Optimizing efficacy and safety of T cell bispecific antibodies: the interdependence of CD3 and tumor antigen binder affinities in FOLR1 and CEACAM5 2 + 1 TCBs

ABSTRACT

T cell bispecific antibodies (TCBs) are an emerging class of cancer therapy that are typically designed for high binding affinity to CD3 and tumor antigen (TA). Using this approach, TCBs have demonstrated significant clinical efficacy, but they have also elicited cytokine release syndrome and off-target on-tumor toxicities. CD3 affinity-attenuation has recently been reported as an approach to maintain efficacy while reducing cytokine release, but the interdependence of CD3 affinity with other factors is often not systematically explored. For this purpose, we generated a series of TCBs comprising CD3 binders with varying affinities and TA binders with either high or low affinities, utilizing FOLR1 and CEACAM5 as tumor targets. The CD3 binders were classified into high, intermediate, low, and very low affine binders based on affinity measurements as well as functionality. Depending on the target, different combinations of binders showed the most advantageous profile of tumor-cell killing while coupled with lower cytokine secretion. For instance, within the FOLR1-TCBs series, CD3^intermed exhibited a favorable profile compared to CD3^high in vitro using cocultures and in vivo using humanized mice. For CEACAM5-TCBs, CD3^low, along with CD3^intermed, showed a favorable profile compared to CD3^high in both in vitro and in vivo settings. Additionally, CD3^low avoided on-target, off-tumor toxicity and reduced cytokine release in transgenic mice. Taken together, reducing cytokine release while maintaining adequate efficacy is possible through CD3 binder affinity attenuation, but optimizing cytokine release profiles by CD3 binder affinity-attenuation is dependent on additional parameters.

Introduction

T cell bispecific antibodies (TCBs) have proven to be a promising class of cancer therapy over the past decade. Substantial resources have been directed toward their development, notably following the success of blinatumomab, a bispecific T cell engager (BiTE).¹ More recent approvals in the hematological malignancies landscape include teclistamab, mosunetuzumab, epcoritamab, glofitamab, talquetamab, elranatamab, odronextamab and linvoseltamab.^2–9 TCBs face several challenges such as on-target, off-tumor toxicity¹⁰ or insufficient T cell infiltration into tumors,¹¹ which appear to represent a more prominent issue for solid tumors than hematological ones.¹² Catumaxomab was the first TCB approved for solid tumors, but its use was restricted to intraperitoneal administration in patients with malignant ascites in epithelial cellular adhesion molecule (EpCAM)-positive carcinomas.¹³ Building on the learnings of earlier generations, the approvals of tebentafusp and tarlatamab have raised hopes of paving the way for subsequent solid tumor-directed TCBs.^14,15

Although TCBs have achieved notable success over the past 15 y, relatively few of the new molecular entities that entered clinical pipelines have been approved. While TCBs demonstrated efficacy in different indications, they often encountered hurdles related to dose-limiting toxicities, such as cytokine release syndrome (CRS).¹⁶ Various strategies to mitigate CRS have been investigated and implemented, including step-up dosing, corticosteroid and tocilizumab administration to reduce the risk of TCB-driven CRS and allow for increased dose levels.^17,18 The preventive use of JAK and mTOR inhibitors was recently suggested as an alternative approach.¹⁹ Alternatively, masking strategies with tumor-dependent activation have been developed to enhance the tumor specificity of TCBs. For example, protease-activated TCBs have been reported to mediate enhanced safety profiles by masking CD3 binders and exclusively cleaving the mask within the tumor.²⁰ Similarly, pH-dependent TCBs have shown the ability to utilize the preferential conditions in the tumor microenvironment (TME), compared to normal tissues, with the goal of avoiding peripheral toxicity.²¹

TCBs are designed to bind concurrently to CD3e on T cells and a tumor antigen (TA), facilitating the formation of an immunological synapse (IS). The stability and maturity of this synapse influences the extent by which the T cell response is orchestrated.²² Faroudi et al. suggested that the threshold for T cell-mediated cytolytic activity differs from that leading to cytokine release, with the former being more sensitive than the latter.²³ A key factor in determining whether T cells are activated toward killing over cytokine secretion is the stability of the IS. To evade the risk of an aggressive T cell response that leads to CRS, one strategy is to lower the IS stability by attenuating the affinity of the anti-CD3 binders of TCBs. This modulation has been shown to favor cytolytic activity over excessive cytokine release, with the aim to improve the safety profile of TCB therapies. Notably, most approved T cell engagers are characterized by high affinity for both CD3 and the target antigen (Supplementary Table S1, reported values and in-house measurements). Nevertheless, recent reports suggest that tumor cell killing with low levels of cytokine release is feasible when reducing CD3 binder affinity.^24–26 Further studies have shown the possibility of decoupling high levels of cytokine release from anti-tumor cytotoxicity.^27–30 This has led to the advancement of clinical candidates adopting this approach, so far primarily against hematological targets.^31,32 In addition to CD3 affinity, Staflin et al. suggested that TA binder affinity determines the efficacy and safety of the anti-HER2/CD3 TCB.³³ However, the functional interplay between the affinities of both the CD3 and TA binders in TCBs and their impact on efficacy, cytokine release, and on-target, off-tumor effects, requires further thorough investigation.

In this study, we aimed to better understand the impact of CD3 affinity, in conjunction with the TA binder affinity, on the overall efficacy-safety relationship of TCBs targeting solid tumor surface antigens. Our investigative strategy involved generating a series of TCBs targeting folate receptor alpha (FOLR1) and carcinoembryonic antigen-related cell adhesion molecule 5 (CEACAM5) in the 2 + 1 “inverted” format that has been applied to cibisatamab, glofitamab, forimtamig, alnuctamab, and others.^34–37 These TCBs bind bivalently to the TA and monovalently to CD3, with the CD3-binding domain positioned inside rather than outside. The bivalent binding to the TA provides the advantage of avidity mediated binding.³⁸ The generated TCBs comprised a set of affinity-attenuated CD3 binders derived from an original SP34-derived parental Clone22, ensuring the same N-terminal binding epitope on CD3e was maintained. For each target, two binders with either high or low affinity were used for FOLR1 and CEACAM5, respectively. CD3 binders were categorized into four groups based on affinity measurement and functionality. Hence, in this study, we conducted an in-depth investigation of a representative binder of each category: CD3^high, CD3^intermed, CD3^low, and CD3^vlow across both tumor targets.

Results

Generating CD3 binders of attenuated affinity

To create antibody variants with reduced binding to CD3e, affinity-attenuated binders were generated using phage display based on the parental SP34-derived CD3e-specific binder Clone22, which has been clinically applied in the context of forimtamig and EGFRvIII-TCB.^36,39 After initial selection and testing for reduced affinity, the most promising variants were further purified and analyzed for their binding kinetics. These included affinity-reduced variants as well as a high affinity variant as a control. Additionally, modifications were applied to remove potential glycosylation sites, and the antibody variants were engineered into monovalent IgG-like constructs and TCBs for further assessment (Supplementary Figure S1). As part of our initial screening of the CD3 binders library, we evaluated binders across a range of affinities. During this process, we identified a functionality loss cutoff at approximately 82.4 nM, beyond which binders demonstrated poor performance in our test systems. The final library of CD3 binders, with varying affinities, was narrowed down to four binders: CD3^high, CD3^intermed, CD3^low, and CD3^vlow (affinities summarized in Table 1; sensorgrams: Supplementary Figure S1D). This categorization was based on surface plasmon resonance (SPR) measurements combined with functional activity assessments and is specific to the SP34-derived binder clones used and aligned with the CD3 affinities of approved TCBs (Supplementary Table S1). These binders were subsequently used in TCB series for two different targets: FOLR1 and CEACAM5.

Table 1.

Association rates (k_a), dissociation rates (k_d) and affinity constants (KD) to recombinant human CD3e-Fc of CD3 binders. Sensorgrams are shown in Supplementary Figure. S1D.

Binder	ka (1/Ms)	kd (1/s)	KD (M)
CD3high	1.11E + 06	1.72E-03	1.56E-09
CD3intermed	8.41E + 05	2.44E-02	2.91E-08
CD3low	8.10E + 05	3.32E-02	4.10E-08
CD3vlow	6.23E + 05	5.13E-02	8.24E-08

CD3^intermed demonstrated a favorable in vitro profile relative to CD3^high in FOLR1-TCBs

First, the FOLR1-TCBs comprising the CD3 binders of varying affinities (CD3^high, CD3^intermed, CD3^low, and CD3^vlow) were investigated in vitro. These TCBs were serially diluted and added to freshly isolated and subsequently frozen human peripheral blood mononuclear cells (PBMCs) from healthy donors. The PBMCs were co-cultured with the FOLR1-positive human ovarian adenocarcinoma cell line, SK-OV-3 at effector-to-target (E:T) ratio of 10:1. The main readouts included anti-tumor TDCC, cytokine secretion, and T-cell activation at 24, 48, and 72 h, respectively.

Despite the affinity attenuation, the TCB comprising the CD3^intermed binder maintained a robust level of anti-tumor TDCC, showing a marginal reduction in overall killing compared to CD3^high in terms of the effective area under the curve (AUCE) across all time points tested (Figure 1(A)). This was associated with more than 4-fold reduction in cytokine secretion under the same conditions (Figure 1(B)). In contrast, CD3^low and CD3^vlow binders could not induce an adequate level of anti-tumor TDCC compared to CD3^intermed. Additionally, CD3^intermed activated T cells equivalently to CD3^high, as evidenced by CD69 upregulation on CD8⁺ T cells and CD4⁺ T cells, as well as Granzyme B intracellular production at saturating concentrations (Figure 1(C-E)). In this instance, CD3^intermed demonstrated a more favorable profile compared to CD3^high, exhibiting a comparable anti-tumor response, similar T cell activation pattern, and reduced cytokine levels.

Figure 1.

CD3^intermed demonstrated a favorable in vitro profile relative to CD3^high in FOLR1-TCBs.

FOLR1-TCBs comprising CD3 binders of varying affinities (CD3^high in black, CD3^intermed in blue, CD3^low in purple, and CD3^vlow in pink) and a FOLR1 binder of high affinity were characterized in vitro.(A) Anti-tumor TDCC assay displayed as time-independent effective area under the curve (AUCE) calculated from three incubation time-points: 24 h, 48 h, and 72 h.⁶³ (B) Cytokine secretion assessment displayed as AUCE calculated from the same three incubation time points (24 h, 48 h, and 72 h), with TNFα used as a representative example of the multiplex analysis. Blue arrow indicates a four-fold reduction in TNFα levels when using CD3^intermed compared to CD3^high. T cell activation assessment shown as the percentage of CD69 upregulation on CD8⁺ (C) and CD4⁺ (D) T cells after 24 h. (E) Granzyme B release shown as the percentage of intracellular release within CD8⁺ T cells after 72 h. Data points presented as filled squares (■) indicate the usage of the high affinity TA binder. TA binder monovalent affinity: 57 nM. Each data point depicts the average of technically duplicated measurements. Data shown are representative of two PBMC donors. The cell line used was SK-OV-3 (antigen binding capacity ~90,000/cell). E:T = 10:1. The maximum concentration of TCBs used in the assays was 5000 pM, with TCBs titrated in a 1:10 serial dilution series. Error bars indicate ± standard deviation (SD). Time-dependent data shown in Supplementary Figure S2.

