Overcoming T cell tolerance to tumor self-antigens through catch-bond engineering

Abstract

T cells are often weakly responsive to tumor self-antigens because of central tolerance, constraining their ability to eliminate tumors. We exploited mechanical force to engineer a weakly reactive T cell receptor (TCR) specific for a nonmutated tumor-associated antigen (TAA), prostatic acid phosphatase (PAP). We identified a catch-bonding "hotspot" whose mutation enhanced T cell activity by increasing TCR-pMHC (peptide-major histocompatibility complex) bond lifetime while preserving physiological affinities and antigen fine specificities. T cells expressing these engineered TCRs showed vastly superior expansion in the tumor, effector phenotypes, and tumor elimination. Crystal structures and molecular dynamics simulations revealed a single amino acid mutation at the catch-bond hotspot primes the TCR for peptide interaction through water reorganization at the TCR-pMHC interface. Catch-bond engineering is a viable biophysically based strategy for transforming tolerized antitumor T cells into potent TCR-T cell therapy killers.

Fig. 1:. TCR catch bond engineering improves the sensitivity of the PAP₂₂/HLA-A2 specific TCR156.

(A) Schematic illustration of the strategy for PAP TCR engineering. Three major steps were designed, each intended to further augment the TCR156 potency. (B-D) Position scanning of TCR156α. (B) Example FACs plot of CD69 expression of SKW-3 cells expressing TCR156 bearing mutations at position 30 and 32 compared to TCR156wt upon stimulation with 10⁻⁷M PAP₂₂. (C) Titration curves of the TCR156 mutants that showed enhanced CD69 expression upon antigen stimulation. (D) Deconvolution of position 30 and 32. The experiments were carried out in duplicate, each experiment was repeated once. (E-F) DNA shuffling library selection. (E) SKW-3 cells expressing the TCR156α DNA shuffling library were selected for five rounds to enrich population that expressed higher CD69 but no increased binding to PAP₂₂/HLA-A2 tetramer compared to the wild-type TCR156 SKW-3 cells. Round 1, 3, and 5 selection are shown. The round 5 selected skw3 T cells did not respond to MAGE-A1/HLA-A2 stimulation (last plot). The selected cells were deep-sequenced and we test the top 12 clones. Peptide titration curve for the best clone S32Qα is shown in (F). (G) Illustration of the two “catch-bond hotspot” on TCR156α CDR1 domain. (H) The hotspot Ser32 was substituted with the additional 16 amino acids. The amino acid substitutes that enhanced the EC50 of the TCR upon peptide titration were then combined with the Ser30Glu mutation to screen for EC50 enhanced clones. The S32Mα and the S30ES32Qα TCR mutants were shown in (H). The grey area indicated the distribution of all the other TCR mutants that exhibited EC50 enhancement during the previous selections. These experiments were repeated once, each experiment included duplicates. (I) The EC50 comparison of the EC50 improved clones from different selection steps shown as Mean ± SEM. ONE-way ANOVA followed by Dunnett for multiple comparisons was performed to compare TCR156 variants to 156wt. * P ≤ 0.05; ** P ≤ 0.01; *** P ≤ 0.001; **** P ≤ 0.0001. These experiments were repeated twice, each experiment included duplicates. Abbreviation: irr. TCR: irrelevant TCR. The baseline CD69 MFIs from the titration curves, which are SKW-3 cultured with T2 cells without peptides, were subtracted.

Fig. 2:. Catch bond engineered TCRs exhibit prolonged bond lifetimes that correlate with enhanced functionality.

(A) Illustration of the BFP experiment setup. (B) Selected TCR variants were subjected to BFP measurement. The bond lifetime between the TCR and PAP/HLA-A*02:01 were measured under zero to 45 pN of force. (C) Comparisons of peak bond lifetime of the TCR variants measured in BFP. Data indicate mean ± SEM. ONE-way ANOVA followed by Dunnett for multiple comparisons was performed to compare peak bond lifetime of different groups. * P ≤ 0.05; ** P ≤ 0.01; *** P ≤ 0.001; **** P ≤ 0.0001 (D) SPR measurement of the TCR variants with the PAP/HLA-A*02:01. The K_d values were calculated and shown for each TCR. (E-G) Correlation of the CD69 EC50 versus peak bond lifetimes of TCR56wt and TCR variants measured by BFP (E), koff by kinetic SPR at low coupling density (F) or (G) Kd measured by steady-state SPR, Spearman coefficient is calculated and shown (data shown in fig. S6).

