Recombinant T cell receptors (TCRs) can be used to redirect naïve T cells to eliminate virally infected or cancerous cells; however, they are plagued by low stability and uneven expression. Here, we use molecular modeling to identify mutations in the TCR constant domains (Cα/Cβ) that increase the unfolding temperature of Cα/Cβ by 20 °C, improve the expression of four separate α/β TCRs by 3- to 10-fold, and improve the assembly and stability of TCRs with poor intrinsic stability. The stabilizing mutations rescue the expression of TCRs destabilized through variable domain mutation. The improved stability and folding of the TCRs reduces glycosylation, perhaps through conformational stabilization that restricts access to N-linked glycosylation enzymes. The Cα/Cβ mutations enables antibody-like expression and assembly of well-behaved bispecific molecules that combine an anti-CD3 antibody with the stabilized TCR. These TCR/CD3 bispecifics can redirect T cells to kill tumor cells with target HLA/peptide on their surfaces in vitro.
T cell receptors (TCRs) are the adaptive immune system’s tool for recognizing and eliminating “non-self” intracellular antigens. These non-self antigens can emerge from viral infection or from genetic alteration. Genetic alterations are key to aberrant cellular function and can lead to diseases such as cancer. α/β TCRs are expressed on both CD8+ effector and CD4+ helper T cells and recognize proteosomally degraded foreign antigens displayed on infected/cancerous cell surfaces when complexed with HLA/MHC1,2. When expressed on CD8+ T effector cells, α/β TCRs interact specifically with Type I MHC/peptide complexes and this interaction induces T cell activation and elimination of cells displaying recognizable non-self antigens.
Many approaches have been developed to harness the exquisite non-self-recognizing properties of α/β TCRs for therapeutic use. Recombinant α/β TCRs have been transduced/transfected into bulk naïve T cells as a means of retargeting these T cells to target tumor associated antigens3,4. Alternately, the use of the soluble extracellular region of the TCR fused to an scFv that recognizes an activating TCR subunit, commonly CD3ε, has been used to redirect endogenous T cells to target tumor cells5. Unlike antibody-based Bispecific T cell Engagers (BiTEs) that rely on the direct recognition of overexpressed antigens on the cell surface6, TCR or TCR-mimic bispecifics can recognize a much larger subset of intracellular and abnormal tumor or viral antigens and thus have broader potential applicability5.
However, TCR assembly and expression is challenging7. The common method of soluble α/β TCR production is expression as inclusion bodies in Escherichia coli (E. coli) followed by resolubilization, refolding/assembly/oxidation, and finally purification at relatively low yields8. To simplify the assembly, some researchers have tried using only the variable domain regions of TCRs in a single chain format or scTv, like an antibody single chain Fv (scFv), for targeting specific HLA/peptide complexes9. While scFvs have shown a propensity for instability, aggregation, and low solubility, scTvs have generally shown even worse expression and stability issues. Great efforts have been made to stabilize scTvs for therapeutic and diagnostic use9. Unfortunately, the high diversity of Vα/Vβ germlines (higher than antibody VH/VL germlines), renders each set of stabilizing designs within scTvs to be unique to the individual Vα/Vβ subunit and unlikely to find general use.
Given that most extracellular proteins are intrinsically glycosylated with complex disulfide pairings, we decided to rely on mammalian expression to generate soluble TCRs. Industrial antibody production has predominately moved to mammalian expression systems10. However, it has been our experience that α/β TCRs express poorly with less reliable assembly than antibodies when expressed in the commonly utilized Chinese hamster ovary (CHO) cell system. Many novel bispecific antibody (BsAb) formats, including those with relevance to soluble α/β TCR bispecifics, may require TCRs to express at antibody-like levels for proper molecular assembly. Bispecific “ImmTac” moieties that recombinantly fuse soluble TCRs to antibody scFvs have typically followed the complex E. coli production process described above5.
We hypothesize that general stabilization of the Cα/Cβ subunit may improve the overall stability and folding of α/β TCRs. Recent studies have shown that strong thermodynamic cooperativity exists between the subunits of α/β TCRs. Cα requires pairing with Cβ in the ER for folding similar to what has been observed for antibody CH1/Cκ subunits11,12. Additionally, many Vα/Vβ subunits are intrinsically unfolded in isolation and require Cα/Cβ for proper folding13. In support of our hypothesis that Cα/Cβ stabilization may generally improve TCR expression and stability, adding a disulfide between the Cα/Cβ domains positively impacts many α/β TCRs14. Therefore, we set out to design a robust Cα/Cβ subunit for general TCR stabilization with the goal of generating TCRs at antibody-like levels that assemble properly.
To identify mutations that stabilize the Cα and Cβ domains, we perform protein design simulations with the molecular modeling software Rosetta15. During a design simulation, Rosetta samples alternative amino acid sequences and sidechain conformations in search of mutations that lower the calculated energy of the protein16. The Rosetta energy function favors amino acids that pack well and form favorable electrostatic and hydrogen bonding interactions while minimizing desolvation costs and torsional strain17. It is common to experimentally test several predictions in search of the best performing mutations because it is difficult to accurately predict how a mutation will affect protein stability18. An alternative strategy for finding mutations that will stabilize a protein is to assemble a multiple sequence alignment (MSA) for the protein family and search for highly conserved amino acids that are not conserved in the protein of interest19. Recently, there has been considerable success in finding stabilizing mutations by combining conservation analysis with energy-based methods like Rosetta20. One potential advantage of the MSA-based approach is that it may capture phenomena, such as the role of a residue in preventing aggregation, that are difficult to capture with a structure-based approach. Here, we test mutations based solely on Rosetta calculations as well as use conservation analysis to filter the results from the design simulations. After screening a host of computationally designed mutations in the Cα/Cβ context, we identify seven mutations that, when combined, significantly improve Cα/Cβ and full-length α/β TCR assembly and expression. These stabilized TCRs can be fused to antibody domains to generate functional BsAbs.
Stabilizing the Cα/Cβ TCR subunit
First, we investigated the thermodynamic properties of a soluble TCR, 1G4_122, and its Vα/Vβ and Cα/Cβ subunits. 1G4_122 binds the NY-ESO-1 antigen21. Using a mammalian expression system, we generated both the Vα/Vβ and Cα/Cβ subunits in the presence and absence of flexible (Gly4Ser×4) linkers that link Vα to Vβ or Cα to Cβ. We also tested the subunit expression and assembly with or without stabilizing interdomain disulfides. Most of the Vα/Vβ and Cα/Cβ constructs we generated either failed to express or failed to assemble, including the single chain variants. The best Vα/Vβ subunit expression was obtained by adding a Vα44/Vβ110 disulfide (homologous to the VH44/VL100 disulfide used to stabilize antibody variable domain fragments or Fvs22), while the best Cα/Cβ expression was obtained using both the known stabilizing disulfide (Cα166/Cβ173) near the top of the Cα/Cβ interface and the native Cα/Cβ disulfide at the C-terminus of the domains (Cα213/Cβ247)14. Differential scanning calorimetry (DSC) experiments showed the (Cα166/Cβ173) disulfide increased the midpoint of thermal unfolding (Tm) of the full-length TCR (both subunits) by 8 °C. When comparing the Vα/Vβ and Cα/Cβ subunit unfolding to the unfolding of the full-length extracellular TCR protein, the individual subunits were clearly both destabilized in the absence of one another (Fig. 1a), agreeing with the thermodynamic cooperativity observed previously13.
To investigate if further stabilization of Cα/Cβ could benefit TCR expression, stability, and assembly when expressing soluble TCRs in mammalian cells, we used molecular modeling simulations to identify mutations that stabilize the constant domains. We scanned over every residue position in the constant domains and used Rosetta to model all possible point mutations (except for cysteine). For each simulation, the TCR was fixed in space and only residues in close proximity to the mutation were allowed to make small movements (“relax”) to accommodate the mutation. Long-range changes in structure were not allowed to reduce noise in the energy calculations. We repeated these local relaxations using the native sequence without any mutations. The predicted change in energy was determined by comparing the Rosetta score of the mutation with the Rosetta score of the native sequence. We then split the mutations into two lists: a list of mutations that are observed in a MSA of TCRs and a list of mutations that are absent from the MSA. We then picked the mutations with the best predicted energies from both lists for experimental screening. This method allowed a conserved mutation to be tested even if it did not have the best Rosetta energy.
