eIg-based bispecific T-cell engagers targeting EGFR: Format matters

ABSTRACT

Bispecific antibodies are molecules with versatile modes of action and applications for therapy. They are commonly developed as T-cell engagers (TCE), which simultaneously target an antigen expressed by tumor cells and CD3 expressed by T-cells, thereby inducing T-cell-mediated target cell killing. There is growing evidence that the molecular composition and valency for the target antigen influence the activity of TCEs. Here, the eIg platform technology was used to generate a set of bispecific TCEs targeting epidermal growth factor receptors (EGFR) and CD3. These molecules either included or lacked an Fc region and exhibited one binding site for CD3 and either one or two binding sites for EGFR (1 + 1 or 2 + 1 formats) utilizing different molecular arrangements of the binding sites. In total, 11 different TCE formats were analyzed for binding to target cells and T cells, T cell-mediated killing of tumor cells, and for the activation of T cells (release of cytokines and proliferation of T-cells). Bivalent binding to EGFR strongly increased binding and T cell-mediated killing. However, the molecular composition and position of the CD3-binding arm also affected target cell killing, cytokine release, and T-cell proliferation. Our findings support that screening of a panel of formats is beneficial to identify the most potent bispecific TCE, and that format matters.

KEYWORDS:

• Bispecific antibodies
• CD3
• EGFR
• T-cell retargeting
• valency

Introduction

Bispecific antibodies are powerful molecules to redirect effector cells of the immune system to tumors. This includes bispecific antibodies targeting a tumor-associated antigen (TAA) on tumor cells and, for example, CD3 on T cells or CD16 expressed by natural killer cells and macrophages.¹ These molecules thereby bring target and effector cells in close proximity, resulting in the formation of an immunological synapse that leads to the release of effector proteins (e.g., granzyme B and perforin) by the effector cell and killing of the tumor cell. A plethora of different bispecific antibody formats targeting TAAs expressed by tumor cells of hematological malignancies or solid tumors have been developed and are in different stages of preclinical development and clinical trials.^2–4 The majority of these molecules utilize CD3 expressed by T cells as trigger molecules to induce T cell-mediated target cell destruction.^5,6 So far, five of these T-cell engagers (TCEs) have been approved for cancer therapy. The first approved TCE was catumaxomab, a hybrid mouse-rat IgG molecule targeting EpCAM and CD3.⁷ Catumaxomab was approved in the European Union (EU) in 2009, but then withdrawn from the market in 2017 for economic reasons. Blinatumomab, a bispecific T-cell engager (BiTE) targeting CD19 and CD3 with an Fc-less tandem scFv format, was approved in the United States (US) in 2014 for the treatment of acute lymphoblastic leukemia (ALL).⁸ In 2022, three additional bispecific TCEs were approved in both the EU and the US, teclistimab targeting BCMA and CD3 for the treatment of multiple myeloma, mosunetuzumab directed against CD20 and CD3 for the treatment of follicular lymphoma, and tebentafusp, an HLAA*02:01-gp100×CD3 ImmTAC for the treatment of uveal melanoma.^9,10

Reducing the risk of severe side effects that can be caused by bispecific TCE used to treat solid tumors is a major challenge. These side effects include the systemic activation of the immune system resulting in cytokine release syndrome (CRS) and “on-target/off-tumor” activities leading to cytotoxicity in healthy tissue with regular target expression.¹¹ Achieving a broad therapeutic window by maintaining a balance between efficacy and severe side effects is a crucial aspect of the development of novel TCE that must be addressed early in the development process.

The differential expression of TAAs in healthy and malignant tissue makes receptors like epidermal growth factor receptor (EGFR) a potential target for the development of novel cancer immunotherapies. EGFR, a member of the ErbB receptor family, is a membrane-bound tyrosine kinase that is overexpressed in a variety of epithelial tumors, e.g., colorectal and lung cancer, correlating with a poor clinical outcome for the patients.¹² Two different classes of molecules have been approved for the treatment of tumors overexpressing EGFR: 1) monoclonal antibodies inhibiting the signaling of the receptor by competing with its ligands, and 2) intracellularly acting tyrosine kinase inhibitors (TKI). Both classes of molecules rely on the effective inhibition of downstream signaling pathways that are activated upon ligand binding followed by receptor dimerization, phosphorylation of tyrosine residues in the signaling domain and activation of downstream signaling pathways by binding of adaptor molecules to the phosphorylated receptor dimers.^13,14 Treatment with EGFR-targeting drugs is restricted to cancers that are not mutated in signaling pathways downstream of EGFR. For example, cetuximab is approved for metastatic colorectal cancer that is wild-type for the downstream signaling molecule KRAS.^15,16

TCEs are capable of bypassing these intracellular resistance mechanisms, and several EGFR-targeting TCEs are currently being investigated in preclinical and clinical trials.⁶ Recent studies have shown that the antibody format strongly influences the activity when designing TCEs underlining the importance of identifying a suitable format to fulfill the desired characteristics.^17–20 For example, 2 + 1 formats with two binding sites for the TAA and one for CD3, to avoid nonspecific activation of T-cells in the absence of TAA-expressing tumor cells,^21,22 have been shown to achieve an avidity-mediated specificity gain for tumors with elevated expression of the TAA.^19,23–25

We recently established our eIg technology as a versatile platform to generate bispecific antibodies of different geometry and valency.²⁶ Here, we applied this eIg technology to generate TCEs targeting EGFR and CD3, modifying valency, geometry, and size of the molecules. These molecules comprise a Fab-like anti-CD3 buildings block (eFab), substituting the C_H1 and C_L domains by heterodimerizing EHD2 (hetEHD2) domains derived from human IgE heavy-chain domain 2, combined with an EGFR-targeting Fab into bi- or trivalent molecules with or without a silenced Fc region. The presence or absence of an Fc region in a fusion protein can affect the activity (e.g., distance between cells in immunological synapse), the pharmacokinetic properties due to FcRn-mediated recycling, as well as the stability and manufacturability. In total, 11 different formats were generated and analyzed for their activity regarding antigen binding, cytotoxicity, and T-cell activation with the aim to identify the most potent bispecific antibody format.

Results

Generation of 1 + 1 eIg and 2 + 1 Fab-eIg variants with N-terminal fusions of Fab or eFab

Based on the eIg technology,²⁶ Fc-comprising, bispecific antibodies for T-cell retargeting were designed with different valencies and geometries. The TAA EGFR is targeted by a humanized variant of cetuximab (hu225)²⁷ and CD3 on T cells by a humanized variant of UCHT1 (huU3). A silenced Fc region (Δab)²⁸ containing the knob-into-hole mutations²⁹ was used as heterodimerization module. While the CD3-binding arm is formed by an eFab using engineered heterodimerization domains derived from human IgE heavy-chain domain 2 (hetEHD2), the EGFR-binding arm is a classical Fab containing C_H1/C_Lκ. The hetEHD2 domains fused to the variable domains of the CD3 antibody differed regarding their N-glycosylation at position 39. The hetEHD2 (EHD2–2) fused to the V_LCD3 is non-glycosylated, resulting from a N39Q mutation, while the hetEHD2 (EHD2–1) fused to the V_HCD3 still contains the N-glycosylation site at position N39. By fusing the anti-CD3 eFab to a Fc(knob) and the anti-EGFR Fab to a Fc(hole), a 1 + 1 eIg was generated (e11-1-1). This molecule was used as a basis to generate three different 2 + 1 Fab-eIg molecules by fusing additional anti-EGFR Fab to the N-termini of the heavy chains, resulting in three trivalent, bispecific antibodies with different geometries (e21-1-1, e21-1-2, and e21-1-3) (Figure 1 a).

