stabilized in this arrangement by hydrophobic interactions, accounting for their remarkable resistance to reducing agents. For each CD3 heterodimer, the paired cystines dramatically redirect the CPs, forcing them to “cross-over” (Figure 3D). This in turn separates and re-positions the TM regions of each subunit under the other subunit in a reciprocal arrangement. Perhaps, more importantly, by securing the positions of the CPs, the cystine pairs rigidify the linker regions of each heterodimeric subunit. Superpositions reveal the remarkable similarity of the two CD3 heterodimers as structural modules (Figure 3D), which can be thought of as semi-autonomous assemblies stabilized by their CPs that assemble with TCR-αβ heterodimers via interactions across three layers, extending from the extracellular space to the membrane (discussed below). Notably, the CPs allow the ectodomains of all three heterodimers to tilt back or forward relative to the long axis of their TM regions, likely facilitating close TM packing (Figures 3D and 3E).

11The tilted geometry of each heterodimeric subunit imposed by the rigid linkers enabled substantial interactions across three structural layers for each subunit pair. In the region of the IgSF domains (layer 1, Figure 4A), CD3-εδ and CD3-εγ were held together by small interfaces comprising just a few residues. This included a three-way interaction close to TCR-αβ wherein a glutamate in CD3-γ (Glu16) slotted between arginines in CD3-δ (Arg42) and CD3-ε (Arg93, supported by a salt- and H-bridge network involving Glu7 of CD3-δ and Tyr91 of CD3-ε, respectively; Figure 4B). The CD3-εγ and CD3-εδ heterodimers were each in turn anchored to the base of the TCR-αβ constant regions: on one side by a contact that had, at its core, a four-way hydrophobic interaction between His206 and Trp239 of TCR-β, Tyr14 of CD3-γ, and Leu68 of CD3-ε′ (Figure 4C), in the center by Ser196 of TCR-β contacting Glu9 of CD3-δ, and, on the other side, by a network of reciprocal side- and main-chain contacts between Arg163 and Asp166 of TCR-α and Glu6 and Arg36 of CD3-δ (Figure 4D). These highly directional interactions positioned CD3-εγ adjacent to TCR-β, and CD3-εδ next to TCR-α, in this way stabilizing the tilted arrangement of TCR-αβ relative to the membrane. N-glycans attached to the TCR are unlikely to have a direct impact on complex assembly. The cryo-EM map yielded “density” only for the Asn-linked GlcNAc moieties, suggesting that N-glycans are not involved in any protein interfaces (Figure S4A).

12Interactions in the linker region (layer 2; Figure 4A) were dominated by TCR-α and involved CD3-ζζ. TCR-α formed three main contacts with the disulfide-linked CD3-ζ dimer (Figure 4E). First, the backbone of Lys212 of TCR-α contacted the N terminus of CD3-ζ. Second, Phe214 of TCR-α formed a main-chain contact with Ser2 of CD3-ζ and a hydrophobic cluster with Phe3 and Leu6 of CD3-ζ as well as Phe246 and Tyr251 of TCR-β. Third, Thr216 and Asn221 of TCR-α bound main-chain atoms of Ser2 and Phe3, as well as Leu5 of CD3-ζ, respectively. In addition, Asn221 of TCR-α established main-chain contacts with Gln73 and Cys75 of CD3-δ (Figure 4E). The side chain of Gln73 also interacted with main-chain atoms of Thr216 in TCR-α. In stark contrast, interactions of TCR-β with the CD3 subunits in this region were limited to just one hydrogen bond with the main chain of Gln83 in CD3-γ, apart from the participation of Phe246 (TCR-β) in the hydrophobic cluster.

13The interlocking arrangements thus established in the folded ectodomain and linker regions were reinforced by interactions within the membrane (layer 3; Figure 4A). As in the unliganded TCR, the TMs of the CD3 dimers packed around a centrally located TCR-αβ TM helical bundle spanning the length of the membrane because interactions in the linker region had placed the CD3-γε TM next to the TCR-β TM, and CD3-δε alongside the TCR-α TM (Figure 5A). However, we observed a pronounced asymmetry in the degree of interdigitation of the TM helices across the membrane (Figure 5B). Close associations within the outer, but not the inner, leaflet of TCR-α and TCR-β TMs with CD3-εδ and CD3-εγ, respectively, favored the well-characterized intra-membrane neutralization of the positive charges of Lys236 of TCR-α (by Asp115 of CD3-ε as well as Asp90 and Thr94 of CD3-δ), Arg231 of TCR-α (by Asp15 of each CD3-ζ chain and main-chain atoms of Cys11 of CD3-ζ), and Lys268 of TCR-β (by Glu100 of CD3-γ; Figure 5C; ). Alongside the prominent salt bridges in the TM regions, two tyrosine residues provide long-range connectivity, involving Tyr262 of TCR-β with Asp15 of CD3-ζ′, and Tyr272 of TCR-β with Thr245 of TCR-α.