Using the isolated Cα/Cβ fragment containing the Cα166/Cβ173 disulfide bond (Fig. 1a), the library of single, computationally designed mutants was generated and expressed in two separate blocks. Increased expression level was used blindly to select mutants for scale-up, purification, and characterization by differential scanning fluorimetry (DSF, Supplementary Table S1). Expression was assessed using an Octet Red with Ni2+-NTA sensortips and purified Cα/Cβ protein containing a N-terminal Cα-histag for the standard curve. Increased expression correlated well with increased protein stability. Roughly half of the single mutants chosen for further evaluation based on expression demonstrated a significant increase in stability over the wild-type (native) Cα/Cβ protein (Supplementary Table S1). The Tms for seven of the identified stabilizing mutants are shown in Fig. 1b and range from +1 to +7 °C over wild-type Cα/Cβ. Five of these mutations were substitutions that are observed in a MSA of TCRs, but the two mutations that produced the biggest gains in Tm, D155P (+7.1 °C) and S139F (+4.4 °C), are not observed in the MSA. Supplementary Fig. S1 shows the experimental outcome of each modeled point mutation with respect to its computational metrics.
Next, the stabilizing mutations were combined into a variant harboring four of the mutations and a variant with all seven mutations. Modeling results suggested each of the mutations was likely compatible with the other mutations (i.e., unlikely to interfere with each other). Adding all seven stabilizing mutations led to a seven-fold increase in expression and 20 °C increase in the Tm over the Cα/Cβ subunit harboring only the Cα166/Cβ173 disulfide (denoted “wild-type”, Table 1, Fig. 1c). Assembly of the wild-type and variant Cα/Cβ subunits was probed on a non-reducing SDS-PAGE gel. The Cα/Cβ variant with seven mutations assembled more efficiently as evidenced by the lack of disulfide laddering and was smaller/more compact, likely due to less spurious glycosylation (described in detail below) via maintenance of a more discretely folded and compact structure (Fig. 1d). Even for the stabilized variant, multiple bands are clear on the SDS-PAGE gel (Fig. 1d) that likely reflect partial glycan occupancy of the three N-linked glycosylation sites in Cα and single N-linked glycosylation site in Cβ.
Lastly, we assessed the impact of the designs on potential immunogenicity of the TCR. Using the EpiVax server, we investigated whether any increase in MHC-peptide complexes were likely for each of the mutations23. Overall, a slight increase in the EpiMatrix score was observed for both Cα and Cβ. Increases in the individual peptide scores were marginal and did not reach the level predicted to result in significant MHC binding. An outlier of concern included the Cα_T150I mutation, whose score went up substantially for one peptide 9-mer and there is strong predicted MHC binding predicted for a second 9-mer peptide that also exists in the wild-type Cα peptide. The Cβ_E134K_H139R mutants are found together on multiple 9-mer peptides and there is a slight increase in the EpiMatrix scores, but none of the scores put these peptides in the high-risk category (i.e., strong predicted MHC binding). Interestingly, the Cβ_D155P and Cβ_S170D variants, which are the two most impactful mutants from a Tm perspective, lower the overall EpiMatrix scores versus wild-type Cα/Cβ.
Characterization of the seven stabilizing Cα/Cβ mutations
Next, we attempted to crystallize both the wild-type and stabilized forms of the Cα/Cβ subunit. N-linked glycans were removed enzymatically before crystallization. Unsurprisingly based on the SDS-PAGE gel results with the proteins (Fig. 1d), only the seven mutant variant yielded protein crystals. We obtained a 1.76 Å crystal structure of the stabilized Cα/Cβ subunit (Supplementary Table S2). The side chains for all seven mutants are resolved in the structure and provide an indication of how the mutations are stabilizing the protein (Figs. 2 and 3).
In the crystal structure and the Rosetta model, phenylalanine 139 (Cα S139F) fills a partially buried cleft in the structure and forms tight packing interactions with the sidechain of Cα 129Q. Notably, the orientation of the interaction leaves the polar groups on the glutamine open for forming hydrogen bonds with water. Despite being solvent exposed, Cβ D155P has the largest impact on protein stability of the seven mutations. The Rosetta energy calculations favor the proline because the residue has backbone torsion angles (ϕ = −73.4°, ψ = 52.3°) compatible with the closed ring of the proline. One benefit of mutating to proline is that proline is more restrictive than aspartic acid in Ramachandran space and therefore loses less entropy when the protein folds24. Isoleucine 150 (Cα T150I) packs tightly with surrounding residues and removal of the threonine does not leave any buried polar groups without a hydrogen-bond partner.
Cα A190T is at the beginning of a turn in a solvent-exposed loop. In addition to increasing solubility on the surface, Rosetta predicted that this threonine would form a stabilizing hydrogen bond to Cα 129Q. However, the conformation of Cα 129Q appears to be influenced away from the hydrogen bonding conformation by another mutation, Cα S139F. Instead, the crystal structure shows the threonine adopting a different rotamer to form a hydrogen bond with the backbone on the other end of the loop’s turn (backbone nitrogen of Cα 193N). Rosetta predicted that Cβ S170D would form a bidentate hydrogen bond across the interface with the sidechain of Cα 171R. The crystal structure shows Cβ S170D hydrogen bonding with the backbone nitrogen atom of Cα 171R (which was not previously participating in a hydrogen bond) and allowing the arginine’s sidechain to become more solvent-exposed. Cβ H139R creates an interface-spanning hydrogen bond with Cα 124D. Rosetta modeled the arginine and aspartic acid to form a bidentate hydrogen bond, however the crystal structure shows that they form just one hydrogen bond so that the arginine can better pack with neighboring residues. The designed lysine at residue 134 on the β chain (Cβ E134K) has high b-factors and is likely interacting with several negatively charged amino acids in the vicinity.
After seeing the sidechain conformations in the crystal structure deviate from the Rosetta model for many of the designed polar interactions, we decided to investigate why Rosetta did not predict the observed sidechain conformations. One possibility is that the sidechain and backbone sampling protocol in Rosetta was unable to sample these conformations. Alternatively, Rosetta may have sampled them but predicted them to be higher in energy. To test between these hypotheses, we used the backbone coordinates from the crystal structure of the stabilized mutant as the starting point for fixed-backbone sidechain prediction simulations. The backbone conformation in our crystal structure is similar (0.338 Å rmsd) to the crystal structure of the wild-type protein, but there are small deviations that may affect sidechain positioning and their ability to form hydrogen bonds. Indeed, this is what we observed, given the crystal structure backbone Rosetta correctly predicted most of the sidechain conformations of the mutants (Supplementary Fig. S2). This result suggests that Rosetta did not correctly predict the sidechain conformations of these mutations in the original design model because it did not sample and/or favor the small backbone perturbations observed in the crystal structure.
Overall, the stability measurements and the crystal structure of the stabilized variant highlight some of the strengths and weaknesses of protein modeling. In general, it is more straightforward to predict the conformation and energetic consequences of hydrophobic mutations in the protein core. The constrained environment as well as the more limited flexibility of hydrophobic side chains allows more accurate prediction of the most favorable rotamer25. Polar amino acids are typically exposed to solvent and in general are longer with more sidechain degrees of freedom. Given the challenges associated with modeling polar residues, making use of information in a MSA may be most useful when mutating polar amino acids. Consistent with this conclusion, all five of the stabilizing mutations that involved mutation to a polar amino acid were present in the MSA, while the stabilizing mutations that were not in the MSA involved mutations to hydrophobic amino acids. The absolute value of the Rosetta Energies (dE) for each of the stabilizing polar mutations was smaller than all the dEs of the mutations chosen without the requirement of being within the MSA. Without the MSA filter, most of the stabilizing polar mutations would not have been included in the group of 50 mutants that were screened.