Figure 1. Biochemical characterization and binding properties of Fc-comprising 1 + 1 eIg and 2 + 1 Fab-eIg variants with N-terminal fusions of Fab or eFab. (a) Schematic illustration of 1 + 1 eIg (e11-1-1) and 2 + 1 Fab-eIg variants (e21-1-1, e21-1-2, e21-1-3). Nomenclature: e: bispecific antibody containing an eFab; first two numbers referring to valency: 11 - monovalent binding to EGFR (blue) and CD3 (red)/21 - bivalent binding to EGFR (blue) and monovalent binding to CD3 (red)/third number referring to presence of Fc-part: 1 - with Fc-part; fourth number referring to geometry: continuous number. The constant domains of the heavy chain are shown in gray, the constant domain of the light chain in white and the hetEHD2 domain in orange. (b) Size-exclusion chromatography by HPLC after preparative size-exclusion chromatography by FPLC. (c) SDS-PAGE analysis under reducing (red.) and non-reducing (n.r.) conditions. (d) EGFR-binding ELISA using 3 µg/mL EGFR-ECD-moFc as antigen and an HRP-conjugated antibody specific for human Fc (Mean ± S.D; n = 3). (e) Cell binding to CD3-expressing Jurkat cells analyzed by flow cytometry using a RPE-conjugated antibody specific for human Fc (Mean ± S.D; n = 3; MFI: median fluorescence intensity).

Set and QC-analysis of 1+1 eIg and of N-terminal fused 2+1 Fab-eIg variants.

By transient co-transfection of the four chains of each bispecific antibody into HEK293-6E cells, the antibodies were produced and purified by protein A chromatography followed by a preparative size-exclusion chromatography (SEC) if necessary to remove higher molecular weight (HMW) species (Figure 1 b). All molecules exhibited a purity of >95% (Table 1). The presence of the two heavy and two light chains, as well as their correct assembly, was confirmed by SDS-PAGE under reducing and non-reducing conditions (Figure 1 c). While all four bispecific antibodies contain two common light chains, each molecule contains a unique combination of two different heavy chains (see Figure 1 a).

Table 1. Productivity and integrity of 1 + 1 eIg and 2 + 1 Fab-eIg variants with N-terminal fusions of Fab or eFab (after preparative SEC).

	e11-1-1	e21-1-1	e21-1-2	e21-1-3
Molecular weight [kDa]	147	194	194	194
Yield [mg/L]	4.4	3.7	2.1	1.1
Purity after preparative SEC [%]	99.6	99.1	95.7	98.2
EC50 EGFR-binding (ELISA) [nM]	1.18 ± 0.52	0.64 ± 0.32	0.74 ± 0.35	0.94 ± 0.41
EC50 CD3-binding (flow cytometry) [nM]	3.3 ± 2.3	39.3 ± 26.6	4.5 ± 2.9	2.3 ± 1.6

Binding of the bispecific molecules to their targets EGFR and CD3 was confirmed by EGFR-binding enzyme-linked immunosorbent assay (ELISA) (Figure 1 d) and by binding to CD3-expressing Jurkat cells using flow cytometry (Figure 1 e). Increasing the valency for EGFR from one binding site (e11-1-1) to two binding sites (e21-1-1 to e21-1-3) resulted in a less than 2-fold increased binding to EGFR in ELISA. In contrast, flow cytometry analysis showed different binding to CD3 with EC₅₀ values between 2.3 and 4.5 nM for e11-1-1, e21-1-2, and e21-1-3, while an about 9- to 17-fold reduced binding was observed for e21-1-1 (39.3 nM). This reduced binding can be explained by a steric interference of binding of the anti-CD3 eFab by the anti-EGFR Fab that is fused to the N-terminus of the anti-CD3 eFab in e21-1-1.

T-cell retargeting by 1 + 1 eIg and 2 + 1 Fab-eIg variants with N-terminal fusions of Fab or eFab

Cell binding (Figure 2 a, d) and cytotoxicity in co-culture assays with human peripheral blood mononuclear cells (PBMCs) (Figure 2 e, h) was analyzed for a panel of four different cancer cell lines with different EGFR expression levels: (1) Squamous cell carcinoma cell line FaDu expressing high levels of EGFR (188,000 ± 26,000 receptors/cell), (2) colorectal cancer cell line LIM1215 expressing intermediate levels of EGFR (55,300 ± 4,700 receptors/cell), (3) breast cancer cell line MCF-7, and (4) colorectal cancer cell line SW620 expressing low levels of EGFR (<11,000 receptors/cell) (Figure S1/Table S1). All 2 + 1 Fab-eIg variants (e21-1-1 to e21-1-3) bound with comparable sub-nanomolar EC₅₀ values to these cell lines due to avidity effects, while reduced binding was observed for the 1 + 1 eIg (e11-1-1) molecule (Table 2). For binding to FaDu, a 2-fold increased binding was observed for the 2 + 1 formats compared to the 1 + 1 format (Figure 2 a). This valency effect was increased when the EGFR expression levels of tumor cells were lower, resulting in an 8- to 17-fold increased binding for the trivalent (e21-1-1 to e21-1-3) molecules compared to the bivalent e11-1-1 molecule using LIM1215, MCF-7, and SW620 cells (Figure 2 b, d). Additionally, the binding efficacy of the 2 + 1 Fab-eIg variants was strongly increased compared to the 1 + 1 eIg.

Figure 2. Cell binding to cancer cell lines, cytotoxicity and T-cell activation of Fc-comprising 1 + 1 eIg and 2 + 1 Fab-eIg variants with N-terminal fusions of Fab or eFab. (a – d) Binding to (a) FaDu, (b) LIM1215, (c) MCF-7 and (d) SW620 analyzed by flow cytometry using a R-PE-conjugated antibody specific for human Fc (Mean ± S.D; n = 3; MFI: median fluorescence intensity). (e – h) Cytotoxic potential of PBMCs co-cultured with the cancer cell lines (e) FaDu, (f) LIM1215, (g) MCF-7 and (h) SW620 in an effector-to-target ratio of 5:1 (Mean ± S.D.; n = 3 - three individual donors). (i + j) Release of (i) IL-2 after 24 h and (j) IFNγ after 48 h by PBMCs co-cultured with FaDu using an effector-to-target ratio of 5:1 analyzed by sandwich ELISA (Mean ± S.D.; n = 3 - three individual donors). (k + l) Proliferation of (k) CD4⁺ and (l) CD8⁺ T-cells determined by CFSE dilution in flow cytometry from co-culture assays with FaDu using an effector-to-target ratio of 5:1 (Mean ± S.D.; n = 3 - three individual donors).

Tumor cell binding and T-cell-mediated activity of 1+1 eIg and of N-terminal fused 2+1 Fab-eIg variants.

Table 2. Bioactivity of 1 + 1 eIg and 2 + 1 Fab-eIg variants with N-terminal fusions of Fab or eFab (n.D. – not determinable).

		e11-1-1	e21-1-1	e21-1-2	e21-1-3
EC50 cell binding [pM]	FaDu	665 ± 219	280 ± 28	321 ± 88	323 ± 138
	LIM1215	2,225 ± 1,400	147 ± 55	249 ± 91	187 ± 76
	MCF-7	456 ± 185	32 ± 7	54 ± 14	57 ± 25
	SW620	215 ± 143	14 ± 8	20 ± 11	13 ± 7
EC50 cytotoxicity [pM]	FaDu	50 ± 59	2 ± 2	3 ± 2	10 ± 10
	LIM1215	432 ± 214	5 ± 2	7 ± 2	27 ± 13
	MCF-7	n.d.	12 ± 7	33 ± 17	n.d.
	SW620	n.d.	102 ± 45	228 ± 75	n.d.
EC50 T-cell activation [pM]	IL-2	7,688 ± 9,730	27 ± 29	22 ± 33	169 ± 231
	IFNγ	593 ± 831	27 ± 19	15 ± 22	210 ± 300
	CD4^+ proliferation	n.d.	2 ± 1	1 ± 1	23 ± 8
	CD8^+ proliferation	23 ± 9	4 ± 1	2 ± 1	39 ± 12

Strong cytotoxicity was observed in co-culture assays with human PBMCs for all four molecules with FaDu and LIM1215 cells (Figure 2 e, f). Here, the 2 + 1 Fab-eIg variants e21-1-1 and e21-1-2 performed equally well, while cytotoxicity was lower for variant e21-1-3. Furthermore, e11-1-1 with only one EGFR binding site exhibited a 17- to 62-fold reduced cytotoxicity on FaDu and LIM1215, respectively, compared to e21-1-1 and e21-1-2. Strong cytotoxicity on MCF-7 cells with low EGFR expression levels was still observed on for e21-1-1 and e21-1-2 (EC₅₀ 12 ± 7 and 33 ± 17 pM; Figure 2 g). Some residual cytotoxicity for these molecules was observed at 1 nM for SW620 with the lowest EGFR levels (Figure 2 h). In contrast, no or only marginal cytotoxicity was observed for e11-1-1 and e21-1-3 for concentrations up to 100 nM and 1 nM, respectively. Taken together, e21-1-1 and e21-1-2 were the most potent molecules regarding cytotoxicity, showing only limited dependency on EGFR expression, while e11-1-1 and e21-1-3 were less potent but exhibited a dependency on EGFR expression levels.