TA binder affinity additionally influenced cytokine release when using CD3 ^intermed

Under the same co-culture conditions described, CD3 binders of varying affinities were tested in the context of a FOLR1 binder of reduced affinity (FOLR1^low) to evaluate the impact of TA binder affinity. Expectedly, the lower TA affinity led to a substantial reduction in killing potency and an overall decrease in absolute anti-tumor TDCC (Figure 2(A)) and T cell activation for both CD3^high and CD3^intermed binders (Figure 2(C-E)). This was coupled with a more than proportional 10-fold reduction in cytokine levels when using the CD3^intermed compared to CD3^high (Figure 2(B)). This example underscores the importance of considering the interdependence of TA binding affinity on the activity of TCBs with CD3 binders of reduced affinities when selecting the optimal lead TCB.

Figure 2.

TA binder affinity additionally influenced cytokine release when using CD3^intermed.

FOLR1-TCBs comprising CD3 binders of varying affinities (CD3^high in black, CD3^intermed in blue, CD3^low in purple, and CD3^vlow in pink) and a FOLR1 binder of low affinity were characterized in vitro. (A) Anti-tumor TDCC assay displayed as time-independent effective area under the curve (AUCE) calculated from three incubation time-points: 24 h, 48 h, and 72 h.⁶³ Blue arrow indicates a minor fold reduction in cytotoxicity levels when using CD3^intermed compared to CD3^high. (B) Cytokine secretion assessment displayed as AUCE calculated from the same three incubation time points (24 h, 48 h, and 72 h), with TNFα used as a representative example of the multiplex analysis. Blue arrow indicates a ten-fold reduction in TNFα levels when using CD3^intermed compared to CD3^high. T cell activation assessment shown as the percentage of CD69 upregulation on CD8⁺ (C) and CD4⁺ (D) T cells after 24 h. (E) Granzyme B release shown as the percentage of intracellular release within CD8⁺ T cells after 72 h. Data points presented as open squares (□) indicate the usage of the low affinity TA binder. TA binder monovalent affinity: 1000 nM. Each data point depicts the average of technically duplicated measurements. Data shown are representative of two PBMC donors. The cell line used was SK-OV-3 (antigen binding capacity ~90,000/cell). E:T = 10:1. The maximum concentration of TCBs used in the assays was 5000 pM, with TCBs titrated in a 1:10 serial dilution series. Error bars indicate ± standard deviation (SD). Time-dependent data shown in Supplementary Figure S3.

Dynamic interplay between target expression and CD3 binder affinities dictated in vitro response of FOLR1-TCBs

In addition to SK-OV-3, HT-29 a 9-fold lower-FOLR1-expressing cell line was used to characterize the FOLR1-TCBs comprising the CD3 binders of varying affinities (CD3^high, CD3^intermed, CD3^low, and CD3^vlow) in vitro. In terms of TDCC potency, when using the FOLR1^high binder, CD3^high and CD3^intermed achieved low-picomolar EC50 values (0.3 pM and 5 pM, respectively) when tested with SK-OV-3. In the presence of HT-29, the EC50 values were 4 pM and 40 pM, favoring CD3^high. However, this was coupled with an almost diminished tumor necrosis factor alpha (TNFα) secretion when using CD3^intermed (Supplementary Figure S4). This could be a particularly advantageous property for targeting moderate- to high-expressing tumors while sparing low-expressing target tissues. A similar pattern was observed when using the FOLR1^low binder, although an almost diminished TDCC was observed in HT-29 when using CD3^intermed. This was likely due to the collective reduction in TA binder affinity, CD3 affinity, and target expression (Supplementary Figure S5).

The rate of CD3 internalization correlated with CD3 binding-affinity in vitro

One of the key features at the TCR-CD3 complex interface is the internalization of CD3 subunits following TCR engagement.⁴⁰ This internalization is subsequently followed by recycling, which allows for the continued availability of the CD3 subunits. While the effect of overall binding affinity on this phenomenon has been reported,⁴¹ the specific impact of CD3 binding affinity on the rates of internalization and recycling has not yet been studied. Using the high affinity FOLR1-TCBs comprising CD3 binders of varying affinities described above, a co-culture of PBMCs with SK-OV-3 at effector-to-target (E:T) ratio of 10:1, was set-up. The main readout was the decreased CD3⁺ MFI signal detected via flow cytometry at the tested time-points (24 h, 48 h, and 72 h), indicating a reduction in surface CD3. As shown in Supplementary Figure S6, high CD3 affinity can lead to a more potent lowering of the CD3⁺ signal, indicating faster CD3 internalization at lower concentrations, reducing the availability of surface CD3. One explanation for this could be that high affinity binders may induce stronger crosslinking and clustering of CD3, which accelerates receptor internalization. In this experiment, we observed a difference in EC50 between CD3^high and CD3^intermed, with the difference becoming more pronounced over time. Although this could be attributed to the lower initial rate of internalization observed with CD3^intermed, it is important to recognize this as a hypothesis-generating finding and consider its potential as a direction for further research.

CD3 affinity modulates the stability and function of the IS

Faroudi et al. described a dual-threshold phenomenon that occurs at the CD3e of TCR.²³ Namely, a downstream trigger for redirecting T cells for tumor cell killing and another for cytokine secretion. In order to activate both cascades simultaneously, they proposed that a stable IS should be formed. Subsequently, reducing CD3 affinity could possibly lead to an IS stable enough to trigger T cell mediated cytotoxicity, but not to induce elevated cytokine levels. Following this hypothesis, we investigated the IS using CD3 binders of varying affinities coupled with the high affinity FOLR1 binder (Figure 3(A-C)). CD8⁺ T cells were co-cultured with CellTrace Violet (CTV)-stained SK-OV-3 (FOLR1⁺) cells, and FOLR1-TCBs comprising CD3^high, CD3^intermed, and CD3^low were added at a concentration of 10 nM. Forty-five minutes later, synapse formation was analyzed using the Amnis ImageStream imaging flow cytometer, according to the gating strategy described by Boushehri et al.⁴² Data from three donors indicated a decrease in the percentage of synapses involving a single CD8⁺ T cell interacting with one tumor cell (SK-OV-3), which was dependent on CD3 affinity (Figure 3(A-B)). The CD3^vlow only induced the formation of synapses that were similar to the no TCB control and was therefore excluded from further analysis. Within the single CD8⁺-SK-OV-3⁺ synapse population, lytic and stimulatory synapses were identified by gating on Granzyme B⁺ and P-CD3ζ⁺, respectively (Figure 3(B)). Granzyme B is generally associated with an increased propensity for T cell-mediated tumor-cell killing, while P-CD3ζ⁺ is a key component of the TCR involved in cytokine release upon T cell activation. To investigate the stability of lytic and stimulatory synapses, we measured the intensity of F-actin, a cytoskeletal marker whose expression and intensity correlate with IS strength and stability, within the synapse mask (see F-actin channel, cyan area) (Figure 3B). The intensity of F-actin was more reduced in stimulatory synapses compared to lytic synapses when using CD3^intermed compared to CD3^high (Figure 3B-C). These findings indicate that while lytic synapse stability is closely associated with cytotoxicity, stimulatory synapse stability does not predict killing activity. Instead, it appears to influence cytokine secretion, as previously noted by Faroudi et al.²³ Combined with our functional in vitro data, this suggests that the threshold for IS stability, enabling sustained cytotoxicity without excessive cytokine secretion, can be achieved at an intermediate CD3 affinity level when using FOLR1-TCB.

Figure 3.

CD3^intermed induced a greater reduction in the stability of stimulatory synapses compared to lytic synapses.

(A) Frequency of single CD8± SK-OV-3 synapses as percentage of total CD8+ T cells among CD3^high, CD3^intermed, and CD3^low. Error bars indicate ± standard error of the mean (SEM), n = 3. (B) Exemplary imaging flow cytometry data of lytic (GrzmB+) and stimulatory (P-CD3ζ+) CD8± SK-OV-3 synapses among CD3^high and CD3^intermed. The cyan area in the F-Actin image shows the synapse mask, and the values in yellow indicate the F-Actin intensity within the synapse mask for the respective image. Images are representative of one experiment using CD8+ T cells from three different donors. (C) Geometric mean fluorescence intensity (gMFI) of F-Actin within the synapse mask of lytic (GrzmB+) and stimulatory (P-CD3ζ+) synapses among CD3^high and CD3^intermed. Statistical significance was computed using two-way ANOVA: *p < 0.05. Error bars indicate ± standard error of the mean (SEM), n = 3. Data points presented as filled squares (■) indicate the usage of the high affinity TA binder. TA binder monovalent affinity: 57 nM.

CD3^intermed inhibited tumor growth of a patient-derived xenograft model while inducing lower cytokine levels as a FOLR1-TCB in humanized nsg mice

FOLR1-TCBs containing CD3 affinity-attenuated binders were tested in vivo for their anti-tumor efficacy and peripheral cytokine secretion profiles. For that purpose, female fully humanized NSG mice were inoculated with a FOLR1-expressing, breast cancer patient-derived xenograft (PDX) BC004 subcutaneously, and after reaching the desired tumor volume, randomized into different treatment arms (11 mice per group). All therapies were administered intravenously (iv) using a dose of 0.5 mg/kg once per week for four treatment cycles. In addition to that, the vehicle control group received histidine buffer at the same frequency. On one hand, FOLR1-TCB comprising CD3^high and CD3^intermed efficiently inhibited tumor growth, whereas CD3^low and CD3^vlow were deemed inefficacious at this dose (Figure 4(B), single-animal data; Supplementary Figure S7). On the other hand, CD3^intermed induced substantially lower levels of peripheral cytokines than CD3^high at all measured time points following the first treatment (Figure 4(C-E)). Furthermore, three mice per group (scouts) were euthanized 3 d after the second infusion (D45) to analyze intra-tumoral infiltration of T cells. Interestingly, the extent of T cell infiltration (Figure 4(F-H)) as well as their activation (Figure 4(I-J)) were similar across the CD3^intermed and CD3^high groups, as indicated also by the upregulation of the co-stimulatory marker CD69 on both CD4 and CD8 T cells.

Figure 4.

CD3^intermed inhibited tumor growth of a patient-derived xenograft model while inducing lower cytokine levels as a FOLR1-TCB in humanized nsg mice.