Impact of the Cα/Cβ designs on full-length TCRs
We evaluated the impact of the Cα/Cβ designs on the expression and stability of unrelated, full-length, soluble TCRs. First, we generated the 1G4_122 anti-NY-ESO-1 TCR with and without the Cα/Cβ designs. Without the CαT166C/CβS173C disulfide, the 1G4_122 TCR expressed poorly in mammalian cells, but with the Cα/Cβ designs, the expression increased over 10-fold and the Tm measured by DSC was 8 °C higher (Table 1, Fig. 4a). We also expressed and characterized three different TCRs targeting MAGE_A3 (pdb 5BRZ; NCBI: α=ABY74337.1/β=ACZ48691.1), HIV (pdb 3VXT; NCBI: α=ABB89050.1/β=ACY74607.1), and WT-1 antigens26,27,28. The expression of the non-stabilized TCRs was poor (59 ± 32 mg/L), while the expression of the stabilized TCRs was significantly improved (280 ± 180 mg/L) (Table 1). For comparison, the average expression of seven different monoclonal antibodies (mAbs), when expressed simultaneously with the TCRs, was 235 ± 280 mg/L. The assembly of two of the TCRs (anti-WT-1 and anti-HIV) was poor based on the lack of distinct bands near 50 kDa expected for differentially glycosylated α/β TCRs, while all TCRs showed proper assembly in the presence of the stabilizing designs. (Supplementary Fig. S3a). DSF analysis of the TCRs demonstrated that stabilized TCRs all had a dominant unfolding Tm between 60 and 70 °C, while the non-stabilized TCRs either demonstrated no clear unfolding event or an unfolding Tm < 60 °C (Supplementary Fig. S3b). When comparing the expression of the TCRs stabilized only by the CαT166C/CβS173C or by the seven mutant designs, in all cases, the seven mutant designs described here significantly improved TCR expression (by several-fold) over the addition of the CαT166C/CβS173C disulfide (Table 1). While the stabilization of the NY-ESO-1 TCR by the seven mutant designs versus the CαT166C/CβS173C disulfide appeared to be comparable, the other three TCRs demonstrated superior stabilization using the seven mutant designs (Table 1).
The seven mutant stabilizing Cα/Cβ designs were also evaluated in combination with the CαT166C/CβS173C disulfide. Adding the seven mutant designs to the CαT166C/CβS173C disulfide resulted in a roughly 3-fold increase in titer and an increase in thermal stability by DSC for all four TCRs (Table 1, Fig. 4b). Interestingly, all four TCRs were more compact as assessed by SDS-PAGE analysis and analytical size exclusion chromatography (SEC, Fig. 4c, d).
To investigate further the more compact nature of the TCRs in the presence of the stabilizing Cα/Cβ mutations, we performed both hydrogen deuterium exchange (HDX) analyses and glycan occupancy studies using peptide mass mapping with the WT-1 TCR (CE10) with and without the stabilizing Cα/Cβ designs. HDX patterns of the CE10 TCR with and without the stabilizing Cα/Cβ mutations was nearly identical except for significant regions of the Cα chain (Fig. 4e). HDX protection of the Cα domain upon stabilization was not localized to the sites of mutation, but more globally to various regions of the Cα fold, including protection at the C-terminus. The wild-type and stabilized CE10 TCR was also evaluated for the presence/occupancy of N-linked glycans before and after enzymatic N-linked and O-linked deglycosylation. A single N-linked glycosylation motif exists in the Vα domain, three in the Cα domain, and one in the Cβ domain. Cα/Cβ stabilization led to a significant reduction in N-linked glycosylation within the CE10 TCR including within Vα, suggesting that stabilizing Cα/Cβ also stabilizes the Vα/Vβ domains (Table 2, Fig. 4f). O-linked glycosylation was probable at the C-terminus of Cα since the C-terminal peptide could not be resolved in the TCR lacking the stabilizing designs. A well-resolved C-terminal peptide lacking O-linked glycosylation was identified upon stabilization of the Cα/Cβ subunit. The Cα C-terminal region is variably present in different TCR crystal structures and absent in the NY-ESO-1 structure we used for the design. In the structure of the stabilized Cα/Cβ subunit, the Cα C-terminus is well resolved. Interestingly, the Cβ H139R mutation forms charge-charge interactions as well as a potential π-stacking interaction across the Cα/Cβ interface that may stabilize/lock-down the conformation of the Cα C-terminus, which also shows enhanced protection from HDX. Thus, the universal decrease in hydrodynamic radius observed for all the stabilized TCRs by SDS-PAGE and SEC was confirmed as a reduction in glycosylation.
Next, we assessed the ability of the stabilizing mutations to enable full-length TCRs to withstand variable domain mutation. Using Rosetta, a limited number of mutations in both the Vα and Vβ domains were generated in the 1G4_122 anti-NY-ESO-1 TCR. Ultimately, one (βN66E) was stabilizing, multiple mutations were neutral, and one was significantly destabilizing (βS49A, Fig. 5a). The Vα and Vβ mutations were combined and this small library of TCR variants were expressed in the absence and presence of the stabilizing Cα/Cβ mutations. A wide range of Tms were observed by DSF for the 1G4_122 TCRs in the presence of a wild-type Cα/Cβ subunit, while the stability of the 1G4_122 TCRs was virtually constant when the stabilizing Cα/Cβ subunit was present (Fig. 5b, c). A slightly higher apparent Tm of the non-stabilized anti-NY-ESO-1 TCR containing βN66E versus that of the TCR containing the seven mutant designs is unlikely a real difference, but instead the result of the overlap of the unfolding transitions of the variable and constant domains within the non-stabilized TCR making it difficult to accurately measure the variable domain Tm. As observed with the TCRs described above, the entire library of 1G4_122 variants achieved a roughly 3-fold increase in titer with the Cα/Cβ designs present (Fig. 5d). Interestingly, the Tms of the 1G4_122 variants ranged from 49 to 65 °C, whereas the Cα/Cβ stabilized library all had Tms tightly clustered around 62 °C (Fig. 5e) and were impervious to variable domain instability. These results suggest the stabilized Cα/Cβ subunit may rescue the expression and stability of TCRs with low intrinsic stability.
Producing soluble TCR/CD3 bispecifics
We believe a major benefit to stabilizing α/β TCRs would be the additional bispecific formats that could be enabled. First generation TCR bispecifics combine the specificity of soluble α/β TCRs with an anti-CD3 scFv in the ImmTac format5. This format enables exquisite redirected lysis potency; however, the major drawbacks from our perspective are what we perceive to be low yield bacterial expression and refolding as well as rapid serum clearance that makes in vivo dosing challenging. With the significant increase in expression and proper folding of α/β TCRs in mammalian cells afforded by the stabilizing the Cα/Cβ subunit, we believe additional bispecific formats should be accessible that include moieties with enhanced pharmacokinetics (PK) as well as avidity to the tumor cells.
We generated both IgG-like and Tandem bispecific antibodies (BsAbs, Fig. 6a). The BsAbs utilized the 1G4_122 α/β TCR that binds the HLA-A2/NY-ESO-1 peptide complex and was engineered to bind with high affinity (1 nM)21. The IgG format requires the expression of two heavy chains that require an antibody Fc heterodimerization design to achieve proper HC assembly. The previously published 7.8.60 heterodimer design was chosen29. We expressed the IgG BsAb with a wild-type or stabilized Cα/Cβ subunit. As we observed with the soluble TCRs, the IgG BsAb containing a stabilized Cα/Cβ subunit expressed over two-fold higher than the construct with a wild-type Cα/Cβ subunit. The IgG BsAb protein lacking the stabilizing mutations was found to be a mixture of TCR/CD3 BsAb and anti-CD3 half-antibody (Fig. 6b). The anti-CD3 half/antibody was observable as a low-molecular weight (LMW) peak by analytical SEC after IgG-Fc affinity purification (Fig. 6b). Both the IgG and Tandem BsAbs with a stabilized Cα/Cβ subunit were secreted as >90% monodisperse (i.e., pure) proteins coming directly from the cell culture based on their profile after a single affinity chromatography step (Fig. 6b). The correct ratio of each of the chains and correct molecular weight of the fully assembled BsAbs was confirmed using SDS-PAGE (Fig. 6c). The purity of the stabilized TCR/CD3 BsAbs directly off an affinity chromatography step is remarkable, even for antibody-based bispecifics30. Both the improved expression and optimal purity would have a big impact on translatability to the clinic and for commercializability of the platform since the cost of biologics manufacturing is high and the removal of product-related impurities can be challenging and result in significantly reduced overall product yields.
