Synthetic Peptides with Inadvertent Chemical Modifications Can Activate Potentially Autoreactive T Cells

Characterization of the peptide epitopes recognized by CD8⁺ T cells is important for understanding the immunobiology of cancer and infectious diseases and for the development of related immunotherapeutics. The use of synthetic peptides has been instrumental in this endeavor (1). T cell recognition is both highly specific, such that single amino acid substitutions in a given cognate epitope can abrogate functional responses, and highly cross-reactive, such that more than a million different peptides can be engaged productively by a single degenerate TCR (2). Epitope mapping is further complicated by the need to define individual restriction elements and the fact that antigenic peptides can be generated by splicing discontinuous segments of the same protein (3, 4).

TCRs can accommodate naturally occurring posttranslational modifications to agonist MHC class I–bound peptide structures arising from deamidation (5, 6), phosphorylation (7), nitration (8), and glycosylation (9). Similar peptide modifications, including citrullination (10) and nitration (11), have been described in the context of MHC class II. T cells can also recognize chemically modified peptides that do not occur in nature (12, 13). We recently described a CD8⁺ T cell clone (6C5) that recognized a peptide (LLSYFGTPT) modified via the addition of a tert-butyl (tBu) group to the tyrosine (Y) residue at position (P) 4 (14). This modification was likely introduced during peptide synthesis as a low-frequency contaminant (∼1%). We further showed that this nonnatural peptide bound poorly to HLA-A*02:01, and computer modeling suggested that the 3-tBu–modified Y residue was a likely contact point for the TCR (14). However, it was not clear why LLSY(3-tBu)FGTPT was immunogenic in the face of competition from unmodified peptides that were more abundant in the initial screening mixture and bound more strongly to HLA-A*02:01.

In this study, we used a variety of approaches to investigate the influence of chemical groups on T cell specificity. Our data revealed that molecular mimicry between chemically modified and naturally occurring peptides enabled the activation of potentially autoreactive T cells, exemplified by clone 6C5.

The CD8⁺ T cell clone 6C5 was originally derived and expanded from a patient with chronic lymphocytic leukemia (CLL) (14). Aliquots of cryopreserved T cells were thawed and rested overnight prior to use in functional assays as described previously (14). The clonality of 6C5 was confirmed by derivation of a single TCRα sequence (TRAV12-13*01/02/CAMSYYNNNDMR/TRAJ43*01) and a single TCRβ sequence (TRBV6-5*01/CASSSSYEQY/TRBJ2-7*01) from extracted mRNA (15).

6C5 T cells were prescreened using sizing scan peptide mixtures (16) and showed a preference for nonamer peptides (data not shown). Subsequent screens were performed using a nonamer combinatorial peptide library (CPL) synthesized in a positional scanning format (PepScan) as described previously (2, 16–18). Briefly, 6 × 10⁴ C1R–HLA-A*02:01 cells were preincubated for 2 h at 37°C with each peptide mixture (100 µM) in 96-well microplates (Corning) containing RPMI 1640 medium supplemented with 100 U/ml penicillin, 100 µg/ml streptomycin, 2 mM l-glutamine, and 2% heat-inactivated FCS (Thermo Fisher Scientific). After the peptide-pulsing phase, 3 × 10⁴ 6C5 T cells were added to each well, and the plates were incubated for 16–18 h at 37°C. All assays were performed in duplicate. Negative control wells lacked exogenous peptides, and positive control wells contained PHA (10 µg/ml; Sigma-Aldrich). Cell-free supernatants were collected after incubation and assayed for MIP-1β content by ELISA (17).

Data were processed as described previously to identify preferred amino acid residues at each P across the peptide backbone (2) and generate a predicted optimal epitope sequence (16, 18). Sequence-driven database searches were then conducted using the PI CPL webtool hosted by the Warwick Systems Biology Centre (18). Peptides were ranked in order of likelihood of recognition according to the agonist likelihood score generated by the PI CPL webtool search, and potential agonists were further stratified according to origin and predicted binding affinity for HLA-A*02:01 (www.iedb.org). The final list of candidate agonists included 11 peptides derived from human viruses and 26 peptides derived from human proteins (self). Three modified versions of LLSYFGTPT were also included as representatives of substitutions predicted to alter the TCR interaction (4Y > 4W), substitutions predicted to have no effect (4Y > 4F), or substitutions predicted to improve peptide binding to HLA-A*02:01 (9T > 9V). These individual peptides (n = 40) were synthesized commercially at ≥95% purity (Mimotopes). All peptides were dissolved in DMSO at 10 mg/ml stock and stored in aliquots at −80°C. Fresh aliquots were used for each experiment, and the final DMSO concentration in culture wells never exceeded 0.4%.

