Heterodimeric antibody that binds to CD3 and GPC3

Abstract

Provided herein are novel GPC3 binding domains and antibodies (e.g., anti-GPC3 x anti-CD3) that contain such GPC3 binding domains. Also provided herein are methods of using such antibodies for the treatment of GPC3-associated cancers.

Description

[Technical field]

[0001] Priority claim This application claims the benefit of U.S. Provisional Patent Application No. 63 / 159,058, filed March 10, 2021, and U.S. Provisional Patent Application No. 63 / 173,127, filed April 9, 2021, the entireties of which are incorporated by reference herein.

[0002] Sequence Listing This application contains a Sequence Listing that has been submitted electronically in ASCII format and is incorporated herein by reference in its entirety. This ASCII copy, created on February 28, 2022, is named 067461-5282-WO_SL.txt and is 853,417 bytes in size. Summary of the Invention

[0003] Provided herein are novel GPC3 binding domains and antibodies (e.g., anti-GPC3 x anti-CD3) that contain such GPC3 binding domains. Also provided herein are methods of using such antibodies for the treatment of GPC3-associated cancers.

[0004] In a first aspect, a composition comprising a GPC3 antigen binding domain (ABD) is provided herein. The GPC3 binding domain is [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3]H1_L1.6 5, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC3]H1.1_L 1, [GPC3]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC3]H1.1_L1.4, [GPC3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC3]H1.1_L1.7, [GPC3 ]H1.1_L1.8, [GPC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [GPC3]H1.1_L1.31, [GPC3]H 1.1_L1.65, [GPC3]H1.1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1.70, [GPC3]H1.1_L1.71, [GPC3]H1.1_L1.72, [GPC3]H1.1_L1.73, [ GPC3]H1.2_L1, [GPC3]H1.2_L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1.2_L1.4, [GPC3]H1.2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2 _L1.7, [GPC3]H1.2_L1.8, [GPC3]H1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [GPC3]H1.2_L1.23, [GPC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1 .2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [G PC3]H1.2_L1.73, [GPC3]H1.3_L1, [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [ GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3]H1.3_L1.7, [GPC3]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1. 10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [GPC3]H1. 3_L1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GP C3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3_L1.71, [GPC3]H1.3_L1.7 2, [GPC3]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1 .2, [GPC3]H1.4_L1.3, [GPC3]H1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_ L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1. 4_L1.10, [GPC3]H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.29, [GPC [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73.

[0005] In some embodiments, the ABD is selected from the group consisting of [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3]H1_L1.65, [GPC 3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3] H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC3]H1.1_L1, [GPC3]H1. 1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC3]H1.1_L1.4, [GPC3]H 1.1_L1.5, [GPC3]H1.1_L1.6, [GPC3]H1.1_L1.7, [GPC3]H1.1_L1.8, [GPC3 ]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [GPC3]H1.1_L1.31, [GPC3]H1.1_L1.65, [GPC3]H1.1 _L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1.70, [GPC3]H1.1_L1.7 1, [GPC3]H1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1.2_L1, [GPC3]H1.2_L 1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1.2_L1.4, [GPC3]H1.2 _L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3]H1.2_L1.8, [GPC3]H1 .2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [GPC3]H1.2_L1.23, [GP C3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.73, [GPC3 ]H1.3_L1, [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3] H1.3_L1.7, [GPC3]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [ GPC3]H1.3_L1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3 _L1.71, [GPC3]H1.3_L1.72, [GPC3]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [GPC3]H1 .4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3] The VH / VL pair is selected from the group consisting of H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.29, [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73.

[0006] In some embodiments, the VH / VL pair is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69. In an exemplary embodiment, the composition is a monoclonal antibody.

[0007] In another aspect, a heterodimeric antibody is provided herein, comprising a) a first monomer, b) a second monomer, and c) a light chain. The first monomer comprises i) an anti-CD3 scFv comprising a first variable light domain, an scFv linker, and a first variable heavy domain, and ii) a first Fc domain, where the scFv is covalently linked to the N-terminus of the first Fc domain using a domain linker. The second monomer comprises a VH2-CH1-hinge-CH2-CH3 monomer, where VH is the second variable heavy domain, and CH2-CH3 is the second Fc domain. In this embodiment, the second variable heavy domain and the second variable light domain form a GPC3 antigen binding domain (ABD).

[0008] In some embodiments, the GPC3 binding domain is, [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3]H1_L1. 65, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70 , [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC3]H1.1_L1, [GP C3]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC3]H1.1_L1.4, [ GPC3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC3]H1.1_L1.7, [GPC3]H1.1_L1.8, [GPC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L 1.23, [GPC3]H1.1_L1.29, [GPC3]H1.1_L1.31, [GPC3]H1.1_L1.65, [GPC3]H 1.1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1.70, [GPC3]H1.1_L 1.71, [GPC3]H1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1.2_L1, [GPC3]H1.2 _L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1.2_L1.4, [GPC3]H1 .2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3]H1.2_L1.8, [GPC3]H 1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [GPC3]H1.2_L1.23, [G PC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.73, [GPC3]H1.3_L1 , [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3]H1.3_L1.7, [GPC3 ]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [GPC3]H1.3_L1.31, [GPC3] H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3_L1.71, [GPC3]H1.3_L1.72, [GPC3 ]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [GPC3]H1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_L 1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3]H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.2 9, [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73.

[0009] In an exemplary embodiment, the GPC3 binding domain is [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3]H1_L1. 5, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC3]H1.1_L1, [GPC 3]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC3]H1.1_L1.4, [G PC3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC3]H1.1_L1.7, [GPC3]H1.1_L1.8, [ GPC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [GPC3]H1.1_L1.31, [GPC3]H1.1_L1.65, [GPC3]H 1.1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1.70, [GPC3]H1.1_L 1.71, [GPC3]H1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1.2_L1, [GPC3]H1.2 _L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1.2_L1.4, [GPC3]H1 .2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3]H1.2_L1.8, [GPC3]H 1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [GPC3]H1.2_L1.23, [G PC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.73, [GPC3 ]H1.3_L1, [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3] H1.3_L1.7, [GPC3]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [ GPC3]H1.3_L1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3 _L1.71, [GPC3]H1.3_L1.72, [GPC3]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [GPC3]H 1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3 ]H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.29, [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73.

[0010] In an exemplary embodiment, the GPC3 binding domain comprises a VH / VL pair selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69.

[0011] In some embodiments, the anti-CD3 scFv comprises a VH and VL pair selected from the group consisting of H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31.

[0012] In some embodiments of the heterodimeric antibody, the scFv linker is a charged scFv linker.

[0013] In some embodiments, the first and second Fc domains are variant Fc domains. In some embodiments, the first and second Fc domains comprise a set of heterodimerization variants selected from the group consisting of those depicted in Figures 1A-1E. In some embodiments, the set of heterodimerization variants selected is selected from the group consisting of S364K / E357Q:L368D / K370S, S364K:L368D / K370S, S364K:L368E / K370S, D401K:T411E / K360E / Q362E, and T366W:T366S / L368A / Y407V, where numbering is according to EU numbering.

[0014] In some embodiments, the first and second monomers further comprise one or more deletion variants, in some embodiments, the one or more deletion variants are E233P / L234V / L235A / G236del / S267K, where the numbering is according to EU numbering.

[0015] In exemplary embodiments, one of the first or second monomers comprises one or more pI variants, hi some embodiments, the one or more pI variants are N208D / Q295E / N384D / Q418E / N421D, where the numbering is according to EU numbering.

[0016] In an exemplary embodiment of a heterodimeric antibody, a first monomer comprises the amino acid variants S364K / E357Q / E233P / L234V / L235A / G236del / S267K and a second monomer comprises the amino acid variants L368D / K370S / N208D / Q295E / N384D / Q418E / N421D / E233P / L234V / L235A / G236del / S267K, where numbering is according to EU numbering.

[0017] In some embodiments, the first and second monomers each further comprise the amino acid variant 428 / 434S, where the numbering is according to EU numbering.

[0018] In another aspect, provided herein is a heterodimeric antibody comprising: a) a first monomer comprising, from N-terminus to C-terminus, a scFv-linker-CH2-CH3, where the scFv is an anti-CD3 scFv and CH2-CH3 is a first Fc domain; b) a second monomer comprising, from N-terminus to C-terminus, a VH-CH1-hinge-CH2-CH3, where CH2-CH3 is a second Fc domain; and c) a light chain comprising VL-CL. The first variant Fc domain comprises the amino acid variant S364K / E357Q, the second variant Fc domain comprises the amino acid variant L368D / K370S, the first and second variant Fc domains each comprise the amino acid variants E233P / L234V / L235A / G236del / S267K, and the CH1-hinge-CH2-CH3 of the second monomer comprises the amino acid variants N208D / Q295E / N384D / Q418E / N421D (EU numbering). Further, the VH and VL form a GPC3 binding domain comprising the variable heavy and variable light domains, respectively, of a GPC3 binding domain selected from [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69, The scFv comprises a variable heavy chain domain and a variable light chain domain of a CD3 binding domain selected from H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31.

[0019] In some embodiments, the first and second variant Fc domains each further comprise the amino acid variant 428 / 434S, where numbering is according to EU numbering.

[0020] In another aspect, provided herein is a heterodimeric antibody comprising a) a first monomer, b) a second monomer, and c) a common light chain. The first monomer comprises, from N-terminus to C-terminus, VH1-CH1-linker1-scFv-linker2-CH2-CH3, where VH1 is a first variable heavy chain domain, scFv is an anti-CD3 scFV, linker1 and linker2 are a first domain linker and a second domain linker, respectively, and CH2-CH3 is a first Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH2-CH1-hinge-CH2-CH3, where VH2 is a second variable heavy chain domain, CH2-CH3 is a second Fc domain, and c) the common light chain comprises a variable light chain domain. The first variable heavy and light domains form a first CLDN6 ABD, and the second variable heavy and light domains form a second CLDN6 ABD.

[0021] In another aspect, provided herein is a heterodimeric antibody comprising a) a first monomer, b) a second monomer, and c) a common light chain. The first monomer comprises, from N-terminus to C-terminus, VH1-CH1-linker1-scFv-linker2-CH2-CH3, where VH1 is a first variable heavy chain domain, scFv is an anti-CD3 scFV, linker1 and linker2 are a first domain linker and a second domain linker, respectively, and CH2-CH3 is a first Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH2-CH1-hinge-CH2-CH3, where VH2 is a second variable heavy chain domain, CH2-CH3 is a second Fc domain, and c) the common light chain comprises a variable light chain domain. The first variable heavy and light domains form a first GPC3 ABD, and the second variable heavy and light domains form a second GPC3 ABD.

[0022] In some embodiments, the first and second GPC3 binding domains are each selected from the group consisting of [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3]H1_L1.65, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC 3]H1.1_L1, [GPC3]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC 3]H1.1_L1.4, [GPC3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC3]H1.1_L1.7, [G PC3]H1.1_L1.8, [GPC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [GPC3]H1.1_L1.31, [GPC3]H1. 1_L1.65, [GPC3]H1.1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1. 70, [GPC3]H1.1_L1.71, [GPC3]H1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1 .2_L1, [GPC3]H1.2_L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1 .2_L1.4, [GPC3]H1.2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3] H1.2_L1.8, [GPC3]H1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [G PC3]H1.2_L1.23, [GPC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.7 3, [GPC3]H1.3_L1, [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3]H1 .3_L1.7, [GPC3]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [GPC3]H1.3_L 1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3_L1.71, [GPC3]H1.3_L1 .72, [GPC3]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [GPC3]H1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3 ]H1.4_L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3]H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4 The set of six CDRs (vhCDR1, vhCDR2, vhCDR3, vlCDR1, vlCDR2, and vlCDR3) from a VH / VL pair selected from the group consisting of [GPC3]H1.4_L1.29, [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73.

[0023] In some embodiments, each of the first and second GPC3 binding domains is selected from the group consisting of [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31 , [GPC3]H1_L1.65, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GP C3]H1.1_L1, [GPC3]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GP C3]H1.1_L1.4, [GPC3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC3]H1.1_L1.7, [G PC3]H1.1_L1.8, [GPC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [GPC3]H1.1_L1.31, [GPC3]H1. 1_L1.65, [GPC3]H1.1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1. 70, [GPC3]H1.1_L1.71, [GPC3]H1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1 .2_L1, [GPC3]H1.2_L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1 .2_L1.4, [GPC3]H1.2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3] H1.2_L1.8, [GPC3]H1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [G PC3]H1.2_L1.23, [GPC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1 .2_L1.73, [GPC3]H1.3_L1, [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3 _L1.6, [GPC3]H1.3_L1.7, [GPC3]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1 .3_L1.29, [GPC3]H1.3_L1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [G PC3]H1.3_L1.71, [GPC3]H1.3_L1.72, [GPC3]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [G PC3]H1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [G In some embodiments, the VH / VL pair is selected from the group consisting of [GPC3]H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.29, [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73. In some embodiments, the VH / VL pair is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69.

[0024] In some embodiments, the scFv has a set of six CDRs (vhCDR1, vhCDR2, vhCDR3, vlCDR1, vlCDR2, and vlCDR3) from a VH / VL pair selected from the group consisting of H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31. In some embodiments, the scFv comprises the variable heavy chain domain and variable light chain domain of any of the following CD3 binding domains: H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31.

[0025] In some embodiments, the heterodimeric antibody is a charged scFv linker. In some embodiments, the scFv linker is a charged scFv linker having the amino acid sequence (GKPGS)4 (SEQ ID NO:1).

[0026] In some embodiments, the first and second Fc domains are variant Fc domains. In some embodiments, the first and second Fc domains comprise a set of heterodimerization variants selected from the group consisting of those depicted in Figures 1A-1E. In some embodiments, the set of heterodimerization variants selected is selected from the group consisting of S364K / E357Q:L368D / K370S, S364K:L368D / K370S, S364K:L368E / K370S, D401K:T411E / K360E / Q362E, and T366W:T366S / L368A / Y407V, where numbering is according to EU numbering.

[0027] In some embodiments, the first and second monomers further comprise one or more deletion variants, in some embodiments, the one or more deletion variants are E233P / L234V / L235A / G236del / S267K, where the numbering is according to EU numbering.

[0028] In exemplary embodiments, one of the first or second monomers comprises one or more pI variants, hi some embodiments, the one or more pI variants are N208D / Q295E / N384D / Q418E / N421D, where the numbering is according to EU numbering.

[0029] In an exemplary embodiment of a heterodimeric antibody, a first monomer comprises the amino acid variants S364K / E357Q / E233P / L234V / L235A / G236del / S267K and a second monomer comprises the amino acid variants L368D / K370S / N208D / Q295E / N384D / Q418E / N421D / E233P / L234V / L235A / G236del / S267K, where numbering is according to EU numbering.

[0030] In some embodiments, the first and second monomers each further comprise the amino acid variant 428 / 434S, where the numbering is according to EU numbering.

[0031] 45. The heterodimeric antibody according to 44, wherein the first and second variant Fc domains each further comprise the amino acid variant 428 / 434S, wherein the numbering is according to EU numbering.

[0032] In another aspect, provided herein is a heterodimeric antibody comprising a) a first monomer, b) a second monomer, and c) a common light chain. The first monomer comprises, from N-terminus to C-terminus, VH1-CH1-linker1-scFv-linker2-CH2-CH3, where scFv is an anti-CD3 scFv and CH2-CH3 is a first Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH1-CH1-hinge-CH2-CH3, where CH2-CH3 is a second Fc domain. The common light chain comprises VL-CL. The first variant Fc domain comprises the amino acid variant S364K / E357Q, the second variant Fc domain comprises the amino acid variant L368D / K370S, the first and second variant Fc domains each comprise the amino acid variants E233P / L234V / L235A / G236del / S267K, and the CH1-hinge-CH2-CH3 of the second monomer comprises the amino acid variants N208D / Q295E / N384D / Q418E / N421D (EU numbering). The VH and VL comprise variable heavy and variable light domains of GPC3 ABD selected from H1.9_L1.187, H1.24_L1.187, H2.91_L1.187, and H1.9_L1.187, and are anti-CD3 The scFv comprises a variable heavy chain domain and a variable light chain domain of a CD3 binding domain selected from H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31. In some embodiments, the first and second variant Fc domains each further comprise the amino acid variant 428 / 434S.

[0033] Also provided herein are nucleic acid compositions comprising nucleic acids encoding the antibodies described herein, expression vector compositions comprising such nucleic acids, host cells for producing the antibodies comprising the expression vector compositions, and methods of producing the antibodies. [Background technology]

[0034] Antibody-based therapeutics have been successfully used to treat a variety of diseases, including cancer. An increasingly popular avenue being explored is the engineering of a single immunoglobulin molecule that simultaneously links two different antigens. Such alternative antibody formats that link two different antigens are often referred to as bispecific antibodies. Because the considerable diversity of antibody variable regions (Fvs) allows the generation of Fvs that recognize virtually any molecule, a typical approach to bispecific antibody generation is to introduce a new variable region into an antibody.

[0035] A particularly useful approach for bispecific antibodies is to engineer a first binding domain that links to CD3 and a second binding domain that links to an antigen that is associated with or upregulated on cancer cells, such that the bispecific antibody redirects CD3+ T cells to destroy cancer cells. Glypican 3 (GPC3), a member of the heparan sulfate proteoglycan family, has previously been reported to be highly expressed in hepatocellular carcinoma (especially the liver) and minimally expressed in healthy tissues. In light of this, it is believed that anti-GPC3 antibodies are useful, for example, for localizing antitumor therapeutics (e.g., chemotherapeutic agents and T cells) to such GPC3-expressing tumors. Previous attempts to target GPC3, such as CD3 bispecific antibodies and CAR-T therapy, have shown some success but have demonstrated adverse events, such as cytokine release syndrome in preclinical or clinical settings. Provided herein are novel bispecific antibodies to CD3 and GPC3 that can localize CD3+ effector T cells to GPC3-expressing tumors and improve therapeutic profiles. [Brief description of the drawings]