The BsAbs were next evaluated for their function. Flow cytometry was performed with Jurkat cells to assess CD3 binding and with 624.38 and Saos-2 tumor cells to assess peptide/MHC binding. Both the 624.38 melanoma line and the Saos-2 sarcoma line have been characterized to express HLA-A231. The TCR/CD3 bispecifics demonstrated insignificant binding in the absence of endogenously added NY-ESO-1 peptide (SLLMWITQC), but strong binding to these cells once the NY-ESO-1 peptide was added to the tumor cells (Fig. 7a). Antigen peptide pulsing is often required induce recombinant T cell-mediated killing of tumor cells in vitro31,32. After adding synthetic NY-ESO-1 peptide to the tumor cells, both the IgG and Tandem BsAbs bound both the 624.38 and Saos-2 lines (Fig. 7a). The Tandem BsAb appeared to bind slightly better to the two separate HLA-A2+ tumor cells compared to the IgG BsAb. Both BsAbs showed a 10- to 30-fold decrease in binding potency to Jurkat cells compared to the bivalent anti-CD3 mAb due to a loss of avidity. This is a commonly observed phenomenon when converting 2-arm binding mAbs to a 1-arm binding mAb or Fab. Given the weaker binding to tumor cells observed for the IgG BsAb, a couple higher potency TCR21 versions of the IgG BsAb were generated to offset the potency loss (Supplementary Fig. S4a). The 1G4_107 TCR variant demonstrated improved potency binding to tumor cells over the 1G4_122 variant by flow cytometry (Supplementary Fig. S4b).
Next the BsAbs were assessed for their ability to redirect T cells to kill the HLA-A2+ tumor cell lines. The BsAbs were titrated onto the 624.38 and Saos-2 tumor cells in the absence and presence of the synthetic NY-ESO-1 peptide. Next, unstimulated primary T cells were added to the cells. Neither BsAb induced T cell redirected lysis in the absence of the NY-ESO-1 peptide. This was not entirely unexpected as primary T cells expressing recombinant TCR targeting NY-ESO-1 have been shown to require peptide pulsing to induce redirected lysis31. Upon addition of the peptide, both BsAbs demonstrated the ability to redirect primary T cells to kill the tumor cells. The TCR/CD3 IgG BsAb demonstrated modest potency (~1 nM EC50) and approximately 50% killing on Saos-2 cells and showed no activity on 624.38 cells over the 48 h of the assay (Fig. 7b). Increasing the affinity of the TCR/CD3 IgG BsAb using a known high affinity variant21 did not increase its potency or saturated level of killing (Supplementary Fig. S4c). The tandem BsAb format, however, was significantly more potent (~0.03 and 0.1 nM EC50s) and effective (approximately 90% and 40% killing) on Saos-2 and 624.38 cells, respectively, over the 48 h of the assay. (Fig. 7b).
To determine whether the tandem BsAb can redirect T cells to kill HLA-A2+ tumor cells without pulsing the NY-ESO-1 peptide exogenously, we evaluated its ability to kill A375 HLA-A2+/NY-ESO-1+ melanoma cells since ImmTacs have been shown to kill this cell line in the absence of pulsed peptide33. The IgG-like BsAb was not evaluated since it demonstrated inferior T cell redirected lysis when pulsing peptide onto HLA-A2+ tumor cells. The affinity (KD) of the 1G4_122 α/β TCR variant in the tandem BsAb was 1 nM21, which is much lower than the affinity of the ImmTAC-NYE shown to kill tumor cells by McCormack and coworkers, which has a reported affinity of 50 pM33. Therefore, we produced a second tandem BsAb using the 1G4_113 α/β TCR variant with 26 pM affinity to the NY-ESO-1/HLA-A2 complex21. Both the 1 nM and 26 pM 1G4/anti-CD3 tandem BsAbs were titrated onto A375 melanoma cells in the presence of expanded primary T cells from two separate donors. Both tandem Fabs could potently kill the A375 cells expressing endogenous NY-ESO-1 in the absence of pulsed peptide (Fig. 7c). As observed with the IgG-like BsAbs in Supplementary Fig. S4b, both tandem Fabs showed similar redirected T cells lysis potency on A375 cells regardless of their affinity differences (Fig. 7c). Thus, the tandem BsAb bifunctionals can potently redirect T cells to kill tumor cell lines endogenously displaying the NY-ESO-1 peptide on HLA-A2.
Based on the data presented here, the stabilized Cα/Cβ subunit will likely improve the expression and stability of most recombinant TCRs including those with low stability variable domains. Compared to traditional bacterial methods of production in inclusion bodies followed by solubilization, refolding, and purification7, the mammalian expression/secretion of fully assembled/oxidized TCRs described here should dramatically increase their ease of production. The stabilized Cα/Cβ subunit led to antibody-like expression levels for TCRs regardless of the presence of the known stabilizing disulfide bond14—an achievement not provided by the disulfide bond in isolation. This increased expression enables the robust production of various TCR/antibody bifunctional molecules. An interesting possible application not evaluated in this report would be to include the stabilized Cα/Cβ subunit within full-length TCRs transgenically induced within primary T cells. Difficulty with chain association and expression have seen minor improvements via the introduction of the disulfide bond, the order of expression of the α/β chains, or introduction of hydrophobics in the transmembrane region34,35. It would be interesting to determine if the stabilized Cα/Cβ subunit described here would generally improve cell-surface TCR expression for cell based-therapy approaches.
The stabilized TCRs also displayed reduced overall glycosylation (both variable and constant domains) compared to their non-stabilized counterparts. Glycosylation has been known to stabilize and solubilize many different proteins including antibody Fc and, in specific cases, antibody Fab moieties36. Glycoengineering has even been used as an approach to stabilize proteins, reduce their propensity to aggregate/misfold, or improve their solubility36. The loss of glycosylation of Cα was shown to dramatically reduce the expression of IgG-like proteins containing a Cα/Cβ subunit37. Thus, we were surprised to find that the stabilizing designs significantly reduced the N-linked glycosylation by roughly 50% per site throughout the entire TCR. Putative O-linked glycosylation at the C-terminus of Cα was also absent in the stabilized form. We hypothesize that stabilizing interactions reduce the conformational fluctuation of these sites within the folding/folded protein, reducing access of the enzymes to decorate the canonical N-linked motifs and O-linked site(s). This includes an N-linked site within Vα, providing further evidence that stabilizing the Cα/Cβ subunit generally stabilizes the entire TCR. Further, full-elimination of the N-linked glycans on Cα via mutation had no impact on protein expression within the stabilized constructs.
Clear geometrical constraints exist for TCR/CD3 bifunctionals’ ability to redirect T cells to lyse tumor cells. We show a substantial difference in overall activity and potency between the IgG and tandem BsAbs described here even though they contained identical soluble TCR and anti-CD3 arms. Such geometrical/size impacts on redirected lysis potency/activity have been demonstrated previously for antibody-based Tumor-Associated Antigen/CD3 BsAbs38,39,40. We are not aware of similar data being shown for TCR-based therapeutics. The flexibility and spacing of the soluble TCR and anti-CD3 Fab subunits appears critical to achieving high potency redirected lysis. Many discussions regarding the size of the antigen and or bispecific construct and their ability to fit within the immune synapse have been raised as a determining factor for their activity41. However, we observed differences between the IgG-like and tandem BsAbs in their ability to bind tumor cells in the absence of the T cells, which complicates the size/immune synapse interpretation. Indeed, enhancing the affinity of both the IgG and tandem Fab BsAb by 20- to 30-fold did not result in significantly improved potency, perhaps because the geometrical constraints of each format dictate the maximum potency that can be observed once a certain affinity is reached. The individual binding characteristics to each cell type may also play a big role in the case of these TCR/CD3 BsAbs. Overall, we believe the significant advantages afforded by the stabilized Cα/Cβ designs could be valuable not only for soluble TCR-based therapeutics, but could be leveraged to improve recombinant α/β TCR expression in cellular systems including those that use designed TCRs as part of cell based therapies.