Initial screens were performed at a peptide dose of 4 µg/ml (∼4 µM). Briefly, 5 × 10⁴ T2 cells, which express HLA-A*02:01, were pulsed with peptide for 1 h at 37°C, washed, and cultured with 6C5 T cells at a ratio of 1:1 for 18 h at 37°C. Negative control wells contained unpulsed T2 cells with DMSO at a concentration of 0.4%, and positive control wells contained T2 cells pulsed with LLSY(3-tBu)FGTPT. Cell-free supernatants were collected after incubation and assayed for MIP-1β content by ELISA (17). Agonist peptides were tested similarly in dose-response experiments at concentrations ranging from 10¹ µg/ml to 10⁻⁶ µg/ml, stratified according to maximal production of MIP-1β. Dose-response curves were generated in Prism version 8 (GraphPad).

T cell functions were assessed using flow cytometry as described previously (14). Briefly, 6C5 T cells were incubated with T2 cells ± peptide (10 µg/ml) in the presence of GolgiStop (0.7 µl/ml; BD Biosciences) and anti-CD107a–FITC (clone H4A3; BD Biosciences). Negative control tubes contained equivalent DMSO or no T2 cells, and positive control tubes contained PMA (10 ng/ml; Sigma-Aldrich) and ionomycin (1.5 µg/ml; Sigma-Aldrich). Cells were incubated for 6 h at 37°C, washed with FACS buffer (Thermo Fisher Scientific), and stained for 15 min at 4°C with anti-CD8a–PerCP-Cy5.5 (clone SK1), anti-CD19–BV510 (clone HIB19), and Zombie NIR (all from BioLegend). After a further wash, cells were fixed/permeabilized using a FIX & PERM Cell Fixation & Permeabilization Kit (Thermo Fisher Scientific), washed again, and stained for 20 min at 4°C with anti–IFN-γ–V450 (clone B27; BD Biosciences), anti–IL-2–allophycocyanin (clone MQ1-17H12; BioLegend), anti–MIP-1β–PE (clone D21-1351; BD Biosciences), and anti–TNF-α–PE-Cy7 (clone MAb11; BioLegend). Data were acquired immediately using a FACSCanto flow cytometer (BD Biosciences) and analyzed using FlowJo software version 9.9.4 (FlowJo). Individual functions were quantified among single, Zombie NIR⁻, and CD19⁻ lymphocytes that expressed CD8.

Peptide binding was assessed indirectly by measuring the upregulation of HLA-A*02:01 on the surface of T2 cells (19). Briefly, 2 × 10⁵ T2 cells were incubated for 18 h at 37°C with each test peptide at a concentration of 50 µg/ml in serum-free RPMI 1640. Negative control tubes contained equivalent DMSO, and positive control tubes contained the Bax_136–144 peptide (IMGWTLDFL). Cells were then stained with anti–HLA-A*02:01–FITC (clone BB7.2; BioLegend). Data were acquired using an Accuri C6 flow cytometer (BD Biosciences) and analyzed using CFlow software (BD Biosciences). Upregulation of HLA-A*02:01 was calculated using the following formula: % increase = [(mean fluorescence with peptide − mean fluorescence without peptide)/(mean fluorescence without peptide) × 100].

Chemically modified peptides were synthesized at ≥95% purity (Mimotopes or PolyPeptide Group). Peptide modifications and sequences were verified by mass spectrometry with collision-induced dissociation using an amaZon SL Ion Trap (Bruker). All peptides were dissolved in DMSO at 10 mg/ml stock and stored in aliquots at −80°C. Fresh aliquots were used for each experiment, and the final DMSO concentration in culture wells never exceeded 0.4%.

