Broadening alloselectivity of T cell receptors by structure guided engineering | Scientific Reports

Nov 06, 2024

Scientific Reports volume 14, Article number: 26851 (2024) Cite this article

Metrics details

Specificity of a T cell receptor (TCR) is determined by the combination of its interactions to the peptide and human leukocyte antigen (HLA). TCR-based therapeutic molecules have to date targeted a single peptide in the context of a single HLA allele. Some peptides are presented on multiple HLA alleles, and by engineering TCRs for specific recognition of more than one allele, there is potential to expand the targetable patient population. Here, as a proof of concept, we studied two TCRs, S2 and S8, binding to the PRAME peptide antigen (ELFSYLIEK) presented by HLA alleles HLA-A*03:01 and HLA-A*11:01. By structure-guided affinity maturation targeting a specific residue on the HLA surface, we show that the affinity of the TCR can be modulated for different alleles. Using a combination of affinity maturation and functional T cell assay, we demonstrate that an engineered TCR can target the same peptide on two different HLA alleles with similar affinity and potency. This work highlights the importance of engineering alloselectivity for designing TCR based therapeutics suitable for differing global populations.

T cells play a vital role in stimulating immune responses to foreign antigens during infections and to self-antigens for cancer surveillance1,2. In some scenarios, T cells recognise and respond to self-antigens presented on healthy cells leading to autoimmune diseases3,4,5. The class I major histocompatibility complex (MHC) presents peptide antigens on the surface of antigen presenting cells (APCs), which are then recognised by T cell receptors (TCRs). In humans, peptides are presented in the context of a polymorphic human leukocyte antigen (HLA) (over 35,000 alleles), which results in differing severity of pathogenic infections among diverse populations6.

Immunotherapies targeting cell-surface peptide-HLA (pHLA) provides a route for biologics to target intracellular antigens7,8,9,10,11. However, the HLA locus is highly polymorphic and varies significantly across different ethnic groups. Therapeutic TCRs typically engage a peptide in the context of a single HLA allotype restricting responses to a specific patient population. To date the development of TCR therapies has focused on antigens presented on the most prevalent HLA alleles, particularly on HLA-A*02:01. Recently, tebentafusp became the first TCR based molecule targeting pHLA that was approved by FDA for treating patients with HLA-A*02:01 positive metastatic uveal melanoma12,13.

One way to expand the potential patient population for a TCR therapeutic is to break HLA-restriction by specifically employing TCR subsets that target antigens presented by non-polymorphic MHC molecules like MR1, CD1 and HLA-E14. An alternate approach is to engineer TCR therapeutics that can engage a peptide antigen presented on more than one HLA allele. Overcoming the single allotype HLA-restriction of a TCR is challenging for two reasons: Firstly, each HLA allele presents a distinct peptide repertoire due to polymorphism in the HLA antigen binding groove, and secondly, around two-thirds of the TCR binding interface comprises the solvent exposed HLA surface. This makes it especially challenging to isolate TCRs with the desired peptide and alloselectivity, and several reports of alloreactive TCRs engaging more than one peptide have been documented15,16,17,18.

Targeting the same peptide on more than one allele is most tractable by pursuing alleles within an HLA superfamily where polymorphisms are minimised in both the TCR accessible HLA surface and the peptide binding groove (fig. S1). One such superfamily is HLA-A3 which includes subtypes HLA-A*03:01, HLA-A*11:01, HLA-A*31:01, HLA-A*33:01 and HLA-A*68:01, members of which are reported to present overlapping peptide repertoires19,20,21. Targeting peptides presented by both HLA-A*03:01 and HLA-A*11:01 would enable a single TCR to potentially address a large, racially diverse population as HLA-A*03:01 is prevalent in western populations (22% and 25% of the population in the US and EU5, respectively) whereas HLA-A*11:01 is prevalent in Asian populations such as China (47%) and Japan (16%).

Here, as a proof of concept, we looked at a peptide derived from the widely studied PRAME (preferentially expressed antigen in melanoma) antigen presented in the context of HLA-A*03:01 and HLA-A*11:01 alleles22. PRAME is a cancer-testis antigen that is homogeneously expressed in a wide range of tumour types but shows minimal expression in normal adult tissues, except for testis and placenta23,24,25. To date, TCR targeting of PRAME has focussed on peptides presented on the HLA-A*02:01 allele, which is most prevalent in Western populations, with clinical studies in multiple cancer indications employing a variety of drug modalities, including T cells with engineered TCR (NCT02743611, NCT03686124, NCT03503968)26 and soluble bispecifics (NCT04262466, NCT05958121)27,28.

In this study, we used immunopeptidomics to validate the dual presentation of a PRAME derived peptide on multiple alleles in the HLA-A3 superfamily. Two TCRs were isolated against the peptide on HLA-A*03:01, and by targeting a single residue on the HLA surface we engineered a TCR to recognise the same peptide on HLA-A*11:01 with similar affinity and potency. The ability to target peptides presented on multiple HLA alleles may expand the potential of TCR-based therapies to address a greater and more diverse patient population.

Using the pan-HLA Class I specific antibody W6/32, we isolated HLA from two PRAME-expressing HLA-A*03:01 cell lines (MEL526, MEL624) and one PRAME-positive HLA-A*11:01 cancer cell line (SKMEL28) (table S1). The same PRAME derived peptide, ELFSYLIEK (residues 190–198), was identified by mass spectrometry in all three cell lines (Fig. 1A). We also identified this peptide in PRAME-positive ovarian cancer tissue heterozygous for HLA-A*03:01 (Fig. 1B). This peptide has been reported previously as an HLA-A*03:01 restricted epitope29. The peptide was estimated to bind to HLA-A*03:01 and HLA-A*11:01 with predicted affinities of 224 nM and 56 nM (netMHCpan 4.130), respectively. Stable pHLA complexes were refolded for both HLA-A*03:01 and HLA-A*11:01 in presence of the peptide (hereafter referred as A3-ELF and A11-ELF). We assessed the stability of the refolded pHLA complexes by measuring binding to high affinity variant of soluble ILT2 (ILT2-c50) using bio-layer interferometry31. ILT2 binds at the interface of HLA heavy chain and β2m and has been previously used to assess intact pHLA complex over time32. The half-life of pHLAs at 25 °C were determined as 5.2 and 4.6 h for A3-ELF and A11-ELF, respectively (fig. S2).

Identification of PRAME peptide antigen by Mass spectrometry. (A) Mirror plot comparing the mass spectrum obtained for ELFSYLIEK native peptide (blue) with the stable isotope labelled spiked in ELFSYLIEK* peptide standard (red) in the SKMEL28 cancer cell line (HLA-A*11:01). (B) Mirror plot comparing the mass spectrum obtained for ELFSYLIEK native peptide (blue) with the stable isotope labelled spiked in ELFSYLIEK* peptide standard (red) in ovarian cancer tissue (HLA-A*03:01; HLA-A*68:01).

We used phage libraries encoding TCR sequences derived from healthy donors’ PBMCs to isolate TCRs against A3-ELF. To guide peptide selectivity for the target antigen, the libraries were counter-selected against streptavidin beads presenting a pool of twenty ubiquitously presented HLA-A*03:01 self-peptides. Since our TCR discovery approach deliberately ignored alloselectivity in favour of peptide contacts, we wanted to assess whether a simple engineering strategy informed by sequence and structural differences between the HLA-A*03:01 and HLA-A*11:01 could affinity enhance recognition of both HLA alleles simultaneously, whilst narrowing affinity differences between them.

Two TCRs, hereafter referred as S8WT and S2WT, were identified after three rounds of panning. As anticipated by the panning approach, both TCRs preferentially bound to A3-ELF. The S8WT TCR bound to A3-ELF and A11-ELF with affinities of 44 µM and 286 µM respectively (Table 1), whereas S2WT bound to A3-ELF with affinity of 21 µM and 72 µM for A11-ELF (Table 2).

HLA-A*03:01 and HLA-A*11:01 share high sequence identity with only seven polymorphic residues (fig. S1B). These polymorphisms can be classified based on their position on the HLA surface. Three of these residues (F/Y9, L/Q156 and E/A152) are in the peptide binding groove and determine peptide selectivity and pHLA stability. Two of the residues, A/D90 and S/P105, reside in loops at the outer edges of the HLA surface and likely have no impact on either peptide or TCR binding19,33. However, D/E161 and T/R163 are on the solvent exposed surface of the HLA α2 helix and have the potential to directly impact TCR binding (fig. S1C,D).

To aid the engineering of alloselectivity in S8WT TCR and to decipher interactions with the HLA polymorphic residues D/E161 and T/R163, we initiated structure determination in complex with A3-ELF and A11-ELF. Crystals were obtained only for the S8WT-A3-ELF complex and the structure was determined to 1.81 Å resolution (Table 3).

