Stress-testing the relationship between T cell receptor/peptide-MHC affinity and cross-reactivity using peptide velcro

Significance

T cells recognize their targets through the T cell receptor (TCR). The affinity of a typical receptor for an agonist peptide-major histocompatibility complex (pMHC) molecule is extremely weak, and TCRs are known to be cross-reactive for related peptides. However, there are known TCR/pMHC interactions that occur at weaker affinities, such as in thymic selection and recognition of self-antigens, yet little is known about the identity of these peptides. We show that TCR/pMHC interactions of extremely low affinities remain highly specific, which informs of the nature of extremely weak affinity ligands. We also show that a peptide “velcro” can induce peptide-dependent T cell activation, providing a method for increasing the potency of a target, which is useful in immunotherapy.

Abstract

T cell receptors (TCRs) bind to peptide-major histocompatibility complex (pMHC) with low affinity (K_d ∼ μM), which is generally assumed to facilitate cross-reactive TCR “scanning” of ligands. To understand the relationship between TCR/pMHC affinity and cross-reactivity, we sought to engineer an additional weak interaction, termed “velcro,” between the TCR and pMHC to probe the specificities of TCRs at relatively low and high affinities. This additional interaction was generated through an eight-amino acid peptide library covalently linked to the N terminus of the MHC-bound peptide. Velcro was selected through an affinity-based isolation and was subsequently shown to enhance the cognate TCR/pMHC affinity in a peptide-dependent manner by ∼10-fold. This was sufficient to convert a nonstimulatory ultra-low-affinity ligand into a stimulatory ligand. An X-ray crystallographic structure revealed how velcro interacts with the TCR. To probe TCR cross-reactivity, we screened TCRs against yeast-displayed pMHC libraries with and without velcro, and found that the peptide cross-reactivity profiles of low-affinity (K_d > 100 μM) and high-affinity (K_d ∼ μM) TCR/pMHC interactions are remarkably similar. The conservation of recognition of the TCR for pMHC across affinities reveals the nature of low-affinity ligands for which there are important biological functions and has implications for understanding the specificities of affinity-matured TCRs.

When the first reliable affinities between the T cell receptor (TCR) and peptide-major histocompatibility complex (pMHC) were measured (1, 2), a major surprise was their low affinities (K_d ∼ μM) in comparison to antibody/antigen interactions (1), which are typically high affinity (K_d ∼ nM). Although there are exceptions of high-affinity engineered TCRs (3), the low affinity between the TCR and pMHC has been validated by hundreds of measurements (4, 5). The TCR/pMHC interface has ∼1,000 to ∼2,000 Ų of buried surface area, but poor shape complementarity between the interacting surfaces (6, 7) has been suggested to explain, in part, the low affinity of TCR/pMHC interactions.

The relatively low affinity of the TCR has been proposed to facilitate “scanning” of the diverse pMHC repertoire (8, 9). TCR cross-reactivity has been attributed to a variety of mechanisms, such as TCR binding in altered docking geometries (10), conformational flexibility in the TCR/pMHC complex (11), peptide molecular mimicry (12), register shifting in peptide ligands (13), and degeneracy at ancillary peptide positions not directly recognized by the TCR (4, 10, 14). Because typical affinities fall within the micromolar affinity range, many interactions are difficult to measure via biophysical measurements or structural characterization, leaving a dearth of knowledge about how TCRs recognize ligands with weak affinities (K_d > 100 μM).

This ability to survey different peptides is important for thymic selection and peripheral surveillance against pathogens or self (4, 15–17). Positively selecting ligands for class I-restricted TCRs are claimed to be as much as an order of magnitude lower affinity than typical TCR ligands for mature T cells (18–23). However, the relationship between activating and positively selecting ligands has been controversial. While some results show that relatively diverse repertoires can develop from a single positively selecting ligand (24–27), indicating TCR degeneracy for peptides, others argue that the repertoire is limited due to peptide specificity for positively selecting ligands (28, 29).

Relevant studies attempting to elucidate the relationship between TCR/pMHC cross-reactivity and affinity have utilized methods that alter the interaction by mutating the TCR/pMHC interface (3, 30, 31), which changes the nature of the binding interaction. We adapted a combinatorial library approach to ask whether we could profile TCR/pMHC cross-reactivity in both low- and high-affinity regimes without making any mutations to the interaction surfaces. To accomplish this, we engineered a weak second interaction site termed “velcro,” independent of the canonical interface between the TCR and pMHC, without mutating the MHC, peptide, or TCR. This molecular velcro provides an ∼10-fold apparent affinity gain that remains dependent upon the cognate TCR/pMHC interaction. When this sequence is linked to a library of peptides displayed on the mouse class II MHC I-E^k, we find that the TCR retains specificities for similarly related peptides, indicating that cross-reactivity is not gained at lower affinities. In addition, velcro is sufficient to transform nonagonist ligands into agonists in a TCR- and peptide-specific manner. These results show that a high degree of TCR/pMHC specificity is maintained even at very low (>100 μM) affinities and provides a method to detect ultra-low-affinity peptides to a TCR of interest.