[0036] [Figure 1A]Shown are useful pairs of sets of Fc heterodimerization variants (including scubariants and PI variants) that result in Fc heterodimerization. There are variants for which there is no corresponding "monomer 2" variant; these are pi variants that can be used alone for either monomer. [Figure 1B] Shown are useful pairs of sets of Fc heterodimerization variants (including scubariants and PI variants) that result in Fc heterodimerization. There are variants for which there is no corresponding "monomer 2" variant; these are pi variants that can be used alone for either monomer. [Figure 1C] Shown are useful pairs of sets of Fc heterodimerization variants (including scubariants and PI variants) that result in Fc heterodimerization. There are variants for which there is no corresponding "monomer 2" variant; these are pi variants that can be used alone for either monomer. [Figure 1D] Shown are useful pairs of sets of Fc heterodimerization variants (including scubariants and PI variants) that result in Fc heterodimerization. There are variants for which there is no corresponding "monomer 2" variant; these are pi variants that can be used alone for either monomer. [Figure 1E] Shown are useful pairs of sets of Fc heterodimerization variants (including scubariants and PI variants) that result in Fc heterodimerization. There are variants for which there is no corresponding "monomer 2" variant; these are pi variants that can be used alone for either monomer. [Diagram 2] A list of constant regions of isosteric variant antibodies and their respective substitutions is provided. pI_(-) indicates low pI variants and pI_(+) indicates high pI variants. These can be optionally and independently combined with other heterodimerization variants of the invention (and other variant types outlined herein). [Diagram 3]Useful deletion variants that eliminate FcγR binding are shown (sometimes referred to as "knockout" or "KO" variants). Generally, deletion variants are found in both monomers, but in some cases, may be in only one monomer. [Figure 4] Particularly useful embodiments of "non-Fv" components of the invention are provided. [Diagram 5] As described herein, several charged scFv linkers are shown that find use in increasing or decreasing the pI of the subject heterodimeric bsAbs that use one or more scFvs as building blocks. (+H) positive linkers are of particular use herein, particularly with the anti-CD3 VL and VH sequences shown herein. A single prior art scFv linker with a single charge is referred to as "Whitlow" from Whitlow et al., Protein Engineering 6(8):989-995 (1993). It should be noted that this linker was used to reduce aggregation and increase proteolytic stability in scFvs. Such charged scFv linkers can be used in any of the subject antibody formats disclosed herein that include scFvs (e.g., 1+1 Fab-scFv-Fc and 2+1 Fab2-scFv-Fc formats). [Figure 6] Some exemplary domain linkers are shown. In some embodiments, these linkers find use in linking a single chain Fv to an Fc chain. In some embodiments, these linkers can be combined. For example, the GGGGS linker (SEQ ID NO:2) can be combined with a "half hinge" linker. [Figure 7A]The following shows the sequences of some useful 1+1 Fab-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, without the Fv sequences (e.g., scFv and Fab side VH): Scaffold 1 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, C220S on the chain with the S364K / E357Q scaffold variant, N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S sqvariant, C220S on the chain with the S364K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S sqvariant, C220S on the chain with the S364K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368E / K370S sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 4 is based on human IgG1 (356E / 358M allotype) and contains the D401K:K360E / Q362E / T411E sqvariant, C220S on the chain with the D401K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the K360E / Q362E / T411E sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains.Scaffold 5 is based on human IgG1 (356D / 358L allotype) and contains the S364K / E357Q:L368D / K370S sq variant, C220S on the chain with the S364K / E357Q sq variant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 6 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S sqvariant, C220S on the chain with the S364K / E357Q sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sqvariant, E233P / L234V / L235A / G236del / S267K deletion variants on both chains as well as the N297A variant on both chains. Scaffold 7 is identical to 6 except the mutation is N297S. Scaffold 8 is based on human IgG4 and contains an S364K / E357Q:L368D / K370S scubariant, an N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scubariant, as well as an S228P (EU numbering, which is S241P in Kabat) variant on both chains that eliminates Fab arm exchange, as known in the art. Scaffold 9 is based on human IgG2 and contains an S364K / E357Q:L368D / K370S scubariant, an N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scubariant. Scaffold 10 is based on human IgG2 and contains the S364K / E357Q:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant as well as the S267K variant on both chains. Scaffold 11 is identical to scaffold 1 except that it contains the M428L / N434S Xtend mutations.Scaffold 12 is based on human IgG1 (356E / 358M allotype) and includes the P217R / P229R / N276K pI variants on one chain with the S364K / E357Q:L368D / K370S scaffold, the C220S, and S364K / E357Q scaffolds, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Included in each of these scaffolds are sequences that are 90, 95, 98, and 99% identical (as defined herein) to the listed sequences and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 additional amino acid substitutions (compared to the "parent" in the figure, which already contains some amino acid modifications compared to the parent human IgG1 (or IgG2 or IgG4 depending on the scaffold), as will be appreciated by those skilled in the art). That is, the recited scaffolds may contain additional amino acid modifications (generally amino acid substitutions) in addition to the sqvariants, pI variants, and deletion variants contained within the scaffolds of this figure. [Figure 7B]The following shows the sequences of some useful 1+1 Fab-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, without the Fv sequences (e.g., scFv and Fab side VH): Scaffold 1 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, C220S on the chain with the S364K / E357Q scaffold variant, N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S sqvariant, C220S on the chain with the S364K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S sqvariant, C220S on the chain with the S364K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368E / K370S sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 4 is based on human IgG1 (356E / 358M allotype) and contains the D401K:K360E / Q362E / T411E sqvariant, C220S on the chain with the D401K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the K360E / Q362E / T411E sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains.Scaffold 5 is based on human IgG1 (356D / 358L allotype) and contains the S364K / E357Q:L368D / K370S sq variant, C220S on the chain with the S364K / E357Q sq variant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 6 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S sqvariant, C220S on the chain with the S364K / E357Q sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sqvariant, E233P / L234V / L235A / G236del / S267K deletion variants on both chains as well as the N297A variant on both chains. Scaffold 7 is identical to 6 except the mutation is N297S. Scaffold 8 is based on human IgG4 and contains an S364K / E357Q:L368D / K370S scubariant, an N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scubariant, as well as an S228P (EU numbering, which is S241P in Kabat) variant on both chains that eliminates Fab arm exchange, as known in the art. Scaffold 9 is based on human IgG2 and contains an S364K / E357Q:L368D / K370S scubariant, an N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scubariant. Scaffold 10 is based on human IgG2 and contains the S364K / E357Q:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant as well as the S267K variant on both chains. Scaffold 11 is identical to scaffold 1 except that it contains the M428L / N434S Xtend mutations.Scaffold 12 is based on human IgG1 (356E / 358M allotype) and includes the P217R / P229R / N276K pI variants on one chain with the S364K / E357Q:L368D / K370S scaffold, the C220S, and S364K / E357Q scaffolds, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Included in each of these scaffolds are sequences that are 90, 95, 98, and 99% identical (as defined herein) to the listed sequences and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 additional amino acid substitutions (compared to the "parent" in the figure, which already contains some amino acid modifications compared to the parent human IgG1 (or IgG2 or IgG4 depending on the scaffold), as will be appreciated by those skilled in the art). That is, the recited scaffolds may contain additional amino acid modifications (generally amino acid substitutions) in addition to the sqvariants, pI variants, and deletion variants contained within the scaffolds of this figure. [Figure 7C]The following shows the sequences of some useful 1+1 Fab-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, without the Fv sequences (e.g., scFv and Fab side VH): Scaffold 1 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, C220S on the chain with the S364K / E357Q scaffold variant, N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S sqvariant, C220S on the chain with the S364K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S sqvariant, C220S on the chain with the S364K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368E / K370S sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 4 is based on human IgG1 (356E / 358M allotype) and contains the D401K:K360E / Q362E / T411E sqvariant, C220S on the chain with the D401K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the K360E / Q362E / T411E sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains.Scaffold 5 is based on human IgG1 (356D / 358L allotype) and contains the S364K / E357Q:L368D / K370S sq variant, C220S on the chain with the S364K / E357Q sq variant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 6 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S sqvariant, C220S on the chain with the S364K / E357Q sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sqvariant, E233P / L234V / L235A / G236del / S267K deletion variants on both chains as well as the N297A variant on both chains. Scaffold 7 is identical to 6 except the mutation is N297S. Scaffold 8 is based on human IgG4 and contains an S364K / E357Q:L368D / K370S scubariant, an N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scubariant, as well as an S228P (EU numbering, which is S241P in Kabat) variant on both chains that eliminates Fab arm exchange, as known in the art. Scaffold 9 is based on human IgG2 and contains an S364K / E357Q:L368D / K370S scubariant, an N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scubariant. Scaffold 10 is based on human IgG2 and contains the S364K / E357Q:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant as well as the S267K variant on both chains. Scaffold 11 is identical to scaffold 1 except that it contains the M428L / N434S Xtend mutations.Scaffold 12 is based on human IgG1 (356E / 358M allotype) and includes the P217R / P229R / N276K pI variants on one chain with the S364K / E357Q:L368D / K370S scaffold, the C220S, and S364K / E357Q scaffolds, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Included in each of these scaffolds are sequences that are 90, 95, 98, and 99% identical (as defined herein) to the listed sequences and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 additional amino acid substitutions (compared to the "parent" in the figure, which already contains some amino acid modifications compared to the parent human IgG1 (or IgG2 or IgG4 depending on the scaffold), as will be appreciated by those skilled in the art). That is, the recited scaffolds may contain additional amino acid modifications (generally amino acid substitutions) in addition to the sqvariants, pI variants, and deletion variants contained within the scaffolds of this figure. [Figure 7D]The following shows the sequences of some useful 1+1 Fab-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, without the Fv sequences (e.g., scFv and Fab side VH): Scaffold 1 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, C220S on the chain with the S364K / E357Q scaffold variant, N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S sqvariant, C220S on the chain with the S364K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S sqvariant, C220S on the chain with the S364K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368E / K370S sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 4 is based on human IgG1 (356E / 358M allotype) and contains the D401K:K360E / Q362E / T411E sqvariant, C220S on the chain with the D401K sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the K360E / Q362E / T411E sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains.Scaffold 5 is based on human IgG1 (356D / 358L allotype) and contains the S364K / E357Q:L368D / K370S sq variant, C220S on the chain with the S364K / E357Q sq variant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 6 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S sqvariant, C220S on the chain with the S364K / E357Q sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sqvariant, E233P / L234V / L235A / G236del / S267K deletion variants on both chains as well as the N297A variant on both chains. Scaffold 7 is identical to 6 except the mutation is N297S. Scaffold 8 is based on human IgG4 and contains an S364K / E357Q:L368D / K370S scubariant, an N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scubariant, as well as an S228P (EU numbering, which is S241P in Kabat) variant on both chains that eliminates Fab arm exchange, as known in the art. Scaffold 9 is based on human IgG2 and contains an S364K / E357Q:L368D / K370S scubariant, an N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scubariant. Scaffold 10 is based on human IgG2 and contains the S364K / E357Q:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant as well as the S267K variant on both chains. Scaffold 11 is identical to scaffold 1 except that it contains the M428L / N434S Xtend mutations.Scaffold 12 is based on human IgG1 (356E / 358M allotype) and includes the P217R / P229R / N276K pI variants on one chain with the S364K / E357Q:L368D / K370S scaffold, the C220S, and S364K / E357Q scaffolds, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Included in each of these scaffolds are sequences that are 90, 95, 98, and 99% identical (as defined herein) to the listed sequences and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 additional amino acid substitutions (compared to the "parent" in the figure, which already contains some amino acid modifications compared to the parent human IgG1 (or IgG2 or IgG4 depending on the scaffold), as will be appreciated by those skilled in the art). That is, the recited scaffolds may contain additional amino acid modifications (generally amino acid substitutions) in addition to the sqvariants, pI variants, and deletion variants contained within the scaffolds of this figure. [Figure 8A]The following shows the sequences of some useful 2+1 Fab2-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, without the Fv sequences (e.g., scFv and Fab side VH): Scaffold 1 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368E / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 4 is based on human IgG1 (356E / 358M allotype) and contains the D401K:K360E / Q362E / T411E sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the K360E / Q362E / T411E sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 5 is based on human IgG1 (356D / 358L allotype) and contains the S364K / E357Q:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains.Scaffold 6 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains, as well as the N297A variant on both chains. Scaffold 7 is identical to 6, except that the mutation is N297S. Scaffold 8 is identical to Scaffold 1, except that it contains the M428L / N434S Xtend mutations. Scaffold 9 is based on human IgG1 (356E / 358M allotype) and includes the S364K / E357Q:L368D / K370S scaffold variant, the P217R / P229R / N276K pI variant on one chain with the S364K / E357Q scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Included in each of these scaffolds are sequences that are 90, 95, 98, and 99% identical (as defined herein) to the listed sequences and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 additional amino acid substitutions (compared to the "parent" in the figure, which already contains some amino acid modifications compared to the parent human IgG1 (or IgG2 or IgG4 depending on the scaffold), as will be appreciated by those skilled in the art). That is, the recited scaffolds may contain additional amino acid modifications (generally amino acid substitutions) in addition to the sqvariants, pI variants, and deletion variants contained within the scaffolds of this figure. [Figure 8B]The following shows the sequences of some useful 2+1 Fab2-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, without the Fv sequences (e.g., scFv and Fab side VH): Scaffold 1 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368E / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 4 is based on human IgG1 (356E / 358M allotype) and contains the D401K:K360E / Q362E / T411E sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the K360E / Q362E / T411E sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 5 is based on human IgG1 (356D / 358L allotype) and contains the S364K / E357Q:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains.Scaffold 6 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains, as well as the N297A variant on both chains. Scaffold 7 is identical to 6, except that the mutation is N297S. Scaffold 8 is identical to Scaffold 1, except that it contains the M428L / N434S Xtend mutations. Scaffold 9 is based on human IgG1 (356E / 358M allotype) and includes the S364K / E357Q:L368D / K370S scaffold variant, the P217R / P229R / N276K pI variant on one chain with the S364K / E357Q scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Included in each of these scaffolds are sequences that are 90, 95, 98, and 99% identical (as defined herein) to the listed sequences and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 additional amino acid substitutions (compared to the "parent" in the figure, which already contains some amino acid modifications compared to the parent human IgG1 (or IgG2 or IgG4 depending on the scaffold), as will be appreciated by those skilled in the art). That is, the recited scaffolds may contain additional amino acid modifications (generally amino acid substitutions) in addition to the sqvariants, pI variants, and deletion variants contained within the scaffolds of this figure. [Figure 8C]The following shows the sequences of some useful 2+1 Fab2-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, without the Fv sequences (e.g., scFv and Fab side VH): Scaffold 1 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368E / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 4 is based on human IgG1 (356E / 358M allotype) and contains the D401K:K360E / Q362E / T411E sqvariant, N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the K360E / Q362E / T411E sqvariant, and E233P / L234V / L235A / G236del / S267K deletion variants on both chains. Scaffold 5 is based on human IgG1 (356D / 358L allotype) and contains the S364K / E357Q:L368D / K370S sq variant, the N208D / Q295E / N384D / Q418E / N421D pI variants on the chain with the L368D / K370S sq variant, and the E233P / L234V / L235A / G236del / S267K deletion variants on both chains.Scaffold 6 is based on human IgG1 (356E / 358M allotype) and contains the S364K / E357Q:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D pI variant on the chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains, as well as the N297A variant on both chains. Scaffold 7 is identical to 6, except that the mutation is N297S. Scaffold 8 is identical to Scaffold 1, except that it contains the M428L / N434S Xtend mutations. Scaffold 9 is based on human IgG1 (356E / 358M allotype) and includes the S364K / E357Q:L368D / K370S scaffold variant, the P217R / P229R / N276K pI variant on one chain with the S364K / E357Q scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Included in each of these scaffolds are sequences that are 90, 95, 98, and 99% identical (as defined herein) to the listed sequences and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 additional amino acid substitutions (compared to the "parent" in the figure, which already contains some amino acid modifications compared to the parent human IgG1 (or IgG2 or IgG4 depending on the scaffold), as will be appreciated by those skilled in the art). That is, the recited scaffolds may contain additional amino acid modifications (generally amino acid substitutions) in addition to the sqvariants, pI variants, and deletion variants contained within the scaffolds of this figure. [Figure 9] Some useful constant light chain domain framework sequences are shown, based on human IgG1, without including Fv sequences (e.g., scFv or Fab). Included herein are constant light chain framework sequences that are 90, 95, 98 and 99% identical (as defined herein) to the listed sequences and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10 additional amino acid modifications. [Figure 10A]5 shows the sequence of an exemplary anti-CD3 scFv suitable for use in the bispecific antibodies of the present invention. The CDRs are underlined, the scFv linker is double underlined (in the sequence, the scFv linker is a positively charged scFv(GKPGS)4 linker (SEQ ID NO:15), but as will be appreciated by those skilled in the art, this linker can be replaced by other linkers, including uncharged or negatively charged linkers, some of which are shown in FIG. 5), and the slashes indicate the boundaries of the variable domains. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figures, these VH and VL sequences can be used in either scFv or Fab format. [Figure 10B] 5 shows the sequence of an exemplary anti-CD3 scFv suitable for use in the bispecific antibodies of the present invention. The CDRs are underlined, the scFv linker is double underlined (in the sequence, the scFv linker is a positively charged scFv(GKPGS)4 linker (SEQ ID NO:15), but as will be appreciated by those skilled in the art, this linker can be replaced by other linkers, including uncharged or negatively charged linkers, some of which are shown in FIG. 5), and the slashes indicate the boundaries of the variable domains. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figures, these VH and VL sequences can be used in either scFv or Fab format. [Figure 10C]5 shows the sequence of an exemplary anti-CD3 scFv suitable for use in the bispecific antibodies of the present invention. The CDRs are underlined, the scFv linker is double underlined (in the sequence, the scFv linker is a positively charged scFv(GKPGS)4 linker (SEQ ID NO:15), but as will be appreciated by those skilled in the art, this linker can be replaced by other linkers, including uncharged or negatively charged linkers, some of which are shown in FIG. 5), and the slashes indicate the boundaries of the variable domains. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figures, these VH and VL sequences can be used in either scFv or Fab format. [Figure 10D] 5 shows the sequence of an exemplary anti-CD3 scFv suitable for use in the bispecific antibodies of the present invention. The CDRs are underlined, the scFv linker is double underlined (in the sequence, the scFv linker is a positively charged scFv(GKPGS)4 linker (SEQ ID NO:15), but as will be appreciated by those skilled in the art, this linker can be replaced by other linkers, including uncharged or negatively charged linkers, some of which are shown in FIG. 5), and the slashes indicate the boundaries of the variable domains. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figures, these VH and VL sequences can be used in either scFv or Fab format. [Figure 10E]5 shows the sequence of an exemplary anti-CD3 scFv suitable for use in the bispecific antibodies of the present invention. The CDRs are underlined, the scFv linker is double underlined (in the sequence, the scFv linker is a positively charged scFv(GKPGS)4 linker (SEQ ID NO:15), but as will be appreciated by those skilled in the art, this linker can be replaced by other linkers, including uncharged or negatively charged linkers, some of which are shown in FIG. 5), and the slashes indicate the boundaries of the variable domains. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figures, these VH and VL sequences can be used in either scFv or Fab format. [Figure 10F] 5 shows the sequence of an exemplary anti-CD3 scFv suitable for use in the bispecific antibodies of the present invention. The CDRs are underlined, the scFv linker is double underlined (in the sequence, the scFv linker is a positively charged scFv(GKPGS)4 linker (SEQ ID NO:15), but as will be appreciated by those skilled in the art, this linker can be replaced by other linkers, including uncharged or negatively charged linkers, some of which are shown in FIG. 5), and the slashes indicate the boundaries of the variable domains. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figures, these VH and VL sequences can be used in either scFv or Fab format. [Figure 11A]1 shows antigen sequences for some of the antigens used in the present invention, including both human and cynomolgus monkey, to facilitate the development of antigen binding domains that bind to both, amenable to clinical development. [Figure 11B] 1 shows antigen sequences for some of the antigens used in the present invention, including both human and cynomolgus monkey, to facilitate the development of antigen binding domains that bind to both, amenable to clinical development. [Figure 12] The variable heavy and variable light sequences of the humanized GPC3-A variants are shown. The CDRs are underlined, and the slash indicates the boundary between the variable region and the constant domain. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. Furthermore, each of the variable heavy domains shown herein can be paired with any other αGPC3-A variable light domain, and each of the variable light domains shown herein can be paired with any other αGPC3-A variable heavy domain. [Figure 13A]The variable heavy and variable light sequences of GPC3-A variants engineered for reduced tendency to degradation (e.g., aspartic acid isomerization and deamidation), regulated GPC3 binding affinity and / or selectivity for high GPC3 expressing cell lines are shown. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. Furthermore, each of the variable heavy domains shown herein can be paired with any other αGPC3 variable light domain, and each of the variable light domains shown herein can be paired with any other αGPC3 variable heavy domain. [Figure 13B] The variable heavy and variable light sequences of GPC3-A variants engineered for reduced tendency to degradation (e.g., aspartic acid isomerization and deamidation), regulated GPC3 binding affinity and / or selectivity for high GPC3 expressing cell lines are shown. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. Furthermore, each of the variable heavy domains shown herein can be paired with any other αGPC3 variable light domain, and each of the variable light domains shown herein can be paired with any other αGPC3 variable heavy domain. [Figure 13C]The variable heavy and variable light sequences of GPC3-A variants engineered for reduced tendency to degradation (e.g., aspartic acid isomerization and deamidation), regulated GPC3 binding affinity and / or selectivity for high GPC3 expressing cell lines are shown. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. Furthermore, each of the variable heavy domains shown herein can be paired with any other αGPC3 variable light domain, and each of the variable light domains shown herein can be paired with any other αGPC3 variable heavy domain. [Figure 13D] The variable heavy and variable light sequences of GPC3-A variants engineered for reduced tendency to degradation (e.g., aspartic acid isomerization and deamidation), regulated GPC3 binding affinity and / or selectivity for high GPC3 expressing cell lines are shown. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. Furthermore, each of the variable heavy domains shown herein can be paired with any other αGPC3 variable light domain, and each of the variable light domains shown herein can be paired with any other αGPC3 variable heavy domain. [Figure 13E]The variable heavy and variable light sequences of GPC3-A variants engineered for reduced tendency to degradation (e.g., aspartic acid isomerization and deamidation), regulated GPC3 binding affinity and / or selectivity for high GPC3 expressing cell lines are shown. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. Furthermore, each of the variable heavy domains shown herein can be paired with any other αGPC3 variable light domain, and each of the variable light domains shown herein can be paired with any other αGPC3 variable heavy domain. [Figure 13F] The variable heavy and variable light sequences of GPC3-A variants engineered for reduced tendency to degradation (e.g., aspartic acid isomerization and deamidation), regulated GPC3 binding affinity and / or selectivity for high GPC3 expressing cell lines are shown. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. Furthermore, each of the variable heavy domains shown herein can be paired with any other αGPC3 variable light domain, and each of the variable light domains shown herein can be paired with any other αGPC3 variable heavy domain. [Figure 13G]The variable heavy and variable light sequences of GPC3-A variants engineered for reduced tendency to degradation (e.g., aspartic acid isomerization and deamidation), regulated GPC3 binding affinity and / or selectivity for high GPC3 expressing cell lines are shown. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for any sequence described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore, not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. Furthermore, each of the variable heavy domains shown herein can be paired with any other αGPC3 variable light domain, and each of the variable light domains shown herein can be paired with any other αGPC3 variable heavy domain. [Figure 14A] Figure 1 shows a bivalent anti-GPC3 mAb with E233P / L234V / L235A / G236del / S267K deletion variant and an exemplary GPC3-A variant formatted as an IgG1 backbone. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for all sequences described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. [Figure 14B]Figure 1 shows a bivalent anti-GPC3 mAb with E233P / L234V / L235A / G236del / S267K deletion variant and an exemplary GPC3-A variant formatted as an IgG1 backbone. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for all sequences described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. [Figure 14C] Figure 1 shows a bivalent anti-GPC3 mAb with E233P / L234V / L235A / G236del / S267K deletion variant and an exemplary GPC3-A variant formatted as an IgG1 backbone. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for all sequences described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. [Figure 14D]Figure 1 shows a bivalent anti-GPC3 mAb with E233P / L234V / L235A / G236del / S267K deletion variant and an exemplary GPC3-A variant formatted as an IgG1 backbone. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for all sequences described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. [Figure 14E] Figure 1 shows a bivalent anti-GPC3 mAb with E233P / L234V / L235A / G236del / S267K deletion variant and an exemplary GPC3-A variant formatted as an IgG1 backbone. The CDRs are underlined and the slash indicates the boundary between the variable region and the constant domain. As is true for all sequences described herein and containing CDRs herein, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 2, and therefore not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. Furthermore, for all sequences in the figure, these VH and VL sequences can be used in either scFv or Fab format. [Figure 15A] Two formats of the invention are shown: Figure 15A shows a "1+1 Fab-scFv-Fc" format, where the first Fab arm binds to GPC3 and the second scFv arm binds to CD3; Figure 15B shows a "2+1 Fab2-scFv-Fc" format, with the first Fab arm binding to GPC3 and the second Fab-scFv arm binding to CD3, where the Fab binds to GPC3 and the scFv binds to CD3. [Figure 15B]Two formats of the invention are shown: Figure 15A shows a "1+1 Fab-scFv-Fc" format, where the first Fab arm binds to GPC3 and the second scFv arm binds to CD3; Figure 15B shows a "2+1 Fab2-scFv-Fc" format, with the first Fab arm binding to GPC3 and the second Fab-scFv arm binding to CD3, where the Fab binds to GPC3 and the scFv binds to CD3. [Figure 16A] Illustrated are sequences of exemplary αGPC3×αCD3 bsAbs in 1+1 Fab-scFv-Fc format and including CD3 High scFv (H1.30_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. Additionally, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 16B] Illustrated are sequences of exemplary αGPC3×αCD3 bsAbs in 1+1 Fab-scFv-Fc format and including CD3 High scFv (H1.30_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. Additionally, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 16C] Illustrated are sequences of exemplary αGPC3×αCD3 bsAbs in 1+1 Fab-scFv-Fc format and including CD3 High scFv (H1.30_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. Additionally, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 17A] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 1+1 Fab-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). The CDRs are underlined and the slash indicates the boundary between the variable region and other chain components (e.g., constant region and domain linker). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 17B]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 1+1 Fab-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). The CDRs are underlined and the slash indicates the boundary between the variable region and other chain components (e.g., constant region and domain linker). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 17C] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 1+1 Fab-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). The CDRs are underlined and the slash indicates the boundary between the variable region and other chain components (e.g., constant region and domain linker). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 17D]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 1+1 Fab-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). The CDRs are underlined and the slash indicates the boundary between the variable region and other chain components (e.g., constant region and domain linker). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 17E] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 1+1 Fab-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). The CDRs are underlined and the slash indicates the boundary between the variable region and other chain components (e.g., constant region and domain linker). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 17F]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 1+1 Fab-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). The CDRs are underlined and the slash indicates the boundary between the variable region and other chain components (e.g., constant region and domain linker). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 17G] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 1+1 Fab-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). The CDRs are underlined and the slash indicates the boundary between the variable region and other chain components (e.g., constant region and domain linker). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 18A]Illustrated are sequences of exemplary αGPC3×αCD3 bsAbs in 2+1 Fab2-scFv-Fc format and including CD3 High scFv (H1.30_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. Additionally, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 18B] Illustrated are sequences of exemplary αGPC3×αCD3 bsAbs in 2+1 Fab2-scFv-Fc format and including CD3 High scFv (H1.30_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. Additionally, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 19A]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 19B] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 19C]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 19D] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 19E]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 19F] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 19G]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 19H] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#1 scFv (H1.32_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 20A]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#2 scFv (H1.89_L1.47 in either VHVL or VLVH orientation). The CDRs are underlined and the slash indicates the boundary between the variable region and other chain components (e.g., constant region and domain linker). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 20B] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#2 scFv (H1.89_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 20C]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#2 scFv (H1.89_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 20D] 1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#2 scFv (H1.89_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 20E]1 shows the sequence of an exemplary αGPC3×αCD3 bsAb, which is in a 2+1 Fab2-scFv-Fc format and includes CD3 High-Int#2 scFv (H1.89_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. In addition, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 21] Illustrated are sequences of exemplary αGPC3-A×αCD3 bsAbs in 2+1 Fab2-scFv-Fc format and including CD3 Intermediate scFv (H1.33_L1.47 in either VHVL or VLVH orientation). CDRs are underlined and slashes indicate the boundaries between the variable regions and other chain components (e.g., constant regions and domain linkers). Note that αGPC3-A×αCD3 bsAbs can utilize variable regions, Fc regions, and constant domains that are 90, 95, 98, and 99% identical (as defined herein) and / or contain 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid substitutions. Additionally, each sequence outlined herein can include or exclude the M428L / N434S variant in one or preferably both Fc domains, which results in a longer half-life in serum. [Figure 22] FIG. 1 shows sequences based on the comparator αGPC3×αCD3 bispecific antibody disclosed in WO 2016 / 047722. [Figure 23]Figure 1 shows the binding affinity of five GPC-A affinity variants H1.1_L1.6, H1.1_L1.29, H1.1_L1.16, H1.1_L1.23, and H1.1_L1.31 (formatted as 1+1 Fab-scFv-Fc with CD3 High-Int#1 scFv) to human and cynomolgus GPC3 antigens. The affinity variants showed a broad range of binding (4-400 nM KD) and were cross-reactive to cynomolgus GPC3. [Figure 24] Binding of three GPC-A affinity variants H1.1_L1.29, H1.1_L1.16, and H1.1_L1.31 (formatted as 1+1 Fab-scFv-Fc with CD3 High-Int#1 scFv) to HepG2 (GPC3high) cells. [Diagram 25] Figure 2 shows the binding of GPC3-A variants engineered to eliminate degradation propensity to human GPC3-A, with some variants showing reduced binding affinity to GPC3. [Figure 26] Figure 2 shows binding of GPC3-A variants engineered to eliminate degradation propensity to human and cynomolgus GPC3-A. Several of the variants showed reduced binding, biphasic binding, or abolished binding to GPC3. [Figure 27] Figure 1 shows A) induction of RTCC and B) T cell activation (as indicated by IFNγ secretion) by the αGPC3 x αCD3 bsAb comparator XENP31308. XENP31308 induces robust RTCC and T cell activation in the presence of GPC3highHepG2, GPC3medHuh7, and GPC3lowHEK293 cells, respectively (effector to target ratio of 10:1). [Figure 28] Shown is A) induction of RTCC and B) T cell activation (as indicated by IFNγ secretion) by the αGPC3×αCD3 bsAb comparator XENP31308 at a 1:1 effector to target ratio. [Figure 29]Figure 1 shows activation of T cells (as indicated by IFNγ release) in the presence of HepG2 (GPC3high) cells by GPC3 binding domains that bind different GPC3 epitopes formatted as 2+1 Fab2-scFv-Fc bsAbs with CD3 High-Int#1. Each of the bsAbs with GPC3 binding domains other than GPC3-A activated T cells less strongly than the bsAb with the GPC3-A binding domain. [Diagram 30] Figure 1 shows the binding of A) GPC3-A H1.1_L1.16 (100 nM) and B) GPC3-A H1.1_L1.31 (400 nM) variants to HepG2 (GPC3high) in the context of 1+1 Fab-scFv-Fc bsAb, 2+1 Fab2-scFv-Fc bsAb, and bivalent monospecific mAb. The bsAb in 2+1 format showed stronger binding compared to the bsAb in 1+1 format (and binding equivalent to the bivalent monospecific mAb). The potency shift from 1+1 to 2+1 format was more pronounced for the lower binding affinity GPC3-A_H1.1_L1.31 (400 nM) variant. [Diagram 31] Figure 1 shows A) induction of RTCC and B) cytokine release by T cells after incubation of HepG2(GPC3high) cells with T cells (effector:target ratio of 10:1) and αGPC3×αCD3 bsAb in 1+1 Fab-scFv-Fc-Fc format with different CD3 binding affinities. Detuning CD3 binding affinity from CD3 High to CD3 High-Int#1 significantly reduces RTCC efficacy and more importantly IFNγ release, potentially mitigating cytokine release syndrome. [Diagram 32]Figure 1 shows A) induction of RTCC and B) cytokine release by T cells following incubation of HepG2 (GPC3high) cells with T cells (effector:target ratio of 10:1) and αGPC3×αCD3 bsAb in 2+1 Fab2-scFv-Fc format with different CD3 binding affinities. Detuning CD3 binding affinity from CD3 High to CD3 High-Int#1 did not significantly reduce potency, whereas further reduction to CD3 High-Int#2 significantly reduced potency. [Diagram 33] Figure 1 shows A) induction of RTCC and B) cytokine secretion in the presence of HepG2 (GPC3high) cells by three GPC3-A affinity variants formatted as 2+1 Fab2-scFv-Fc with CD3 High-Int#2[VLVH]scFv. The bsAbs induced RTCC in a manner that correlated with their reduced affinity, with XENP37625 with 70nm GPC3 affinity variant showing the strongest potency and XENP37626 with 400nM GPC3 affinity variant showing the weakest potency. Surprisingly, the H1.1_L1.23 200nM affinity variant did not appear to be significantly different from the 400nM affinity variant. [Diagram 34] Figure 2 shows induction of RTCC in A) HepG2 cells and B) HEK293 cells by additional αGPC3 x αCD3 bsAbs of the invention (XENP36935, XENP37430, XENP34364, XENP34920, XENP35843, XENP36939, and XENP37433) at an effector to target ratio of 1:1. [Diagram 35]Figure 1 shows A) HepG2 cells and B) HEK293 cells at an effector to target ratio of 1:1 and induction of IFNγ release by T cells in the presence of additional αGPC3×αCD3 bsAbs of the invention (XENP36935, XENP37430, XENP34364, XENP34920, XENP35843, XENP36939, and XENP37433). XENP37430 and XENP34920 showed the highest potency against HepG2, while XENP36939 had the lowest potency. The potency of each bsAb in inducing IFNγ release was proportional to their potency in inducing RTCC. The majority of bsAbs selectively induced killing (and cytokine release) in the presence of HepG2 (GPChigh) over HEK293 (GPClow). [Diagram 36] Figure 1 shows induction of RTCC in HepG2 by αGPC3 x αCD3 bsAbs of the invention (XENP36935, XENP37430, XENP37625, XENP35843, XENP34920, XENP37624, XENP36939, XENP37433, and XENP37626) and comparator bsAb XENP31308 at an effector to target ratio of 10:1. BsAbs XENP34920, XENP37430, XENP37625, and XENP37624 showed similar potency (EC50 below 100ng / mL) as comparator XENP31308. The other bsAbs XENP35843, XENP36935, XENP37433, and XENP37626 were much less potent but were able to achieve effective killing at higher concentrations. XENP36939 showed no killing. [Figure 37] Figure 1 shows induction of RTCC in HepG2 by αGPC3 x αCD3 bsAbs of the invention (XENP36935, XENP37430, XENP37625, XENP35843, XENP34920, XENP37624, XENP36939, XENP37433, and XENP37626) at an effector to target ratio of 1:1 and the comparator XENP31308. [Figure 38]Figure 1 shows induction of RTCC in Huh7 by αGPC3 x αCD3 bsAbs of the invention (XENP36935, XENP37430, XENP37625, XENP35843, XENP34920, XENP37624, XENP36939, XENP37433, and XENP37626) at an effector to target ratio of 10:1 and the comparator XENP31308. [Figure 39] Figure 1 shows the growth of GPC3high HepG2 tumors over time in a 2D Incucyte model. A treatment concentration of 10μg / mL of each test article and an E:F ratio of 10:1 were used. As seen in other Incucyte studies, XENP36939 showed no significant activity and efficacy was similar to the PBS control. However, most other studies, including XENP34920, XENP37430, XENP36935, XENP37624, and XENP37625, showed similar efficacy to the comparator XENP31308. [Diagram 40] Figure 1 shows the growth of GPC3med Huh7 tumors over time in a 2D Incucyte model. A treatment concentration of 10ug / mL of each test article was used. None of the bsAbs (XENP36939, XENP37433, or XENP37626) with 400nM GPC3 binders were able to significantly inhibit cell proliferation. However, all 2+1 bsAbs with either 70nM or 100nM GPC3 binders were able to prevent proliferation with similar efficacy as the comparator. [Diagram 41]Results from a 3D spheroid RTCC model using Incucyte are shown. 1,000 HepG2 cells (with a doubling time of 24 hours) are seeded into wells and grown for 72-96 hours, then 40,000 PBMCs are added with the indicated bsAbs at a concentration of 10ug / mL (resulting in an E:T ratio of approximately 3:1). The intensity of each signal (tumor or immune cell) was integrated across each well. The curves for XENP34920 and XENP37624 were more similar than expected considering that XENP37625 has a lower affinity CD3 arm (High-Int#2) compared to XENP36935 (High-In#1). After 144 hours, all bsAbs except XENP36939 show 100% efficacy. [Diagram 42] Results are shown for a 3D spheroid RTCC model using Incucyte. 1,000 HepG2 cells (with a doubling time of 24 hours) are seeded into wells and grown for 72-96 hours, then 40,000 PBMCs are added with the indicated bsAbs at a concentration of 10ug / mL (resulting in an E:T ratio of approximately 3:1). The intensity of each signal (tumor or immune cell) was integrated across each well. Interestingly, one of the weaker affinity 400nM GPC3 binders showed the highest level of proliferation. With the exception of the 1+1 constructs XENP35843 and XENP36939, the remaining bsAbs showed similar or higher levels of immune cell proliferation than the comparator XENP31308. [Diagram 43]Photographs of results from a 3D spheroid RTCC model using Incucyte are shown. 1,000 HepG2 cells (with a doubling time of 24 hours) are seeded into wells and grown for 72-96 hours, then 40,000 PBMCs are added with the indicated bsAbs at a concentration of 10ug / mL (resulting in an E:T ratio of approximately 3:1). The target tumor cell spheroids can be seen most clearly in the PBS control column, with the PBMCs clustered around the spheroids and remaining unchanged over time. Comparing the PBS control column to columns treated with XENP31308, XENP37625, XENP34920, and XENP36935, the proliferation of PBMCs as immune cells kill tumor cells can be visualized. [Figure 44A] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Figure 44B] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Figure 44C] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Fig.44D] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Figure 44E] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Fig.44F] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Figure 44G] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Fig.44H] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Fig.44I]Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. [Fig.44J] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and a second anti-antigen binding domain. Additionally, mAb-Fv, mAb-scFv, central scFv, central Fv, one-armed central scFv, one-scFv-mAb, scFv-mAb, and dual scFv are all shown. For all scFv domains shown, they can be either variable heavy chain-(optional linker)-variable light chain from N-terminus to C-terminus, or vice versa. Additionally, in the case of one-arm scFv-mAb, the scFv can be attached either to the N-terminus of the heavy chain monomer or to the N-terminus of the light chain. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0037] The present invention provides heterodimeric bispecific antibodies that bind to human CD3ε and human GPC3.