14Intra-membrane associations of the CD3-ζ homodimer complete the assembly. By packing against TCR-α, CD3-ζζ allowed neutralization of Arg231 of the α chain (by Asp15 of each CD3-ζ chain) but positioned the two CD3-ζ chains in very different environments (Figure 5B). One CD3-ζ polypeptide was displaced from the rest of the TM assembly, especially in the cytosolic leaflet, where it was likely surrounded by unstructured lipids, reminiscent of intra-membrane proteases. The other CD3-ζ chain formed very few protein-protein contacts at all. Instead, unexpectedly, extra density was seen in the outer leaflet of the membrane sandwiched between the TMs of CD3-ζ and CD3-γ, which we assigned to cholesterol (Figure 5D). Notably, although production of the complex for structural analysis required all three expression plasmids, we observed that pMHC-binding TCR complexes formed readily in CHO cells after we specifically deleted CD3-ζ (Figure S1H), suggesting that ζ-chain homodimer incorporation comprises a late step in TCR assembly. This is in marked contrast to normal T cells as well as T cell hybridomas and leukemic T cell lines, which express very low levels of TCR in the absence of CD3-ζ. Evidently, receptor editing is far more stringent in T cells than in CHO cells. Sterol lipids, intercalated among the TM regions, may nevertheless have an important role in stabilizing a TCR-αβ/CD3-δγε2 intermediate while also contributing to a composite docking site for the ζ-chain homodimer.

15Our structure of the pMHC/TCR complex allowed us to examine the effects of pMHC binding on the fully assembled receptor. Comparing this structure with the backbone conformation of the cryo-EM reconstruction at 3.7 Å of the unliganded TCR indicated that the TCR is remarkably resistant to any form of ligand-induced structural rearrangement. The root-mean-square-deviation (RMSD) between the two structures over all Cα atoms was 1.2 Å (Figures 6A and S5). Notably, although crystallographic and NMR studies had suggested that pMHC binding induces changes in the region of the TCR-Cβ FG loop (; Figure 6B), the TCR-Cα AB loop (;; Figure 6C) and the TCR-Cβ αA helix (;; Figure 6D), these regions were essentially identical, exhibiting Cα-RMSDs of 0.7, 0.5, and 1.0 Å, respectively. Moreover, despite large differences in construct design and sample preparation, there were only minor displacements of the C termini of the membrane-spanning regions, with the largest being observed for the tips of peripherally located CD3-ε (3.5 Å for Cα-Lys131) and CD3-ε′ (2.5 Å for Cα-Lys134; Figure 6E).

16The GPa3b17 TCR binds with high affinity to the melanoma-derived gp100/HLA-A2 complex (KD = 13 pM). As such, it was theoretically possible that the high affinity of the interaction “traps” functionally relevant dynamics that are important for signaling in response to typical, lower-affinity ligands (KD = 1−100 μM). To investigate this we used in silico mutagenesis to return the GPa3b17 TCR to the parental “WT” gp100 TCR sequence, which binds gp100/HLA-A2 with an affinity of KD = 22 μM (mutated residues in TCR-αβ are listed in the STAR Methods section). We then performed 4 μs-long atomistic MD simulations of gp100/HLA-A2 in complex with partial (i.e., TCR-αβ) and fully assembled models of the GPa3b17 and WT TCRs, respectively (see STAR Methods). The fully assembled simulation systems were embedded in a realistic model of the plasma membrane.

17RMSDs and root-mean-square fluctuations of the complex subunits revealed that the subdomains of the GPa3b17 and WT TCR complexes were stable throughout the simulations (Figure S6), with two local exceptions: one of the two CD3-ζ TM helices unraveled slightly in both fully assembled TCR simulations, possibly owing to truncations of the simulation models at the N termini of the CD3-ζ cytosolic regions, and the CD3-ε and CD3-ε′ D β strands (residues 52–56) exhibited a tendency to dislodge from their respective β sheets in the fully assembled TCR simulations (Video S1). We hypothesize that the binding of UCHT1 Fab to the preceding loop has a stabilizing effect on this region that is not captured in our simulation model without the Fab. As expected, the simulations showed that the interaction of gp100/HLA-A2 with the WT TCR was more transient than the interaction of gp100/HLA-A2 with GPa3b17. Both the partial and fully assembled GPa3b17 TCRs maintained stable interactions with gp100/HLA-A2 (Figure S7A). In contrast, only the fully assembled WT TCR remained stably bound to gp100/HLA-A2, with the pMHC interaction of the partial WT TCR loosening during the simulation. However, both the GPa3b17 and WT TCR-αβ subdomains exhibited similar internal dynamics and mean simulation structures throughout, with near-identical distance matrices of TCR-αβ residue pairs (Figures S7B and S7C). Likewise, the TM domains of the complex exhibited similar dynamics in the simulations of the fully assembled GPa3b17 and parental TCRs (Figure S6). Therefore, the enhanced affinity of the GPa3b17 receptor was not associated with marked changes in the internal stability or dynamics of the ligand-bound complex. These data indicate that lower-affinity TCRs also resist ligand-induced structural rearrangements.

18Finally, by docking CD8 to pMHC in the liganded structure, in the manner of the complex formed by the soluble truncated ectodomains of murine H-2Dd and CD8αβ, we were able to reconsider the role of co-receptors in signaling by the TCR. We found that CD8 binds comfortably to the “underside” of HLA-A2 in the pMHC/TCR complex (Figure 7A). The “locking” of the binding site in such a position by TCR engagement of pMHC would favor associations with any protein that is tethered to the same membrane as the receptor and interacts in trans with the apposing cell, such as CD8. This may explain the temporal sequence of events leading to assembly of the ternary pMHC/TCR/CD8 (or CD4) complex, with the coreceptors arriving last.