(A) Experimental design of the in vivo efficacy experiment. CD34+ humanized NSG mice were inoculated with a FOLR1-expressing, human, patient-derived breast cancer xenograft (BC004) prior to treatment administration. Each group consisted of 11 mice, receiving either the vehicle (histidine buffer) as a control or FOLR1-TCBs with CD3 binders of varying affinities, administered intravenously. All treated mice received the same dose 0.5 mg/kg once per week for four treatment cycles. Scouts (n = 3/group) were euthanized 72 h after the second therapy for analysis. (B) Tumor volumes measured by a caliper three times per week throughout the experiment to assess the efficacy of the therapies in inhibiting tumor growth. Dashed lines indicate experiment days on which treatments were administered. Statistical significance was determined using an unpaired, 2-tailed t test with Welch’s correction: *p < 0.05. Blood samples were collected at specified time points after the first therapy (1 h, 4 h, 48 h, and 96 h) and 1 h after the second therapy (3 mice bled per time point) for cytokine release assessment. IP-10 (C), MIG (D), and IL-6 (E) are representative examples from the multiplex analysis. (F-J) Intratumoral T cell infiltration and activation were assessed in scouts 72 h after the second therapy. Three mice were euthanized, and their tumors were processed and analyzed via flow cytometry to determine the total count of T cells, CD8, CD4, and CD69 upregulation on both cell types. Significance was computed using one-way ANOVA with Tukey’s multiple comparison test: *p < 0.05. Data points presented as filled squares (■) indicate the usage of the high affinity TA binder. TA binder monovalent affinity: 57 nM. Error bars indicate ± standard error of the mean (SEM).

These results confirmed the favorable profile that CD3^intermed binder demonstrated in vitro, showing significantly comparable levels of anti-tumor efficacy to CD3^high with substantially lower levels of peripheral cytokine release. It is important to highlight that, although the difference in affinity between these CD3^intermed and CD3^low binders is not large (29.1 nM and 41 nM respectively), the latter failed to achieve anti-tumor efficacy in this experiment. This underscores the necessity of considering functional activity in addition to affinity values when evaluating and categorizing CD3 binders. Next, we aimed to explore the impact of CD3 affinity on a second TA.

CD3^intermed and CD3^low exhibited enhanced in vitro profiles as compared to CD3^high in CEACAM5-TCBs

Similar to the FOLR1-TCB series, CEACAM5-TCBs comprising the same CD3 binders of varying affinities (CD3^high, CD3^intermed, CD3^low, and CD3^vlow) were characterized in vitro to verify the findings observed when using FOLR1-TCBs. The exact assay conditions and readouts were applied, with the only exception being co-culturing the PBMCs with LS-180, a CEACAM5-positive human colorectal adenocarcinoma cell line. In contrast to FOLR1, CD3^intermed and CD3^low -to a slightly lower extent-induced comparable anti-tumor TDCC to CD3^high (Figure 5(A)). This was accompanied by a considerably greater reduction in cytokine secretion than CD3^high (5- to 10-fold, respectively, Figure 5(B-D)). Additionally, CD3^intermed maintained an equivalent level of T cell activation to CD3^high. However, CD3^low induced a lower level of T cell activation expressed as CD69 and CD25 upregulation on CD8⁺ cells (Figure 5(E-F)). These findings portray the possible difference in functional response that could occur across different targets, even when using the same set of CD3 binders. In line with our approach for the FOLR1-TCB series, the same co-culture was conducted using a lower affine variant of the CEACAM5 binder (CEACAM5^low) to assess the additional impact of TA affinity. Compared to FOLR1-TCBs comprising low affinity TA binder, the low affinity CEACAM5-TCBs did not show such a strong reduction in potency of killing, cytokine secretion and activation. CEACAM5^low led to a slight overall decrease in cytotoxic activity and T cell activation, which was more pronounced with CD3^low (Figure 6(A,E-F)). On the other hand, this was coupled with a 10- to 20-fold reduction in cytokine levels in favor of CD3^intermed and CD3^low, respectively, when compared to CD3^high (Figure 6(B-D)).

Figure 5.

CD3^intermed and CD3^low exhibited enhanced in vitro profiles as compared to CD3^high in CEACAM5 TCBs.

CEACAM5-TCBs comprising CD3 binders of varying affinities (CD3^high in black, CD3^intermed in blue, CD3^low in purple, and CD3^vlow in pink) and a CEACAM5 binder of high affinity were characterized in vitro. (A) Anti-tumor TDCC assay displayed as time-independent effective area under the curve (AUCE) calculated from three incubation time-points: 24 h, 48 h, and 72 h.⁶³ Cytokine secretion assessment displayed as AUCE calculated from the same three incubation time points (24 h, 48 h, and 72 h), with TNFα (B), IFNγ (C), and IL-6 (D) used as a representative example of the multiplex analysis. T cell activation assessment shown as the percentage of CD69 upregulation after 24 h (E) and CD25 upregulation after 72 h (F) on CD8+ T cells. Data points presented as filled squares (■) indicate the usage of the high affinity TA binder. TA binder monovalent affinity: 0.1 nM. Each data point depicts the average of technically duplicated measurements. Data shown are representative of two PBMC donors. The cell line used was LS-180 (antigen binding capacity ~76,000/cell). E:T = 10:1. The maximum concentration of TCBs used in the assays was 5000 pM, with TCBs titrated in a 1:10 serial dilution series. Error bars indicate ± standard deviation (SD). Time-dependent data shown in Supplementary Figure S8.

Figure 6.

TA binder affinity additionally influenced cytokine release when using CD3^intermed and CD3^low as CEACAM5-TCBs.

CEACAM5-TCBs comprising CD3 binders of varying affinities (CD3^high in black, CD3^intermed in blue, CD3^low in purple, and CD3^vlow in pink) and a CEACAM5 binder of low affinity were characterized in vitro. (A) Anti-tumor TDCC assay displayed as time-independent effective area under the curve (AUCE) calculated from three incubation time-points: 24 h, 48 h, and 72 h.⁶³ Cytokine secretion assessment displayed as AUCE calculated from the same three incubation time points (24 h, 48 h, and 72 h), with TNFα (B), IFNγ (C), and IL-6 (D) used as a representative example of the multiplex analysis. T cell activation assessment shown as the percentage of CD69 upregulation after 24 h (E) and CD25 upregulation after 72 h (F) on CD8+ T cells. Data points presented as open squares (□) indicate the usage of the low affinity TA binder. TA binder monovalent affinity: 5.5 nM. Each data point depicts the average of technically duplicated measurements. Data shown are representative of two PBMC donors. The cell line used was LS-180 (antigen binding capacity ~76,000/cell). E:T = 10:1. The maximum concentration of TCBs used in the assays was 5000 pM, with TCBs titrated in a 1:10 serial dilution series. Error bars indicate ± standard deviation (SD). Time-dependent data shown in Supplementary Figure S9.

Affinity-attenuated CD3 binders decreased cytokine secretion with minimal on-target, off-tumor effects in a 3D co-culture model of healthy jejunal organoids

To evaluate the potential on-target, off-tumor effects of the tested CEACAM5-TCBs in CEACAM5-positive normal tissues of the gastrointestinal (GI) tract, we used a 3D co-culture model. Healthy human jejunum-derived organoids were co-cultured with human PBMCs from a healthy donor. CEACAM5-TCBs comprising CD3 binders of varying affinities (CD3^high, CD3^intermed, CD3^low, and CD3^vlow) as well as high or low CEACAM5 binders were diluted to the following concentrations: 50,000 pM, 5000 pM, and 500 pM. Lastly, the prediluted TCBs were added to the co-culture and incubated for 72 h. The main readouts included cytokine secretion assessment and measurements of organoid cytotoxicity. Cytotoxicity was assessed using two methods: imaging to detect fluorescence of early apoptotic cells and measuring lactate dehydrogenase (LDH) release in the supernatant. While CD3^high groups induced the highest level of TDCC irrespective of the TA binder affinity (Figure 7(B)), TCBs with CD3 affinity-attenuated binders exhibited minimal cytotoxic effects on target cells. This was evident in both the mean fluorescence intensity quantification as well as LDH assessment (Figure 7(C-D)). More importantly, CD3 affinity-attenuated binders maintained a baseline level of cytokines that was nearly equivalent to the control group, regardless of the TCB concentration, whereas CD3^high induced the highest levels of TNFα (Figure 7(E)) and IL-2 (Figure 7(F)). No major differences were observed across the CEACAM5^high and CEACAM5^low binder groups, coupled with CD3^high. The limited difference in response across both high and low target affinity sets could be attributed to the relatively narrow range of their monovalent affinities (KD: 0.1 nM vs. 5.5 nM). This narrow range likely diminished the impact of target binder affinity in the presence of a complex in vitro model.

Figure 7.

Affinity-attenuated CD3 binders decreased cytokine secretion with minimal on-target, off-tumor effects in a 3D co-culture model of healthy jejunal organoids.

(A) Experimental design of the 3D organoid co-culture (picture created with BioRender). Healthy jejunal organoids were co-cultured with human PBMCs before administering the CEACAM5-TCBs at three concentrations (50000 pM, 5000 pM, and 500 pM) and incubating for 72 h. (B) Imaging assessment of co-culture conditions using the Opera from PerkinElmer. Green fluorescence indicates activated Caspase 3/7, denoting early apoptotic cells. Highest signals were detected in the presence of CD3^highbinder (left-hand side panels). Healthy organoid cytotoxicity shown as intensity image region mean (C, Fold change from baseline), or lactase dehydrogenase increase (D, Fold change from baseline), is the highest when using CD3^high compared to affinity-attenuated CD3 binders. Cytokine secretion assessment displayed as observed cytokine concentration (pg/ml) calculated from the same incubation time point (72 h), with TNFα (E) and IL-2 (F) used as representative examples from the multiplex analysis. Each data point depicts the average of technically triplicated measurements. Error bars indicate ± standard deviation (SD).

CD3^intermed and CD3^low demonstrated anti-tumor efficacy with minimal cytokine release as CEACAM5-TCBs in humanized BRGS-CD47 mice

To assess their in vivo efficacy and cytokine release, CEACAM5-TCBs containing CD3 affinity-attenuated binders were tested using female fully humanized mice. All mice were inoculated with a CEACAM5-expressing, human tumor cell line (HPAF-II) subcutaneously and after reaching the desired tumor volume, randomized into different treatment arms (eight mice per group). All therapies were administered iv at a 1 mg/kg dose, once per week for four treatment cycles. In addition to that, the vehicle control group received histidine buffer at the same frequency (Figure 8(A)). In terms of efficacy, TCBs comprising CD3^intermed and CD3^high successfully induced the best tumor growth inhibition (TGI), whereas CD3^low showed slightly weaker albeit significant levels of TGI. As anticipated, CD3^vlow induced the lowest level of anti-tumor efficacy (Figure 8(B), single-animal data; Supplementary Figure S10). Along with that, CD3^intermed and CD3^low displayed markedly lower levels of peripheral cytokines than CD3^high at 4 h after the first treatment, which was the same level of cytokines produced in the vehicle group (Figure 7(C-E)). To investigate the intra-tumoral infiltration of T cells, similar to the FOLR1-TCBs efficacy experiment, samples from three scouts were analyzed after they were euthanized 72 h following the second therapy (D24). Interestingly, T cells managed to infiltrate the tumor when treated with CD3^low binder, though in slightly lower amounts than CD3^intermed and CD3^high (Figure 7(F-H)). In this experiment, the extent of T cell activation among the efficacious groups appeared to depend on CD3 affinity, with the lowest CD69 upregulation levels illustrated by CD3^low binder (Figure 7(I-J)). Hence, in the context of CEACAM5-TCBs, we not only observe an advantage in favor of the CD3^intermed binder but also in favor of CD3^low, which successfully inhibited tumor growth with minimal levels of cytokines compared to CD3^high.