To analyze the activation of T cells by the four molecules, cytokine release assays (IL-2 and IFNγ) and a T-cell proliferation assay were performed with co-culture of PBMCs and FaDu (Figure 2 i, l). Both IL-2 and IFNγ release were stronger for e21-1-1 and e21-1-2 compared to e11-1-1 and e21-1-3. At a concentration of 1 nM, where all molecules showed strong cytotoxicity on FaDu cells (>60% killing), the released IL-2 concentration was 6- to 3-fold lower for e11-1-1 and e21-1-3 (180 pg/mL and 290 pg/mL) compared to e21-1-1 and e21-1-2 (920 pg/mL and 1,010 pg/mL), respectively (Figure 2 i). This was also observed in the IFNγ release assay, where e21-1-1 and e21-1-2 (2,350 pg/mL and 2,040 pg/mL) showed 3- to 4-fold increased cytokine release compared to e11-1-1 and e21-1-3 (740 pg/mL and 600 pg/mL) (Figure 2 j). Furthermore, the 2 + 1 Fab-eIg variants e21-1-1 and e21-1-2 showed comparable T-cell proliferation for both CD4⁺ and CD8⁺ T-cells, while the activity was reduced 20- to 23-fold for e21-1-3 (Figure 2 k, l). For T-cell proliferation, a target-dependent activation by all bispecific antibodies was shown (Figure S2 a, b). In summary, cytokine release and T-cell proliferation were highest for e21-1-1 and e21-1-2, while reduced activities was observed for the e11-1-1 and e21-1-3.

Generation of 2 + 1 eIg-Fab variants with C-terminal fusions of Fab or eFab

As the geometry of the 2 + 1 Fab-eIg molecules strongly influenced the activity of bispecific antibodies for T-cell retargeting to EGFR-expressing cancer cell lines, three additional trivalent, bispecific antibodies with two binding sites for EGFR and one binding site for CD3 were generated. Therefore, anti-EGFR Fab or anti-CD3 eFab fragments were fused to the C-termini of the light (e21-1-4) or heavy chain (e21-1-5, e21-1-6) (Figure 3 a). After protein A purification, molecules exhibited a purity of >94% (Figure 3 b, Table 3). SDS-PAGE analysis (Figure 3 c) under reducing and non-reducing conditions confirmed the presence of the different polypeptide chains and correct assembly (see Figure 3 a).

Figure 3. Biochemical characterization and binding properties of Fc-comprising 2 + 1 eIg-Fab variants with C-terminal fusions of Fab or eFab. (a) Schematic illustration of 2 + 1 eIg-Fab variants (e21-1-4, e21-1–5, e21-1-6). Nomenclature: e: bispecific antibody containing an eFab; first two numbers referring to valency: 21 - bivalent binding to EGFR (blue) and monovalent binding to CD3 (red)/third number referring to presence of Fc-part: 1 - with Fc-part; fourth number referring to geometry: continuous number. (b) Size-exclusion chromatography by HPLC. (c) SDS-PAGE analysis under non-reducing (n.r.) and reducing (red.) conditions. (d) EGFR-binding ELISA using 3 µg/mL EGFR-ECD-moFc as antigen and an HRP-conjugated antibody specific for human Fc (Mean ± S.D; n = 3). (e) Cell binding to CD3-expressing Jurkat cells analyzed by flow cytometry using a R-PE-conjugated antibody specific for human Fc (Mean ± S.D; n = 3; MFI: median fluorescence intensity).

Set and QC-analysis of C-terminal fused 2+1 eIg-Fab variants.

Table 3. Productivity and integrity of 2 + 1 eIg-Fab variants with C-terminal fusions of Fab or eFab.

	e21-1-4	e21-1-5	e21-1-6
Molecular weight [kDa]	194	194	194
Yield [mg/L]	2.8	2.6	4.4
Purity after protein A [%]	95.2	98.5	94.3
EC50 EGFR-binding (ELISA) [nM]	0.41 ± 0.04	0.43 ± 0.04	0.59 ± 0.06
EC50 CD3-binding (flow cytometry) [nM]	11.2 ± 3.8	32.8 ± 11.6	5.6 ± 2.1

Binding of the three constructs to EGFR investigated by ELISA showed similar EC₅₀ values for all three 2 + 1 eIg-Fab variants independent of the geometry (Figure 3 d). In contrast, binding to CD3 on Jurkat cells showed differential-binding behavior (Figure 3 e), with strongest binding for e21-1-6 (EC₅₀ 5.6 ± 2.0 nM) followed by e21-1-4 (EC₅₀ 11.2 ± 3.8 nM) and e21-1-5 (EC₅₀ 32.8 ± 11.6 nM) (Table 3). This 3- and 6-fold reduced binding for e21-1-5 to CD3 in comparison to the other two variants can be explained, as for e21-1-1, by a steric interference for binding of the anti-CD3 eFab that is fused to the C-terminus of the Fc part.

T-cell retargeting by 2 + 1 eIg-Fab variants with C-terminal fusions of Fab or eFab

Binding of the 2 + 1 eIg-Fab variants with C-terminal fusions of Fab or eFab to EGFR-expressing cancer cell lines FaDu, LIM1215, MCF-7, and SW620 was independent of the geometry, but dependent on the EGFR expression levels (Figure 4 a – d).

Figure 4. Cell binding to cancer cell lines, cytotoxicity and T-cell activation of Fc-comprising 2 + 1 eIg-Fab variants with C-terminal fusions of Fab or eFab. (a – d) Binding to (a) FaDu, (b) LIM1215, (c) MCF-7 and (d) SW620 analyzed by flow cytometry using a R-PE-conjugated antibody specific for human Fc (Mean ± S.D; n = 3; MFI: median fluorescence intensity). (e – h) Cytotoxic potential of PBMCs co-cultured with the cancer cell lines (e) FaDu, (f) LIM1215, (g) MCF-7 and (h) SW620 in an effector-to-target ratio of 5:1 (Mean ± S.D.; n = 3 - three individual donors). (i + j) Release of (i) IL-2 after 24 h and (j) IFNγ after 48 h by PBMCs co-cultured with FaDu using an effector-to-target ratio of 5:1 analyzed by sandwich ELISA (Mean ± S.D.; n = 3 - three individual donors). (k + l) Proliferation of (k) CD4⁺ and (l) CD8⁺ T-cells determined by CFSE dilution in flow cytometry from co-culture assays with FaDu using an effector-to-target ratio of 5:1 (Mean ± S.D.; n = 3 - three individual donors).

Tumor cell binding and T-cell-mediated activity of C-terminal fused 2+1 eIg-Fab variants.

All three variants effectively induced cytotoxicity in co-culture assays with PBMCs for FaDu and LIM1215 (Figure 4 e, f). On FaDu, e21-1-4 was most effective, while e21-1-5 and e21-1-6 showed an 11- and 5-fold lower cytotoxicity, respectively (Figure 4 e). Similar results were observed on LIM1215 cells with a 10- or 3-fold decreased EC₅₀ value for e21-1-4 compared to e21-1-5 or e21-1-6, respectively (Figure 4 f). Cytotoxicity was dependent on the EGFR expression levels with less than 25% cytotoxicity at the highest analyzed concentration of 0.5 nM on MCF-7 and SW620 expressing low amounts of EGFR (Figure 4g, h).