Results

Generating Velcro Using a Low-Affinity 2B4 TCR Peptide Ligand.

We have previously studied the structural basis of interactions and cross-reactivities of TCRs restricted to the murine class II MHC I-E^k (4, 32). These studies identified hundreds of TCR ligands, the majority of which resembled the previously known cognate antigen sequence. While we identified many low-affinity ligands (with K_d values >100–200 μM), these selections potentially excluded peptides of even lower affinity. These ligands can inform us of the nature of low-affinity peptides relevant in TCR repertoire development and maintenance, such as tonic signaling (33), coagonism (22), and positive selection (20).

To improve the sensitivity of our selections, we sought to engineer a small peptide sequence that formed a second, very weak interaction site with the TCR, structurally independent from the TCR/pMHC binding interface. The avidity gain afforded by the two-site binding interaction between the MHC and the TCR (i.e., site 1, the cognate TCR/pMHC interface; site 2, the velcro contact) would provide an affinity gain for bona fide TCR/pMHC interactions sufficient to search for lower affinity peptides (Fig. 1A). If such a velcro sequence were linked to a pMHC of interest, the avidity would increase the effective affinity and, potentially, activity of the pMHC. Importantly, the velcro should increase the affinity of the TCR/pMHC interaction in a peptide-dependent manner so as to not inadvertently engineer a synthetic superantigen-like molecule that could nonspecifically activate T cells by strongly cross-linking TCR and MHC regardless of the presented peptide (34). Thus, the peptide velcro should not recover null peptide antigens from the combinatorial selections.

Fig. 1.

Design and isolation of velcro using yeast display of pep17–I-E^k. (A) Conceptual schematic of velcro. (B) 2B4 TCR tetramer staining of uninduced yeast (negative control) or a yeast-displayed I-E^k platform presenting either cognate MCC peptide or pep17. (C) Design of the velcro library fused to pep17–I-E^k. Dotted lines represent covalent linkers. (D) c-Myc enrichment of the yeast bearing a velcro library over two rounds of selection. (E) 2B4 fluorescent TCR tetramer staining of library rounds 3 and 4 costained with c-Myc. (F) 2B4 fluorescent TCR tetramer staining of the three converged velcros displayed with pep17–I-E^k.

To engineer the velcro, we exploited the fact that many yeast clones previously identified in peptide–I-E^k libraries enriched when selected with 2B4 TCR-coated streptavidin beads yet did not stain with TCR tetramers (4), likely due to very weak TCR/pMHC interactions (Fig. 1B). One such candidate was the 17th most abundant peptide from the selections (pep17; ADSLSFFSSSIK) (Fig. 1B), which shows some homology to the cognate epitope moth cytochrome C (MCC; ADLIAYLKQATK) and similarities to an MCC-related peptide (2A; ADPVAFFSSAIK), whose structure was previously determined in complex with 2B4 TCR in the context of I-E^k (4).

Our strategy to create velcro was to select for a short peptide that, when linked to pep17, restored 2B4 TCR tetramer staining. We therefore introduced an eight-amino acid random peptide library to the N terminus of pep17, linked via a six-amino acid Gly-Ser sequence (4, 15–17) (Fig. 1C). We hypothesized that this would allow for the creation of a binding interface large enough to improve the binding of pep17–I-E^k to 2B4 TCR without creating an interface so large as to make binding to 2B4 TCR peptide-independent. The six-amino acid linker between pep17 and the random library was sufficiently long to allow sampling of much of the TCR surface by the random peptide library, but it was not directed toward any given potential binding site on the TCR.

The 2B4 TCR was used to positively select velcro sequences using pep17 fixed in the I-E^k display. After five rounds of selection (three rounds using TCR bound to streptavidin-coated beads and two using TCR tetramers), we observed enrichment of the library for an epitope tag marking display of the pMHC protein complex (Fig. 1D) and recovery of 2B4 TCR tetramer staining of the library (Fig. 1E and SI Appendix, Fig. S1A). Sequencing 12 yeast clones from a sorted population of tetramer-positive yeasts revealed a conserved motif along the first six of eight amino acid positions of the random peptide library [YX(V/I)VP(D/E)]. When we examined 2B4 tetramer staining of clones containing the velcro sequences linked to pep17, we observed that all three velcro sequences rescued tetramer staining to a level comparable to the wild-type MCC peptide (Fig. 1 B and F). Velcro clone 17lib-6 (YVVVPDGT) displayed the brightest TCR tetramer staining of the three sequences (Fig. 1F), and this was the sequence chosen for further characterization. Interestingly, the remaining nine clones either had a mutation in the MHC (V65αI) or a mutation in the peptide [S(−2)P] that restored tetramer staining to varying degrees (SI Appendix, Fig. S1 B and C).

Velcro Conjugated to TCR-Specific Peptides Improves Affinity and Cell Activity.