Figure 8.

CD3intermed and CD3low demonstrated anti-tumor efficacy with minimal cytokine release as CEACAM5 TCBs in humanized BRGS mice(A) Experimental design of the in vivo efficacy experiment (picture created with BioRender). Humanized BRGS-CD47 mice were inoculated with a CEACAM5-expressing, human, tumor cell line (HPAF-II: antigen binding capacity ~100,000/cell) prior to treatment administration. Each group consisted of eight mice, receiving either the vehicle (histidine buffer) as a control or CEACAM5-TCBs with CD3 binders of varying affinities, administered intravenously. All treated mice received the same dose 1 mg/kg once per week for four treatment cycles. Scouts (n = 3/group) were euthanized 72 h after the second therapy for analysis. (b) Tumor volumes measured by a caliper three times per week throughout the experiment to assess the efficacy of the therapies in inhibiting tumor growth. Dashed lines indicate experiment days on which treatments were administered. Statistical significance was determined using an unpaired, 2-tailed t test with Welch’s correction: *p < 0.05; ***p < 0.0001. Blood samples were collected at four hours (4 h) after the first therapy (3 mice bled per time point) for cytokine release assessment. IP-10 (c), mig (d), and IL-6 (e) are representative examples from the multiplex analysis. Statistical significance was computed using one-way anova with Tukey’s multiple comparison test: *p < 0.05; **p < 0.01; ***p < 0.0001; ****p < 0.0001. (F-J) intratumoral T cell infiltration and activation were assessed in scouts 72 h after the second therapy. Three mice were euthanized, and their tumors were processed and analyzed via flow cytometry to determine the total count of T cells, CD8, CD4, and CD69 upregulation on both cell types. Data points presented as filled squares (■) indicate the usage of the high affinity TA binder. Error bars indicate ± standard error of the mean (SEM).

Higher target binding affinity enhances the activity of TCBs comprising CD3^low binder

Out of the affinity-attenuated CD3 binders, only CD3^intermed elucidated a favorable profile of anti-tumor TDCC and cytokine release in comparison to CD3^high within the FOLR1-TCB series. All binders beyond the CD3^intermed affinity level were deemed inefficacious, possibly due to the relatively low monovalent target affinity (ca. 57 nM) (Figure 9(A)). However, when CD3^low was coupled with a high affine CEACAM5 binder (ca. 0.1 nM), it demonstrated a comparable cytotoxic response to both CD3^high and CD3^intermed (Figure 9(B)). This was also demonstrated through the in vitro to in vivo translation for both FOLR1- and CEACAM5-TCBs. In these cases, our in vitro potency findings were recapitulated in our in vivo efficacy readouts (Supplementary Figure S11). Similarly, CD3^low exhibited in vitro cytotoxicity that was on par with CD3^high when used as a TYRP1 TCB (monovalent KD ~0.9 nM, Figure 9(C)). This is especially interesting owing to the further reduction of cytokines that CD3^low achieved, beyond the already-reduced levels produced by CD3^intermed (Figure 9(D-F)). These findings illustrate the interdependent relationship between target and CD3 affinities to achieve the optimal balance of efficacy and safety in TCBs.

Figure 9.

Higher target binding affinity enhances the activity of CD3^low binder.

Comparison of TCBs comprising CD3 binders of varying affinities (CD3^high in black, CD3^intermed in blue, CD3^low in purple, and CD3^vlow in pink) in vitro with different TAs. T cell mediated tumor cytotoxicity assay displayed as relative luminescence unit (RLU) after a 48-h incubation. (A) FOLR1 TCBs. (B) CEACAM5-TCBs. (C) TYRP1-TCBs. Cytokine secretion assessment displayed as AUC calculated from the incubation time point (48 h) with TNFα used as a representative example from the multiplex analysis. (D) FOLR1-TCBs. (B) CEACAM5-TCBs. (C) TYRP1-TCBs. Data points presented as filled squares (■) indicate the usage of the high affinity TA binder. TA binders’ monovalent affinities are depicted on the figure. Each data point depicts the average of technically duplicated measurements. Data shown are representative of two PBMC donors. The cell line used were SK-OV-3 (FOLR1+), LS-180 (CEACAM5+), and M150543 (TYRP1+). E:T = 10:1. The maximum concentration of TCBs used in the assays was 50,000 pM, with TCBs titrated in a 1:10 serial dilution series. Error bars indicate ± standard deviation (SD). AUCE was calculated using GraphPad Prism 10.

CD3^low reduced cytokine release and evaded on-target off-tumor toxicity in huCD3xhuCEACAM5 transgenic mouse model

To assess the effect of CEACAM5-TCBs on normal tissue, we conducted an in vivo experiment using a huCD3xhuCEACAM5 transgenic mouse model, which exhibits target expression in the GI tract and human CD3 on circulating mouse T cells.⁴³ Six mice per treatment group were administered the respective TCB iv at 0.5 mg/kg dose, except for the CD3^high group which was dosed with 0.1 mg/kg for tolerability reasons. Additionally, the vehicle control group received histidine buffer via iv infusion. Seven days after a single dose treatment, all mice were euthanized, and the relevant organs were collected for histopathological assessment. The histopathological assessment involved identifying parameters indicative of microscopic changes in the small and large intestines, such as apoptosis of single enterocytes in the bottom of the intestinal crypts, variably associated with mixed cell infiltrates in the lamina propria and submucosa, as well as regenerative epithelial hyperplasia. These findings were similarly observed in mice treated with CD3^high and CD3^intermed with higher incidence and severity of regenerative epithelial hyperplasia in those treated with TCB comprising CD3^intermed binder (Table 2). This could be attributed to the higher dose of the TCB comprising CD3^intermed compared to CD3^high, leading to a similar level of tissue alteration. In contrast, CD3^low did not induce any changes in both organs investigated, exhibiting an identical profile to the vehicle control group. Most intriguingly, CD3^low peripheral cytokine levels were equivalent to those produced by the vehicle group, at various time points, even when administered at a dose five times higher than CD3^high (Figure 10C-F). Meanwhile, CD3^intermed displayed a cytokine profile similar to that induced by CD3^high, despite the latter being injected at a lower dose. This showcases the possibility to attain an optimized cytokine profile without inducing on-target off-tumor effects using this modality, even when the TA is expressed in normal tissues.

Table 2.

Incidence and severity of histopathological findings in huCd3xhuceacam5 transgenic mice when treated with CEACAM5-TCBs comprising CD3 binders of varying affinities.

Group	Vehicle	CD3high	CD3intermed	CD3low	CD3vlow
Small intestine
Number examined	6	6	6	6	6
Apoptosis of single enterocytes
Not Present	6	1	2	6	6
Minimal	–	5	2	–	–
Slight	–	–	2	–	–
Mixed cell infiltrate lamina propria/submucosa
Not Present	6	1	0	6	6
Minimal	–	5	2	–	–
Slight	–	–	4	–	–
Epithelial hyperplasia, regenerative
Not Present	6	6	3	6	6
Slight	–	–	1	–	–
Moderate	–	–	1	–	–
Marked	–	–	1	–	–
Large intestine
Number examined	6	6	6	6	6
Apoptosis of single enterocytes
Not Present	5	5	4	5	6
Minimal	1	1	2	1	–
Mixed cell infiltrate lamina propria/submucosa
Not Present	6	5	6	6	6
Minimal	–	1	–	–	–

Figure 10.

CD3^low reduced cytokine release and evaded on-target off-tumor toxicity in huCd3xhuceacam5 transgenic mouse model.

(A) Experimental design of the in vivoexperiment assessing the effect of CEACAM5-TCBs on normal tissue (picture created with BioRender). Each group consisted of six mice, receiving either the vehicle (histidine buffer) as a control or CEACAM5-TCBs with CD3 binders of varying affinities, administered intravenously. Mice treated with TCBs comprising the CD3^intermed, CD3^low and CD3^vlow binders received the same dose 0.5 mg/kg, while mice treated with TCB comprising the CD3^high binder were dosed at 0.1 mg/kg for tolerability reasons. Seven days after a single dose treatment, all mice were euthanized, and the intestinal tract was collected for histopathological assessment. (B) FFPE embedded staining of the small intestine: top left (1, Vehicle), top right (2, CD3^high), bottom left (3, CD3^intermed), and bottom right (4, CD3^low). HE 20x magnification. Detailed severity and incidence in Table 2. Mice were bled intravenously for cytokine assessment at different time-points (4 h–96 h) after the single dose treatment (2–3 mice bled per time point). TNFα (C), IFNγ (D), IL-6 (E) and IL-2 (F) are representative examples from the multiplex analysis. Data points presented as filled squares (■) indicate the usage of the high affinity TA binder. TA binder monovalent affinity: 0.1 nM. Statistical significance was computed using two-way ANOVA with Tukey’s multiple comparison test: *p < 0.05; **p < 0.01. Error bars indicate ± standard error of the mean (SEM).

CD3 affinity has no major impact on pharmacokinetics

Pharmacokinetic (PK) parameters were evaluated to investigate the impact of the selected CD3 binders on PK properties in vivo. Time-dependent exposure of FOLR1- and CEACAM5-TCBs was determined using mouse serum acquired at various time-points within the corresponding experiments (Figure 11). The mouse serum samples were analyzed using an ELISA assay using a biotinylated version of the anti-Fc(P329G) IgG (an antibody that specifically binds human IgG₁ Fc (P329G), the backbone of the test TCBs). In summary, we could illustrate that CD3 affinity has no major impact on PK, including TCB exposure and other PK parameters such as clearance, half-life, and Cmax (Table 3). This was shown at five (Figure 11A) or four (Figure 11B-C) different time points after the first therapy in the corresponding studies. Notably, these findings were confirmed across three different mouse models: humanized NSG mice (A), humanized BRGS-CD47 mice (B), and transgenic mice (C), irrespective of the TA expression on normal tissue or its absence. It is worth emphasizing that the lack of impact of CD3 affinity on PK was observed at efficacious doses for some of the TCBs used as we previously described. This contrasts with published reports suggesting that higher CD3 affinity resulted in lower systemic exposure of TCBs due to their infiltration into CD3⁺ T cell-rich tissues,³⁵ and that affinity-attenuated CD3 binders led to two-to-three fold increase in systemic exposure in monkeys.^29,44

Figure 11.

CD3 affinity has no major impact on PK.

TCB concentration (exposure) was measured at a series of time points in vivo to assess the impact of CD3 affinity on PK in: (A) HuNSG mice using FOLR1-TCBs that were administered at 0.5 mg/kg. Blood samples were collected at 1 h, 4 h, 48 h, 96 h, and 168 h after the first therapy (3 mice bled per time point). (B) HuBRGs mice using CEACAM5-TCBs that were administered at 1 mg/kg. Blood samples were collected at 4 h, 48 h, 96 h, and 168 h after the first therapy (3 mice bled per time point). (C) huCD3xhuCEACAM5 transgenic mice using CEACAM5-TCBs that were administered at 0.5 mg/kg (except CD3^high: 0.1 mg/kg for tolerability reasons). Blood samples were collected at 4 h, 48 h, 96 h, and 168 h after the first therapy (3 mice bled per time point). Data points presented as filled squares (■) indicate the usage of the high affinity TA binder. Error bars indicate ± standard error of the mean (SEM).