Soluble peptide–HLA-A*02:01 complexes were generated and purified as described previously (20, 21). Purified proteins were set up for crystallization trials in a sitting-drop vapor-diffusion arrangement. Crystallization screen and protein solutions were dispensed in a standard automation-compatible 96-well microplate using an ARI Phoenix Robot (Alpha Biotech). Screens were performed using PACT Premier (Molecular Dimensions) and TOPS (22). All crystallization trials were conducted at 20°C. The best crystals were obtained with PACT Premier condition G06 (0.2 M Na formate, 0.1 M Bis-Tris propane, and 20% polyethylene glycol [PEG] 3350 [pH 7.5]) for IIGWMWIPV–HLA-A*02:01, PACT Premier condition D02 (0.1 M MMT buffer and 25% PEG 1500 [pH 5.0]) for LLGWVFAQV–HLA-A*02:01, and TOPS condition A12 (0.1 M cacodylate buffer, 5% glycerol, and 25% PEG 4000 [pH 6.0]) for LLSY(3-tBu)FGTPT–HLA-A*02:01 (Supplemental Table I).

Crystals were harvested into thin plastic loops and cryocooled in liquid nitrogen for transfer to the Diamond Light Source (Didcot, UK). Diffraction data were collected in single-axis rotation at 100 K on beamlines I02 and I03, and data reduction was completed automatically through XDS called from within XIA2 (23, 24). Data were scaled and merged using AIMLESS and TRUNCATE in the CCP4 suite (25). The structures were solved using molecular replacement in PHASER (26), with Protein Data Bank entry 5EU5 as the starting model (27), and subsequently refined using REFMAC5 (28). Data collection and refinement statistics are summarized in Supplemental Table I. The final model was adjusted to match density maps in COOT (29). Atomic coordinates were deposited in the Protein Data Bank (https://www.rcsb.org/) under accession codes 6Z9V for IIGWMWIPV–HLA-A*02:01, 6Z9W for LLGWVFAQV–HLA-A*02:01, and 6Z9X for LLSY(3-tBu)FGTPT–HLA-A*02:01.

Specific tests are indicated in the relevant figure legends. All statistical comparisons were performed using Prism version 8 (GraphPad).

The human CD8⁺ T cell clone 6C5 has been shown to recognize the synthetic peptide LLSY(3-tBu)FGTPT but not the unmodified peptide LLSYFGTPT (14). To investigate the fine specificity of this phenomenon, we tested the agonist properties of peptides modified with larger (di-tBu) or smaller chemical groups (Me) attached to the Y residue at P4 (Fig. 1A). As expected, the tBu-modified peptide (LLSY-tBu) induced multiple effector functions, namely degranulation, measured via the surface mobilization of CD107a, and the production of IFN-γ, IL-2, MIP-1β, and TNF-α (Fig. 1B). In contrast, the di-tBu–modified (LLSY-di-tBu) and Me-modified peptides (LLSY-Me) were largely inert, although the latter induced a small increase in the production of MIP-1β (Fig. 1B). Earlier modeling of the LLSY(3-tBu)FGTPT–HLA-A*02:01 complex suggested that the tBu-modified Y residue pointed up toward the expected P of the TCR (14). This model also predicted that a bulky amino acid might be able to mimic the structure of 3-tBu-Y. In our assays, however, a peptide containing tryptophan (W) in place of Y at P4 (LLSWFGTPT) failed to elicit any effector functions (Fig. 1B). Collectively, these data suggested that the 6C5 TCR was highly specific for the tBu modification at P4.

Although clone 6C5 was originally derived from a patient with CLL, no reactivity was detected against autologous or HLA-A*02:01–matched CLL cells (14). In an attempt to discover the natural epitope, we screened clone 6C5 against a nonamer CPL presented by C1R cells expressing HLA-A*02:01 (2). As expected for peptides that bind HLA-A*02:01, there was a preference for hydrophobic residues at P2 and P9 (Fig. 2A). There also appeared to be constraints on the number of amino acids that could be recognized in the middle portion of the peptide (P3–P6). This was particularly evident at P3, where alanine (A) and glycine (G) were the preferred residues, and at P4, where there was a clear preference for W (Fig. 2A).

The predicted optimal agonist sequence based on preferred residues across the entire peptide backbone (P1–P9) was IIGWMWIPV (Fig. 2B). No direct matches for this sequence were identified in a database search of human viral pathogens and self-proteins (18). However, a list of >400 peptide sequences ranked in order of likelihood of recognition was generated using this approach (data not shown), and these potential agonists were further refined to a list of 37 candidate peptides based on predicted binding affinity for HLA-A*02:01. These candidate peptides included 11 nonamer sequences derived from common human viruses and 26 nonamer sequences derived from self-proteins (Table I). Three modified versions of LLSYFGTPT were also included for downstream testing as representatives of substitutions predicted to alter the TCR interaction (4Y > 4W), substitutions predicted to have no effect (4Y > 4F), or substitutions predicted to improve peptide binding to HLA-A*02:01 (9T > 9V) (Table I).

In functional assays, seven of the natural peptides activated clone 6C5 more potently than LLSY(3-tBu)FGTPT, and all of these agonists were derived from self-proteins (Fig. 3). None of the variant peptides with substitutions at P4 (LLSWFGTPT and LLSFFGTPT) or P9 (LLSYFGTPV) were recognized in parallel assays (Fig. 3), although as predicted empirically, the 9T > 9V substitution enhanced binding to HLA-A*02:01 (Supplemental Fig. 1A).

Of note, two peptides that elicited functional responses in the initial screens (LTAYMLLAI and FMAYFGVSA) were subsequently found to be inert using high-purity (>95%) preparations (Supplemental Fig. 1B). Mass spectrometric analysis of the false-positive LTAYMLLAI peptide (initial purity 35%) further indicated artifactual tBu modification of the Y residue at P4 (data not shown). These findings suggested that tBu modification enabled the functional recognition of largely unrelated peptides via the 6C5 TCR.

It was also notable that peptide immunogenicity did not correlate with surface stabilization of HLA-A*02:01 (Supplemental Fig. 1C). For example, some peptides that bound strongly to HLA-A*02:01, such as ILAYVFPGV and SLGWVLTA (Supplemental Fig. 1A), failed to elicit a T cell response (Fig. 3). As a prerequisite for antigenicity, however, all seven of the agonist peptides that activated clone 6C5 more potently than LLSY(3-tBu)FGTPT (Fig. 3) were able to stabilize HLA-A*02:01 (Fig. 4A, Supplemental Fig. 1A).

To confirm our initial results, we tested a smaller panel of 14 peptides across a wider range of concentrations (Fig. 4B). This panel included peptides identified as more potent (e.g., LLGWVFAQV and MLAWIFLPI), equivalently potent (e.g., ILGWIFVPI and FLAYFLVSI), or less potent (e.g., YMAYMFLTL) than LLSY(3-tBu)FGTPT. In line with the initial screens, we found that seven of these peptides, including the predicted optimal agonist (IIGWMWIPV), activated clone 6C5 to a greater extent than LLSY(3-tBu)FGTPT. The six most potent peptides had EC₅₀ values up to 600-fold higher than LLSY(3-tBu)FGTPT based on the production of MIP-1β (Supplemental Fig. 2).

In further experiments, we used flow cytometry to measure other effector functions elicited by all agonist peptides in the confirmatory panel (n = 12) (Fig. 4C). Similar response patterns were observed for the seven most potent agonists, with minor variations in the production of IL-2 and TNF-α (Fig. 4C). The predicted optimal agonist (IIGWMWIPV) elicited the strongest responses (Fig. 4C). In contrast, peptides that elicited suboptimal levels of MIP-1β, namely LLGWMLSQV, YVAWFLVFA, and ILGWIFVPI, largely failed to induce degranulation or the production of IFN-γ, IL-2, or TNF-α (Fig. 4C).

Collectively, these experiments identified seven agonist ligands that activated clone 6C5 more potently than LLSY(3-tBu)FGTPT. Shared features of these ligands included an A or a G residue at P3, a W residue at P4, and a phenylalanine (F) residue at P6. Otherwise, these agonist peptides were largely dissimilar, with minimal linear sequence homology to LLSY(3-tBu)FGTPT.

In an attempt to understand how ligands with disparate linear sequences could trigger the same TCR, we solved the binary structures of LLSY(3-tBu)FGTPT and two more potent agonists, namely the predicted optimal peptide (IIGWMWIPV) and a self-peptide derived from tissue factor precursor (LLGWVFAQV), in complex with HLA-A*02:01.

All three peptides adopted a typical bulged-out conformation, and two conformations were observed for each peptide–HLA-A*02:01 complex (Fig. 5A–C). The maximum resolution values were 2.01 Å for IIGWMWIPV–HLA-A*02:01, 2.70 Å for LLGWVFAQV–HLA-A*02:01, and 2.70 Å for LLSY(3-tBu)FGTPT–HLA-A*02:01, and the corresponding R-work/R-free values were 0.176/0.216 for IIGWMWIPV–HLA-A*02:01, 0.20/0.28 for LLGWVFAQV–HLA-A*02:01, and 0.186/0.241 for LLSY(3-tBu)FGTPT–HLA-A*02:01 (Supplemental Table I). The potential effects of crystal packing on the corresponding peptide conformers are detailed in Supplemental Fig. 3A–C, and pairwise comparisons between each copy of each structure are shown in Supplemental Fig. 3D.

The main-chain trace of all six peptide copies superimposed on each other in bound form revealed similar central conformations (Fig. 6A). In particular, the amino acid residues at P4 all pointed upward away from the peptide-binding groove of HLA-A*02:01, consistent with the earlier in silico model (14). The bulk of the tBu-modified Y residue appeared to be similar to that of the W residue at P4. A preference for flexible nonpolar side-chains, namely F, methionine (M), or valine, suggested that P5 was also a potential contact residue for the 6C5 TCR (Fig. 5A–C). In contrast, the orientation of residues at P6 was more consistent with a secondary anchor role, potentially stabilizing peptide binding to HLA-A*02:01 (Fig. 6B). Previous studies have shown that the P6 residue can associate with the P3 side-chain to support the central bulge of nonamer peptides bound to HLA-A*02:01 (21, 27). This support role could not be achieved by the short side-chain of serine in conjunction with G at P6 in the bound conformation of LLSY(3-tBu)FGTPT, allowing the central bulge to subside slightly in the binary structure compared with IIGWMWIPV–HLA-A*02:01 and LLGWVFAQV–HLA-A*02:01.

Although similar in terms of overall conformation, some differences were apparent between the two copies of each bound peptide, most notably with respect to the side-chains at P4, P5, and P6. The largest deviation was observed for IIGWMWIPV. In this structure, the W at P4, the M at P5, and the W at P6 adopted opposing conformations in complex with HLA-A*02:01 (Fig. 5B).

Collectively, these data suggested a common mode of immunogenicity involving the central bulged region of each peptide, likely focused on P4 and P5, and further indicated that the tBu-modified Y residue mimicked the bulky nature of W at P4 (Figs. 6C, 7).

In this study, we used functional assays, CPL screens, and x-ray crystallography to identify and characterize natural ligands for the orphan 6C5 TCR, which was previously shown to recognize the chemically modified Bax_161–170 peptide LLSY(3-tBu)FGTPT (14). We identified seven peptides that activated clone 6C5 more potently than LLSY(3-tBu)FGTPT. All of these peptides were derived from self-proteins and bound HLA-A*02:01. Structural analyses further revealed that the tBu-modified Y residue in the LLSY(3-tBu)FGTPT–HLA-A*02:01 complex mimicked the unmodified W residue in two more potent agonist complexes, namely IIGWMWIPV–HLA-A*02:01 and LLGWVFAQV–HLA-A*02:01.

Multiple rounds of in vitro stimulation with a peptide mixture containing LLSY(3-tBu)FGTPT were required to generate clone 6C5 (14, 30). It therefore seemed likely that the parent clonotype was relatively infrequent in the peripheral circulation and that recognition of the chemically modified LLSY(3-tBu)FGTPT epitope, which is not known to occur in nature, was a consequence of cross-reactivity at the level of the TCR (12). To identify the primary specificity of clone 6C5, we used functional screens in combination with a nonamer CPL (2, 16–18). This approach has been used previously with advanced bioinformatics to identify candidate peptide ligands for orphan TCRs (18, 31, 32).