NY-ESO-1 1G4_122 soluble TCR, and the individual NY-ESO-1 Vα/Vβ and Cα/Cβ subunits were originally generated using gBlocks (Integrated DNA Technologies, IDT) and subcloned into mammalian expression vectors (Lonza) using recombinase cloning (In-Fusion, Clontech). Sequences for 1G4_122,107, and 113 were derived using the 2F53 crystal structure42 and the CDRs described elsewhere21. A murine kappa light chain leader sequence was used to drive secretion of each protein and 8xHistags were added recombinantly to the N-termini of the α-chains. DNA sequences were derived using IDT optimized codon algorithms.
Single mutant libraries were created using a site-directed mutagenesis protocol. Briefly, the protocol employs the supercoiled double-stranded DNA vector and two synthetic oligonucleotide primers (generated at IDT) containing the desired mutation(s). The oligonucleotide primers, each complementary to opposite strands of the vector, are extended during thermal cycling by the DNA polymerase (HotStar HiFidelity Kit, Qiagen) to generate an entirely new mutated plasmid. Following temperature cycling, the product is treated with Dpn I enzyme to eliminate the methylated, non-mutated plasmid that was prepared using E. coli (New England BioLabs). Each newly generated mutant plasmids is then transformed into Top 10 E. coli competent cells (Life Technologies). Colonies were picked, cultured, miniprepped (Qiagen), and sequence verified. Mutant combinations were generally synthesized as gBlocks.
TCR/CD3 constructs were created using overlapping PCR of each fragment and recombinase cloning. The chimeric SP-34 anti-CD3 heavy and light chain sequences were published previously38.
Protein expression and purification
α/β constructs were co-transfected for transient expression in either 24 well plates (1 mL scale) or individual flasks (100–200 mL scale) as described previously43. Briefly, plasmids were transfected in CHO K1SV KO cells maintained in a DMEM-based medium with 8–12 mM L-Glutamine (LM-Growth, SAFC Cat#59202C-100). Cells were maintained in shake flasks at 37 °C, 6–8% CO2 in an 80% humidified incubator. Transient transfection of plasmids was performed using Polyethylenimine Max (PEI Max from Polysciences Cat# 24765-2). CHO cells were seeded at 1.5e6 cells/mL. DNA (1.6 mg/L coding plasmid + 1.6 mg/L herring sperm DNA) and PEI Max (8 mg/L) were sequentially added to the cells. Cultures were maintained at 32 °C in an incubator with 6–8% CO2 and 80–90% humidity with shaking at 250 rpm on a ¾ inch shaking diameter shaker. Secreted protein was harvested by centrifugation of the cell cultures for 5 min. Supernatants were passed through 2 μM filters for purification44,45. Small scale expressions of TCR and TCR fragments were quantified using Ni2+-NTA tips (Pall Forte Bio) reading on the Octet RED384 System. Purified TCR Cα/Cβ and full-length α/β TCR proteins were used to generate standard curves.
TCR/TCR fragment His-tag proteins were purified using a Ni2+ immobilized resin (cOmpleteTM His-Tag, Millipore Sigma) and an AKTA Pure system (GE Healthcare). The column was equilibrated with at least 5 column volumes of binding buffer (20 mM Tris-HCl pH 8.0, 0.5 M NaCl) at a flow rate of 1 mL/min for 1 mL columns and 5 mL/min for 5 mL columns, respectively. After passing the CHO expression supernatant over the column to capture the His-tagged TCRs or TCR fragments, the TCR proteins were eluted with ten column volumes of elution buffer (20 mM Tris-HCl pH8.0, 0.5 M NaCl, 250 mM Imidazole). Imidazole was later removed by passing the proteins through a preparative size exclusion column (SRT-10C SEC300) or via dialysis against a phosphate buffered saline solution (PBS) at pH 7.2–7.4 using VIVASPIN 6 concentrators with a 10,000 MW cutoff. IgG-Fc containing proteins were purified using mAbSure resin (GE Healthcare) and preparative size exclusion. Sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) was performed using 4–12% Bis Tris gels according to the manufacturer (Life Technologies). Reductions and alkylations were performed with 1 mM dithiothreitol (DTT) and 1 mM N-ethylmaleimide (NEM).
Differential scanning fluorimetry and calorimetry
Protein unfolding was monitored on the Lightcycler 480II RT PCR machine (Roche) using SYPRO Orange dye (Invitrogen, S6651). Excitation and emission filters were set at 465 nm and 580 nm, respectively. The temperature was ramped from 25 °C to 95 °C at a rate of 1 °C/s. Stock solutions of protein were stored in running buffer (1 X PBS pH 7.4).
Final assay conditions were 0.2 mg/mL protein and a 500-fold Sypro dilution in running buffer. The experiment was performed by diluting the proteins to 0.4 mg/ml and diluting Sypro 250-fold, then combining the diluted protein and Sypro in equal parts in a 96-well plate (final volume: 30 μL per well). The samples were divided into 6 μL triplicates on a 384-multiwell assay plate (Roche, 04-729-749-001).
The midpoint of thermal unfolding (Tm) determination for each protein was performed on the LightCycler Thermal Shift Analysis Software (Roche) using the first derivative method. The analysis software smoothed the raw fluorescence data and the Tm was collected by determining the temperature where the upward slope of fluorescence vs. temperature was the steepest (i.e., temperature where first derivative of melt curve was maximal).
For DSC, thermal unfolding profiles were measured using an automated capillary DSC (capDSC, MicroCal/Malvern). Scans were performed at 1.5 °C/min using the low feedback mode. Protein concentrations were 1 mg/mL. The scan range was 20–110 °C. All 96-well plates containing protein were stored within the instrument at 6 °C. Unfolding transition midpoints (Tms) were determined by fitting the curves using the non-two-state model with the Origin 7 software provided by the manufacturer.
Rosetta energy predictions
We used the TCR crystal structure with PDB code 2F53 and removed chains A-C, leaving only the alpha and beta chains. The remaining structure was relaxed using the FastRelax protocol in Rosetta with coordinate constraints (energy penalties that grow as a backbone atom deviates from its starting position) on every residue’s CA atom46. All possible mutations (except for mutations to cysteine) were also simulated using the FastRelax protocol. FastRelax consists of two sub-protocols: (1) fixed-backbone rotamer substitutions (i.e., sidechain conformation sampling) and (2) all-atom minimization of backbone and sidechain torsion angles. The first sub-protocol generates a rotamer library for each residue position and stochastically samples rotamers at user-allowed positions. The lowest-energy combination of rotamers is assigned to the protein after this sampling is repeated many times (hundreds of thousands). The second sub-protocol uses gradient-based minimization of torsion angles to relax the structure into its local minimum in the Rosetta energy landscape. We used the energy function REF2015 for this study47,48. Certain considerations were made to prevent noise in the Rosetta energy predictions. For FastRelax’s first sub-protocol, we only allowed residues within 10 Å of the mutation to sample alternate sidechain conformations. We added coordinate constraints to all residue positions in the second sub-protocol to prevent the backbone from varying much from the starting structure. After FastRelax completed, we allowed one final run of all-atom minimization without coordinate constraints; allowing the protein to relax into its local, unconstrainted energy minimum. Our Rosetta script is provided in the supplementary information. We ran ten independent trajectories for every possible mutation and used the average score of the three lowest-scoring trajectories as the representative score for that mutation. The version identifier (git sha1) of our Rosetta version was 360d0b3a2bc3d9e08489bb9c292d85681bbc0cbd, released on 4 June 2017.
Multiple sequence alignment
We assembled a list of 39 CA and 53 CB sequences by hand-curating a MSA of TCR sequences from various organisms. We assembled two lists of TCR-similar sequences by passing the sequences of the CA and CB chains of the 2F53 structure into BLAST using NCBI’s database of non-redundant protein sequences (169,859,411 sequences in database), allowing only sequences from animals. Both lists were purged by hand of any sequences that did not appear to be a TCR either because of poor alignment or because of the sequence’s label. We partitioned the list of mutations deemed stabilizing by Rosetta into two lists: (1) mutations present in the MSA and (2) mutations absent from the MSA. We selected the 30 mutations from list 1 and 25 mutations from list 2 for experimental testing. Selection decisions were made primarily by Rosetta score but some candidates were removed for the sake of increasing design diversity. To aid the decision process we made use of a figure that displayed two computational metrics for all possible single point mutants in the constant domains (Supplementary Fig. S1): the calculated Rosetta score for the mutant and a log odds score based on the mutation representation in the MSA. By splitting the mutations into two lists, we could promote mutations that were deemed lower-tier by Rosetta if they have been shown to be compatible with any T-Cell Receptor structure.