[0038] A. Overview Anti-bispecific antibodies that simultaneously link CD3 and tumor antigen targets are used to redirect T cells to attack and lyse target tumor cells. Examples include Bite® and DART formats that monovalently link CD3 and tumor antigens. Although CD3 targeting approaches have shown considerable promise, a common side effect of such treatments is the associated production of cytokines, often resulting in toxic cytokine release syndrome. The anti-CD3 binding domain of the bispecific antibody links to all T cells, thus recruiting a high cytokine-producing CD4 T cell subset. Furthermore, CD4 T cell subsets include regulatory T cells, whose recruitment and proliferation can result in immune suppression and negatively impact long-term tumor suppression. Furthermore, these formats do not contain an Fc domain and exhibit a very short serum half-life in patients.

[0039] Provided herein are novel anti-CD3 x anti-GPC3 (also referred to as anti-GPC3 x anti-CD3, αCD3 x αGPC3, αGPC3 x αCD3, or simply GPC3 x CD3) heterodimeric bispecific antibodies and methods of using such antibodies for the treatment of cancer. In particular, provided herein are anti-CD3, anti-GPC3 bispecific antibodies in various formats. These bispecific antibodies are useful for the treatment of cancer, particularly cancers with increased GPC3 expression, such as renal cell carcinoma. Such antibodies are used to target CD3+ effector T cells to GPC3+ tumors, thereby enabling the CD3+ effector T cells to attack and lyse GPC3+ tumors.

[0040] Further, in some embodiments, the present disclosure provides bispecific antibodies with different binding affinities for human CD3 that can alter or reduce potential side effects of anti-CD3 therapy. That is, in some embodiments, the antibodies described herein are "strong" or "high affinity" binders for CD3 (e.g., heavy and light chain variable domains shown as H1.30_L1.47 (optionally including a charged linker as required)) and provide antibody constructs that include an anti-CD3 antigen binding domain that also binds to GPC3. In other embodiments, the antibodies described herein provide antibody constructs that include an anti-CD3 antigen binding domain that is a "light" or "low affinity" binder for CD3. Additional embodiments provide antibody constructs that include an anti-CD3 antigen binding domain with intermediate or "medium" affinity for CD3 that also binds to GPC3. While a large number of anti-CD3 antigen binding domains (ABDs) can be used, particularly useful embodiments use six different anti-CD3 ABDs, but in two scFv orientations as described herein. Affinity is generally measured using a Biacore assay.

[0041] It is understood that the "high, medium, low" anti-CD3 sequences provided herein can be used in a variety of heterodimerization formats, as described herein. In general, due to potential side effects of T cell recruitment, exemplary embodiments use formats that only bind monovalently to CD3, as shown in Figures 15A and 15B, and in the formats shown herein, it is the CD3 ABD that is the scFv, as described more fully herein. In contrast, the subject bispecific antibodies can bind GPC3 either monovalently (e.g., Figure 15A) or bivalently (e.g., Figure 15B).

[0042] Compositions comprising GPC3 binding domains are provided herein, including antibodies having such GPC3 binding domains (e.g., GPC3xCD3 bispecific antibodies). Subject antibodies comprising such GPC3 binding domains advantageously induce a variety of different immune responses depending on the specific GPC3 binding domain used. For example, subject antibodies show different selectivity for cells with different GPC3 expression, potency for GPC3-expressing cells, ability to induce cytokine release, and sensitivity to soluble GPC3. Such GPC3 binding domains and related antibodies are used, for example, to treat GPC3-associated cancers.

[0043] Thus, in one aspect, heterodimeric antibodies are provided herein that bind to two different antigens, e.g., the antibodies are "bispecific" in that they bind to two different target antigens, typically GPC3 and CD3, as described herein. These heterodimeric antibodies can bind to these target antigens either monovalently (e.g., there is a single antigen-binding domain, such as a pair of variable heavy and variable light domains) or bivalently (there are two antigen-binding domains, each of which binds to an antigen independently). In some embodiments, the heterodimeric antibodies provided herein comprise one CD3-binding domain and one GPC3-binding domain (e.g., heterodimeric antibodies in the "1+1 Fab-scFv-Fc" format described herein). In other embodiments, the heterodimeric antibodies provided herein comprise one CD3-binding domain and two GPC3-binding domains (e.g., heterodimeric antibodies in the "2+1 Fab2-scFv-Fc" format described herein). The heterodimeric antibodies provided herein are based on the use of different monomers that contain amino acid substitutions on the homodimer that "skew" the formation of the heterodimer, as outlined more fully below, and are linked to "pI variants" that allow for simple purification of the heterodimer away from the homodimer, as also outlined below. The heterodimeric bispecific antibodies provided generally rely on the use of engineered or variant Fc domains that can self-assemble in the production cell to produce the heterodimeric protein, and methods to generate and purify such heterodimeric proteins.

[0044] B. Nomenclature The antibodies provided herein are listed in several different formats. In some cases, each monomer of a particular antibody is given a unique "XENP" number, although longer sequences may contain shorter ones, as understood in the art. For example, the "scFv-Fc" monomer of a 1+1 Fab-scFv-Fc format antibody may have a first XENP number, but the scFv domain itself has a different XENP number. Some molecules have three polypeptides, so the XENP number is used as the name along with the components. Thus, the molecule XENP33744, which is a 2+1 Fab2-scFv-Fc format, contains three sequences (see Figure 18) of a "Fab-Fc heavy chain" monomer, 2) a "Fab-scFv-Fc heavy chain" monomer, and 3) a "light chain" monomer or equivalent, which one of skill in the art can easily identify through sequence alignment. These XENP numbers are in the sequence listing as well as the identifiers and used in the figures. In addition, one molecule containing three components generates multiple sequence identifiers. For example, a list of Fabs includes the complete heavy chain sequence, the variable heavy chain domain sequence, and the three CDRs of the variable heavy chain domain sequence, the complete light chain sequence, the variable light chain domain sequence, and the three CDRs of the variable light chain domain sequence. A Fab-scFv-Fc monomer includes the full length sequence, the variable heavy chain domain sequence, the three heavy chain CDR sequences, and the scFv sequence (including the scFv variable heavy chain domain sequence, the scFv variable light chain domain sequence, and the scFv linker). Note that in the molecules with some scFv domains herein, a single charged scFv linker (+H) is used, but others can be used. Furthermore, the nomenclature of certain antigen binding domains (e.g., GPC3 and CD3 binding domains) uses a format of the type "Hx.xx_Ly.yy", where the numbers are unique identifiers for the particular variable chain sequence. Thus, the Fv domain of the antigen binding domain is "H1L1", indicating that the variable heavy domain H1 is combined with the light domain L1. When these sequences are used as scFvs, the designation "H1L1" indicates that the variable heavy domain H1 is combined with the light domain L1 in the N-to-C-terminus VH-linker-VL orientation.This molecule, having the same sequences of heavy and light variable domains but in reverse order (VL-linker-VH orientation, N-terminus to C-terminus), will be designated "L1_H1.1". Similarly, different constructs can be "mixed and matched" heavy and light chains, as will be apparent from the sequence listing and figures.

[0045] Furthermore, the bispecific antibodies of the present invention may be referred to herein as "anti-CD3 x anti-GPC3", "αCD3 x αGPC3", "αGPC3 x αCD3", or simply as "GPC3 x CD3". Most of the formats used as scFvs have anti-CD3 ABD as the scFv, but the order of the antigens is not limiting, as discussed below.

[0046] C.Definition In order to facilitate a more complete understanding of this application, certain definitions are provided below. Such definitions are intended to encompass grammatical equivalents.

[0047] As used herein, "GPC3" refers to a protein belonging to the claudin family. The GPC3 sequence is shown, for example, in Figure 11. The ABD of the present invention binds to human GPC3.

[0048] As used herein, "removal" refers to reduction or elimination of activity. Thus, for example, "ablation of FcγR binding" means that an Fc region amino acid variant has less than 50% of the starting binding compared to an Fc region not containing the particular variant, preferably with greater than 70-80-90-95-98% loss of activity, and generally activity below detectable binding levels in Biacore, SPR, or BLI assays. Particularly useful in removing FcγR binding are those shown in FIG. 3, which are generally added to both monomers.

[0049] As used herein, "ADCC" or "antibody-dependent cell-mediated cytotoxicity" refers to a cell-mediated reaction in which non-specific cytotoxic cells expressing FcγR recognize bound antibody on a target cell and subsequently cause lysis of the target cell. ADCC correlates with binding to FcγRIIIa, and increased binding to FcγRIIIa results in increased ADCC activity.

[0050] As used herein, "ADCP" or antibody-dependent cell-mediated phagocytosis refers to a cell-mediated reaction in which non-specific phagocytes expressing FcγR recognize bound antibody on a target cell and subsequently cause phagocytosis of the target cell.

[0051] As used herein, the term "antibody" is used generically. Antibodies according to the invention can take several formats, as described herein, including conventional antibodies as well as antibody derivatives, fragments, and mimetics, including several bispecific formats as described herein.

[0052] Conventional immunoglobulin (Ig) antibodies are "Y" shaped tetramers. Each tetramer typically consists of two identical pairs of polypeptide chains, each pair having one "light" chain monomer (typically having a molecular weight of about 25 kDa) and one "heavy" chain monomer (typically having a molecular weight of about 50-70 kDa).

[0053] Other useful antibody formats include, but are not limited to, the 1+1 Fab-scFv-Fc and 2+1 Fab-scFv-Fc antibody formats described herein and shown in FIG. 15, as well as the "mAb-Fv", "mAb-scFv", "central Fv", "one-arm scFv-mAb", "scFv-mAb", "dual scFv", and "trident" format antibodies discussed below and shown in FIG. 44.

[0054] An antibody heavy chain typically comprises a variable heavy (VH) domain, which comprises vhCDR1-3, and an Fc domain, which comprises CH2-CH3 monomers. In some embodiments, the antibody heavy chain comprises a hinge and a CH1 domain. Conventional antibody heavy chains are monomers organized from N-terminus to C-terminus: VH-CH1-hinge-CH2-CH3. The CH1-hinge-CH2-CH3 are collectively referred to as the heavy chain "constant domain" or "constant region" and there are five different categories or "isotypes" (IgA, IgD, IgG, IgE, and IgM). Thus, as used herein, "isotype" refers to any subclass of immunoglobulins defined by the chemical and antigenic characteristics of their constant regions. It is understood that therapeutic antibodies may also include hybrids of isotypes and / or subclasses. For example, the antibodies described herein include the use of human IgG1 / G2 hybrids, as set forth in US Patent Application Publication No. 2009 / 0163699, which is incorporated by reference.

[0055] In some embodiments, the antibodies provided herein comprise an IgG isotype constant domain with several subclasses including, but not limited to, IgG1, IgG2, IgG3, and IgG4. In the IgG subclass of immunoglobulins, there are several immunoglobulin domains in the heavy chain. By "immunoglobulin (Ig) domain" herein is meant a region of an immunoglobulin with distinct tertiary structures. The heavy chain domains including the constant heavy chain CH domain and the hinge domain are of interest in the antibodies described herein. In the context of IgG antibodies, IgG isotypes each have three CH regions. Thus, the "CH" domains in the context of IgG are as follows: "CH1" refers to positions 118-220 according to the EU index as in Kabat; "CH2" refers to positions 237-340 according to the EU index as in Kabat, and "CH3" refers to positions 341-447 according to the EU index as in Kabat. As shown herein and described below, the pI variants can be in one or more CH regions as well as the hinge region discussed below.

[0056] It should be noted that IgG1 has different allotypes with polymorphisms at 356 (D or E) and 358 (L or M). The sequences depicted herein use the 356D / 358M allotype, but other allotypes are included herein. That is, any sequence comprising an IgG1 Fc domain included herein can have 356E / 358L instead of the 356D / 358M allotype. It is understood that therapeutic antibodies can also comprise isotype and / or subclass hybrids. For example, as shown in US Patent Application Publication No. 2009 / 0163699, which is incorporated by reference, the present antibodies comprise IgG1 / IgG2 hybrids in some embodiments.

[0057] As used herein, "Fc" or "Fc region" or "Fc domain" refers to a polypeptide comprising the constant region of an antibody, in some cases excluding all or a portion of the first constant region immunoglobulin domain (e.g., CH1), and in some cases optionally including all or a portion of the hinge. In the case of an IgG, the Fc domain comprises immunoglobulin domains CH2 and CH3 (Cγ2 and Cγ3), and optionally all or a portion of the hinge region between CH1 (Cγ1) and CH2 (Cγ2). Thus, in some cases, the Fc domain comprises, from N-terminus to C-terminus, CH2-CH3 and hinge-CH2-CH3. In some embodiments, the Fc domain is from human IgG1, IgG2, IgG3, or IgG4, with human IgG1 hinge-CH2-CH3, and IgG4 hinge-CH2-CH3 finding particular use in many embodiments. Additionally, in the case of human IgG1 Fc domains, the hinge frequently comprises a C220S amino acid substitution. Moreover, in the case of human IgG4 Fc domains, the hinge frequently comprises a S228P amino acid substitution. Although the boundaries of the Fc region may vary, the human IgG heavy chain Fc region is usually defined to include residues E216, C226 or A231 at its carboxyl-terminus, numbering according to the EU index of Kabat. In some embodiments, amino acid modifications are made to the Fc region to, for example, alter binding to one or more FcγR or FcRn, as described more fully below.

[0058] By "heavy chain constant region" herein is meant the CH1-hinge-CH2-CH3 portion of an antibody (or fragment thereof), excluding the variable heavy domain, which in the EU numbering for human IgG1 is amino acids 118 to 447. By "heavy chain constant region fragment" herein is meant a heavy chain constant region that has fewer amino acids from either or both the N-terminus and C-terminus, but still retains the ability to form a dimer with another heavy chain constant region.

[0059] Another type of Ig domain of the heavy chain is the hinge region. By "hinge" or "hinge region" or "antibody hinge region" or "hinge domain" herein is meant a flexible polypeptide comprising the amino acids between the first and second constant domains of an antibody. Structurally, the IgG CH1 domain ends at EU position 215 and the IgG CH2 domain begins at residue EU position 231. Thus, for IgG, the antibody hinge is defined herein to include positions 216 (E216 in IgG1) to 230 (p230 in IgG1), numbering according to the EU index as in Kabat. In some cases, "hinge fragments" are used that contain fewer amino acids at either or both the N-terminus and C-terminus of the hinge domain. As described herein, pI variants can be made in the hinge region as well. Many of the antibodies herein have at least one cysteine ​​at position 220 according to EU numbering (hinge region) replaced with a serine. Generally, this modification is on the "scFv monomer" side for most of the sequences shown herein, but can also be on the "Fab monomer" side to reduce disulfide formation, or both. Specifically included within the sequences herein are those in which one or both of these cysteines have been replaced (C220S).

[0060] As will be appreciated by those of skill in the art, the exact numbering and arrangement of heavy chain constant region domains may vary between different numbering systems. A useful comparison of the EU and Kabat heavy chain constant region numbering is as follows: see Edelman et al., 1969, Proc Natl Acad Sci USA 63:78-85, and Kabat et al., 1991, Sequences of Proteins of Immunological Interest, 5th Ed., United States Public Health Service, National Institutes of Health, Bethesda, which are incorporated by reference in their entireties.

[0061] [Table 1]

[0062] An antibody light chain generally comprises two domains: a variable light domain (VL) that contains the light chain CDRs vlCDR1-3, and a constant light domain (often referred to as CL or Cκ). Antibody light chains are usually organized from the N-terminus to the C-terminus: VL-CL.

[0063] As used herein, "antigen binding domain" or "ABD" refers to a set of six complementarity determining regions (CDRs) that, when present as part of a polypeptide sequence, specifically binds to a target antigen as discussed herein (e.g., GPC3 or CD3). As known in the art, these CDRs generally exist as a first set of variable heavy chain CDRs (vhCDRs or VHCDRs) and a second set of variable light chain CDRs (vlCDRs or VLCDRs), each of which contains three CDRs, namely, vhCDR1, vhCDR2, vhCDR3 variable heavy chain CDRs, and vlCDR1, vlCDR2, vlCDR3 vhCDR3 variable light chain CDRs. The CDRs are present in the variable heavy chain domain (vhCDR1-3) and the variable light chain domain (vlCDR1-3). The variable heavy chain domain and the variable light chain domain form the Fv region.

[0064] The antibodies described herein provide a number of different CDR sets. In this case, a "complete CDR set" includes three variable light chain CDRs and three variable heavy chain CDRs, such as vlCDR1, vlCDR2, vlCDR3, vhCDR1, vhCDR2, and vhCDR3. These may be part of a larger variable light chain or variable heavy chain domain, respectively. In addition, as more fully outlined herein, the variable heavy chain domain and the variable light chain domain may be on separate polypeptide chains when heavy and light chains are used (e.g., when Fabs are used) or on a single polypeptide chain in the case of scFv sequences.

[0065] As will be appreciated by those skilled in the art, the exact numbering and arrangement of the CDRs may vary between different numbering systems. However, it should be understood that the disclosure of a variable heavy chain sequence and / or a variable light chain sequence includes the disclosure of the associated (unique) CDRs. Thus, the disclosure of each variable heavy chain region is the disclosure of the vhCDRs (e.g., vhCDR1, vhCDR2, and vhCDR3), and the disclosure of each variable light chain region is the disclosure of the vlCDRs (e.g., vlCDR1, vlCDR2, and vlCDR3). A useful comparison of CDR numbering follows, see Lafranc et al., Dev. Comp. Immunol. 27(1):55-77 (2003).

[0066] [Table 2]

[0067] Throughout this specification, the Kabat numbering system is generally used when referring to residues within the variable domain (approximately residues 1-107 for light chain variable regions and residues 1-113 for heavy chain variable regions), while the EU numbering system is for the Fc region (see, e.g., Kabat et al. (1991), supra).

[0068] CDRs contribute to antigen binding, or more specifically, to the formation of the antigen binding domain and epitope binding site of an antibody. "Epitope" refers to a determinant that interacts with a specific antigen binding site in the variable region of an antibody molecule known as a paratope. An epitope is a grouping of molecules, such as amino acids or sugar side chains, that usually have specific structural characteristics, as well as specific charge characteristics. A single antigen may have more than one epitope.

[0069] An epitope may include amino acid residues that are directly involved in binding (also referred to as the immunodominant components of the epitope) and other amino acid residues that are not directly involved in binding, e.g., amino acid residues that are effectively blocked by a specific antigen-binding peptide, in other words, amino acid residues that lie within the footprint of a specific antigen-binding peptide.

[0070] Epitopes may be either conformational or linear. Conformational epitopes are generated by the spatial juxtaposition of amino acids from different segments of a linear polypeptide chain. Linear epitopes are those generated by adjacent amino acid residues within a polypeptide chain. Conformational and nonconformational epitopes can be distinguished in that the binding to the former but not the latter is lost in the presence of denaturing solvents.

[0071] An epitope typically comprises at least 3, more usually at least 5, or 8-10 amino acids in a unique spatial conformation. Antibodies that recognize the same epitope can be verified in a simple immunoassay showing the ability of one antibody to block the binding of another antibody to a target antigen, e.g., "binning". As outlined below, the present disclosure includes not only the recited antigen binding domains and antibodies herein, but also those that compete for binding to the epitope bound by the recited antigen binding domains.

[0072] In some embodiments, the six CDRs of an antigen-binding domain are contributed by a variable heavy domain and a variable light domain. In the "Fab" format, the set of six CDRs is contributed by two different polypeptide sequences, a variable heavy domain (vh or VH; comprising vhCDR1, vhCDR2, and vhCDR3), and a variable light domain (vl or VL; comprising vlCDR1, vlCDR2, and vlCDR3), with the C-terminus of the vh domain attached to the N-terminus of the CH1 domain of the heavy chain and the C-terminus of the vl domain attached to the N-terminus of the constant light domain (thus forming the light chain). In the scFv format, the vh and vl domains are covalently linked into a single polypeptide sequence by the use of a linker ("scFv linker"), generally as outlined herein, which can be (starting from the N-terminus) either vh-linker-vl or vl-linker-vh, with the former generally being preferred (with optional domain linkers on either side, depending on the format used (e.g., from Figure 44)). Generally, the C-terminus of the scFv domain is attached to the N-terminus of the hinge of the second monomer.

[0073] As used herein, a "variable region" or "variable domain" refers to a region of an immunoglobulin that includes one or more Ig domains substantially encoded by any of the Vκ, Vλ, and / or VH genes that constitute the kappa, lambda, and heavy chain immunoglobulin loci, respectively, and contains CDRs that confer antigen specificity. Thus, a "variable heavy domain" pairs with a "variable light domain" to form an antigen-binding domain ("ABD"). In addition, each variable domain includes three hypervariable regions ("complementarity determining regions", "CDRs") (VHCDR1, VHCDR2, and VHCDR3 for the variable heavy domain and VLCDR1, VLCDR2, and VLCDR3 for the variable light domain) and four framework (FR) regions, arranged from amino-terminus to carboxy-terminus in the following order: FR1-CDR1-FR2-CDR2-FR3-CDR3-FR4. The hypervariable regions generally correspond to approximately amino acid residues 24-34 (LCDR1; "L" indicates light chain), 50-56 (LCDR2), and 89-97 (LCDR3) in the light chain variable region and approximately 31-35B (HCDR1; "H" indicates heavy chain), 50-65 (HCDR2), and 95-102 (HCDR3) in the heavy chain variable region; Kabat et al., SEQUENCES OF PROTEINS OF IMMUNOLOGICAL INTEREST, 5th Ed. Public Health Service, National Institutes of Health, Bethesda, Md. (1991)) and / or residues forming hypervariable loops (e.g., residues 26-32 (LCDR1), 50-52 (LCDR2), and 91-96 (LCDR3) in the light chain variable region, and 26-32 (HCDR1), 53-55 (HCDR2), and 96-101 (HCDR3) in the heavy chain variable region; Chothia and Lesk (1987) J. Mol. Biol. 196:901-917. Particular CDRs of the invention are set out in Table 2.

[0074] As used herein, "Fab" or "Fab region" refers to a polypeptide that generally comprises the VH, CH1, VL, and CL immunoglobulin domains on two different polypeptide chains (e.g., VH-CH1 on one chain, VL-CL on the other chain). Fab can refer to this region alone or in the context of the bispecific antibodies described herein. In the context of Fab, Fab includes the Fv region in addition to the CH1 and CL domains.

[0075] As used herein, "Fv" or "Fv fragment" or "Fv region" refers to a polypeptide comprising the VL and VH domains of the ABD. The Fv region can be formatted as both a Fab (which, as noted above, are two distinct polypeptides that also generally comprise a constant region as outlined above) and an scFv, where the VL and VH domains combine (generally with a linker as discussed herein) to form an scFv.

[0076] "Single-chain Fv" or "scFv" herein generally refers to a variable heavy domain covalently linked to a variable light domain, forming a scFv or scFv domain using the scFv linker discussed herein. The scFv domain may be in either orientation, from N-terminus to C-terminus (VH-linker-VL or VL-linker-VH). In the sequences shown in the sequence listing and figures, the order of the VH and VL domains is indicated in the name. For example, H.X_L.Y means, from N-terminus to C-terminus, VH-linker-VL and L.Y_H.X means, VL-linker-VH. However, the disclosure of any "H_L" pair is meant to include them in either order.

[0077] Some embodiments of the subject antibodies provided herein comprise at least one scFv domain, which, although not naturally occurring, generally comprises a variable heavy domain and a variable light domain linked together by an scFv linker. As outlined herein, the scFv domain is generally oriented N-terminus to C-terminus as VH-scFv linker-VL, although this can be reversed for any scFv domain (or domain constructed using vh and vl sequences from a Fab) to VL-scFv linker-VH, with optional linkers at one or both ends depending on the format.

[0078] By "modification" herein is meant substitution, insertion and / or deletion of amino acids in a polypeptide sequence or alteration to a moiety chemically linked to a protein. For example, a modification may be an altered carbohydrate or PEG structure attached to a protein. By "amino acid modification" herein is meant substitution, insertion and / or deletion of amino acids in a polypeptide sequence. For clarity, unless otherwise stated, amino acid modifications are always to amino acids encoded by DNA, e.g., the 20 amino acids for which DNA and RNA have codons.

[0079] An "amino acid substitution" or "substitution" herein refers to replacing an amino acid at a particular position in a parent polypeptide sequence with a different amino acid. In particular, in some embodiments, the substitution is for an amino acid that does not naturally occur (either not naturally occurring in an organism or not occurring in any organism) at the particular position. For example, the substitution E272Y refers to a variant polypeptide, in this case an Fc variant, in which glutamic acid at position 272 is replaced with tyrosine. For clarity, a protein that has been engineered to change the nucleic acid coding sequence but not change the starting amino acid (e.g., replacing CGG (which codes for arginine) with CGA (which still codes for arginine) to increase expression levels in the host organism) is not an "amino acid substitution". That is, if a protein has the same amino acid at that particular starting position, despite the creation of a new gene that codes for the same protein, it is not an amino acid substitution.

[0080] As used herein, "amino acid insertion" or "insertion" refers to the addition of an amino acid sequence at a particular position in a parent polypeptide sequence. For example, -233E or 233E indicates the insertion of a glutamic acid after position 233 and before position 234. Furthermore, -233ADE or A233ADE indicates the insertion of AlaAspGlu after position 233 and before position 234.

[0081] As used herein, "amino acid deletion" or "deletion" refers to the removal of an amino acid sequence at a particular position in a parent polypeptide sequence. For example, E233- or E233#, E233() or E233del indicates the deletion of glutamic acid at position 233. Additionally, EDA233- or EDA233# indicates the deletion of the sequence GluAspAla beginning at position 233.

[0082] As used herein, "variant protein" or "protein variant" or "variant" refers to a protein that differs from that of a parent protein based on at least one amino acid modification. A protein variant has at least one amino acid modification compared to the parent protein, but not so many that the variant protein does not align with the parent protein using an alignment program such as described below. In general, the variant proteins outlined herein (such as variant Fc domains) are generally at least 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, or 99% identical to the parent protein using an alignment program such as BLAST, described below. As used herein, "variant" also refers to certain amino acid modifications that confer certain functions (e.g., "heterodimerization variants", "pI variants", "deletion variants", etc.).

[0083] As described below, in some embodiments, the parent polypeptide, e.g., the Fc parent polypeptide, is a human wild-type sequence, such as a heavy chain constant domain or Fc region from IgG1, IgG2, IgG3, or IgG4, although human sequences having variants can also function as "parent polypeptides", including, for example, the IgG1 / 2 hybrids of US Patent Application Publication No. 2006 / 0134105. The sequences of the protein variants described herein preferably have at least about 80% identity, most preferably at least about 90% identity, and more preferably at least about 95-98-99% identity to the parent protein sequence. Thus, as used herein, an "antibody variant" or "variant antibody" refers to an antibody that differs from a parent antibody by the basis of at least one amino acid modification; as used herein, an "IgG variant" or "variant IgG" refers to an antibody that differs from a parent IgG (again, often derived from a human IgG sequence) by the basis of at least one amino acid modification; and as used herein, an "immunoglobulin variant" or "variant immunoglobulin" refers to an immunoglobulin sequence that differs from that of the parent immunoglobulin sequence by the basis of at least one amino acid modification.