The structure revealed that S8WT TCR binds canonically to A3-ELF with a docking angle of 62° and its CDRs spread evenly across both sides of the pHLA (Fig. 2A,B). S8WT engages with a patch of peptide residues (S4-E8) with contacts coming from CDR1α (Y32α), CDR3α (G97α, S98α and Q99α), CDR2β (Y51β) and CDR3β (G97β, G98β, H100β and N101β) including polar contacts to peptide S4, Y5 and E8 side chains (Fig. 2C). The S8WT footprint on HLA helix 1 came from CDR3α and CDR1-3β, burying a surface area of 437 Å2, while contacts to HLA helix 2 came from CDR1-3α and CDR3β, burying a surface area of 245 Å2 (fig. S3).

The S8WT-A3-ELF complex showed that CDR3α was aligned closest to the solvent exposed HLA polymorphic residues D161 and T163 but made no direct contact to either of these residues. The orientation of the CDR3α loop meant it would be difficult to engineer contact to D161 as it was furthest away, but HLA T163 was within reach (Fig. 2D) and could potentially act as HLA hotspot to narrow affinity differences between A3-ELF and A11-ELF.

To confirm the direct role of TCR contacts to HLA T/R163 in modulating alloselectivity and rule out indirect impact of the polymorphisms that modulate the peptide interactions with HLA, we prepared the following single and triple mutants of A3-ELF and A11-ELF: A3T163R-ELF; A11R163T-ELF; A3F9Y/E152A/L156Q-ELF (hereafter referred as A33M-ELF) and A11Y9F/A152E/Q156L-ELF (hereafter referred as A113M-ELF). The residues selected for the triple mutants align along the peptide binding groove and cannot directly engage with the TCR.

S8WT showed a ~ 6-fold higher affinity for A3-ELF over A11-ELF (Table 1). Using the HLA mutants, we observed that S8WT affinity for A3T163R-ELF decreased by ~ 2.7 fold and affinity for A11R163T-ELF increased by ~ 2.5 fold, compared to A3-ELF and A11-ELF, respectively. This again showed the critical role of position 163 in regulating alloselectivity between the alleles. From the structure of S8WT-A3-ELF we observed that the HLA T163 was likely to interact with nearby CDR3α residues. Modelling suggested that the longer side chain of CDR3α R96α (AVRQRGSQGNLI) was driving alloselectivity by clashes with HLA R163. To test this, we generated a phage library encoding diversity in the CDR3α. The position of CDR3α indicated mutations in this loop could enhance affinity to both A3-ELF and A11-ELF whilst narrowing the selectivity window such that there is a limited affinity difference.

Structural analysis of S8 TCRs. (A) Overall view of the S8WT-A3-ELF complex. The peptide is shown as sticks (pink) and HLA (orange), β2m (brown), TCRα (cyan) and TCRβ (green) are shown as cartoons. (B) Top view showing the position of CDRs (cartoon tubes) on the pHLA surface. (C) Peptide interactions from S8WT TCR. CDR residues that are within 4 Å from the peptide are shown as sticks. The dotted lines indicate polar contacts. (D) S8WT CDR3α residues that are near HLA T163 in the S8WT-A3-ELF complex are highlighted. (E) The HLA-A*11:01 residues (orange) that are within 4 Å radius of CDR3α W96α in the S89F3-A11-ELF complex are shown as sticks. The mutated residues in S89F3 are shown in red. Images were generated using Pymol v2.6.0.

Phage panning results identified CDR3α variants where position R96α was changed to a tryptophan. As the side chain conformation of tryptophan can potentially reach HLA T163, a clone containing the tryptophan mutation was selected for further analysis. This clone (hereafter referred to as S89F3) had an additional Q99Lα mutation further down the CDR3α. It is noteworthy that the Q99Lα mutation in CDR3α was near the peptide and away from HLA T163.

S89F3 exhibited similar affinities for A3-ELF and A11-ELF (7.5 µM and 14.5 µM respectively). This narrowing of the affinity window for S89F3 can be attributed to contacts with R163 as we observed a KD of 4.2 µM for A3T163R-ELF, which is ~ 1.7 fold higher than for A3-ELF, and a KD of 35 µM for A11R163T-ELF, which is ~ 2.4-fold lower than A11-ELF.

To understand the molecular details of increased preference for A11-ELF by S89F3, we solved the structure of S89F3-A11-ELF to a resolution of 2.49 Å. Overlay of S8WT-A3-ELF and S89F3-A11-ELF complexes indicated that W96α in S89F3 occupied a similar backbone position as R96α in S8WT without any significant change in the CDR3α loop conformation. The W96α side chain conformation varied between two molecules in the crystallographic asymmetric unit. In one molecule, the W96α side chain conformation was parallel to the HLA surface and in the other it was a mix of parallel and perpendicular conformations. The W96α side chain made strong van der Waals contacts with HLA R163 in both conformations, and additionally with HLA A158 and Y159 in the perpendicular conformation (Fig. 2E). The interactions observed between W96α, and HLA R163 in A11-ELF was not feasible with HLA T163 side chain in A3-ELF. The second mutation (Q99Lα) in S89F3 was wedged between peptide S4 and L6 and surrounded by HLA N66, A69 and R65. This side chain orientation was similar to Q99α in S8WT-A3-ELF, but crucially L99α made hydrophobic contacts with peptide L6 which could potentially impact S89F3 peptide specificity (Fig. 2E).

Based on the structure, we predicted that W96α in S89F3 was likely the main driver for the reduced affinity gap between A3-ELF and A11-ELF. To ascertain this, we prepared an S8 single mutant R96Wα (S8W96), and tested affinities for A3-ELF, A11-ELF, A3T163R-ELF, A11R163T-ELF, A33M-ELF, and A113M-ELF (fig. S4). S8W96 affinities for A3-ELF and A11-ELF were 52 µM and 89 µM respectively, which revealed a slightly lowered affinity for A3-ELF and ~ 3-fold higher affinity for A11-ELF, when compared with S8WT. Similarly, S8W96 exhibited ~ 3.5-fold higher affinity for A3T163R-ELF and slightly lower affinity for A11R163T-ELF, when compared with S8WT. These results unambiguously indicated CDR3α W96α was responsible for driving the HLA R163 preference.

To assess the impact of engineering and ensure the engineered TCRs have the requisite specificity we sought to characterise the peptide repertoire engaged by the S8WT and S89F3 TCRs. For this purpose, we used single chain peptide-HLA (scHLA) phage libraries. A nine amino acid peptide library was generated with diversity at the DNA level of approximately 10 × 1010 variants. These variants were cloned into phagemids encoding either HLA-A*11:01 or HLA*A3:01 in a widely adopted single chain trimer format (Fig. 3A). Each of the pHLA libraries were expanded separately, quantified by spectroscopy, and mixed at equimolar concentrations.

Following two rounds of panning with the S8WT or S89F3 TCR, pHLAs were sequenced by Illumina Miseq. For each TCR, the 100 most enriched peptides were clustered based on amino acid similarity (Fig. 3B). The S8WT enriched 76/100 peptides on HLA-A03:01 which accounted for 93% of peptides by frequency. The largest cluster, with the consensus sequence RIFEHKAEV was the most consistent with the target ELFSYLIEK peptide. An aromatic residue at the third peptide position and Glu at position 8 appear to be strict requirements for recognition by the S8WT TCR, however the central residues were dissimilar suggesting that the initial TCR isolated may not have optimal specificity for the target PRAME peptide.

Reflecting the reduction in affinity difference between the two alleles, the 100 most enriched peptides to the allo-engineered affinity enhanced variant S89F3 TCR were more evenly shared between the two alleles 45 peptides (65% frequency) and 55 peptides (45% frequency) on HLA-A03:01 and HLA A11:01 respectively (Fig. 3C). S89F3 incorporates R96Wα and Q99Lα, both of which increased hydrophobicity of the TCR paratope. The peptides were grouped into three distinct clusters all of which showed marked increase in hydrophobic residues at peptide positions 4–7. The cluster most consistent with the target ELFSYLIEK peptide had the consensus sequence SQGYWLVEK which excluded charged residues Asp, Glu, Arg, Lys at positions 4–7 whilst retaining the selectivity for an aromatic residue (Phe/Tyr/Trp/His) at position 3 and Glu at position 8.

scHLA panning for S8WTand S89F3. (A) Peptide-HLA phage libraries (HLA-A*03:01; HLA-A*11:01) were generated using a disulphide stabilised single chain trimer format and combined at equimolar phage titre. (B, C) Peptide specificity profiles were generated from the 100 most enriched peptides clustered based on amino acid similarity for S8WT and S89F3 respectively. For each cluster the relative number of peptides from each HLA allele are shown with their summed frequencies determined by NGS shown in brackets. Amino acids are coloured according to side chain properties: red (negative), blue (positive), black (hydrophobic), green (aromatic), purple (polar) and orange (small non-polar). The target peptide is shown on top of sequence logo for comparison.