Fig. 3:. T cell profiling of the TCR156 variants.

(A) Membrane expression of CD107a, the cytotoxic degranulation marker, on primary human T cells after the T cells were cultured with PC3-PAP-A2 for 4 hours at an E:T ratio of 4:1. The experiments were performed in duplicates and repeated at least once with human PBMCs from 3 donors. Data indicate mean ± SEM. (B-D) Primary human T cells expressing TCR156 variants were co-cultured with PC3-PAP-A2 for 16 hours at an E:T ratio of 4:1. Intracelluar staining was performed to evaluate the production of Granzyme B (B), IFNγ (C) and TNFα (D). Experiments were performed in duplicates and repeated at least once with human PBMCs from 3-4 donors. Data indicate mean ± SEM. (E-F) T cell proliferation upon stimulation with PC3-PAP-A2 for 72 hours at an E:T ratio of 4:1. The T cells were labeled with CellTrace dye and the proliferated cells and the number of proliferation cycles were tracked by flow cytometry. Representative plots are shown in (E). The percentage of original T cells that underwent proliferated were compared in (F). Experiments were performed in duplicates with human PBMCs from 3 donors. Data indicate mean ± SEM in (F). (G-I) in vitro exhaustion assay with repetitive PAP antigen stimulation. 2x10⁴ PC3-PAP-A2 cells were added every 2 days to a starting culture of 4x10⁴ primary T cells with TCR156 variants as illustrated in (G). The exhaustion states were determined on day 7-8. (H) Percentage of different IRs were depicted in Pie charts. (I) The expression of the transcription factor TOX in the T cells 12 days after repetitive stimulation. The exhaustion assay was performed with 3 human donors, each in duplicates. ONE-way ANOVA followed by Dunnett for multiple comparisons was performed to compare the TCR156 variants to 156wt for plots A-D, F and I. * 0.01< P ≤ 0.05; ** 0.001< P ≤ 0.01; *** 0.001< P ≤ 0.001; **** P ≤ 0.0001 (J) Radar plot showed the overall improvement of cytotoxicity (represented by Granzyme B), IFNγ and/or TNFα, proliferation potency and resistance to exhaustion (TOX⁻ T cells) of the TCR156 mutants compared to wild-type TCR. Abbreviation: ntrd: non-transduced

Fig. 4:. Anti-tumor potencies of the TCR156 variants.

(A-D) In vitro killing of the TCR156 variants expressing human PBMCs measured in Incucyte assay under different conditions. (A) E:T ratio of 1:1; (B) Different E:T ratios as indicated were tested for TCR156 variants S32Mα and S30E32Qα; (C) TCR156 variants S32Mα and S30E32Qα were stimulated by PC3-PAP-A2 repetitively every 2 days. (D) Transduced CD4 and CD8 T cells expressing wt, S32Mα, S30E32Qα TCRs were cultured with PC3-PAP at E:T ratio of 1:1. These experiments were repeated using PBMCs from at least three healthy individuals; each experiment included triplicates. (E) Timeline of adoptive T cell transfer in NSG mice. (F) Tumor size measurement of mice treated with different TCR expressing human PBMCs either as average (left panel) or individual mice (right panels). (G) Immunohistochemistry staining of the excised tumor samples after adoptive T cell transfer. The tumor experiment was repeated in three independent experiments using PBMCs from three healthy individuals, each experiment included 5 animals per group.

Fig. 5:. Single-cell RNA-sequencing analysis of the TILs from the PC3-PAP-A2 xenograft in NSG mice.