[0084] As used herein, "Fc variant" or "variant Fc" refers to a protein comprising an amino acid modification in the Fc domain. The modification may be an addition, deletion, or substitution. Fc variants are defined according to the amino acid modifications that compose them. Thus, for example, N434S or 434S is an Fc variant with a substitution serine at position 434 relative to the parent Fc polypeptide, where the numbering is according to the EU index. Similarly, M428L / N434S defines an Fc variant with substitutions M428L and N434S relative to the parent Fc polypeptide. The identity of the WT amino acids may not be specified, in which case the aforementioned variant is referred to as 428L / 434S. It is noted that the order in which the substitutions are provided is arbitrary, i.e., for example, 428L / 434S is the same Fc variant as 434S / 428L, etc. For all positions discussed in the present invention relating to antibodies or derivatives and fragments thereof (e.g., Fc domains), unless otherwise specified, the numbering of amino acid positions is according to the EU index. "EU index" or "Kabat-like EU index" or "EU numbering" scheme refers to the numbering of the EU antibody (Edelman et al., 1969, Proc Natl Acad Sci USA 63:78-85, hereby incorporated by reference in its entirety).

[0085] Generally, the variant Fc domain has at least about 80, 85, 90, 95, 97, 98 or 99 percent identity to the corresponding parent human IgG Fc domain (using the identity algorithm discussed below (in one embodiment the BLAST algorithm known in the art) with default parameters). Alternatively, the variant Fc domain may have 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20 amino acid modifications relative to the parent Fc domain. Alternatively, the variant Fc domain may have up to 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20 amino acid modifications compared to the parent Fc domain. Additionally, as discussed herein, the variant Fc domains described herein still retain the ability to form dimers with another Fc domain as measured using known techniques as described herein, such as non-denaturing gel electrophoresis.

[0086] "Protein" herein means at least two covalently attached amino acids, including proteins, polypeptides, oligopeptides, and peptides. Additionally, the polypeptides from which the antibodies described herein are made can include synthetic derivatization of one or more side chains or termini, glycosylation, pegylation, circular permutation, cyclization, linkers to other molecules, fusion to proteins or protein domains, and addition of peptide tags or labels.

[0087] As used herein, "residue" refers to the position in the protein and its associated amino acid identity. For example, asparagine 297 (also called Asn297 or N297) is the residue at position 297 in human antibody IgG1.

[0088] As used herein, "IgG subclass modification" or "isotype modification" refers to an amino acid modification that converts one amino acid of one IgG isotype to the corresponding amino acid of a different, matching IgG isotype. For example, since IgG1 contains a tyrosine and IgG2 contains a phenylalanine at EU position 296, an F296Y substitution in IgG2 is considered to be an IgG subclass modification.

[0089] As used herein, a "non-naturally occurring modification" refers to an amino acid modification that is not isotypic. For example, the substitution 434S in IgG1, IgG2, IgG3, or IgG4 (or hybrids thereof) is considered to be a non-naturally occurring modification, since none of the human IgGs contain a serine at position 434.

[0090] As used herein, "amino acid" and "amino acid identity" refer to one of the 20 naturally occurring amino acids encoded by DNA and RNA.

[0091] As used herein, "effector function" refers to a biochemical event that results from the interaction of an antibody Fc region with an Fc receptor or ligand. Effector functions include, but are not limited to, ADCC, ADCP, and CDC.

[0092] As used herein, "IgG Fc ligand" refers to any biologically derived molecule, preferably a polypeptide, that binds to the Fc region of an IgG antibody to form an Fc / Fc ligand complex. Fc ligands include, but are not limited to, FcγRI, FcγRII, FcγRIII, FcRn, C1q, C3, mannan-binding lectin, mannose receptor, staphylococcal protein A, streptococcal protein G, and viral FcγR. Fc ligands also include Fc receptor homologs (FcRH), a family of Fc receptors that are homologous to FcγR (Davis et al., 2002, Immunological Reviews 190:123-136, incorporated by reference in its entirety). Fc ligands may include yet to be discovered molecules that bind to Fc. Particular IgG Fc ligands are FcRn and Fc gamma receptors. As used herein, "Fc ligand" refers to any biologically derived molecule, preferably a polypeptide, that binds to the Fc region of an antibody to form an Fc / Fc ligand complex.

[0093] As used herein, "Fc gamma receptor," "FcγR," or "Fc gamma R" refers to any member of a family of proteins that binds to the IgG antibody Fc region and is encoded by the FcγR gene. In humans, this family includes FcγRI (CD64) including isoforms FcγRIa, FcγRIb, and FcγRIc, FcγRII (CD32) including isoforms FcγRIIa (including allotypes H131 and R131), FcγRIIb (including FcγRIIb-1 and FcγRIIb-2), and FcγRIIc, and FcγRIII (CD16) including isoforms FcγRIIIa (including allotypes V158 and F158), and FcγRIIIb (including allotypes FcγRIIb-NA1 and FcγRIIb-NA2) (Jefferis et al., 2002, Immunol Lett 82:57-65, entirely incorporated by reference), as well as any undiscovered human FcγR or FcγR isoform or allotype. FcγR can be from any organism, including, but not limited to, human, mouse, rat, rabbit, and monkey. Mouse FcγRs include, but are not limited to, FcγRI (CD64), FcγRII (CD32), FcγRIII (CD16), and FcγRIII-2 (CD16-2), as well as any undiscovered mouse FcγR or FcγR isoform or allotype.

[0094] As used herein, "FcRn" or "neonatal Fc receptor" refers to a protein that binds to the Fc region of an IgG antibody and is at least partially encoded by the FcRn gene. FcRn can be from any organism, including but not limited to human, mouse, rat, rabbit, and monkey. As known in the art, a functional FcRn protein comprises two polypeptides, often referred to as heavy and light chains. The light chain is beta-2-microglobulin, and the heavy chain is encoded by the FcRn gene. Unless otherwise stated herein, FcRn protein or FcRn protein refers to the complex of FcRn heavy chain and beta-2-microglobulin. Various FcRn variants can be used to increase binding to the FcRn receptor, and in some cases to increase serum half-life. An "FcRn variant" is one that increases binding to FcRn, and suitable FcRn variants are listed below.

[0095] A "parent polypeptide" as used herein refers to a starting polypeptide that is subsequently modified to generate a variant. A parent polypeptide may be a naturally occurring polypeptide or a variant or engineered version of a naturally occurring polypeptide. Thus, a "parent immunoglobulin" as used herein refers to an unmodified immunoglobulin polypeptide that is modified to produce a variant, and a "parent antibody" as used herein refers to an unmodified antibody that is modified to generate a variant antibody. Note that a "parent antibody" includes known commercially available recombinantly produced antibodies, as outlined below. In this context, a "parent Fc domain" is one that is relative to the recited variant, and thus a "variant human IgG1 Fc domain" is compared to the parent Fc domain of human IgG1, a "variant human IgG4 Fc domain" is compared to the parent Fc domain human IgG4, etc.

[0096] As used herein, "position" refers to a location in a sequence of a protein. Positions may be numbered sequentially or according to an established format, such as the EU index for antibody numbering.

[0097] "Target antigen" as used herein means a molecule that is specifically bound by an antibody binding domain comprising the variable region of a given antibody.

[0098] "Strandedness" in reference to the monomers of heterodimeric antibodies described herein means that the heterodimerization variants are incorporated into each monomer such that they retain the ability to "match" and form heterodimers, similar to the double-stranded DNA that they "match". For example, if several pI variants are engineered into monomer A (e.g., to increase the pI), a "charge-paired" steric variant that can be similarly utilized does not interfere with the pI variant, e.g., the pI-increased charge variant is placed in the same "strand" or "monomer", retaining both functionality. Similarly, for "skewed" variants that come in a set of pairs, as outlined in more detail below, one skilled in the art will take pI into account when deciding which strand or monomer one set of pairs will go into, so that pI separation is also maximized using the skewed pI.

[0099] As used herein, "target cell" means a cell that expresses a target antigen.

[0100] By "host cell" in the context of producing bispecific antibodies with the antibodies described herein is meant a cell that contains exogenous nucleic acid encoding the components of the bispecific antibody and is capable of expressing the bispecific antibody under suitable conditions. Suitable host cells are described below.

[0101] By "wild-type or WT" herein is meant an amino acid or nucleotide sequence found in nature, including allelic variations. A WT protein has an amino acid or nucleotide sequence that has not been intentionally modified.

[0102] Provided herein are several antigen-binding domains that have sequence identity to human antibody domains. Sequence identity between two similar sequences (e.g., antibody variable domains) can be determined using the methods of Smith, TF & Waterman, MS (1981) "Comparison Of Biosequences," Adv. Appl. Math. 2: 482 [local homology algorithm], Needleman, SB & Wunsch, CD. (1970) "A General Method Applicable To The Search For Similarities In The Amino Acid Sequence Of Two Proteins," J. Mol. Biol. 48: 443 [homology-alignment algorithm], Pearson, WR & Lipman, DJ (1988) "Improved Tools For Biological Sequence Comparison," Proc. Natl. Acad. Sci. (USA) 85: 2444 [similarity search method], or Altschul, SF et al, (1990) "Basic Local Alignment Search Tool," J. Mol. Biol. 215: 403-10 ["BLAST" algorithm (https: http: / / blast.ncbi.nlm.nih.gov / Blast.cgi) algorithm. When using any of the aforementioned algorithms, default parameters (window length, gap penalty, etc.) are used. In one embodiment, sequence identity is measured using the BLAST algorithm with default parameters.

[0103] The antibodies described herein are generally isolated or recombinant. "Isolated" as used to describe the various polypeptides disclosed herein means a polypeptide that has been identified, separated, and / or recovered from the cell or cell culture in which it is expressed. Ordinarily, an isolated polypeptide will be prepared by at least one purification step. "Isolated antibody" refers to an antibody that is substantially free of other antibodies having different antigen specificities. "Recombinant" means that the antibodies are produced using recombinant nucleic acid technology in an exogenous host cell, and may also be isolated.

[0104] "Specific binding" to a particular antigen or epitope, or "specifically binds to" or "specific for" a particular antigen or epitope, refers to binding that is measurably different from non-specific interactions. Specific binding can be measured, for example, by determining the binding of a molecule compared to the binding of a control molecule, which is generally a molecule of similar structure that has no binding activity. For example, specific binding can be determined by competition with a control molecule that is similar to the target.

[0105] Specific binding to a particular antigen or epitope is, for example, at least about 10 -4 M, at least about 10 -5 M, at least about 10 -6 M, at least about 10 -7 M, at least about 10 -8 M, at least about 10 -9 M, alternatively at least about 10 -10 M, at least about 10 -11 M, at least about 10 -12 A specific binding characteristic can be exhibited by an antibody having a KD of 20, 50, 100, 500, 1000, 5,000, 10,000, or more times that of a control molecule, where KD refers to the off-rate of a particular antibody-antigen interaction. Typically, an antibody that specifically binds to an antigen will have a KD of 20, 50, 100, 500, 1000, 5,000, 10,000, or more times that of a control molecule, relative to the antigen or epitope.

[0106] Specific binding to a particular antigen or epitope may also be exhibited, for example, by an antibody having a K or K for the antigen or epitope that is at least 20-fold, 50-fold, 100-fold, 500-fold, 1000-fold, 5,000-fold, 10,000-fold, or more times higher than a control, where K or K refers to the association rate of a particular antibody-antigen interaction. Binding affinity is typically measured using Biacore, SPR, or BLI assays.

[0107] D. Antibodies of the Invention The present invention provides antibodies, including monoclonal and bispecific antibodies, that bind to human GPC3 (note that to facilitate preclinical testing, many if not most of the exemplified antibodies also bind to cynomolgus monkey GPC3, although this is not required in all embodiments). In particular, bispecific antibodies are provided that bind to CD3 and GPC3 and take a variety of formats, as described more fully below.

[0108] 1. Antibodies The antibodies provided herein contain different antibody domains, as described more fully below. As described herein and known in the art, the antibodies described herein contain different domains within the heavy and light chains, which may also overlap. These domains include, but are not limited to, Fc domain, CH1 domain, CH2 domain, CH3 domain, hinge domain, heavy chain constant domain (CH1-hinge-Fc domain or CH1-hinge-CH2-CH3), variable heavy chain domain, variable light chain domain, light chain constant domain, Fab domain, and scFv domain.

[0109] In particular, the format shown in Figures 15A and 15B is an antibody that is commonly referred to as a "heterodimeric antibody," meaning that the protein has at least two related Fc sequences self-assembled into a heterodimeric Fc domain and at least two Fv regions, whether as a Fab or as an scFv.

[0110] a. Chimeric and Humanized Antibodies In certain embodiments, the antibodies described herein comprise a heavy chain variable region derived from a particular germline heavy chain immunoglobulin gene and / or a light chain variable region derived from a particular germline light chain immunoglobulin gene. For example, such antibodies may comprise or consist of a human antibody comprising a heavy or light chain variable region that is the "product of" or "derived from" a particular germline sequence. A human antibody that is the "product of" or "derived from" a human germline immunoglobulin sequence can be identified as such (using the methods outlined herein) by comparing the amino acid sequence of the human antibody to the amino acid sequences of human germline immunoglobulins and selecting the human germline immunoglobulin that is closest in sequence to the human antibody (i.e., highest percent identity). A human antibody that is the "product of" or "derived from" a particular human germline immunoglobulin sequence can contain amino acid differences compared to the germline sequence, for example, due to naturally occurring somatic mutations or deliberate introduction of site-specific mutations. However, a humanized antibody is typically at least 90% identical in amino acid sequence to an amino acid sequence encoded by a human germline immunoglobulin gene and contains amino acid residues that identify the antibody as derived from a human sequence when compared to germline immunoglobulin amino acid sequences of other species (e.g., mouse germline sequences). In certain cases, a humanized antibody may be at least 95, 96, 97, 98, or 99% identical in amino acid sequence to an amino acid sequence encoded by a germline immunoglobulin gene, or even at least 96%, 97%, 98%, or 99% identical in amino acid sequence to an amino acid sequence encoded by a germline immunoglobulin gene. Typically, a humanized antibody derived from a particular human germline sequence will exhibit no more than 10-20 amino acid differences from the amino acid sequence encoded by a human germline immunoglobulin gene (prior to the introduction of any skew, pI, and deletion variants described herein, i.e., the number of variants is generally small prior to the introduction of the variants described herein).In certain cases, a humanized antibody may display no more than 5 amino acids, or even no more than 4, 3, 2, or 1 amino acid difference from the amino acid sequence encoded by a germline immunoglobulin gene (again, prior to the introduction of any skew, pI, and deletion variants described herein, i.e., prior to the introduction of the variants described herein, the number of variants is generally small).

[0111] In one embodiment, the parent antibody has been affinity matured as known in the art. Structure-based methods can be used for humanization and affinity maturation, for example, as described in U.S. Patent Application No. 11 / 004,590. Selection-based methods can be used to humanize and / or affinity mature antibody variable regions, including, but not limited to, those described in Wu et al., 1999, J. Mol. Biol. 294:151-162; Baca et al., 1997, J. Biol. Chem. 272(16):10678-10684; Rosok et al., 1996, J. Biol. Chem. 271(37):22611-22618; Rader et al., 1998, Proc. Natl. Acad. Sci. USA 95:8910-8915; Krauss et al., 2003, Protein Engineering 16(10): 753-759, all of which are incorporated by reference. Other humanization methods may involve grafting only portions of the CDRs, including, but not limited to, those described in U.S. Patent Application Serial No. 09 / 810,510; Tan et al., 2002, J. Immunol. 169:1119-1125; De Pascalis et al., 2002, J. Immunol. 169:3076-3084, the entire contents of which are incorporated by reference.

[0112] 2. Heterodimeric antibodies In an exemplary embodiment, the bispecific antibodies provided herein are heterodimeric bispecific antibodies comprising two variant Fc domain sequences, such variant Fc domains comprising amino acid modifications to facilitate self-assembly and / or purification of the heterodimeric antibody.

[0113] A continuing problem in antibody technology is the desire for "bispecific" antibodies that bind two different antigens simultaneously, generally bringing the different antigens into close proximity and providing new functions and new therapeutic approaches. Generally, these antibodies are made by including genes for each heavy and light chain in a host cell. This generally results in the formation of two homodimers (AA and BB (not including the issue of light chain heterodimers)) as well as the desired heterodimer (AB). However, a major obstacle in bispecific antibody formation is the difficulty of biasing the desired heterodimer formation over homodimer formation and / or purifying the heterodimeric antibody away from homodimers.

[0114] There are many mechanisms that can be used to generate the heterodimeric antibody of interest. In addition, as will be appreciated by those skilled in the art, these different mechanisms can be combined to ensure high heterodimerization. Amino acid modifications that facilitate the production and purification of heterodimers are generally collectively referred to as "heterodimerization variants." As discussed below, heterodimerization variants include "skew" variants (e.g., "knob and hole" and "charge pair" variants, as described below) as well as "pI variants," which allow purification of heterodimers away from homodimers. As generally described in U.S. Pat. No. 9,605,084, which is incorporated by reference in its entirety, and specifically as follows for the discussion of heterodimerization variants, mechanisms useful for heterodimerization include "knobs and holes" ("KIH") as described in U.S. Pat. No. 9,605,084, "electrostatic steering" or "charge pairs" as described in U.S. Pat. No. 9,605,084, the pI variants as described in U.S. Pat. No. 9,605,084, and further general Fc variants as outlined in U.S. Pat. No. 9,605,084 and below.

[0115] Heterodimerization variants useful for the formation and purification of the subject heterodimeric antibodies (e.g., bispecific antibodies) are discussed in further detail below.

[0116] a. Scuba liant In some embodiments, the heterodimeric antibody comprises a scuba wardant, which is one or more amino acid modifications in the first Fc domain (A) and / or the second Fc domain (B), that promote the formation of Fc homodimers (Fc dimers comprising the first and second Fc domains; AB) over Fc homodimers (Fc dimers comprising two of the first Fc domains or two of the second Fc domains; AA or BB). Suitable scuba wardants are generally included in Figure 29 of US Patent Publication No. 2016 / 0355608, which is incorporated herein by reference for its disclosure of scuba wardant in particular, as well as in Figure 1.

[0117] Thus, suitable pairs of Fc heterodimerization variants that allow for the formation of heterodimeric Fc regions are shown in Figure 1. Thus, a first Fc domain has a first Fc heterodimerization variant and a second Fc domain has a second Fc heterodimerization variant selected from the pairs in Figure 1.

[0118] One mechanism may be optionally used, commonly referred to in the art as "knobs and holes," which refers to amino acid manipulations that result in steric effects that favor heterodimer formation and not homodimer formation. This may be referred to as "knobs and holes," as described in U.S. Patent Application Serial No. 61 / 596,846; Ridgway et al., Protein Engineering 9(7):617 (1996); Atwell et al., J. Mol. Biol. 1997 270:26; and U.S. Patent No. 8,216,805, all of which are incorporated herein by reference in their entireties. These figures identify a number of "monomer A-monomer B" pairs that rely on "knobs and holes." Additionally, as described in Merchant et al., Nature Biotech. 16:677 (1998), these "knobs and holes" mutations can be combined with disulfide bonds to skew formation toward heterodimerization.

[0119] A further mechanism that finds use in generating heterodimers may be referred to as "electrostatic steering" as described in Gunasekaran et al., J. Biol. Chem. 285(25):19637 (2010), which is incorporated herein by reference in its entirety. This may also be referred to herein as "charge pairing." In this embodiment, electrostatics are used to skew the formation toward heterodimerization. As will be appreciated by those skilled in the art, these may also affect pI, i.e. purification, and therefore may be considered pI variants in some cases. However, because they were generated to force heterodimerization and were not used as a purification tool, they are classified as "steric variants." These include, but are not limited to, D221E / P228E / L368E paired with D221R / P228R / K409R (e.g., these are a "matched set of monomers"), and C220E / P228E / 368E paired with C220R / E224R / P228R / K409R.

[0120] In some embodiments, the scuba variants advantageously and simultaneously promote heterodimerization based on both the "knob and hole" mechanism and the "electrostatic steering" mechanism. In some embodiments, the heterodimeric antibody comprises one or more sets of such heterodimerizing scuba variants. These variants are "pairs" of "sets". That is, pairs of one set are incorporated into the first monomer and pairs of the other set are incorporated into the second monomer. It should be noted that these sets do not necessarily behave as "knob and hole" variants, but have a one-to-one correspondence between residues of one monomer and residues of the other monomer. That is, these pairs of sets may instead form an interface between the two monomers that promotes the formation of heterodimers and not homodimers, and the percentage of heterodimers that form spontaneously under biological conditions is more than 90% instead of the expected 50% (25% homodimer A / A: 50% heterodimer A / B: 25% homodimer B / B). Exemplary heterodimerization "skew" variants are shown in FIG. In exemplary embodiments, the heterodimeric antibody comprises an Fc heterodimerization variant as the set S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K / E357L, K370S:S364K / E357Q, or T366S / L368A / Y407V:T366W (optionally including a bridging disulfide, T366S / L368A / Y407V / Y349C:T366W / S354C), all of which are "skewed" variant amino acid substitution sets of Fc heterodimerization variants. In an exemplary embodiment, the heterodimeric antibody comprises the "S364K / E357Q:L368D / K370S" amino acid substitution set. In terms of nomenclature, the pair "S364K / E357Q:L368D / K370S" means that one of the monomers comprises an Fc domain that comprises the amino acid substitutions S364K and E357Q, and the other monomer comprises an Fc domain that comprises the amino acid substitutions L368D and K370S. As mentioned above, the "twistiness" of these pairs depends on the starting pI.

[0121] In some embodiments, the scubariants provided herein are optionally and independently incorporated into one or both of the first and second Fc domains of a heterodimeric antibody along with any other modifications, including but not limited to other scubariants (see, e.g., FIG. 37 of US Patent Application Publication No. 2012 / 0149876, particularly those incorporated herein by reference for their disclosure of scubariants), pI variants, isotype variants, FcRn variants, deletion variants, etc. Furthermore, individual modifications can also be independently and optionally included or excluded from the heterodimeric antibody.

[0122] The additional monomer A and monomer B variants can be optionally and independently combined in any amount with other variants, such as the pI variants outlined herein or other conformational variants shown in Figure 37 of US Patent Application Publication No. 2012 / 0149876, the figures, legends and sequence numbers of which are expressly incorporated herein by reference.

[0123] In some embodiments, the conformational mutations outlined herein can optionally and independently incorporate any pI mutations (or other mutations such as Fc mutations, FcRn mutations, etc.) into one or both monomers and can independently optionally be included or excluded from the antibody proteins described herein.

[0124] A list of suitable scuba variants showing some pairs that are particularly useful in many embodiments is shown in Figure 1. Sets of pairs that are particularly useful in many embodiments include, but are not limited to, S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K / E357L, and 370S:S364K / E357Q. In terms of nomenclature, the pair "S364K / E357Q:L368D / K370S" means that one monomer has the double variant set S364K / E357Q and the other has the double variant set L368D / K370S.

[0125] b. pI (isoelectric point) variants of heterodimers In some embodiments, the heterodimeric antibody comprises a purification variant that advantageously allows separation of the heterodimeric antibody (eg, an anti-GPC3 x anti-CD3 bispecific antibody) from homodimeric proteins.

[0126] There are several basic mechanisms that can facilitate the purification of heterodimeric antibodies. For example, modification of one or both of the antibody heavy chain monomers A and B such that each monomer has a different pI allows for isoelectric purification of the heterodimeric AB antibody from the monomeric AA and BB proteins. Alternatively, some scaffold formats, such as the "1+1 Fab-scFv-Fc" and "2+1 Fab2-scFv-Fc" formats, also allow for separation based on size. As mentioned above, it is also possible to "skew" the formation of heterodimers over homodimers using scuba variants. Thus, the combination of heterodimerizing scuba variants and pi variants finds particular use in the heterodimeric antibodies provided herein.

[0127] Additionally, depending on the format of the heterodimeric antibody, pi variants contained within the monomeric constant regions and / or Fc domains and / or domain linkers can be used, as outlined more fully below. In some embodiments, the heterodimeric antibody contains additional modifications for alternative functions, such as Fc, FcRn and KO variants, that can also create pi alterations.

[0128] In some embodiments, the subject heterodimeric antibodies provided herein include at least one monomer having one or more modifications that alter the pI of the monomer (i.e., a "pI variant"). In general, as will be appreciated by those of skill in the art, there are two general categories of pI variants: those that increase the pI of a protein (basic changes) and those that decrease the pI of a protein (acidic changes). As described herein, all combinations of these variants can be made, where one monomer can be wild type or a variant that does not exhibit a pI that is significantly different from wild type, and the other can be either more basic or more acidic. Alternatively, each monomer is altered, one to be more basic and one to be more acidic.

[0129] Depending on the format of the heterodimeric antibody, the pI variants may be included within the constant domains and / or Fc domains of the monomers, or either charged linkers, domain linkers or scFv linkers may be used. That is, antibody formats utilizing scFvs, such as "1+1 Fab-scFv-Fc", may include a charged scFv linker (either positive or negative) that provides an additional pI boost for purification purposes. As will be appreciated by those skilled in the art, the antibodies described herein also provide pI variants in one or both of the monomers, and / or charged domain linkers, although some 1+1 Fab-scFv-Fc formats are useful with only a charged scFv linker, without additional pI tuning. In addition, further amino acid engineering for alternative functionality may also impart pI changes, such as Fc, FcRn, and KO variants.

[0130] In subject heterodimeric antibodies that utilize pI as a separation mechanism to allow purification of the heterodimeric protein, amino acid variants are introduced into one or both of the monomeric polypeptides. That is, the pI of one of the monomers (referred to herein for simplicity as "monomer A") can be engineered away from monomer B, or the pI of both monomers A and B can be altered, increasing the pI of monomer A and decreasing the pI of monomer B. As outlined more fully below, pI changes in either or both monomers can be made by removing or adding a charged residue (e.g., substituting a neutral amino acid with a positive or negative amino acid residue, e.g., glycine to glutamic acid), by changing a charged residue from positive or negative to the opposite charge (aspartic acid to lysine), or by changing a charged residue to a neutral residue (e.g., loss of charge, lysine to serine). Some of these variants are shown in Figures 1 and 2.

[0131] Thus, in some embodiments, the subject heterodimeric antibodies contain amino acid modifications in the constant region that alter the isoelectric point (pI) of at least one, if not both, of the monomers of the dimeric protein by incorporating an amino acid substitution (a "pI variant" or "pI substitution") into one or both of the monomers. As demonstrated herein, separation of the heterodimer from the two homodimers can be achieved when the pI of the two monomers differs by as little as 0.1 pH units, with 0.2, 0.3, 0.4, and 0.5 or more differing being all used in the antibodies described herein.

[0132] As will be appreciated by those of skill in the art, the number of pI mutations to be included in each or both monomers to obtain good separation will depend in part on the starting pI of the components, e.g., the starting pI of the scFv and Fab of interest in 1+1 Fab-scFv-Fc and 2+1 Fab2-scFv-Fc formats. That is, to determine which monomers to engineer or in which "direction" (e.g., more positive or more negative), the Fv sequences of the two target antigens are calculated and a decision made therefrom. As is known in the art, different Fvs will have different starting pIs utilized in the antibodies described herein. Generally, as outlined herein, pIs are engineered to result in a total pI difference of each monomer of at least about 0.1 log, with 0.2-0.5 being preferred as outlined herein.

[0133] The use of constant regions of the heavy chains provides a more modular approach to designing and purifying bispecific proteins, including antibodies, when pI variants are used to achieve heterodimerization. Thus, in some embodiments, heterodimerization variants (including skew and pI heterodimerization variants) are not included in the variable region, and therefore each individual antibody must be engineered. In addition, in some embodiments, the potential for immunogenicity resulting from pI variants is significantly reduced by importing pI variants from different IgG isotypes, such that the pI is changed without introducing significant immunogenicity. Thus, a further problem to be solved is the elucidation of low pI constant domains with high human sequence content, e.g., minimizing or avoiding non-human residues at any particular position. Instead of or in addition to isotype substitutions, the potential for immunogenicity resulting from pI variants is significantly reduced by utilizing isosteric substitutions (e.g., Asn to Asp and Gln to Glu).

[0134] As discussed below, potential collateral benefits of this pi engineering are also increased serum half-life and increased FcRn ligation. That is, as described in US Patent Application Publication No. 2012 / 0028304 (hereby incorporated by reference in its entirety), lowering the pi of antibody constant domains (including those found in antibodies and Fc fusions) can result in longer serum retention in vivo. These pi variants to extend serum half-life also facilitate pi changes for purification.