Modeling Cα/Cβ mutant combinations
After the best performing point mutations were determined experimentally, we ran Rosetta simulations to verify that each mutation was compatible with every other mutation. We separately applied each mutation to the pre-relaxed 2F53 structure and ran fixed-backbone sidechain repacking. We performed the same simulations with every pair of mutations and ensured that the Rosetta energy of the pair was not dissimilar to the sum of the energies of the mutations individually.
Purified protein was crystallized at 8 °C using the sitting-drop vapor-diffusion method. Crystals were obtained by mixing one part protein solution at 15 mg/mL (10 mM Tris-HCl pH 7.5, and 150 mM sodium chloride) with one part reservoir solution containing 23% PEG 4 K, 300 mM magnesium sulfate, and 10% glycerol. Drops were immediately seeded with crushed crystals grown from 15% PEG 3350, and 100 mM magnesium formate. Single, plate-shaped crystals were observed within one week. These were harvested into a drop of reservoir solution containing 15% glycerol, and flash frozen in liquid nitrogen.
X-ray data collection and structure determination
Synchrotron X-ray diffraction data were collected on a single crystal at the Advanced Photon Source on beamline 31-ID-D (LRL-CAT). The resulting data were integrated using autoPROC49, and merged and scaled with SCALA50. The structure was solved by Molecular Replacement with Phaser51, using the TCR constant domains of a previously solved TCR-pMHC complex (Protein Databank accession code 2F53) as the search model42. Several rounds of model building and refinement were conducted with COOT52 and REFMAC553, as implemented in the CCP4 software package v. 7.054. The final model was validated with MolProbity55. Additional data collection and refinement details can be found in Supplementary Table S2 of the Supplementary Information.
The wild-type and stabilized CE10 TCRs were denatured, reduced, cysteine-alkylated, and digested for the analyses. Briefly, 100 μg of TCR variant was denatured in 6 M guanidine prior to reduction with 10 mM dithiothreitol at 37 °C for 30 min. Next, 15 mM iodoacetamide was added and the reaction was allowed to proceed for 45 min in the dark. Each sample was then buffer-exchanged into 10 mM TRIS, pH 7.5 using spin filters. 5 μg of trypsin was added to the reduced/alkylated sample and incubated for 4 h at 37 °C. The trypsinized samples were divided into two equal parts. The first sample was deglycosylated by adding 1 μL PNGaseF (ProZyme) and 5 μL 10x digestion buffer and incubating overnight at room temperature, while the second sample was left untreated.
To determine the glycan occupancy, an aliquot of PNGase F treated and untreated tryptic digest was injected onto an Agilent 1290 UPLC coupled to an Agilent 5210 QTOF Mass Spectrometer for peptide mapping and mass alignment. Briefly, 5 μg TCR digest was injected on to a Waters 150 mm × 2.1 mm 1.7 μm C18 CSH UPLC column and gradient separated from 2% B (0.1% formic acid in acetonitrile) to 35% B over 30 min. The eluting peptides were injected into the mass spectrometer for mass analysis with a scan range from 200 to 2000 m/z, source temperature of 350 °C, capillary voltage of 4 kV, fragmentor at 250 V, capillary exit at 75 V, source gas at 40 PSI, and cone gas at 12 PSI. The peptide mass spectra was aligned to the protein sequence by Mass Hunter/Bioconfirm 7.0 software with a mass accuracy of 5 ppm resulting in over 90% sequence coverage. To measure glycan occupancy, deamidation of Asn was added as a variable modification and each predicted glycan was checked manually by extracted ion chromatogram (EIC) with 5 ppm mass accuracy. The ratio of Asn to Asp was calculated using the EIC and reported as percent glycan occupancy: Asn being unoccupied and Asp being occupied since the enzymatic deglycosylation converts Asn to Asp.
Hydrogen/deuterium exchange coupled with mass spectrometry
HDX-MS experiments were performed on a Waters nanoACQUITY system with HDX technology56, including a LEAP HDX robotic liquid handling system. The deuterium exchange experiment was initiated by adding 55 μL of D2O buffer containing 0.1x PBS to 5 μl of each protein (concentration was 25 μM) at 15 ˚C for various amounts of time (0 s, 10 s, 1 min, 10 min, 60 min, and 240 min). The reaction was quenched using equal volume of was 0.32 M TCEP, 0.1 M phosphate pH 2.5 for 2 min at 1 ˚C. 50 µL of the quenched reaction was injected on to an on-line pepsin column (Waters BEH Enzymate) at 14˚C, using 0.2% formic acid in water as the mobile phase at a flow rate of 100 µL/ min for 4 min. The resulting peptic peptides were then separated on a C18 column (Waters, Acquity UPLC BEH C18, 1.7 μm, 1.0 mm × 50 mm) fit with a Vanguard trap column using a 3–85% acetonitrile (containing 0.2% formic acid) gradient over 10 min at a flow rate of 50 μL/min. The separated peptides were directed into a Waters Xevo G2 time-of-flight (qTOF) mass spectrometer. The mass spectrometer was set to collect data in the MSE, ESI+ mode; in a mass acquisition range of m/z 255.00–1950.00; with a scan time of 0.5 s. The Xevo G2 was calibrated with Glu-fibrinopeptide prior to use. All acquired data was mass corrected using a 2 µg/ml solution of LeuEnk in 50% ACN, 50% H2O and 0.1% FA at a flow rate of 5 µl/min every 30 s (m/z of 556.2771). The peptides were identified by Waters Protein Lynx Global Server 3.02. The processing parameters were set to low energy threshold at 100.0 counts, an elevated energy threshold at 50.0 counts and an intensity threshold at 1500.0 counts. The resulting peptide list was imported to Waters DynamX 3.0 software, with threshold of 5 ppm mass, 20% fragments ions per peptide based on peptide length. The relative deuterium incorporation for each peptide was determined by processing the MS data for deuterated samples along with the non-deuterated control in DynamX 3.0 (Waters Corporation).
T cell redirected lysis assays
Saos-2 and A375 cells were from ATCC (Cat#HTB-85 and CRL-1619, respectively) and the 624.38 cells were from the NCI/NIH DTP, DCTD Tumor Repository57. All three lines were cultured in Complete Media (RPMI 1640 /10% fetal bovine serum (FBS) Corning/gentamicin—Gibco). Primary naïve T cells were from StemExpress (Cat#PB03020C, Lot#1804170175; Donor#D001003581). For the assay, the tumor cells were removed from their culture flask using Accutase (Innovative Technologies) and spun down for 10 min at 170g. The tumor cells were resuspended in Complete Media and seeded at 5000 cells/well in 96-well black, clear bottom plates (Perkin Elmer) and SLLMWITQC (NY-ESO-1) peptide (synthesized by CPC Scientific), 100 μL total volume, for 3 h. After 3 h, 50 μL of the peptide solution was removed from each well. TCR/CD3 bifunctionals (50 μL/well at 4×) or control antibodies (in-house, recombinantly-produced anti-CD3 chimeric SP3438 and irrelevant anti-CD3 IgG158) were then titrated onto the cells using 1:10 serial dilutions starting at 200 nM (2×) and ending at 0.002 nM (2×) in triplicate for 30 min. During this period, primary T cells were thawed and washed twice in Complete Media and gentamicin. T cells (100 μL) were then added at 50 K cells/well (200 μL total volume). After the T cells were added the TCR/CD3 bifunctionals were 1×(starting at 100 nM and ending at 0.001 nM) and the peptide was at 1.5 μg/ml for the duration of the experiment. Non-peptides plates were treated the same as those with peptide except for the absence of peptide in the procedure. Cells were incubated for 48 h at 5% CO2 and 37 °C. At 48 h, the plates were washed gently (twice) with serum free RPMI 1640. For assays with the A375 cells, primary T cells (Donor 1: StemCell Cat#70024, Lot#190981591C, Donor 110044380; and Donor 2: StemExpress Cat#PB03020C, Donor 1804170165) were thawed and resuspended in Complete Media (RPMI 1640 /10% FBS Corning/gentamicin—Gibco) containing anti-CD28 (BD Biosciences 555725, CD28.27, 2.5 μg/mL) and IL-2 (R&D systems 202-IL/CF, 2 ng/mL) and expanded by being placed in flasks precoated overnight with anti-CD3 (BD Biosciences Cat#555329, UCHT1, 5 μg/mL). T cell expansion was allowed to proceed between 4 and 14 days before use. At day 7, T cells were re-stimulated by transfer into a new flask precoated with UCHT1 and containing fresh Complete Media supplemented with anti-CD28 and IL-2. 100 K expanded T cells/well were washed 2 times in Complete Media and added to each well with A375 tumor cells. Prior to the addition of the T cells, the tandem BsAbs were titrated into the plates with tumor cells starting at 500 nM and ending at 10 pM. The T cell/tumor cell/BsAb (or MAb) mixtures were allowed to incubate for 24 h, then the plates were washed gently (twice) with serum free RPMI 1640. Finally, to determine the amount of tumor cells alive at the end of the incubations, 100 μL RPMI 1640 and 100 μL Cell Titer Glo reagent (Promega) were added, mixed for 2 min on a shaker (slowly) and incubated for 10 min in the dark. Lastly, the luminescence was read on an EnVision 2130 multilabel reader.