The predicted optimal agonist ligand based on residue preferences across the peptide backbone was IIGWMWIPV. Although the corresponding synthetic peptide was a potent activator of clone 6C5, no direct match for this sequence was found in a database search of human viral pathogens and self-proteins (18). CPL-driven database searches nonetheless identified several candidate ligands with peptide sequences derived mostly from intracellular proteins. All of these proteins were encoded by ubiquitously expressed genes with no known links to carcinogenesis. Moreover, peptides derived from some of these proteins, including sodium/myo-inositol cotransporter 2, vacuolar protein sorting-associated protein 8 homolog, and galactosylgalactosylxylosylprotein 3-β-glucuronosyltransferase, have been identified in previous studies of the global immunopeptidome, albeit not in the context of HLA-A*02:01 (33). It is also important to note that clone 6C5 did not respond functionally to CD40L-activated HLA-A*02:01⁺ CLL cells (14). Further work is therefore required to determine the extent to which these agonist peptides are naturally presented on the cell surface bound to HLA-A*02:01.

All of the candidate ligands were more potent agonists than the tBu-modified peptide, likely reflecting greater Ag density as a consequence of enhanced binding to HLA-A*02:01 and/or a better fit with the 6C5 TCR. The latter supposition requires further investigation to define the precise modes of engagement and the potential for flexible interactions that may favor the recognition of particular conformers identified in the binary state (34, 35). Despite some conserved features at the linear sequence level, these naturally occurring agonist peptides were largely dissimilar to LLSY(3-tBu)FGTPT, highlighting the value of unbiased ligand identification using CPLs (2). However, the binary structures of two of these agonist ligands (IIGWMWIPV and LLGWVFAQV) revealed conformational similarities with LLSY(3-tBu)FGTPT–HLA-A*02:01, providing initial insights into the specificity of the 6C5 TCR.

Our data showed that neither LLSY(3-Me)FGTPT nor LLSY(3,5-di-tBu)FGTPT were able to activate clone 6C5, suggesting that the size of the chemical modification was an important determinant of recognition via the TCR. Moreover, the unmodified peptide could not be converted into an agonist ligand via a 4Y > 4W substitution, despite a strong preference for W at P4 in the CPL screens and the similarity in size between W and 3-tBu-Y. This observation suggested that neighboring peptide residues were key determinants of activation. In line with these results, the structural data indicated that the 3-Me moiety would likely be too small and that the 3,5-di-tBu moiety would likely be too big to trigger a functional response, assuming a necessity for overall shape complementarity between P4 and the 6C5 TCR. Similarly, the LLSWFGTPT peptide likely failed to activate clone 6C5 as a consequence of the F residue at P5, which presumably breached the clearance needed after the W side-chain to make appropriate contacts with the TCR. In contrast, the LLGWVFAQV peptide was likely a potent activator of clone 6C5 as a consequence of the valine residue at P5, the side-chain of which tucked in underneath the central peptide bulge toward the TCR, whereas the IIGWMWIPV peptide was likely a potent activator of clone 6C5 as a consequence of the M residue at P5, the flexible side-chain of which was free to swivel out of the way, preserving the central peptide bulge toward the TCR (21).

TCR degeneracy is thought to be essential for comprehensive immune coverage (36, 37) and likely underlies the phenomena of alloreactivity and autoimmunity (38, 39) via mechanisms that include conformational adaptability (35, 40–42), “hotspot” recognition (43–47), and molecular mimicry (48, 49). Similar off-target effects can also complicate adoptive gene transfer, especially with high-affinity TCRs (50, 51). We have extended these paradigms by demonstrating that synthetic peptides with inadvertent chemical modifications can mimic endogenous ligands and activate potentially autoreactive T cells. Although the original isolation of clone 6C5 was serendipitous, our results will likely have important implications for peptide-based immunotherapies and vaccines, specifically emphasizing the need for rigorous manufacturing processes and quality controls to mitigate the risk of adverse effects in vivo (52).

We thank the staff at the Diamond Light Source for assistance with data collection.

D.K.C. is currently an employee of Immunocore. R.R. is currently an employee of Kite Pharma. The other authors have no financial conflicts of interest.