Antigens were modeled and measured in Pymol using coordinates derived from PDB IDs 7JIC (CD19), 6AMT (HLA-A*02), or 1Z8L (PSMA). For antigens without a complete experimental structure (CEA, ICAM1, MSLN), starting coordinates were derived from Alphafold (22). All structures were relaxed and re-modeled using Rosetta with the ref2015 score function and the Pyrosetta interface (23).

scFvs were designed using flexible (G4S)3-GG linker to connect the VH and VL domains. All third-generation activator CAR constructs contained the CD8a or EGF-like hinges fused to CD28 TM, as well as CD28, 4-1BB, and CD3z ICDs. All blocker receptor constructs contained either the LIR-1 or EGF-like hinges fused to the LIR-1 TM and ICD. Templates used for target protein mRNA synthesis contained 5′ T7 promoter followed by the V kappa 1 signal peptide and codons encoding either HLA, MSLN, ICAM-1, CEA, or PSMA. For FLAG-based blocking assays, a FLAG sequence (DYKDDDDK) was inserted N-terminally to the mature protein sequence. All DNA constructs were assembled using Golden Gate Assembly. DNA templates were either amplified by PCR or linearized by restriction enzyme digest, then mRNA was synthesized using the HiScribe T7 ARCA mRNA kit (New England Biolabs). The in vitro synthesized mRNA was purified using the Monarch RNA Cleanup Kit (New England Biolabs), eluted in 1 mM sodium acetate, and stored at −80°C.

Surface plasmon resonance experiments were performed with a Biacore 3000 instrument using CM5 sensor chips. In all experiments, recombinantly expressed scFvs were immobilized on the sensor chip via standard amine coupling and monovalent antigens were injected as analyte. Experiments were performed at 25°C in 20mM HEPES (pH7.4), 150mM NaCl, and 0.005% surfactant P-20. Injected analyte concentrations spanned 1-1000nM. Data were processed using BiaEvaluation 4.

On the day of transfection, target cells (HeLa or K562, depending on experiment) were counted, washed with 1x PBS and resuspended to 1.1e7vc/mL in SE (for HeLa cells) or SF (for K562 cells) transfection buffer (Lonza). Target antigen mRNA was serially diluted 2-fold across 15 points in SE or SF transfection buffer in 96-well v-bottom plates. Target cells were added to each well containing the mRNA at a concentration of 1.33e7vc/mL. The mRNA/cell mixture was transferred to a 16-well Lonza 4D cuvette and electroporated according to the manufacturer’s protocol established for target cell line. Post-transfection, the cells were immediately placed into MEM growth media containing serum and seeded into rows of 384-well culture plates at a density of 5,000–10,000cells/well, depending on experiment. Remaining transfected cells were seeded into separate 96-well plates for expression testing by flow cytometry. Plates were cultured for >16 hours at 37°C and 5% CO2.

Jurkat-NFAT luciferase cells were counted, washed with 1x PBS and 2e6 viable cells were resuspended in 120ml of R2 buffer (ThermoFisher Scientific) containing 4-8 μg of a 1:1 DNA mixture encoding appropriate activator and blocker receptor constructs. Cells were transfected with the Neon Transfection System (Invitrogen) using the 100 µl format with E2 Buffer (1500 Volts, 10 width, 3 pulses). Cells were cultured overnight at 37°C, 5% CO2 in RPMI containing 20% FBS for co-culture assays the following day. 5,000-10,000 activating and blocker receptor expressing Jurkat-NFAT luciferase cells were combined with 5,000-10,000 transfected target cells expressing a fixed amount of activator antigen as described above in triplicate wells of a 96-well plate. The co-culture plates were incubated at 37°C, 5% CO2 for 6 hours. Twenty microliters of luciferase substrate (BPS Biosciences) was added to each well of the plate. The plate was then incubated at room temp for 15 minutes and read on a Tecan M1000 luminescent plate reader with 100ms integration time/well. Percent inhibition was interpreted as the ratio between luminescence from Jurkat cells co-cultured with target cells treated with the highest mRNA concentration of blocker target, and luminescence from Jurkat cells co-cultured with target cells expressing no blocker target.