[0135] In addition, the pI variants provide additional benefits to the analysis and quality control process of bispecific antibodies, as the ability to eliminate, minimize, and distinguish when homodimers are present is significant. Similarly, the ability to reliably test the reproducibility of heterodimeric antibody production is important.

[0136] Generally, the particular embodiment used relies on a set of variants including scubariants, which in combination with pI variants that increase the pI difference between the two monomers, favor heterodimer formation in preference to homodimer formation, and facilitate purification of the heterodimer by removal of the homodimer.

[0137] Exemplary combinations of pI variants are shown in Figures 1 and 2, and in Figure 30 of US Patent Application Publication No. 2016 / 0355608, all of which are incorporated herein by reference in their entirety, and in particular for their disclosure of pI variants. Preferred combinations of pI variants are shown in Figures 1 and 2. As outlined herein and shown in the figures, these changes are shown relative to IgG1, but all isotypes can be modified in this manner, as can isotype hybrids. R133E and R133Q can also be used when the heavy chain constant domain is derived from IgG2-4.

[0138] Thus, in some embodiments, one monomer has a set of substitutions from Figure 2 and the other monomer has a charged linker (which can be selected from those shown in Figures 5 and 6, either in the format of an scFv or a charged scFv linker as that monomer comprises a charged domain linker as shown in the format).

[0139] In some embodiments, the modifications occur in the hinge of the Fc domain, including positions 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229 and 230, based on EU numbering. Thus, pI mutations, particularly substitutions, can be made at one or more of positions 216-230, with 1, 2, 3, 4 or 5 mutations finding use. Similarly, all possible combinations are contemplated, alone or together with other pI variants in other domains.

[0140] Particular substitutions used to reduce the pI of the hinge domain include, but are not limited to, a deletion at position 221, a non-native valine or threonine at position 222, a deletion at position 223, a non-native glutamic acid at position 224, a deletion at position 225, a deletion at position 235, and a deletion or non-native alanine at position 236. In some cases, only pI substitutions are made in the hinge domain, while in other instances these substitutions are added in any combination to other pI variants in other domains.

[0141] In some embodiments, mutations can be made in the CH2 region including positions 233, 234, 235, 236, 274, 296, 300, 309, 320, 322, 326, 327, 334 and 339 based on EU numbering. Note that changes at 233-236 can be made to increase effector function (along with 327A) in an IgG2 backbone. Similarly, all possible combinations of these 14 positions can be made, e.g., = can include variant Fc domains with 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10 CH2 pI substitutions.

[0142] Specific substitutions used to reduce the pI of the CH2 domain include, but are not limited to, a non-native glutamine or glutamic acid at position 274, a non-native phenylalanine at position 296, a non-native phenylalanine at position 300, a non-native valine at position 309, a non-native glutamic acid at position 320, a non-native glutamic acid at position 322, a non-native glutamic acid at position 326, a non-native glycine at position 327, a native glutamic acid at position 334, a non-native threonine at position 339, and all possible combinations within CH2 and with other domains.

[0143] In this embodiment, modifications may be independently and optionally selected from positions 355, 359, 362, 384, 389, 392, 397, 418, 419, 444, and 447 (EU numbering) of the CH3 region. Specific substitutions used to reduce the pI of the CH3 domain include, but are not limited to, a non-native glutamine or glutamic acid at position 355, a non-native serine at position 384, a non-native asparagine or glutamic acid at position 392, a non-native methionine at position 397, a non-native glutamic acid at position 419, a non-native glutamic acid at position 359, a non-native glutamic acid at position 362, a non-native glutamic acid at position 389, a non-native glutamic acid at position 418, a non-native glutamic acid at position 444, and a deletion or non-native aspartic acid at position 447.

[0144] In general, as will be appreciated by those of skill in the art, there are two general categories of pI variants: those that increase the pI of a protein (basic changes) and those that decrease the pI of a protein (acidic changes). As described herein, all combinations of these variants can be made, with one monomer being wild type or a variant that does not exhibit a pI that differs significantly from wild type, and the other being either more basic or more acidic. Alternatively, each monomer is altered, one to be more basic and one to be more acidic.

[0145] A preferred combination of pI variants is shown in Figure 2. As outlined herein and shown in the figures, these changes are shown relative to IgG1, however, all isotypes can be modified in this manner, as can isotype hybrids. R133E and R133Q can also be used when the heavy chain constant domain is derived from IgG2-4.

[0146] In one embodiment, for example in the format of Figures 15A and 44, a preferred combination of pI variants has one monomer (negative Fab side) comprising the 208D / 295E / 384D / 418E / 421D variants (N208D / Q295E / N384D / Q418E / N421D for human IgG1) and a second monomer (positive scFv side) comprising a positively charged scFv linker comprising (GKPGS)4 (SEQ ID NO:15). However, as will be appreciated by those skilled in the art, the first monomer comprises a CH1 domain comprising position 208. Thus, in constructs that do not contain a CH1 domain (e.g. in the case of heterodimeric antibodies that do not utilize a CH1 domain in one of the domains, e.g. in the case of dual scFv formats or "one-arm" formats as shown in Figure 44C or Figure 44D), a preferred negative pI variant Fc set comprises the 295E / 384D / 418E / 421D variant (for human IgG1, Q295E / N384D / Q418E / N421D).

[0147] Thus, in some embodiments, one monomer has a set of substitutions from Figure 4 and the other monomer has a charged linker (which can be selected from those shown in Figures 5 and 6, either in the format of an scFv or a charged scFv linker as that monomer comprises a charged domain linker as shown in the format).

[0148] C isotype variants Furthermore, many embodiments of the antibodies described herein rely on the "import" of pI amino acids at specific positions from one IgG isotype to another, thus reducing or eliminating the possibility of introducing undesirable immunogenicity into the variant. Some of these are shown in FIG. 21 of US Patent Application Publication No. 2014 / 0370013, incorporated herein by reference. That is, IgG1 is a common isotype for therapeutic antibodies for a variety of reasons, including high effector function. However, the heavy constant region of IgG1 has a higher pI than that of IgG2 (8.10 vs. 7.31). By introducing IgG2 residues into the IgG1 backbone at specific positions, the pI of the resulting monomer is lowered (or increased) and also exhibits a longer serum half-life. For example, IgG1 has glycine (pI 5.97) at position 137 and IgG2 has glutamic acid (pI 3.22), and importing glutamic acid affects the pI of the resulting protein. As described below, several amino acid substitutions are generally required to significantly affect the pI of the variant antibody, however, it should be noted that even changes in the IgG2 molecule can allow for increased serum half-life, as discussed below.

[0149] In other embodiments, non-isotypic amino acid changes are made to reduce the overall charge state of the resulting protein (e.g., by changing from a high pI amino acid to a low pI amino acid) or to allow for tuning of the structure, such as for stability, as described in more detail below.

[0150] In addition, by pI engineering both the heavy and light chain constant domains, significant changes can be seen in each monomer of the heterodimer. As discussed herein, the pI of the two monomers differs by at least 0.5, which may allow for separation by ion exchange chromatography or isoelectric focusing, or other methods sensitive to isoelectric point.

[0151] d. Calculation of pI The pI of each monomer depends on the pI of the variant heavy chain constant domain and the pI of the entire monomer, which may include the variant heavy chain constant domain and the fusion partner. Thus, in some embodiments, the change in pI is calculated based on the variant heavy chain constant domain using the chart in Figure 19 of US Patent Application Publication No. 2014 / 0370013. As discussed herein, which monomers to engineer is generally determined by the inherent pI of the Fv region and the scaffold region. Alternatively, the pI of each monomer can be compared.

[0152] e. pI variants that also confer better FcRn in vivo binding If the pI variants reduce the pI of the monomer, they may have the added advantage of improving serum retention in vivo.

[0153] Although still under investigation, it is believed that Fc regions have a longer half-life in vivo because binding to FcRn at pH 6 in endosomes sequesters Fc (Ghetie and Ward, 1997 Immunol Today. 18(12):592-598, incorporated by reference in its entirety). The endosomal compartment then recycles Fc to the cell surface. Once the compartment opens to the extracellular space, a higher pH of about 7.4 induces the release of Fc into the blood. In mice, Dall'Acqua et al. showed that Fc mutants with increased FcRn binding at pH 6 and pH 7.4 indeed have the same serum concentration and half-life as wild-type Fc (Dall'Acqua et al., 2002, J. Immunol. 169:5171-5180, incorporated by reference in its entirety). It is believed that the increased affinity of Fc for FcRn at pH 7.4 prevents the release of Fc into the blood. Therefore, Fc mutations that increase the half-life of Fc in vivo would ideally increase FcRn binding at lower pH while still allowing release of Fc at higher pH. The amino acid histidine changes its charge state in the pH range of 6.0 to 7.4. It is therefore not surprising to find His residues at key positions in the Fc / FcRn complex.

[0154] Recently, it has been suggested that antibodies with variable regions with lower isoelectric points may also have longer serum half-lives (Igawa et al., 2010 PEDS.23(5):385-392, incorporated by reference in its entirety). However, this mechanism is still poorly understood. Furthermore, variable regions vary from antibody to antibody. As described herein, constant region variants with reduced pI and extended half-life would provide a more modular approach to improving the pharmacokinetic properties of antibodies.

[0155] F. Additional Fc variants for additional functionality In addition to the heterodimerization variants outlined above, there are a number of useful Fc amino acid modifications that can be made for a variety of reasons, including but not limited to, altering binding to one or more FcγR receptors, altered binding to FcRn, as outlined below.

[0156] Thus, the antibodies provided herein (heterodimers, as well as homodimers) can include such amino acid modifications with or without the heterodimerization variants (e.g., pI variants and conformational variants) outlined herein. Each set of variants can independently and optionally include or exclude certain heterodimeric proteins.

[0157] (i) FcγR variants Thus, there are several useful Fc substitutions that can be made to modify binding to one or more of the FcγR receptors. In certain embodiments, the subject antibody comprises a modification that alters binding to one or more FcγR receptors (i.e., "FcγR variants"). Substitutions that result in increased binding as well as decreased binding can be useful. For example, increased binding to FcγRIIIa is generally known to result in ADCC (antibody-dependent cell-mediated cytotoxicity, a cell-mediated reaction in which non-specific cytotoxic cells expressing FcγR recognize bound antibodies on target cells and subsequently cause lysis of the target cells). Similarly, decreased binding to FcγRIIb (an inhibitory receptor) can also be beneficial under some circumstances. Amino acid substitutions used in the antibodies described herein include those listed in U.S. Patent Nos. 8,188,321 (particularly FIG. 41) and 8,084,582, and U.S. Patent Application Publication Nos. 20060235208 and 20070148170, all of which are expressly incorporated herein by reference in their entirety, and in particular with respect to the variants disclosed therein. Particular variants used include, but are not limited to, 236A, 239D, 239E, 332E, 332D, 239D / 332E, 267D, 267E, 328F, 267E / 328F, 236A / 332E, 239D / 332E / 330Y, 239D / 332E / 330L, 243A, 243L, 264A, 264V, and 299T.

[0158] Furthermore, as specifically disclosed in U.S. Patent Application Serial No. 12 / 341,769, which is incorporated herein by reference in its entirety, there are additional Fc substitutions that are used to increase binding to the FcRn receptor and increase serum half-life, including, but not limited to, 434S, 434A, 428L, 308F, 259I, 428L / 434S, 428L / 434A, 259I / 308F, 436I / 428L, 436I or V / 434S, 436V / 428L, and 259I / 308F / 428L. Such modifications may be included in one or both Fc domains of the subject antibodies.

[0159] (ii) Deletion variants Similarly, another category of functional variants are "FcγR ablated variants" or "Fc knockout (FcKO or KO)" variants. In these embodiments, for some therapeutic applications, it is desirable to reduce or eliminate normal binding of the Fc domain to one or more or all Fcγ receptors (e.g., FcγR1, FcγRIIa, FcγRIIb, FcγRIIIa, etc.) to avoid additional mechanisms of action. That is, for example, in many embodiments, particularly in the use of bispecific antibodies that bind monovalently to CD3, it is generally desirable to ablate FcγRIIIa binding to eliminate or significantly reduce ADCC activity, and one of the Fc domains comprises one or more Fcγ receptor ablated variants. These deletion variants are shown in Figure 14, and each can be independently and optionally included or excluded in preferred embodiments with a deletion variant selected from the group consisting of G236R / L328R, E233P / L234V / L235A / G236del / S239K, E233P / L234V / L235A / G236del / S267K, E233P / L234V / L235A / G236del / S239K / A327G, E233P / L234V / L235A / G236del / S267K / A327G, and E233P / L234V / L235A / G236del. Note that the deletion variants referred to herein eliminate FcγR binding, but generally do not eliminate FcRn binding.

[0160] As is known in the art, the Fc domain of human IgG1 has the highest binding to Fcγ receptors, therefore, when the constant domain (or Fc domain) of the heterodimeric antibody scaffold is IgG1, deletion variants can be used. Alternatively, or in addition to deletion variants of the IgG1 background, a mutation at glycosylation position 297 (generally to A or S) can, for example, significantly eliminate binding to FcγRIIIa. Human IgG2 and IgG4 have naturally reduced binding to Fcγ receptors, so these scaffolds can be used with or without deletion variants.

[0161] E. Combinations of Heterodimers and Fc Variants As will be appreciated by those of skill in the art, all of the listed heterodimerization variants (including scubariants and / or pI variants) can be optionally and independently combined as long as they maintain their "twistiness" or "monomer split". In some embodiments, the heterodimeric antibodies provided herein comprise a combination of heterodimerization scubariants, isosteric pI substitutions, and FcKO variants, as shown in Figure 4. Furthermore, all of these variants can be combined in any of the heterodimerization formats.

[0162] In the case of pI variants, while embodiments that find particular use are shown in the figures, other combinations can be generated following the basic rule of altering the pI difference between the two monomers to facilitate purification.

[0163] Additionally, any of the heterodimerization variants, skews, and pIs can be independently and optionally combined with Fc ablation variants, Fc variants, FcRn variants, as generally outlined herein.

[0164] Exemplary combinations of variants included in some embodiments of heterodimeric 1+1 Fab-scFv-Fc and 2+1 Fab2-scFv-Fc format antibodies are included in Figure 4. In certain embodiments, the antibody is a heterodimeric 1+1 Fab-scFv-Fc or 2+1 Fab2-scFv-Fc format antibody, as shown in Figures 15A and 15B.

[0165] F. Anti-GPC3 x anti-CD3 bispecific antibody In another aspect, provided herein is a bispecific antibody, anti-GPC3×anti-CD3 (also referred to herein as "αGPC3×αCD3" or simply "GPC3×CD3"). Such an antibody comprises at least one GPC3 binding domain and at least one CD3 binding domain. In some embodiments, the bispecific αGPC3×αCD3 provided herein selectively an immune response at tumor sites expressing GPC3.

[0166] Please note that unless specified herein, the order of the antigens listed in the name does not confer structure, i.e., a GPC3xCD3 1+1 Fab-scFv-Fc antibody may bind scFv to GPC3 or CD3, but in some cases the order dictates structure as indicated.

[0167] As outlined in more detail herein, these combinations of ABDs can be in a variety of formats, generally with one ABD being in Fab format and the other in scFv format, as outlined below. Exemplary formats for use in the bispecific antibodies provided herein include 1+1 Fab-scFv-Fc and 2+1 Fab2-scFv-Fv formats (see, e.g., Figures 15A and 15B). Other useful antibody formats include, but are not limited to, antibodies in the "mAb-Fv", "mAb-scFv", "central Fv", "one-arm scFv-mAb", "scFv-mAb", "dual scFv", and "trident" formats, as shown in Figure 44 and described more fully below.

[0168] Moreover, typically one of the ABDs comprises an scFv as outlined herein in an N- to C-terminal orientation of VH-scFv linker-VL or VL-scFv linker-VH, depending on the format, one or both of the other ABDs is typically a Fab comprising a VH domain (typically as a component of a heavy chain) on one protein chain and a VL (typically as a component of a light chain) on another protein chain.

[0169] As will be appreciated by those of skill in the art, any set of six CDRs or VH and VL domains can be made into scFv or Fab format, which is then added to heavy and light chain constant domains, where the heavy chain constant domain contains mutations (including within the CH1 and Fc domains). The scFv sequences contained in the sequence listing utilize specific charged linkers, however, uncharged or other charged linkers can be used, including those shown in Figures 5 and 6, as outlined herein.

[0170] In addition, as noted above, the numbering used in the sequence listing for the identification of the CDRs is Kabat, however, different numbering can be used, which will alter the amino acid sequence of the CDRs as shown in Table 2.

[0171] Further variants can be made for all variable heavy and light chain domains described herein. As outlined herein, in some embodiments, the set of six CDRs can have 0, 1, 2, 3, 4 or 5 amino acid modifications (with amino acid substitutions finding particular use), as well as variations in the framework regions of the variable heavy and light chain domains, as long as the framework (excluding the CDRs) retains at least about 80, 85 or 90% identity with a human germline sequence selected from those listed in Figure 1 of US Patent No. 7,657,380 (the figures and legends of which are incorporated herein by reference in their entirety). Thus, for example, the same CDRs described herein can be combined with different framework sequences derived from human germline sequences, as long as the framework regions retain at least 80, 85 or 90% identity with a human germline sequence selected from those listed in Figure 1 of US Patent No. 7,657,380. Alternatively, the CDRs may have amino acid modifications (e.g., 1, 2, 3, 4 or 5 amino acid modifications in the set of CDRs (i.e., the CDRs may be modified as long as the total number of changes in the set of 6 CDRs is less than 6 amino acid modifications, any combination of CDRs may be altered, e.g., there may be one change in vlCDR1, two changes in vhCDR2, no changes in vhCDR3, etc.)) and may have variations in the framework regions as long as the framework regions retain at least 80, 85, or 90% identity to a human germline sequence selected from those listed in Figure 1 of U.S. Pat. No. 7,657,380.

[0172] As discussed herein, the subject heterodimeric antibodies comprise two antigen binding domains (ABDs), each of which binds to GPC3 or CD3. As outlined herein, these heterodimeric antibodies can be bispecific and bivalent (each antigen is bound by a single ABD, e.g., in the format shown in FIG. 15A), or bispecific and trivalent (e.g., one antigen is bound by a single ABD and the other by two ABDs, as shown in FIG. 15B).

[0173] Moreover, typically one of the ABDs comprises an scFv as outlined herein in an N- to C-terminal orientation of VH-scFv linker-VL or VL-scFv linker-VH, depending on the format, one or both of the other ABDs is typically a Fab comprising a VH domain (typically as a component of a heavy chain) on one protein chain and a VL (typically as a component of a light chain) on another protein chain.

[0174] The present disclosure provides several ABDs as outlined below. As will be appreciated by those skilled in the art, any set of six CDRs or VH and VL domains can be in scFv or Fab format, which is then added to heavy and light chain constant domains, the heavy chain constant domains including variants (including within the CH1 and Fc domains). The scFv sequences included in the sequence listing utilize a particular charged linker, however, uncharged or other charged linkers can be used, including those shown in Figure 5, as outlined herein.

[0175] In addition, as noted above, the numbering used in the sequence listing for the identification of the CDRs is Kabat, however, different numbering can be used, which will alter the amino acid sequence of the CDRs as shown in Table 2.

[0176] Further variants can be made for all variable heavy and light chain domains described herein. As outlined herein, in some embodiments, the set of six CDRs can have 0, 1, 2, 3, 4 or 5 amino acid modifications (with amino acid substitutions finding particular use), as well as variations in the framework regions of the variable heavy and light chain domains, as long as the framework (excluding the CDRs) retains at least about 80, 85 or 90% identity with a human germline sequence selected from those listed in Figure 1 of US Patent No. 7,657,380 (the figures and legends of which are incorporated herein by reference in their entirety). Thus, for example, the same CDRs described herein can be combined with different framework sequences derived from human germline sequences, as long as the framework regions retain at least 80, 85 or 90% identity with a human germline sequence selected from those listed in Figure 1 of US Patent No. 7,657,380. Alternatively, the CDRs may have amino acid modifications (e.g., 1, 2, 3, 4 or 5 amino acid modifications in the set of CDRs (i.e., the CDRs may be modified so long as the total number of changes in a set of 6 CDRs is less than 6 amino acid modifications, and any combination of CDRs may be altered, e.g., 1 change in VH CDR1, 2 changes in VH CDR2, no changes in VH CDR3, etc.)) as well as variations in the framework regions as long as the framework regions retain at least 80, 85, or 90% identity to a human germline sequence selected from those listed in Figure 1 of U.S. Pat. No. 7,657,380.

[0177] 1. GPC3 antigen-binding domain Provided herein are monoclonal and bispecific antibodies comprising an antigen-binding domain that binds to human GPC3. A suitable set of six CDRs and / or VH and VL domains is shown in Figures 12 and 13. In some embodiments, the heterodimeric antibody is an antibody in 1+1 Fab-scFv-Fc or 2+1 Fab2-scFv-Fv format, although any of the formats outlined below can be used.

[0178] In some embodiments, the anti-GPC3 ABD has a set of vhCDRs selected from vhCDR1, vhCDR2, and vhCDR3 sequences from a VH selected from the group consisting of H1, H1, H1.1, H1.2, H1.3, and H1.4, the sequences of which are shown in Figures 12 and 13.

[0179] In some embodiments, the VH domain of the anti-GPC3 ABD is selected from the group consisting of H1, H1, H1.1, H1.2, H1.3 and H1.4, the sequences of which are shown in Figures 12 and 13.

[0180] In some embodiments, the anti-GPC3 ABD has a set of vlCDRs selected from vlCDR1, vlCDR2 and vlCDR3 sequences from a VL selected from the group consisting of L1, L1.1, L1.2, L1.3, L1.4, L1.5, L1.6, L1.7, L1.8, L1.9, L1.10, L1.16, L1.23, L1.29, L1.31.L1.65, L1.66, L1.68, L1.69, L1.70, L1.71, L1.72 and L1.73, the sequences of which are shown in Figures 12 and 13.

[0181] In some embodiments, the VL of the GPC3 ABD is selected from the group consisting of L1, L1.1, L1.2, L1.3, L1.4, L1.5, L1.6, L1.7, L1.8, L1.9, L1.10, L1.16, L1.23, L1.29, L1.31.L1.65, L1.66, L1.68, L1.69, L1.70, L1.71, L1.72 and L1.73, the sequences of which are shown in Figures 12 and 13.

[0182] したがって、[GPC3]H1_L1、[GPC3]H1_L1.1、[GPC3]H1_L1.2、[GPC3]H1_L1.3、[GPC3]H1_L1.4、[GPC3]H1_L1.5、[GPC3]H1_L1.6、[GPC3]H1_L1.7、[GPC3]H1_L1.8、[GPC3]H1_L1.9、[GPC3]H1_L1.10、[GPC3]H1_L1.16、[GPC3]H1_L1.23、[GPC3]H1_L1.29、[GPC3]H1_L1.31、[GPC3]H1_L1.65、[GPC3]H1_L1.66、[GPC3]H1_L1.67、[GPC3]H1_L1.68、[GPC3]H1_L1.70、[GPC3]H1_L1.71、[GPC3]H1_L1.72、[GPC3]H1_L1.73、[GPC3]H1.1_L1、[GPC3]H1.1_L1.1、[GPC3]H1.1_L1.2、[GPC3]H1.1_L1.3、[GPC3]H1.1_L1.4、[GPC3]H1.1_L1.5、[GPC3]H1.1_L1.6、[GPC3]H1.1_L1.7、[GPC3]H1.1_L1.8、[GPC3]H1.1_L1.9、[GPC3]H1.1_L1.10、[GPC3]H1.1_L1.16、[GPC3]H1.1_L1.23、[GPC3]H1.1_L1.29、[GPC3]H1.1_L1.31、[GPC3]H1.1_L1.65、[GPC3]H1.1_L1.66、L1.67、[GPC3]H1.1_L1.68、[GPC3]H1.1_L1.70、[GPC3]H1.1_L1.71、[GPC3]H1.1_L1.72、[GPC3]H1.1_L1.73、[GPC3]H1.2_L1、[GPC3]H1.2_L1.1、[GPC3]H1.2_L1.2、[GPC3]H1.2_L1.3、[GPC3]H1.2_L1.4、[GPC3]H1.2_L1.5、[GPC3]H1.2_L1.6、[GPC3]H1.2_L1.7、[GPC3]H1.2_L1.8、[GPC3]H1.2_L1.9、[GPC3]H1.2_L1.10、[GPC3]H1.2_L1.16、[GPC3]H1.2_L1.23、[GPC3]H1.2_L1.29、[GPC3]H1.2_L1.31、[GPC3]H1.2_L1.65、[GPC3]H1.2_L1.66、[GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.73, [GPC3]H1.3_L1, [GPC3]H1.3_L1. 1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3]H1.3_L1.7, [GPC3]H1.3_L1.8, [G PC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [GPC3]H1.3_L1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3_L1.71, [GPC3]H1.3_L1.72, [GPC3]H1.3_L1.7 3, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [GPC3]H1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_L1.6, [GPC 3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3]H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.29, [GPC3 Included herein are GPC3 ABDs having a set of six CDRs (vhCDR1, vhCDR2, vhCDR3, vlCDR1, vlCDR2, and vlCDR3) from a VH / VL pair selected from the group consisting of [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73.

[0183] Furthermore, [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3]H1_L1.65, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC3]H1.1_L1, [GPC3]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC3]H1.1_L1.4, [GPC3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC3]H1.1_L1.7, [GPC3]H1.1_L1.8, [GPC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [GPC3]H1.1_L1.31, [GPC3]H1.1_L1.65, [GPC3]H1.1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1.70, [GPC3]H1.1_L1.71, [GPC3]H1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1.2_L1, [GPC3]H1.2_L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1.2_L1.4, [GPC3]H1.2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3]H1.2_L1.8, [GPC3]H1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [GPC3]H1.2_L1.23, [GPC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.73, [GPC3]H1.3_L1, [GPC3]H 1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3]H1.3_L1.7, [GPC 3]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [GPC3]H1.3_L1 .31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3_L1.71, [GP C3]H1.3_L1.72, [GPC3]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [GPC3]H1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3]H1.4_L Included herein are GPC3 ABDs having a VH / VL pair selected from the group consisting of [GPC3]H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.29, [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73.

[0184] In certain embodiments, the VH / VL pair is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69.

[0185] In a specific embodiment, the VH / VL pair is Fab and is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69.

[0186] As will be appreciated by those of skill in the art, a suitable GPC3 binding domain may comprise a set of six CDRs as shown in the figures, either as underlined or as identified using other alignments within the VH and VL sequences of those shown in Figures 12 and 13, if a different numbering scheme is used as described herein and shown in Table 2. Suitable ABDs may also comprise these sequences and the entire VH and VL sequences as shown in the figures, used as scFv or Fab. In many of the embodiments herein that include an Fv against GPC3, it is the Fab monomer that is linked to GPC3.

[0187] In addition to the parent CDR set disclosed in the figures and sequence listing that form the ABD for GPC3, variant GPC3 ABDs are provided herein that have CDRs that include at least one modification of the GPC3 ABD CDRs disclosed herein (e.g., Figures 12 and 13, and the sequence listing). In one embodiment, the GPC3 ABD of the subject heterodimeric antibodies comprises a set of six CDRs that have 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 amino acid modifications compared to the six CDRs of the VH / VL pair of the GPC3 binding domains described herein, such as the figures and sequence listing. In an exemplary embodiment, the GPC3 ABD of the subject heterodimeric antibodies comprises a set of six CDRs that have 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 amino acid modifications compared to the six CDRs of one of the following GPC3 binding domains: [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69. In certain embodiments, the GPC3 ABD of the subject antibody can bind to GPC3 as measured by at least one of Biacore, surface plasmon resonance (SPR), BLI (biolayer interferometry, e.g., Octet Assay) assays, and / or flow cytometry, the latter finding particular use in many embodiments. In certain embodiments, the GPC3 ABD can bind to human GPC3 (see FIG. 11). In some cases, each variant CDR has no more than one or two amino acid changes, with no more than one per CDR being particularly useful.