To ascertain whether broadly equivalent affinity and selectivity of S89F3 for A3-ELF and A11-ELF were also reflected functionally, we performed a T cell activation assay with S8WT or S89F3 transduced into a TCR-null NFAT-luciferase engineered Jurkat clone J1601 (fig. S5,S6). The CD3+ population was FACS-sorted and tested for stimulation using an artificial antigen presenting cell (aAPC) bead system (Fig. 4A). The beads were coated with either A3-ELF or A11-ELF and anti-CD28 antibody for functional activation of respective TCR-bearing Jurkats. For S8WT, the activation of T cells, monitored through bioluminescence, showed higher activity for A3-ELF than for A11-ELF and aligned with the observed 6.5-fold difference in affinity (Fig. 4B,C). However, for the affinity matured and alloselectivity engineered S89F3 TCR, the activation levels against A11-ELF were elevated to those seen for A3-ELF. This is consistent with the broadly equivalent affinity of the TCR for the two alleles. These results clearly show that the affinity of S8 TCR can be modulated enabling similar T cell activation profiles on A3-ELF and A11-ELF.

Cellular assay using aAPCs and Jurkat cells expressing S8 TCRs: (A) Schematic representation of the aAPC co-culture. Streptavidin coated magnetic beads were incubated with biotinylated pHLA and biotinylated anti-CD28 antibodies to mimic cellular stimulation. (B) NFAT-Luciferase TCR-null Jurkat cells transduced with S8WT or S89F3 TCR were cocultured with aAPC beads. T cell activation was assessed using NFAT-luciferase readout. aAPCs coated with an irrelevant pHLA (IRR) and positive phytohemagglutinin (PHA) stimulation were used as controls. Mean of three independent experiments (± SEM) is shown. (C) Statistical analysis of the area under the curve using a paired t-test with Wilcoxon matched pairs signed rank test.

To extend the proof of concept, we aimed to replicate the alloselective engineering of a second TCR (S2). To define the TCR paratope we determined the structures of S2WT-A3-ELF and S2WT-A11-ELF complexes to resolutions of 2.34 Å and 1.97 Å, respectively (Table 3). The structure of S2WT-A3-ELF complex revealed that S2WT binds in a typical diagonal orientation with a docking angle of 50° and the alignment slightly tilted and shifted towards HLA α1 helix (Fig. 5A,B). The S2WT engages the central peptide residues, S4 to E8, dominated by polar contacts to Y5 and E8 side chains and the L6 backbone (Fig. 5C). Multiple CDR loops of the S2WT TCR form an extensive interface with HLA helix 1, whereas there are few contacts with HLA helix 2 burying a surface area of 455 Å2 and 73 Å2 respectively (fig. S7A,B).

The S2WT-A11-ELF structure revealed that the S2WT binds to A11-ELF in a similar orientation to A3-ELF (Fig. 5A). Superposition of the two complexes showed very high structural similarity with an all-atom RMSD of 0.974 Å. The peptide presentation and conformations of CDRs were also highly similar (Fig. 5B). Many of the peptide interactions from S2WT to A3-ELF and A11-ELF are the same (Fig. 5C). Minor differences were observed between HLA and ELF peptide interactions in the two complexes, notably from the HLA polymorphic residues (fig. S8).

An overlay of S2WT-A3-ELF and S8WT-A3-ELF complexes shows that the different binding orientations resulted in dissimilar positioning of CDRs, leading to varied footprints on the two HLA helices (fig. S7C,D). The S8WT footprint on the HLA helices was more balanced compared to the S2WT-A3-ELF complex, as the tilt and shift towards HLA helix 1 observed in the latter was less pronounced in the S8WT-A3-ELF complex.

Structural analysis of S2 TCRs. (A) Cartoon overlay of S2WT TCR bound to A3-ELF and A11-ELF. The peptides are shown as sticks (pink) and HLA (orange), β2m (brown), TCRα (grey) and TCRβ (blue) are shown as cartoons. The chains in S2WT-A11-ELF complex are shown in darker shades. (B) Overlay as in A with top view showing the position of CDRs (cartoon tubes) on the pHLA surfaces. (C) Peptide interactions from S2WT TCR from overlay as in A. CDR residues that are within 4 Å from the peptide are highlighted. The dotted lines indicate polar contacts. (D, E) S2WT CDR1α residues that are near HLA residue 163 (T163 in S2WT-A3-ELF and R163 in S2WT-A11-ELF) are shown as sticks. (E) Overlay of S2WT-A3-ELF and S2198-A3-ELF complexes showing HLA residues near S2198 W30α. The mutated residues in S2198 are shown in yellow. The arrows indicate the different CDR1α conformations. Images were generated using Pymol v2.6.0.

The S2WT TCR showed ~ 3-fold higher affinity for A3-ELF over A11-ELF. Comparative analysis of complex structures revealed that the only polymorphic HLA residue accessed directly by the S2WT TCR is T/R163 (Fig. 5D,E). In S2WT-A11-ELF, HLA R163 forms a salt bridge with peptide E1 and a hydrogen bond to S29α in the CDR1α (DSAIYN) (Fig. 5E). Whereas in S2WT-A3-ELF, the shorter HLA T163 sidechain makes H-bonds with peptide E1 but does not reach CDR1α (Fig. 5D). The positioning of CDR1α indicated mutations at the tip would increase contacts to HLA T163 (A3-ELF). Using a similar strategy adopted for S8WT, we aimed to identify CDR mutations that target HLA position 163 to narrow the affinity gap between A3-ELF and A11-ELF and broadening selectivity for the two alleles. To engineer this contact point, we focused on mutagenesis of CDR1α residues and subjected them to a single round of affinity maturation using phage display.

We designed an NNK library that allowed all 6 residues of the germline encoded CDR1α (DSAIYN) to be modified and panned against A3-ELF. Counter selection with A11-ELF was not performed as we did not wish to exclude mutations that enhance affinity to both variants. For the CDR1α library output, all variants identified involved mutations to position 29 in CDR1α and the flanking residues on either side (DSAIYN). Around 79% of identified clones had either ‘LKW’ or ‘LRW’ (hereafter referred as S2198) as the first three CDR1α residues.

The S2198 affinities for A3-ELF/A11-ELF were 76 nM and 1227 nM, respectively (Table 2). This represents an affinity enhancement of > 280-fold for A3-ELF and > 59-fold for A11-ELF when compared with S2WT. However, by focussing on contacts localised to the HLA polymorphic residue 163, selectivity window between A3-ELF and A11-ELF was increased up to ~ 16-fold and not narrowed as we originally intended.

To understand the molecular details of increased preference for A3-ELF by S2198, we solved the complex structure to a resolution of 2.05 Å. While most contacts remained unaltered as expected, the introduced mutations significantly altered the conformation of the CDR1α. This rearrangement resulted in W30α being the closest residue to HLA helix2 as compared to S29α in S2WT (Fig. 5F). Specifically, W30α engaged HLA via an extensive interface and a H-bond to the HLA T163. Crucially, S2198 affinity of 479 nM for A3T163R-ELF was 7.3 times lower than for A3-ELF (Table 2). Mutating the three other differences in the HLA binding groove to HLA-A*11:01 residues had almost no impact on the binding (fig. S9). This indicates that T163R mutation is potentially introducing clashes and driving the HLA alloselectivity in this engineered variant. As the narrowed affinity window was not attained, no further experiment on this TCR was carried out.

TCR based therapeutics typically target a single peptide antigen in the context of a single HLA allele27,34,35. Efficacy is dependent on expression of both the antigen and HLA allele which limits the patient cohort eligible for a potential treatment. This may compound disparities as many TCR drug development efforts are directed toward HLA-A*02:01 which is highly prevalent in Western populations36.

Alloreactivity in TCRs which lack peptide specificity and engage different peptides on multiple alleles is seen as an unwanted consequence of thymic selection, where negative selection to self-pHLA is restricted to peptides in the context of HLA alleles in that individual. To our knowledge this is the first example of a TCR isolated in the context of a single HLA-allele that has been engineered to engage the same peptide on an alternate HLA-allele while retaining reactivity with the original pHLA.

Wild type TCRs recognise pHLA with affinity in the µM range. For developing potent soluble bispecifics, TCR affinity engineering whilst maintaining high specificity is required to target cells which typically display tens to hundreds of peptide copies on the surface37,38,39.