Primary human T cells were transduced with two lentiviral constructs encoding the CD19 CAR and MSLN blocker. Cells were grown in G-Rex plates according to the manufacturer’s instructions in X-VIVO 15 media (Lonza Bioscience) supplemented with 1% human serum and 300 IU/mL IL-2. After 10 days, samples were taken from each well, counted, and stained with recombinant CD19 and MSLN to estimate transduction efficiency. Transduced cells were initially enriched using biotinylated MSLN, and subsequent cell enrichment was performed with recombinant MSLN to remove activator-only cells.

Renilla luciferase (Rluc)(+) RFP(+) Raji target cells were used to enable visualization by ImageXpress Micro (Molecular Devices) and terminal luminescence measurements. Rluc(+) RFP(+) Raji targets were co-cultured with Tmod T cells at effective E:T ratios ranging from 0.001-2.7 (0.01-27 actual E:T) for 48 hours. Using Renilla luciferase substrate (Promega), relative luminescence values were captured, and a specific killing percentage was calculated. MetaXpress software was used to calculate RFP(+) surface area between conditions and verify T cell killing (Data not shown).

Approximately 100,000 cells were collected from each transfected group for flow cytometric analysis. The cells were washed three times in cold PBS containing 1% BSA, then incubated with the corresponding primary antibody for 45 minutes on ice using the manufacturers’ recommended staining concentrations. To prepare antigen probes, biotinylated proteins were incubated with PE-conjugated streptavidin at a 4:1 ratio for 20 minutes at room temperature, and quenched using RPMI media containing 2% v/v FBS and diluted to 10 μg/mL using PBS/BSA. The cells were then washed 3 times in PBS/BSA and incubated with secondary antibody (see below) using the manufacturers’ recommended staining concentrations. Cells were then loaded for flow cytometry data acquisition using a BD FACSCanto (BD Biosciences) and FACSDiva software (BD Biosciences). Forward scatter (FSC) vs side scatter (SSC) measurements were used to gate for single cells. Flow cytometric analysis was performed using FlowJo software.

Quantitative analysis of antibody-binding capacity was performed using the QIFIKIT (Dako) according to manufacturer’s instructions. Briefly, during flow cytometry sample preparation, 50 μL of resuspended QIFI bead slurry was incubated with secondary antibody using the same protocol as the sample cells. After 3 washes in PBS/BSA, the QIFI beads were loaded on the FACS Canto, using the same parameters as the sample cells being interrogated. Using FlowJo, median fluorescence intensity (MFI) was calculated for each of 6 species of QIFI beads containing different densities of antibody-binding sites, then converted into antibody-binding capacity using values provided by the manufacturer. A calibration curve was generated from these values, which was then used to convert sample cell MFI values into antibody-binding capacities.

Antibodies used in this study: Mouse anti-huHLA-A*02 (BD Biosciences, Cat# 551230); F(ab’)2-Goat anti-Mouse IgG (H+L) cross-adsorbed secondary antibody, Alexa Fluor 647 (Invitrogen, Cat# A21237); Mouse anti-huMSLN (R&D Systems, Cat# MAB32653); Mouse anti-huICAM-1 (R&D Systems, Cat# MAB720); Mouse anti-FMC63, PE (Acro Biosystems, Cat#FM3-HPY53); biotinylated human mesothelin (296-580) protein, His, Avitag (Acro Biosystems, Cat#MSN-H82E9); PE Streptavidin (BioLegend, Cat# 405245).

CAR-T platforms have generally been more successful when targeting small proteins over large and bulky antigens (15, 24). We hypothesized that the same principles would apply to the inhibitory blocker module used for Tmod; i.e., smaller blocker antigens (B-antigens) would translate to improved inhibitory responses. The first iterations of the Tmod blocker were developed to target the relatively small (55Å axial length; Figure 1B ) extracellular domain of HLA-A*02, which can trigger exceptional inhibitory behavior (8, 21). However, expansion into larger patient populations independent of HLA LOH status requires the ability to engage non-HLA targets (25).