Differential activation of T cells was observed depending on the geometry. e21-1-4 (740 pg/mL IL-2/4,000 pg/mL IFNγ) revealed the strongest release of IL-2 and IFNγ, while this was 5- and 3-fold reduced for e21-1-5 (140 pg/mL IL-2/1,470 pg/mL IFNγ) and 2.7- and 1.6-fold reduced for e21-1-6 (270 pg/mL IL-2/2,540 pg/mL IFNγ) at a protein concentration of 5 nM, respectively (Figure 4 i, j). In line, variant e21-1-4 induced the strongest proliferation for both CD4⁺ and CD8⁺ T-cells (Figure 4 k, l). The weakest induction of T-cell proliferation was observed for variant e21-1-5, being 5- and 9-fold reduced compared to e21-1-4 for CD4⁺ and CD8⁺ T-cells, respectively (Table 4). All three TCEs showed a target-dependent induction of T-cell proliferation (Figure S2 c, d).

Table 4. Bioactivity of 2 + 1 eIg-Fab variants with C-terminal fusions of Fab or eFab (n.d. – not determinable).

		e21-1-4	e21-1-5	e21-1-6
EC50 cell binding [pM]	FaDu	258 ± 59	256 ± 60	316 ± 68
	LIM1215	111 ± 65	103 ± 63	156 ± 87
	MCF-7	32 ± 40	22 ± 32	47 ± 54
	SW620	18 ± 6	10 ± 8	19 ± 9
EC50 cytotoxicity [pM]	FaDu	7 ± 4	75 ± 72	38 ± 28
	LIM1215	5 ± 3	51 ± 28	14 ± 6
	MCF-7	204 ± 203	n.d.	n.d.
	SW620	105 ± 140	n.d.	n.d.
EC50 T-cell activation [pM]	IL-2	9 ± 7	40 ± 34	45 ± 30
	IFNγ	35 ± 13	301 ± 143	132 ± 72
	CD4^+ proliferation	8 ± 3	37 ± 12	9 ± 4
	CD8^+ proliferation	2 ± 0	18 ± 4	7 ± 1

Of the three molecules, the most potent one regarding cytotoxicity, e21-1-4, also induced the strongest release of IL-2 and IFNγ and the most potent activity regarding T-cell proliferation. Variants e21-1-5 and e21-1-6 were less potent in inducing cytotoxicity, showed only limited cytotoxicity on cells with low EGFR expression, and a strongly reduced release of cytokines as well as reduced proliferation of T-cells.

Generation of Fc-less 1 + 1 and 2 + 1 Fab-eFab variants

To investigate the retargeting of T cells to EGFR-expressing tumor cells using smaller molecules, Fc-less Fab-eFab variants were generated (Figure 5 a). The bispecific Fc-less 1 + 1 Fab-eFab variant e11-0-1 is monovalent for binding to both EGFR and CD3. By fusion of one additional anti-EGFR Fab arm to the C- or N-termini of an anti-CD3 eFab, three 2 + 1 Fab-eFab variants were constructed being bivalent for binding to EGFR and monovalent for binding to CD3 (e21-0-1 to e21-0-3).

Figure 5. Biochemical characterization and binding properties of Fc-less 1 + 1 and 2 + 1 Fab-eFab variants. (a) Schematic illustration of 1 + 1 (e11-0-1) and 2 + 1 (e21-0-1, e21-0-2, e21-0-3) Fab-eFab variants. Nomenclature: e: bispecific antibody containing an eFab; first two numbers referring to valency: 11 - monovalent binding to EGFR (blue) and CD3 (red)/21 - bivalent binding to EGFR (blue) and monovalent binding to CD3 (red)/third number referring to presence of Fc-part: 0 - without Fc-part; fourth number referring to geometry: continuous number. (b) Size-exclusion chromatography by HPLC after preparative size-exclusion chromatography by FPLC. (c) SDS-PAGE analysis under non-reducing (n.r.) and reducing (red.) conditions. (d) EGFR-binding ELISA using 3 µg/mL EGFR-ECD-moFc as antigen and an HRP-conjugated antibody specific for human Fab (Mean ± S.D; n = 3). (e) Cell binding to CD3-expressing Jurkat cells analyzed by flow cytometry using an FITC-conjugated antibody specific for human Fab (Mean ± S.D; n = 3; MFI: median fluorescence intensity).

Set and QC-analysis of Fc-less 1+1 and 2+1 Fab-eFab variants.

The three individual chains of each Fab-eFab molecule were transiently co-transfected into HEK2936E cells, and the antibodies were purified from the supernatant by CaptureSelect C_H1-XL affinity chromatography. To remove both HMW and lower molecular weight (LMW) species, preparative SEC was included for all four molecules. Analytical SEC of the protein after two-step purification showed a purity of >90% for 1 + 1 Fab-eFab e1101 and the 2 + 1 Fab-eFab variant e21-0-1, while the 2 + 1 Fab-eFab variants e21-0-2 and e21-0-3 exhibited a purity of 88.3% and 80.2%, respectively, mainly due to some LMW species (Figure 5 b). SDS-PAGE analysis under reducing and non-reducing conditions confirmed the presence of the individual chains and assembly into intact molecules (Figure 5 c). For the bivalent e11-0-1, two chains comprising the variable domains of the light chains (V_LCD3-hetEHD2–2 and V_LEGFR-C_L) and one chain comprising both variable domains of the heavy chains (V_HEGFR-C_H1-V_HCD3-hetEHD2–1) are visible. The trivalent e21-0-1 to e21-0-3 are formed by one light chain (V_LEGFR-C_L) and two different chains comprising V_HEGFR-C_H1 fused either to V_LCD3-hetEHD2–2 or to V_HCD3-EHD2-1 (Figure 5 a). The additional chain of the trivalent molecules runs at approximately 50 kDa, slightly below the one chain that is shared with e11-0-1.

Binding of the Fab-eFab molecules to EGFR was investigated by ELISA (Figure 5 d). Compared to the 1 + 1 Fab-eFab, the 2 + 1 Fab-eFab variants being bivalent for EGFR exhibited a 7- to 11-fold increase of binding to EGFR (Table 5). Though all Fab-eFab variants consist of one binding site for CD3, strong differences for binding to CD3 were seen on CD3-expressing Jurkat cells (Figure 5 e). The strongest binding was observed for e21-0-3 (EC₅₀ 3.5 ± 3.2 nM), followed by e11-0-1 and e21-0-2 (EC₅₀ 11.3 ± 8.0 and 16.1 ± 12.0 nM) (Table 5). The 2 + 1 Fab-eFab variant e21-0-1 bound 32-fold weaker compared to the strongest binder e21-0-3. While no N-terminal fusion to the anti-CD3 eFab is present in e21-0-3, there are one (e11-0-1 and e21-0-2) or two (e21-0-1) anti-EGFR Fab fused to the N-termini of this eFab in the other three molecules, explaining the differential-binding behavior to CD3 due to steric hindrance. Similar effects for binding to CD3 were shown in prior results for the Fc-comprising molecules e21-1-1 and e21-1-5 where the CD3 binding eFab is sterically hindered for binding due to an N-terminal fusion of an anti-EGFR Fab or an Fc part, respectively.

Table 5. Productivity and integrity of Fc-less 1 + 1 and 2 + 1 Fab-eFab variants (after preparative SEC).

	e11-0-1	e21-0-1	e21-0-2	e21-0-3
Molecular weight [kDa]	97	145	145	145
Yield [mg/L]	0.8	1.0	0.9	1.0
Purity after preparative SEC [%]	91.5	91.6	88.3	80.2
EC50 EGFR-binding (ELISA) [nM]	2.02 ± 0.29	0.18 ± 0.04	0.21 ± 0.04	0.28 ± 0.04
EC50 CD3-binding (flow cytometry) [nM]	11.3 ± 8.0	111.5 ± 91.4	16.1 ± 12.0	3.5 ± 3.2

T-cell retargeting by Fc-less 1 + 1 and 2 + 1 Fab-eFab variants

Next, binding to cancer cell lines with different EGFR expression levels was investigated (Figure 6 a d). Strong valency effects were observed when comparing the 1 + 1 Fab-eFab e11-0-1 to the 2 + 1 Fab-eFab variants. On both FaDu and LIM1215, binding of e11-0-1 was 4- to 29-fold reduced when comparing EC₅₀ values to the 2 + 1 Fab-eFab variants (Figure 6 a, b). Weak binding of all Fc-less molecules was observed on MCF-7 and SW620 cell lines with low EGFR expression (Figure 6 c, d).