Our primary consideration was to introduce affinity enhancing mutations towards the HLA, whilst maintaining a broad energetic footprint and retaining peptide specificity40. When trying to introduce an additional selectivity parameter we had two options, either (1) isolate TCRs which already had the requisite peptide and HLA allele selectivity, or (2) engineer either peptide or HLA selectivity into TCRs isolated against a single pHLA allele. One limitation of incorporating alloselectivity into an initial TCR discovery screen is that the global binding footprint may be shaped by the HLA mismatches compromising peptide specificity. During isolation of TCRs the binding modes are most diverse. We observe that adding additional allele selectivity requirements at the TCR discovery stage limits the successful identification of peptide specific TCRs (not shown).

Intuitively it is more challenging to engineer peptide selectivity than to engineer contacts to another HLA allele. This is because typically the self-peptides that drive off-target recognition of a given TCR are not known and can be highly divergent in sequence, whereas HLA alleles with overlapping peptide repertoires belong to the same HLA-superfamily. Recently, it has been shown that alloselective TCR mimic antibodies could be isolated from libraries encoding natural antibody diversity41. The authors were able to isolate an antibody that targets a peptide discovered on HLA-A*24:02, that was able to recognise the same peptide presented by the closely related allotype HLA-A*23:01. Initially the antibody showed some cross-reactivity to an off-target peptide mimetic and the requisite peptide specificity was introduced through engineering41,42.

In the present study, we identified two TCRs, S8WT and S2WT, with typical affinity and potency for their target antigen and using crystal structures we were able to identify the TCR paratope engaging critical HLA polymorphisms. As HLA residue 163 was the only polymorphic position that can be directly engaged by S8WT and S2WT, we were able to target this position during affinity maturation and modulate the affinity window between A3-ELF and A11-ELF. S89F3 and S8W96 exhibited a significantly narrowed affinity window between the two alleles when compared to S8WT (6.5-fold for S8 WT vs. 1.9-fold for S89F3). The structure of the S89F3-A11-ELF complex showed W96α made significant contacts to HLA helix 2 residues, especially to R163. Using chimeric HLA molecules, we were able to demonstrate that a single TCR point mutation was sufficient to fine tune alloselectivity such that both alleles are recognised with comparable affinities. Similarly, Papadaki et al.., used chimeric HLA class I molecules, incorporating novel peptide selectivity into distinct HLA allotypes as tool molecules to characterise the relative peptide and HLA selectivity of pHLA targeting molecules43. Further they suggested that these synthetic HLA molecules could be used as a single immunogen to isolate antibodies recognising a peptide presented by multiple HLA alleles43. It will be interesting to see whether combining both desired peptide and alloselectivity into a single immunogen can generate TCR mimic antibodies with the requisite specificity.

A substantial portion of peripheral T cells are thought to be autoreactive due to TCR engagement with self pHLA44,45. Affinity engineered TCRs risk enhancing recognition of self-peptides further which can lead to clinical toxicity46. Identifying and characterising responses to self-peptides forms an important part of an immunotherapeutic TCR safety profile47. In this study, we utilised a previously developed scHLA phage display approach to characterise the peptide landscape engaged by S8WT and S89F3 TCRs on both HLA-A*03:01 and HLA-A*11:018. As expected, the S8WT TCR preferentially enriched peptides on HLA-A*03:01 whereas the allo-engineered S89F3 variant enriched peptides from both alleles. Whilst the S8WT TCR enriched peptides that shared some features with the target ELF peptide, the central residues in the scHLA specificity profiles were chemically and structurally dissimilar. The S89F3 variant, with two point mutations in CDR3β, enhanced peptide selectivity by increasing the hydrophobicity of the TCR paratope. This was reflected in the peptide specificity profile of this variant which excluded charged residues from the central positions. Engineering peptide specificity typically involves optimising contacts at multiple interaction hotspots between the TCR CDRs and accessible peptide. The tolerated hotspot residue would ideally be restricted to the same amino acid as the target peptide. However, it is more typical that the tolerance is limited to amino acids which share structural and chemical similarity48. Further affinity and specificity enhancing mutations as well as extensive safety testing would be necessary before S8 TCR can be considered for use as a therapeutic targeting moiety.

In contrast to S89F3, S2198 increased the preferential binding for A3-ELF over A11-ELF even further (~ 3.4 fold for S2WT vs. ~ 16 fold for S2198). The structure of S2198-A3-ELF revealed rearrangement of the CDR1α loop conformation which provided additional contacts to HLA T163 and to peptide E1. The proximity of W30α in S2198 to HLA position 163 would introduce steric clashes with R163 in A11-ELF explaining the increased affinity gap between A3-ELF and A11-ELF.

It was striking that in both S8 and S2 mutants, a tryptophan binding to a similar HLA location played a vital role in HLA T/R163 selectivity but with opposite effects. CDR3α W96α in S89F3 exhibited increased preference for HLA R163 over T163 whereas CDR1α W30α in S2198 showed increased preference for HLA T163 over R163. It is possible that the positioning of the tryptophan residue with respect to HLA and the subtle differences in interactions with the HLA residues could have led to differing alloselectivity. The W96α conformation in the S89F3-A11-ELF complex was wedged between HLA R163 and A158 with the NE1 atom facing away from HLA position 163 whereas the W30α conformation in the S2198-A3-ELF complex was much closer to HLA T163 with a H-bond from the NE1 atom. These observations indicate HLA R163 was better accommodated by S89F3 than by S2198.

TCR-null Jurkat luciferase-reporter cells expressing S8WT and stimulated using bead-based artificial APCs, showed higher activation against A3-ELF as compared to A11-ELF while S89F3 TCR exhibited increased potency against the A11-ELF, as expected given the similar trend observed with affinity. This confirmed that findings from the binding experiments using soluble TCRs were translated in the functional assay using full length membrane bound TCRs on the surface of T cells.

Whilst the main determinant of peptide presentation will be the affinity of the peptide for that particular allele, which can typically be predicted in silico, a critical step in the study design was experimental validation that the target peptide was expressed on both HLA-A*03:01 and A*11:01 alleles. This is because several factors could affect peptide processing and HLA presentation and it is not possible to predict with accuracy whether a peptide will be presented. Additionally, whilst being able to equalise TCR affinity to a cross-presented peptide, several factors influence the relative target densities, and it remains likely that the same peptide will be presented at different levels in patients depending on the HLA allele. One of the main factors affecting target density is differences in HLA expression level which is modulated at both genetic (homozygous vs. heterozygous) and epigenetic (mRNA expression) level49. Other factors that could impact relative peptide density include pHLA stability and HLA dependence on tapasin50.

An additional factor that can impact the success of multi-allele approach is the peptide conformation. Differing conformation of cross-presented peptides will likely affect TCR recognition and binding kinetics. Comparison of peptide cross-presentation when bound to S2 or S8 TCR showed identical backbone conformation and highly similar side chain conformations (fig. S8, Fig. 2D-E). However, in the absence of A3-ELF and A11-ELF structures in the TCR unbound state, it is unclear whether the shared binding geometry is a prerequisite for or consequence of TCR binding. To assess the general applicability of this approach we identified other examples of peptide cross-presentation in the protein data bank (PDB) on multiple HLA class 1 alleles (table S2) and applied structural similarity based supertype classification51. For three different peptides (HIV RT313, HIV Nef and KRAS) we observed that the peptide conformation was broadly conserved, when presented on multiple alleles within the HLA A03 supertype (fig. S11). We observed that cross-presented peptides on closely related HLAs typically adopt similar geometry, however, this is dependent on the peptide. In the case of HLA-B*14:02 and B*27:05, where the peptide-binding groove has mismatches at 18 HLA positions, peptide RRRWRRLTV derived from LMP2 adopted a completely novel sidechain and backbone geometry whilst a second peptide IRAAPPPLF from pCatA adopted an almost identical conformation52.

Altogether, these results have potential implications for the development of TCR therapeutics (as bispecific or TCR-T) targeting two HLA alleles. A clear limitation for expanding this approach to other alleles is the need for similar HLA surface sequence, especially for those residues that are vital for TCR interactions. Also, variations in residues lining the peptide binding pocket in HLA can have implications for peptide stability and can indirectly impact TCR binding, kinetics, and potency. Conversely, high sequence dissimilarity among alleles, or alleles from different superfamily are unlikely to present the same peptide antigen. Nevertheless, this work presents a step forward for TCR engineering demonstrating a simple engineering strategy targeting positions mismatched between target HLAs is highly tractable. Developments in identification of disease specific antigens in the context of many varied HLA alleles and their structural presentation using experimental methods, combined with accurate predictions using deep learning tools, will guide future TCR (and antibody) engineering with potential to target two or more HLA allotypes.