19The structure of a TCR that binds to pMHC in an arrangement that is the reverse of what is typically seen was recently determined. This positions CD8 at the “top” of the pMHC rather than underneath, a location that is likely to be less accessible than in the canonical arrangement owing to displacement of the binding site by ∼50 Å (Figures S4B and S4C). Although probably flexible, it seems unlikely that the extended mucin-like regions of CD8 easily accommodate binding to pMHC in both positions. CD4, which is comprised of concatenated IgSF domains, is even less likely to have the necessary length variability. These considerations suggest a second explanation for the failure of the reverse-topology receptor to signal. In addition to the inability of CD8 to deliver kinases to the cytosolic CD3 chains in this arrangement, signaling could be prevented by the inability of CD8 to dock efficiently with the preformed pMHC/TCR complex, owing to the unique geometry of the complex.

20We determined the structure of a fully assembled TCR in complex with a cognate pMHC ligand. Because we could avoid chemical cross-linking, our data reveal that a pMHC-bound TCR is remarkably stable. We provide a near-atomic view of how the TCR achieves this conformational stability, which involves (1) asymmetric, multivalent contacts between the CD3-εδ and CD3-εγ ectodomains as well as the constant regions and CPs of the TCR-αβ subunits consistent with mutational and NMR studies, (2) the assembly of the CD3 TMs around a centrally located TCR-αβ TM bundle in an arrangement that is relatively loose but crucially favors charged TM-residue neutralization, and (3) an intercalated sterol lipid that supports packing of the TM bundle in the region of the outer membrane leaflet. Our work emphasizes, especially, the role of conserved, interacting cystines in the CPs of the CD3-εδ and CD3-εγ subunits in rigidifying the two CD3 heterodimers, and the similarity of these heterodimers as structural units. The essential role of the cystines explains the impact of mutating these residues on TCR assembly and signaling, and thymocyte development. To our knowledge, this arrangement of paired, hydrophobically interacting cystines is unique to the TCR.

21We speculate that the likely rigid conformations of the CD3-εδ, CD3-εγ, and TCR-αβ subunits facilitate their docking. The structural similarity of the CD3 heterodimers also suggests that in some circumstances, they might exhibit a degree of interchangeability during receptor assembly. We predict γδTCRs and the pre-TCR, each of which can assemble in the absence of CD3-δ, and the TCRs of chickens, which have just two CD3 genes (encoding ε and a single γ/δ ortholog), to have the same overall arrangement of subunits we observe here. The signaling subunits of the BCR, Ig-α and Ig-β, which lack equivalent cysteine residues in their CPs, likely assemble in a different manner (see, e.g., ). The multivalent nature of the subunit interactions, and our finding that the formation of ligand binding-competent TCRs requires all subunits to be present except CD3-ζ, suggest an initial “all-in or none-in” mode of assembly of the heterodimeric subunits, followed by incorporation of CD3-ζζ. Being largely buried, sterol lipid incorporation into the TCR is different to that observed for G protein-coupled receptors, wherein cholesterol is attached to the outer surface of the TM bundle. Cholesterol is proposed to stabilize non-signaling TCRs but might instead have a structural role, consistent with the signaling-inhibitory properties of cholesterol sulfate, which could affect complex stability. Our finding that sterol lipid, presumably cholesterol, is intercalated within the TM regions of the liganded GPa3b17 TCR is mirrored by recent work showing cholesterol to be similarly located in an unliganded TCR. However, it seems highly unlikely that cholesterol would have a gating role in signaling as proposed by if it is present in both structures.

22Most importantly, this initial view of a ligand-bound receptor reveals the underpinnings of TCR function. Comparison with the unliganded structure suggested that except for local changes at the ligand-binding site, the TCR resists perturbation and interacts as a “rigid body” with gp100/HLA-A2. Previously, it was proposed that pMHC binding triggers allosteric rearrangements of the CD3-ζ subunits or TCR-α and TCR-β constant regions, but this is not confirmed here. “Dynamic allostery,” i.e., the transmission of signals to distal sites via changes in protein dynamics in the absence of overt structural transitions, warrants further investigation. Nonetheless, the TCR does not exhibit the intrinsic conformational flexibility needed for allostery-based signaling in the manner of, e.g., G protein-coupled receptors. It needs to be acknowledged that these conclusions rest on an analysis of a high-affinity TCR/pMHC complex, although our MD simulations suggest that the receptor’s structural dynamics are not fundamentally altered by more typical, lower-affinity ligand interactions.