To investigate the effectiveness of a blocker module reprogrammed to target B-antigens of different heights, we compared blockers directed at two non-HLA antigens with the established HLA-A*02 blocker (8). We used an established CD19 CAR with the FMC63 scFv in a 3^rd-generation backbone as the activator (26, 27), paired with three different blocker constructs targeting membrane-distal epitopes of three distinct B-antigens (HLA-A*02, MSLN, and ICAM-1; Figure 1B ). The affinities of the scFvs for their cognate B-antigens ranged from 30 to 200 nM ( Supplementary 1 ). These model B-antigens cover a range of ~100Å axial lengths and were intended to explore the modularity of the Tmod platform regarding B-antigen height.

We measured the effectiveness of each blocker by evaluating the ligand-dependent inhibition of the CD19-mediated CAR activation. The genes for each receptor combination were co-transfected into Jurkat cells expressing an NFAT-luciferase reporter. Jurkat cells were co-cultured with target cell lines expressing a fixed level of CD19 and increasing levels of B-antigen (generated by transfection with increasing amounts of B-antigen synthetic mRNA) to generate dose-inhibition curves. We characterized the level of inhibition with two parameters: IC₅₀ (B-antigen concentration at half-maximal inhibition) and I_max (maximum percentage of inhibition).

As expected, the blocker targeting the HLA-A*02 B-antigen efficiently inhibited Jurkat cell activation in a concentration-dependent manner. Indeed, at the highest HLA-A*02 mRNA concentration, this blocker module inhibited approximately 90% of the CD19 CAR activation signal ( Figure 1C ). In contrast, blockers targeting MSLN and ICAM-1 were less effective at inhibiting the CD19 CAR (40% and 20%, respectively). The relative amount of inhibition was anti-correlated with the height of the blocker target, with the small HLA-A*02 (55Å) outperforming both the larger MSLN (90Å) and ICAM-1 (185Å). As other basic parameters of CAR function (e.g., affinity and expression; Supplementary 1 ) do not explain the decrease in blocking function, these data suggests that the larger blocker targets contribute to an overall receptor/ligand complex height (combined height or C_B; Figure 1A ; Table 1 ) that compromises Tmod function.

The sharp decrease in blocking function associated with larger B-antigens implies that there is an upper limit to combined height (C_B) for an effective blocker. This is analogous to the reported restrictions on the activation synapse (C_A) observed with CARs and TCRs (28), although antigen size is not the sole predictor of T cell antigen response (12, 29). One strategy shown to improve T cell antigen sensitivity is the deliberate targeting of membrane-proximal epitopes (15, 30). Therefore, we considered how the proximity of the B-antigen epitope (E_B) relative to the target cell membrane may also influence blocking function, presumably by altering C_B ( Figure 2A ). However, targeting membrane-proximal epitopes of long or bulky antigens may also limit epitope accessibility, and therefore receptor engagement, by creating the potential for steric interference of the B-antigen distal region with the effector cell membrane. We thus also consider the length of the epitope distal region on the blocker antigen (X_B).

After comparing the length of ICAM-1 to native TCR:pMHC complexes ( Figure 1B ; Supplementary 2 ), we noted that the high X_B values of membrane-proximal ICAM binders may introduce a steric clash with the effector cell membrane. Therefore, we selected the mid-sized MSLN B-antigen from above to explore these variables. We screened 36 anti-MSLN LBDs with the intention to diversify E_B (31) and used MSLN truncations to functionally define and classify the membrane-distal and membrane-proximal binders ( Supplementary 3 ). After conversion into the blocker format, we assessed blocking potencies by the Jurkat activation assay described above. Clustering anti-MSLN blockers by epitope revealed that blockers targeting membrane-proximal MSLN epitopes were generally more effective at blocking the CAR activation signal compared to those targeting membrane-distal epitopes (p< 0.001, Figures 2B, C ). On average, blockers generated from membrane-proximal MSLN binders inhibited over 70% of the maximum activation signal when co-cultured with CD19(+) target cells transfected with MSLN mRNA. This contrasted with the reduced potencies of blockers derived from membrane-distal MSLN binders (I_max ~70% vs 50% inhibition; IC₅₀ ~60 ng vs 250 ng MSLN mRNA) ( Figure 2C ). Although blocker function in these experiments was not normalized for expression level and affinity, the data here imply that the inhibitory mechanism of the blocker has a significant dependence on the epitope distance from the membrane.