(A) Experimental setup for isolation of TILs from different groups of T cells transduced with non-engineered (i.e., wild-type; TCR156) or three variants of catch bond engineered TCRs. An irrelevant TCR (F5) was also used as a negative control. For each experimental group, TILs collected from three individual animals were pooled (n=3) and sequenced. (B) Uniform Manifold Approximation and Projection (UMAP) representation of TILs from all groups combined. The TILs sequenced in this study were mapped onto a reference pan-caner T-cell atlas (52) to cluster the cells according to their differentiation states. (C) Heatmap of the representative genes for each cluster. (D) UMAP of individual groups. (E) Pie chart indicating the proportion of each cluster in individual groups. (F) The distribution of different clusters on the pseudotime axis for the different TIL groups. (G) The effector scores for each clusters (see methods for effector score calculation). (H) Correlation of the BFP peak bond lifetime and the median effector scores of different TCR156 variants. (I) Scaled expression levels of representative genes for T-cell effector function. (J) Geneset enrichment analysis indicated the pathway differences between the TCR catch-bond engineered TILs and the wild-type TILs were E2F targets and G2M checkpoints. (K) The Cell-cycle scores for each cluster. (L) Correlation of the BFP peak bond lifetime and the median Cell-cycle scores of different TCR156 variants. (M) Scaled expression levels of representative genes associated with proliferation.

Fig. 6:. Structural environment of a catch bond hotspot at the TCR-pMHC interface.

(A) Crystal structures of wild-type TCR156 and mutants superimposed on HLA-A2 (B) CDR footprint of TCR156 variants aligned on PAP-A2. Wild-type (sky blue), S32Hα (red), S32Mα (magenta), S32Qα (blue), S30E32Qα (pink) The Cα r.m.s.d. for the TCRs range between 0.298 – 0.369 Å (C-F) (top) Details of boxed region in A showing closest approach of Ser32α mutants to PAP₂₂ peptide. TCR β chain is omitted for clarity. Bound water molecules resolved in the electron density maps are shown as red spheres. Buried surface area between residue 32 and the peptide/HLA-A2 is: wt: 24 Ų; S32H: 160 Ų; S32M: 180 Ų; S32Q: 153 Ų; (bottom) Contact plots of variant residues at position 32 of the alpha chain with coloring as in panel A. Hydrogen bonds are shown as dashed green lines, water molecules as large spheres, and van der Waals contacts are indicated by rays.

Fig. 7:. The catch bond hotspot mutation primes the TCR for peptide interaction.

(A-E) TCR–peptide interface dynamics in MD simulation. (A) Inset showing the Vα hotspot region of interest with the TCRβ subunit hidden. (B) Representative simulation snapshots show water-mediated hydrogen-bonding networks that form in WT simulations and a network of protein–peptide hydrogen bonds that form in S32M simulations. Note that these protein–peptide hydrogen bonds can also form in WT simulations. (C) Box-and-whisker plots show the average number of hydrogen bonds formed between Ser4 and either Asn92 or Asn93 on TCRα at any given frame, averaged across each independent simulation. Ten simulations analyzed for each condition. P < 0.01 (D) Overlay of simulation frames down sampled every 200 ns across ten independent simulations of either complex. Displayed waters are those in a bounding sphere of radius 4.5 Å, centered on the crystallographic Met32 sulfur atom after global alignment of all simulation frames. (E) Box-and-whisker plots of the average number of waters within the sphere across ten simulations of either complex. P < 1 × 10⁻⁶ (F-G) A2-peptide yeast library selection of TCR156wt and S32Mα. (F) Yeast library design of HLA-A*02:01 presenting randomized 9mer peptides with restricted amino acids at the anchor position 2 and 9. (G) Heat map of round 3 (TCR156wt) or 4 (S32Mα) selected peptides. Darker blue indicates the preferred amino acid at the position. Boxed amino acids represents the PAP₂₂ epitope TLMSAMTNL. Serine at position 4 has been circled in red. (H) Differences in EC50 for 156wt and S32Mα TCRs when responding to wild-type PAP22 versus PAP22-Ala4. PAP-S4A led to enhanced EC50 for TCR156wt, but reduced EC50 for S32Mα. The changes are shown in logarithmic scale. (I-J) BFP measurements of 156wt and S32Mα with the PAP₂₂-S4A variant. (I) The bond lifetime between the TCRs and alanine variants on PAP/A2 were measured under zero to around 30 pN of force. (J) Comparisons of peak bond lifetime of the PAP/A2 variants measured in BFP. Data indicate mean ± SEM. Student T test was performed to compare peak bond lifetime between different groups. **** P ≤ 0.0001