To characterize velcro’s effects on 2B4–I-E^k binding affinity and T cell activation, we synthesized velcro (YVVVPDGT) with a high affinity (MCC), a low affinity (pep17), and a nonbinding peptide (5c.2) to 2B4 (Fig. 2). We exchanged each peptide into recombinantly expressed I-E^k and assayed each pMHC for binding to 2B4 TCR by surface plasmon resonance (SPR) (Fig. 2 A and B). Velcro increased the affinity of pep17 by ∼10-fold, from 159 μM without velcro to 15.4 μM with velcro present (Fig. 2 A and B). A corresponding fivefold increase in affinity was observed when velcro was linked to MCC, demonstrating that velcro’s effect is not dependent upon the identity of the pMHC used to select for the sequence (Fig. 2 A and B). When velcro was linked to 5c.2, which has no detectable affinity for the 2B4 TCR, there was still no binding observed; therefore, velcro does not convert a nonbinding pMHC into a TCR ligand (Fig. 2 A and B). We repeated this set of binding experiments with 5c.c7, a TCR that shares MCC–I-E^k as a cognate antigen and interacts with the 5c.2 peptide but does not interact with pep17 (Fig. 2 A and B). Velcro modestly enhances the affinity of 5c.c7 for both MCC and 5c.2, less so compared with 2B4 for MCC; however, velcro does not result in a detectable affinity for 5c.c7 when fused to pep17–I-E^k. These data confirm that velcro does not convert a nonbinding pMHC into a TCR ligand but, in fact, can enhance the affinity of a TCR for a peptide ligand.

Fig. 2.

Characterization of velcro. (A) Table of 3D K_d values and EC₅₀ values of 2B4 and 5c.c7 TCRs interacting with peptide–I-E^k with and without velcro. Underlined amino acids are peptide anchor positions for the MHC. N/A, not applicable; NB, nonbinder; ND, not determined. The K_d and EC₅₀ values are determined by SPR (B) and stimulation (C) assays measuring CD69 up-regulation after 24 h on a 2B4 hybridoma cell line mixed with an I-E^k–expressing CH27 cell line. Experiments are done as single experiments.

We next examined if affixing velcro to peptides potentiates TCR activation. Using a 2B4-expressing T cell hybridoma, we observed that velcro was able to convert pep17 from a nonagonist peptide into an agonist more potent than MCC based on CD69 activation (Fig. 2C). We see that velcro does not nonspecifically induce 2B4 TCR activity when linked to a noninteracting peptide, as in the case for the known 2B4 nonbinder MCC K99E–I-E^k (35) (Fig. 2C).

Structural Characterization of Velcro–Pep17–I-E^k with 2B4 TCR.

Our initial characterization of velcro suggested that its binding epitope on the TCR was independent of the TCR/pMHC interface, as velcro could potentiate affinity and signaling for known peptide binders across different TCRs. To determine the molecular basis of velcro’s interaction with TCR, we solved the crystal structure of velcro-pep17 presented by I-E^k bound to the 2B4 TCR via molecular replacement (Fig. 3A and SI Appendix, Table S1). Upon inspection of the TCR surface, we observed an area of unmodeled electron density located underneath the 2B4 Vβ-chain (Fig. 3 A–C). This density was sufficient to model the first six amino acids of velcro (YVVVPD) (Fig. 3 B and C), whose placement was validated by generating a simulated annealing omit map (Fig. 3C). This omit map removes the bias of the placement of velcro to ensure accurate structure determination.

Fig. 3.

Structure of velcro–pep17–I-E^k in complex with 2B4 TCR. (A) Overall structure of the entire velcro–pep17–I-E^k in complex with the 2B4 TCR. (B) Molecular interactions between velcro and the 2B4β-chain (orange) and 2B4α-chain (red). (C) mF_o-DF_c omit map contoured at 3σ to verify placement of velcro. (D) Model comparison of 2B4 and 5c.c7 [PDB ID code 3QJH (32)] TCRs in proximity to velcro.

Velcro primarily forms contacts with the TCR Vβ C′′ strand and the helix between the C′′ and D strands (Fig. 3B). The N-terminal Tyr of velcro forms a hydrogen bond with the carboxylate group of 2B4 Glu-62β. Val-2 and Val-4 of velcro each form hydrogen bonds with the C′′ strand of the TCR: Val-2 binds with Leu55β and Gln57β, while Val-4 interacts with Gln57β (Fig. 3B). Velcro also makes contacts with 2B4α, creating a hydrogen bond between the carbonyl oxygen of Pro-5 and the amide group of Asn95α of the TCR CDR3α loop (Fig. 3D). The velcro buries 528 Å of its 1,095 Å total solvent-accessible surface area in binding to 2B4. Interestingly, when we align the 5c.c7/5.c2–I-E^k structure to 2B4/velcro–pep17–I-E^k, the corresponding residue on the 5c.c7 α-chain (Lys99α) is 7.3 Å away from Pro-5, compared with the 3.4-Å distance to Asn95α on 2B4. Given that the TCRβ epitope that velcro binds is common between 2B4 and 5c.c7 TCRs, it is possible that this hydrogen bond is the basis of the attenuated effect on affinities observed for 5c.c7 (Fig. 2 A and B). The interface between velcro and the TCR is largely independent of the TCR/pMHC interaction site, as there is a sufficiently long linker (between the designed residues and the two C-terminal residues from the library) to allow for some degree of conformational flexibility of the N terminus of the peptide relative to velcro. Together, this allows velcro to enhance binding independent of the peptide bound in the MHC or the relative TCR/MHC orientation; however, the previous results show that this does not lead to peptide-independent recognition by the TCR.