Table 3.

Summary of pk parameters using FOLR1-TCBs comprising CD3 binders of varying affinities in humanized NSG mice.

	Cl	AUC0-last	Cmax
	(mL/day/kg)	(day*µg/mL)	(µg/mL)
CD3high	24.2	15.4	6.5
CD3intermed	23.3	15.9	6.1
CD3low	19	18.6	6.3
CD3vlow	25	16.4	7.2

Discussion

Historically, T cell-based therapies were designed with high affinities for both effector target and tumor target to elicit a strong immune response. This concept has been clinically validated, demonstrating effective anti-tumor efficacy with several TCBs bearing high affinity CD3 binders being approved. Nonetheless, such TCBs can face challenges related to cytokine release. For that purpose, recent reports suggested that reducing CD3 affinity could stimulate lower levels of cytokines while maintaining on-target tumor killing.^24–26 In this study, we characterized the impact of the anti-CD3 affinity on the activity of FOLR1- and CEACAM5-TCBs, and highlighted its interdependency with other parameters, aiming to identify a profile that is both safe and efficacious for these TCBs.

Our stepwise approach involved identifying key parameters that could influence the optimum functionality of TCBs. We maintained most parameters constant while selectively varying a few. Initially, our strategy relied on deriving CD3 binders of different affinities from a primary parental clone, ensuring that the binding epitope was sustained. Similarly, low-affinity TA binders were developed from their high-affinity counterparts targeting FOLR1 and CEACAM5. The same TCB format (2 + 1 “inverted” format) was applied across experiments to eliminate the potential impact of TCB geometry on our results. Moreover, we ensured that the expression levels of target cell lines in vitro were uniform across all targets. We used the same biological PBMC donors and kept incubation time points as well as assay readouts consistent. For in vivo efficacy assessment, we made use of hematopoietic stem cell (HSC)-engrafted fully humanized mice instead of a PBMC-transfer model, which is associated with the development of strong graft-versus-host disease. This condition limits the experimental window and results in a non-translatable activated T cell phenotype, potentially leading to an overestimation of anti-tumor efficacy.^45–47

In this study, our results confirmed the impact of CD3 affinity on both in vitro and in vivo activity of FOLR1- and CEACAM5-TCBs. In terms of efficacy, only CD3^intermed among the affinity-attenuated binders mediated TGI as a FOLR1-TCB in vivo. In contrast, CD3^low was inefficacious even when dosed at concentrations exceeding expected saturation levels. However, when targeting CEACAM5 both CD3^intermed and CD3^low achieved comparable TGI, consistent across in vitro and in vivo assessments (Supplementary Figure S11). TA binder affinity also played a role in dictating anti-tumor activity. The combination of CD3^low with the CEACAM5^high binder achieved anti-tumor cytotoxicity both in vitro and in vivo, whereas this effect was not observed when CD3^low was combined with the FOLR1^high binder. This can be attributed to the elevated monovalent affinity of the CEACAM5^high binder (0.1 nM) compared to that of FOLR1^high (57 nM).

In line with published reports, CD3^high mediated the highest levels of cytokine secretion both in vitro and in vivo in comparison to CD3^intermed and CD3^low regardless of the target. This highlighted the importance of lowering CD3 affinity to achieve a safe functional profile of TCBs. Even though CEACAM5 was expressed in normal tissue in vivo in transgenic mice, CD3^low managed to induce the lowest levels of cytokines compared to CD3^high, despite the on-target, off-tumor target expression. It is worth highlighting that, despite being administered at a dose five times higher for tolerability reasons, CD3^intermed induced the same cytokine levels as CD3^high.

Mechanistically, the role of the IS in T cell-mediated cytotoxicity has been a subject of interest for over 20 y. Purbhoo et al. suggested that full synapse stability and maturation are not a prerequisite for T cell-mediated cytotoxic function.²² This prompted the ongoing pursuit of sufficient yet balanced T cell activation patterns using antibody engineering approaches to improve the properties of TCBs. CD3 affinity reduction is one of the approaches used to achieve this phenomenon. In our study, we illustrated that the stability of the IS is CD3-affinity dependent. Intriguingly, the level of synapse stability that CD3^intermed achieved was sufficient to sustain a suitable anti-tumor cytotoxicity both in vitro and in vivo. This came with the added benefit of triggering a weaker cytokine profile than that of CD3^high, which supports the hypothesis proposed by Faroudi et al. Synapse instability with CD3^vlow was reflected in its performance in vitro and in vivo, an issue that is additionally attributable to the relatively weak monovalent affinity of FOLR1^high.

TA abundance is one of the factors crucial for dictating TCB functionality, as well as selecting the suitable affinities of the targeting arms. In this study, we explored how the TCBs comprising CD3 and TA binders of varying affinities behave in the presence of different levels of target expression in vitro. Even at the highest expression level when using CEACAM5-TCBs, affinity-attenuated CD3 binders (CD3^intermed & CD3^low) induced lower TNFα levels than CD3^high, while maintaining satisfactory TDCC levels, especially for CD3^intermed. This advantage was more obvious at an intermediate expression level using LS-180. Whereas CD3^high was the only functioning binder at the lowest expression level, giving rise to the possibly beneficial therapeutic window offered through affinity-tuned CD3 binders (Supplementary Figure S12). While these findings were generated using high-affinity TA binder, similar findings were realized when reducing the TA affinity, although with an overall reduction of TDCC and additional TNFα reduction in favor of the affinity-attenuated CD3 binders (Supplementary Figure S13), highlighting the necessity of accounting for the interdependence of target expression levels together with the CD3 and TA binder affinities.

It is important to consider additional factors influencing TCB activity, such as the location of the target binding epitope, whether membrane-distal or proximal. Proximal epitope binding is often associated with enhanced potency, increased tumor-killing, and improved synapse formation^48,49, .⁵⁰ In contrast, distal binding is linked to reduced anti-tumor efficacy, but carries a lower risk of excessive T-cell engagement. In our experiments using CEACAM5 as a tumor target, the selected TA binder was reported to bind to a proximal epitope. To harness the potential of proximal binding through affinity attenuation, we initially investigated diverse combinations of CD3 and TA binders with varying affinities. Our findings illustrated a “seesaw-relationship” between the affinities of both arms, resulting in reduced cytokine release while maintaining anti-tumor TDCC (in vitro AUCE summary: Supplementary Table S2), compared to conventionally used high-affinity binders (Supplementary Figure S14). Hence, the possibility of binding to proximal or distal epitopes of the target should be considered when selecting suitable affinity ranges for both CD3 and TA binders during drug development. The nature of the tumor target is another factor to consider when interpreting preclinical data sets. Our study focused exclusively on solid tumor targets to avoid possible discrepancies that could arise from the higher accessibility of hematological tumors.⁵¹

Furthermore, the choice of suitable animal models to characterize CD3 binders of reduced affinity is often overlooked. A common theme in published reports in this field is demonstrating potent anti-tumor efficacy with low CRS using PBMC-transfer mouse models to assess the efficacy of their molecular entities.^{26–30,52–54} Here, we relied on HSC-engrafted humanized mouse models for the efficacy assessment of the modified TCBs, specifically CD34⁺ huNSG and huBRGS-CD47 mouse models. These models allow a more accurate and realistic representation of human T cell biology, and other immune cells that influence the cytokine release cascade, including dendritic cells, monocytes, macrophages and neutrophils.⁵⁵ In our experiments using stem-cell humanized mice, the TCB comprising the CD3 binder of the lowest affinity (CD3^vlow) lost its anti-tumor activity regardless of the TA. This contrasts with published reports that claim maintaining efficacy with low cytokine release when using TCBs with low CD3 affinity in PBMC-transfer mice. For example, CEA/CD3-v2, which contains an affinity-modulated CD3 binder showed superior efficacy when combined with atezolizumab in huPBMC-NCG mice, where a mixture of LS-174T tumor cells and human PBMCs were injected before treatment.²⁸ Haber et al. described that CD3 arms of weaker affinity reduced tumor size with lower cytokine levels in NSG mice injected with human PBMCs.²⁹

Following the development of the F2B low-affinity CD3 binder,²⁶ several antibodies targeting different antigens were generated using this binder and were assessed similarly in vivo. Specifically, TNB-486 (anti-CD19) and TNB-928B (anti-FRα) were reported to induce tumor regression assessed in mice injected with human PBMCs prior to the first therapy, whereas TNB-585 (anti-PSMA) caused dose dependent TGI using resting pan-T cells or preactivated PBMCs.^27,30,52 The observed differences may be explained by the use of distinct humanized in vivo models. In the PBMC-transfer model, xeno-activated human T cells exhibit a low threshold for reactivation, even with lower CD3 affinities. In contrast, resting human T cells from the HSC-transfer model do not display this characteristic since they have undergone murine thymic selection de novo and thus being tolerant against the mouse tissue. Utilizing HSC-humanized mice could therefore reduce the risk of overestimating the efficacy of low-affinity CD3 binders by preventing the exaggerated immune response that might result from the preactivation of injected PBMCs. Another advantage over the PBMC-transfer is eliminating the possibility of early-onset graft-versus-host disease, allowing for longer study duration and extensive efficacy assessments.⁵⁶ That being said, we recognize some of the limitations of humanized mice, primarily concerning the reduced presence of human cell subsets like myeloid cells and natural killer (NK) cells, alongside the limited heterogeneity of the TME.^57,58 Nevertheless, the integration of PDX in humanized mice offers an enhanced TME that better reflects tumor heterogeneity, which is necessary to bridge the gap between preclinical data and clinical outcomes.⁵⁷ By leveraging these state-of-the-art models, we reconfirmed our in vitro findings in vivo and established a high in vitro to in vivo translatability across data sets. Apart from the primary aim of understanding the contribution of selected parameters to on-target toxicity while maintaining potency and efficacy, we investigated the impact of the CD3 binder affinity on PK properties. Previous studies emphasized the effect CD3 affinity poses on biodistribution and exposure. Namely, higher anti-CD3 affinity led to the relative distribution of TCBs to T cell-rich tissues rather than the tumor in mice.⁵⁹ In that instance, the main focus was on the impact of CD3 affinity on differential biodistribution in huCD3e-TG mice, where TCBs with high CD3 affinity showed significantly greater distribution to the lymph nodes and spleen compared to TCBs with low CD3 affinity. Similarly, MUC16-targeted TCBs of higher CD3 affinity localized in both the spleen and the MUC16⁺ tumor, whereas TCBs of weaker CD3 affinity accumulated only in the tumor.²⁹ In addition, TCBs displayed CD3 affinity-related differential patterns of systemic PK, as reported in cynomolgus monkeys. However, it is worth highlighting that these studies were conducted with TCBs targeting hematological antigens.²⁹ In contrast to the findings mentioned above, our CD3 binders illustrated similar exposure levels and PK parameters in humanized mice whether used as FOLR1 or CEACAM5-TCBs. In huCD3xhuCEACAM5 transgenic mice, affinity-attenuated binders also achieved comparable exposure levels. This suggests the limited impact of CD3 affinity on PK when combined with TA binders spanning a broad affinity range in the 2 + 1 format, as demonstrated by our TCBs targeting solid TAs.