As mentioned above, targeting membrane-proximal epitopes of large antigens must consider the extracellular antigen bulk beyond the epitope. Indeed, the full-length MSLN protein extends ~50% further from the target cell membrane than HLA-A*02 ( Figure 1B ), and this additional mass of extracellular domain (X_B, Figure 2A ) may interfere with blocking, regardless of epitope position. To directly interrogate the effect of the excess MSLN extracellular domain (X_B), we selected the most sensitive MSLN blocker from above and repeated the inhibition experiment with a truncated form of MSLN ( Figure 2B ). Although the truncated MSLN variant expressed at comparable levels relative to the full-length construct (data not shown), the selected blocker was more sensitive to the truncated form ( Figure 2D ). Together, these data suggest that, in addition to the height of the B-antigen (T_B), the location of the epitope (E_B) may also influence blocking function ( Figures 1A , 2A ). However, although targeting membrane-proximal epitopes shortens E_B, it also leads to a proportionally larger X_B value that may sterically prevent the formation of the blocker complex.

In the previous sections, we described how antigen height and epitope location may influence formation of the blocker complex and subsequent blocking function. Although targeting membrane-proximal epitopes of small antigens is a valid strategy for antigen selection, this is not always feasible from a therapeutic standpoint. Therefore, we investigated the height of the blocker receptor itself (R_B) and how this parameter affects function. This posed a challenge, as the hinges typically used for CAR design are flexible, complicating the estimation of R_B ( Supplementary 2 ) (10, 32).

To thoroughly define the relationship between receptor length (R_B) and blocker function, we leveraged modular domain repeats that allow for estimation of the distance of the LBD from the T cell membrane surface based on the number of repeats ( Figure 3A ). Epidermal growth factor (EGF)-like domains, characterized by three intradomain disulfide bonds, were particularly attractive due to the high and uniform expression levels when incorporated as hinge subunits ( Supplementary 4 ). Thus, we designed a series of blocker hinges that utilized up to 7 EGF-like domain repeats (extracted from LRP-1; uniprotID: Q07954, aa 4147–4409) to increase the distance between the MSLN ligand-binding domain and the T cell surface in 20-30Å increments ( Figure 3A ; Table 2 ). To characterize the geometric relationship between CAR and blocker, we also incorporated the same rigid hinge series into the CD19 CAR. Blocking efficiencies of each hinge combination were used to generate a 7x7 functional matrix, enabling a systematic evaluation of different activator/blocker geometries.

To compare blocking strength, we utilized a simplified inhibition assay in which we transfected each hinge combination into Jurkat cells and co-cultured these cells with either MSLN(+) or MSLN(-) K562 target cells engineered to overexpress CD19. We then calculated the percent decrease in activation signal associated with MSLN expression. Populating a matrix with these values revealed that short blocker hinges paired with long activator hinges led to significantly improved blocking, whereas short activators paired with long blockers completely abrogated blocker function ( Figure 3B ). Comparing two distinct hinge configurations in a MSLN titration co-culture yielded the full span of blocking profiles ( Figures 3C, D ). Furthermore, structural models suggest that the short CD19 activators sterically prevent the blocker from engaging the MSLN B-antigen (i.e., the blocker complex is too large; Figure 3C ), whereas the hinge combination that best compensated for differences in the target antigens led to the most efficient blocking profile ( Figure 3D ).

However, although blocking was most effective with long activators, extended activators were also associated with a decrease in activation sensitivity ( Supplementary 4 ). This is consistent with reports highlighting that longer CAR hinges can negatively influence T cell activation (14), and these data suggest that the preference for short activator hinges imposes a constraint on blocker function. To confirm this observation in primary T cells, we transduced PBMCs with the two Tmod constructs presented in Figures 3C and D . Consistent with our observations in the Jurkat inhibition assay, PBMCs transduced with the short-hinged activator and long-hinged blocker responded strongly to CD19(+)MSLN(-) target cells, but also killed CD19(+)MSLN(+) target cells – indicating a failure to protect cells expressing the MSLN blocker antigen ( Figure 4A ). In contrast, PBMCs expressing long-hinged activators and short-hinged blockers were less effective at killing the CD19-only target cells, but completely spared the cells expressing both CD19 and MSLN ( Figure 4B ). Together, these data suggest that one approach to achieve compatibility with a suboptimal Tmod antigen pair is to modulate receptor lengths, leveraging a trade-off between selectivity and sensitivity.