Immortalized cancer cell lines MEL624 (NCI), MEL526 (NCI) and SKMEL28 (ICLC) were cultured according to suppliers’ instructions and lysed in buffer containing non-ionic detergent NP-40 (ThermoFisher) followed by centrifugation to remove cell debris. Flash frozen surgically resected primary tumour specimens were commercially obtained and milled under cryogenic conditions. Small portions of pulverized tissue were collected for HLA typing by NGS (Next Generation Sequencing) (15 to 20 mg). The bulk of the pulverized tumour material was lysed in buffer containing non-ionic detergent NP-40 followed by centrifugation to remove cell debris. HLA class I peptide complexes were enriched by affinity chromatography from lysed cell lines and tissue specimens by passing over resin containing anti-HLA-class I W6/32 antibody immobilized on a protein-A-Agarose scaffold. HLA-peptide complexes were eluted in 0.5% trifluoroacetic acid (TFA)/ 5% acetonitrile (ACN), desalted by reversed phase solid phase extraction (Sep-Pak C18, Waters), reduced in volume by vacuum centrifugation, and stored at -80° C until LC-MS/MS analysis.

HLA peptides were reconstituted in 0.1% TFA/5% ACN, separated by reversed phase liquid chromatography (LC) and analysed by tandem mass spectrometry (MS/MS). Briefly, purified pHLA complexes were loaded onto an Acclaim™ PepMap™ 100 trap column (100 μm x 2 cm, ThermoFisher); separated using an EASY-Spray™ column (75 μm x 50 cm, ThermoFisher) with mobile phase A (0.1% formic acid in water), and B (0.1% formic acid in ACN) at a flowrate of 250 nL/min; eluted into an EASY-Spray™ ionization source (ThermoFisher) and analysed on an Orbitrap Fusion™ Lumos™ Tribrid™ Mass Spectrometer (ThermoFisher). MS/MS data was acquired with data dependent analysis (DDA) with the following settings: full MS1 scan recorded at 120 K resolution (AGC 4E5, 50 ms) after quadrupole isolation (300–1200 m/z range); precursor ions selected for MS2 with a quadrupole isolation window of 1.6 Da (Orbitrap Fusion™) or 1.2 Da (Orbitrap Fusion™ Lumos™), fragmented by HCD with 28% collision energy, and analysed in the Orbitrap at 30 K resolution (AGC 2E4, 240 ms). Stable isotope labelled (SIL) peptides (JPT technologies, Berlin, Germany) were introduced into each sample at an exact molar amount of 100 femtomole immediately prior to MS analysis. MS/MS data was processed with PEAKS7.5 (Bioinformatics Solutions) and searched against the Uniprot human protein database with no enzyme restriction and a tolerance of 5 ppm for MS1 and 0.02 Da for MS2.

TCRs specific to A3-ELF were isolated from in-house naïve TCR phage display libraries11. Two TCRs (S2WT and S8WT) with confirmed binding to A3-ELF and A11-ELF by surface plasmon resonance were selected for affinity maturation. TCR affinity maturation was performed using phage display methodology53,54. For S2WT, a NNK library was designed that allowed all 6 residues of CDR1α to modify. For S8WT, we adopted the soft library approach such that all 12 residues in CDR3α (AVRQRGSQGNLI) were covered in a single library and had approximately 3% variability at each nucleotide position. TCR phage libraries underwent two rounds of selection using 100 nM A3-ELF immobilised on streptavidin beads. Prior to pan 2 the libraries were counter-selected against 200 nM pHLA A3 streptavidin beads presenting a pool of twenty validated HLA-A*03:01 peptides- KLYEQLSGK, VLYPSAQEK, VLYENPNLK, ALFLTLTTK, SVASPFTSK, ALANVSIEK, SLYKDAMQY, VVYAPLSKK, QLYWSHPRK, LLFANQTEK, QLYSALANK, SLSSPLNPK, ILGPMFSGK, KLLDPIREK, LLYEKNLVK, RLYQHAVEY, QLYKEQLAK, VLYDRVLKY, SLFSNVVTK, GLFGKTVPK.

The four mutants HLA-A*03:01 T163R, HLA-A*03:01 F9Y E152A L156Q, HLA-A*11:01 R163T and HLA-A*11:01 Y9F A152E Q156L were made by site directed mutagenesis of HLA-A*03:01 and HLA-A*11:01 using a method described previously55.

The wildtype HLA-A*03:01 and HLA*11:01 with and without C-terminal AVI-tag, the four HLA mutants with a C-terminal AVI-tag, and B2m were expressed as inclusion bodies in E. coli. The pHLA complexes were refolded by dilution as described previously56 with PRAME peptide (ELFSYLIEK) (Peptide Protein Research Ltd). Following refolding, pHLA complexes were dialysed into 20 mM Tris pH 8.1 and purified by anion exchange chromatography using POROS HQ resin (Applied Biosystems). The purified pHLA complexes with AVI-tags were biotinylated using BirA biotin-protein ligase bulk reaction kit (Avidity) as per manufacturer’s protocol. All pHLAs were further purified by size exclusion chromatography using a Superdex S75 Increase column (Cytiva).

The half-lives of pHLAs were measured at 25 °C by bio-layer interferometry (BLI) using an Octet96. Biotinylated pHLA was loaded onto streptavidin coated pins to a response of 1 nM and the maximal binding response of 5 µM ILT2 c5031 was measured every 30 min for four hours. Fresh pins were used for each repeat to generate the triplicate data. The half-lives were calculated using a non-linear fit of the percentage of binding relative to time using GraphPad Prism.

Recombinant expression and purification of untagged and AviTag™ S2 and S8 TCRs and their variants were carried out as explained previously2,57: Sequences are provided in the supplementary material. Briefly, alpha and beta chains of TCRs containing an interchain disulphide bond were cloned into pGMT7 vector and the protein chains were expressed separately using Rosetta DE3 E. coli cells grown in autoinduction media containing ampicillin. Inclusion bodies of TCRα and TCRβ were isolated, mixed (1.5:1 molar ratio) and denatured using buffer containing 50 mM Tris pH 8.1, 100 mM Sodium Chloride, 6 M Guanidine and 20 mM Dithiothreitol (DTT). The mixture was refolded by diluting to a final protein concentration of 60 mg/L into a buffer containing 100 mM Tris pH 8.1, 4 M Urea, 400 mM L-Arginine, 1.9 mM Cystamine and 6.5 mM Cysteamine redox couple. The refolded TCRs were then purified by anion exchange using Poros50 HQ column. Biotinylation of AviTag™ TCRs were carried out as explained before for pHLAs. All samples were further purified by gel filtration using Superdex 75 GL column equilibrated in PBS buffer.

Binding analyses were performed using a Biacore T200 equipped with a CM5 sensor chip as reported previously58. All experiments were performed at 25 °C and using PBS buffer supplemented with surfactant P20. The biotinylated pHLAs were coupled to the streptavidin coated until the response units reached ~ 1000. Biotin was used to block the remainder of the chip surface. The TCRs were then injected at six concentrations following a series of two-fold dilutions with a top concentration ~ 10x above expected KD (S2WT 150 − 4.69 µM- S8WT 250 − 7.8 µM) at 10 µ/min. The triplicate data, generated using repeat injections over the same surface, were processed using one site total least squares fit in GraphPad Prism 9.0.

S2 and S8 TCRs and were mixed with either A3-ELF or A11-ELF in equimolar ratio and concentrated to 8–10 mg/ml and buffer exchanged to 10 mM Tris pH 8.0, 20 mM NaCl. Sitting drops were set up containing 150 nl of protein solution and 150 nl of reservoir solution in MRC crystallisation plates using the Gryphon robot (ART Robbins) and incubated at 20 °C in Rock Imager 1000 (Formulatrix). Crystals were cryoprotected using reservoir solution containing 30% ethylene glycol and flash cooled in liquid N2. Diffraction data were collected at beamlines I04 and I04-1 at the Diamond Light Source, UK. Datasets used for structure solutions were collected from crystals grown in the following crystallisation conditions:

S2WT-A3-ELF: 50 mM HEPES pH 7.0, 13% w/v PEG 8000.

S2WT-A11-ELF: 0.1 M Amino acids (0.02 M Sodium Glutamate; 0.02 M Alanine; 0.02 M Glycine; 0.02 M Lysine hydrochloride; 0.02 M Serine), 0.1 M Buffer system 1 (Imidazole; MES monohydrate) pH 6.5, 50% Precipitant mix 2 (20% v/v Ethylene glycol; 10% w/v PEG 8000).

S2198-A3-ELF: 50 mM MOPS pH 6.7, 14% w/v PEG 8000.

S8WT-A3-ELF: 50 mM Magnesium acetate, 20 mM MOPS pH 7.2, 50 mM Sodium chloride, 12% w/v PEG 8000.