[0188] In some embodiments, the GPC3 ABD of the subject antibody comprises six CDRs that are at least 90, 95, 97, 98, or 99% identical to the six CDRs of the GPC3 ABD described herein, including the figures and sequence listings. In exemplary embodiments, the GPC3 ABD of the subject antibody comprises six CDRs that are at least 90, 95, 97, 98, or 99% identical to the six CDRs of one of the following VH / VL pairs of GPC3 binding domains: [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69. In certain embodiments, the GPC3 ABD of the subject antibody can bind to GPC3 as measured by at least one of Biacore, surface plasmon resonance (SPR), BLI (biolayer interferometry, e.g., octet assay) assays, and / or flow cytometry. In certain embodiments, the GPC3 ABD can bind to human GPC3 antigen (see FIG. 11).

[0189] In another exemplary embodiment, the GPC3 ABD of the subject antibody comprises a variable heavy (VH) domain and a variable light (VL) domain of any one of the VH / VL pairs of the GPC3 ABD binding domains described herein, including the figures and sequence listings (e.g., Figures 12 and 13).

[0190] In some embodiments, the subject antibodies comprise a GPC3 ABD comprising a variable heavy and / or variable light domain that is a variant of the VH and VL domains of the GPC3 ABD disclosed herein. In one embodiment, the VH and / or VL domains of the variant have 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid changes from the VH and / or VL domains of the GPC3 ABD described herein, including the figures and sequence listings. In an exemplary embodiment, the VH and / or VL domains of the variant have 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid changes from the VH and / or VL domains from one of the following VH / VL pairs of GPC3 binding domains: [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69. In some embodiments, the changes are in the VH domains shown in Figures 12 and 13. In some embodiments, the changes are in the VL domain depicted in Figure 12 and Figure 13. In some embodiments, the changes are in the VH and VL domains depicted in Figure 12 and Figure 13. In some embodiments, the one or more amino acid changes are in the VH and / or VL framework regions (FR1, FR2, FR3, and / or FR4). In some embodiments, the one or more amino acid changes are in one or more CDRs. In certain embodiments, the GPC3 ABD of the subject antibody can bind to GPC3 as measured by at least one of Biacore, surface plasmon resonance (SPR), BLI (biolayer interferometry, e.g., octet assay) assay, and / or flow cytometry. In certain embodiments, the GPC3 ABD can bind to human GPC3 antigen (see Figure 11).

[0191] In one embodiment, the VH and / or VL domains of the variant are at least 90, 95, 97, 98, or 99% identical to the VH and / or VL of the GPC3 ABD described herein, including the Figures and Sequence Listings. In an exemplary embodiment, the VH and / or VL domains of the variant are at least 90, 95, 97, 98, or 99% identical to the VH and / or VL of one of the following VH / VL pairs of GPC3 binding domains: [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69. In some embodiments, the GPC3 ABD comprises a VH that is at least 90, 95, 97, 98, or 99% identical to the VH domain shown in Figures 12 and 13. In some embodiments, the GPC3 ABD comprises a VL that is at least 90, 95, 97, 98, or 99% identical to the VL domain shown in Figures 12 and 13. In some embodiments, the GPC3 ABD comprises a VH and VL that are at least 90, 95, 97, 98, or 99% identical to the VH and VL domains shown in Figures 12 and 13. In certain embodiments, the GPC3 ABD of the subject antibody can bind to GPC3 as measured by at least one of Biacore, surface plasmon resonance (SPR), BLI (biolayer interferometry, e.g., Octet Assay) assays, and / or flow cytometry. In certain embodiments, the GPC3 ABD can bind to the human GPC3 antigen (see Figure 11).

[0192] 2. CD3 antigen-binding domain The heterodimeric bispecifics of the invention (eg, anti-GPC 3×anti-CD3 antibodies) also comprise an ABD that binds human epsilon CD3 (CD3ε).

[0193] A suitable set of six CDRs and / or VH and VL domains and scFv sequences is shown in FIG. CD3 binding domain sequences of particular use include, but are not limited to, anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 H1.31_L1.47, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 L1.47_H1.31, as shown in FIG. 10. As shown in FIG. 10, when the anti-CD3 ABD is an scFv domain, the VH and VL domains can be in either orientation.

[0194] As will be appreciated by those of skill in the art, a suitable CD3 binding domain may comprise the set of six CDRs shown in Figure 10 as underlined, or as identified using other alignments within the VH and VL sequences of those shown in Figures 10A-10F if a different numbering scheme is used, as described herein and shown in Table 2. A suitable ABD may also comprise these sequences and the entire VH and VL sequences as shown in the figures, used as an scFv or Fab. In many of the embodiments herein that include an Fv against CD3, it is the scFv monomer that is linked to CD3.

[0195] In addition to the parent CDR set disclosed in the Figures and Sequence Listing that form the ABD for CD3, variant CD3 ABDs are provided herein having CDRs that include at least one modification of a CD3 ABD CDR disclosed herein (e.g., (Figure 10 and Sequence Listing). In one embodiment, the CD3 ABD of a subject heterodimeric antibody (e.g., an anti-GPC3 x anti-CD3 antibody) comprises a set of six CDRs that have 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 amino acid modifications compared to the six CDRs of the CD3 ABD described herein, such as in the Figures and Sequence Listing. In an exemplary embodiment, the CD3 ABD of a subject heterodimeric antibody comprises the following CD3 binding domains: anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 In certain embodiments, the subject antibodies include a set of six CDRs having 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 amino acid modifications compared to the six CDRs of one of anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 L1.47_H1.31 (FIG. 10). The ABD is capable of binding to the CD3 antigen as measured by at least one of Biacore, surface plasmon resonance (SPR), flow cytometry, and / or BLI (biolayer interferometry, e.g., Octet Assay) assays, the latter finding particular use in many embodiments. In certain embodiments, the CD3ABD is capable of binding to human CD3.

[0196] In some embodiments, the CD3 ABD of the subject antibodies comprises six CDRs that are at least 90, 95, 97, 98, or 99% identical to the six CDRs of the CD3 ABDs described herein, including the Figures and Sequence Listings. In exemplary embodiments, the CD3 ABD of the subject antibodies comprises the following CD3 binding domains: anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 H1.31_L1.47, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 L1.47_H1.31 (FIG. 10). In certain embodiments, the CD3 ABD is capable of binding to CD3 as measured by at least one of Biacore, surface plasmon resonance (SPR), flow cytometry, and / or BLI (biolayer interferometry, e.g., Octet Assay) assays, the latter finding particular use in many embodiments. In certain embodiments, the CD3 ABD is capable of binding to the human CD3 antigen.

[0197] In another exemplary embodiment, the CD3 ABD of a subject antibody comprises the variable heavy (VH) and variable light (VL) domains of any one of the CD3 binding domains described herein, including the figures and sequence listings.

[0198] In some embodiments, the subject antibodies comprise a CD3 ABD that comprises a variable heavy and / or variable light domain that is a variant of the VH and VL domains of the CD3 ABD disclosed herein, in one embodiment, the variant VH and / or VL domain has 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10 amino acid changes from the VH and / or VL domains of the CD3 ABD described herein, including in the Figures and Sequence Listings. In exemplary embodiments, the VH and / or VL domains of the variants comprise the following CD3 binding domains: anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 H1.31_L1.47, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 L1.47_H1.31 (FIG. 10). In some embodiments, the changes are in the VH domain shown in FIG. 10. In some embodiments, the changes are in the VL domain shown in FIG. 10. In some embodiments, the changes are in the VH and VL domains shown in FIG. 10. In some embodiments, the one or more amino acid changes are in the VH and / or VL framework regions (FR1, FR2, FR3, and / or FR4). In some embodiments, the one or more amino acid changes are in one or more CDRs. In certain embodiments, the CD3 ABD of the subject antibody is capable of binding to CD3 as measured by at least one of Biacore, surface plasmon resonance (SPR), flow cytometry, and / or BLI (biolayer interferometry, e.g., Octet Assay) assays, the latter finding particular use in many embodiments. In certain embodiments, the CD3 ABD is capable of binding to the human CD3 antigen.

[0199] In one embodiment, the VH and / or VL domain of the variant is at least 90, 95, 97, 98 or 99% identical to the VH and / or VL of a CD3 ABD described herein, including in the Figures and Sequence Listing. In exemplary embodiments, the VH and / or VL domains of the variants comprise the following CD3 binding domains: anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 H1.31_L1.47, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 L1.47_H1.31 ( FIG. 10 ). In some embodiments, the CD3 ABD comprises a VH that is at least 90, 95, 97, 98, or 99% identical to the VH domain depicted in FIG. 10. In some embodiments, the CD3 ABD comprises a VL that is at least 90, 95, 97, 98, or 99% identical to the VL domain depicted in FIG. 10. In some embodiments, the CD3 ABD comprises a VH and VL that are at least 90, 95, 97, 98, or 99% identical to the VH and VL domains depicted in FIG. 10. In certain embodiments, the CD3 ABD is capable of binding to CD3 as measured by at least one of Biacore, surface plasmon resonance (SPR), flow cytometry, and / or BLI (biolayer interferometry, e.g., Octet Assay) assays, the latter finding particular use in many embodiments. In certain embodiments, the CD3 ABD is capable of binding to the human CD3 antigen.

[0200] In addition to the αCD3 ABD in FIG. 10, additional ABDs useful in the present invention include those shown in FIGS. 14 and 15 of WO 2014 / 145806, which are expressly incorporated herein in their entirety, including their figures and legends.

[0201] 3. Linker As shown herein, there are several suitable linkers (used as either domain linkers or scFv linkers) that can be used to covalently link (including traditional peptide bonds produced by recombinant techniques) the enumerated domains (e.g., scFv, Fab, Fc domains, etc.). Exemplary linkers for attaching the domains of the subject antibodies to one another are shown in FIG. 6. In some embodiments, the linker peptide may primarily comprise the following amino acid residues: Gly, Ser, Ala, or Thr. The linker peptide should be of sufficient length to link the two molecules in such a way that they assume the correct conformation relative to one another so that they retain the desired activity. In one embodiment, the linker is about 1-50 amino acids long, preferably about 1-30 amino acids long. In one embodiment, linkers of 1-20 amino acids long may be used, with about 5 to about 10 amino acids finding use in some embodiments. Useful linkers include, for example, glycine-serine polymers, glycine-alanine polymers, alanine-serine polymers, including (GS), (GSGGS) (SEQ ID NO: 3), (GGGGS) (SEQ ID NO: 2), and (GGGS) (SEQ ID NO: 4), where n is an integer of at least 1 (and typically 3-4), as well as other flexible linkers, some of which are shown in Figures 5 and 6. Alternatively, a variety of non-proteinaceous polymers may find use as linkers, including, but not limited to, polyethylene glycol (PEG), polypropylene glycol, polyoxyalkylenes, or copolymers of polyethylene glycol and polypropylene glycol.

[0202] Other linker sequences include any sequence of the CL / CH1 domain of any length, but may not include all residues of the CL / CH1 domain, for example, the first 5-12 amino acid residues of the CL / CH1 domain. Kor Cλ. The linker may be derived from an immunoglobulin heavy chain of any isotype, including, for example, γ1, Cγ2, Cγ3, Cγ4, Cal, Ca2, Cδ, Cε, and Cμ. Linker sequences may also be derived from other proteins, such as Ig-like proteins (e.g., TCR, FcR, KIR), sequences from hinge regions, and other naturally occurring sequences from other proteins.

[0203] In some embodiments, the linker is a "domain linker" used to link together any two domains as outlined herein. For example, in FIG. 15B, there may be a domain linker that connects the C-terminus of the CH1 domain of the Fab to the N-terminus of the scFv, and another optional domain linker connects the C-terminus of the scFv to the CH2 domain (although in many embodiments, a hinge is used as this domain linker). While any suitable linker can be used, many embodiments utilize glycine-serine polymers as domain linkers, including, for example, (GS)n, (GSGGS)n (SEQ ID NO:3), (GGGGS)n (SEQ ID NO:2), and (GGGS)n (SEQ ID NO:4), where n is an integer of at least 1 (generally 3-4-5), as well as any peptide sequence that allows recombinant attachment of two domains with sufficient length and flexibility so that each domain retains its biological function. In some cases, charged domain linkers can be used, as used in some embodiments of scFv linkers, with attention to the "degree of twist" as outlined below. Exemplary useful domain linkers are shown in FIG.

[0204] With particular reference to the domain linkers used to connect the scFv domain to the Fc domain in a "2+1" format, there are several domain linkers that find particular use, including the "full hinge C220S variant," "flex half hinge," "charged half hinge 1," and "charged half hinge 2," as shown in FIG.

[0205] In some embodiments, the linker is a "scFv linker" used to covalently link the VH and VL domains discussed herein. In many cases, the scFv linker is a charged scFv linker, some of which are shown in FIG. 5. Thus, in some embodiments, the antibodies described herein further provide a charged scFv linker to facilitate separation of the pI between the first and second monomers. That is, by incorporating either a positively or negatively charged scFv linker (or both in the case of scaffolds that use scFvs on different monomers), this allows for altering the pI of the monomers that contain the charged linker without further altering the Fc domain. These charged linkers can be substituted into any scFv containing standard linkers. Again, as will be appreciated by those skilled in the art, a charged scFv linker is used on the correct "chain" or monomer according to the desired change in pI. For example, as discussed herein, to generate a 1+1 Fab-scFv-Fc format heterodimeric antibody, the original pI of the Fv region for each of the desired antigen binding domains is calculated and one is selected to generate the scFv, and depending on the pI, either a positive or negative linker is selected.

[0206] Charged domain linkers can also be used to increase the pI separation of the antibody monomers described herein, and thus those included in FIG. 5 can be used in any embodiment herein in which a linker is utilized.

[0207] G. Useful Formats of the Invention As will be appreciated by those of skill in the art and described in more detail below, the bispecific heterodimeric antibodies provided herein can be in a wide variety of configurations, as generally shown in FIG. 15 as well as FIG. 44. Some figures show "single-ended" configurations, with one type of specificity on one "arm" of the molecule and a different specificity on the other "arm". Other figures show "dual-ended" configurations, with at least one type of specificity on the "top" of the molecule and one or more different specificities on the "bottom" of the molecule. Thus, in some embodiments, novel immunoglobulin compositions are directed to simultaneously linking different first and second antigens.

[0208] As will be appreciated by those skilled in the art, the heterodimeric formats of the present invention may have different valencies and may be bispecific. That is, the heterodimeric antibodies of the antibodies described herein may be bivalent and bispecific, where one target tumor antigen (e.g., CD3) is bound by one binding domain and the other target tumor antigen (e.g., GPC3) is bound by a second binding domain. The heterodimeric antibodies may also be trivalent and bispecific, where a first antigen is bound by two binding domains and a second antigen is bound by a second binding domain. As outlined herein, when CD3 is one of the target antigens, it is preferred that CD3 is bound only monovalently to reduce potential side effects.

[0209] The antibodies described herein utilize an anti-CD3 antigen binding domain in combination with an anti-GPC3 binding domain. As will be appreciated by those skilled in the art, any collection of anti-CD3 CDRs, anti-CD3 variable light and heavy domains, Fabs and scFvs as shown in any of the figures can be used. Similarly, any of the anti-GPC3 antigen binding domains can be used, and any of the CDRs, variable light and heavy domains, Fabs and scFvs can be used, optionally and independently, in any combination as shown in any of the figures.

[0210] 1.1+1Fab-scFv-Fc format One heterodimeric scaffold that finds particular use with the antibodies described herein is the "1+1 Fab-scFv-Fc" or "bottle opener" format as shown in Figure 15A, with an exemplary combination of a CD3 binding domain and a tumor target antigen (GPC3) binding domain. In this embodiment, one heavy chain monomer of the antibody comprises a single chain Fv ("scFv" as defined below) and an Fc domain. The scFv comprises a variable heavy domain (VH1) and a variable light domain (VL1), where VH1 is linked to VL1 using a flexible scFv linker (see, e.g., Figure 5). The scFv is linked to the heavy chain using a domain linker (see, e.g., Figure 6). The other heavy chain monomer is a "normal" heavy chain (VH-CH1-hinge-CH2-CH3). The 1+1 Fab-scFv-Fc also comprises a light chain that interacts with VH-CH1 to form a Fab. This structure is sometimes referred to herein as the "bottle opener" (format) due to its broad visual resemblance to a bottle opener. The two heavy chain monomers are held together through the use of amino acid variants (e.g., the heterodimerization variants described above) in the constant regions (e.g., the Fc domain, the CH1 domain and / or the hinge region) that promote the formation of heterodimeric antibodies, as described in more detail below.

[0211] The current "1+1 Fab-scFv-Fc" format has several distinct advantages. As is known in the art, antibody analogs that rely on two scFv constructs often have stability and aggregation problems that can be mitigated by the addition of "normal" heavy and light chain pairing in the antibodies described herein. In addition, in contrast to formats that rely on two heavy chains and two light chains, there is no problem with mispairing of heavy and light chains (e.g., heavy chain 1 and light chain 2).

[0212] Many of the embodiments outlined herein generally rely on 1+1 Fab-scFv-Fc or "bottle opener" format antibodies, which comprise a first monomer comprising a scFv, which comprises variable heavy and variable light domains covalently linked with a scFv linker (often charged, but not always), and the scFv is covalently linked to a first Fc domain, usually via a domain linker. The domain linker may be charged or uncharged, and may be exogenous or endogenous (e.g., all or part of a natural hinge domain). Any suitable linker may be used to attach the scFv to the N-terminus of the first Fc domain. In some embodiments, the domain linker is selected from the domain linkers in FIG. 6. The second monomer of the 1+1 Fab-scFv-Fc or "bottle opener" format is a heavy chain, and the composition further comprises a light chain.

[0213] Generally, in many preferred embodiments, the scFv is the CD3 binding domain and the Fab forms the GPC3 binding domain. An exemplary anti-GPC3 x anti-CD3 bispecific antibody in the 1+1 Fab-scFv-Fc format is shown in Figure 15A. An exemplary anti-GPC3 x anti-CD3 bispecific antibody in the 1+1 Fab-scFv-Fc format is shown in Figures 16 and 17.

[0214] In addition, the Fc domains of the antibodies described herein generally comprise sqvariants (e.g., the set of amino acid substitutions shown in FIG. 1 , where particularly useful sqvariants are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K / E35 5, and the heavy chain comprises a pI variant (including those shown in FIG. 2).

[0215] In certain embodiments, the 1+1 Fab-scFv-Fc scaffold format comprises a first monomer comprising an scFv domain linker-CH2-CH3 monomer, a second monomer comprising a first variable heavy domain-CH1-hinge-CH2-CH3, and a third monomer comprising a first variable light domain. In some embodiments, the CH2-CH3 of the first monomer is a first variant Fc domain and the CH2-CH3 of the second monomer is a second variant Fc domain. In some embodiments, the scFv comprises an scFv variable heavy domain and an scFv variable light domain that form a CD3 binding moiety. In certain embodiments, the scFv variable heavy domain and the scFv variable light domain are covalently linked using a scFv linker (which is often, but not all, charged). See, e.g., FIG. 5. In some embodiments, the first variable heavy domain and the first variable light domain form a GPC3 binding domain.

[0216] In some embodiments, the 1+1 Fab-scFv-Fc format comprises a skewer variant, a pI variant, and a deletion variant. Thus, some embodiments comprise a 1+1 Fab-scFv-Fc format, which comprises a) a first monomer ("scFv monomer") comprising a charged scFv linker (in some embodiments the +H sequence of FIG. 5 is preferred), a skewer variant S364K / E357Q, a deletion variant E233P / L234V / L235A / G236del / S267K, and an scFv that binds to CD3 as outlined herein; b) a skewer variant S364K / E357Q, a deletion variant E233P / L234V / L235A / G236del / S267K, and a CD3-binding scFv as outlined herein; a second monomer ("Fab monomer") comprising the variants L368D / K370S, pI variants N208D / Q295E / N384D / Q418E / N421D, deletion variants E233P / L234V / L235A / G236del / S267K, and a variable heavy domain; and c) a light chain comprising a variable light domain (VL) and a constant light domain (CL), numbering according to EU numbering. The variable heavy domain and the variable light domain constitute the GPC3 binding portion.

[0217] Any suitable CD3 ABD can be included in the 1+1 Fab-scFv-Fc format antibody, including those provided herein. CD3 binding domain sequences that find particular use in these embodiments include, but are not limited to, H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31 or variants thereof, and those depicted in Figure 10 and Figures 14 and 15 of WO 2014 / 145806, which are incorporated herein by reference, including the legend.

[0218] Any suitable GPC3 ABD can be included in the 1+1 Fab-scFv-Fc format antibody, including those provided herein. GPC3 ABDs that are particularly useful in these embodiments include [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3] H1_L1.65, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC3]H1.1_L1, [GPC3]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC3]H1.1_L1.4, [GPC3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC 3]H1.1_L1.7, [GPC3]H1.1_L1.8, [GPC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [ GPC3]H1.1_L1.31, [GPC3]H1.1_L1.65, [GPC3]H1.1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1.70, [GPC3]H1.1_L1.71, [GPC3]H 1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1.2_L1, [GPC3]H1.2_L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1.2_L1.4, [GPC3]H1. 2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3]H1.2_L1.8, [GPC3]H1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [GPC3]H1.2_L1.23, [GPC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.73, [GPC3]H1.3_L1, [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3]H1.3_L1.7, [GPC3]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [GPC3]H1.3_L1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3_L1.71, [GPC3]H1.3_L1.72, [GPC3]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [GPC3]H1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3]H1.4_L1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.29, [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73, or a variant thereof.

[0219] In certain embodiments, the VH / VL pair is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69, or a variant thereof.

[0220] In some embodiments, the 1+1 Fab-scFv-Fc format comprises a scFv variant, a pI variant, a deletion variant and an FcRn variant. Thus, some embodiments comprise a 1+1 Fab-scFv-Fc format, which comprises a) a first monomer ("scFv monomer") comprising a charged scFv linker (in some embodiments the +H sequence of FIG. 5 is preferred), a scFv variant S364K / E357Q, a deletion variant E233P / L234V / L235A / G236del / S267K, an FcRn variant M428L / N434S, and an scFv that binds to CD3 as outlined herein; b) a scFv variant ("scFv monomer") comprising a scFv variant S364K / E357Q, a deletion variant E233P / L234V / L235A / G236del / S267K, an FcRn variant M428L / N434S, and an FcRn variant M428L / N434S that binds to CD3 as outlined herein; and c) a light chain comprising a variable light domain (VL) and a constant light domain (CL), numbering according to EU numbering. The variable heavy domain and the variable light domain constitute the GPC3 binding domain. CD3 binding domain sequences that find particular use in these embodiments include, but are not limited to, H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31, or variants thereof, and those depicted in FIG. 10. GPC3 binding domain sequences that are particularly useful in these embodiments include, but are not limited to, an αGPC3 ABD VH / VL pair selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69.

[0221] Particularly useful combinations of GPC3 and CD3 sequences for use in antibodies in the 1+1 format are disclosed, for example, in FIGS.

[0222] Figures 7A-7D show some exemplary Fc domain sequences useful in 1+1 Fab-scFv-Fc format antibodies. The "monomer 1" sequences shown in Figures 7A-7D generally refer to the Fc domain of the "Fab-Fc heavy chain," and the "monomer 2" sequences refer to the Fc domain of the "scFv-Fc heavy chain." Additionally, Figure 9 shows useful CL sequences that can be used in this format.

[0223] In some embodiments, any of the VH and VL sequences shown herein (including all VH and VL sequences shown in the figures and sequence listing, including those directed against GPC3) can be added to the bottle opener scaffold format of Figures 7A-7D as the "Fab side" using any of the anti-CD3 scFv sequences shown in the figures and sequence listing.

[0224] For bottle opener scaffold 1 from FIG. 7A (optionally including the 428L / 434S variant), CD binding domain sequences that find particular use in these embodiments include, but are not limited to, CD3 binding domains anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47 and anti-CD3 H1.31_L1.47 attached as the scFv side of the scaffold shown in FIG. 7.

[0225] 2. mAb-Fv One heterodimeric scaffold that finds particular use with the antibodies described herein is the mAb-Fv format (Figure 44G). In this embodiment, the format relies on the use of an "additional" variable heavy domain C-terminally attached to one monomer and an "additional" variable light domain C-terminally attached to the other monomer, thereby forming a third antigen-binding domain, with the Fab portions of the two monomers binding to GPC3 and the "additional" scFv domain binding to CD3.

[0226] In this embodiment, the first monomer comprises a first heavy chain comprising a first constant heavy chain domain comprising a first variable heavy chain domain and a first Fc domain, and has a first variable light chain domain covalently linked to the C-terminus of the first Fc domain using a domain linker (VH1-CH1-hinge-CH2-CH3-[optional linker]-VL2). The second monomer comprises a second variable heavy domain of a second constant heavy domain comprising a second Fc domain and a third variable heavy domain covalently linked to the C-terminus of the second Fc domain using a domain linker (vh1-CH1-hinge-CH2-CH3-[optional linker]-VH2. The two C-terminally linked variable domains constitute an Fv that binds to CD3 (as it is less desirable to have bivalent CD3 binding). This embodiment further utilizes a common light chain comprising a variable light domain and a constant light domain, which associates with the heavy chains to form two identical Fabs that link to GPC3. For many of the embodiments described herein, these constructs include scubariants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0227] The antibodies described herein provide a mAb-Fv format in which the CD3 binding domain sequence is as shown in Figure 10 or a variant thereof. The antibodies described herein provide a mAb-Fv format in which the GPC3 binding domain sequence is as shown in Figures 12 and 13 or a variant thereof.

[0228] In addition, the Fc domain of the mAb-Fv format can be modified with a sqvariant (e.g., the set of amino acid substitutions shown in FIG. 1, where particularly useful sqvariants are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K / E357Q ... L368E / K370S:S364K, L368E / K370S:S364K, L368E / K370S:S364K, L368E / K370S:S364K, L368E / K370S:S364K, L368E / K370S:S364K, L368E / K370S:S364K, L368E / K370S:S364K, L, K370S:S364K / E357Q, T366S / L368A / Y407V:T366W, and T366S / L368A / Y407V / Y349C:T366W / S354C), optionally a deletion variant (including those shown in FIG. 3), optionally a charged scFv linker (including those shown in FIG. 5), and the heavy chain comprises a pI variant (including those shown in FIG. 2).

[0229] In some embodiments, the mAb-Fv format comprises a scubariant, a pI variant, and a deletion variant. Thus, some embodiments comprise a mAb-Fv format comprising: a) a first monomer comprising a scubariant S364K / E357Q, a deletion variant E233P / L234V / L235A / G236del / S267K, and a first variable heavy chain domain constituting an Fv that binds to GPC3, together with a first variable light chain domain of the light chain, and a second variable heavy chain domain; b) a scubariant L368D / K370S, a pI variant N208D, and a deletion variant E233P / L234V / L235A / G236del / S267K, and a first variable light chain domain of the light chain; a second monomer comprising a first variable heavy chain domain which, together with the first variable light chain domain, constitutes an Fv that binds to GPC3 as outlined herein, and a second variable light chain which, together with the second variable heavy chain domain, forms an Fv(ABD) that binds to CD3; and c) a light chain comprising the first variable light chain domain and a constant light chain domain.

[0230] In some embodiments, the mAb-Fv format includes scubariants, pI variants, deletion variants, and FcRn variants. Thus, some embodiments include a mAb-Fv format, which includes a) a first monomer comprising a scubariant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variants M428L / N434S, and a first variable heavy chain domain constituting an Fv that binds to GPC3, together with a second variable heavy chain domain; b) a scubariant L368D / K370S, pI variants N208D / Q208D ... a second monomer comprising a first variable heavy chain domain which, together with the first variable light chain domain, constitutes an Fv that binds GPC3 as outlined herein, and a second variable light chain which, together with the second variable heavy chain domain of the first monomer, forms an Fv(ABD) that binds CD3; and c) a light chain comprising the first variable light chain domain and a constant light chain domain.