Velcro Enables Highly Sensitive TCR Selections of a Peptide–I-E^k Library.

We next sought to determine how the improvement of affinity by addition of velcro affected the pMHC recognition landscape for three Vβ3 TCRs of interest that had been previously used to select peptides from an I-E^k library (4). We created a peptide library displayed by I-E^k on yeast, randomizing every position from P(−2) to P10, with the outermost residues and anchor residues given only limited diversity to ensure the majority of the peptides in the library were able to be displayed by the MHC, as described previously (4) (Fig. 4A). Importantly, this library had the YVVVPDGT velcro sequence fused to the N terminus of the peptide through a Gly-Ser linker, providing an affinity enhancement for any selected peptide (Fig. 4A). This would potentially allow peptide sequences to be discovered that were previously below the sensitivity floor of the selections.

Fig. 4.

TCR peptide specificity is conserved between low- and high-affinity interactions. (A) Schematic of yeast-display I-E^k peptide library design. (Left Inset) MCC peptide structure with the 2B4 CDR3 loops [PDB ID code 3QIB (32)]. (Right Inset) Library design to accommodate a diverse peptide pool presented by I-E^k. (B) Plot showing peptide enrichment over rounds of TCR selection. (C) Heat maps generated from round 3 sequencing data incorporating peptide abundance, identifying the selected probability distribution of amino acid identities per position of the peptide selected by 2B4, 226, and 5c.c7 TCRs with and without velcro. Selection results without velcro were reprinted with permission from ref. 4. Outlined boxes identify the amino acids of the cognate antigen MCC.

We conducted four rounds of selection with the 2B4 TCR, the TCR for which velcro was originally selected, as well as for the related MCC–I-E^k–specific 226 and 5c.c7 TCRs. Like in previous selection experiments done without velcro (4), peptides were enriched in the selection by round 3 based on the number of unique peptides by rounds of selection and Simpson’s diversity index (Fig. 4B and SI Appendix, Fig. S2A). After analyzing the deep sequencing data at round 3 of the selection, taking into account frequency according to deep-sequencing reads, we saw little deviation from the allowed residues previously described for the 226 and 5c.c7 TCRs (4) (Fig. 4C and SI Appendix, Fig. S2B). There is a general consistency of the amino acid preferences for selections with and without velcro, which indicates a conservation of cross-reactivity at lower and higher affinity ends of the TCR/pMHC binding spectrum.

Conserved Peptide Recognition Between “Low”- and “High”-Affinity Interactions.

For the 2B4 TCR, there was a subtle change in allowed residues of the peptide when velcro was added to the selections (Fig. 5A). Whereas 2B4 was previously the most peptide-restrictive of the three tested I-E^k–restricted TCRs, velcro seemingly relaxed the composition of the allowed residues along the peptide (Fig. 5A). For example, Arg, Val, Ile, and Phe were now observed at the central P5 TCR contact where, previously, 2B4 was restricted to Lys, Met, and Ser (4) (Fig. 5A); however, these substitutions are fairly conservative expansions of the residues previously observed. Velcro selections accommodated additional residues at P8, including a predominant Ile or Leu, which was previously allowed but not favored (Fig. 5A). A common feature across the TCR is the broadening of allowed residues at MHC contact residues P4 and P6 (Fig. 5 A–C).

Fig. 5.

Quantification of TCR specificity between low- and high-affinity interactions. Results of the 2B4 (A), 226 (B), and 5c.c7 (C) selections with (Bottom) and without (Middle) velcro. Selection results without velcro were reprinted with permission from ref. 4. Bar graphs represent amino acids per position of the peptide selected at round 3 with a cutoff of 1% abundance. (Top) Structures of MCC with either 2B4 TCR [PDB ID code 3QIB (32)] or 226 TCR [PDB ID code 3QIU (32)]. No structure exists for MCC–I-E^k with 5c.c7. (D–F) Word logos of positions P(−1) to P9 of the peptide from the library are displayed using the unique peptides from round 3, disregarding the frequency of each peptide selected by either 2B4, 226, or 5c.c7. (G–I) 2B4, 226, and 5c.c7 average Hamming distances as box plots of peptides selected from the nonvelcro selections (green), velcro selections (red), and combined datasets (blue). Outliers are displayed as individual dots and are determined as being less than quartile 1 or greater than quartile 3 by 1.5-fold the interquartile range. The numbers of data points listed in order of nonvelcro, velcro, and combined selections, respectively, for 2B4 (5,767, 490, 6,257), 226 (13,990, 1,439, 15,429), and 5c.c7 (6,562, 126, 6,688) TCRs are shown.

The above analyses take into account the enrichment of a sequence from the selection data. When we consider all of the unique sequences in round 3 of the selection between nonvelcro and velcro selections regardless of frequency, we see that the selected populations for each TCR have similar peptide compositions (Fig. 5 D–F). In fact, identical peptide sequences were selected in both velcro and nonvelcro selections per TCR (SI Appendix, Fig. S2C).