In summary, the model CD19/MSLN target system explored above suggests that several geometric parameters (length of the receptor, antigen, and overall complex) are directly related to T cell function. These predictors contextualize the data for the Tmod antigen pairs examined in section 1 (CD19/HLA-A*02, CD19/MSLN, CD19/ICAM-1), and offer a rational explanation for the superior inhibitory performance of smaller B-antigens. Furthermore, the observations suggest a set of design principles to predict and improve blocker performance: (i) the blocker complex should be less than or equal to the activator complex and (ii): the B-antigen epitope must be accessible to the blocker on the cell surface, following engagement of A-Antigen by the activating CAR ( Supplementary 6 ).

We next explored the generalizability of these design principles and engineering strategies when targeting a more extreme height disparity with different activator and blocker antigens. For the A-antigen, we selected the well-documented tumor associated antigen PSMA, for which many CARs targeting the membrane-distal region are available (33, 34). For the B-antigens, we considered HLA-A*02, ICAM-1, and a series of CEACAM5 variants to fully diversify the B-antigen axial length. To control variability between distinct blocker LBDs due to differences in expression, affinity, and epitope accessibility, we developed a blocker directed towards an engineered FLAG tag present on the N-terminus of each B-antigen target ( Figure 5A ) (36). Thus, any differences observed in Jurkat activation should be a direct result of differences in the target antigen construct.

Consistent with our observations presented in Figure 1C , the ability of the FLAG-blocker to inhibit the PSMA activation signal was closely correlated with B-antigen axial length (R=0.9; Figure 5B ). Jurkat cells expressing the PSMA activator and FLAG-directed blocker only inhibited ~30% of the activation signal when co-cultured with PSMA(+)/FLAG-CEA(+) HeLa target cells (relative to co-culture with PSMA(+)/FLAG(-) target cells). In contrast, the same Jurkat cells inhibited over 80% of the maximum activation signal when co-cultured with target cells expressing the smallest FLAG-tagged B-antigens ( Figure 5B ). Differences in B-antigen expression did not explain these differences in function, as each B-antigen expressed at similar, high levels (>500,000 antigen-binding capacity (ABC) quantified by QIFI) ( Supplementary 5A ).

Leveraging this diverse functional behavior in response to antigen height, we then attempted to optimize blocking function for each FLAG-tagged B-antigen with the modular EGF hinges described in section 3. We incorporated a series of modular hinges into both the PSMA activator and FLAG blocker to generate a 4x4 matrix of Jurkat cells expressing both receptors. The Jurkat cells expressing long activators and short blockers were generally more capable of inhibiting the maximum activation signal when co-cultured with PSMA(+)/FLAG(+) HeLa target cell lines, regardless of B-antigen length ( Supplementary 5B ). Indeed, the hinge extensions improved Tmod function with even the largest FLAG-CEA and FLAG-ICAM-1 B-antigens, although the resultant long activator hinge reduced the baseline signal.

To compare these datasets in a single analytical framework we estimated the length of each receptor-ligand complex using a conversion that sums the axial length of the receptor ( Table 2 ) with the axial length of the target antigen ( Table 3 ). After calculating the maximum percentage of inhibition (I_max), we plotted the measured values against the difference in estimated length between the activation and blocking complexes (ΔC_B-A, Supplementary 5C ). Despite the cumulative inaccuracies associated with estimating the length of both receptor and antigen, the data followed a general nonlinear trend with peak inhibition occurring when ΔC_B-A approaches zero. Plotting all the data together revealed an approximate Gaussian distribution, even with limited data points on the lefthand (i.e., activator smaller than blocker) side ( Figure 5C ). Thus, any gross mismatch between activator and blocker complex has the potential to limit Tmod function, but rational receptor design can correct for suboptimal conditions.