[0231] 3. mAb-scFv One heterodimeric scaffold that finds particular use with the antibodies described herein is the mAb-scFv format (Figure 44H). In this embodiment, the format relies on the use of C-terminal attachment of an scFv to one of the monomers, thereby forming a third antigen-binding domain, with the Fab portions of the two monomers binding to GPC3 and the "additional" scFv domain binding to CD3. Thus, the first monomer comprises a first heavy chain (comprising a variable heavy domain and a constant domain) with a covalently attached scFv at its C-terminus comprising an scFv variable light domain, an scFv linker, and an scFv variable heavy domain in either orientation (VH1-CH1-hinge-CH2-CH3-[optional linker]-VH2-scFv linker-VL2 or VH1-CH1-hinge-CH2-CH3-[optional linker]-VL2-scFv linker-VH2). This embodiment further utilizes a common light chain comprising a variable light domain and a constant light domain, which associates with the heavy chain to form two identical Fabs that bind to GPC3. For many of the embodiments described herein, these constructs include scubariants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0232] The antibodies described herein provide a mAb-scFv format in which the CD binding domain sequence is as shown in Figures 10A-10F or a variant thereof, and the GPC3 binding domain sequence is as shown in Figures 12 and 13 or a variant thereof.

[0233] In addition, the Fc domain of the mAb-scFv format may be modified with a scFv variant (e.g., the set of amino acid substitutions shown in FIG. 1 , where particularly useful scFv variants are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S ... 5, and the heavy chain comprises a pI variant (including those shown in FIG. 2).

[0234] In some embodiments, the mAb-scFv format includes scFv variants, pI variants, and deletion variants. Thus, some embodiments include a mAb-scFv format, which comprises a) a first monomer comprising a scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, and a variable heavy chain domain constituting an Fv that binds GPC3 as outlined herein together with a variable light chain domain of the common light chain, and an scFv domain that binds CD3; b) a first monomer comprising a scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, and a variable light chain domain of the common light chain, and a variable heavy chain domain constituting an Fv that binds GPC3 as outlined herein; a second monomer comprising a variable heavy chain domain constituting an Fv that binds to GPC3 as outlined herein together with the variable light chain domain of the cuvariant L368D / K370S, the pI variants N208D / Q295E / N384D / Q418E / N421D, the deletion variants E233P / L234V / L235A / G236del / S267K, and the variable light chain domain of the common light chain, and c) a common light chain comprising a variable light chain domain and a constant light chain domain.

[0235] In some embodiments, the mAb-scFv format includes scFv variants, pI variants, deletion variants, and FcRn variants. Thus, some embodiments include a mAb-scFv format, which comprises a) a first monomer comprising a scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variants M428L / N434S, and a variable heavy chain domain constituting an Fv that binds GPC3 as outlined herein together with a variable light chain domain of the common light chain, and an scFv domain that binds CD3; b) a) a variable heavy chain domain constituting an Fv that binds to GPC3 as outlined herein together with the variable light chain of the scubariant L368D / K370S, the pI variants N208D / Q295E / N384D / Q418E / N421D, the deletion variants E233P / L234V / L235A / G236del / S267K, the FcRn variant M428L / N434S, and the common light chain, and c) a common light chain comprising a variable light chain domain and a constant light chain domain.

[0236] 4.2+1Fab2-scFv-Fc Format One heterodimeric scaffold that finds particular use with the antibodies described herein is the "2+1 Fab2-scFv-Fc" format (also referred to as the "Central-scFv format" in previous related filings) shown in FIG. 15B in an exemplary combination of a CD3 binding domain and two tumor target antigen (GPC3) binding domains. In this embodiment, the format relies on the use of an inserted scFv domain, thereby forming a third antigen binding domain, with the Fab portions of the two monomers binding to GPC3 and the "additional" scFv domain binding to CD3. The scFv domain is inserted between the Fc domain and the CH1-Fv region of one of the monomers, thereby providing the third antigen binding domain. As described, GPC3×CD3 bispecific antibodies with a 2+1 Fab2-scFv-Fc format are potent in inducing redirected T-cytotoxicity in a cellular environment expressing low levels of GPC3. Furthermore, as shown in the examples, GPC3xCD3 bispecific antibodies with 2+1 Fab2-scFv-Fc format allow for "fine-tuning" of immune responses, since such antibodies exhibit a wide variety of different properties depending on the GPC3 and / or CD3 binding domains used. For example, such antibodies exhibit selectivity for cells with different GPC3 expression, potency against GPC3-expressing cells, ability to induce cytokine release, and different sensitivity to soluble GPC3. These GPC3 antibodies are used, for example, in the treatment of GPC3-associated cancers.

[0237] In this embodiment, one monomer comprises a first heavy chain comprising a first variable heavy domain, a CH1 domain (and optional hinge) and an Fc domain, and an scFv comprising an scFv variable light domain, an scFv linker and an scFv variable heavy domain. The scFv is covalently linked between the C-terminus of the CH1 domain of the heavy chain constant domain and the N-terminus of the first Fc domain using an optional domain linker (VH1-CH1-[optional linker]-VH2-scFv linker-VL2-[optional linker including hinge]-CH2-CH3, or in the opposite orientation to the scFv, VH1-CH1-[optional linker]-VL2-scFv linker-VH2-[optional linker including hinge]-CH2-CH3). The optional linker can be any suitable peptide linker, such as, for example, the domain linkers included in FIG. 6. In some embodiments, the optional linker is a hinge or a fragment thereof. The other monomer is a standard Fab side (e.g., VH1-CH1-hinge-CH2-CH3). This embodiment further utilizes a common light chain that includes a variable light domain and a constant light domain, which associates with the heavy chain to form two identical Fabs that bind to GPC3. For many of the embodiments described herein, these constructs include scubariants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0238] In one embodiment, an antibody in 2+1 Fab2-scFv-Fc format comprises an scFv having the VH and VL of the CD3 binding domain sequence shown in Figure 10. In one embodiment, an antibody in 2+1 Fab2-scFv-F format comprises two Fabs having the VH and VL of the GPC3 binding domain sequence shown in Figures 12 and 13.

[0239] In an exemplary embodiment, a 2+1 Fab2-scFv-Fc that finds particular use in these embodiments is GPC3 binding domains of GPC3xCD3 bispecific antibodies include, but are not limited to, H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31, or variants thereof, and those depicted in Figure 10 and Figures 14 and 15 of WO 2014 / 145806, which are incorporated herein by reference, including the legend.

[0240] Any suitable GPC3 ABD can be included in the 2+1 Fab2-scFv-Fc format antibody, including those provided herein. GPC3 ABDs that are particularly useful in these embodiments include [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3] H1_L1.65, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC3]H1.1_L1, [GPC3]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC3]H1.1_L1.4, [GPC3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC 3]H1.1_L1.7, [GPC3]H1.1_L1.8, [GPC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [ GPC3]H1.1_L1.31, [GPC3]H1.1_L1.65, [GPC3]H1.1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1.70, [GPC3]H1.1_L1.71, [GPC3]H 1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1.2_L1, [GPC3]H1.2_L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1.2_L1.4, [GPC3]H1. 2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3]H1.2_L1.8, [GPC3]H1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [GPC3]H1.2_L1.23, [GPC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GP C3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1. 70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.73, [GPC3]H1 .3_L1, [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1 .3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3]H1.3_L1.7, [GPC3] H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [G PC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [GPC3]H1.3_L1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H 1.3_L1.70, [GPC3]H1.3_L1.71, [GPC3]H1.3_L1.72, [GPC3]H1.3_L1.73, [ GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [ GPC3]H1.4_L1.4, [GPC3]H1.4_L1.5, [GPC3]H1.4_L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3]H1.4_L 1.16, [GPC3]H1.4_L1.23, [GPC3]H1.4_L1.29, [GPC3]H1.4_L1.31, [GPC3]H1.4_L1.65, [GPC3]H1.4_L1.66, [GPC3]H1.4_L1.67, [GPC3]H1.4_L1.68, [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73, or variants thereof.

[0241] In certain embodiments, the VH / VL pair is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69, or a variant thereof.

[0242] In addition, the Fc domain of the 2+1 Fab2-scFv-Fc format may be a scavariant (e.g., a set of amino acid substitutions as shown in FIG. 1, where particularly useful scavariant are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K / E35 5, and the heavy chain comprises a pI variant (including those shown in FIG. 2).

[0243] In some embodiments, the 2+1 Fab2-scFv-Fc format antibody comprises scFv variants, pI variants and deletion variants. Thus, some embodiments comprise a 2+1 Fab2-scFv-Fc format, which comprises: a) a first monomer (Fab-scFv-Fc side) comprising a scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, and a variable heavy chain domain constituting an Fv that binds to GPC3 as outlined herein, together with a variable light chain domain of the common light chain, and an scFv domain that binds to CD3; b) a first monomer (Fab-scFv-Fc side) comprising a scFv variant L368D / a) a second monomer (Fab-Fc side) comprising a variable heavy chain domain constituting an Fv that binds to GPC3 as outlined herein together with the variable light chain domain of the common light chain domain, K370S, pI variants N208D / Q295E / N384D / Q418E / N421D, deletion variants E233P / L234V / L235A / G236del / S267K, and the variable light chain domain of the common light chain domain, and c) a common light chain comprising a variable light chain domain and a constant light chain domain, numbering according to EU numbering. In some embodiments, the VH / VL pair of αGPC3 is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69.

[0244] In some embodiments, the 1+1 Fab2-scFv-Fc format antibody comprises scFvians, pI variants, deletion variants and FcRn variants. Thus, some embodiments comprise a 2+1 Fab2-scFv-Fc format, which comprises a) a first monomer (Fab-scFv-Fc side) comprising a scFvian S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variants M428L / N434S, and a variable heavy chain domain constituting an Fv that binds to GPC3 as outlined herein, together with a variable light chain domain of the common light chain, and a scFv domain that binds to CD3; b) a first monomer (Fab-scFv-Fc side) comprising a scFvian S368D / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variants M428L / N434S, and a variable light chain domain of the common light chain; a) a second monomer (Fab-Fc side) comprising a variable heavy chain domain constituting an Fv that binds to GPC3 as outlined herein together with the variable light chain domain of the common light chain domain, K370S, pI variants N208D / Q295E / N384D / Q418E / N421D, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variant M428L / N434S, and the variable light chain domain of the common light chain domain, and c) a common light chain comprising a variable light chain domain and a constant light chain domain, numbering according to EU numbering. In some embodiments, the VH / VL pair of αGPC3 is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69. CD3 binding domain sequences that find particular use in these embodiments include, but are not limited to, H1.30_L1.47, H1.32_L1.47, H1.89_L1.47, H1.90_L1.47, H1.33_L1.47, H1.31_L1.47, L1.47_H1.30, L1.47_H1.30, L1.47_H1.32, L1.47_H1.89, L1.47_H1.90, L1.47_H1.33, and L1.47_H1.31, or variants thereof.

[0245] Figures 8A-8C show some exemplary Fc domain sequences useful in the 2+1 Fab2-scFv-Fc format. The "monomer 1" sequences shown in Figures 8A-8C generally refer to the Fc domain of the "Fab-Fc heavy chain," and the "monomer 2" sequences refer to the Fc domain of the "Fab-scFv-Fc heavy chain." Additionally, Figure 9 shows useful CL sequences that can be used in this format.

[0246] Exemplary anti-GPC3 x anti-CD3 2+1 Fab2-scFv-Fc format antibodies are shown in Figures 18-21.

[0247] 5. Central-Fv One heterodimeric scaffold that finds particular use with the antibodies described herein is the central Fv format (Figure 44I). In this embodiment, the format relies on the use of an inserted Fv domain (i.e., a central Fv domain) to form an "additional" third antigen-binding domain, where the Fab portions of the two monomers bind to GPC3 and the "additional" central Fv domain binds to CD3. The "additional" central Fv domain is inserted between the Fc domain and the CH1-Fv region of the monomer, thus providing a third antigen-binding domain (i.e., an "additional" central Fv domain), where each monomer comprises a component of the "additional" central Fv domain (i.e., one monomer comprises the variable heavy chain domain and the other comprises the variable light chain domain of the "additional" central Fv domain).

[0248] In this embodiment, one monomer comprises a first heavy chain comprising a first variable heavy domain, a CH1 domain, and an Fc domain, and an additional variable light domain. The light chain domain is covalently linked between the C-terminus of the CH1 domain of the heavy chain constant domain and the N-terminus of the first Fc domain using a domain linker (VH1-CH1-[optional linker]-VL2-hinge-CH2-CH3). The other monomer comprises a first heavy chain comprising a first variable heavy domain, a CH1 domain, and an Fc domain, and an additional variable heavy domain (VH1-CH1-[optional linker]-VH2-hinge-CH2-CH3). The light chain domain is covalently linked between the C-terminus of the CH1 domain of the heavy chain constant domain and the N-terminus of the first Fc domain using a domain linker.

[0249] This embodiment utilizes a common light chain comprising a variable light domain and a constant light domain, which associates with a heavy chain to form two identical Fabs, each of which binds to GPC3. For many of the embodiments described herein, these constructs include scubariants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0250] The antibodies described herein provide a central Fv format, in which the CD3 binding domain sequence is as shown in Figure 10 or a variant thereof, and the GPC3 binding domain sequence is as shown in Figures 12 and 13 or a variant thereof.

[0251] 6. One-arm central scFv One heterodimeric scaffold that finds particular use in the antibodies described herein is the one-arm central scFv format (Figure 44C). In this embodiment, one monomer contains only an Fc domain, and the other monomer contains a Fab domain (first antigen-binding domain), an scFv domain (second antigen-binding domain), and an Fc domain, where the scFv domain is inserted between the Fc domain and the Fc domain. In this format, the Fab portion binds to one receptor target and the scFv binds to another. In this format, the Fab portion binds to GPC3 and the scFv binds to CD3, or vice versa.

[0252] In this embodiment, one monomer comprises a first heavy chain comprising a first variable heavy domain, a CH1 domain and an Fc domain, and the scFv comprises an scFv variable light domain, an scFv linker and an scFv variable heavy domain. The scFv is covalently linked between the C-terminus of the CH1 domain of the heavy chain constant domain and the N-terminus of the first Fc domain using a domain linker in either of the following orientations: VH1-CH1-[optional domain linker]-VH2-scFv linker-VL2-[optional domain linker]-CH2-CH3 or VH1-CH1-[optional domain linker]-VL2-scFv linker-VH2-[optional domain linker]-CH2-CH3. The second monomer comprises an Fc domain (CH2-CH3). This embodiment further utilizes a light chain comprising a variable light domain and a constant light domain that associate with the heavy chain to form a Fab.

[0253] For many of the embodiments described herein, these constructs include sq variants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0254] The antibodies described herein provide a central Fv format, in which the CD3 binding domain sequence is as shown in Figure 10 or a variant thereof, and the GPC3 binding domain sequence is as shown in Figures 12 and 13 or a variant thereof.

[0255] Additionally, the Fc domain of the one-arm central scFv format can be modified with a scFv variant (e.g., the set of amino acid substitutions shown in FIG. 1, where particularly useful scFv variants are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K, and T366S / L368A / Y407V / Y349C:T366W / S354C), optionally a deletion variant (including those shown in FIG. 3), optionally a charged scFv linker (including those shown in FIG. 5), and the heavy chain comprises a pI variant (including those shown in FIG. 2).

[0256] In some embodiments, the one-arm central scFv format includes scV variants, pI variants, and deletion variants. Thus, some embodiments of the one-arm central scFv format comprise: a) a first monomer comprising a variable heavy chain domain constituting an Fv that binds GPC3 as outlined herein together with the scFv variants S364K / E357Q, the deletion variants E233P / L234V / L235A / G236del / S267K, and the variable light chain domain of the common light chain, and an scFv domain that binds CD3; b) a second monomer comprising an Fc domain with the scFv variants L368D / K370S, the pI variants N208D / Q295E / N384D / Q418E / N421D, the deletion variants E233P / L234V / L235A / G236del / S267K, and c) a light chain comprising a variable light chain domain and a constant light chain domain.

[0257] In some embodiments, the one-arm central scFv format comprises scFv variants, pI variants, deletion variants, and FcRn variants. Thus, some embodiments of the one-arm central scFv format comprise a) a scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variants M428L / N434S, and a variable heavy chain domain that constitutes an Fv that binds GPC3 as outlined herein together with a variable light chain domain of the common light chain, and a scFv that binds CD3. a) a first monomer comprising an Fv domain; b) a second monomer comprising an Fc domain having the sq variant L368D / K370S, the pI variant N208D / Q295E / N384D / Q418E / N421D, the deletion variant E233P / L234V / L235A / G236del / S267K, and the FcRn variant M428L / N434S; and c) a light chain comprising a variable light domain and a constant light domain.

[0258] 7. One-arm scFv-mAb One heterodimeric scaffold that finds particular use in the antibodies described herein is the one-arm scFv-mAb format (Figure 44D). In this embodiment, one monomer contains only the Fc domain, while the other monomer uses a scFv domain attached to the N-terminus of the heavy chain, generally through the use of a linker: VH-scFv linker-VL-[any domain linker]-CH1-hinge-CH2-CH3 or (opposite orientation) VL-scFv linker-VH-[any domain linker]-CH1-hinge-CH2-CH3. In this format, the Fab portion binds to GPC3, respectively, and the scFv binds to CD3. This embodiment further utilizes a light chain that includes a variable light domain and a constant light domain that associate with the heavy chain to form a Fab. For many of the embodiments described herein, these constructs include scFv variants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0259] The antibodies described herein provide a one-arm scFv-mAb format in which the CD3 binding domain sequence is as shown in Figure 10 or a variant thereof, and the GPC3 binding domain sequence is as shown in Figures 12 and 13 or a variant thereof.

[0260] In addition, the Fc domain of the one-arm scFv-mAb format may be modified with a scFv variant (e.g., the set of amino acid substitutions shown in FIG. 1, where particularly useful scFv variants are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K / E 5, and the heavy chain comprises a pI variant (including those shown in FIG. 2).

[0261] In some embodiments, the one-arm scFv-mAb format comprises a scFv variant, a pI variant, and a deletion variant. Thus, some embodiments of the one-arm scFv-mAb format comprise: a) a first monomer comprising a variable heavy chain domain constituting an Fv that binds GPC3 as outlined herein together with the scFv variants S364K / E357Q, the deletion variants E233P / L234V / L235A / G236del / S267K, and the variable light chain domain of the common light chain, and an scFv domain that binds CD3; b) a second monomer comprising an Fc domain with the scFv variants L368D / K370S, the pI variants N208D / Q295E / N384D / Q418E / N421D, the deletion variants E233P / L234V / L235A / G236del / S267K, and c) a light chain comprising the variable light chain domain and the constant light chain domain.

[0262] In some embodiments, the one-arm scFv-mAb format comprises scFv variants, pI variants, deletion variants, and FcRn variants. Thus, some embodiments of the one-arm scFv-mAb format include: a) scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variants M428L / N434S, and a variable heavy chain domain that constitutes an Fv that binds GPC3 as outlined herein together with a variable light chain domain of the common light chain, and an scFv that binds CD3. v domain; b) a second monomer comprising an Fc domain having the sq variant L368D / K370S, the pI variant N208D / Q295E / N384D / Q418E / N421D, the deletion variant E233P / L234V / L235A / G236del / S267K, and the FcRn variant M428L / N434S; and c) a light chain comprising a variable light domain and a constant light domain.

[0263] 8.scFv-mAb One heterodimeric scaffold that finds particular use with the antibodies described herein is the mAb-scFv format (Figure 44E). In this embodiment, the format relies on the use of N-terminal attachment of an scFv to one of the monomers, thereby forming a third antigen-binding domain, with the Fab portions of the two monomers binding to GPC3 and an "additional" scFv domain binding to CD3.

[0264] In this embodiment, the first monomer comprises a first heavy chain (comprising a variable heavy domain and a constant domain) with an N-terminal covalently linked scFv comprising an scFv variable light domain, an scFv linker and an scFv variable heavy domain, in either orientation ((VH1-scFv linker-VL1-[optional domain linker]-VH2-CH1-hinge-CH2-CH3) or (with an scFv in the opposite orientation) ((VL1-scFv linker-VH1-[optional domain linker]-VH2-CH1-hinge-CH2-CH3)). This embodiment further utilizes a common light chain comprising a variable light domain and a constant light domain, which associates with the heavy chain to form two identical Fabs that are linked to GPC3. For many of the embodiments described herein, these constructs include scavariant, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0265] The antibodies described herein provide an scFv-mAb format in which the CD3 binding domain sequence is as shown in Figure 10 or a variant thereof, and the GPC3 binding domain sequence is as shown in Figures 12 and 13 or a variant thereof.

[0266] In addition, the Fc domain of the scFv-mAb format may be modified with a scFv variant (e.g., the set of amino acid substitutions shown in FIG. 1 , where particularly useful scFv variants are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S ... 5, and the heavy chain comprises a pI variant (including those shown in FIG. 2).

[0267] In some embodiments, the scFv-mAb format includes scFv variants, pI variants, and deletion variants. Thus, some embodiments include an scFv-mAb format, which comprises a) a first monomer comprising the scFv variant S364K / E357Q, the deletion variants E233P / L234V / L235A / G236del / S267K, and a variable heavy chain domain constituting an Fv that binds GPC3 as outlined herein together with the variable light chain domain of the common light chain, and an scFv domain that binds CD3; a second monomer comprising a variable heavy chain domain constituting an Fv that binds to GPC3 as outlined herein together with the variable light chain domain of the cuvariant L368D / K370S, the pI variants N208D / Q295E / N384D / Q418E / N421D, the deletion variants E233P / L234V / L235A / G236del / S267K, and the variable light chain domain of the common light chain, and c) a common light chain comprising a variable light chain domain and a constant light chain domain.

[0268] In some embodiments, the scFv-mAb format includes scFv variants, pI variants, deletion variants, and FcRn variants. Thus, some embodiments include an scFv-mAb format, which comprises: a) a first monomer comprising a scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variants M428L / N434S, and a variable heavy chain domain constituting an Fv that binds GPC3 as outlined herein together with a variable light chain domain of the common light chain, and an scFv domain that binds CD3; b) a) a variable heavy chain domain constituting an Fv that binds to GPC3 as outlined herein together with the variable light chain of the scubariant L368D / K370S, the pI variants N208D / Q295E / N384D / Q418E / N421D, the deletion variants E233P / L234V / L235A / G236del / S267K, the FcRn variant M428L / N434S, and the common light chain, and c) a common light chain comprising a variable light chain domain and a constant light chain domain.

[0269] 9. Dual scFv format The antibodies described herein also provide a dual scFv format (Figure 44B), as known in the art. In this embodiment, the GPC3 x CD3 heterodimeric bispecific antibody is composed of two scFv-Fc monomers, either both in the (VH-scFv linker-VL-[optional domain linker]-CH2-CH3) format or in the (VL-scFv linker-VH-[optional domain linker]-CH2-CH3) format, or with one monomer in one orientation and the other in the other orientation.

[0270] The antibodies described herein provide a dual scFv format, in which the CD3 binding domain sequence is as shown in Figure 10 or a variant thereof, and the GPC3 binding domain sequence is as shown in Figures 12 and 13 or a variant thereof.

[0271] In some embodiments, the dual scFv format comprises a scFv variant, a pI variant, and a deletion variant. Thus, some embodiments include a dual scFv format comprising a) a first monomer comprising a scFv variant S364K / E357Q, a deletion variant E233P / L234V / L235A / G236del / S267K, and a first scFv that binds to either CD3 or GPC3, and b) a second monomer comprising a scFv variant L368D / K370S, a pI variant N208D / Q295E / N384D / Q418E / N421D, and a deletion variant.

[0272] E233P / L234V / L235A / G236del / S267K, and a second scFv that binds either CD3 or GPC3. In some embodiments, the dual scFv format comprises a scFv variant, a pI variant, a deletion variant, and an FcRn variant. In some embodiments, the dual scFv format comprises a scFv variant, a pI variant, and a deletion variant. Thus, some embodiments include a dual scFv format comprising: a) a first monomer comprising the scFv variant S364K / E357Q, the deletion variants E233P / L234V / L235A / G236del / S267K, the FcRn variants M428L / N434S, and a first scFv that binds either CD3 or GPC3; and b) a second monomer comprising the scFv variant L368D / K370S, the pI variants N208D / Q295E / N384D / Q418E / N421D, the deletion variants E233P / L234V / L235A / G236del / S267K, the FcRn variants M428L / N434S, and a second scFv that binds either CD3 or GPC3.

[0273] 10. Non-heterodimeric bispecific antibodies As will be understood by those skilled in the art, the anti-GPC3 x anti-CD3 antibodies provided herein can also be included in a non-heterodimeric bispecific format (see Figure 44J). In this format, the anti-GPC3 x anti-CD3 comprises 1) a first monomer comprising VH1-CH1-hinge-CH2-CH3, 2) a second monomer comprising VH2-CH1-hinge-CH2-CH3, 3) a first light chain comprising VL1-CL, and 4) a second light chain comprising VL2-CL. In such an embodiment, VH1 and VL1 form a first antigen-binding domain, and VH2 and VL2 form a second antigen-binding domain. One of the first antigen-binding domain or the second antigen-binding domain binds to GPC3, and the other antigen-binding domain binds to CD3.

[0274] Any suitable GPC3 binding domain and CD3 binding domain may be included in the anti-GPC3 x anti-CD3 antibody in a non-heterodimeric bispecific antibody format, including the GPC3 binding domain and CD3 binding domain provided herein and associated VH and VL or variants thereof (see, e.g., Figures 10, 12, and 13).

[0275] 11. Trident Format In some embodiments, the bispecific antibodies described herein are in the "Trident" format generally as described in WO2015 / 184203, the entire contents of which are incorporated herein by reference, in particular the figures, illustrations, definitions, and sequences of the "Heterodimer-Promoting Domains" or "HPDs" including the "K-coil" and "E-coil" sequences. Trident relies on the use of two different HPDs as structural building blocks that associate to form a heterodimeric structure. In this embodiment, the Trident format comprises a "conventional" heavy and light chain (e.g., VH1-CH1-hinge-CH2-CH3 and VL1-CL), a third chain comprising a first "Diabody-type Linking Domain" or "DART®", VH2-(linker)-VL3-HPD1, and a fourth chain comprising a second DART®, VH3-(linker)-(linker)-VL2-HPD2. VH1 and VL1 form a first ABD, VH2 and VL2 form a second ABD, and VH3 and VL3 form a third ABD. As shown in Figure 1K, in some cases, the second and third ABDs link, for example bivalently, to the same antigen, in this example typically GPC3, and the first ABD links monovalently to CD3.

[0276] Any suitable GPC3 binding domain and CD3 binding domain may be included in the anti-GPC3 x anti-CD3 antibody in the Trident bispecific antibody format, including any of the GPC3 binding domains and CD3 binding domains provided herein and associated VH and VL or variants thereof (see, e.g., Figures 10, 12, and 13).

[0277] 12. Monospecific, monoclonal antibodies As will be appreciated by those skilled in the art, the novel Fv sequences outlined herein can also be used in both monospecific antibodies (e.g., "traditional monoclonal antibodies") or non-hetero-bispecific formats. Thus, in some embodiments, the antibodies described herein provide monoclonal (monospecific) antibodies comprising the six CDRs and / or vh and vl sequences from the figures, generally with IgG1, IgG2, IgG3, or IgG4 constant regions, IgG1, IgG2, and IgG4 (including the IgG4 constant region containing the amino acid substitution S228P), and find particular use in some embodiments. That is, any sequence designated herein as "H_L" can be linked to the constant region of a human IgG1 antibody.