Figure 6. Cell binding to cancer cell lines, cytotoxicity and T-cell activation of Fc-less 1 + 1 and 2 + 1 Fab-eFab variants. (a – d) Binding to (a) FaDu, (b) LIM1215, (c) MCF-7 and (d) SW620 analyzed by flow cytometry using a FITC-conjugated antibody specific for human Fab (Mean ± S.D; n = 3; MFI: median fluorescence intensity). (e – h) Cytotoxic potential of PBMCs co-cultured with the cancer cell lines (e) FaDu, (f) LIM1215, (g) MCF-7 and (h) SW620 in an effector-to-target ratio of 5:1 (Mean ± S.D.; n = 3 - three individual donors). (i + j) Release of (i) IL-2 after 24 h and (j) IFNγ after 48 h by PBMCs co-cultured with FaDu using an effector-to-target ratio of 5:1 analyzed by sandwich ELISA (Mean ± S.D.; n = 3 - three individual donors). (k + l) Proliferation of (k) CD4⁺ and (l) CD8⁺ T-cells determined by CFSE dilution in flow cytometry from co-culture assays with FaDu using an effector-to-target ratio of 5:1 (Mean ± S.D.; n = 3 - three individual donors).

Tumor cell binding and T-cell-mediated activity of Fc-less 1+1 and 2+1 Fab-eFab variants.

The dependency of cytotoxicity on EGFR expression levels was investigated in co-culture assays with different cancer cell lines and PBMCs. All Fab-eFab constructs lead to an effective killing of cancer cells regardless of the EGFR expression levels at a maximum concentration of 500 pM (Figure 6 e h). The increased valency for binding to EGFR resulted in a 4- to 24-fold increased cytotoxicity on FaDu and LIM1215 matching with the valency effects observed for cell binding between the 1 + 1 and the 2 + 1 Fab-eFab variants (Figure 6 e, f; Table 6). Increased activity based on increased valency was observed in a similar range for the 2 + 1 Fab-eFab variants on MCF-7 and SW620 (Figure 6 g, h). Though all tested Fab-eFab variants are highly potent, also killing cells with low EGFR expression, a dependency on EGFR expression was seen for e11-0-1 (Figure 6 ef). Using a concentration of 50 pM revealed 70% and 50% cytotoxicity for FaDu and LIM1215 cells, respectively, while this concentration showed less than 10% cell death for MCF-7 or SW620 cells.

Table 6. Bioactivity of Fc-less 1 + 1 and 2 + 1 Fab-eFab variants (n.D. – not determinable).

		e11-0-1	e21-0-1	e21-0-2	e21-0-3
EC50 cell binding [pM]	FaDu	2,386 ± 478	228 ± 41	440 ± 76	597 ± 109
	LIM1215	2,667 ± 214	93 ± 13	203 ± 27	310 ± 38
	MCF-7	691 ± 456	22 ± 18	60 ± 34	137 ± 91
	SW620	545 ± 209	16 ± 10	19 ± 10	65 ± 25
EC50 cytotoxicity [pM]	FaDu	21 ± 11	5 ± 2	4 ± 2	1 ± 1
	LIM1215	47 ± 25	2 ± 1	2 ± 1	2 ± 1
	MCF-7	n.d.	14 ± 3	25 ± 7	37 ± 15
	SW620	n.d.	8 ± 4	21 ± 12	16 ± 7
EC50 T-cell activation [pM]	IL-2	28 ± 16	9 ± 9	6 ± 6	4 ± 4
	IFNγ	241 ± 128	39 ± 19	37 ± 21	40 ± 24
	CD4^+ proliferation	23 ± 9	4 ± 2	4 ± 2	3 ± 1
	CD8^+ proliferation	14 ± 4	2 ± 1	2 ± 1	1 ± 1

Activation of T cells by the Fab-eFab constructs was investigated in co-culture assays using PBMCs and FaDu cells (Figure 6 i l). As observed in the cytotoxicity assays, the 2 + 1 FabeFab variants showed increased activity compared to the 1 + 1 Fab-eFab. For the release of IL-2 and IFNγ, comparable cytokine levels at a concentration of 5 nM were measured (Figure 6 i, j). When comparing the EC₅₀ values between the Fab-eFab constructs, e11-0-1 showed a 3- and 7-fold reduced release of IL-2 and IFNγ compared to the 2 + 1 Fab-eFab variants, respectively (Table 6). All constructs effectively induced proliferation of both CD4⁺ and CD8⁺ T-cells (Figure 6 k, l). The 2 + 1 Fab-eFab variants showed a 6- to 14-fold increased proliferation of both CD4⁺ and CD8⁺ T-cells compared to e11-0-1. Additionally, a target-dependent induction of T-cell proliferation was observed for all Fab-eFab constructs (Figure S2 e, f).

Side-by-side comparison for the most promising formats

Based on the results from the three different panels of EGFRxCD3 bispecific antibodies, a selection of the most promising candidates was used for a side-by-side comparison. This selection included both the most potent formats that were identified, as well as formats that showed beneficial characteristics regarding EGFR level-dependent cytotoxicity and limited cytokine release.

From the 1 + 1 eIg and 2 + 1 Fab-eIg panel (Figures 1, 2), the most potent format 21-1-1 with favorable production and purification properties showing highly effective cytotoxicity on FaDu and LIM1215, but still inducing cytotoxicity on MCF-7 and SW620 with low EGFR expression, was chosen. Additionally, the 1 + 1 eIg e11-1-1 having favorable characteristics regarding productivity and purity but showing the strongest EGFR level-dependent cytotoxicity accompanied with reduced release of cytokines was selected. From the three 2 + 1 eIg-Fab variants (Figures 3, 4), e21-1-5 was selected, showing an EGFR level-dependent cytotoxicity as well as limited cytokine levels at a concentration of 0.5 nM, where it still effectively killed both FaDu and LIM1215 (>60% killing). Additionally, it was purified by a one-step protein A chromatography resulting in a purity of 98.5%. From the Fc-less Fab-eFab variants (Figures 5, 6) we selected e11-0-1 and the highly potent 2 + 1 Fab-eFab variant e21-0-1 being about one order of magnitude more active than e11-0-1. This was the only Fc-less molecule that showed an EGFR expression-dependent killing of tumor cells.

The five selected candidates, e11-1-1, e21-1-1, e21-1-5, e11-0-1, and e21-0-1, were compared side-by-side for thermal stability, cell binding, and cytotoxicity on two additional cell lines, the colorectal cancer cell lines DiFi and HCT-116 (Figure 7 c + d, f + g). While DiFi express very high levels of EGFR (955,000 ± 109,000 receptors/cell) about 5-fold higher than FaDu (188,000 ± 26,000 receptors/cell), HCT-116 express intermediate levels (17,300 ± 1,700) about 3-fold lower than LIM1215 (55,300 ± 4,700 receptors/cell) (Figure S1). Furthermore, the panel included KRAS-wild-type and KRAS-mutated (HCT-116; KRAS G13D;³⁰ cell lines to demonstrate that T cell-mediated killing is independent of EGFR signal inhibition. By further including SW620, the EGFR level-dependent cytotoxicity was investigated for a cell line with very low EGFR expression (Figure 7 e, h). The release of IL-2 was analyzed using PBMCs of three different donors (Figure 7 b). Finally, the pharmacokinetic properties of the bispecific antibodies e21-0-1 (Fc-less), e11-1-1, and e21-1-1 (both containing an Fc-part) in comparison to the monospecific EGFR-targeting antibody were analyzed in mice.