23Since current cryo-EM technology does not allow the pMHC/TCR complex to be subjected to forces, caution must be exercised with respect to their role in signaling. Here, we note only that theories of mechanotransduction (e.g., ) must account for signaling by all ligands. Our structure reveals that pMHC and agonistic antibodies would each “pull” on different parts of the TCR and along different trajectories, making it unlikely that both classes of ligands could trigger a shared sequence of conformational rearrangements. Moreover, forces acting via the TCR were only barely measurable for T cells interacting with fluid-phase SLBs presenting TCR ligands. Our finding that untethered pMHC and Fab adducts that only slow the diffusion of the receptor are capable of triggering signaling also suggests that strong forces are unnecessary. Being essentially unchanged by pMHC binding, the GPa3b17 TCR is also unlikely to exhibit any new tendency to oligomerize after interacting with ligands, as has been suggested. Instead, contrary to general expectation, the TCR adopts a stable structure that appears to be primed for ligand engagement and signaling without an apparent need for spontaneous structural rearrangements. To the best of our knowledge, in this respect, the TCR differs from all other classes of signaling receptors. Overall, the structural data are consistent with the notion that a TCR ligand may need only to bind the receptor, holding the TCR in regions of the membrane that favor signaling. This explains both the intrinsic structural diversity of the TCR and its ability to be triggered by ligands that engage it in so many profoundly different ways.

24The pMHC binding site on the TCR is tilted and located on the “side” of the receptor owing to CD3 locking the TCR-αβ subunits at an angle of 59° relative to the cell surface. Positioning the ligand-binding region in this way would constrain de novo “head-to-head” encounters of the TCR- and pMHC-binding sites in a direction orthogonal to the respective cell surfaces. Similarly, a binding site at the “top” of the TCR would hinder binding if the inter-membrane distance was pre-set by similarly sized adhesion proteins such as CD2 and CD58, and only lateral movement was possible. We propose that the important role of CD2 and CD58, or of other similarly sized cell-cell interacting proteins, is to create an initial overlap in the positions of the TCR and pMHC-binding sites (Figure 7B). Along with the remarkably fast two-dimensional pMHC/TCR binding kinetics, pre-alignment of the membranes by CD2 and CD58 would facilitate antigen scanning by the TCR. In turn, formation of the pMHC/TCR complex could favor co-receptor recruitment. The structural data therefore explain the key role of small adhesion proteins in, e.g., tumor killing by T cells. More importantly, access to the structure of a fully assembled ligand-bound receptor brings us substantially closer to solving the enigma of TCR triggering.

25Caveats of our study are (1) that we draw inferences about TCR signaling from a receptor bound to a soluble pMHC ligand, (2) that we removed the cytosolic regions of the complex, and (3) that we used a high-affinity TCR to assemble a pMHC/TCR complex that was stable enough for cryo-EM analysis. Future work is needed to determine whether the TCR is as resistant to structural rearrangements when it is ligated by a membrane tethered ligand as it is when it binds a soluble pMHC. Previous work has demonstrated that the cytosolic regions of an unliganded TCR are unstructured, and it will be important to determine whether these regions are also unstructured in fully assembled, ligand-bound complexes, ideally in the presence of lipids reflecting the composition of the T cell plasma membrane. Finally, the possibility that all TCR interactions are essentially rigid body in character needs to be tested via the determination of structures of other fully assembled, ligand-bound TCRs, including those with more typical ligand affinities, as soon as this becomes possible.

26Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Simon J. Davis (simon.davis@imm.ox.ac.uk).

27Plasmids generated in this study have been deposited at Addgene (www.addgene.org). All other unique/stable reagents generated in this study are available from the lead contact with a completed Materials Transfer Agreement.

28Rosetta 2 (DE3) pLysS bacteria were grown in suspension culture in 2.5 L Erlenmeyer flasks at 37 °C, 200 rpm in LB medium supplemented with 50 μg/mL kanamycin and 34 μg/mL chloramphenicol.

29These cells were grown in adherent cell culture in T175 flasks incubated at 37 °C, 5% CO2 in DMEM, supplemented with 10% (v/v) FBS, 1% (v/v) Penicillin/Streptomycin/Neomycin (PSN; final concentrations 50 U/ml Penicillin, 50 μg/ml Streptomycin, 100 μg/ml Neomycin), 2 mM L-glutamine, and 1 mM sodium pyruvate.

30CHO-S cells were maintained in suspension cell culture in 2 L Erlenmeyer flasks incubated at 37 °C, 8% CO2, 85% humidity, 125 rpm in Freestyle CHO Expression Medium, supplemented with 8 mM L-glutamine.

31HEK-293T cells were grown in adherent cell culture in T75 flasks incubated at 37 °C, 5% CO2 in DMEM, supplemented with 10% (v/v) FBS, 1% (v/v) PSN, and 2 mM L-glutamine.

32These cells were kept in suspension cell culture in T25 flasks incubated at 37 °C, 5% CO2 in RPMI, supplemented with 10% (v/v) FBS, 1% (v/v) PSN, 2 mM L-glutamine, and 1 mM sodium pyruvate. The cells were maintained at a density of approximately 105-106 cells/ml and split the day before experiments.

33THP-1 cells were maintained in suspension cell culture in T25 flasks incubated at 37 °C, 5% CO2 in RPMI, supplemented with 10% (v/v) FBS, 1% (v/v) PSN, 2 mM L-glutamine, 1 mM sodium pyruvate, and 10 mM Hepes, pH 7.3.

34All the cell lines used in this study were purchased from established suppliers who authenticate the cells before distribution, except for the CHO-K1 (D28-W1) line, which was obtained from Lonza Biologics (formerly Celltech Ltd) in 1988 and for which we no longer hold records.

35The affinity-matured GPa3b17 TCR is encoded by the following gene segments: the TCR-α V domain is encoded by TRAV17∗01 and TRAJ29∗01 gene segments and the C region by the TRAC∗01 gene segment; the TCR-β V domain is encoded by TRBV19∗01, TRBJ2-7∗01, and TRBD1∗01 gene segments and the C region by the TRBC1∗01 segment. DNA encoding full-length TCR-α and TCR-β polypeptides, and truncated forms of CD3-ε (residues 1-158), CD3-γ (1-144), CD3-ζ (1-57), and CD3-δ (1-132) fused to GFP (1-238), each separated as required by 2A ribosome-skipping sites (i.e., F2A (GSG)VKQTLNFDLLKLAGDVESNPGP, T2A (GSG)EGRGSLLTCGDVEENPGP, E2A (GSG)QCTNYALLKLAGDVESNPGP), were cloned into the pHR-SIN vector alone or in combination to create the following expression constructs: pHR-αγεζ, pHR-βεζ, and pHR-δGFP (Figure S1C). Lentivirus was made by transiently transfecting HEK-293T cells (RRID:CVCL_0063; ATCC), grown in DMEM (Sigma) supplemented with antibiotics and L-glutamine (Sigma), with the pMD.G and p8.91 packaging vectors and each of the pHR-SIN constructs, using Genejuice (Merck) according to the manufacturer’s instructions. Forty-eight to 72 hrs after transfection, the supernatants were harvested and 0.2 μm filtered (Millipore) to remove cell debris. 1 x 106 Freestyle Chinese hamster ovary cells (CHO-S; RRID:CVCL_7183; ThermoFisher) in 1 mL of Freestyle CHO Expression Medium (Gibco) supplemented with L-glutamine, were then transduced overnight with 2 mL of pHR-αγεζ and pHR-βεζ lentiviruses and 1 mL of the pHR-δGFP lentivirus. The cells were recovered with 5 mL of Freestyle CHO Expression Medium supplemented with L-glutamine the next day and grown as attached cells. Four days later, 1 x 106 of these cells were re-transduced with the expression constructs before being converted back to suspension cells via culturing in an orbital shaker at 125 rpm and 37 °C. Expression was confirmed after seven days using flow cytometry, by incubating the TCR-expressing cells with gp100/HLA-A2 tagged with Alexa Fluor 647 (Molecular Probes), and/or PE-labelled UCHT1 antibody (Biolegend) for 30 min at 4 °C in the dark. All flow cytometric analysis was conducted on an Attune NxT flow cytometer (ThermoFisher), and the data analyzed with FlowJo software.

36DNA encoding residues 25-304 of the heavy chain of HLA-A2, flanked by six histidine residues at the N-terminus and by an HPC-4 epitope tag (‘C-Tag’: EDQVDPRLIDGK) at the C terminus, and residues 22-119 of the light chain (β2-microglobulin) were each cloned into the pET28a(+) vector (Novagen) to yield the pET28a(+)-HLA-A2(Heavy) and pET28a(+)-HLA-A2(Light) expression vectors. Expression of each of the polypeptides and refolding of HPC-4 tagged HLA-A2 in the presence of gp100 peptide (YLEPGPVTV), were undertaken as described. Briefly, inclusion bodies released by cell disruption at 28,000 psi, 10 °C (Constant Systems) were centrifuged at 15,000g for 10 min at 4 °C. The washed, pelleted inclusion bodies were solubilized in 50 mM Tris/HCl pH 8.0, 10 mM EDTA, 8 M urea overnight at room temperature, before being clarified by centrifugation at 15,000g for 10 min at 4 °C and 0.2 μm filtration (Millipore) and concentrated to 15 mg/mL. Seventy-five milligrams of the MHC heavy chain inclusion bodies and 30 mg of light chain inclusion bodies in 50 mM Tris/HCl pH 8.0, 100 mM NaCl, 8 M urea, were denatured with 10 mM DTT at 37 °C for 30 min. The two polypeptides, along with 15 mg of gp100 peptide in DMSO, were then added to 100 mM Tris/HCl pH 8.0, 2 mM EDTA, and 1.2 M L-arginine in 12 kDa MWCO cellulose dialysis tubing (Sigma) and dialyzed against de-ionized water (1 change after 72 h) and then 10 mM Tris/HCl pH 8.0 (after 80 h). After 96 h, the soluble gp100/HLA-A2 was concentrated using a Vivaflow 200, 10 kDa MWCO PES concentrator and Amicon Ultra-15 10 kDa MWCO Ultracel-10 membrane (Millipore) and purified by SEC on a Superdex 200 HiLoad 16/600 GL column (GE Healthcare) pre-equilibrated with PBS pH 8.0 supplemented with 0.05% (w/v) sodium azide. The gp100/HLA-A2-containing fractions were pooled and concentrated to 1 mg/mL.

37DNA encoding anti-HPC-4 antibody heavy and light chains, codon-optimized for expression in CHO cells was cloned into separate pHR-IRESEm plasmids; a 6x histidine tag at the C terminus of the heavy chain was also encoded to facilitate antibody purification. Lentiviruses were generated by separate transient transfections of HEK-293T cells of each construct using Genejuice. Forty-eight to 72 h after transfection, two milliliters of the 0.2 μM filtered supernatant containing lentiviruses was used to transduce 1 x 106 CHO-K1 cells overnight. After four days recovery in 5 mL DMEM supplemented with 10% FBS (Gibco), antibiotics and 2 mM L-glutamine, the supernatant was removed, and the cells used to seed large-scale culture for supernatant harvesting every 4-5 days. Anti-HPC-4 antibody was affinity purified by nickel-chelation chromatography (Qiagen). Absorbance at 280 nm and 12% acrylamide Coomassie Blue-stained analytical gels were used to confirm antibody quality. Sixteen liters of supernatant yielded 1g of antibody. The antibody was buffer exchanged into 20 mM Hepes pH 7.4, 150 mM NaCl, 0.05% (w/v) sodium azide using SnakeSkin dialysis tubing (Thermo Scientific) before being coupled to CNBr-activated Sepharose according to the manufacturer’s instructions (GE Healthcare).

38CHO-S cells expressing the TCR were grown in Freestyle CHO Expression Medium supplemented with L-glutamine in 1 L cultures in 2.5 L Erlenmeyer flasks (Corning) and harvested when they reached a density of 2.8 x 106 cells/mL. The cells were recovered by centrifugation at 500g for 10 min at 4 °C. The cells were then washed with cold PBS and re-pelleted before being resuspended and an aliquot stained with gp100/HLA-A2 for 1 h at 4 °C to allow sustained expression to be confirmed by flow cytometry. Cells were washed with cold PBS and pelleted for storage at -80 °C.

39Cells harvested from a total of 10-50 L of culture were re-suspended in 30 mM Tris/HCL pH 8.0, 750 mM NaCl supplemented with Roche cOmplete™ Protease Inhibitor Cocktail, 2.5 U/μL Benzonase Nuclease (Sigma) and 10 μg/mL DNase I (Roche), and disrupted at 5,000 psi using a benchtop 1.1-KW cell disruptor (Constant Systems), cooled to 10 °C with a Frigomix R (Sartorius-Stedim). The cell lysate was centrifuged at 600g for 10 min at 4 °C. The supernatant was transferred to 70 mL Beckman Coulter centrifuge tubes (cat. 355622) and centrifuged using a Type 45 TI rotor at 15,000g for 5 min at 4 °C. The supernatant was collected and re-centrifuged at 100,000g for 60 min at 4 °C to pellet the membranes. The membrane fraction was solubilized into 20 mM HEPES pH 7.5 (Gibco), 500 mM NaCl, 15% (v/v) glycerol (MP Biomedicals) supplemented with Roche cOmplete™ Protease Inhibitor Cocktail, 1% (w/v) GDN (Anatrace), 2.5 U/μL Benzonase Nuclease, 1 mM CaCl2, by rotation at 5 rpm overnight at 4 °C. The soluble fraction was isolated by centrifugation at 142,000g for 4 h at 4 °C, 5 μm and 0.45 μm filtered (Sartorius), and passed over 2 mL of anti-HPC-4 antibody-coupled Sepharose beads in a 10 mL column (VWR) pre-equilibrated with solubilization buffer without detergent. The resin was washed with 20 mM Hepes pH 7.5, 500 mM NaCl supplemented with 1 mM CaCl2, 2 mM ATP (Enzo Life Sciences), 0.15% (w/v) GDN, and 20 mM Hepes pH 7.5, 500 mM NaCl, 1 mM CaCl2, 2 mM ATP, and 0.05% (w/v) GDN. The TCR complex was eluted using 200 μg/mL C-Tag peptide (Ne Biotech) in elution buffer comprised of 20 mM Hepes pH 7.5, supplemented with 500 mM NaCl, 1 mM EDTA, and 0.05% (w/v) GDN. Finally, the eluted complex was purified by SEC on a Superdex 200 HiLoad 16/600 GL column (GE Healthcare) in the elution buffer. Analytical, high-resolution SEC was performed using an ACQUITY UPLC Protein BEH200 SEC column (Waters).

40Tests of GPa3b17 TCR-induced calcium signaling were undertaken using SLBs. Glass coverslips (25 mm, thickness no. 1.5; VWR) were prepared by incubation in a 3:1 mixture of pure sulfuric acid/hydrogen peroxide (30% solution in water, Merck) at room temperature overnight, then rinsed thoroughly in MQ water, dried rapidly and plasma cleaned for 1 min. SLBs were formed by vesicle fusion in CultureWell 50-well silicon covers (Grace Bio-Labs) which were cut to size and placed on the cleaned coverslips. Small unilamellar vesicle (SUV) lipid preparations were produced by mixing 98% POPC and 2% DGS-NTA-Ni2+ (Avanti Polar Lipids) molar solutions in chloroform in a 1.5 mL amber glass vial (Merck) and dried under a stream of nitrogen. The lipids were then resuspended in 0.22 μm filtered PBS to a final concentration of 1 mg/mL, vortexed for 30 s and tip sonicated on ice for 30 min. Five microliters of the SUV mixture was added to each well with 5 μL of filtered PBS and incubated at room temperature for 30 min. Each well was then washed five times with 5 μL of filtered PBS before adding 5 μL of protein mixture and incubating for 1 h at room temperature. Refolded and purified gp100/HLA-A2 was produced with 2 x 6 histidine C-terminal tags as described above for binding to the bilayers. Protein mixtures were prepared with the gp100/HLA-A2 at 0.2 μg/mL final concentration and a ‘null’, irrelevant pMHC (HLA-A2 presenting the 9V peptide of NY-ESO; ) at a final concentration of 10 μg/mL to saturate all available nickel sites on the SLB. Immediately before use, each bilayer was washed 10 times with 5 μL filtered PBS.

41To prepare the cells for calcium signaling experiments, 1 x 106 GPa3b17 TCR-expressing Jurkat T cells were placed in a 1.5 mL Eppendorf tube, pelleted, and resuspended in 100 μL RPMI, 100 μL HBS, and 1 μL Fluo-4 AM calcium dye (25 μg/mL final concentration; ThermoFisher). The cells were then incubated at 37 °C and 5% CO2 for 5 min before being pelleted and washed twice in 500 μL PBS. The cells were finally resuspended in 200 μL pre-warmed PBS before imaging. Approximately 5 μL of the cell suspension was added to each SLB-containing well for imaging on a Zeiss LSM 880 Inverted Microscope with Airyscan (Carl Zeiss), using a 10x objective to capture several hundred to a thousand cells in the field of view. Fluo-4 was excited using an Argon 488 nm laser at 10% power and images were taken every second for a total of 10 min without averaging. The fraction of cells triggering under each condition was analyzed by a bespoke MATLAB® script (written by Jane Humphrey and Aleks Ponjavic; Klenerman Laboratory, Department of Chemistry, Cambridge). Comparisons were made with Jurkat T cells expressing the 1G4 TCR that interacts with SLBs loaded with soluble HLA-A2 presenting the 9V (SLLMWITQV) NY-ESO peptide.

42Tests of intermediate signaling induced by the affinity matured GPa3b17 TCR were undertaken by measuring CD69 upregulation. GPa3b17 TCR-expressing Jurkat T cells were incubated overnight with wild-type THP-1 APCs (ATCC) and THP-1 cells whose expression of β2-microglobulin was prevented using CRISPR/Cas9 targeting, in the presence and absence of 100 μM gp100 peptide. CD69 expression was then analyzed by flow cytometry.

43Purified TCR was concentrated to 2 mg/mL, mixed with UCHT1 Fab (1:5 molar ratio), and incubated for 30 min on ice. Three microliters of protein solution was deposited on a graphene-coated UltraAuFoil R1.2/1.3 grid (Quantifoil). Excess protein was blotted away for six seconds with a force of 25 at 4 °C and 100% relative humidity before plunge-freezing in liquid ethane using a Vitrobot Mark IV (Thermo Scientific). EZ transfer graphene was obtained commercially (Graphenea) and deposited onto the grid by floating in a water bath, and the sacrificial polymer layer removed according to the manufacturer’s instructions. 50 nM 1-pyrenecarboxylic acid in isopropanol was used to increase hydrophilicity before rinsing with pure isopropanol and air drying. Despite a reduction in image contrast, use of graphene proved crucial for stabilizing the TCR, as vitrification on unsupported grids did not yield sufficiently intact particles. However, graphene alone was insufficient to adequately stabilize the TCR for high resolution imaging, which was only achieved with the addition of UCHT1 Fab. It should be noted that virtually no density for UCHT1 Fab was observed in the final cryo-EM reconstruction, even though biochemical analysis confirmed UCHT1 binding in a 2:1 stoichiometry (Figures S1E–S1G), and weak density associated with the Fab could be discerned in reference-free 2D classifications of the gp100/HLA-A2/TCR particles (Figure S2G). Thus, dominant contributions of UCHT1 to particle alignment seem unlikely. Rather, we speculate that binding of UCHT1 Fab to the peripherally located CD3-ϵ subunits shielded the TCR from unfavorable interactions with the air-water or graphene-water interfaces enabling high-resolution structure determination. Vitrified grids were imaged in a Titan Krios (FEI) equipped with a K2 Summit direct electron detector operating in counting mode. Zero-loss imaging was performed with a 20 eV slit width using a GIF Quantum SE post-column energy filter (Gatan). Micrographs were collected in SerialEM as 70-frame movie stacks at a nominal magnification of 130,000×, corresponding to a calibrated pixel size of 1.05 Å/pixel at the specimen level. Cryo-EM data collection parameters are summarized in Table S1.

44Movie stacks were imported into cryoSPARC and subjected to Patch Motion correction followed by Patch CTF-estimation. A subset of 100 randomly selected micrographs was used to train a picking model in TOPAZ, which was subsequently applied to the entire dataset. TOPAZ-picked particles were extracted with a box size of 384 pixels and subjected to multi-class ab initio reconstruction (class similarity 0). The ab initio map corresponding to the lone ‘good’ class, with clear protein features, as well as two ‘junk’ maps, lacking discernable protein features, were selected as references for downstream hetero refinement of the entire particle stack. Hetero refinement was then iterated until 98% of the remaining particles classified into the good class. The final image stack of 154,408 particles was subsequently subjected to non-uniform refinement and CTF refinement in cryoSPARC resulting in a cryo-EM map at 3.08 Å resolution. The cryo-EM processing workflow is illustrated in Figure S2B.