A metric used to quantify the relative similarity of selected peptides within a selection is Hamming distance, which identifies the number of amino acid differences by pairwise comparisons. We compared all-by-all unique peptides via Hamming distance within the nonvelcro selections, velcro selections, and combined datasets (Fig. 5 G–I and SI Appendix, Fig. S3 A–C). When we compare the Hamming distance of the combined dataset with each of the selections with or without velcro, the Hamming distance does not dramatically differ, showing that the peptides selected between nonvelcro and velcro selections are similar. As a reference, when we determine the ratio of average Hamming distance to peptide length of all possible amino acid permutations, we see that the ratio approaches ∼0.95, indicating that a completely random set of 13mers will approach an average Hamming distance of ∼12.35 (SI Appendix, Fig. S3D). Thus, the velcro- and nonvelcro-selected peptides are more similar to one another than a completely random set of peptides.

A QY Family Peptide is Converted into a Potent T Cell Agonist with Velcro.

While most selected sequences have apparent homology with those peptides selected without the aid of velcro, we used covariation analysis to identify sets of sequences that differ from the MCC-like motif. We found a small family of peptides that contain P3 Gln and P8 Tyr, with some variation allowed at P5 for the 2B4 selections (Fig. 6A). This specific peptide family was not present for the velcro selections with the 226 or 5c.c7 TCRs.

Fig. 6.

Identification of an additional family of peptides with stimulatory potential. (A) Covariation analysis of P3 and P8 of the peptides identified from the 2B4 selections (red), 226 selections (blue), and 5c.c7 selections (green) without and with velcro, respectively. The outlined box represents the QY family. (B) SPR of MCC and QYpep2 with velcro. RU, resonance unit. (C) Activation of a 2B4 hybridoma via CD69 from a coculture with CH27 cells loaded with peptide as analyzed by flow cytometry. MFI, mean fluorescence intensity. (D) Summary table of affinity and activity data. Underlined amino acids are peptide anchor positions for the MHC. N/A, not applicable; ND, not determined.

To determine if the family of peptides discovered within the velcro selections was functionally active, we examined one of the peptides with the P3 Gln/P8 Tyr motif QYpep2 for binding and ability to induce T cell signaling. We found that velcro was able to enhance the affinity of QYpep2 for the 2B4 TCR from >1 mM to 35 μM, which is well within the range of T cell agonists (Fig. 6 B and C). Correspondingly, QYpep2-velcro activated the 2B4 hybridoma with the same functional activity as MCC via CD69 activation (Fig. 6C). Interestingly, velcro provided no significant improvement to the activity of MCC (Fig. 6C), suggesting that MCC signaling potency may be limited by an affinity ceiling for T cell signaling potency, consistent with previous observations (36, 37), or a factor independent of TCR/pMHC affinity, such as the efficiency of peptide loading onto the MHCs on the antigen-presenting cell (APC) surface. These data suggest that velcro could be used as a general strategy to discover new T cell agonist peptides as well as to increase the functional avidity of low-affinity peptides (Fig. 6D).

Discussion

It has generally been assumed that the low affinity of TCR/pMHC interactions enables both scanning and cross-reactivity (8, 9). Here, we utilized a protein engineering approach to generate a de novo peptide, termed velcro, which enabled us to measure the pMHC cross-reactivity of a TCR in both high- and low-affinity interactions without altering the interaction interface of either protein. Surprisingly, we find that increasing the apparent affinity of the TCR/pMHC interaction does not significantly alter the cross-reactivity profile of a TCR. Based on the analysis of this model system of known TCR/pMHC interactions, the low affinity of the TCR/pMHC binding interaction does not endow the TCR with promiscuous peptide recognition.

While T cells interact with different antigens with a wide range of affinities (38), even “high-affinity” TCR ligands bind to TCRs with objectively low affinities compared with many other protein/protein receptor/ligand interactions (e.g., antibody/antigen, growth factor receptor, cytokine receptors) (39, 40). The relationship between affinity and specificity has not been studied extensively, despite the diversity of protein/protein interactions (41). Intuitively, one might expect a correlation between the affinity and the specificity of protein/protein interactions (42). We observe that the cross-reactivity for high-affinity TCR ligands is maintained compared with low-affinity TCR ligands. Our data show conservation of peptide motifs between the selections with and without velcro, which implies that cross-reactivity of TCR/pMHC interactions throughout a wide range of affinities is mediated through defined interactions. Understanding the relationship between TCR/pMHC cross-reactivity and affinity is of paramount importance in the use of TCR-based therapies. It is important to consider the implications of affinity maturation on TCR cross-reactivity and the introduction of novel specificities that may occur across affinity regimes (43–45). Mutations to the TCR that create higher affinities to peptides may introduce new specificities that can propagate to lower affinity peptides. We posit that velcro could be used as a tool to probe the specificities of engineered receptors.