Figure 7. Cell binding to cancer cell lines, cytotoxicity and IL-2 release for selected set of formats based on the eIg technology. (a) Schematic illustration for selected formats including e11-1-1, e21–11, e21-1-5, e11-0-1 and e21-0-1. (b) Release of IL-2 after 24 h by PBMCs co-cultured with DiFi using an effector-to-target ratio of 5:1 analyzed by sandwich ELISA (Mean ± S.D.; n = 3 - three individual donors). (c – e) Binding to (c) DiFi, (d) HCT-116 and (e) SW620 analyzed by flow cytometry using a FITC-conjugated antibody specific for human Fab (Mean ± S.D; n = 3; MFI: median fluorescence intensity). (f – h) Cytotoxic potential of PBMCs co-cultured with (f) DiFi, (g) HCT-116 and (h) SW620 in an effector-to-target ratio of 5:1 (Mean ± S.D.; n = 3 - three individual donors).

Tumor cell binding and T-cell-mediated activity of a selected set of different variants based on the eIg technology.

Thermal stability of the five different molecules was determined by dynamic light scattering with aggregation points between 71°C and 72°C, which is slightly increased compared to cetuximab with an aggregation point of 65°C (Figure S3).

Cell binding to the cancer cell lines DiFi, HCT-116 and SW620 revealed an avidity effect for the tested formats. Thus, all three 2 + 1 formats showed similar cell binding and the two 1 + 1 showed similar but weaker cell binding as the 2 + 1 formats (Figure 7 c e, Table 7). This avidity effect increased with reduced EGFR levels of the tumor cell. The 2 + 1 formats bound 3- to 4-fold stronger on DiFi, while this difference increased to at least 14-fold for HCT-116 and SW620 (Table 7).

Table 7. Bioactivity of side-by-side comparison for selected formats (n.D. – not determinable).

		e11-1-1	e21-1-1	e21-1-5	e11-0-1	e21-0-1
EC50 cell binding [pM]	DiFi	9,232 ± 2,459	2,563 ± 821	2,651 ± 904	8,213 ± 2,482	2,101 ± 727
	HCT-116	1,882 ± 728	78 ± 15	83 ± 13	1,869 ± 444	38 ± 9
	SW620	602 ± 879	32 ± 41	42 ± 57	841 ± 1,259	31 ± 51
EC50 cytotoxicity [pM]	DiFi	0.88 ± 0.56	0.39 ± 0.30	8.18 ± 5.95	0.10 ± 0.13	0.30 ± 0.22
	HCT-116	431.6 ± 335.8	1.61 ± 0.31	23.81 ± 24.08	18.80 ± 4.34	0.66 ± 0.10
	SW620	n.d.	29.72 ± 26.19	n.d.	325.7 ± 171.2	5.27 ± 3.05
EC50 T-cell activation [pM]	IL-2	37.4 ± 81.8	18.2 ± 20.7	n.d.	2.7 ± 4.1	16.7 ± 16.8

Cytotoxicity studies on these cell lines showed that while most constructs performed equally well on DiFi with very high EGFR expression (EC₅₀ values<1 pM, except for e2115), a very broad range of activity was found on HCT-116 with intermediate EGFR expression and SW620 with low EGFR levels (Table 7). For HCT116, e21-0-1 showed the highest activity (EC₅₀ 0.7 ± 0.1 pM reaching >90% killing at 1 nM), while e11-1-1 showed the lowest activity (EC₅₀ 432 ± 336 pM, reaching ~ 50% killing at 10 nM) (Figure 7 g, Table 7). As observed before, e11-1-1 and e21-1-5 showed promising characteristics regarding an EGFR level-dependent cytotoxicity leading to >90% killing of the cells at 10 nM and 1 nM on DiFi, respectively, and to only marginal cytotoxicity on SW620 for these concentrations (Figure 7 f, h). Looking at a concentration of 10 pM, a similar trend was shown for the e11-0-1 format with strong killing on DiFi with an EC₅₀ value of 0.1 ± 0.1 pM and only minimal cytotoxicity on SW620 with an EC₅₀ value of 326 ± 171 pM (Figure 7 f, h). The two most active molecules remained e21-1-1 and the e21-0-1 with strong cytotoxicity on all cell lines. Of these two molecules, the Fc-less e21-0-1 showed the strongest activity in all cytotoxicity studies (Figure 7 f – h). This molecule also induced the strongest release of IL-2 compared to the other constructs at a concentration of 1 nM, where all formats showed comparable cytotoxicity on DiFi. The strongest release of IL-2 was observed for the three constructs e21-1-1, e11-0-1, and e21-0-1, while strongly reduced levels were measured for e11-1-1 and e21-1-5 (Figure 7 b). Taken together, the favorable characteristics of the two formats e11-1-1 and e21-1-5 observed in the initial characterizations are supported by the data from the side-by-side comparison. Here, these molecules were highly active on cancer cells expressing high levels of EGFR, while only minimal killing was observed for cancer cells with low EGFR expression level (<11,000 receptors/cell). At concentrations leading to highly potent target cell killing, reduced IL-2 release was detected.

The Fc-comprising molecules e11-1-1 and e21-1-1 exhibited an Ig-like pharmacokinetic profile with a terminal half-life of 110–122 hours, while the Fc-less e21-0-1 had a strongly reduced terminal half-life of 15.3 hours (Table S2; Figure S4).

Discussion

EGFR is a target over-expressed by various cancer cell types, including colorectal cancer, head, and neck squamous cell carcinoma, and non-small cell lung cancer,¹² but it also plays a role in early development and is expressed on normal tissues such as skin, liver, and gut.^31,32 The ability to discriminate cancer cells from healthy cells, i.e., a low or absent activity against cells expressing moderate to low levels of EGFR, is critical for efficient and safe therapy with TCEs. In this study, the eIg technology²⁶ was successfully applied to generate a panel of bispecific antibodies for T-cell retargeting to EGFR-expressing cells, varying in valency for EGFR, the position of the CD3 binding arm, and the presence or absence of an Fc region, with the aim to identify molecules with favorable activity profiles.

Our panel of molecules included two bivalent bispecific molecules (1 + 1 formats) either comprising an Fc region (e11-1-1) or lacking an Fc region (e11-0-1). All other molecules were trivalent bispecific, possessing two binding sites for EGFR and one for CD3 (2 + 1 formats) with different geometries and again lacking or possessing an Fc region. Binding studies with a panel of tumor cell lines expressing low to high EGFR levels revealed little dependency of cell-binding potency on EGFR levels for the 1 + 1 formats, in accordance with a monovalent binding mode, while the 2 + 1 formats showed, compared to cells expressing high levels of EGFR (>100.000 receptors/cell), strongly increased binding to cells expressing moderate to low EGFR levels, both regarding efficacy and potency. This finding supports an avidity-driven gain in cell binding, especially on cells with lower EGFR expression levels favoring a bivalent binding of the antibodies, in line with findings for other 1 + 1 and 2 + 1 TCEs, e.g., targeting HER2 or HER3, both receptors being members of the ErbB receptor family.^19,33

Interestingly, we observed that binding to CD3 on Jurkat cells was also affected by the format and position of the CD3 binding arm. Molecules with a freely accessible CD3 binding site, i.e., located at the N-terminal end of a polypeptide chain (e11-1-1, e21-1-1, e21-1-3, e21-1-4, e21-1-6, and e21-0-3) bound with EC₅₀ values in the low nanomolar range. In contrast, molecules where the CD3 binding arm was fused to the C-terminus of one (e21-1-1, e21-1-5, e11-0-1, and e21-0-2) or two other chains (e21-0-3) showed a reduced binding to Jurkat cells. This can be explained by a sterical interference of binding to CD3 by the chains fused to the N-terminus of the CD3 binding eFab.