One of the limitations of our study was the affinity gap between our TA binders. For FOLR1, the affinity difference was substantial (57 nM vs. 1000 nM), which led to a diminished response when coupled with CD3 binders of low and very low affinity. In contrast, for CEACAM5, their affinity difference was 55-fold (0.1 nM vs. 5.5 nM), but both values were within a high affinity range. It is not surprising that the difference in response between the two affinity sets was minimal, given the additional influence of avidity.⁶⁰ Hence, exploring a range of TA binders at appropriate affinity intervals could help identify various TCB activity profiles for further optimization. Furthermore, exploring the impact of TCB design on mechanistic factors, such as T cell exhaustion and memory T cell formation, particularly in the context of affinity attenuation, warrants further investigation.^57,61

Taken together, our study illustrates the critical role that CD3 affinity plays in achieving the sweet spot of decoupling T cell-mediated cytotoxicity from high levels of cytokine release, in conjunction with the chosen TA binder. As evident in our data, it is feasible to achieve target cell cytotoxicity as well as anti-tumor efficacy without the need for the typically aggravated T cell response elicited by CD3 binders of high affinity. Nevertheless, the selection of the optimal affinity of the CD3 binder depends on the TA binder, and the complete elimination of cytokine release was not possible without losing anti-tumor efficacy in humanized mice. These findings are specific to the CD3 epitope studied here and underscore the potential need for further optimization when exploring other CD3 epitopes or binder backbones. Additionally, cross-comparisons with binders targeting different CD3 epitopes should be cautiously interpreted, as differences in epitope location and binding dynamics could significantly influence TCB functionality.

In conclusion, reducing cytokine release while maintaining adequate efficacy is possible through CD3 binder affinity attenuation; however, optimizing cytokine release profiles by CD3 binder affinity-attenuation is dependent on additional parameters. Depending on the target, a tailored combination of CD3 and TA binder affinities might have to be selected accordingly. Establishing a “seesaw-relationship” between TA and CD3 binders’ affinity could be beneficial to achieve optimized safety-efficacy profiles.

Materials and Methods

Generation of CD3 binders of attenuated affinity

CD3 antigen used for phage display selection

The parental CD3e-specific antibody Clone22 binds to the extracellular domain (ECD) of CD3e with an affinity of 4 nM. For the generation of affinity-reduced variants of Clone22 by phage display, an Avi tag-biotinylated recombinant soluble antigen consisting of a heterodimeric CD3(epsilon)/CD3(delta) – Fc fusion construct (knob into hole) was cloned, produced and purified.

Selection of affinity matured A5H1EL1D-derived antibodies

For the generation of affinity-reduced CD3e-specific antibodies, a previously generated Clone22-derived library was used. This library was originally built to identify Clone22 binder derivatives without hotspots and contains randomized positions in the variable heavy (VH) region.

Selection of affinity reduced Clone22-derived clones

For the selection of affinity-reduced clones, a phage display selection was performed using the recombinant soluble CD3-Fc antigen described previously. The panning procedure was performed in solution according to the following steps:

1. Binding of phagemid particles to 100 nM biotinylated CD3 antigen for 30 min in a total volume of 1 mL.

2. Capture of biotinylated CD3-Fc antigen and specifically bound phage particles by addition of 5.4 × 10⁷ streptavidin-coated magnetic beads (Invitrogen, 11205D) for 10 min.

3. Washing of beads using 5x 1 ml phosphate-buffered saline (PBS)/Tween 20 (Sigma-Aldrich, P9416) and 3 × 1 mL PBS (Gibco, 10,010,023). During this procedure, the washing buffer of the last washing step was harvested, stored, and later used for screening of low affinity binders.

4. Elution of phage particles by the addition of 1 mL 100 mM TEA (Sigma-Aldrich, T0886) for 10 min and neutralization by adding 500 µL 1 M Tris/HCl (Sigma-Aldrich, T3253), pH 7.4.

5. Infection of exponentially growing E. coli TG1 bacteria

6. Super-infection with helper phage VCSM13 (Agilent Technologies, 200,251), followed by PEG/NaCl (Sigma-Aldrich, P1458; S7653) precipitation of phagemid particles to be used in subsequent selection rounds.

Selections were carried out over two rounds using constant antigen concentrations (100 nM). In both rounds, the washing buffer from the last washing steps was harvested and used for the infection of bacteria and subsequent screening for low-affinity binders.

Specific CD3 binders of both the eluates and the washing buffer of the last wash step were identified by ELISA as follows: 100 µL of 100 nM biotinylated CD3-Fc antigen (produced in-house) per well were coated on NeutrAvidin plates (Pierce, 15,121). Fab-containing bacterial supernatants were added and binding Fabs were detected via their Flag-tags using an anti-Flag/HRP secondary antibody (Sigma-Aldrich, A8592) Clones that were ELISA-positive were further tested by SPR.

Identification of affinity-reduced Clone22-derived variants by SPR

To further characterize the ELISA-positive clones, the off-rate was measured by SPR using a Proteon XPR36 machine at 25°C. The results were compared with the parental humanized clone, which was used as a reference in the same experiment.

About 7000 RUs of a polyclonal anti-human Fab antibody (Sigma-Aldrich, I5260) were immobilized on all six channels of a GLM chip (Bio-Rad, 176–5012) by Amine coupling (Na acetate (Sigma-Aldrich, S2889)

pH 4.5, 25 ml/min, 240 s) (vertical orientation). Each antibody-containing bacterial supernatant was filtered and 3-fold diluted with PBS and then injected for 360 s at 25 μl/min to achieve immobilization levels of between 100 and 400 response units (RU) in vertical orientation.

Injection of CD3-Fc: For one-shot kinetics measurements, injection direction was changed to horizontal orientation and a two-fold dilution series of CD3-Fc (varying concentration ranges between 100 and 6.25 nM) was injected simultaneously at 50 μl/min along separate channels 1–5, with association times of 240 s, and dissociation times of 1000 s. Buffer (PBST) was injected along the sixth channel to provide an “in-line” blank for referencing. Regeneration was performed by two pulses of 10 mM glycine (Sigma-Aldrich, G7126) pH 1.5 and 50 mM NaOH (Sigma-Aldrich, S8045) for 30 s at 90 μl/min (horizontal orientation). Dissociation rate constants (koff) were calculated using a simple one-to-one Langmuir binding model in ProteOn Manager v3.1 software by simultaneously fitting the sensorgrams.

Analysis of the koff rates revealed ten binders that showed significantly reduced dissociation rate constants compared to Clone22. While the variable light (VL) sequence was kept constant in this library and contained the same sequence as Clone22, the randomized VH regions of these clones were sequenced.

Fab purification of affinity-reduced Clone22 clones

To further characterize the most promising affinity-reduced clone variants, the respective Fab fragments were purified for the analysis of the kinetic parameters. For each clone, a 500 mL culture was inoculated with bacteria harboring the corresponding phagemid and induced with 1 mM IPTG (Sigma-Aldrich, I6758) at an OD600 0.9. Afterward, the cultures were incubated at 25°C overnight and harvested by centrifugation. After incubation of the resuspended pellet for 20 min in 25 mL PPB buffer (30 mM Tris-HCl pH8, 1 mM EDTA (Sigma-Aldrich, E9884), 20% sucrose (Sigma-Aldrich, S0389)), bacteria were centrifuged again, and the supernatant was harvested. This incubation step was repeated once with 25 mL of a 5 mM MgSO4 (Sigma-Aldrich, M7506) solution. The supernatants of both incubation steps were pooled, filtered and loaded on an immobilized metal affinity chromatography column (His gravitrap, GE Healthcare). Subsequently, the column was washed with 40 mL of washing buffer (500 mM NaCl, 20 mM imidazole (Sigma-Aldrich, I2399), 20 mM NaH2PO4 pH 7.4). After the elution (500 mM NaCl, 500 mM imidazole, 20 mM NaH2PO4 (Sigma-Aldrich, S8282) pH 7.4), the eluate was re-buffered in PBS using PD10 columns (GE Healthcare). The yield of purified protein was in the range of 300 to 500 μg/L. Generation of 2 + 1 TCBs

Cloning and production of proteins were performed at Evitria (Switzerland) and WuXi Biologics (China). The corresponding cDNAs were cloned into Evitria’s or WuXi’s vector system using conventional (non-PCR based) cloning techniques. Plasmid DNA was prepared under low-endotoxin conditions based on anion exchange chromatography. DNA concentration was determined by measuring the absorption at a wavelength of 260 nm. Correctness of the sequences was verified with Sanger sequencing (with two sequencing reactions per plasmid). Suspension-adapted CHO K1 cells (originally received from ATCC and adapted to serum-free growth in suspension culture at Evitria or WuXi) were used for production. Cells were grown in commercially available, chemically defined media (Gibco™ CD OptiCHO™ Medium; 12,681–011). Transfection was performed using a proprietary transfection reagent. Post-transfection, cells were cultivated under the following conditions: 36.5°C with 6% carbon dioxide. The supernatant was harvested by centrifugation and subsequent filtration (0.2 μm filter).

Proteins were purified in-house from filtered cell culture supernatants referring to standard protocols. In brief, Fc-containing proteins were purified from cell culture supernatants by Protein A-affinity chromatography (equilibration buffer: 20 mM sodium citrate (Sigma-Aldrich, S4641), 20 mM sodium phosphate (Sigma-Aldrich, S7907), pH 7.5; elution buffer: 20 mM sodium citrate, pH 3.0). Elution was achieved at pH 3.0 followed by immediate pH neutralization of the sample. The protein was concentrated by centrifugation (Millipore Amicon® ULTRA-15, UFC903096), and aggregated protein was separated from monomeric protein by size exclusion chromatography in 20 mM histidine (Sigma-Aldrich, H8000), 140 mM sodium chloride, pH 6.0. The concentrations of purified proteins were determined by measuring the absorption at 280 nm using the mass extinction coefficient calculated on the basis of the amino acid sequence according to Pace et al.⁶² Purity and molecular weight of the proteins were analyzed by CE-SDS in the presence and absence of a reducing agent using a LabChipGXII or LabChip GX Touch (Perkin Elmer, CLS760675). Determination of the aggregate content was performed by HPLC chromatography at 25°C using analytical size-exclusion column (TSKgel G3000SWXL, 0008541 or UP-SW3000, 0023480) equilibrated in running buffer (200 mM KH2PO4 (Sigma-Aldrich, P0662), 250 mM KCl (Sigma-Aldrich, P9333) pH 6.2, 0.02% NaN3 (Sigma-Aldrich, S2002)). Quantification of Fc-containing proteins in supernatants was performed by Protein A HPLC on an Agilent HPLC system with UV detector. Supernatants were injected on POROS 20 A (Applied Biosystems, 1–2259-12), washed with 10 mM Tris, 50 mM glycine, 100 mM NaCl, pH 8.0 and eluted in the same buffer at pH 2.0. The eluted peak area at 280 nm was integrated and converted to concentration by use of a calibration curve with standards analyzed in the same run. All TCBs were produced with a final-product batch quality with consistently low endotoxin concentrations, optimal protein purity (deduced from CE-SDS main peak), as well as adequate purification yields.