Cell culture was performed as described above. The day before running flow cytometry, T25 flasks were seeded with Saos-2 or 624.38 cells in Growth buffer (RPMI 1640 /10% fetal bovine serum (FBS) Corning/gentamicin—Gibco). The next morning, the Saos-2 and 624.38 tumor cells were washed once with Growth buffer. Peptide was added to the tumor cells (or not as a negative control) at 6 μg/mL for 3 h at 37 °C, 5% CO2. Next, excess peptide was aspirated off and the cells were washed 3X with PBS buffer. The tumor cells were lifted from the T25 flasks using Accutase (Innovative Technologies Cat#AT104). All subsequent steps were performed on ice. The Jurkat cells grow in suspension (ATCC Cat#TIB-152). Cells were transferred to centrifuge tubes and pelleted by centrifugation at 170g for 7 min. Henceforth, “Wash buffer” was PBS/2% FBS/0.05% NaN3 /10% Normal Goat Serum with extra 10% FBS. “Blocking buffer” was Wash buffer supplemented with Human BD Fc Block (BD Biosciences Cat#564220). The cells were resuspended in Blocking buffer for 15 min, pelleted, Washed 3×, and resuspended in Wash buffer before adding 50 μL of the cells (0.2 × 106 cells/well for tumor cells and 0.5 × 106 cells/well for Jurkat cells) to 96-well plates (Corning 3799). For the tumor cells, the TCR/CD3 bifunctionals and an anti-CD3 mAb (in-house, recombinantly-produced anti-CD3 chimeric SP34 hIgG1_N297Q38) were added to the wells at 30, 3, 0.3, and 0.03 μg/mL and incubated 45 min. For Jurkat cells, the TCR/CD3 bifunctional was titrated starting at 30 μg/mL using 3-fold dilutions. The cells were pelleted, washed, and the supernatants were aspirated again before adding 100 μL R-Phycoerythrin-conjugated Goat anti-Human Lambda (1:500 dilution, Southern Biotechnology Cat#2070-09, lot#B4419-P219) in Wash buffer for 45 min. The cells were pelleted and washed again. Finally, the cells were resuspended in Wash buffer with 1:1000 PI (Molecular Probes Cat#P3566) and covered with foil. The cells were then acquired on a Becton Dickinson LSRFortessa flow cytometer with BD FACSDiva software (v8.0.1) and data were analyzed using FlowJo version 10.5.3. The gating strategies for distinguishing live versus dead cells and single cells versus doublets/multiplets that were used for Jurkat, 624.38, and Saos-2 cells are provided in Supplementary Fig. S5-S7, respectively.
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source data are available in the Source Data file. The coordinates for the TCR constant domain structure, and the corresponding structure factors, have been deposited in the Protein Data Bank (http://www.rcsb.org) under accession code 6U07 [http://www.rcsb.org/structure/6U07]
All other relevant data are available upon reasonable request.
Custom Scripts for running computational single residue mutations in Rosetta are provided in the Supplementary Information as well as in the Rosetta Commons: rosetta_scripts_scripts/scripts/public/point_mutant_scan/point_mutant_scan.Froning_et_al.xml. To run the scripts, users must have access to Rosetta software [https://www.rosettacommons.org/software/license-and-download]. The scripts can also be found at [https://gist.github.com/JackMaguire/6f33119ff9cc4e46e16dc860cdf306e2
Davis, M. M. & Bjorkman, P. J. T-cell antigen receptor genes and T-cell recognition. Nature 334, 395–402 (1988).
Heemels, M. T. & Ploegh, H. Generation, translocation, and presentation of MHC class I-restricted peptides. Annu. Rev. Biochem. 64, 463–491 (1995).
Riley, R. S., June, C. H., Langer, R. & Mitchell, M. J. Delivery technologies for cancer immunotherapy. Nat. Rev. Drug Discov. 18, 175–196 (2019).
Parkhurst, M. et al. Isolation of T-cell receptors specifically reactive with mutated tumor-associated antigens from tumor-infiltrating lymphocytes based on CD137 expression. Clin. Cancer Res. 23, 2491–2505 (2017).
Liddy, N. et al. Monoclonal TCR-redirected tumor cell killing. Nat. Med. 18, 980–987 (2012).
Baeuerle, P. A. & Reinhardt, C. Bispecific T-cell engaging antibodies for cancer therapy. Cancer Res. 69, 4941–4944 (2009).
Wilson, I. A. & Garcia, K. C. T-cell receptor structure and TCR complexes. Curr. Opin. Struct. Biol. 7, 839–848 (1997).
van Boxel, G. I. et al. Some lessons from the systematic production and structural analysis of soluble (alpha)(beta) T-cell receptors. J. Immunol. Methods 350, 14–21 (2009).
Stone, J. D., Chervin, A. S., Aggen, D. H. & Kranz, D. M. T cell receptor engineering. Methods Enzymol. 503, 189–222 (2012).
Shukla, A. A., Wolfe, L. S., Mostafa, S. S. & Norman, C. Evolving trends in mAb production processes. Bioeng. Transl. Med. 2, 58–69 (2017).
Feige, M. J. et al. An unfolded CH1 domain controls the assembly and secretion of IgG antibodies. Mol. Cell 34, 569–579 (2009).
Toughiri, R. et al. Comparing domain interactions within antibody Fabs with kappa and lambda light chains. MAbs 8, 1276–1285 (2016).
Feige, M. J., Behnke, J., Mittag, T. & Hendershot, L. M. Dimerization-dependent folding underlies assembly control of the clonotypic alphabetaT cell receptor chains. J. Biol. Chem. 290, 26821–26831 (2015).
Boulter, J. M. et al. Stable, soluble T-cell receptor molecules for crystallization and therapeutics. Protein Eng. 16, 707–711 (2003).
Leaver-Fay, A. et al. ROSETTA3: an object-oriented software suite for the simulation and design of macromolecules. Methods Enzymol. 487, 545–574 (2011).
Kuhlman, B. & Baker, D. Native protein sequences are close to optimal for their structures. Proc. Natl Acad. Sci. USA 97, 10383–10388 (2000).
Alford, R. F. et al. The Rosetta all-atom energy function for macromolecular modeling and design. J. Chem. Theory Comput 13, 3031–3048 (2017).
Potapov, V., Cohen, M. & Schreiber, G. Assessing computational methods for predicting protein stability upon mutation: good on average but not in the details. Protein Eng. Des. Sel. 22, 553–560 (2009).
Magliery, T. J. Protein stability: computation, sequence statistics, and new experimental methods. Curr. Opin. Struct. Biol. 33, 161–168 (2015).
Goldenzweig, A. et al. Automated structure- and sequence-based design of proteins for high bacterial expression and stability. Mol. Cell 63, 337–346 (2016).
Li, Y. et al. Directed evolution of human T-cell receptors with picomolar affinities by phage display. Nat. Biotechnol. 23, 349–354 (2005).
Brinkmann, U., Reiter, Y., Jung, S. H., Lee, B. & Pastan, I. A recombinant immunotoxin containing a disulfide-stabilized Fv fragment. Proc. Natl Acad. Sci. USA 90, 7538–7542 (1993).
De Groot, A. S. & Martin, W. Reducing risk, improving outcomes: bioengineering less immunogenic protein therapeutics. Clin. Immunol. 131, 189–201 (2009).
Matthews, B. W., Nicholson, H. & Becktel, W. J. Enhanced protein thermostability from site-directed mutations that decrease the entropy of unfolding. Proc. Natl Acad. Sci. USA 84, 6663–6667 (1987).
Leaver-Fay, A. et al. Scientific benchmarks for guiding macromolecular energy function improvement. Methods Enzymol. 523, 109–143 (2013).
Raman, M. C. et al. Direct molecular mimicry enables off-target cardiovascular toxicity by an enhanced affinity TCR designed for cancer immunotherapy. Sci. Rep. 6, 18851 (2016).
Shimizu, A. et al. Structure of TCR and antigen complexes at an immunodominant CTL epitope in HIV-1 infection. Sci. Rep. 3, 3097 (2013).
Richman, S. A. et al. Structural features of T cell receptor variable regions that enhance domain stability and enable expression as single-chain ValphaVbeta fragments. Mol. Immunol. 46, 902–916 (2009).
Leaver-Fay, A. et al. Computationally designed bispecific antibodies using negative state repertoires. Structure 24, 641–651 (2016).
Demarest, S. J. & Glaser, S. M. Antibody therapeutics, antibody engineering, and the merits of protein stability. Curr. Opin. Drug Discov. Dev. 11, 675–687 (2008).
Zhao, Y. et al. Primary human lymphocytes transduced with NY-ESO-1 antigen-specific TCR genes recognize and kill diverse human tumor cell lines. J. Immunol. 174, 4415–4423 (2005).
Zhao, Y. et al. High-affinity TCRs generated by phage display provide CD4+ T cells with the ability to recognize and kill tumor cell lines. J. Immunol. 179, 5845–5854 (2007).
McCormack, E. et al. Bi-specific TCR-anti CD3 redirected T-cell targeting of NY-ESO-1- and LAGE-1-positive tumors. Cancer Immunol. Immunother. 62, 773–785 (2013).
Jin, B. Y. et al. Engineered T cells targeting E7 mediate regression of human papillomavirus cancers in a murine model. JCI Insight 3, e99488 (2018).
Scholten, K. B. et al. Promiscuous behavior of HPV16E6 specific T cell receptor beta chains hampers functional expression in TCR transgenic T cells, which can be restored in part by genetic modification. Cell Oncol. 32, 43–56 (2010).
Zhou, Q. & Qiu, H. The mechanistic impact of N-glycosylation on stability, pharmacokinetics, and immunogenicity of therapeutic proteins. J. Pharm. Sci. 108, 1366–1377 (2019).
Wu, X. et al. Protein design of IgG/TCR chimeras for the co-expression of Fab-like moieties within bispecific antibodies. MAbs 7, 364–376 (2015).
Wu, X. et al. Fab-based bispecific antibody formats with robust biophysical properties and biological activity. MAbs 7, 470–482 (2015).
Bluemel, C. et al. Epitope distance to the target cell membrane and antigen size determine the potency of T cell-mediated lysis by BiTE antibodies specific for a large melanoma surface antigen. Cancer Immunol. Immunother. 59, 1197–1209 (2010).
Li, J. et al. Membrane-proximal epitope facilitates efficient T cell synapse formation by anti-FcRH5/CD3 and is a requirement for myeloma cell killing. Cancer Cell 31, 383–395 (2017).
Davis, S. J. & van der Merwe, P. A. The kinetic-segregation model: TCR triggering and beyond. Nat. Immunol. 7, 803–809 (2006).
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).
Rajendra, Y. et al. Transient and stable CHO expression, purification and characterization of novel hetero-dimeric bispecific IgG antibodies. Biotechnol. Prog. 33, 469–477 (2017).
Lewis, S. M. et al. Generation of bispecific IgG antibodies by structure-based design of an orthogonal Fab interface. Nat. Biotechnol. 32, 191–198 (2014).
Froning, K. J. et al. Computational design of a specific heavy chain/kappa light chain interface for expressing fully IgG bispecific antibodies. Protein Sci. 26, 2021–2038 (2017).
Conway, P., Tyka, M. D., DiMaio, F., Konerding, D. E. & Baker, D. Relaxation of backbone bond geometry improves protein energy landscape modeling. Protein Sci. 23, 47–55 (2014).
Park, H. et al. Simultaneous optimization of biomolecular energy functions on features from small molecules and macromolecules. J. Chem. Theory Comput. 12, 6201–6212 (2016).
O’Meara, M. J. et al. Combined covalent-electrostatic model of hydrogen bonding improves structure prediction with Rosetta. J. Chem. Theory Comput. 11, 609–622 (2015).
Vonrhein, C. et al. Data processing and analysis with the autoPROC toolbox. Acta Crystallogr. D: Biol. Crystallogr. 67, 293–302 (2011).
Evans, P. R. An introduction to data reduction: space-group determination, scaling and intensity statistics. Acta Crystallogr. D: Biol. Crystallogr. 67, 282–292 (2011).
McCoy, A. J. et al. Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674 (2007).
Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. D: Biol. Crystallogr. 66, 486–501 (2010).
Murshudov, G. N. et al. REFMAC5 for the refinement of macromolecular crystal structures. Acta Crystallogr. D: Biol. Crystallogr. 67, 355–367 (2011).
Winn, M. D. et al. Overview of the CCP4 suite and current developments. Acta Crystallogr. D: Biol. Crystallogr. 67, 235–242 (2011).
Williams, C. J. et al. MolProbity: More and better reference data for improved all-atom structure validation. Protein Sci. 27, 293–315 (2018).
Wales, T. E., Fadgen, K. E., Gerhardt, G. C. & Engen, J. R. High-speed and high-resolution UPLC separation at zero degrees Celsius. Anal. Chem. 80, 6815–6820 (2008).
Rivoltini, L. et al. Quantitative correlation between HLA class I allele expression and recognition of melanoma cells by antigen-specific cytotoxic T lymphocytes. Cancer Res. 55, 3149–3157 (1995).
Forero-Torres, A. et al. Results of a phase 1 study of AME-133v (LY2469298), an Fc-engineered humanized monoclonal anti-CD20 antibody, in FcgammaRIIIa-genotyped patients with previously treated follicular lymphoma. Clin. Cancer Res. 18, 1395–1403 (2012).
The authors would like to thank Shyama Yallapragada for her help with DNA construction and preparation, Nichole Niemela and Bon Gil Koo for CHO cell transfections, culture, and small scale purification, Michael Bacica for initial LCMS characterization, Carina Torres for flow cytometry discussions, and Andrew Leaver-Fay for helpful discussions. This research used resources of the Advanced Photon Source (APS), a U.S. Department of Energy (DOE) Office of Science User Facility operated for the DOE Office of Science by Argonne National Laboratory under Contract No. DE-AC02-06CH11357. We thank Jordi Benach, John Koss, and Laura Morisco for data collection support at the Lilly Research Laboratories Collaborative Access Team (LRL-CAT), beamline Sector 31, of the APS.
K.F., A.S., F.H., S.C., K.W., A.F., X.W., H.A., E.C., J.F., A.R.H., D.B., M.H., and S.J.D. are employees of Eli Lilly and Company. Eli Lilly and Company and the University of North Carolina at Chapel Hill filed a joint patent application (still unpublished) related to the TCR constant domain mutations and S.J.D., B.K., J.M., and K.F. are listed as inventors.
Peer review information Nature Communications thanks Nathan Lewis, Sarel Fleishman and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Froning, K., Maguire, J., Sereno, A. et al. Computational stabilization of T cell receptors allows pairing with antibodies to form bispecifics. Nat Commun 11, 2330 (2020). https://doi.org/10.1038/s41467-020-16231-7