All molecules showed a positive correlation between EGFR expression and target cell killing. Of note, such a correlation is not always observed for TCEs. Thus, a correlation was confirmed for targets such as HER2 and EphA2, while for targets such as CD20 and MCSP no correlation was found. This might be due to intracellular protection mechanisms or the fact that already minimal numbers of target antigens are sufficient to induce potent T cell-mediated target cell killing.³⁴

As EGFR is a TAA that is not solely expressed by cancer cells, it is crucial to define a minimal EGFR expression level necessary to effectively induce cytotoxicity in cancer cells, but low or absent in healthy tissues. Strong killing of LIM1215 with 55,300 EGFR per cell (64% killing at 1 nM), reduced killing of HCT116 with 17,300 EGFR per cell (29% killing at 1 nM), and the minimal killing of MCF-7 with 6,100 EGFR per cell (3% killing at 1 nM) was observed for e1111, indicating a threshold of ~10,000 EGFR per cell for this TCE. This is within the same range that has been identified for an Fc-comprising 2 + 1 TCE based on the Roche’s CrossMab technology directed against the TAA CEA expressed in low levels on the apical surface of glandular epithelia in the gastrointestinal tract. For this molecule (cibisatamab, RO6958688), a threshold of 10,000 receptors per cell was described for effective cytotoxicity based on in vitro results, and manageable toxic effects were observed in a first clinical study (NCT02324257).^25,35 However, our study also revealed that cytotoxicity is strongly influenced by format and valency and, thus, the threshold for target cell killing seems to be lower for other formats, such as some of the 2 + 1 formats.

The potency of target cell killing by our TCEs was strongly affected by the presence or absence of an Fc region. All Fc-less TCEs were very potent in killing target cells, especially the 2 + 1 formats (e21-0-1, e21-0-2, and e21-0-3), which killed EGFR^high target cells (DiFi, FaDu) with EC₅₀ values between 0.1 and 4 pM and were even highly potent against EGFR^low target cells (MCF-7, SW620). In contrast, the presence of an Fc region reduced killing activity. This can be explained by a sterical interference of the Fc region to form a tight immunological synapse and by differences in the distance between the EGFR and CD3 binding sites in these molecules.¹⁷ The latter might also explain differences seen for the e21-1-5 fusion protein, i.e., where the CD3 binding is located at the C-terminus of an anti-EGFR IgG. This format showed reduced target cell killing activity compared to other 2 + 1 formats where the CD3 arm is closer to the EGFR binding arms. This is in line with findings for molecules such as IgG-Fynomer fusion proteins, where the CD3-binding Fynomer was either fused to the N- or C-terminus of an anti-HER2 heavy chain. Here, the N-terminal fusion, mediating a closer contact between HER2 and CD3, was more active in target cell killing than the C-terminal fusion.³⁶ Similarly, 2 + 2 fusions of an anti-CD3 single-chain variable fragment (scFv) to the C-terminus of either the light or heavy chain of an anti-GD2 antibody demonstrated higher target cell killing activity for light-chain fusion in which the CD3 scFvs are closer to the GD2-binding Fab arms.³⁷ Furthermore, comparing 1 + 1 GD2×CD3 IgG-light chain fusions, the study revealed that molecules presenting the CD3 arm in cis, i.e., fused to the anti-GD2 light chain, were more potent in vivo than molecules with the CD3 arm present in trans (on the opposing dummy Fab arm) though in vitro cytotoxicity was similar, supporting the notion that interdomain spacing and spatial orientation affect potency of bispecific TCEs.

In our study, the Fc-less TCEs were the most potent molecules regarding cytotoxicity, but with the disadvantage of also killing cells with low EGFR expression and inducing the highest release of cytokines, which increases the risk of severe side effects by CRS. Multiple bispecific antibodies for T-cell retargeting have been designed in the past also using anti-EGFR binding domains derived from cetuximab. In one study, a 1 + 1 EGFRxCD3 tandem Fab, having a similar geometry as e11-0-1, was compared to a tandem scFv (BiTE). Although the tandem Fab showed about 10-fold reduced activity for killing of A431 with high EGFR expression compared to the BiTE molecule, the thermal stability of the tandem Fab molecule was increased.³⁸ Previously, a 2 + 1 EGFRxCD3 TriFab molecule, having a similar geometry as e21-0-1 Fab-eFab format, was described to be highly flexible so that binding to CD3 is possible even though two anti-EGFR Fab arms are fused to the N-termini of the anti-CD3 binding site. This molecule showed potent cell killing for A431 expressing high levels of EGFR and inducing robust cytokine release.²⁴

Release of proinflammatory cytokines from activated T-cells may pose limitations regarding safety profiles.¹¹ Thus, CRS was frequently observed in clinical trials of bispecific TCEs.^39,40 To avoid activation of accessory immune cells through FcγR interactions, silenced Fc regions are routinely used in immuno-modulatory therapeutic antibodies.⁴¹ In our case, we used the FcΔab mutations in our Fc-comprising TCEs, comprising residues in the lower hinge and C_H2 domain derived from human IgG2 and IgG4.²⁸ These mutations have been shown to strongly reduce binding to various Fcγ receptors and C1q, thus silencing Fc-mediated effector functions.⁴² However, target-cell-mediated binding of bispecific TCEs and engagement of T cells leads to a direct activation of the T cells and release of cytokines. For our TCEs, we observed a correlation between T cell-mediated target cell killing and activation of the T-cells as determined by cytokine release and T-cell proliferation, which essentially correlated with binding to CD3-expressing Jurkat cells. However, it is meanwhile established that TCE-induced cytokine release can be uncoupled from T cell-mediated cytotoxicity and that this requires different activation thresholds for the formation of lytic and stimulatory synapses.^43–45 Lowering the affinity for CD3 might be an option to reduce cytokine release while maintaining potent target cell killing. This was, for example, shown for a bispecific TCE combining an anti-PSMA antibody with a unique, low-affinity anti-CD3 arm, which mediated potent killing of PSMA+ tumor cells with minimal cytokine release and reduced activation of Tregs.⁴⁶ Of note, reducing the affinity for CD3 also affects tissue distribution. Thus, a strong correlation between CD3 affinity and accumulation in T cell-rich tissues was described for an HER2×CD3 TCE.⁴⁷ Here, a higher affinity for CD3 reduced system exposure and shifted the distribution of the TCE away from the tumor to T cell-containing tissues. Furthermore, in an Fc-comprising 1 + 1 TCE format targeting HER2, favorable characteristics were shown for reduced affinities for both the TAA and CD3.⁴⁸ This indicates that efficient target cell killing can be achieved even for molecules with reduced binding for both antigens. This is in line with findings that a mature immunological synapse is formed by about 10 complexes between a peptide loaded MHCI and the TCR, i.e., by only a few contact points.⁴⁹

Our comparative data revealed for some of the formats a strong target cell killing activity over a large EGFR expression range, including killing of cells with rather low EGFR levels, which correlated with rather strong T-cell activation and cytokine release. On the other hand, some of the formats, such as e11-1-1 and e21-1-5, exhibited a more differential killing profile. Thus, these two formats showed strong and almost complete killing of target cells expressing high EGFR levels (e.g., DiFi and FaDu), while killing of cells with moderate EGFR levels was strongly reduced (e.g., HCT-116) and no or only marginal killing was observed for low EGFR-expressing cells, such as SW620. Hence, these formats might exhibit a favorable “on-target/off-tumor” profile in vivo. Modification of the EGFR binding affinity to achieve an avidity-based specificity gain for the trivalent bispecific molecules might be a further option to increase the therapeutic window. For example, the influence of affinity for the TAA on the potency of 2 + 1 bispecific TCEs has been shown for P-cadherin and HER2.^33,50

Of note, two formats, e11-1-1 and e21-1-5, also showed rather low levels of cytokine release over the entire concentration range tested (up to 10 nM). These two formats might therefore represent suitable candidates for further development of EGFR-targeting TCEs, capable of killing tumor cells with high and moderate EGFR levels, while sparing low expressing cells and causing only minor cytokine release. Of course, these findings have to be evaluated and confirmed in further in vivo studies investigating efficacy and safety.