While both CD4 and CD8 coreceptors provide additional signaling capacity through the recruitment of Lck (46, 47), they also provide additional stability to the TCR/pMHC interaction (48). We suggest that velcro acts as a surrogate coreceptor by generating peptide-dependent avidity and increased signaling capacity to the TCR/pMHC interaction much like CD4 and CD8. Several studies using mice deficient in CD4 and CD8 coreceptors suggest that the specificity of the TCR/pMHC interaction is primarily a result of coreceptor engagement (49, 50). Here, we examine the specificity of the same TCR/pMHC interactions in a regime analogous to coreceptor-dependent and independent states. We observe that the TCRs exhibit striking pMHC specificity in both the absence and presence of the velcro. Indeed, the specificity of the TCRs examined for their pMHC ligands in the absence of the velcro is supportive of intrinsic affinity, as well as germline-constrained docking solutions between TCR and MHC (10). Second, the observation of relaxation to a broader set of residues at neutral or MHC-facing positions of the peptide is consistent with findings previously made (51). These variant peptides required CD8 coreceptor to induce T cell activation, whereas those with unaltered residues at MHC-facing residues did not. Like CD8, velcro provides an affinity gain to convert nonagonist ligands to agonist ligands.

The engineered velcro achieves an affinity gain of approximately fivefold to 10-fold between a TCR and peptide agonist and activates T cells in a peptide-dependent manner. The affinity floor for TCR/pMHC interactions is currently unknown, and there is no tractable and systematic method to identify and characterize interactions within the affinity regime of these “weak” recognition events. For a given TCR, identifying previously undetectable peptide ligands can be used to understand weak recognition events of self-antigen in positive selection, co-agonism, and tonic signaling. Previous work in studying the T cell specificity in positive selection has been controversial (24–26, 28, 29, 52). Whether or not T cells are seeing specific peptides during positive selection has not been definitively answered. However, the results identifying the exquisite specificity of the TCR at ultra-low affinities suggest that positive selection is governed by peptide-specific recognition resembling that of an agonist.

The results obtained in this study (i) demonstrate the ability to engineer a protein/protein interaction site to alter the apparent affinity of the TCR/pMHC binding event and (ii) highlight the relationship between affinity and specificity/cross-reactivity in a weak affinity interaction, with important implications in protein engineering, T cell biology, and the nature of affinity and specificity in protein/protein interactions. This work can be extended by generating a “universal velcro” by engineering a weak interaction with the TCR constant region. Additionally, this method can be useful to study cross-reactivities of human TCRs for which there are clinical uses and/or human TCRs that potentiate responses to specific antigens.

Methods

Expression of TCRs.

TCRs were generated as done previously (4). Briefly, V_mC_h chimeric TCRs were generated containing an engineered C-domain disulfide and cloned into the pAcGP67a insect expression vector (BD Biosciences), with a C-terminal 3C protease site followed by either an acidic GCN4-zipper-biotin acceptor peptide-6xHis tag or a C-terminal basic GCN4 zipper-6xHis tag. The vectors were used to generate baculovirus in SF9 cells via cotransfection of Bestbac 2.0 DNA (Expression Systems) with Cellfectin II (Life Technologies). Separate viruses were generated for each chain and used to coinfect 2 mL of High Five cells to titrate for a 1:1 stoichiometric expression of TCR chains as determined by SDS/PAGE gel and Coomassie staining.

TCRs were then expressed in large scale using 1 L of High Five cells coinfected for 48 h at 28 °C. Supernatant was collected and treated with 100 mM Tris⋅HCl (pH 8.0), 1 mM NiCl₂, and 5 mM CaCl₂ to precipitate contaminants. The supernatant was treated to batch Ni-NTA purification (Qiagen) at room temperature for 3 h. Nickel resin was washed twice with 20 mM imidazole in 1× Hepes-buffered saline (pH 7.2) before being eluted with 200 mM imidazole. The elution was concentrated in a 30-kDa filter and biotinylated overnight at 4 °C using BirA ligase, 100 μM biotin, 50 mM bicine (pH 8.3), 10 mM ATP, and 10 mM magnesium acetate. Protein was purified by size exclusion chromatography using a Superdex 200 column with an AKTAPurifier (GE Healthcare). The 2B4 TCR used for crystallography and SPR was not treated with biotinylation reagents, but cleaved with 3C protease and carboxypeptidase A at 4 °C overnight.

Expression and Library Generation of the Yeast-Displayed I-E^k.

The I-E^k yeast-displayed construct was designed as previously described (4) (Fig. 1C) utilizing the native Aga2p/Aga1p protein/protein interaction for display. A myc epitope tag was used to monitor expression of the library. To generate the various libraries, degenerate oligos were designed using NNK codons to encode the 20 amino acids. Libraries were constructed as described previously (4), with the exception that a BamHI restriction site was introduced in the linker between the velcro sequence and the peptide displayed by the MHC so that the velcro itself could serve as the homologous recombination sequence. (NNK)₈ codons were used to generate the velcro library, which was linked to the N terminus of the peptide using GGSGSG. For the peptide library, positions were limited to a restricted set of amino acids using codon sequence RMANNKVTT(NNK)₇AAARVA (Fig. 4A). The DNA library is generated using a nested PCR, in which the first PCR generates the DNA library from a stop codon cassette and the second adds flanking nucleotides to overlap with the cut vector. Both reactions used 30 cycles with a 0:10 ratio at 95 °C, a 0:30 ratio at 60 °C, and a 1:30 extension phase at 72 °C. The library (100 μg) and cut vector (20 μg) were coelectroporated into chemically competent EBY100 cells. The library diversity is calculated based on colony formation from limited dilutions using SDCAA plates.