S89F3-A11-ELF: 150 mM Magnesium acetate, 20 mM MOPS pH 7.2, 12% w/v PEG 8000.

The diffraction data were integrated and scaled using the xia259 automated processing pipeline implementing XDS60 and XSCALE for S2WT-A3-ELF, S2WT-A11-ELF, S8WT-A3-ELF, and S89F3-A11-ELF datasets. The diffraction data was integrated and scaled using the xia259 automated processing pipeline implementing DIALS61 for S2198-A3-ELF dataset. All the structures were solved using molecular replacement using Phaser62 within the CCP4 suite63. For the S2WT, alpha chain model from PDB 5BRZ and beta chain model from PDB 5XOT were used as search models. For the S8WT, PDB 5BRZ was used as search model. For HLA-A*0301 and HLA-A*1101, PDB 3RL1 and PDB 1Q94 (without the peptides) were used as search models, respectively. The model was built using iterative cycles of manual model building in COOT64 and refinement using Refmac65. The stereochemical properties and validation of the models were assessed using PDB-REDO66 and molprobity67. Data collection and refinement statistics were given in Table 3. Buried surface area and TCR docking geometry statistics based on those described previously68 were generated using Molecular Operating Environment (Chemical Computing Group). The structural figures were generated using Pymol (Schrödinger).

Peptide HLA libraries were generated in a single chain format with peptide-β2m-HLA-A*03:01 or HLA-A*11:01 displayed on the surface of phage as disulfide trapped single chain trimers. Briefly, a randomized 9-mer peptide library consisting of 1 × 1010 peptide diversity was synthesized (Twist Biosciences) and cloned into a phagemid scHLA construct using a pelB leader sequence and C-terminal coat protein pIII9. This phagemid library was introduced by electroporation into E. coli TG1 cells with KM13 helper phage to enable monovalent display with an estimated library size of 6.6 × 109 colonies69. Diversity was confirmed post electroporation by next generation sequencing of the initial library. This confirmed that all 20 amino acids were represented at every position with a flat distribution of 5% (max 5.8%, min 4.5%).

Streptavidin-coated paramagnetic beads (Dynabeads™ M-280, Thermo Fisher Scientific) were saturated with biotinylated S8WT or S89F3 and phage selections were performed as described previously9. Phagemid DNA was isolated from glycerol stocks of panning outputs by miniprep (Qiagen). Region of interest encompassing a synthetic 9-mer peptide sequence and the HLA-A*03:01/-A*11:01 variable segment was amplified by 15 cycles of PCR. Dual indices and adapters for Illumina sequencing were added by second round of PCR using high fidelity DNA polymerase (NEBNext Ultra II Q5 Master Mix) and multiplex oligos (NEBNext Multiplex Oligos for Illumina (Dual Index Primers Set 1)). Final libraries were cleaned up and size selected using 0.7x volume of AMPure XP beads (Beckman Coulter). Libraries were pooled at equimolar ratios, spiked with 10% PhiX and paired end (PE) sequenced (2 × 300 bp) on MiSeq (MiSeq 600 cycles v3 kit, Illumina).

Following quality based trimming and filtering of raw PE reads using BBDuk with parameters qtrim = r, trimq = 20, minlength = 80 (BBMap), sequences coding for 9-mer peptides were extracted using Cutadapt70 from R1 reads and associated HLA allele sequences from R2 reads. For each unique combination of peptide and HLA allele, reads were tallied to obtain peptide/allele counts. For S8WT, following two rounds of panning 208,366 reads passed QC and were mapped to either HLA-A*03:01 or HLA-A*11:01, which encoded 6,393 and 4,007 unique peptides respectively. For the affinity enhanced variant S89F3, following two rounds of panning, 453,116 reads passed QC and were mapped to either HLA-A*03:01 or HLA-A*11:01, which encoded 23,598 and 16,192 unique peptides respectively.

The 100 most enriched peptides from each variant were clustered based on their amino acid similarity averaged across all positions. A distance matrix was generated based on pairwise distance and defined as the average of the BLOSUM scores when each position is compared between the two peptides and visualised using tSNE. For example, to compare a peptide 123,456,789 to another peptide abcdefghi, we take the average BLOSUM62 scores for the first position (1 vs. a = Z), second position (2 vs. b = C), and so on. A heatmap of the distance matrix was generated and then the values were used to run a tSNE clustering procedure. Cluster identification was performed using HDBscan71 and three clusters were identified for each TCR and reported as sequence logos.

To prevent mispairing between the introduced TCR α and β chain and the endogenous TCR α and β chain, endogenous TCR genes were inactivated in NFAT-Luciferase Jurkat cells (Promega) using CRISPR-Cas9 by knocking out both alpha and beta TCR constant chains. More specifically, cells were sequentially electroporated with Cas9 protein (Thermo Fisher) along with synthetic gRNAs (sgRNA) (Thermo Fisher) against alpha (AGAGTCTCTCAGCTGGTACA) and beta (GGAGAATGACGAGTGGACCC) constants chains. For optimal editing, 200,000 cells were electroporated with 240ng sgRNA and 1250ng Cas9 protein using Neon Transfection System (Thermo Fisher) according to manufacturer`s instructions.

Cells were screened for the absence of surface CD3 (BD Bioscience, Clone SKY7, BB515), TRBC1 (Novus Biologicals, Clone JOVI.1, PE) expression by flow cytometry. TCR null cell population were single cell sorted to achieve monoclonality. Absence of wild type sequences and the presence of indels were confirmed by sanger sequencing followed by sequence trace decomposition using CRISP-ID (http://crispid.gbiomed.kuleuven.be/) and TIDE (https://tide.nki.nl/).

15 million HEK-293T cells were transfected with 10 µg TCRαβ lentiviral expression construct, 12 µg pRSV.REV, 12 µg pMDLg and 12 µg pMDg.2 using PEI-Max (Polysciences) following manufacturer`s protocol. Lentiviral supernatant was collected at 72 h after transfection, spun down at 500 g for 10 min and clarified through a 0.45-µm syringe filter. Cleared supernatant was precipitated with Lenti-X concentrator (Takara-Clontech) according to manufacturer`s instructions and incubated overnight at 4°C. Samples were centrifuged next day at 1500 g for 3 h at 4°C. The pellet containing lenti-viral particles was gently resuspended in complete RPMI medium and added to Jurkat cells in RPMI medium containing 10% FBS and 12 µg/ml polybrene (Merck) Samples were centrifuged at 700 g for 2 h at 32°C. Three days after viral transduction, cells were analysed for cell surface CD3 to measure transduction efficiencies.

Flow cytometric analysis was performed on BD LSRFortessa X-20 Cell Analyzer. Surface staining for flow cytometry was performed pelleting cells at 400 g for 5 min and resuspending in 50 µl of FACS buffer (5% FBS in PBS) with antibodies and viability dye for 30 min at 4 °C in the dark. Cells were washed once in FACS buffer before resuspension.

To enrich CD3+ cells post transduction, 2–5 million cells were stained with viability dye (1/1000) (eFluor780) and with anti-human CD3 antibody (1/100), (BD Bioscience, Clone SKY7, BB515) or TRBC1 (1/100) (Novus Biologicals, Clone JOVI.1, PE). Cells were then sorted on a BD FACSAria Fusion platform for live CD3+ cells. The sorted Jurkat cells were recovered and expanded in RPMI medium with 10% FBS, 2mM L-glutamine, 1mM sodium pyruvate, 1X MEM non-essential amino acid solution and 200 mg/ml hygromycin.

SPHERO™ Streptavidin 4.0–4.5 μm magnetic particles (Spherotech) were coated with 18 pM biotinylated anti-CD28 antibody (BioLegend) and varying concentrations of refolded pHLA complexes (10-fold dilutions from 18 pM) in PBS for 1 h at 4 °C. Coating was followed by two washes of PBS and resuspension in assay medium (RPMI-1640 medium containing 10% fetal bovine serum, NEAA and Sodium pyruvate (Gibco)).

TCR-null NFAT-Luciferase Jurkat cells transduced with either S8WT or S89F3 were co-cultured with aAPC targets at a 1:1 E: T ratio and incubated at 37 °C, 5% CO2 for 6 h in opaque, white plates. Bio-glo luciferase assay substrate (Promega) was then added to the co-culture plate according to the manufacturer’s instructions. NFAT activity was determined by measuring luminescence using a Clariostar plus plate reader (BMG LabTech). NFAT activity was normalised to unstimulated controls. Statistical analyses were performed using GraphPad Prism. Data are displayed as the mean ± SEM and were analysed using a paired t test with Wilcoxon matched pairs signed rank test.