Our study showed that Fc-comprising formats exhibit an Ig-like pharmacokinetic profile, while an Fc-less molecule was rapidly cleared from circulation. This is presumably not due to different thermal stability of the different molecules, as similar aggregation points were determined for Fc-comprising and Fc-less formats. The presence or absence of an Fc region seems to be less critical for bioactivity since formats such as e21-1-1 and e21-0-1 exhibited similar potency in mediating target cell killing, especially on cells with high and medium EGFR levels. The most potent molecules comprised two binding sites for EGFR, with the CD3-binding site close to the EGFR binding sites. This was, for example, prominent for e21-1-5, with the CD3-binding arm at the C-terminus of one heavy chain, which exhibited a strongly reduced killing activity compared to e21-1-1, with the CD3-binding site as part of a Fab-eFab arm. This supports previous findings that a shorter distance between target antigen and CD3 mediated by the bispecific antibody results in increased target cell killing due to the ability to form tight immunological synapses, especially on cells with lower copy numbers of target antigens.¹⁸ It should be emphasized, however, that other parameters, such as affinities for target antigen and CD3, antigen size, epitope location on the antigen and distance from the cell membrane can all have a profound effect on target cell killing.^{17,18,51–53}

In summary, we established the eIg technology as a versatile platform to generate bispecific TCEs with different molecular architectures to select those which show an efficient target-cell-dependent T-cell activation and T cell-mediated target cell killing while keeping cytokine release at a minimum. Our study demonstrates and further supports the finding that valency, size, and geometry strongly influence the functional properties of bispecific TCEs, underlining that format matters.

Materials and methods

Antibody production and purification

Heavy and light chains of the bispecific antibodies were cloned into variants of pSecTagA vectors and transiently produced in HEK293-6E suspension cells as described previously.²⁰ Bispecific antibodies containing an Fc part were purified by protein A affinity chromatography (Cytiva), eluted with 100 mM glycine pH 3.5, and dialyzed against phosphate-buffered saline (PBS) at 4°C. Bispecific antibodies without an Fc part were purified by CaptureSelect C_H1-XL affinity chromatography (ThermoFisher), eluted with 100 mM glycine pH 4.0, and dialyzed against PBS at 4°C. For e11-1-1, e21-1-1, e21-1-2, e21-1-3, e11-0-1, e2101, e21-0-2, and e21-0-3, proteins were further purified with a preparative SEC by FPLC.

Antibody characterization

Bispecific antibodies were analyzed by SDS-PAGE (3 µg for non-reducing/6 µg for reducing conditions) using 12% polyacrylamide gels and by SEC using a Waters 2695 HPLC in combination with a TSKgel SuperSW mAb HR column (Tosoh Bioscience) at a flow rate of 0.5 (Figure 1) or 0.4 mL/min (Figures 3 and 5) as described previously.²⁰ The determination of aggregation points of the antibodies was performed using dynamic light scattering (ZetaSizer Nano ZS, Malvern). Approximately 100 µg of purified protein was diluted to a total volume of 1 ml and analyzed. The aggregation point was defined as the temperature where the light scattering increased.

Enzyme-linked immunosorbent assay

Binding of bispecific antibodies to EGFR-ECD-moFc (extracellular domain of EGFR: aa 25–645 with C-terminal fusion to a murine IgG2a-Fc part) was analyzed by ELISA as described previously.²⁶

Flow cytometry analysis

Binding of bispecific antibodies to CD3-expressing Jurkat cells or tumor cells (DiFi, FaDu, HCT-116, LIM1215, MCF-7, SW620) was analyzed by flow cytometry as described previously.²⁰ Bound antibodies were detected with a R-PE-labeled anti-human Fc antibody (for antibodies containing an Fc-part (109-115-098, Jackson ImmunoResearch) or an FITC-labeled anti-human Fab antibody (F5512, Sigma Aldrich)) using a MACSQuant VYB (Miltenyi Biotec). Data were analyzed using FlowJo (Tree Star), and relative MFI was calculated as follows: relative MFI = ((MFI_sample - (MFI_detection-MFI_cells))/MFI_cells).

Cytotoxicity

Cytotoxicity assays were performed as described previously²⁰ using an effector-to-target ratio of 5:1 (20,000 tumor cells (DiFi, FaDu, HCT-116, LIM1215, MCF-7, and SW620) and 100,000 PBMCs per well in a 96-well plate).

IL-2 and IFNγ release

IL-2 and IFNγ release were analyzed 24 h and 48 h, respectively, after treatment with bispecific antibodies as described previously²⁰ using an effector-to-target ratio of 5:1 (20,000 tumor cells (FaDu, DiFi) and 100,000 PBMCs per well in a 96-well plate). Cytokine levels in the supernatant were quantified by sandwich ELISA according to instructions of manufacturer (IL-2/IFNγ Duo Set ELISA; R&D Systems).

T-cell proliferation

T-cell proliferation was analyzed 6 d after treatment with bispecific antibodies as described previously²⁰ using an effector-to-target ratio of 5:1 (20,000 FaDu cells and 100,000 carboxyfluorescein succinimidyl ester (CFSE)-labeled PBMCs per well in a 96-well plate). Effector cells were stained for CD3, CD4, and CD8 (anti-CD3-PerCP-Vio700 (REA613), anti-CD4-VioBlue (REA623), anti-CD8-PE-Vio770 (REA734); Miltenyi Biotec) and analyzed by flow cytometry using a MACSQuant Analyzer 10 (Miltenyi Biotec). Data were analyzed using FlowJo (Tree Star). As the antibody used for the detection of CD3 (REA613) competes with UCHT1 (parental antibody of huU3 applied in the bispecific antibodies), no gating for CD3+ T-cells was performed, and classification of CD4⁺ or CD8⁺ T-cells was solely based on the staining with anti-CD4-VioBlue and anti-CD8-PE-Vio770 antibodies.

Animal experiments

All animal studies were approved by the University of Stuttgart and governmental authorities. For the determination of the pharmacokinetic profile of the different antibodies, the proteins were administered by intravenous infusion, and blood samples were taken after 3 minutes, 1 hour, 6 hours, 24 hours, 72 hours, and 168 hours. After incubation on ice for 20 minutes, samples were centrifuged (16,000 × g, 4°C, 20 min) and stored at −20°C until analysis. Serum concentrations of antibodies were determined by ELISA using EGFR-ECD-moFc as immobilized antigen. Bound antibodies were detected with horseradish peroxidase-labeled anti-human Fc (Sigma Aldrich, A0170) or anti-human Fab (Sigma Aldrich, A2093) secondary antibody. The pharmacokinetics data were calculated with PKSolver.

Statistics

All data are represented as mean ± SD for n = 3, analyzed by One-way ANOVA with Tukey posttest (GraphPad Prism 7). For co-culture experiments with PBMCs from healthy donors (cytotoxicity, T-cell activation), three different donors were analyzed.

Abbrrevations

Table

ALL	acute lymphoblastic leukemia
BiTE	bispecific T-cell engager
CD	cluster of differentiation
CRS	cytokine release syndrome
EGFR	epidermal growth factor receptor
eIg/eFab	bispecific Ig/Fab domain containing hetEHD2
ErbB	erythroblastic leukemia viral oncogene homolog
Fab	fragment antigen binding
Fc	fragment crystallizable
hetEHD2	EHD2 domain with C102S mutation (EHD2–1) or C14S/N39Q mutations (EHD2–2) of IgE
HMW	higher molecular weight
Ig	immunoglobulin
KRAS	Kirsten rat sarcoma virus
LMW	lower molecular weight
PBMC	peripheral blood mononuclear cell
scFv	single-chain fragment variable
SEC	size-exclusion chromatography
TAA	tumor-associated antigen
TCE	T-cell engager
TKI	tyrosine kinase inhibitor

Author contributions

L.K., A.K.S., S.K., N.A., and O.S. performed cloning, protein expression and purification, biochemical analysis, binding studies, and bioactivity assays. L.K., A.K.S., S.K., R.E.K., and O.S. analyzed and interpreted the data. L.K., N.A., R.E.K. and O.S. were responsible for experimental design and supervised the work. L.K., R.E.K. and O.S. wrote the manuscript. All authors read and approved the final manuscript.

Additional information

Competing interests: R.E.K. and O.S. are named inventors on a patent application covering the eIg technology. The data are available and can be sent upon request.

Acknowledgments

We would like to thank Nadine Heidel, Sabine Münkel, Beatrice Reiser, Vanessa Schmitt, Laura Walisch, and Alexandra Kraske for their excellent technical assistance.

Disclosure statement

R.E.K. and O.S. are named inventors on a patent application covering the eIg technology. No potential conflict of interest was reported by the other authors.

Supplementary data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2023.2183540

Additional information

Funding

The author(s) reported that there is no funding associated with the work featured in this article.