Yeast libraries were cultured at 1 × 10⁷ yeasts per milliliter in SDCAA (pH 4.5) at 30 °C and induced for protein expression in SGCAA (pH 4.5) at 20 °C for 2–3 d. Yeasts are collected for selection and subsequently grown overnight in SDCAA before repeating induction. Libraries are frozen in 2% glycerol and 0.67% yeast nitrogen base at 10-fold diversity of the library and stored at −80 °C.

Selection of Velcro from a Random Library.

Yeasts expressing peptides selected by the 2B4 TCR (4) were tested for their ability to stain with 500 nM 2B4 TCR tetramers using streptavidin-647. Pep17 was fixed in the yeast construct to allow selection for velcro peptides. The library was selected for three rounds using 2B4 TCR streptavidin-coated magnetic beads (Miltenyi), followed by two rounds with 400 nM TCR tetramer. The first selection used 10-fold the diversity of the yeast library. Yeasts were washed in 1× PBS, 1 mM EDTA, and 0.5% BSA (PBE). Yeasts were then resuspended in 10 mL of PBE with 250 μL of streptavidin-coated magnetic beads for negative selection for 1 h at 4 °C. Yeasts were passed through a cell strainer into an LS column (Miltenyi) attached to a magnetic stand (Miltenyi) and washed three times with 3 mL of PBE, and the flow-through was used for further selection. The yeasts were resuspended in 10 mL of PBE with preincubated 250-μL streptavidin-coated magnetic beads and 400 nM biotinylated TCR for 15 min at 4 °C. The yeasts and TCR-coated beads were allowed to mix for 3 h at 4 °C before being washed in PBE and placed over an LS column. After three washes with 3 mL of PBE, the elution was collected, washed in SDCAA, and cultured in SDCAA overnight. Approximately 30 million yeast cells were resuspended in SGCAA for protein induction. The two rounds of selection with 400 nM TCR tetramer used anti-647 magnetic beads (Miltenyi) instead of streptavidin-coated beads. These selections were made similar to the previous selections, except for the use of anti-647 beads in the negative selection, generation of the streptavidin-647 TCR tetramer, and addition of TCR tetramer in the positive selection for 2.5 h at 4 °C followed by a wash in PBE and incubation with anti-647 beads for 20 min at 4 °C. All rounds of selection were monitored using either anti-Myc (Cell Signaling) and/or 100 nM or 400 nM TCR tetramer staining.

After the final round of selection, the yeasts were grown in SDCAA, induced in SGCAA, and tetramer-stained with varying concentrations of TCR. The sample stained with 500 nM tetramer was sorted using a FACSJazz cell sorter (BD Biosciences), and the top 5% of the TCR⁺Myc⁺ population was sorted. The DNA was extracted from the sorted yeasts by miniprep (Zymo Research) and cloned into DH5α cells, and 12 individual colonies were sequenced using Sanger sequencing. The original yeast clones were stained with 500 nM 2B4 tetramer.

Velcro-Assisted pMHC Selections.

Nonvelcro selection data were taken from Birnbaum et al. (4). Velcro selections were done using three rounds of selection by streptavidin-coated magnetic beads and one round of TCR tetramer and anti–647-coated magnetic beads. Following the final selection, 5E7 yeasts were taken from all rounds of the selection for DNA miniprep and subsequently used for preparation for deep sequencing.

DNA was prepared using a series of two PCRs as done previously (4). The first PCR adds a selected 6-nt barcode and random 8mer nucleotide sequence, and the second adds the Illumina adapters. The final amplicon design contains Illumina P5-Truseq read 1-(N₈)-Barcode-pMHC-(N₈)-Truseq read 2-IlluminaP7. The primers amplify from the velcro sequence to the MHC β1 domain to amplify a short product <150 bp. The DNA library was purified by agarose gel purification, quantified by nanodrop and/or a Bioanalyzer (Agilent Genomics), and deep-sequenced (biosample accession nos. in the National Center for Biotechnology Information Sequence Read Archive are SAMN08728345, SAMN08728346, and SAMN08728347). The DNA library was prepared per a 2 × 150 V2 kit using the low-diversity library protocol. Data were analyzed using PandaSeq to generate paired end reads and deconvoluted by barcode in Geneious. For quality control, all sequences were ensured to have the YVVVPDGT velcro sequence. Any sequences with less than 10 reads and located only one nucleotide away from another sequence were aggregated to accommodate sequencing errors. The sequencing round counts were tabulated for every peptide across the rounds. Any sequences containing the nucleotide sequence of the same pep17 used to generate the velcro sequence were manually removed, as this was likely contamination.