X-ray data of S2wt-A3-ELF, S2wt-A11-ELF, S2198-A3-ELF, S8wt-A3-ELF, and S89F3-A11-ELF complexes were deposited at the Protein Data Bank (PDB) under accession codes 8RYM, 8RYN, 8RYO, 8RYP and 8RYQ, respectively.

Varela-Rohena, A. et al. Control of HIV-1 immune escape by CD8 T cells expressing enhanced T-cell receptor. Nat. Med. 14, 1390–1395 (2008).

CAS PubMed PubMed Central Google Scholar

Cole, D. K. et al. Germ line-governed recognition of a cancer epitope by an immunodominant human T-cell receptor*. J. Biol. Chem. 284, 27281–27289 (2009).

CAS PubMed PubMed Central Google Scholar

Bulek, A. M. et al. Structural basis for the killing of human beta cells by CD8+ T cells in type 1 diabetes. Nat. Immunol. 13, 283–289 (2012).

CAS PubMed PubMed Central Google Scholar

Ishigaki, K. et al. HLA autoimmune risk alleles restrict the hypervariable region of T cell receptors. Nat. Genet. 1–10. https://doi.org/10.1038/s41588-022-01032-z (2022).

Cole, D. K. et al. Hotspot autoimmune T cell receptor binding underlies pathogen and insulin peptide cross-reactivity. J. Clin. Investig. 126, 2191–2204 (2016).

PubMed PubMed Central Google Scholar

Barker, D. J. et al. The IPD-IMGT/HLA database. Nucleic Acids Res. 51, D1053–D1060 (2022).

PubMed Central Google Scholar

Rapoport, A. P. et al. NY-ESO-1 specific TCR engineered T-cells mediate sustained antigen-specific antitumor effects in myeloma. Nat. Med. 21, 914–921 (2015).

CAS PubMed PubMed Central Google Scholar

Coles, C. H. et al. TCRs with distinct specificity profiles use different binding modes to engage an identical Peptide–HLA complex. J. Immunol. 204, ji1900915 (2020).

Google Scholar

Coles, C. H. et al. T cell receptor interactions with human leukocyte antigen govern indirect peptide selectivity for the cancer testis antigen MAGE-A4. J. Biol. Chem. 295, 11486–11494 (2020).

CAS PubMed PubMed Central Google Scholar

Middleton, M. R. et al. A TCR/anti-CD3 bispecific fusion protein targeting gp100, potently activated antitumor immune responses in patients with metastatic melanoma. Clin. Cancer Res. 26, 5869–5878 (2020). Tebentafusp.

CAS PubMed PubMed Central Google Scholar

Fergusson, J. R. et al. Immune-mobilizing monoclonal T cell receptors mediate specific and rapid elimination of hepatitis B-infected cells. Hepatology. 72, 1528–1540 (2020).

CAS PubMed Google Scholar

Dhillon, S. Tebentafusp: First approval. Drugs 82, 703–710 (2022).

CAS PubMed Google Scholar

Nathan, P. et al. Overall survival benefit with tebentafusp in metastatic uveal melanoma. New. Engl. J. Med. 385, 1196–1206 (2021).

CAS PubMed Google Scholar

Ogg, G., Cerundolo, V. & McMichael, A. J. Capturing the antigen landscape: HLA-E, CD1 and MR1. Curr. Opin. Immunol. 59, 121–129 (2019).

CAS PubMed Google Scholar

Simpson, A. A. et al. Structural and energetic evidence for highly peptide-specific tumor antigen targeting via allo-MHC restriction. Proc. Natl. Acad. Sci. 108, 21176–21181 (2011).

ADS CAS PubMed PubMed Central Google Scholar

Wang, Y. et al. How an alloreactive T-cell receptor achieves peptide and MHC specificity. Proc. Natl. Acad. Sci. 114, E4792–E4801 (2017).

CAS PubMed PubMed Central Google Scholar

Macdonald, W. A. et al. T cell allorecognition via molecular mimicry. Immunity. 31, 897–908 (2009).

CAS PubMed Google Scholar

Singh, N. K. et al. A class-mismatched TCR bypasses MHC restriction via an unorthodox but fully functional binding geometry. Nat. Commun. 13, 7189 (2022).

ADS CAS PubMed PubMed Central Google Scholar

Zhang, S. et al. Structural basis of cross-allele presentation by HLA-A*0301 and HLA-A*1101 revealed by two HIV-derived peptide complexes. Mol. Immunol. 49, 395–401 (2011).

CAS PubMed Google Scholar

Groot, N. G. et al. The HLA A03 supertype and several pan species major histocompatibility complex class I a allotypes share a preference for binding positively charged residues in the F pocket: Implications for controlling retroviral infections. J. Virol. 94, (2020).

Hansasuta, P. et al. Recognition of HLA-A3 and HLA‐A11 by KIR3DL2 is peptide‐specific. Eur. J. Immunol. 34, 1673–1679 (2004).

CAS PubMed Google Scholar

Stanojevic, M. et al. Identification of novel HLA-restricted preferentially expressed antigen in melanoma peptides to facilitate off-the-shelf tumor-associated antigen-specific T-cell therapies. Cytotherapy. 23, 694–703 (2021).

CAS PubMed PubMed Central Google Scholar

Naik, A. et al. Cancer testis antigen PRAME: An anti-cancer target with immunomodulatory potential. J. Cell. Mol. Med. 25, 10376–10388 (2021).

CAS PubMed PubMed Central Google Scholar

Wadelin, F. et al. Leucine-rich repeat protein PRAME: Expression, potential functions and clinical implications for leukaemia. Mol. Cancer. 9, 226–226 (2010).

PubMed PubMed Central Google Scholar

Hermes, N., Kewitz, S. & Staege, M. S. Preferentially expressed antigen in melanoma (PRAME) and the PRAME family of leucine-rich repeat proteins. Curr. Cancer Drug Targets. 16, 400–414 (2015).

Google Scholar

Sanderson, J. P. et al. Preclinical evaluation of an affinity-enhanced MAGE-A4-specific T-cell receptor for adoptive T-cell therapy. Oncoimmunology. 9, 1682381 (2019).

PubMed PubMed Central Google Scholar

Robinson, R. A., McMurran, C., McCully, M. L. & Cole, D. K. Engineering soluble T-cell receptors for therapy. Febs J. 288, 6159–6173 (2021).

CAS PubMed PubMed Central Google Scholar

Lowe, K. L. et al. Novel TCR-based biologics: Mobilising T cells to warm ‘cold’ tumours. Cancer Treat. Rev. 77, 35–43 (2019).

CAS PubMed Google Scholar

Kessler, J. H. et al. Antigen processing by nardilysin and thimet oligopeptidase generates cytotoxic T cell epitopes. Nat. Immunol. 12, 45–53 (2011).

CAS PubMed Google Scholar

Reynisson, B., Alvarez, B., Paul, S., Peters, B. & Nielsen, M. NetMHCpan-4.1 and NetMHCIIpan-4.0: Improved predictions of MHC antigen presentation by concurrent motif deconvolution and integration of MS MHC eluted ligand data. Nucleic Acids Res. 48, gkaa379 (2020).

Google Scholar

Moysey, R. K. et al. High affinity soluble ILT2 receptor: A potent inhibitor of CD8+ T cell activation. Protein Cell. 1, 1118–1127 (2010).

CAS PubMed Google Scholar

Barber, C. et al. Structure-guided stabilization of pathogen‐derived peptide‐HLA‐E complexes using non‐natural amino acids conserves native TCR recognition. Eur. J. Immunol. 52, 618–632 (2022).

CAS PubMed PubMed Central Google Scholar

McMahon, R. M. et al. Structure of HLA-A*0301 in complex with a peptide of proteolipid protein: Insights into the role of HLA-A alleles in susceptibility to multiple sclerosis. Acta Crystallogr. Sect. D Biol. Crystallogr. 67, 447–454 (2011).

ADS CAS Google Scholar

Goebeler, M. E. & Bargou, R. C. T cell-engaging therapies—BiTEs and beyond. Nat. Rev. Clin. Oncol. 17, 418–434 (2020).

PubMed Google Scholar

Klebanoff, C. A., Chandran, S. S., Baker, B. M., Quezada, S. A. & Ribas A. T cell receptor therapeutics: Immunological targeting of the intracellular cancer proteome. Nat. Rev. Drug Discov. 1–22. https://doi.org/10.1038/s41573-023-00809-z (2023).

Smithy, J. W., Blouin, A., Diamond, L. C. & Postow, M. Ensuring equity in the era of HLA-restricted cancer therapeutics. J. Immunother. Cancer. 10, e005600 (2022).

PubMed PubMed Central Google Scholar

Robertson, I. B. et al. Tuning the potency and selectivity of ImmTAC molecules by affinity modulation. Clin. Exp. Immunol. 215, 105–119 (2023).