CD3 affinity measurement in the TCB format using SPR

Affinities of the CD3 binders in the FOLR1-TCB format to recombinant human CD3-Fc (heterodimer of the CD3 epsilon and CD3 delta chains fused on a human Fc) were assessed by SPR. SPR was performed on a Biacore 8K⁺ at 25°C with HEPES-buffered saline (HBS)-EP running buffer (0.01 M HEPES pH 7.4, 0.15 M NaCl, 3 mM EDTA, 0.05% Surfactant P20, Cytiva, BR100826). Anti-Fc(P329G) IgG (an antibody that specifically binds human IgG1 Fc(P329G)); “anti-PG antibody” (produced in-house) was directly immobilized on a CM3 chip (Cytiva, 29,104,990) at pH 5.0 using the standard amine coupling kit (Cytiva, BR100050). After activation of the sensor surface with a 1:1 mixture of 0.4 M 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (EDC; Sigma-Aldrich, E7750) and 0.1 M N-hydroxysuccinimide (NHS; Sigma-Aldrich, 130,672), 10 μg/ml of anti-PG antibody (diluted in 10 mM acetate pH 5.0, produced in-house) was injected for 200 s with a flow rate of 10 μl/min. After blocking with 1 M ethanolamine-HCl (Sigma-Aldrich, E4128) pH 8.5, the coupling procedure led to approximately 2500 RU anti-PG antibody surface density. FOLR1-TCBs comprising the various CD3 binders were captured for 100 s (flow: 10 μl/min) with a concentration of 5 nM. Recombinant human CD3-Fc was injected at various concentrations (300–0.046 nM, 1:3 dilution) at a flow rate of 30 μl/min through the flow cells. Association and dissociation were monitored for 240 s and 800 s, respectively. The chip surface was regenerated after every cycle by using two injections (60 s each) of 10 mM glycine pH 2. Bulk refractive index differences were corrected by subtracting the response obtained on the reference flow cell. The affinity constants were derived from the kinetic rate constants by fitting to a 1:1 Langmuir interaction model by using Biacore Insight Evaluation Software 4.0 (Cytiva) and are shown in Table 1.

To measure CD3 binding kinetics of approved T cell engagers (Supplementary Table S1), we used SPR on a Biacore 8K instrument (Cytiva). HBS containing 0.05% PS20 was used as running and dilution buffer with the temperature maintained at 25°C. For antibody capturing, a mouse anti-human IgG Fc-PGLALA (produced in-house) was immobilized on an HC30M sensor chip (Xantec bioanalytics GmbH) using standard amine coupling chemistry. After capturing the human anti-CD3 IgGs at the surface, human CD3ε/δ was injected onto the surface with increasing concentrations up to 600 nM using the single cycle kinetics setup. The association steps were monitored for 100 s, dissociation was monitored for 600 s followed by a surface regeneration with injecting 10 mM NaOH for 60 s to remove any non-covalently bound proteins. Bulk refractive index differences were corrected by subtracting blank injections and the response from the reference flow cell without captured antibody. Data analysis and curve fitting were conducted using the 1:1 Langmuir binding model within the Biacore evaluation software.

Binding affinities and kinetics of monovalent anti-CD3 antibodies were investigated by SPR using a Biacore T200 instrument (Cytiva) at 25°C. Biotinylated human CD3ε/δ was immobilized on a Neutravidin sensor chip (Xantec bioanalytics GmbH, NAHLC30M) with a surface density of approximately 10 resonance units (RU). As a running and dilution buffer, HBS containing 0.05% Tween-20 was used. The monovalent anti-CD3 antibodies were injected onto the surface with a concentration series up to 300 or 900 nM for 90 s and 60 s, dissociation was monitored for 300 s and 60 s, respectively. Subsequently, the surface was regenerated by injecting 10 mM glycine pH 2.0 for 60 s. Bulk refractive index differences were corrected by subtracting blank injections and by subtracting the response obtained from the reference flow cell without CD3ε/δ. Curve fitting was performed using the 1:1 Langmuir binding model within the Biacore evaluation software.

Cell lines

The following cell lines were used in either in vitro assays or in vivo experiments and were acquired from ATCC: SK-OV-3 (HTB-77™), HT-29 (HTB-38), LS-180 (CL-187), and HPAF-II (CRL-1997). MKN-45 was obtained from Cytion. All cell lines were thawed, cultured, and frozen according to the provider’s protocols.

Antigen binding capacity of cell lines

The expression levels of the corresponding targets were determined using QIFIKIT® (Dako, K0078) and were presented as antigen binding capacity (ABC) as follows: for FOLR1, SK-OV-3 (90,000) and HT-29 (10,000); and for CEACAM5, MKN-45 (361,000), LS-180 (76,000), and HT-29 (4,500).

In vitro assays

FOLR1- and CEACAM5-TCBs were tested in a co-culture of freshly isolated and subsequently frozen human PBMCs from healthy donors and TA-bearing target cell lines.

Briefly, PBMCs were prepared by Histopaque density centrifugation of fresh blood obtained from healthy human donors. Fresh blood was diluted with sterile PBS and layered over a Histopaque gradient (Sigma, H889). After centrifugation (450 g, 30 min, room temperature (RT)), the plasma above the PBMC-containing interface was discarded, and PBMCs were transferred into a new Falcon tube subsequently filled with 50 mL of PBS. The mixture was centrifuged (400 g, 10 min, RT), the supernatant discarded and the PBMC pellet washed twice with sterile PBS (centrifugation steps 350 g, 10 min). The resulting PBMC population was counted automatically (ViCell) and stored in culture medium (RPMI 1640 medium (11875093) containing 10% fetal bovine serum (FBS; 26,140,079), 1% sodium pyruvate (11360070), 1% GlutaMax, 1% (35050061) non-essential-amino-acids (11140035); all purchased from Gibco) at 37°C and 5% CO₂ in the cell incubator until further use (no longer than 24 h). Target cells were harvested with TrypLE™ Express (Gibco, 12,605–010) washed and plated at a density of 30.000 cells/well using U-bottom 96-well plates (Greiner bio-one, 650,185). PBMCs were added to target cells at a final effector-to-target ratio (E:T) of 10:1, before introducing pre-diluted TCBs (serial dilution, maximum concentration 5–50 nM) in technical duplicates. The main readouts included anti-tumor TDCC, cytokine secretion, and T-cell activation at 24, 48, and 72 h, respectively.

Anti-tumor T cell-dependent cellular cytotoxicity

Target cell killing was assessed after an incubation for 24, 48, and 72 h under the previously mentioned conditions, by the quantification of protease released into cell supernatants by apoptotic/necrotic cells using the CytoTox-Glo™ Cytotoxicity Assay (Promega, G9290) according to the manufacturer’s instruction. Data was acquired using the PerkinElmer Envision 2104 Multilabel Reader as relative luminescence units. To seamlessly compare the response across different time-points, data were plotted as time-independent AUCE for each tested TCB concentration. This approach was adapted from Van De Vyver et al.⁶³

T cell activation assessment by flow cytometry

Activation of CD8 and CD4 T cells upon anti-tumor (TDCC) mediated by the TCBs was assessed by flow cytometry using antibodies recognizing the T cell activation markers CD25 and CD69. After 24-, 48- and 72-h incubations, co-cultured cells were centrifuged at (300 g, 5 min) and washed twice with fluorescence-activated cell sorting (FACS) buffer (produced in-house). Surface staining of obtained single cells was performed using FACS buffer containing the following antibodies: CD4 FITC (BioLegend, 317,408), CD8 BV605 (BioLegend, 301,040), CD45 Alexa Fluor 700 (BioLegend, 368,514), CD25 BUV395 (BD, 564,034), and CD69 APC-Cy7 (BioLegend, 310,914), according to the manufacturers’ instructions. Following surface staining of T cell markers, intracellular staining for Granzyme B Alexa Fluor 647 (BioLegend, 515,406) was performed as per the manufacturer’s instructions. Cells were washed twice with 150 μl of FACS buffer per well. After the final centrifugation step, samples were resuspended in 150 μl of FACS buffer. Finally, the samples were acquired using the BD FACSymphony™ A3 Cell Analyzer, and the results were analyzed using FlowJo.

Cytokine secretion assessment

The supernatant was captured, and subsequently frozen at −80°C for further analysis was assessed using the Bio-Plex Pro Human Cytokine 8-plex Assay kit (Bio-Rad, M50000007A) according to the manufacturer’s instructions. This assay was performed at the same time-points described above. Briefly, the beads provided by the manufacturer were seeded before the supernatant, controls, and standards were added to the wells. After an incubation step (on the shaker, 30 min, RT), the samples were washed three times using the wash buffer provided, before adding the detection antibody provided in the kit. The same incubation step took place before washing the samples three times and adding the detecting agent (Streptavidin-PE) provided in the kit. After a final incubation step, the cells were washed three times and resuspended in 125 μl of the assay buffer provided. The samples resuspended were acquired via the Bio-Plex 200 system. Similar to the anti-TDCC assessment, data were plotted as time-independent AUCE for each tested TCB concentration. This approach was adapted from Van De Vyver et al.⁶³

IS formation assessment and imaging flow cytometry

Our method for studying synapse formation between immune and target cells was adapted from Boushehri et al.⁴² Human CD8⁺ T cells were first isolated from PBMCs of healthy human donors using the EasySep™ Human CD8⁺ T Cell Isolation Kit from STEMCELL Technologies (cat # 17,953). In the meantime, SK-OV-3 target cells were labeled with CTV (ThermoFisher, C34557) according to the manufacturer’s instructions using a concentration of 0.5 µM. Afterward, live/dead staining of CD8⁺ T and SK-OV-3 cells was performed separately using the fixable viability dye eF780 for 15 min at RT (eBioscience, 65–0865-14). Cells were then re-suspended in RPMI 1640 medium supplemented with 10% FBS (PAN-Biotech, P30-2006), 5% Penicillin-Streptomycin (Gibco, 15,140–122) and 2 mM L-glutamine (PAN-Biotech, P04-80100), and then transferred at an E:T cell ratio of 1:1 into a well of a 96-well round bottom plate (250.000 cells per well). Subsequently, TCBs were added at a concentration of 10 nM to the cells. To strengthen the conjugate formation between target and effector cells, the plate was centrifuged (300 g, 1 min), then incubated at 37°C for 45 min. The cells were then fixed immediately for 15 min at RT followed by permeabilization using the Foxp3/Transcription factor staining buffer set (eBioscience, 00–5523-00). Intracellular staining was performed for 40 min at 4°C in permeabilization buffer containing antibodies and fluorescent dyes: CD8-PE (clone SK1; Biolegend, 344,706), Phalloidin AF594 (ThermoFisher, A12381), GrzmB-AF488 (QA18A28; Biolegend, 396,424), and P-CD3ζ Y142-AF647 (K25-407.69, BD, 558,489). After washing, the cells were resuspended in 30 µl of FACS buffer (PBS supplemented with 2% FBS). Cells were acquired on an Amnis ImageStream^X Mark II Imaging Flow Cytometer (Luminex). Images of 15.000 cells were collected per sample at 60x magnification on a low-speed setting. IDEAS software (version 6.2.187.0, EMD Millipore) was used for data analysis. To identify and quantify immune synapses the gating strategy was adapted from Boushehri et al.⁴² Briefly, cells were first gated on in-focus live⁺ CD8⁺ CTV⁺ cells. From this population, images of single CD8⁺ T cells and single SK-OV-3 cells were selected using the area and aspect ratio feature. Next, to exclude non-interacting cells, the CD8 intensity within a self-created synapse mask was determined. The synapse mask was defined as a combination of the morphology CD8 and CTV mask with a dilation of 2. Only synapses that indicated a CD8 signal in the mask were gated. Finally, images showing CD8⁺-SK-OV-3 cells in one layer were excluded by using the height and area feature of the brightfield, ensuring accurate identification of interacting single CD8⁺-SK-OV-3 immune synapses (Figure 3B).