[0278] In some embodiments, the monospecific antibody is selected from the group consisting of [GPC3]H1_L1, [GPC3]H1_L1.1, [GPC3]H1_L1.2, [GPC3]H1_L1.3, [GPC3]H1_L1.4, [GPC3]H1_L1.5, [GPC3]H1_L1.6, [GPC3]H1_L1.7, [GPC3]H1_L1.8, [GPC3]H1_L1.9, [GPC3]H1_L1.10, [GPC3]H1_L1.16, [GPC3]H1_L1.23, [GPC3]H1_L1.29, [GPC3]H1_L1.31, [GPC3]H1_L1.65, [GPC3]H1_L1.66, [GPC3]H1_L1.67, [GPC3]H1_L1.68, [GPC3]H1_L1.70, [GPC3]H1_L1.71, [GPC3]H1_L1.72, [GPC3]H1_L1.73, [GPC3]H1.1_L1, [GPC3 ]H1.1_L1.1, [GPC3]H1.1_L1.2, [GPC3]H1.1_L1.3, [GPC3]H1.1_L1.4, [GP C3]H1.1_L1.5, [GPC3]H1.1_L1.6, [GPC3]H1.1_L1.7, [GPC3]H1.1_L1.8, [G PC3]H1.1_L1.9, [GPC3]H1.1_L1.10, [GPC3]H1.1_L1.16, [GPC3]H1.1_L1.23, [GPC3]H1.1_L1.29, [GPC3]H1.1_L1.31, [GPC3]H1.1_L1.65, [GPC3]H1 .1_L1.66, L1.67, [GPC3]H1.1_L1.68, [GPC3]H1.1_L1.70, [GPC3]H1.1_L1 .71, [GPC3]H1.1_L1.72, [GPC3]H1.1_L1.73, [GPC3]H1.2_L1, [GPC3]H1.2_ L1.1, [GPC3]H1.2_L1.2, [GPC3]H1.2_L1.3, [GPC3]H1.2_L1.4, [GPC3]H1.2_L1.5, [GPC3]H1.2_L1.6, [GPC3]H1.2_L1.7, [GPC3]H1.2_L1.8, [GPC3]H 1.2_L1.9, [GPC3]H1.2_L1.10, [GPC3]H1.2_L1.16, [GPC3]H1.2_L1.23, [G PC3]H1.2_L1.29, [GPC3]H1.2_L1.31, [GPC3]H1.2_L1.65, [GPC3]H1.2_L1.66, [GPC3]H1.2_L1.67, [GPC3]H1.2_L1.68, [GPC3]H1.2_L1.70, [GPC3]H1.2_L1.71, [GPC3]H1.2_L1.72, [GPC3]H1.2_L1.73, [GPC3]H1 .3_L1, [GPC3]H1.3_L1.1, [GPC3]H1.3_L1.2, [GPC3]H1.3_L1.3, [GPC3]H1.3_L1.4, [GPC3]H1.3_L1.5, [GPC3]H1.3_L1.6, [GPC3]H1.3_L 1.7, [GPC3]H1.3_L1.8, [GPC3]H1.3_L1.9, [GPC3]H1.3_L1.10, [GPC3]H1.3_L1.16, [GPC3]H1.3_L1.23, [GPC3]H1.3_L1.29, [GPC3]H1.3 _L1.31, [GPC3]H1.3_L1.65, [GPC3]H1.3_L1.66, [GPC3]H1.3_L1.67, [GPC3]H1.3_L1.68, [GPC3]H1.3_L1.70, [GPC3]H1.3_L1.71, [GPC3 ]H1.3_L1.72, [GPC3]H1.3_L1.73, [GPC3]H1.4_L1, [GPC3]H1.4_L1.1, [GPC3]H1.4_L1.2, [GPC3]H1.4_L1.3, [GPC3]H1.4_L1.4, [GPC3]H 1.4_L1.5, [GPC3]H1.4_L1.6, [GPC3]H1.4_L1.7, [GPC3]H1.4_L1.8, [GPC3]H1.4_L1.9, [GPC3]H1.4_L1.10, [GPC3]H1.4_L1.16, [GPC3]H and [GPC3]H1.4_L1.70, [GPC3]H1.4_L1.71, [GPC3]H1.4_L1.72, and [GPC3]H1.4_L1.73, or variants thereof.

[0279] In certain embodiments, the VH / VL pair is selected from the group consisting of [GPC3]H1.1_L1.16 and [GPC3]H1.1_L1.69, or a variant thereof.

[0280] H. Specific Embodiments of the Invention The invention specifically provides 1+1 and 2+1 formats that bind to CD3 and GPC3.Specific embodiments include XENP38086 (Xtend analog to XENP34920, meaning that this Xtend analog has the same sequence except for containing 428L / 434S on each Fc domain), XENP38087 (Xtend analog to XENP36935, supra), and XENP38232 (Xtend analog to XENP37625, supra).

[0281] 1.1+1 format In a particular 1+1 format embodiment, the αGPC3 ABD is a Fab and has the VH / VL pair [GPC3]H1.1_L1.16, and the αCD3 ABD is anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 H1.31_L1.47, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 The scFv is selected from the group consisting of L1.47_H1.31.

[0282] In a particular 1+1 format embodiment, the αGPC3 ABD is a Fab and has the VH / VL pair [GPC3]H1.1_L1.69, and the αCD3 ABD is anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 H1.31_L1.47, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 The scFv is selected from the group consisting of L1.47_H1.31.

[0283] Specific embodiments that find use in some applications include those listed in Figures 16 and 17 and include, but are not limited to, XENP21971, XENP34364, XENP34365, XENP34367, XENP34368, XENP35843, XENP36140, XENP36931, XENP36932, XENP36933, XENP36934, XENP36935, XENP36936, XENP36937, XENP36938, XENP36939, XENP36941, and XENP38087.

[0284] 2.2+1 format In a particular embodiment of the 2+1 format, the αGPC3 ABD is a Fab and has the VH / VL pair [GPC3]H1.1_L1.16, and the αCD3 ABD is anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 H1.31_L1.47, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 The scFv is selected from the group consisting of L1.47_H1.31.

[0285] In a particular embodiment of the 2+1 format, the αGPC3 ABD is a Fab and has the VH / VL pair [GPC3]H1.1_L1.69, and the αCD3 ABD is anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.47, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, anti-CD3 H1.31_L1.47, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.30, anti-CD3 L1.47_H1.32, anti-CD3 L1.47_H1.89, anti-CD3 L1.47_H1.90, anti-CD3 L1.47_H1.33, and anti-CD3 The scFv is selected from the group consisting of L1.47_H1.31.

[0286] Specific embodiments that find use in some applications include those listed in Figures 18, 19, 20, and 21 and include, but are not limited to, XENP33744, XENP33745, XENP27259, XENP33746, XENP34919, XENP34920, XENP34921, XENP34922, XENP34923, XENP35840, XENP35840, XENP37246, XENP37247, XENP38086, XENP33747, XENP35841, XENP37624, XENP37625, XENP37626, XENP38232, XENP37430, XENP37433, and XENP33748.

[0287] II. Nucleic Acids of the Invention The present disclosure further provides nucleic acid compositions encoding the anti-GPC3 antibodies provided herein, including, but not limited to, anti-GPC3 x anti-CD3 bispecific antibodies and GPC3 monospecific antibodies.

[0288] As will be appreciated by those skilled in the art, the nucleic acid composition will depend on the format and scaffold of the heterodimeric protein. Thus, for example, if three amino acid sequences are required for the format, such as a 1+1 Fab-scFv-F format (e.g., a first amino acid monomer comprising an Fc domain and scFv, a second amino acid monomer comprising a heavy chain and a light chain), the three nucleic acid sequences can be incorporated into one or more expression vectors for expression. Similarly, in some formats (e.g., a dual scFv format as disclosed in Figure 44), only two nucleic acids are required, which can also be incorporated into one or more expression vectors.

[0289] As is known in the art, the nucleic acids encoding the components of the antibodies described herein can be incorporated into expression vectors, as is known in the art and depending on the host cell used to produce the heterodimeric antibodies described herein. In general, the nucleic acids are operably linked to any number of regulatory elements (promoter, origin of replication, selectable marker, ribosome binding site, inducer, etc.). Expression vectors can be extrachromosomal or integrating vectors.

[0290] The nucleic acids and / or expression vectors described herein are then transformed into any number of different types of host cells known in the art, including mammalian cells, bacterial cells, yeast cells, insect cells and / or fungal cells, although in many embodiments mammalian cells (e.g., CHO cells) are used.

[0291] In some embodiments, the nucleic acids encoding each monomer and, depending on the format, any nucleic acid encoding the light chain are each contained in a single expression vector, generally under different or the same promoter control. In an embodiment specifically used in the antibodies described herein, each of these two or three nucleic acids is contained in a different expression vector. As shown herein and in 62 / 025,931, which is incorporated herein by reference, different vector ratios can be used to promote heterodimer formation. That is, surprisingly, (in many of the embodiments herein having three polypeptides, including heterodimeric antibodies) the protein contains a 1:1:2 ratio of first monomer:second monomer:light chain, which is not the ratio that gives the best results.

[0292] The heterodimeric antibodies described herein are produced by culturing host cells containing expression vectors, as is well known in the art. Once produced, conventional antibody purification steps are performed, including an ion exchange chromatography step. As discussed herein, the pI of the two monomers differs by at least 0.5, which may allow for separation by ion exchange chromatography or isoelectric focusing, or other methods sensitive to isoelectric point. That is, the inclusion of pI substitutions that change the isoelectric point (pI) of each monomer such that the pI of each monomer has a different pI, and the heterodimer also has a different pI, facilitates isoelectric purification (e.g., anion exchange column, cation exchange column) of the "1+1 Fab-scFv-Fc" and "2+1" heterodimers. These substitutions also aid in the determination and monitoring of contaminating dual scFv-Fc and mAb homodimers after purification (e.g., IEF gels, cIEF, and analytical IEX columns).

[0293] III. Biological and Biochemical Functions of Heterodimeric Bispecific Antibodies In general, the bispecific GPC3xCD3 antibodies described herein are administered to patients with cancer and efficacy is assessed in several ways as described herein. Thus, standard assays of efficacy such as assessment of cancer burden, tumor size, presence or extent of metastases can be performed, but immuno-oncological treatments can also be evaluated based on immune status assessment. This can be done in several ways, including both in vitro and in vivo assays.

[0294] IV.Treatment Once generated, the antibody compositions described herein find use in a number of applications, including cancers such as liver cancer, and thus the antibody heterodimeric compositions described herein find use in the treatment of such GPC3-positive cancers.

[0295] V. Antibody Compositions for In Vivo Administration Formulations of antibodies for use in accordance with the methods described herein are prepared in the form of a lyophilized formulation or aqueous solution by mixing the antibody of the desired purity with any pharma- ceutically acceptable carrier, excipient, or stabilizer (Remington's Pharmaceutical Sciences 16th edition, Osol, A. Ed.

[1980] ).

[0296] VI. Administration Modalities The antibodies and chemotherapeutic agents described herein are administered to a subject according to known methods, such as intravenous administration as a bolus or by continuous infusion over a period of time.

[0297] VII. Treatment Modalities In the method of the present invention, treatment is used to provide a positive therapeutic response for a disease or condition. "Positive therapeutic response" refers to the improvement of a disease or condition and / or the improvement of symptoms associated with the disease or condition. For example, a positive therapeutic response can refer to one or more of the following improvements in disease: (1) a decrease in the number of tumor cells; (2) an increase in tumor cell death; (3) inhibition of tumor cell survival; (5) inhibition (i.e., some slowing down, preferably stopping) of tumor growth; (6) an increase in patient survival rate; and (7) some relief from one or more symptoms associated with the disease or condition.

[0298] Positive therapeutic response in any given disease or condition can be determined by standardized response criteria specific to that disease or condition. Tumor response can be evaluated for changes in tumor morphology (i.e., tumor burden, tumor size, etc.) using screening techniques such as magnetic resonance imaging (MRI) scan, X-ray, computed tomography (CT) scan, bone scan, endoscopy, and tumor biopsy sampling, including bone marrow aspiration (BMA) and counting of circulating tumor cells.

[0299] In addition to these positive therapeutic responses, subjects undergoing treatment may experience the beneficial effect of amelioration of symptoms associated with the disease.

[0300] Treatment according to the present disclosure includes a "therapeutically effective amount" of the pharmaceutical agent used. A "therapeutically effective amount" refers to an amount effective, at dosages and for periods of time necessary, to achieve the desired therapeutic result.

[0301] A therapeutically effective amount can vary depending on factors such as the disease state, age, sex, and weight of the individual, as well as the ability of the agent to elicit a desired response in the individual. A therapeutically effective amount is also one in which any toxic or detrimental effects of the antibody or antibody portion are outweighed by the therapeutically beneficial effects.

[0302] A "therapeutically effective amount" for tumor therapy can also be measured by its ability to stabilize disease progression. The ability of a compound to inhibit cancer may be evaluated in an animal model system predictive of efficacy in human tumors.

[0303] Alternatively, this property of the composition may be evaluated by testing the ability of the compound to inhibit cell proliferation or induce apoptosis by in vitro assays known to those of skill in the art. A therapeutically effective amount of a therapeutic compound may reduce tumor size or otherwise alleviate symptoms in a subject. One of skill in the art would be able to determine such an amount based on factors such as the size of the subject, the severity of the subject's symptoms, and the particular composition or route of administration selected.

[0304] The dosage regimen is adjusted to provide the optimum desired response (e.g., therapeutic response). For example, a single bolus may be administered, several divided doses may be administered over time, or the dose may be proportionally increased or decreased as indicated by the exigencies of the therapeutic situation. Parenteral compositions may be formulated in unit dosage form for ease of administration and uniformity of dosage. Unit dosage form as used herein refers to a physically discrete unit suitable as a single dosage for the subject to be treated. Each unit contains a predetermined amount of active compound calculated to produce the desired therapeutic effect in association with the required pharmaceutical carrier.

[0305] The specifications for the unit dosage forms of the present disclosure are influenced by and directly dependent on (a) the unique characteristics of the active compound and the particular therapeutic effect to be achieved, and (b) the limitations inherent in the art of formulating such active compounds for the treatment of susceptibility in individuals.

[0306] The effective dosage and dosing regimen of the bispecific antibodies described herein depends on the disease or condition being treated and can be determined by one skilled in the art.

[0307] An exemplary, non-limiting range for a therapeutically effective amount of a bispecific antibody used in the antibodies described herein is about 0.1-100 mg / kg.

[0308] All cited references are expressly incorporated by reference herein in their entirety.

[0309] While particular embodiments of the present disclosure have been described above for purposes of illustration, those skilled in the art will appreciate that many changes in detail may be made without departing from the invention as set forth in the appended claims. EXAMPLES

[0310] A. Example 1: Engineering of an αGPC3×αCD3 Bispecific Antibody Several formats of αGPC3×αCD3 bispecific antibodies (bsAbs) have been devised, exemplary formats of which are outlined below and in FIG. 15.

[0311] One such format is the 1+1 Fab-scFv-Fc format, which comprises a single-chain Fv ("scFv") covalently linked to a first heterodimeric Fc domain, a heavy chain variable region (VH) covalently linked to a complementary second heterodimeric Fc domain, and a light chain (LC) transfected separately such that a Fab domain is formed with the variable heavy domain.

[0312] Another format is the 2+1 Fab-scFv-Fc format, which comprises a VH domain covalently linked to CH1, which is covalently linked to a scFv, which is covalently linked to a first heterodimeric Fc domain (VH-CH1-scFv-Fc), the VH domain covalently linked to a complementary second heterodimeric Fc domain, and a LC transfected separately such that a Fab domain is formed with the VH domain.

[0313] The DNA coding strand for the αGPC3×αCD3 bsAb was generated by standard gene synthesis followed by isothermal cloning (Gibson assembly) or subcloning into pTT5 expression vectors containing fusion partners (e.g., domain linkers shown in FIG. 6 and / or scaffolds shown in FIG. 7-8). DNA was transfected into HEK293E cells for expression. Exemplary αGPC3×αCD3 bsAb sequences in 1+1 Fab-scFv-Fc and 2+1 Fab2-scFv-Fc formats are shown in FIG. 16-21, respectively.

[0314] B. Example 2: Engineering the GPC3-binding domain 2A: Humanization of the mouse GPC3 binding domain A mouse clone that binds to the C-terminal peptide of GPC3 (amino acids 524-563) of the GPC3 protein, located near the cell membrane, was humanized using string content optimization (see, e.g., U.S. Patent No. 7,657,380, issued Feb. 2, 2010). The sequences of the humanized variants are designated GPC3-A and are shown in FIG. 12. The variants have reduced propensity for degradation (e.g., aspartate isomerization and deamidation) and modulated GPC3 binding affinity and / or selectivity for high GPC3 expressing cell lines. Sequences for exemplary such variants are shown in FIG. 13.

[0315] 2B: Modulation of GPC3 binding affinity After humanization, GPC3 binding arms were engineered with single point mutations in the variable light domain to generate a ladder of GPC3 binding affinities. An exemplary sequence is shown in FIG. 13. 73 variable light domain variants were engineered and paired with GPC3-A H1.1 (see Example 3C for description of GPC3-A H1.1 variants). GPC3-A variants were engineered in a 1+1 Fab-scFv-Fc bsAb format and GPC3 antigen binding was measured using Octet, a biolayer interferometry (BLI)-based method. Octet experimental steps generally include immobilization (capture of ligand on biosensor), association (immersion of ligand-coated biosensor into well containing analyte), and dissociation (returning biosensor to well containing buffer). To facilitate clinical development (e.g., by investigating therapeutics in model animals), it is useful for the binding domain to be cross-reactive to cynomolgus antigens, and therefore binding to both human and cynomolgus GPC3 antigens was investigated. His-tagged human GPC3 and cynomolgus GPC3 were captured on a HIS1K sensor and then soaked in multiple concentrations of the bispecific antibody. The resulting dissociation constants (K D ) are shown in Figure 23. A wide range of KD values ​​was obtained, from the high affinity variant H1.1_L1.6 (4 nM) to the medium affinity variants H1.1_L1.16 and H1.1_L1.29 (100 nM and 70 nM, respectively) to the low affinity variant H1.1_L1.31 (400 nM).

[0316] Next, the binding of the variants to cell surface GPC3 was investigated. Exemplary affinity variants H1.1_L1.29, H1.1_L1.16, and H1.1_L1.31 were engineered in a 1+1 Fab-scFv-Fc format. bsAbs XENP36941 (L1.29, 70 nM GPC3 affinity), XENP35843 (L1.16, 100 nM GPC3 affinity), and XENP36140 (L1.31, 400 nM GPC3 affinity) were bound to the indicated concentrations of GPC3. highThe antibodies were incubated with HepG2 cells for 1 hour at 4° C. The cells were then washed and stained with a secondary antibody (typically anti-human Fc AlexaFlour647) for 1 hour at 4° C. After two further washes, the cells were analyzed by flow cytometry. The resulting binding curves shown in FIG. 24 indicate that cell binding correlates with GPC3 affinity, with XENP36941 (70 nM variable light domain) showing the strongest binding and XENP36140 (400 nM variable light domain) showing the weakest binding.

[0317] 2C: Procedures for removing residues prone to degradation The sequence of the humanized GPC3-A antibody was examined for residues prone to degradation. The GPC3-A H1 heavy chain CDR2 contained D52 / P52a (Kabat numbering) as an aspartate isomerization motif. In addition, the L1 variable light chain domain contained N28 / G29 (Kabat numbering) as a deamidation motif. Therefore, to investigate whether the tendency could be eliminated without affecting GPC3 binding, a library was generated with mutations at these residues. The binding data is shown in Figure 25. The D52E mutation in the variable heavy chain variant H1.1 eliminated the aspartate isomerization tendency while maintaining similar GPC3 binding (compared to H1) (binding of D52N was reduced, whereas D52Q and D52S were tolerated). G29T in the variable light chain variant L1.6 removed the deamidation tendency (compared to L1) while maintaining similar GPC3 binding (N28Q, N28Y, N28S, and N28H significantly reduced binding, while G29A and G29K were tolerated). It should be noted that different mutations may be required to remove the deamidation tendency from affinity variants. Binding data for exemplary variants is shown in FIG. 26. The G29A mutation was used in the L1.69 variant to remove the deamidation tendency from the L1.29 variant (the N28Y mutation reduced binding, but G29K was also tolerated), and the G29K mutation was used in the L1.73 variant to remove the deamidation tendency from the L1.31 variant (the N28Y and G29A mutations resulted in low or aberrant responses). The sequences of the variants described here are shown in FIG. 13.

[0318] C. Example 3: Preparation and optimization of αGPC3×αCD3 bsAb αGPC3×αCD3 bsAbs were engineered with various affinity-tuned GPC3 and CD3 binding domains, as well as with different GPC3 binding valencies, and generated as generally described in Example 1 to optimize redirected T cell cytotoxicity (RTCC), selectivity, and potential therapeutic index.

[0319] We used the RTCC assay to investigate the potential of the αGPC3 × αCD3 bispecific antibody (bsAb) to redirect CD3+ effector T cells to destroy GPC3-expressing cell lines. high Target cells were HepG2 cells (hepatic hepatocellular carcinoma line) and GPC3 med Huh7 cells (also a hepatocellular carcinoma line) and / or GPC3 as target cells low The assay was performed using HEK293 cells (a surrogate for cells outside the tumor environment) as target cells. Two methods of redirected T cell cytotoxicity (RTCC) assay were used: a flow cytometry-based method and a lactate dehydrogenase (LDH)-based method. For the flow cytometry-based RTCC assay, target cells were incubated with human PBMCs and test substances at the indicated effector-to-target cell ratios at 37°C. After incubation, cells were stained with Aqua Zombie dye for 15 min at room temperature. Cells were then washed, stained with antibodies for cell surface markers, and analyzed by flow cytometry. Induction of RTCC was determined using Zombie Aqua staining of CSFE+ target cells, and T cell activation and degranulation were determined by expression of CD107a, CD25, and CD69 on lymphocytes. For the LDH-based RTCC assay, cells and the indicated concentrations of test substances are incubated in flat-bottom 96-well plates in a total volume of 200 μl at 37° C. for 48 or 72 h. Cells were then lysed and mixed with substrate using the Promega CytoTox-one kit according to its protocol. Plates were read using a Wallac instrument. Cytokines were measured using the V-PLEX Inflammatory Panel 1 Human Kit. It should be noted that some data sets are from the same experiment, since several engineering approaches were investigated simultaneously.

[0320] Throughout this section, the αGPC3×αCD3 bsAb disclosed in WO 2016 / 047722 and the comparator bsAb XENP31308 based on the sequence shown in FIG. 22 were also used to benchmark the novel bsAbs of the present invention. XENP31308 was characterized in the RTCC assay as generally described above. Data shown in FIG. 27 show that at an effector:target ratio of 10:1, XENP31308 induced strong RTCC and cytokine secretion in both high and low GPC3 expressing cell lines. In another experiment using the 1:1 effector:target ratio data shown in FIG. 28, XENP31308 barely induces RTCC in the presence of the low GPC3 expressing cell line HEK293, but still strongly induces cytokine secretion. GPC3 low Since the cell lines are surrogates for healthy tissue, maintaining low or no death and cytokine release in the presence of low GPC3 expressing cell lines is important as this may lead to reduced toxicity and cytokine storm in the clinical setting. low The potent killing of cell lines indicates the possibility of on-target off-tissue killing in vivo, which may cause unwanted toxicity.

[0321] 3A: Investigating the effect of GPC3 binding domains binding to different GPC3 epitopes GPC3 is membrane bound and consists of an α subunit and a β subunit. The α subunit (residues 25-358) can be cleaved and released as soluble GPC3, while the β subunit remains attached to the cell membrane. As described in Example 2A, the GPC3 binding domain GPC3-A binds to the C-terminal residues 524-563 of GPC3, which is part of the β subunit. Additional GPC3 binding domains have been described that bind to the N-terminal residues 359-524 of the β or α subunit. To investigate the effect of binding to different GPC3 epitopes, αGPC3×αCD3 bsAbs in a 1+1 format with CD3 High or a 2+1 format with CD3 High-Int#1 scFv were engineered with different GPC3 binding domains that bind to different GPC3 epitopes. RTCC assays were performed with bsAbs and HepG2 cells (effector:target ratio of 10:1). The data in Figure 29 (showing activity of the bsAbs as indicated by induction of IFNγ release) indicate that each of the additional epitopes was less potent than GPC3-A in inducing T cell activity. Some of the difference may be due to differences in binding affinity (between clones) or avidity (between formats), but it is more likely that the GPC3 epitope contributes to the reduced potency.

[0322] 3B: Investigating the effect of GPC3 binding strength It was hypothesized that a bsAb in 2+1 Fab2-scFv-Fc format with bivalent binding to GPC3 may be useful to achieve higher potency with a lower affinity binding domain due to the improved avidity from the two binding domains. The avidity benefit of this format may confer selectivity to cell lines with higher target expression and reduce the possibility of on-target off-tissue (e.g., healthy tissues expressing low levels of GPC3) effects in vivo. Furthermore, high levels of soluble GPC3 may be present in vivo and may act as an antigen sink for GPC3-targeted drugs. The lower GPC3 binding affinity, coupled with bivalent GPC3 binding, may help overcome the sink by binding to cells preferably at a higher a...

Claims

1. A composition comprising a GPC3 antigen-binding domain (ABD), wherein the GPC3 antigen-binding domain is 1) a variable heavy domain (VH) having an amino acid sequence selected from H1.1 (SEQ ID NO: 181), H1 (SEQ ID NO: 173), H1.2 (SEQ ID NO: 185), H1.3 (SEQ ID NO: 189), and H1.4 (SEQ ID NO: 193); 2) L1.16 (SEQ ID NO: 237), L1 (SEQ ID NO: 177), L1.1 (SEQ ID NO: 197), L1.2 (SEQ ID NO: 201), L1.3 (SEQ ID NO: 205), L1.4 (SEQ ID NO: 209), L1.5 (SEQ ID NO: 213), L1.6 (SEQ ID NO: 217), L1.7 (SEQ ID NO: 221), L1.8 (SEQ ID NO: 225), L1.9 (SEQ ID NO: 229), L1.10 (SEQ ID NO: 233), L1.23 (SEQ ID NO: 241), L1.29 ( and a variable light domain (VL) having an amino acid sequence selected from L1.31 (SEQ ID NO:245), L1.31 (SEQ ID NO:249), L1.65 (SEQ ID NO:253), L1.66 (SEQ ID NO:257), L1.67 (SEQ ID NO:261), L1.68 (SEQ ID NO:265), L1.69 (SEQ ID NO:269), L1.70 (SEQ ID NO:273), L1.71 (SEQ ID NO:277), L1.72 (SEQ ID NO:281), and L1.73 (SEQ ID NO:285); A composition comprising:

2. a) a variable light chain domain having the amino acid sequence of SEQ ID NO: 237 and a variable heavy chain domain having the amino acid sequence of SEQ ID NO: 181; b) a variable light domain having the amino acid sequence of SEQ ID NO: 177 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; c) a variable light domain having the amino acid sequence of SEQ ID NO: 177 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; d) a variable light domain having the amino acid sequence of SEQ ID NO: 177 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 185; e) a variable light domain having the amino acid sequence of SEQ ID NO: 177 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 189; f) a variable light domain having the amino acid sequence of SEQ ID NO: 177 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 193; g) a variable light domain having the amino acid sequence of SEQ ID NO: 197 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; h) a variable light domain having the amino acid sequence of SEQ ID NO: 201 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; i) a variable light domain having the amino acid sequence of SEQ ID NO: 205 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; j) a variable light domain having the amino acid sequence of SEQ ID NO: 209 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; k) a variable light domain having the amino acid sequence of SEQ ID NO: 213 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; l) a variable light domain having the amino acid sequence of SEQ ID NO: 217 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; m) a variable light domain having the amino acid sequence of SEQ ID NO: 221 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; n) a variable light domain having the amino acid sequence of SEQ ID NO: 225 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; o) a variable light domain having the amino acid sequence of SEQ ID NO: 229 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; p) a variable light domain having the amino acid sequence of SEQ ID NO: 233 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 173; q) a variable light domain having the amino acid sequence of SEQ ID NO: 249 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; r) a variable light domain having the amino acid sequence of SEQ ID NO: 241 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; s) a variable light domain having the amino acid sequence of SEQ ID NO: 245 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; t) a variable light domain having the amino acid sequence of SEQ ID NO: 253 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; u) a variable light domain having the amino acid sequence of SEQ ID NO: 257 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; v) a variable light domain having the amino acid sequence of SEQ ID NO: 261 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; w) a variable light domain having the amino acid sequence of SEQ ID NO: 265 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; x) a variable light domain having the amino acid sequence of SEQ ID NO: 269 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; y) a variable light domain having the amino acid sequence of SEQ ID NO: 273 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; z) a variable light domain having the amino acid sequence of SEQ ID NO: 277 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; a1) a variable light chain domain having the amino acid sequence of SEQ ID NO: 281 and a variable heavy chain domain having the amino acid sequence of SEQ ID NO: 181, and a2) a variable light domain having the amino acid sequence of SEQ ID NO: 285 and a variable heavy domain having the amino acid sequence of SEQ ID NO: 181; A composition comprising a GPC3 antigen-binding domain comprising a variable light chain domain and a variable heavy chain domain selected from:

Images