45An initial model was generated using RosettaCM with PDB entries 1SY6, 1XIW, 5EU6, and 6JXR as starting templates. All non-protein components and antibody fragments were removed from the starting models prior to density-guided rebuilding in RosettaCM against the final cryo-EM map in real space. Rosetta scripts were adapted from the DiMaio Lab repository (University of Washington, https://dimaiolab.ipd.uw.edu/software/). The highest ranked Rosetta models by geometry and density scores were visually inspected and manually adjusted in COOT. Iterative rebuilding in COOT was followed by density-guided relaxation in Rosetta. Finally, real-space refinement was performed using Phenix. The refinement strategy included global minimization, local grid search, and ADP refinement, with default and secondary-structure restraints enabled. Refinement and validation statistics are summarized in Table S1.

46We performed molecular dynamics simulations of four simulation systems (A–D). System A (GPa3b17 partial) comprised gp100/HLA-A2 in complex with a truncated model of the high affinity GPa3b17 TCR, comprising TCR-α residues 9-201 and TCR-β residues 2-243. System B [wild-type (WT) partial] comprised gp100/HLA-A2 in complex with a truncated model of the parental WT TCR, comprising TCR-α residues 9-201 and TCR-β residues 2-243, with the following residues mutated to represent the parental TCR: αS95D, αM98L, αQ99V, βW51Q, βA52I, βQ53V, βG54N, βW96I, βA98G. System C (GPa3b17 full) comprised the fully assembled high-affinity gp100/HLA-A2/GPa3b17 TCR model as determined in this work, embedded in an asymmetric lipid bilayer representing the plasma membrane, following. The outward-facing leaflet had membrane composition POPC (33.3 mol%), PSM (33.3 mol%), and cholesterol (33.3 mol%); the cytosolic leaflet had membrane composition POPC (35 mol%), POPE (25 mol%), POPS (20 mol%), and cholesterol (20 mol%). System D (WT full) comprised the fully assembled gp100/HLA-A2/GPa3b17 TCR model as determined in this work, with mutations in TCR-α and TCR-β as in system (B), and a lipid membrane composition as in (C).

47All simulations were performed with GROMACS 2020.6 using the CHARMM36m forcefield for all solutes in combination with the TIP3P water model. We set up the simulation systems with CHARMM-GUI. All systems were solvated in a water box of starting dimensions 15.7 x 15.7 x 15.7 nm3 for systems A and B, and 13.3 x 13.3 x 22.5 nm3 for systems C and D. The lipid bilayer in C and D was orientated in the xy plane. We added Na+ and Cl- ions at 150 mM concentration. We energy-minimized all systems by steepest descent until convergence (tolerance: 1000 kJ/[mol nm]). All systems were first equilibrated in the NVT ensemble for 125 ps. For all simulations, the temperature was maintained at 300 K using the v-rescale thermostat with τt = 1 ps. Following the NVT equilibration, we equilibrated the systems in the NPT ensemble for 1.25 ns. The pressure was maintained at 1 bar using isotropic pressure coupling (compressibility K = 4.5x10-5 bar-1) with the Berendsen barostat (; τp=1 ps) for systems A and B, and semi-isotropic pressure coupling for systems C and D. We performed production simulations in the NPT ensemble for 1 μs, with pressure maintained at 1 bar using the Parrinello-Rahman barostat with τp = 5 ps. The first 100 ns of the production simulations were discarded as further equilibration. For equilibration simulations, we applied position restraints to the protein heavy atoms and, for systems C and D, lipid heavy atoms. For production simulations, no external restraints were applied. We used the python packages MDAnalysis and NumPy for the analysis of the protein systems.

48The shape complementarity was calculated with the CCP4 program sc. UCSF ChimeraX was employed to determine the buried surface area, using the default probe radius of 1.4 Å.

49Structure figures were prepared using UCSF Chimera and PyMOL (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC), except for Figure 7, which was prepared with the program Illustrate, using membrane coordinates from the OPM (; PDB ID: 6JXR). To depict the CD2-CD58 complex, structures of the CD2 ectodomain (PDB ID: 1HNF) and of a CD58 (domain 1)/CD2 (domain 2) chimera (PDB ID: 1CCZ) were superimposed onto the structure of the CD2/CD58 complex comprising the N-terminal domains only (PDB ID: 1QA9). To illustrate CD8αβ binding, the MHC α3 domain of the CD8αβ-H-2Dd complex structure (PDB ID: 3DMM) was superimposed onto the same domain of the gp100/HLA-A2/GPa3b17 TCR complex. To analyze the position of the CD8 binding site in the canonical and reversed pMHC I/TCR docking polarities, TCR-αβ of the B17.C1-pMHC I (PDB ID: 7JWJ), B17.R1-pMHC I (PDB ID: 5SWZ), and B17.R2-pMHC I (PDB ID: 7JWI) structures were superimposed onto TCR-αβ of our gp100/HLA-A2/GPa3b17 TCR complex.

50The data shown in Figures S1A and S1B are representative of 3-4 replicate experiments. Reported resolutions of the cryo-EM map are based upon the 0.143 Fourier Shell Correlation criterion. Model validation statistics (Table S1) were computed as implemented in MolProbity of the Phenix software package.