Mouse models

Female humanized mice (CD34⁺ NSG and BRGS-CD47, both purchased from Jackson Laboratory) were used in assessing the efficacy, cytokine release, and PK properties of FOLR1-TCBs and CEACAM5 TCBs, respectively. Additionally, female huCD3xhuCEACAM5 transgenic mice (purchased from Charles Rivers) were used to assess cytokine release and PK properties of CEACAM5 TCBs when the target is expressed in normal tissues. Mice were maintained under specific-pathogen-free conditions according to committed guidelines (GV-Solas; Felasa; TierschG). After their arrival, the animals were maintained for a week to acclimate them to the new environment. During that time, health and wellbeing monitoring was carried out daily.

Tumor models

1×10⁶ cells of the FOLR1⁺ patient-derived xenograft, BC004 (purchased from OncoTest, Freiburg, Germany) or the CEACAM5⁺ cell line, HPAF-II (ATCC) were injected subcutaneously at a volume of 100 μl on the left flank. The injection volume of both tumors consisted of growth factor-reduced (GFR) Matrigel (Corning, 356,230) and RPMI 1640 medium (Gibco, 11,875,093) of equal ratios.

In vivo efficacy assessment

Treatment started after the tumor volume reached 150–200 mm³. Tumor volume was measured three times per week using a caliper according to the following formula: tumor volume (1⁄2 [length × width²]). Body weight measurement was performed twice per week. After mice randomization, mice were injected with corresponding TCBs (i.v., tail vain, 200 μl) once a week for the course of treatment (4 treatment cycles). The vehicle control group was injected with a histidine buffer (20 mM histidine, 140 mM NaCl, 0.01% Tween 20, pH 6.0). All mice were terminated 2 d after the fourth treatment. Our dosing strategy was guided by previous studies, including those of other established TCBs such as glofitamab,⁶⁴ which provided a strong foundation for determining efficacious doses for FOLR1-TCBs. For CEACAM5-TCBs, we implemented a one-fold increase in the dose (1 mg/kg instead of the 0.5 mg/kg used for FOLR1-TCBs) to account for the potential reduced activity observed with the 0.5 mg/kg dose in the context of the affinity-attenuated binders. Mice were bled (iv) for cytokine and PK assessment at different time-points (2–3 mice bled per time point) either after the first (4 h) or the first and second treatment cycles (1 h-168 h post treatment). Based on previous experience with the mouse model used, the peak of cytokine secretion consistently occurs after the first therapy, with subsequent therapies showing lower or no detectable cytokine secretion. Blood was collected in micro sample tubes containing serum gel (SARSTEDT, 41.1500.005) and was centrifuged (15.000 g, 10 min, 4°C), before freezing the serum samples at −80°C. Cytokine assessment was performed using the Bio-Plex Pro Human Immunotherapy Panel, 20-Plex kit (Bio-Rad, 12,007,975) following the manufacturer’s instructions, with data acquisition conducted on the Bio-Plex 200 system.

Ex vivo analysis of tumors via flow cytometry

Three mice per group (also known as scouts) were euthanized 3 d after the second infusion to analyze intra-tumoral infiltration of T cells within efficacy experiments. Unlike cytokine secretion, intra-tumoral T cell infiltration requires longer time points in order to occur and be measured. Time points after at least the second therapy have been found optimum in this context.⁶⁵ Briefly, tumors were cut into small pieces and then transferred into GentleMACS C-tubes (Miltenyi Biotec, 130–093-237) containing RPMI 1640 (Gibco), collagenase D (1 mg/ml; Sigma-Aldrich/Roche, 11,088,858,001), and DNase I (50 μg/ml, Sigma-Aldrich/Roche, 4,716,728,001), and dissociated using a GentleMACS Octo Dissociator (Miltenyi Biotec). Samples were incubated at 37°C for enzymatic digestion, and the GentleMACS was used to obtain a single-cell suspension. Each dissociated tumor suspension was filtered through a 70 μm cell strainer and washed.⁴⁷ Surface staining of obtained single cells was performed using FACS buffer containing the following antibodies: CD4 FITC (BioLegend, 317,408), CD8 BV605 (BioLegend, 301,040), CD45 Alexa Fluor 700 (BioLegend, 368,514), CD25 BUV395 (BD, 564,034), and CD69 APC-Cy7 (BioLegend, 310,914), according to the manufacturers’ instructions. Following surface staining of T cell markers, intracellular staining for Granzyme B Alexa Fluor 647 (BioLegend, 515,406) was performed as per the manufacturer’s instructions. Cells were washed twice with 150 μl of FACS buffer per well. After the final centrifugation step, samples were resuspended in 150 μl of FACS buffer. Finally, the samples were acquired using the BD FACSymphony™ A3 Cell Analyzer, and the results were analyzed using FlowJo.

In vivo assessment of the impact of CEACAM5-TCBs on normal tissue

The huCD3xhuCEACAM5 transgenic mouse model, which exhibits target expression, among others, in the GI tract and human CD3 on circulating mouse T cells, was used for the assessment of CEACAM5-TCBs. Six mice per treatment group were administered the respective TCB iv at 0.5 mg/kg dose, except for the CD3^high group which was dosed with 0.1 mg/kg for tolerability reasons. Additionally, the vehicle control group received a histidine buffer administered iv. Mice were bled (iv) for cytokine and PK assessment at different time-points (4 h–168 h) after the single dose treatment (2–3 mice bled per time point). Seven days after a single dose treatment, all mice were euthanized, and the relevant organs were collected for histopathological assessment. Tissues collected at necropsy were fixed in 10% buffered formalin immediately after necropsy, followed by trimming and paraffin embedding. The histopathological assessment was conducted on histological sections stained with Hematoxylin and Eosin by a Board-Certified Pathologist.

Cytokine assessment was performed using the Bio-Plex Pro Mouse Chemokine Panel 31-Plex kit (Bio-Rad, 12,009,159) following the manufacturer’s instructions, with data acquisition conducted on the Bio-Plex 200 system.

In vivo PK assessment

The mouse serum samples were analyzed using an ELISA assay using a biotinylated version of the anti-PG antibody, an anti-Fc(P329G) IgG (an antibody that specifically binds human IgG₁ Fc(P329G), the backbone of used TCBs; produced in-house). After that, the samples were acquired using the TECAN Freedom EVO robotic system (TECAN life sciences). The PK evaluation was conducted using the standard non-compartmental analysis.

In addition to TCB exposure measured in this assay, additional PK parameters were calculated using Phoenix® version 8.3.4.295 (Certara).

In-vitro 3D co-culture using healthy jejunal organoids

Human intestinal tissue samples were obtained, and experimental procedures performed by Human Tissue and Cell Research (HTCR)-Services GmbH, within the framework of its biobanking service that operates on behalf of the nonprofit HTCR Foundation (Munich, Germany), including informed patient consent. The framework of the HTCR Foundation has been approved by the ethics commission of the Faculty of Medicine in the Ludwig Maximilian University (no. 025–12) and the Bavarian State Medical Association (no. 11,142).

Assay set-up

Healthy human jejunal organoids were co-cultured with human PBMCs from a healthy donor. CEACAM5 expression was confirmed using immunohistochemistry. CEACAM5-TCBs, containing CD3 binders of varying affinities and high or low CEACAM5 binders, were then added. TCBs were diluted to concentrations of 50 nM, 5 nM, and 0.5 nM and introduced to the co-culture, which was incubated for 72 h in a medium containing 1:1,000 Cell Event Caspase 3/7 Detection Reagent (Invitrogen, C10423).⁶⁶ Each condition was tested in technical triplicates. The primary readouts included cytokine secretion and organoid cytotoxicity measurements.

Imaging

Apoptosis indicated by the Cell Event Caspase 3/7 Detection Reagent was assessed by imaging the plates at 72 h. Samples were imaged in confocal mode with a 5x Air Objective on the Opera Phenix (PerkinElmer) covering approximately a 325 µm z-stack, starting at −120 µm. 25 µm was set as the minimum distance between z-stacks with autofocus, two-peak; binning, 2. Bright-field and AlexaFluor 488 were selected as channels and per well, four fields were acquired. CO2 was set to 5% and temperature to 37°C. Fluorescence signal intensity of the Caspase 3/7 was quantified using Opera Harmony software v.4.9 (PerkinElmer). Briefly, segmentation of organoids was done using ‘Find Image Region’ based on the bright-field signal only, followed by ‘Select Region’ and ‘Select Population.’ Next, the Caspase 3/7 fluorescence signal on AF488 was determined using ‘Calculate Intensity Properties.’

LDH assay

Supernatants were collected after 72 h of incubation, and a Roche cytotoxicity detection kit was applied per the manufacturer’s instructions. A standard curve was prepared, and supernatants were diluted in PBS and incubated with the reaction mix in the dark (30 min, RT). Absorbance at 490 nm was then measured using a PerkinElmer EnVision 2104 Multilabel Reader.⁶⁷

Cytokine secretion assessment

Supernatants were captured and subsequently frozen at −80°C for further analysis and were assessed using the Bio-Plex Pro Human Cytokine 8-plex Assay kit (Bio-Rad, M50000007A) according to the manufacturer’s instructions after 72 h of incubation.

Statistical analysis

Statistical significance of the in vivo TGI curves was determined using an unpaired, 2-tailed t-test with Welch’s correction. Statistical significance for the ex vivo tumor analysis and cytokine release assessment was computed using one-way ANOVA with Tukey’s multiple comparison test, performed with GraphPad Prism 10 software. Statistical significance was computed using two-way ANOVA for the in vitro IS formation assay as well as the cytokine release assessment in the huCD3xhuCEACAM5 transgenic mouse model. Significance levels were indicated as follows: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001. Figures were plotted using GraphPad Prism 10.

Funding Statement

This work was supported by the Funding for this research was provided by F. Hoffmann-La Roche Ltd..

Disclosure statement

All authors, with the exception of A. Odermatt, are or were employees of Roche at the time this study was conducted and may hold stock or stock options in the company. O. Abdelmotaleb, A. Schneider, I. Waldhauer, J. Sam, T. Hofer, M. Lechmann, A. Freimoser-Grundschober, A.M. Giusti, C. Gassner, S. Maersch, A. Bransi, P. Bruenker, S. Colombetti, and C. Klein also report holding patents and/or royalties from Roche. A. Odermatt declares no competing interests.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2025.2584374