Data were processed and analyzed using Perl and R (version 3.4.1). For analyses to consider the frequency of sequencing reads per peptide, a cutoff of 1% was implemented. Simpson’s diversity index was calculated using the following formula:

$$D=\raisebox{1ex}{${\displaystyle \sum }n\left(n-1\right)$}\!\left/ \!\raisebox{-1ex}{$N\left(N-1\right)$}\right.$$

in which D = Simpson’s diversity index, n = sequencing count of a peptide in a given round, and N = sequencing count of all peptides in a given round. The Simpson’s diversity index was calculated for all unique peptides in round 3 of the selections. MATLAB (version R2014a; MathWorks) was used to determine Hamming distances between unique peptides from round 3 of the selection. To determine the average Hamming distance of a selection, each peptide is compared with all other peptides within the group, and each pairwise Hamming distance is calculated, summed, and averaged by the total number of comparisons. The velcro and nonvelcro selection datasets were combined before the average Hamming distance was calculated. To determine the average Hamming distance of a completely random set of sequences, all permutations of peptides of various lengths (2–13 amino acids) with a given set of available amino acids (2–20 amino acids) were simulated. Data points were purposefully not calculated due to runtime complexity.

Affinity Measurements Between TCR and Velcro.

Recombinant, biotinylated I-E^k with CLIP peptide fused to the N terminus of the MHC β-chain was created as described previously (4). Briefly, CLIP–I-E^k was expressed as two constructs, one having CLIP–I-E^k β and the other construct, I-E^k α, having acidic/basic zippers as used for the TCRs. The peptide was separated from the β-chain using a Gly-Ser linker containing a thrombin cleavage site. The constructs were used to produce protein in the same manner as the TCRs described above. Peptides were synthesized both with and without the selected velcro sequence (Genscript) and were exchanged into the MHC by adding 10 μM peptide and thrombin at 37 °C for 1 h, followed by a 37 °C incubation overnight in 30 mM citrate buffer (pH 6.2). The reaction was then neutralized with 40 mM Hepes buffer (pH 7.2). Protein was purified by size exclusion using a Superdex 200 column and AKTAPurifier.

SPR measurements were conducted on a Biacore T100 system using streptavidin chips with the MHC immobilized at low density (∼100–200 resonance units) to avoid rebinding essentially as described previously (4). MCC K99E–I-E^k, a pMHC with no detectable binding affinity for 2B4 or 5c.c7, was immobilized on the reference channel of the streptavidin chip. SPR runs were done in HBS-P+ buffer (10 mM Hepes pH 7.4, 150 mM NaCl, 0.5% vol/vol surfactant P20) with 0.1% BSA to reduce nonspecific binding. All measurements were made with threefold serial dilutions of insect-expressed TCR with a 60-s association followed by a 600-s dissociation at a 10–30 μL⋅min⁻¹ flow rate. The K_d was determined using equilibrium measurements. Measurements of previously reported affinities were validated using our SPR measurements (4), which ensures our affinity measurements and analyses are accurate.

Functional Validation of Velcro Linked to Peptide.

A T cell hybridoma line expressing the 2B4 TCR was stimulated using the I-E^k–expressing CH27 APCs pulsed with peptides of interest for 3 h at 37 °C as previously described (4). The cells were cocultured overnight at a 1:2 T cell/APC ratio with a titration series of each peptide starting at 50 μM. Cells were then centrifuged, washed in PBE twice, and stained with anti–CD69-PE (clone H1.2F3; eBiosciences) and anti–CD4-APC (clone GK1.5; eBioscience). Cells were fixed in PBS containing 1.6% paraformaldehyde for 15 min at room temperature and washed once before being analyzed by flow cytometry using an Accuri C6 instrument (BD Biosciences).

Structural Characterization of Velcro–Pep17–I-E^k in Complex with 2B4 TCR.

The velcro was linked to the I-E^k β insect expression construct with the N terminus of the velcro sequence cloned to be flush with the gp67 leader peptide. The 2B4 TCR and velcro–pep17–I-E^k were both expressed via recombinant baculovirus infection of High Five cells as previously described, cleaved with 3C protease to remove acidic/basic leucine zippers that aided in proper chain pairing and carboxypeptidase A, and purified via size exclusion. The purified material was mixed at a 1:1 molar ratio and concentrated to ∼15 mg/mL for crystallography. Crystals formed in sitting drop trays in 18–20% PEG 3350 and 20 nM Na/K phosphate. Crystals were flash-frozen in liquid nitrogen in mother liquor containing 25% ethylene glycol, and datasets were collected at the Stanford Synchrotron Radiation Lightsource using beamline 11-1. Data were processed using XDS/XSCALE (53) and were solved with molecular replacement using the 2B4 TCR (with CDR3 loops deleted) and I-E^k (with peptide deleted) from the previously solved 2B4–2A–I-E^k structure [Protein Data Bank (PDB) ID code 4P2O] using Phaser and iteratively refined with Phenix (54) and Coot (55). To confirm placement of the velcro, an mF_o-DF_c map was calculated using Phenix by omitting the peptide residues and performing simulated annealing before calculating the map with bulk solvent excluded from the omitted region. The crystallographic model and reduced data have been deposited in the PDB (ID code 6BGA).