PubMed Central Google Scholar

Poole, A. et al. Therapeutic high affinity T cell receptor targeting a KRASG12D cancer neoantigen. Nat. Commun. 13, 5333 (2022).

ADS CAS PubMed PubMed Central Google Scholar

Liddy, N. et al. Monoclonal TCR-redirected tumor cell killing. Nat. Med. 18, 980–987 (2012).

CAS PubMed Google Scholar

Blevins, S. J. et al. How structural adaptability exists alongside HLA-A2 bias in the human αβ TCR repertoire. Proc. Natl. Acad. Sci. 113, E1276–E1285 (2016).

Sun, Y. et al. Structural principles of peptide-centric chimeric antigen receptor recognition guide therapeutic expansion. Sci. Immunol. 8, (2023).

Yarmarkovich, M. et al. Targeting of intracellular oncoproteins with peptide-centric CARs. Nature. 623, 820–827 (2023).

ADS CAS PubMed PubMed Central Google Scholar

Papadaki, G. F. et al. Decoupling peptide binding from T cell receptor recognition with engineered chimeric MHC-I molecules. Front. Immunol. 14, 1116906 (2023).

CAS PubMed PubMed Central Google Scholar

Wooldridge, L. et al. A single autoimmune T cell receptor recognizes more than a million different peptides*. J. Biol. Chem. 287, 1168–1177 (2012).

CAS PubMed Google Scholar

Richards, D. M. et al. The contained self-reactive peripheral T cell repertoire: Size, diversity, and cellular composition. J. Immunol. 195, 2067–2079 (2015).

CAS PubMed Google Scholar

Cameron, B. J. et al. Identification of a titin-derived HLA-A1–presented peptide as a cross-reactive target for engineered MAGE A3–directed T cells. Sci. Transl Med. 5, 197ra103 (2013).

PubMed PubMed Central Google Scholar

Raman, M. C. C. et al. Direct molecular mimicry enables off-target cardiovascular toxicity by an enhanced affinity TCR designed for cancer immunotherapy. Sci. Rep-uk. 6, 18851 (2016).

ADS CAS Google Scholar

Singh, N. K. et al. Emerging concepts in TCR specificity: Rationalizing and (maybe) predicting outcomes. J. Immunol. 199, 2203–2213 (2017).

CAS PubMed Google Scholar

Ramsuran, V. et al. Epigenetic regulation of differential HLA-A allelic expression levels. Hum. Mol. Genet. 24, 4268–4275 (2015).

CAS PubMed PubMed Central Google Scholar

Belicha-Villanueva, A. et al. Differential contribution of TAP and tapasin to HLA class I antigen expression. Immunology. 124, 112–120 (2008).

CAS PubMed PubMed Central Google Scholar

Shen, Y., Parks, J. M. & Smith, J. C. HLA Class I supertype classification based on structural similarity. J. Immunol. 210, 103–114 (2023).

CAS PubMed Google Scholar

Kumar, P. et al. Structural basis for T cell alloreactivity among three HLA-B14 and HLA-B27 Antigens*. J. Biol. Chem. 284, 29784–29797 (2009).

CAS PubMed PubMed Central Google Scholar

Li, Y. et al. Directed evolution of human T-cell receptors with picomolar affinities by phage display. Nat. Biotechnol. 23, 349–354 (2005).

CAS PubMed Google Scholar

Dunn, S. M. et al. Directed evolution of human T cell receptor CDR2 residues by phage display dramatically enhances affinity for cognate peptide-MHC without increasing apparent cross‐reactivity. Protein Sci. 15, 710–721 (2006).

CAS PubMed PubMed Central Google Scholar

Liu, H. & Naismith, J. H. An efficient one-step site-directed deletion, insertion, single and multiple-site plasmid mutagenesis protocol. Bmc Biotechnol. 8, 91–91 (2008).

PubMed PubMed Central Google Scholar

Garboczi, D. N., Hung, D. T. & Wiley, D. C. HLA-A2-peptide complexes: Refolding and crystallization of molecules expressed in Escherichia coli and complexed with single antigenic peptides. Proc. Natl. Acad. Sci. 89, 3429–3433 (1992).

ADS CAS PubMed PubMed Central Google Scholar

Boulter, J. M. et al. Stable, soluble T-cell receptor molecules for crystallization and therapeutics. Protein Eng. Des. Sel. 16, 707–711 (2003).

CAS Google Scholar

Cole, D. K. et al. Computational design and crystal structure of an enhanced affinity mutant human CD8 αα coreceptor. Proteins Struct. Funct. Bioinform. 67, 65–74 (2007).

CAS Google Scholar

Winter, G. xia2: An expert system for macromolecular crystallography data reduction. J. Appl. Crystallogr. 43, 186–190 (2010).

ADS CAS Google Scholar

Kabsch, W. X. D. S. Acta Crystallogr. Sect. D Biol. Crystallogr. 66, 125–132 (2010).

ADS CAS Google Scholar

Winter, G. et al. DIALS: Implementation and evaluation of a new integration package. Acta Crystallogr. Sect. D Struct. Biology. 74, 85–97 (2018).

ADS CAS Google Scholar

McCoy, A. J. et al. Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674 (2007).

ADS CAS PubMed PubMed Central Google Scholar

Ballard, C. et al. CCP4: A resource for macromolecular crystallography. Acta Crystallogr. Sect. Found. Adv. 70, C1723–C1723 (2014).

Google Scholar

Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. Sect. D Biol. Crystallogr. 66, 486–501 (2010).

ADS CAS Google Scholar

Murshudov, G. N. et al. REFMAC5 for the refinement of macromolecular crystal structures. Acta Crystallogr. Sect. D: Biol. Crystallogr. 67, 355–367 (2011).

ADS CAS Google Scholar

Joosten, R. P., Long, F., Murshudov, G. N. & Perrakis, A. The PDB_REDO server for macromolecular structure model optimization. Iucrj. 1, 213–220 (2014).

CAS PubMed PubMed Central Google Scholar

Williams, C. J. et al. MolProbity: More and better reference data for improved all-atom structure validation. Protein Sci. 27, 293–315 (2018).

CAS PubMed Google Scholar

Rudolph, M. G., Stanfield, R. L. & Wilson, I. A. How TCRS bind MHCS, peptides, and coreceptors. Annu. Rev. Immunol. 24, 419–466 (2006).

CAS PubMed Google Scholar

Lee, C. M. Y., Iorno, N., Sierro, F. & Christ, D. Selection of human antibody fragments by phage display. Nat. Protoc. 2, 3001–3008 (2007).

CAS PubMed Google Scholar

Martin, M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnetJ. 17, 10–12 (2011).

Google Scholar

Malzer, C. & Baum, M. A. Hybrid approach to hierarchical density-based cluster selection. arXiv. https://doi.org/10.48550/arxiv.1911.02282 (2019).

Download references

We thank Diamond Light Source, UK for access to beamlines (proposals in22870 and in28224) and beamline staff for support with X-ray data collection. We thank Tristan Vaughan, Milos Aleksic, and JoAnn Suzich for critically reading the manuscript.

Immunocore Ltd, 92 Milton Park, Abingdon, Oxfordshire, OX14 4RY, UK

Vijaykumar Karuppiah, Dhaval Sangani, Lorraine Whaley, Robert Pengelly, Pelin Uluocak, Ricardo J. Carreira, Miriam Hock, Pietro Della Cristina, Paulina Bartasun, Paula Dobrinic, Nicola Smith, Keir Barnbrook, Ross A. Robinson & Stephen Harper

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

V.K. and S.H. conceived the project. V.K. purified TCRs and generated crystallography and binding affinity data. K.B, M.H. and R.P. purified wild type and chimeric pHLAs. D.S., L.W. and P.U. engineered Jurkat cell lines and performed cellular assays. R.J.C. analysed mass spectrometry data and prepared Fig. 1. P.D.C and P.B screened TCR mutants by phage display. P.D., N.S. and R.A.R., contributed to study design and analysis. V.K. and S.H. analysed the data and wrote the manuscript with input from all authors.

Correspondence to Vijaykumar Karuppiah or Stephen Harper.

All authors are current or former employees of Immunocore Ltd and may hold Immunocore stock options.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Below is the link to the electronic supplementary material.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

Karuppiah, V., Sangani, D., Whaley, L. et al. Broadening alloselectivity of T cell receptors by structure guided engineering. Sci Rep 14, 26851 (2024). https://doi.org/10.1038/s41598-024-75140-7

Download citation

Received: 11 February 2024

Accepted: 03 October 2024

Published: 06 November 2024

DOI: https://doi.org/10.1038/s41598-024-75140-7

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative