Heterodimeric antibody that binds to claudin 18.2 and CD3

Abstract

The present invention is directed to antibodies, including novel antigen-binding domains and heterodimeric antibodies, that bind to claudin 18.2 (CLDN18.2).

Description

[Technical field]

[0001] Priority This application claims priority to U.S. Provisional Patent Application No. 63 / 210,787, filed June 15, 2021, which is incorporated by reference in its entirety for all purposes.

[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 Jun. 13, 2022, is named 067461-5232-WO_SL.txt and is 1,371,138 bytes in size. [Background technology]

[0003] Antibody-based therapeutics have been successful in treating 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 production of Fvs that recognize virtually any molecule, a typical approach to bispecific antibody generation is to introduce a new variable region into an antibody.

[0004] A particularly useful bispecific antibody approach is to engineer a first binding domain that binds to CD3 and a second binding domain that binds to an antigen that is associated with or upregulated on cancer cells, such that the bispecific antibody redirects CD3+ T cells to destroy the cancer cells.

[0005] Claudins are a family of integral tight junction membrane proteins. Claudin 18 (or CLDN18) is one such protein in the claudin family. Claudin 18 can be expressed as two different splice variants. Claudin 18 isoform A2 splice variant (or CLDN18.2) is expressed on gastric cells, particularly on cells of the gastric epithelium. Of note, CLDN18.2 has also been previously reported to be highly expressed in some cancers, including gastric cancer, esophageal cancer, and pancreatic cancer. In view of this, anti-CLDN18.2 antibodies are useful, for example, for localizing anti-tumor therapeutics (e.g., chemotherapeutic agents and T cells) to such CLDN18.2-expressing tumors.

[0006] Provided herein are novel bispecific antibodies against CD3 and CLDN18.2 that are capable of localizing CD3+ effector T cells to CLDN18.2-expressing tumors. Summary of the Invention

[0007] Accordingly, provided herein are claudin 18.2 (CLDN18.2) antigen-binding domains and anti-CLDN18.2 antibodies (eg, bispecific antibodies).

[0008] In one aspect, provided herein is a heterodimeric antibody comprising a first monomer, a second monomer, and a common light chain. The first monomer comprises, from N-terminus to C-terminus, VH-CH1-domain linker-scFv-domain linker-CH2-CH3, where CH2-CH3 is a first variant Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH-CH1-hinge-CH2-CH3, where CH2-CH3 is a second variant Fc domain. The common light chain comprises VL-CL. In such antibodies, one of the variant Fc domains comprises the amino acid substitutions N208D / Q295E / N384D / Q418E / N421D, one of the first and second variant Fc domains each comprises the amino acid substitutions E233P / L234V / L235A / G236del / S267K, one of the variant Fc domains comprises the amino acid substitutions S364K / E357Q and the other variant Fc domain comprises the amino acid substitutions L368D / K370S, numbering according to the EU index as in Kabat, the VH domain comprises SEQ ID NO: 81 or SEQ ID NO: 82 and the VL domain comprises SEQ ID NO: 84. In some embodiments, the scFv has a sequence selected from the group consisting of SEQ ID NO:87, SEQ ID NO:88, SEQ ID NO:98, SEQ ID NO:99, SEQ ID NO:109, SEQ ID NO:110, SEQ ID NO:120, SEQ ID NO:121, SEQ ID NO:131, SEQ ID NO:132, SEQ ID NO:142, SEQ ID NO:143.

[0009] In some embodiments, the antibody is selected from the group consisting of XENP24647, XENP31729, XENP24649, XENP31723, XENP31725, XENP31727, XENP29476, XENP29478, XENP31724, XENP31726, XENP31728, XENP29477, XENP29479, XENC10101, XENC10102, XENC10103, XENC 10104, XENC10105, XENC10106, XENC10107, XENC10108, XENC10109, XENC10110, XENC10111, XENC10112, XENC 10113, XENC10114, XENC10115, XENC10116, XENC10117, XENC10118, XENC10119, XENC10120, XENC10121, XENC 10122, XENC10123, XENC10124, XENC10125, XENC10126, XENC10127, XENC10128, XENC10129, XENC10130, XENC 10131, XENC10132, XENC10133, XENC10134, XENC10135, XENC10136, XENC10137, XENC10138, XENC10139, XENC 10140, XENC10141, XENC10142, XENC10143, XENC10144, XENC10145, XENC10146, XENC10147, XENC10148, XENC10149, XENC10150, XENC10151, XENC10152, XENC10153, XENC10154, XENC10155, and XENC10156.

[0010] In another aspect, provided herein is a composition comprising an anti-CLDN18.2 antigen binding domain (ABD), the ABD comprising a) a variable heavy chain domain comprising SEQ ID NO: 81, and b) a variable light chain domain comprising SEQ ID NO: 84.

[0011] In another aspect, provided herein is a composition comprising an anti-CLDN18.2 antigen binding domain (ABD), the ABD comprising a) a variable heavy chain domain comprising SEQ ID NO: 82, and b) a variable light chain domain comprising SEQ ID NO: 84.

[0012] In one aspect, provided herein is a heterodimeric antibody comprising a first monomer, a second monomer, and a third monomer. The first monomer comprises, from N-terminus to C-terminus, scFv-domain linker-CH2-CH3, which is a first variant Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH-CH1-hinge-CH2-CH3, which is a second variant Fc domain. The third monomer comprises VL-CL. In such antibodies, one of the variant Fc domains comprises the amino acid substitutions N208D / Q295E / N384D / Q418E / N421D, one of the first and second variant Fc domains each comprises the amino acid substitutions E233P / L234V / L235A / G236del / S267K, one of the variant Fc domains comprises the amino acid substitutions S364K / E357Q and the other variant Fc domain comprises the amino acid substitutions L368D / K370S, numbering according to the EU index as in Kabat. In some embodiments, the scFv has a sequence selected from the group consisting of SEQ ID NO:87, SEQ ID NO:88, SEQ ID NO:98, SEQ ID NO:99, SEQ ID NO:109, SEQ ID NO:110, SEQ ID NO:120, SEQ ID NO:121, SEQ ID NO:131, SEQ ID NO:132, SEQ ID NO:142, SEQ ID NO:143. In some embodiments, the variable heavy (VH) domain comprises SEQ ID NO:81 or SEQ ID NO:82 and the variable light (Vl) domain comprises SEQ ID NO:84.

[0013] In some embodiments, the CH1-hinge-CH2-CH3 component of the second heavy chain comprises SEQ ID NO: 32, the first variant Fc domain comprises SEQ ID NO: 33, and the constant light domain comprises SEQ ID NO: 74. In some embodiments, the antibody is selected from the group consisting of XENP29472, XENP29473, XENP29474, and XENP29475.

[0014] In another aspect, provided herein is a heterodimeric antibody comprising a first monomer, a second monomer, and a common light chain. The first monomer comprises, from N-terminus to C-terminus, VH1-CH1-hinge-CH2-CH3-domain linker-VH2, where CH2-CH3 is a first variant Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH1-CH1-hinge-CH2-CH3-domain linker-VL2, where CH2-CH3 is a second variant Fc domain. The common light chain comprises VL1-CL. In such antibodies, one of the variant Fc domains comprises the amino acid substitutions N208D / Q295E / N384D / Q418E / N421D, the first and second variant Fc domains each comprise the amino acid substitutions E233P / L234V / L235A / G236del / S267K, one of the variant Fc domains comprises the amino acid substitutions S364K / E357Q and the other variant Fc domain comprises the amino acid substitutions L368D / K370S, numbering according to the EU index as in Kabat. VH2 and VL2 are selected from the pairs consisting of SEQ ID NO:89 and SEQ ID NO:93, SEQ ID NO:100 and SEQ ID NO:104, SEQ ID NO:111 and SEQ ID NO:115, SEQ ID NO:122 and SEQ ID NO:126, SEQ ID NO:133 and SEQ ID NO:137, SEQ ID NO:144 and SEQ ID NO:148. Furthermore, the VH1 domain comprises SEQ ID NO:81 or SEQ ID NO:82, and the VL1 domain comprises SEQ ID NO:84.

[0015] In another aspect, provided herein is a heterodimeric antibody comprising a first monomer, a second monomer, and a common light chain. The first monomer comprises, from N-terminus to C-terminus, VH1-CH1-hinge-CH2-CH3-domain linker-scFv, where CH2-CH3 is a first variant Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH1-CH1-hinge-CH2-CH3, where CH2-CH3 is a second variant Fc domain. The common light chain comprises VL1-CL. In such antibodies, one of the variant Fc domains comprises the amino acid substitutions N208D / Q295E / N384D / Q418E / N421D, the first and second variant Fc domains each comprise the amino acid substitutions E233P / L234V / L235A / G236del / S267K, one of the variant Fc domains comprises the amino acid substitutions S364K / E357Q and the other variant Fc domain comprises the amino acid substitutions L368D / K370S, numbering according to the EU index as in Kabat. The scFv has a sequence selected from the group consisting of SEQ ID NO:87, SEQ ID NO:88, SEQ ID NO:98, SEQ ID NO:99, SEQ ID NO:109, SEQ ID NO:110, SEQ ID NO:120, SEQ ID NO:121, SEQ ID NO:131, SEQ ID NO:132, SEQ ID NO:142, SEQ ID NO:143. Furthermore, the VH1 domain comprises SEQ ID NO:81 or SEQ ID NO:82, and the VL1 domain comprises SEQ ID NO:84.

[0016] In one aspect, provided herein is a heterodimeric antibody comprising a first monomer, a second monomer, and a common light chain. The first monomer comprises, from N-terminus to C-terminus, VH1-CH1-domain linker-VH2-domain linker-CH2-CH3, where CH2-CH3 is a first variant Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH1-CH1-domain linker-VL2-domain linker-CH2-CH3, where CH2-CH3 is a second variant Fc domain. The common light chain comprises VL1-CL. In such antibodies, one of the variant Fc domains comprises the amino acid substitutions N208D / Q295E / N384D / Q418E / N421D, the first and second variant Fc domains each comprise the amino acid substitutions E233P / L234V / L235A / G236del / S267K, one of the variant Fc domains comprises the amino acid substitutions S364K / E357Q and the other variant Fc domain comprises the amino acid substitutions L368D / K370S, numbering according to the EU index as in Kabat. VH2 and VL2 are selected from the pairs consisting of SEQ ID NO:89 and SEQ ID NO:93, SEQ ID NO:100 and SEQ ID NO:104, SEQ ID NO:111 and SEQ ID NO:115, SEQ ID NO:122 and SEQ ID NO:126, SEQ ID NO:133 and SEQ ID NO:137, SEQ ID NO:144 and SEQ ID NO:148. Furthermore, the VH1 domain comprises SEQ ID NO:81 or SEQ ID NO:82, and the VL1 domain comprises SEQ ID NO:84.

[0017] In another aspect, provided herein is a heterodimeric antibody comprising a first monomer, a second monomer, and a light chain. The first monomer comprises, from N-terminus to C-terminus, VH-CH1-domain linker-scFv-domain linker-CH2-CH3, where CH2-CH3 is a first variant Fc domain. The second monomer comprises a second variant Fc domain comprising CH2-CH3. The light chain comprises VL-CL. In such antibodies, one of the variant Fc domains comprises the amino acid substitutions N208D / Q295E / N384D / Q418E / N421D, the first and second variant Fc domains each comprise the amino acid substitutions E233P / L234V / L235A / G236del / S267K, one of the variant Fc domains comprises the amino acid substitutions S364K / E357Q and the other variant Fc domain comprises the amino acid substitutions L368D / K370S, numbering according to the EU index as in Kabat. The scFv has a sequence selected from the group consisting of SEQ ID NO:87, SEQ ID NO:88, SEQ ID NO:98, SEQ ID NO:99, SEQ ID NO:109, SEQ ID NO:110, SEQ ID NO:120, SEQ ID NO:121, SEQ ID NO:131, SEQ ID NO:132, SEQ ID NO:142, SEQ ID NO:143. Furthermore, the VH domain comprises SEQ ID NO:81 or SEQ ID NO:82, and the VL domain comprises SEQ ID NO:84.

[0018] In one aspect, provided herein is a heterodimeric antibody comprising a first monomer, a second monomer, and a light chain. The first monomer comprises, from N-terminus to C-terminus, scFv-domain linker-VH-CH1-hinge-CH2-CH3, where CH2-CH3 is a first variant Fc domain. The second monomer comprises a second variant Fc domain comprising CH2-CH3. The light chain comprises VL-CL. In such antibodies, one of the variant Fc domains comprises the amino acid substitutions N208D / Q295E / N384D / Q418E / N421D, the first and second variant Fc domains each comprise the amino acid substitutions E233P / L234V / L235A / G236del / S267K, one of the variant Fc domains comprises the amino acid substitutions S364K / E357Q and the other variant Fc domain comprises the amino acid substitutions L368D / K370S, numbering according to the EU index as in Kabat. The scFv has a sequence selected from the group consisting of SEQ ID NO:87, SEQ ID NO:88, SEQ ID NO:98, SEQ ID NO:99, SEQ ID NO:109, SEQ ID NO:110, SEQ ID NO:120, SEQ ID NO:121, SEQ ID NO:131, SEQ ID NO:132, SEQ ID NO:142, SEQ ID NO:143. Furthermore, the VH domain comprises SEQ ID NO:81 or SEQ ID NO:82, and the VL domain comprises SEQ ID NO:84.

[0019] In another aspect, provided herein is a heterodimeric antibody comprising a first monomer, a second monomer, and a common light chain. The first monomer comprises, from N-terminus to C-terminus, scFv-domain linker-VH-CH1-hinge-CH2-CH3, where CH2-CH3 is a first variant Fc domain. The second monomer comprises, from N-terminus to C-terminus, VH-CH1-hinge-CH2-CH3, where CH2-CH3 is a second variant Fc domain. The common light chain comprises VL-CL. In such antibodies, one of the variant Fc domains comprises the amino acid substitutions N208D / Q295E / N384D / Q418E / N421D, the first and second variant Fc domains each comprise the amino acid substitutions E233P / L234V / L235A / G236del / S267K, one of the variant Fc domains comprises the amino acid substitutions S364K / E357Q and the other variant Fc domain comprises the amino acid substitutions L368D / K370S, numbering according to the EU index as in Kabat. The scFv has a sequence selected from the group consisting of SEQ ID NO:87, SEQ ID NO:88, SEQ ID NO:98, SEQ ID NO:99, SEQ ID NO:109, SEQ ID NO:110, SEQ ID NO:120, SEQ ID NO:121, SEQ ID NO:131, SEQ ID NO:132, SEQ ID NO:142, SEQ ID NO:143. Furthermore, the VH domain comprises SEQ ID NO:81 or SEQ ID NO:82, and the VL domain comprises SEQ ID NO:84.

[0020] In another aspect, provided herein is a nucleic acid composition comprising a nucleic acid encoding any of the heterodimeric antibodies or antigen-binding domains described herein.

[0021] In yet another aspect, provided herein is an expression vector comprising any of the nucleic acids described herein.

[0022] In one aspect, provided herein is a host cell transformed with any of the expression vectors or nucleic acids described herein.

[0023] In another aspect, provided herein is a method of making a subject heterodimeric antibody or antigen-binding domain described herein, the method comprising culturing a host cell transformed with any of the expression vectors or nucleic acids described herein under conditions in which the antibody or antigen-binding domain is expressed, and recovering the antibody or antigen-binding domain.

[0024] In some embodiments, provided herein is a method of treating cancer, comprising administering to a patient in need of treatment any one of the subject antibodies described herein. In some embodiments, the cancer is gastric cancer. [Brief description of the drawings]

[0025] [Figure 1A] Useful pairs of Fc heterodimerization variant sets (including skewed and pI variants) are shown. 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] Useful pairs of Fc heterodimerization variant sets (including skewed and pI variants) are shown. 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] Useful pairs of Fc heterodimerization variant sets (including skewed and pI variants) are shown. 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] Useful pairs of Fc heterodimerization variant sets (including skewed and pI variants) are shown. 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]Useful pairs of Fc heterodimerization variant sets (including skewed and pI variants) are shown. 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 on both monomers, although in some cases they may only be on one monomer. [Figure 4] Particularly useful embodiments of "non-Fv" components of the invention are provided. [Diagram 5] Shown are several charged scFv linkers that are used to increase or decrease the pI of heterodimeric antibodies that utilize one or more scFvs as building blocks. (+H) positive linkers are specifically used herein. A single prior art scFv linker with a single charge is called "Whitlow" (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. It should also be noted that any or all of these linkers may be used as optional domain linkers discussed herein, in particular those listed as "additional scFv linkers" that are uncharged. [Figure 6A]Shown are sequences of some useful 1+1 Fab-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, lacking the Fv sequence (e.g., scFv and Fab side VH). That is, the " / " slash at the beginning of the "Fab-Fc side" indicates that the C-terminus of the VH as outlined herein is attached at that point. Similarly, the " / " slash at the beginning of the "scFv-Fc side" indicates that the C-terminus of the scFv (e.g., VH-scFv linker-VL or VL-scFv linker-VH) as outlined herein is attached at that point. Scaffold 1 is based on human IgG1 (356E / 358M 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 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 that 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 6B]Shown are sequences of some useful 1+1 Fab-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, lacking the Fv sequence (e.g., scFv and Fab side VH). That is, the " / " slash at the beginning of the "Fab-Fc side" indicates that the C-terminus of the VH as outlined herein is attached at that point. Similarly, the " / " slash at the beginning of the "scFv-Fc side" indicates that the C-terminus of the scFv (e.g., VH-scFv linker-VL or VL-scFv linker-VH) as outlined herein is attached at that point. Scaffold 1 is based on human IgG1 (356E / 358M 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 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 that 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 6C]Shown are sequences of some useful 1+1 Fab-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, lacking the Fv sequence (e.g., scFv and Fab side VH). That is, the " / " slash at the beginning of the "Fab-Fc side" indicates that the C-terminus of the VH as outlined herein is attached at that point. Similarly, the " / " slash at the beginning of the "scFv-Fc side" indicates that the C-terminus of the scFv (e.g., VH-scFv linker-VL or VL-scFv linker-VH) as outlined herein is attached at that point. Scaffold 1 is based on human IgG1 (356E / 358M 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 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 that 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 6D]Shown are sequences of some useful 1+1 Fab-scFv-Fc bispecific antibody format heavy chain scaffolds based on human IgG1, lacking the Fv sequence (e.g., scFv and Fab side VH). That is, the " / " slash at the beginning of the "Fab-Fc side" indicates that the C-terminus of the VH as outlined herein is attached at that point. Similarly, the " / " slash at the beginning of the "scFv-Fc side" indicates that the C-terminus of the scFv (e.g., VH-scFv linker-VL or VL-scFv linker-VH) as outlined herein is attached at that point. Scaffold 1 is based on human IgG1 (356E / 358M 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 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 that 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 7A]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 and linker sequences. Thus, for example, at the "Fab-Fc side", (VH domain / " indicates that the C-terminus of the VH domain is attached at the position where the CH1 domain of the heavy chain begins. Similarly, at the beginning of the "Fab-scFv-Fc side", "VH-CH1-domain linker1-scFv-domain linker2 / " indicates that these sequences are variants of the CH2-CH3 domain. Scaffold 1 is based on human IgG1 (356E / 358M allotype) with S364K / E357Q:L368D / K370S scavariant, N208D / Q295E / N384D / Q418E / N421D on chain with L368D / K370S scavariant. As discussed herein, the scFv domain can be in either orientation -VH-scFv linker-VL- or -VL-scFv linker-VH- as discussed herein. In addition, preferred domain linkers for "Domain Linker 2" are those listed in FIG. 5 as "Useful Domain Linkers". Note that the sequence identifiers are for the listed sequences only. Scaffold 1 is based on human IgG1 (356E / 358M allotype) with S364K / E357Q:L368D / K370S scavariant, N208D / Q295E / N384D / Q418E / N421D on chain with L368D / K370S scavariant. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D on one chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D scaffold variant on the chain with the L368E / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant 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 7B]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 and linker sequences. Thus, for example, at the "Fab-Fc side", (VH domain / " indicates that the C-terminus of the VH domain is attached at the position where the CH1 domain of the heavy chain begins. Similarly, at the beginning of the "Fab-scFv-Fc side", "VH-CH1-domain linker1-scFv-domain linker2 / " indicates that these sequences are variants of the CH2-CH3 domain. Scaffold 1 is based on human IgG1 (356E / 358M allotype) with S364K / E357Q:L368D / K370S scavariant, N208D / Q295E / N384D / Q418E / N421D on chain with L368D / K370S scavariant. As discussed herein, the scFv domain can be in either orientation -VH-scFv linker-VL- or -VL-scFv linker-VH- as discussed herein. In addition, preferred domain linkers for "Domain Linker 2" are those listed in FIG. 5 as "Useful Domain Linkers". Note that the sequence identifiers are for the listed sequences only. Scaffold 1 is based on human IgG1 (356E / 358M allotype) with S364K / E357Q:L368D / K370S scavariant, N208D / Q295E / N384D / Q418E / N421D on chain with L368D / K370S scavariant. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D on one chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D scaffold variant on the chain with the L368E / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant 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 7C]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 and linker sequences. Thus, for example, at the "Fab-Fc side", (VH domain / " indicates that the C-terminus of the VH domain is attached at the position where the CH1 domain of the heavy chain begins. Similarly, at the beginning of the "Fab-scFv-Fc side", "VH-CH1-domain linker1-scFv-domain linker2 / " indicates that these sequences are variants of the CH2-CH3 domain. Scaffold 1 is based on human IgG1 (356E / 358M allotype) with S364K / E357Q:L368D / K370S scavariant, N208D / Q295E / N384D / Q418E / N421D on chain with L368D / K370S scavariant. As discussed herein, the scFv domain can be in either orientation -VH-scFv linker-VL- or -VL-scFv linker-VH- as discussed herein. In addition, preferred domain linkers for "Domain Linker 2" are those listed in FIG. 5 as "Useful Domain Linkers". Note that the sequence identifiers are for the listed sequences only. Scaffold 1 is based on human IgG1 (356E / 358M allotype) with S364K / E357Q:L368D / K370S scavariant, N208D / Q295E / N384D / Q418E / N421D on chain with L368D / K370S scavariant. Scaffold 2 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368D / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D on one chain with the L368D / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant on both chains. Scaffold 3 is based on human IgG1 (356E / 358M allotype) and contains the S364K:L368E / K370S scaffold variant, the N208D / Q295E / N384D / Q418E / N421D scaffold variant on the chain with the L368E / K370S scaffold variant, and the E233P / L234V / L235A / G236del / S267K deletion variant 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 8] FIG. 1 shows the "non-Fv" scaffolds of the cognate light chains (i.e., constant light chains) used in the 1+1 Fab-scFv-Fc and 2+1 Fab2-scFv-Fc bispecific antibodies of the invention. [Figure 9] 1 shows the sequences of A) human claudin 18 isoform A2 (CLDN18.2), B) mouse claudin 18 isoform A2.1, C) mouse claudin 18 isoform A2.2, and D) cynomolgus claudin 18 to facilitate the development of antigen-binding domains that are cross-reactive to facilitate clinical development. [Figure 10] 1 shows an exemplary anti-CLDN18.2 ABD variable heavy and variable light chains used in the anti-CLDN18.2×anti-CD3 bispecific antibody of the present invention. The CDRs are underlined. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, and therefore, not only the underlined CDRs but also CDRs contained within the VH and VL domains using other numbering systems are included herein. The H0 and L0 variable heavy chains are murine, and the H1, H2, and L1 variable light chains are humanized. [Figure 11] The sequence of a murine anti-CLDN18.2 antibody with a deletion variant (E233P / L234V / L235A / G236del / S267K, "IgG1_PVA_ / S267K") is shown. The CDRs are underlined. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. [Figure 12A]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: 10), 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 slash indicates the boundary of the variable domain. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. The scFv sequence is shown in both orientations as shown. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, 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 the scFv or Fab format. [Figure 12B] 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: 10), 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 slash indicates the boundary of the variable domain. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. The scFv sequence is shown in both orientations as shown. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, 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 the scFv or Fab format. [Figure 12C] 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: 10), 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 slash indicates the boundary of the variable domain. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. The scFv sequence is shown in both orientations as shown. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, 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 the scFv or Fab format. [Figure 12D]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: 10), 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 slash indicates the boundary of the variable domain. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. The scFv sequence is shown in both orientations as shown. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, 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 the scFv or Fab format. [Figure 12E] 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: 10), 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 slash indicates the boundary of the variable domain. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. The scFv sequence is shown in both orientations as shown. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, 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 the scFv or Fab format. [Figure 12F] 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: 10), 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 slash indicates the boundary of the variable domain. In addition, the naming convention indicates the orientation of the scFv from the N-terminus to the C-terminus. The scFv sequence is shown in both orientations as shown. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, 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 the scFv or Fab format. [Figure 13A] Two useful formats of the present invention are shown. Figure 13A shows a "1+1 Fab-scFv-Fc" format, where a first Fab arm binds CLDN18.2 and a second scFv arm binds CD3. Figure 13B shows a "2+1 Fab2-scFv-Fc" format, with a first Fab arm that binds CLDN18.2 and a second Fab-scFv arm that binds CD3, where the Fab binds CLDN18.2 and the scFv binds CD3. It should be noted that additional formats can be used as well, as generally shown in Figure 42 herein. [Figure 13B]Two useful formats of the present invention are shown. Figure 13A shows a "1+1 Fab-scFv-Fc" format, where a first Fab arm binds CLDN18.2 and a second scFv arm binds CD3. Figure 13B shows a "2+1 Fab2-scFv-Fc" format, with a first Fab arm that binds CLDN18.2 and a second Fab-scFv arm that binds CD3, where the Fab binds CLDN18.2 and the scFv binds CD3. It should be noted that additional formats can be used as well, as generally shown in Figure 42 herein. [Figure 14] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in 1+1 Fab-scFv-Fc format. The antibody is named using the Fab variable region first and the scFv variable region second, separated by a dash. The CDRs are underlined and slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15A] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15B]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15C] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15D]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15E] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15F]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15G] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15H]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15I] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15J]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15K] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15L]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15M] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15N]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15O] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15P]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15Q] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15R]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15S] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15T]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15U] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15V]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15W] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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. [Fig. 15X]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15Y] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15Z]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AA] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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. [Diagram 15AB]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AC] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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. [Fig. 15A]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AE] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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. [Fig. 15AF]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15A-G] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AH]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AI] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AJ]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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. [Fig. 15AK] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AL]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AM] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AN]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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. [Fig. 15AO] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AP]1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 15AQ] 1 shows the amino acid sequence of a prototype anti-CLDN18.2 x anti-CD3 bispecific antibody (based on the murine CLDN18.2 ABD shown in FIG. 10) in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 16] 1 shows the amino acid sequence of a control anti-RSV x high CD3 bispecific antibody in 1 + 1 Fab-scFv-Fc format. The antibody is named with the Fab variable region first and the scFv variable region second separated by a dash. The CDRs are underlined and the slash indicates the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), but this can be reversed. 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 17A]Binding of prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with murine CLDN18.2 ABDs XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), XENP24648 (2+1 Fab2-scFv-Fc with CD3-high-medium #1), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium) to A) KP-4 cells and B) NUGC-4 cells. Controls included anti-RSV x anti-CD3 bispecific antibodies, cells only, and secondary antibody only. The data show that the prototype anti-CLDN18.2 x anti-CD3 bsAbs bound to NUGC-4 cells in a dose-dependent manner, with close to baseline binding to KP-4 cells at all concentrations tested. Notably, the bsAbs in the "2+1 Fab2-scFv-Fc" format (i.e., XENP24647, XENP24648, and XENP24949) bound to NUGC-4 cells much more potently than the bsAbs in the "1+1 Fab-scFv-Fc" format (i.e., XENP24645 and XENP24646), due to the extra avidity carried by the "2+1 Fab2-scFv-Fc" format. [Figure 17B]Binding of prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with murine CLDN18.2 ABDs XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), XENP24648 (2+1 Fab2-scFv-Fc with CD3-high-medium #1), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium) to A) KP-4 cells and B) NUGC-4 cells. Controls included anti-RSV x anti-CD3 bispecific antibodies, cells only, and secondary antibody only. The data show that the prototype anti-CLDN18.2 x anti-CD3 bsAbs bound to NUGC-4 cells in a dose-dependent manner, with close to baseline binding to KP-4 cells at all concentrations tested. Notably, the bsAbs in the "2+1 Fab2-scFv-Fc" format (i.e., XENP24647, XENP24648, and XENP24949) bound to NUGC-4 cells much more potently than the bsAbs in the "1+1 Fab-scFv-Fc" format (i.e., XENP24645 and XENP24646), due to the extra avidity carried by the "2+1 Fab2-scFv-Fc" format. [Figure 18A] 14 shows induction of RTCC on A) KP-4 cells and B) NUGC-4 cells by prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with murine CLDN18.2 ABD XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), XENP24648 (2+1 Fab2-scFv-Fc with CD3-high medium #1), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Data show that the prototypic anti-CLDN18.2 x anti-CD3 bsAb dose-dependently induced RTCC on NUGC-4 but not KP-4 cells. [Figure 18B]14 shows induction of RTCC on A) KP-4 cells and B) NUGC-4 cells by prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with murine CLDN18.2 ABD XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), XENP24648 (2+1 Fab2-scFv-Fc with CD3-high medium #1), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Data show that the prototypic anti-CLDN18.2 x anti-CD3 bsAb dose-dependently induced RTCC on NUGC-4 but not KP-4 cells. [Figure 19A] Figure 1 shows expression levels on A) NUGC-4 cells and B) SNU-601 cells as determined by flow cytometry. The data show that SNU-601 expresses more CLDN18.2 than NUGC-4. [Figure 19B] Figure 1 shows expression levels on A) NUGC-4 cells and B) SNU-601 cells as determined by flow cytometry. The data show that SNU-601 expresses more CLDN18.2 than NUGC-4. [Figure 20A]Binding of prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with murine CLDN18.2 ABDs XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium) to A) NUGC-4 cells and B) SNU-601 cells. Controls included anti-RSV x anti-CD3 bispecific antibodies, cells only, and secondary antibody only. The data show that each of the bsAbs bound to both NUGC-4 and SNU-601 in a dose-dependent manner, with higher maximal binding to SNU-601 cells than to NUGC-4, which is consistent with the respective CLDN18.2 expression levels on each cell line. [Figure 20B] Binding of prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with murine CLDN18.2 ABDs XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium) to A) NUGC-4 cells and B) SNU-601 cells. Controls included anti-RSV x anti-CD3 bispecific antibodies, cells only, and secondary antibody only. The data show that each of the bsAbs bound to both NUGC-4 and SNU-601 in a dose-dependent manner, with higher maximal binding to SNU-601 cells than to NUGC-4, which is consistent with the respective CLDN18.2 expression levels on each cell line. [Figure 21A]Figure 1 shows induction of RTCC (as indicated by a decrease in viable target cells) on A) NUGC-4 cells and B) SNU-601 cells after 24 hours of incubation with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show that the bsAb enhances killing of target cells (i.e., NUGC-4 and SNU-601 cells) as indicated by a reduction in viable cells. [Figure 21B] Figure 1 shows induction of RTCC (as indicated by a decrease in viable target cells) on A) NUGC-4 cells and B) SNU-601 cells after 24 hours of incubation with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show that the bsAb enhances killing of target cells (i.e., NUGC-4 and SNU-601 cells) as indicated by a reduction in viable cells. [Figure 22A]Figure 1 shows induction of RTCC (as indicated by an increase in dead target cells) on A) NUGC-4 cells and B) SNU-601 cells after 24 hours of incubation with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show that bsAb enhances killing of target cells (i.e., NUGC-4 and SNU-601 cells) as indicated by an increase in dead / dying cells. [Figure 22B] Figure 1 shows induction of RTCC (as indicated by an increase in dead target cells) on A) NUGC-4 cells and B) SNU-601 cells after 24 hours of incubation with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show that bsAb enhances killing of target cells (i.e., NUGC-4 and SNU-601 cells) as indicated by an increase in dead / dying cells. [Figure 23A]Figure 1 shows induction of RTCC (as indicated by a decrease in viable target cells) on A) NUGC-4 cells and B) SNU-601 cells after 48 hours of incubation with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show that bsAb enhances killing of target cells (i.e., NUGC-4 and SNU-601 cells) as indicated by a reduction in viable cells. [Figure 23B] Figure 1 shows induction of RTCC (as indicated by a decrease in viable target cells) on A) NUGC-4 cells and B) SNU-601 cells after 48 hours of incubation with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show that bsAb enhances killing of target cells (i.e., NUGC-4 and SNU-601 cells) as indicated by a reduction in viable cells. [Figure 24A]Figure 1 shows induction of RTCC (as indicated by an increase in dead target cells) on A) NUGC-4 cells and B) SNU-601 cells after 48 hours of incubation with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show that bsAb enhances killing of target cells (i.e., NUGC-4 and SNU-601 cells) as indicated by an increase in dead / dying cells. [Figure 24B] Figure 1 shows induction of RTCC (as indicated by an increase in dead target cells) on A) NUGC-4 cells and B) SNU-601 cells after 48 hours of incubation with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium #1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show that bsAb enhances killing of target cells (i.e., NUGC-4 and SNU-601 cells) as indicated by an increase in dead / dying cells. [Figure 25A]Figure 1 shows the percentage of CD4+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of NUGC-4 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 25B] Figure 1 shows the percentage of CD4+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of NUGC-4 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 25C]Figure 1 shows the percentage of CD4+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of NUGC-4 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 26A] Figure 1 shows the percentage of CD8+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of NUGC-4 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 26B]Figure 1 shows the percentage of CD8+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of NUGC-4 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 26C] Figure 1 shows the percentage of CD8+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of NUGC-4 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 27A]Figure 1 shows the percentage of CD4+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of SNU-601 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 27B] Figure 1 shows the percentage of CD4+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of SNU-601 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 27C]Figure 1 shows the percentage of CD4+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of SNU-601 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 28A] Figure 1 shows the percentage of CD8+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of SNU-601 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 28B]Figure 1 shows the percentage of CD8+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of SNU-601 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 28C] Figure 1 shows the percentage of CD8+ T cells expressing A) CD69, B) CD25, and C) CD107a after 48 hours of incubation of SNU-601 cells with human PBMCs (effector to target cell ratio of 20:1) and the following prototype anti-CLDN18.2 x anti-CD3 bispecific antibodies with mouse CLDN18.2 ABD: XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24646 (1+1 Fab-scFv-Fc with CD3-high-medium#1), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), and XENP24649 (2+1 Fab2-scFv-Fc with CD3-medium). Controls included anti-RSV x anti-CD3 bispecific antibodies, target cells only, and target and effector cells only. The data show trends consistent with RTCC, ie, for example, higher affinity CD3 binding and / or bivalent CLDN18.2 binding enhances T cell activation and degranulation. [Figure 29]1 shows the sequence of an anti-CLDN18.2 antibody with a humanized variable region with deleted variants (E233P / L234V / L235A / G236del / S267K, "IgG1_PVA_ / S267K"). The CDRs are underlined. As is true for any sequence described herein that contains CDRs, the exact identification of the CDR positions may vary slightly depending on the numbering used as shown in Table 1, and therefore, not only the underlined CDRs, but also CDRs contained within the VH and VL domains using other numbering systems are included herein. [Diagram 30] 1 shows the germline identity of the humanized CLDN18.2 ABD compared to the murine CLDN18.2 ABD. [Figure 31A] 1 shows the amino acid sequence of an anti-CLDN18.2 x anti-CD3 bispecific antibody with humanized variable regions in a 1+1 Fab-scFv-Fc format. The antibody is named using the Fab variable region first and the scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 31B] 1 shows the amino acid sequence of an anti-CLDN18.2 x anti-CD3 bispecific antibody with humanized variable regions in a 1+1 Fab-scFv-Fc format. The antibody is named using the Fab variable region first and the scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 31C]1 shows the amino acid sequence of an anti-CLDN18.2 x anti-CD3 bispecific antibody with humanized variable regions in a 1+1 Fab-scFv-Fc format. The antibody is named using the Fab variable region first and the scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 32A] 1 shows the amino acid sequence of an anti-CLDN18.2 x anti-CD3 bispecific antibody with humanized variable regions in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 32B] 1 shows the amino acid sequence of an anti-CLDN18.2 x anti-CD3 bispecific antibody with humanized variable regions in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 32C]1 shows the amino acid sequence of an anti-CLDN18.2 x anti-CD3 bispecific antibody with humanized variable regions in a 2+1 Fab2-scFv-Fc format. The antibody is named using the Fab variable region first and the Fab-scFv variable region second, separated by a dash. The CDRs are underlined and the slashes indicate the boundaries of the variable regions. The scFv domains have a VH-scFv linker-VL orientation (N-terminus to C-terminus), although this can be reversed. 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 33A] Figure 1 shows expression levels on A) SNU-601 cells and B) SNU-601(2E4) cells (enriched for the CLDN18.2 expressing population) as determined by flow cytometry. The data show that SNU-601(2E4) contains a substantially higher population of CLDN18.2+ cells. The experiments in this section are performed using SNU-601(2E4) cells. [Figure 33B] Figure 1 shows expression levels on A) SNU-601 cells and B) SNU-601(2E4) cells (enriched for the CLDN18.2 expressing population) as determined by flow cytometry. The data show that SNU-601(2E4) contains a substantially higher population of CLDN18.2+ cells. The experiments in this section are performed using SNU-601(2E4) cells. [Diagram 34]The following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABD: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), 1 shows the binding of XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1) to SNU-601 (2E4) cells. Controls used were XENP24644 (H0L0 CLDN18.2, bivalent mAb), XENP29470 (H1L1 CLDN18.2, bivalent mAb), XENP29471 (H2L1 CLDN18.2, bivalent mAb), XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. The data show that the bsAb with humanized CLDN18.2 ABD had similar binding to SNU-601 (2E4) cells as the bsAb with murine CLDN18.2 ABD, indicating that humanization preserved the binding efficacy of the bsAb. Notably, bsAbs in the "2+1 Fab2-scFv-Fc" format showed similar binding to the bivalent anti-CLDN18.2 mAb.In addition, the data show that bsAbs based on H1L1 humanized variants (e.g., XENP29472, XENP29474, XENP28476, and XENP29478) retained binding better than bsAbs based on H2L1 humanized variants (e.g., XENP29473, XENP29475, XENP29477, and XENP29479). [Figure 35A]Figure 1 shows induction of RTCC on SNU-601(2E4) cells, as indicated by A) a decrease in the number of CFSE+ SNU-601(2E4), B) the percentage of CFSE+ SNU-601(2E4) cells stained with Zombie Aqua, and C) Zombie Aqua MFI on CFSE+ SNU-601(2E4) cells after incubation of CFSE-labeled SNU-601(2E4) for 24 hours with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABD: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (H2L1 1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high#1), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD and CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD and CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high#1) Fab2-scFv-Fc), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab2-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. Consistent with the binding data, humanization preserved the induction of RTCC by bsAbs with humanized CLDN18.2 ABD. [Figure 35B]Figure 1 shows induction of RTCC on SNU-601(2E4) cells, as indicated by A) a decrease in the number of CFSE+ SNU-601(2E4), B) the percentage of CFSE+ SNU-601(2E4) cells stained with Zombie Aqua, and C) Zombie Aqua MFI on CFSE+ SNU-601(2E4) cells after incubation of CFSE-labeled SNU-601(2E4) for 24 hours with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABD: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (H2L1 1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high#1), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD and CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD and CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high#1) Fab2-scFv-Fc), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab2-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. Consistent with the binding data, humanization preserved the induction of RTCC by bsAbs with humanized CLDN18.2 ABD. [Figure 35C]Figure 1 shows induction of RTCC on SNU-601(2E4) cells, as indicated by A) a decrease in the number of CFSE+ SNU-601(2E4), B) the percentage of CFSE+ SNU-601(2E4) cells stained with Zombie Aqua, and C) Zombie Aqua MFI on CFSE+ SNU-601(2E4) cells after incubation of CFSE-labeled SNU-601(2E4) for 24 hours with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABD: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (H2L1 1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high#1), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD and CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD and CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD and CD3-high#1) Fab2-scFv-Fc), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab2-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. Consistent with the binding data, humanization preserved the induction of RTCC by bsAbs with humanized CLDN18.2 ABD. [Figure 36A]Figure 1 shows CD4+ T cell activation after 24-hour incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABDs [as indicated by A) CD69 MFI on CD4+ T cells, and B) percentage of CD4+ T cells expressing CD69]: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29479 ... Fab-scFv-Fc), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. Consistent with the binding data, humanization preserved T cell activation by bsAb with humanized CLDN18.2 ABD. [Figure 36B]Figure 1 shows CD4+ T cell activation after 24-hour incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABDs [as indicated by A) CD69 MFI on CD4+ T cells, and B) percentage of CD4+ T cells expressing CD69]: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29479 ... Fab-scFv-Fc), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. Consistent with the binding data, humanization preserved T cell activation by bsAb with humanized CLDN18.2 ABD. [Figure 37A]Figure 1 shows active degranulation of CD4+ T cells after 24 h incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABDs [as indicated by A) CD107a MFI on CD4+ T cells, and B) percentage of CD4+ T cells expressing CD107a]: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29479 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29470 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29471 (1+1 Fab-scFv Fab-scFv-Fc), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. [Figure 37B]Figure 1 shows active degranulation of CD4+ T cells after 24 h incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABDs [as indicated by A) CD107a MFI on CD4+ T cells, and B) percentage of CD4+ T cells expressing CD107a]: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29479 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29470 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29471 (1+1 Fab-scFv Fab-scFv-Fc), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. [Figure 38A]Figure 1 shows activation of CD8+ T cells after 24-hour incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABDs [as indicated by A) CD69 MFI on CD8+ T cells, and B) percentage of CD8+ T cells expressing CD69]: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29479 ... Fab-scFv-Fc), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. Consistent with the binding data, humanization preserved T cell activation by bsAb with humanized CLDN18.2 ABD. [Figure 38B]Figure 1 shows activation of CD8+ T cells after 24-hour incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABDs [as indicated by A) CD69 MFI on CD8+ T cells, and B) percentage of CD8+ T cells expressing CD69]: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29479 ... Fab-scFv-Fc), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. Consistent with the binding data, humanization preserved T cell activation by bsAb with humanized CLDN18.2 ABD. [Figure 39A]Figure 1 shows active degranulation of CD8+ T cells after 24 h incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABDs [as indicated by A) CD107a MFI on CD8+ T cells, and B) percentage of CD8+ T cells expressing CD107a]: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29479 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29470 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29471 (1+1 Fab-scFv Fab-scFv-Fc), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. [Figure 39B]Figure 1 shows active degranulation of CD8+ T cells after 24 h incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABDs [as indicated by A) CD107a MFI on CD8+ T cells, and B) percentage of CD8+ T cells expressing CD107a]: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29479 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29470 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29471 (1+1 Fab-scFv Fab-scFv-Fc), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high), XENP24647 (2+1 Fab2-scFv-Fc with CD3-high), cells only, and secondary antibody only. [Figure 40A]Figure 1 shows A) IFNγ and B) TNFα secretion after 24 h incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABD: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29479 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29470 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29471 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3- XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high #1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high #1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high) and XENP24647 (2+1 Fab2-scFv-Fc with CD3-high). Consistent with the binding data, humanization preserved the induction of cytokine secretion by bsAb with humanized CLDN18.2 ABD. [Figure 40B]Figure 1 shows A) IFNγ and B) TNFα secretion after 24 h incubation of SNU-601(2E4) with human PBMCs (effector to target cell ratio of 10:1) and the following anti-CLDN18.2 x anti-CD3 bispecific antibodies with humanized CLDN18.2 ABD: XENP29472 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29473 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29474 (1+1 Fab-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high#1), XENP29475 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29476 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29477 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29478 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29479 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29470 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high#1), XENP29471 (1+1 Fab-scFv-Fc with H2L1 CLDN18.2 ABD, CD3- XENP29476 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high), XENP29477 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high), XENP29478 (2+1 Fab2-scFv-Fc with H1L1 CLDN18.2 ABD, CD3-high #1), and XENP29479 (2+1 Fab2-scFv-Fc with H2L1 CLDN18.2 ABD, CD3-high #1). Controls used were XENP24645 (1+1 Fab-scFv-Fc with CD3-high) and XENP24647 (2+1 Fab2-scFv-Fc with CD3-high). Consistent with the binding data, humanization preserved the induction of cytokine secretion by bsAb with humanized CLDN18.2 ABD. [Figure 41A]Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41B] Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41C] Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41D]Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41E] Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Fig.41F] Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41G]Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Fig. 41H] Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Fig.41I] Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41J]Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41K] Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41L] Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 41M]Shown are sequences of several useful 2+1 Fab2-scFv-Fc bispecific antibody formats based on human IgG1, containing scFvs of six different anti-CD3 ABDs in both orientations, as well as sequences that include and exclude the "XTEND®" FcRn variant 428L / 434S (sometimes referred to as the "LS" variant), but exclude the ABDs of other antigens. That is, chain 1 of each set is a -CH1-hinge-CH2-CH3 sequence (the "Fab-Fc side") to which a VH1 sequence can be added (e.g., the C-terminus of VH1 is joined with a slash " / ") to form the complete heavy chain VH1-CH1-hinge-CH2-CH3. Chain 2 of each set is an anti-CD3 scFv-domain linker-CH2-CH3 to which VH1-CH1-optional domain linker- is added (e.g., the C-terminus of the domain linker is joined with a slash " / ") to form the complete chain VH1-CH1-domain linker-scFv-domain linker-CH2-CH3. In this embodiment, the scFv can be in either orientation such that the complete chain is either VH1-CH1-domain linker-VH2-scFv linker-VL2-domain linker-CH2-CH3 or VH1-CH1-domain linker-VL2-scFv linker-VH2-domain linker-CH2-CH3 (note that the sequences in Figure 41 include both options). Chain 3 is an LC domain to which VL1 can be added to form VL1-CL. CDRs are underlined, linkers are double underlined, and domain boundaries are denoted with slashes. It should be noted that all of the chain 2 sequences contain the sequence GGGGSGGGGSKTHTCPPCP (SEQ ID NO: 29), a "flexible half-hinge" domain linker, as the domain linker between the C-terminus of the scFv and the N-terminus of the CH2 domain. However, this linker can be replaced in any of the "useful domain linkers" of FIG. 5 in any of the chain 2 sequences of FIG. 41, in some embodiments replacing SEQ ID NO: 29 with SEQ ID NO: 28, the "full hinge C220S variant". Each of these sequences includes preferred skew, pI, and deletion variants. [Figure 42A] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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 42B] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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 42C] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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.42D] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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 42E] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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.42F]Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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 42G] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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. 42H] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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.42I] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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.42J] Several formats of the invention are shown. The first is a 1+1 Fab-scFv-Fc format with a first and second anti-antigen binding domain. In addition, 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. In addition, 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

[0026] 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, the CD4 T cell subset includes regulatory T cells, whose recruitment and proliferation may result in immune suppression and negatively impact long-term tumor suppression. In addition, these formats do not contain an Fc domain and show a very short serum half-life in patients.

[0027] In particular, anti-CD3, anti-CLDN18.2 bispecific antibodies in various formats are provided herein, such as those shown in Figures 13 and 42. These bispecific antibodies are useful in the treatment of cancer, particularly cancers such as gastric, esophageal, and pancreatic cancer. Such antibodies are used to target CD3+ effector T cells to CLDN18.2+ tumors, thereby enabling the CD3+ effector T cells to attack and lyse CLDN18.2+ tumors.

[0028] In addition, the present invention provides bispecific antibodies with different binding affinities for human CD3 that can modify or reduce potential side effects of anti-CD3 therapy. That is, in some embodiments, the present invention provides antibody constructs comprising an anti-CD3 antigen binding domain that is a "strong" or "high affinity" binder for CD3 (e.g., one example is the heavy and light chain variable domains shown as H1.30_L1.47 (optionally including a charged linker as appropriate)) and also binds to CLDN18.2. In other embodiments, the present invention provides antibody constructs comprising an anti-CD3 antigen binding domain that is a "light" or "low affinity" binder for CD3. Additional embodiments provide antibody constructs comprising an anti-CD3 antigen binding domain with intermediate or "medium" affinity for CD3 that also binds to CD38. While a large number of anti-CD3 antigen binding domains (ABDs) can be used, a particularly useful embodiment uses six different anti-CD3 ABDs, but in two scFv orientations as described herein. Affinity is generally measured using a Biacore assay.

[0029] It is understood that the "high, medium, low" anti-CD3 sequences of the invention can be used in a variety of heterodimerization formats as shown in the figures. In general, due to potential side effects of T cell recruitment, preferred embodiments utilize formats that only bind monovalently to CD3, as shown in Figures 13A and 13B, and in the formats shown herein, it is the CD3 ABD that is the scFv, as described more fully herein. In contrast, bispecific antibodies of the invention can bind CLDN18.2 either monovalently (e.g., Figure 13A) or bivalently (e.g., Figure 13B).

[0030] 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, e.g., CD3 and CLDN18.2, 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). The heterodimeric antibodies provided herein are based on the use of different monomers that contain amino acid substitutions that "skew" the formation of heterodimers on homodimers, as outlined more fully below, and are coupled with "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 a production cell to produce a heterodimeric protein, and methods to generate and purify such heterodimeric proteins.

[0031] C. Nomenclature The bispecific antibodies of the present invention are listed in several different formats. Each polypeptide is given a unique "XENP" number, but longer sequences may contain shorter sequences as understood in the art. For example, the heavy chain of the scFv side monomer of a 1+1 Fab-scFv-Fc format of a given sequence will have a first XENP number, while the scFv domain will have 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 XENP29472, which is a bottle opener format, contains three sequences: XENP29472 chain 1, XENP29472 chain 2, and XENP29472 chain 3 (Figure 31A). These XENP numbers are in the sequence listing as well as the identifiers and are used in the figures. In addition, one molecule containing three components generates multiple sequence identifiers. For example, a list of Fab monomers has the full length sequence, the variable heavy chain sequence, and the three CDRs of the variable heavy chain sequence. The light chain has a full length sequence, a variable light chain sequence, and three CDRs of the variable light chain sequence, and the scFv-Fc domain has a full length sequence, a scFv sequence, a variable light chain sequence, three light chain CDRs, a scFv linker, a variable heavy chain sequence, and three heavy chain CDRs. Note that all molecules herein with scFv domains use a single charged scFv linker (+H), but others can be used. In addition, the nomenclature of a particular variable domain uses a format of "Hx.xx_Ly.yy" type, where the number is a unique identifier of the particular variable chain sequence. Thus, the variable domain on the Fab side of XENP29472 is "H1L1", which indicates that the variable heavy chain domain H1 is combined with the light chain domain L1. When these sequences are used as scFvs, the name "H1L1" indicates that the variable heavy chain domain H1 is combined with the light chain domain L1, in a VH-linker-VL orientation from the N-terminus to the C-terminus. This molecule, with identical sequences of heavy and light chain variable domains but in reverse order, is designated "L1H1". Similarly, as is evident from the sequence listing and figures, different constructs can be "mixed and matched" heavy and light chains.

[0032] D. 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.

[0033] 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.

[0034] 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.

[0035] 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.

[0036] "Antigen binding domain" or "ABD" herein refers to a set of six Complementary Determining Regions (CDRs) that, when present as part of a polypeptide sequence, specifically bind to a target antigen as discussed herein. Thus, a "checkpoint antigen binding domain" binds to a target checkpoint antigen as outlined herein. As known in the art, these CDRs are generally present as a first set of variable heavy chain CDRs (VHCDR or VHCDR) and a second set of variable light chain CDRs (VLCDR or VLCDR), each of which includes three CDRs: VHCDR1, VHCDR2, VHCDR3 for the heavy chain, and VLCDR1, VLCDR2, and VLCDR3 for the light chain. The CDRs are present in the variable heavy chain domain and the variable light chain domain, respectively, and together form the Fv region. (See Table 1 and related discussion above for CDR numbering scheme). Thus, in some cases, the six CDRs of an antigen binding domain are contributed by the variable heavy chain and the variable light chain domain. In the "Fab" format, the set of six CDRs is contributed by two different polypeptide sequences, a variable heavy chain domain (VH or VH, containing VHCDR1, VHCDR2, and VHCDR3), and a variable light chain domain (VL or VL, containing 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 chain domain (thus forming the light chain). In the scFv format, the VH and VL domains are covalently linked into a single polypeptide sequence through the use of a linker ("scFv linker"), generally as outlined herein, which can be either (starting from the N-terminus) VH-linker-VL or VL-linker-VH, with the former generally being preferred (including optional domain linkers on either side, depending on the format used (e.g. from Figure 13). Generally, the C-terminus of the scFv domain is attached to the N-terminus of the hinge of the second monomer.

[0037] 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.

[0038] 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.

[0039] 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. Additionally, -233ADE or A233ADE indicates the insertion of AlaAspGlu after position 233 and before position 234.

[0040] 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.

[0041] 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.

[0042] 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, an "antibody variant" or "variant antibody" as used herein means an antibody that differs from a parent antibody by at least one amino acid modification, an "IgG variant" or "variant IgG" as used herein means an antibody that differs from a parent IgG (again, often derived from a human IgG sequence) by at least one amino acid modification, and an "immunoglobulin variant" or "variant immunoglobulin" as used herein means an immunoglobulin sequence that differs from that of the parent immunoglobulin sequence by at least one amino acid modification. An "Fc variant" or "variant Fc" as used herein means a protein that comprises an amino acid modification in the Fc domain compared to the Fc domain of human IgG1, IgG2, or IgG4.

[0043] The Fc variants of the present invention 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 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, N434S / M428L is the same Fc variant as M428L / N434S, etc. For all positions considered in the present invention with respect to antibodies, the numbering of the amino acid positions is according to the EU index, unless otherwise stated. The EU index, or EU index as in Kabat, or EU numbering scheme, refers to the numbering of EU antibodies. Kabat et al. collected a large number of primary sequences of heavy and light chain variable regions. Based on the degree of sequence conservation, Kabat et al. classified and compiled a list of each primary sequence into CDR and framework (see SEQUENCES OF IMMUNOLOGICAL INTEREST, 5th edition, NIH publication, No. 91-3242, EA Kabat et al., which is incorporated by reference in its entirety). See also Edelman et al., 1969, Proc Natl Acad Sci USA 63:78-85, which is incorporated by reference in its entirety. Modifications can be additions, deletions, or substitutions.

[0044] "Protein" herein means at least two covalently attached amino acids, including proteins, polypeptides, oligopeptides, and peptides. In addition, the polypeptides constituting the antibodies of the present invention may 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.

[0045] 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.

[0046] 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 a bispecific antibody of the invention. In the context of Fab, Fab includes the Fv region in addition to the CH1 and CL domains.

[0047] 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.

[0048] "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 an scFv linker as discussed herein. The scFv domain can be in either orientation (VH-linker-VL or VL-linker-VH) from N-terminus to C-terminus. 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.

[0049] 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.

[0050] 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.

[0051] As used herein, "amino acid" and "amino acid identity" refer to one of the 20 naturally occurring amino acids encoded by DNA and RNA.

[0052] 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.

[0053] 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 homolog (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.

[0054] 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, but is not limited to, FcγRI (CD64), which includes isoforms FcγRIa, FcγRIb, and FcγRIC; FcγRII (CD32), which includes isoforms FcγRIIa (including allotypes H131 and R131), FcγRIIb (including FcγRIIb-1 and FcγRIIb-2), and FcγRIIc; and FcγRIII (CD16), which includes 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.

[0055] 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 β-2-microglobulin, and the heavy chain is encoded by the FcRn gene. Unless otherwise stated herein, FcRn or FcRn protein refers to the complex of FcRn heavy chain and β-2-microglobulin. Various FcRn variants are used to increase binding to the FcRn receptor, and in some cases to increase serum half-life. "FcRn variants" are those that increase binding to the FcRn receptor, and suitable FcRn variants are listed below.

[0056] 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 generate a variant, and a "parent antibody" as used herein refers to an unmodified antibody that is modified to generate a variant antibody. It should be noted 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 related to the listed 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, and so on.

[0057] As used herein, "Fc" or "Fc region" or "Fc domain" refers to a polypeptide comprising the CH2-CH3 domain of an IgG molecule, optionally including the hinge. In the EU numbering of human IgG1, the CH2-CH3 domain comprises amino acids 231-447, and the hinge is 216-230. Thus, the definition of "Fc domain" includes both amino acids 231-447 (CH2-CH3) or 216-447 (hinge-CH2-CH3), or fragments thereof. An "Fc fragment" in this context may contain fewer amino acids at either or both the N-terminus and C-terminus, but still retains the ability to form a dimer with another Fc domain or Fc fragment, typically detectable using standard size-based methods (e.g., non-denaturing chromatography, size exclusion chromatography, etc.). Human IgG Fc domains are particularly useful in the present invention and may be Fc domains from human IgG1, IgG2, or IgG4.

[0058] A "variant Fc region" comprises amino acid modifications compared to the parent Fc domain. Thus, a "variant human IgG1 Fc domain" comprises amino acid modifications (typically amino acid substitutions, although in the case of truncation variants this includes amino acid deletions) compared to the human IgG1 Fc domain. Typically, a variant Fc domain has at least about 80, 85, 90, 95, 97, 98 or 99 percent identity (using the identity algorithm discussed below, using default parameters (in one embodiment utilizing the BLAST algorithm known in the art)) to the corresponding parent human IgG Fc domain. Alternatively, a 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 compared 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. In addition, as discussed herein, the variant Fc domains herein still retain the ability to form dimers with another Fc domain as measured using known techniques described herein, such as non-denaturing gel electrophoresis.

[0059] 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.

[0060] 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.

[0061] "Target antigen," as used herein, refers to a molecule that is specifically bound by an antibody binding domain that comprises the variable region of a given antibody. As discussed below, in this case, the target antigen is a checkpoint inhibitor protein.

[0062] "Strandedness" in the context of the monomers of the heterodimeric antibodies of the invention herein refers to the incorporation of heterodimerization variants into each monomer such that they retain the ability to "match" and form heterodimers, similar to the two strands of 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 will not interfere with the pI variant, e.g., the pI-increased charge variant will be 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 the pI into account when deciding which strand or monomer one set of pairs will go into, so that the pI separation is also maximized using the skewed pI.

[0063] As used herein, "target cell" means a cell that expresses a target antigen.

[0064] By "host cell" in the context of producing a bispecific antibody according to the invention 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.

[0065] 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 includes 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 the amino terminus to the carboxy terminus in the following order: FR1-CDR1-FR2-CDR2-FR3-CDR3-FR4.

[0066] 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.

[0067] The present invention provides several antigen-binding domains that have sequence identity with human antibody domains. Sequence identity between two similar sequences (e.g., antibody variable domains) can be determined by 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 Algorithm." Tool," J. Mol. Biol. 215:403-10 ["BLAST" algorithm (see https: / / blast.ncbi.nlm.nih.gov / Blast.cgi). When using any of the aforementioned algorithms, default parameters (window length, gap penalties, etc.) are used. In one embodiment, sequence identity is measured using the BLAST algorithm using default parameters.

[0068] The antibodies of the present invention are generally isolated or recombinant. "Isolated" when 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. Usually, 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 they may also be isolated.

[0069] "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.

[0070] 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.

[0071] 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.

[0072] E. Antibodies In one aspect, provided herein are bispecific antibodies that bind to CLDN18.2 and CD3 in various formats as outlined below and generally depicted in Figures 13 and 42. These bispecific heterodimeric antibodies comprise a CLDN18.2 binding domain. In certain embodiments, the CLDN18.2 binding domain comprises the VHCDR1, VHCDR2, VHCDR3, VLCDR1, VLCDR2, and VLCDR3 sequences of a CLDN18.2 binding domain selected from the group consisting of those depicted in Figure 10. In some embodiments, the CLDN18.2 binding domain comprises the underlined VHCDR1, VHCDR2, VHCDR3, VLCDR1, VLCDR2, and VLCDR3 sequences of a CLDN18.2 binding domain selected from those depicted in Figure 10.

[0073] These bispecific heterodimeric antibodies bind to CLDN18.2 and CD3. Such antibodies comprise a CD3 binding domain and at least one CLDN18.2 binding domain. Any suitable CLDN18.2 binding domain can be included in the anti-CLDN18.2×anti-CD3 bispecific antibody. In some embodiments, the anti-CLDN18.2×anti-CD3 bispecific antibody comprises one, two, three, four or more CLDN18.2 binding domains, including but not limited to those depicted in FIG. 10. In certain embodiments, the anti-CLDN18.2×anti-CD3 antibody comprises a CLDN18.2 binding domain comprising the VHCDR1, VHCDR2, VHCDR3, VLCDR1, VLCDR2, and VLCDR3 sequences of a CLDN18.2 binding domain selected from the group consisting of those depicted in FIG. 10. In some embodiments, the anti-CLDN18.2x anti-CD3 antibody comprises a CLDN18.2 binding domain comprising the underlined VHCDR1, VHCDR2, VHCDR3, VLCDR1, VLCDR2, and VLCDR3 sequences of a CLDN18.2 domain selected from the group consisting of those depicted in Figure 10. In some embodiments, the anti-CLDN18.2x anti-CD3 antibody comprises a CLDN18.2 binding domain comprising the variable heavy and variable light domains of a CLDN18.2 binding domain selected from the group consisting of those depicted in Figure 10. In exemplary embodiments, the anti-CLDN18.2x anti-CD3 antibody comprises an anti-CLDN18.2 H1L1 or H2L1 binding domain.

[0074] The anti-CLDN18.2x anti-CD3 antibodies provided herein can comprise any suitable CD3 binding domain. In certain embodiments, the anti-CLDN18.2x anti-CD3 antibodies comprise a CD3 binding domain comprising the VHCDR1, VHCDR2, VHCDR3, VLCDR1, VLCDR2, and VLCDR3 sequences of a CD3 binding domain selected from the group consisting of those depicted in Figure 12. In some embodiments, the anti-CLDN18.2x anti-CD3 antibodies comprise a CD3 binding domain comprising the underlined VHCDR1, VHCDR2, VHCDR3, VLCDR1, VLCDR2, and VLCDR3 sequences of a CD3 binding domain selected from the group consisting of those depicted in Figure 12. In some embodiments, the anti-CLDN18.2x anti-CD3 antibodies comprise a CD3 binding domain comprising the variable heavy chain domain and the variable light chain domain of a CD3 binding domain selected from the group consisting of those depicted in Figure 12. In some embodiments, the CD3 binding domain is selected from anti-CD3 H1.30_L1.47, anti-CD3 H1.32_L1.47, anti-CD3 H1.89_L1.48, anti-CD3 H1.90_L1.47, anti-CD3 H1.33_L1.47, and anti-CD3 H1.31_L1.47. As outlined herein, these anti-CD3 antigen binding domains (CD3-ABD) can be used in either orientation of the scFv format (e.g., from N-terminus to C-terminus, VH-scFv linker-VL or VL-scFv linker-VH).

[0075] As used herein, the term "antibody" is used generically. Antibodies used in the present invention can take several formats, as described herein, including conventional antibodies, as well as antibody derivatives, fragments, and mimetics, as described herein.

[0076] Conventional antibody structural units typically comprise a tetramer. Each tetramer typically consists of two identical pairs of polypeptide chains, each pair having one "light" chain (typically with a molecular weight of about 25 kDa) and one "heavy" chain (typically with a molecular weight of about 50-70 kDa). Human light chains are classified as kappa light chains and lambda light chains. The present invention is directed to the IgG class, which has several subclasses, including but not limited to IgG1, IgG2, IgG3, and IgG4. It should be noted that IgG1 has different allotypes with polymorphisms at 356 (D or E) and 358 (L or M). The sequences shown 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.

[0077] In addition, many of the antibodies herein have at least one cysteine ​​at position 220 replaced by a serine. Generally, this 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).

[0078] 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 can also include hybrids of isotypes and / or subclasses. For example, the present invention includes the use of human IgG1 / G2 hybrids, as shown in U.S. Publication No. 2009 / 0163699, which is incorporated by reference.

[0079] 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 described below.

[0080] 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 is as follows, see Lafranc et al., Dev. Comp. Immunol. 27(1):55-77 (2003).

[0081] [Table 1]

[0082] 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).

[0083] 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 comprise positions 216 (E216 in IgG1) to 230 (p230 in IgG1), with numbering according to the EU index as in Kabat. In some cases, "hinge fragments" are used that include fewer amino acids at either or both the N-terminus and C-terminus of the hinge domain. As discussed below, the hinge may also be a domain linker. In these embodiments, useful hinge-based domain linkers are shown in Figure 5, and in particular SEQ ID NO: 28 may be particularly used in "1+1 Fab-scFv-Fc" as well as "2+1 Fab2-scFv-Fc" formats. As described herein, pI variants can be made in the hinge region as well.

[0084] A light chain generally comprises two domains: a variable light domain (which contains the light chain CDRs and, together with the variable heavy domain, forms the Fv region), and a constant light region (often referred to as CL or Cκ).

[0085] Another region of interest for additional substitutions, as outlined below, is the Fc region.

[0086] The present invention provides a number of different CDR sets. In this case, a "complete CDR set" includes three variable light chain and three variable heavy chain CDRs, e.g., 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.

[0087] CDRs contribute to antigen binding, or more specifically, to the formation of the 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.

[0088] An epitope can 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, such as amino acid residues that are effectively blocked by the specific antigen-binding peptide, in other words, amino acid residues that are within the footprint of the specific antigen-binding peptide.

[0089] Epitopes may be either conformational or linear. Conformational epitopes are generated by spatially juxtaposed 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.

[0090] 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 invention 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.

[0091] Thus, the present invention provides different antibody domains. As described herein and known in the art, the heterodimeric antibodies of the present invention comprise 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.

[0092] Thus, an "Fc domain" comprises a -CH2-CH3 domain, and optionally a hinge domain (-hinge domain-CH2-CH3) (again, many embodiments rely on the hinge domain from human IgG1 with the C220S variant). In embodiments herein, particularly for the "1+1" format, when an scFv is linked to an Fc domain, the C-terminus of the scFv construct is linked to all or part of the hinge of the Fc domain. For example, it is generally linked to the sequence EPKS (SEQ ID NO: 473), which is the start of the hinge. In some cases, for example in the "2+1" format, the domain linker can be a combination of flexible linker amino acids as well as part or the entire hinge. For example, SEQ ID NO: 29 is such an example.

[0093] Embodiments of the invention include at least one scFv domain, which is not naturally occurring, but generally includes a variable heavy domain and a variable light domain linked together by an scFv linker. As outlined herein, the scFv domain is generally oriented from 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 either or both ends depending on the format (see generally Figures 13 and 42).

[0094] (a) Linker As shown herein, there are several suitable linkers (used as either domain linkers or scFv linkers) that can be used to covalently link the enumerated domains, including traditional peptide bonds generated by recombinant techniques. 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 each other 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, a linker of 1-20 amino acids long may be used, and in some embodiments, about 5 to about 10 amino acids are used. Useful linkers include, for example, glycine-serine polymers, glycine-alanine polymers, alanine-serine polymers, including (GS), (GSGGS) (SEQ ID NO: 474), (GGGGS) (SEQ ID NO: 475), and (GGGS) (SEQ ID NO: 476), 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 Figure 5. Alternatively, a variety of non-proteinaceous polymers can be used as linkers, including, but not limited to, polyethylene glycol (PEG), polypropylene glycol, polyoxyalkylenes, or copolymers of polyethylene glycol and polypropylene glycol.

[0095] Other linker sequences may include any sequence of any length of the CL / CH1 domain, such as the first 5-12 amino acid residues of the CL / CH1 domain, but not all residues of the CL / CH1 domain. Linkers may be derived from immunoglobulin light chains, such as Cκ or Cλ. Linkers may be derived from immunoglobulin heavy chains of any isotype, including, for example, γ1, Cγ2, Cγ3, Cγ4, Cα1, Cα2, Cδ, Cε, and Cμ. Linker sequences may also be derived from other proteins, such as Ig-like proteins (e.g., TCR, FcR, KIR), sequences derived from hinge regions, and other naturally occurring sequences from other proteins.

[0096] In some embodiments, the linker is a "domain linker" used to link together any two domains as outlined herein. For example, in FIG. 42F, there may be a domain linker connecting the C-terminus of the CH1 domain of the Fab to the N-terminus of the scFv, and another optional domain linker connecting 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: 474), (GGGGS)n (SEQ ID NO: 475), and (GGGS)n (SEQ ID NO: 476), 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.

[0097] With particular reference to domain linkers used to join scFv domains to Fc domains in a "2+1" format, there are several domain linkers that are of particular use, including SEQ ID NO:28, SEQ ID NO:29, SEQ ID NO:30, SEQ ID NO:31.

[0098] 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, the present invention further provides 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 using 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. Also, 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.

[0099] Charged domain linkers can also be used to increase the pI separation of the monomers of the present invention, and thus those included in FIG. 5 can be used in any embodiment herein in which a linker is utilized.

[0100] Specifically, the format shown in FIG. 1 is an antibody that is typically 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.

[0101] F. Chimeric and Humanized Antibodies In certain embodiments, the antibodies of the invention 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 (the number of variants is generally small prior to the introduction of any skew, pI, and deletion variants herein, i.e., prior to the introduction of the variants of the invention).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 the germline immunoglobulin gene (again, prior to the introduction of the skew, pI, and deletion variants herein, i.e., prior to the introduction of the variants of the present invention, the number of variants is generally small).

[0102] 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 in their entireties. 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, which are incorporated by reference in their entireties.

[0103] G. Heterodimeric Antibodies Thus, in some embodiments, the subject antibodies are heterodimeric antibodies that rely on the use of two different heavy chain variant Fc sequences that will self-assemble to form a heterodimeric Fc domain and a heterodimeric antibody.

[0104] The present invention is directed to novel constructs to provide heterodimeric antibodies that allow binding to more than one antigen or ligand, for example to allow bispecific binding (e.g., anti-CLDN18.2 binding and anti-CD3 binding). Heterodimeric antibody constructs are based on the self-assembly of two Fc domains of the heavy chains of an antibody, e.g., two "monomers" that assemble into a "dimer". Heterodimeric antibodies are created by varying the amino acid sequence of each monomer, as discussed more fully below. Thus, the present invention is generally directed to the creation of heterodimeric antibodies, which can simultaneously link antigens (e.g., CLDN18.2 and CD3) in several ways, relying on amino acid variants in the constant regions that are different on each chain to promote heterodimer formation and / or to facilitate purification of the heterodimer over homodimers.

[0105] Thus, the present invention provides bispecific antibodies. In some embodiments, the present invention provides bispecific antibodies comprising a CLDN18.2 binding domain. In some embodiments, the bispecific antibody is an anti-CLDN18.2 x anti-CD3 bispecific antibody. 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 in purifying heterodimeric antibodies away from homodimeric antibodies and / or biasing towards heterodimer formation over homodimer formation.

[0106] There are several mechanisms that can be used to generate the heterodimers of the present invention. In addition, as will be appreciated by those skilled in the art, these mechanisms can be combined to ensure high heterodimerization. Thus, amino acid variants that result in the production of heterodimers are referred to as "heterodimerization variants." As discussed below, heterodimerization variants can include conformational variants (e.g., "knob-and-hole" or "skew" variants, and "charge pair" variants, as described below) as well as "pI variants," which allow for purification of homodimers away from heterodimers. As generally described in WO 2014 / 145806, which is incorporated by reference in its entirety, and specifically described below for the discussion of "heterodimerization variants," useful mechanisms for heterodimerization include "knobs and holes" (sometimes referred to herein as "skew" variants (see discussion in WO 2014 / 145806), "electrostatic steering" or "charge pairing" as described in WO 2014 / 145806, the pI variants described in WO 2014 / 145806, as well as general additional Fc variants as outlined in WO 2014 / 145806 and below).

[0107] In the present invention, there are several basic mechanisms that can facilitate the purification of heterodimeric antibodies. One relies on the use of pI variants, where each monomer has a different pI, thus allowing isoelectric purification of AA, AB and BB dimeric proteins. Alternatively, some scaffolds, such as the "1+1 Fab-scFv-Fc" format, also allow for size-based separation. As further outlined below, it is also possible to "skew" the formation of heterodimers over homodimers. Thus, the combination of conformational heterodimerization variants and pI variants or charge pair variants is of particular use in the present invention.

[0108] In general, embodiments specifically used in the present invention rely 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, facilitating purification of the heterodimer away from the homodimer.

[0109] In addition, depending on the format of the heterodimeric antibody, as outlined more fully below, either the pI variants can be included within the constant domains and / or Fc domains of the monomers, or a charged linker can be used, either a domain linker or a scFv linker. That is, scaffolds utilizing scFv, such as the "1+1 Fab-scFv-Fc" format, can 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, some 1+1 Fab-scFv-Fc formats are useful with only a charged scFv linker and no additional pI adjustment, but the present invention also provides pI variants in one or both of the monomers, and / or charged domain linkers. In addition, additional amino acid engineering for alternative functionality can also impart pI changes, such as Fc, FcRn, and KO variants.

[0110] In the present invention, which utilizes pI as a separation mechanism to allow purification of heterodimeric proteins, amino acid variants can be 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, the pI of either or both monomers can be altered by removing or adding a charged residue (e.g., replacing a neutral amino acid with a positively or negatively charged amino acid residue, e.g., glycine to glutamic acid), changing a charged residue from positive or negative to the opposite charge (aspartic acid to lysine), or changing a charged residue to a neutral residue (e.g., loss of charge, lysine to serine). Some of these variants are shown in the figures.

[0111] Thus, this embodiment of the invention provides for creating a sufficient pI change in at least one monomer such that heterodimers can be separated from homodimers. As will be appreciated by those of skill in the art and discussed further below, this can be done by using a "wild type" heavy chain constant region and a variant region engineered to increase or decrease its pI (wt A-+B+ or wt A--B), or by increasing one region and decreasing the other (A+-B- or A-B+).

[0112] Thus, in general, a component of some embodiments of the present invention are amino acid variants in the constant region of an antibody, intended to modify the isoelectric point (pi) of at least one, if not both, of the monomers of a dimeric protein by incorporating an amino acid substitution ("pi variant" or "pi substitution") into one or both of the monomers to form a "pi antibody"). As shown herein, separation of a heterodimer from 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 of use in the present invention.

[0113] As will be appreciated by those of skill in the art, the number of pI variants 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., in a 1+1 Fab-scFv-Fc format, the starting pI of the scFv and Fab of interest. That is, to determine which monomer 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 from there. As known in the art, different Fvs will have different starting pIs to be utilized in the present invention. 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.

[0114] Additionally, as will be appreciated by those of skill in the art and outlined herein, in some embodiments, heterodimers can be separated from homodimers based on size. For example, some of the formats allow for separation of heterodimers and homodimers based on size, as shown in FIG.

[0115] 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, dimerization variants (including scubariants or purified dimerization variants) are not included in the variable region, so each individual antibody must be engineered. In addition, in some embodiments, the possibility of 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, an additional 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.

[0116] Potential collateral benefits of this pI engineering are also increased serum half-life and increased FcRn binding. That is, as described in U.S. Patent Application Serial No. 13 / 194,904 (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.

[0117] In addition, it should be noted that the pI variants of the dimerization variants provide additional benefits to the analysis and quality control process of bispecific antibodies, due to their remarkable ability to eliminate, minimize, or distinguish when homodimers are present. Similarly, the ability to reliably test the reproducibility of heterodimeric antibody production is important.

[0118] Heterodimerization variants The present invention provides heterodimeric proteins, including heterodimeric antibodies in a variety of formats, that utilize heterodimerization variants to enable heterodimer formation and / or purification from homodimers.

[0119] There are several suitable pairs of sets of heterodimerizing scubariants. These variants are "pairs" of "sets"; that is, a pair from one set is incorporated into the first monomer and a pair from the other set is incorporated into the second monomer. It should be noted that these sets do not necessarily behave as "knobs-in-holes" variants, but rather there is a one-to-one correspondence between residues on one monomer and residues on the other monomer. That is, these pairs of sets form an interface between the two monomers that promotes the formation of heterodimers and does not promote the formation of homodimers, allowing the percentage of heterodimers that form spontaneously under biological conditions to exceed 90%, rather than the expected 50% (25% homodimer A / A: 50% heterodimer A / B: 25% homodimer B / B).

[0120] Stereoscopic variants In some embodiments, heterodimer formation can be promoted by the addition of conformational variants, i.e., by varying the amino acids in each heavy chain, different heavy chains are more likely to associate to form heterodimeric structures than to form homodimers with the same Fc amino acid sequence. Suitable conformational variants are included in FIG. 1.

[0121] One mechanism may optionally be 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 also 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; U.S. Patent No. 8,216,805, all of which are incorporated herein by reference in their entireties. These figures identify several "monomer A-monomer B" pairs that rely on "knobs and holes." In addition, these "knob-and-hole" mutations can be combined with disulfide bonds to distort formation towards heterodimerization, as described by Merchant et al., Nature Biotech. 16:677 (1998).

[0122] The additional mechanism used to generate heterodimers is sometimes 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 is sometimes referred to herein as "charge pairing". In this embodiment, electrostatics are used to skew the formation towards 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 means, 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.

[0123] 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.

[0124] In some embodiments, the conformational variants outlined herein can optionally and independently incorporate any pI variants (or other variants such as Fc variants, FcRn variants, etc.) into one or both monomers and can optionally and independently be included or excluded from the proteins of the invention.

[0125] A list of suitable scubariants can be found in Figure 1, and Figure 4 shows some pairs that are particularly useful in many embodiments. 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 K370S: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.

[0126] Heterodimer pI variants 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 differs significantly 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.

[0127] 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, however, all isotypes can be altered 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.

[0128] In one embodiment, for example in the format of Figures 42A, E, F, G, H, and I, a preferred combination of pI variants has one monomer (negative Fab side) that comprises the 208D / 295E / 384D / 418E / 421D variant (N208D / Q295E / N384D / Q418E / N421D for human IgG1), and a second monomer (positive scFv side) that comprises a positively charged scFv linker that comprises (GKPGS)4 (SEQ ID NO: 10). However, as will be appreciated by those skilled in the art, the first monomer comprises a CH1 domain that comprises position 208. Thus, in constructs that do not include a CH1 domain (e.g., for antibodies that do not utilize a CH1 domain on one of the domains, e.g., in a dual scFv format or a "one-arm" format such as those shown in Figure 42B, C, or D), a preferred negative pI variant Fc set includes the 295E / 384D / 418E / 421D variant (for human IgG1, Q295E / N384D / Q418E / N421D).

[0129] Thus, in some embodiments, one monomer has a set of substitutions from Figure 2 and the other monomer has a charged linker (either a charged scFv linker format as that monomer comprises an scFv or charged domain linker as the format dictates, which can be selected from those shown in Figure 5).

[0130] Isotype variants In addition, many embodiments of the invention 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 undesired immunogenicity into the variant. Some of these are shown in FIG. 21 of US Patent Application Publication No. 2014 / 0370013, incorporated herein by reference. Namely, 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 in addition exhibits a longer serum half-life. For example, IgG1 has glycine (pI 5.97) at position 137, while 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.

[0131] 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.

[0132] 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 differing by at least 0.5 can allow for separation by ion exchange chromatography or isoelectric focusing, or other methods sensitive to isoelectric point.

[0133] Calculate 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.

[0134] pI variants that also confer better FcRn binding in vivo If the pI variants reduce the pI of the monomer, they may have the added advantage of improving serum retention in vivo.

[0135] Although still under investigation, it is believed that Fc regions have a longer half-life in vivo because Fc is sequestered upon binding to FcRn at pH 6 in endosomes (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 back into the blood. In mice, Dall'Acqua et al. showed that Fc variants with increased FcRn binding at pH 6 and pH 7.4 actually have reduced serum concentrations and the same 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 back 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.

[0136] 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.

[0137] Additional Fc variants for additional functionality In addition to pI amino acid variants, 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, altering binding to the FcRn receptor, etc.

[0138] Thus, the proteins of the invention can include amino acid modifications including the heterodimerization variants outlined herein, including pI variants and conformational variants, each set of variants can independently and optionally include or exclude from any particular heterodimeric protein.

[0139] FcγR variants Thus, there are several useful Fc substitutions that can be made to alter binding to one or more of the FcγR receptors. Substitutions that result in increased binding as well as decreased binding can be useful. For example, increased binding to FcγRIIIa is known to generally increase 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 be beneficial in some circumstances. Amino acid substitutions for use in the present invention include those listed in U.S. Patent Application No. 11 / 124,620 (particularly FIG. 41), U.S. Patent Application No. 11 / 174,287, U.S. Patent Application No. 11 / 396,495, and U.S. Patent Application No. 11 / 538,406, all of which are expressly incorporated herein by reference in their entirety, and specifically for the variants disclosed therein. Specific 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.

[0140] Additionally, as specifically disclosed in U.S. Patent Application 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, 259I / 308F, 436I / 428L, 436I or V / 434S, 436V / 428L, and 259I / 308F / 428L.

[0141] Removal variant Similarly, another category of functional variants are "FcγR-depleted variants" or "Fc knock out (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 eliminate FcγRIIIa binding to eliminate or significantly reduce ADCC activity, and one of the Fc domains comprises one or more Fcγ receptor-depleted variants. These deletion variants are shown in Figure 3 and each can be independently and optionally included or excluded, with preferred embodiments utilizing deletion variants 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. It should be noted that the deletion variants referred to herein eliminate FcγR binding, but generally do not eliminate FcRn binding.

[0142] 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.

[0143] Heterodimer and Fc variant combinations As will be appreciated by one 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." Additionally, all of these variants can be combined in any of the heterodimerization formats.

[0144] In the case of pI variants, while specific embodiments that may be used 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.

[0145] 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.

[0146] H. Useful Formats of the Invention As will be appreciated by those of skill in the art and discussed more fully below, the heterodimeric fusion proteins of the invention can adopt a wide variety of configurations, as generally shown in Figures 13 and 42. 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, the present invention relates to novel immunoglobulin compositions that simultaneously link different first or second antigens.

[0147] As will be understood 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 present invention 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., CLDN18.2) 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.

[0148] The present invention utilizes an anti-CD3 antigen binding domain in combination with an anti-CLDN18.2 binding domain. As will be appreciated by one of skill 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-CLDN18.2 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 (e.g., Figures 8-10).

[0149] 1+1 Fab-scFv-Fc format One heterodimeric scaffold of particular use in the present invention is the "1+1 Fab-scFv-Fc" format of Figures 13A and 42A. In this embodiment, one heavy chain of the antibody comprises a single chain Fv ("scFv" as defined herein) and the other heavy chain is a "regular" Fab format comprising a variable heavy chain and a light chain. This structure has also been referred to in previous related applications as the "triple F" format (scFv-Fab-Fc) or the "bottle opener" format due to its rough visual resemblance to a bottle opener. The two chains are brought together by the use of amino acid variants in the constant regions (e.g., Fc domain, CH1 domain and / or hinge region) that promote the formation of heterodimeric antibodies, as described more fully below.

[0150] There are some distinct advantages to the current "1+1 Fab-scFv-Fc" format. As is known in the art, antibody analogs that rely on two scFv constructs often have stability and aggregation problems, which can be mitigated in the present invention by adding "normal" heavy and light chain pairing. 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).

[0151] Many of the embodiments outlined herein generally rely on a bottle opener format antibody comprising a first monomer comprising an scFv, which comprises variable heavy and variable light domains covalently linked using an scFv linker (charged in many, but not all, cases), with the scFv usually being covalently linked to the N-terminus of a first Fc domain through a domain linker (which can be either charged or uncharged, as outlined herein). The second monomer in the bottle opener format is a heavy chain, and the composition further comprises a light chain.

[0152] Generally, in many preferred embodiments, the scFv is the CD3 binding domain and the Fab forms the CLDN18.2 binding domain.

[0153] In addition, the Fc domains of the invention comprise scFv variants (e.g., the set of amino acid substitutions shown in Figures 1 and 4, where particularly useful scFv variants are selected from the group consisting of S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K / E357L, and K370S:S364K / E357Q), optionally deletion variants (including those shown in Figure 3), optionally charged scFv linkers (including those shown in Figure 5), and the heavy chain comprises pI variants (including those shown in Figure 2).

[0154] In some embodiments, the bottle opener format includes scFv variants, pI variants, and truncation variants. Thus, some embodiments include a) a first monomer ("scFv monomer") comprising a charged scFv linker (in some embodiments the +H sequence of FIG. 5 is preferred), scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, and an Fv that binds CD3 as outlined herein; b) scFv variant L368D / K370 S, a second monomer ("Fab monomer") comprising a variable heavy chain domain that constitutes an Fv that binds to CLDN18.2 as outlined herein, together with pI variants N208D / Q295E / N384D / Q418E / N421D, deletion variants E233P / L234V / L235A / G236del / S267K, and a variable light chain domain, and c) a light chain.

[0155] Exemplary variable heavy and light domains of an scFv that binds CD3 are included in Figure 12. Exemplary variable heavy and light domains of an Fv that binds CLDN18.2 are included in Figure 10.

[0156] In some embodiments, the bottle opener format includes scFv variants, pI variants, truncation variants, and FcRn variants. Thus, some embodiments include a) a first monomer ("scFv monomer") comprising a charged scFv linker (in some embodiments the +H sequence of FIG. 2 is preferred), scFv variant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variant M428L / N434S, and an Fv that binds CD3 as outlined herein; b) a scFv variant L368D / K370 S, pI variants N208D / Q295E / N384D / Q418E / N421D, deletion variants E233P / L234V / L235A / G236del / S267K, FcRn variant M428L / N434S, and a second monomer ("Fab monomer") comprising a variable heavy chain domain that constitutes an Fv that binds to CLDN18.2 as outlined herein, together with a variable light chain domain, and c) a light chain.

[0157] Exemplary variable heavy and light domains of an scFv that binds CD3 are included in Figure 12. Exemplary variable heavy and light domains of an Fv that binds CLDN18.2 are included in Figure 10.

[0158] Figure 6 shows some exemplary bottle opener "skeleton" sequences lacking Fv sequences that can be used in the present invention. 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 to CLDN18.2) can be added as the "Fab side" to the bottle opener scaffold format of Figure 6 using any of the anti-CD3 scFv sequences shown in the figures and sequence listing.

[0159] For bottle opener scaffold 1 from FIG. 6 (optionally including the 428L / 434S variant), CD binding domain sequences of 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, as well as those shown in FIG. 12 attached as the scFv side of the scaffold shown in FIG. 6.

[0160] A particularly useful CLDN18.2 and CD3 sequence combination for use with bottle opener backbone 1 from FIG. 6 (optionally including the 428L / 434S variant) is disclosed in FIG.

[0161] mAb-Fv One heterodimeric scaffold of particular use in the present invention is the mAb-Fv format shown in Figure 42G. 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, where the Fab portions of the two monomers bind to CLDN18.2 and the "additional" scFv domain binds to CD3.

[0162] In this embodiment, the first monomer comprises a first heavy chain comprising a first variable heavy domain and a first constant heavy domain comprising a first Fc domain, and has a first variable light 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 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 that associates with the heavy chain to form two identical Fabs that bind CLDN18.2. For many of the embodiments described herein, these constructs include scubariants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0163] The present invention provides a mAb-Fv format in which the CD3 binding domain sequence is as shown in Figure 12. The present invention provides a mAb-Fv format in which the CLDN18.2 binding domain sequence is as shown in Figure 10.

[0164] Additionally, the Fc domain of the mAb-Fv format may be modified with a sqvariant (e.g., the set of amino acid substitutions shown in Figures 1 and 4, where particularly useful sqvariants are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D / K370S:S364K / E35 5, wherein the heavy chain comprises a pI variant (including those shown in FIG. 2), 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).

[0165] In some embodiments, the mAb-Fv format comprises scubariants, pI variants, and deletion variants. Thus, some embodiments comprise a mAb-Fv format, which comprises: a) a first monomer comprising a scubariant S364K / E357Q, deletion variants E233P / L234V / L235A / G236del / S267K, and a first variable light domain of the light chain, together with a first variable heavy domain and a second variable heavy domain, which constitute an Fv that binds to CLDN18.2; b) a scubariant L368D / K370S, pI variant N208D, and a second variable heavy domain; / Q295E / N384D / Q418E / N421D, deletion variants E233P / L234V / L235A / G236del / S267K, and a first variable heavy chain domain which, together with the first variable light chain domain, constitutes an Fv that binds to CLDN18.2 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.

[0166] 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 that constitutes an Fv that binds to CLDN18.2, and a second variable heavy chain domain, together with a first variable light chain domain of the light chain, and a second variable heavy chain domain, b) a scubariant L368D / K370S, pI variants N208D / Q29 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 CLDN18.2 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 to CD3; and c) a light chain comprising the first variable light chain domain and a constant light chain domain.

[0167] mAb-scFv A heterodimeric scaffold of particular use in the present invention is the mAb-scFv format shown in Figure 42H. 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 CLDN18.2 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 the 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 that associates with the heavy chain to form two identical Fabs that bind to CLDN18.2. For many of the embodiments described herein, these constructs include scubariants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0168] The present invention provides a mAb-scFv format in which the CD binding domain sequence is as shown in FIG. 12 and the CLDN18.2 binding domain sequence is as shown in FIG.

[0169] In addition, the Fc domain of the mAb-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 / 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).

[0170] 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 CLDN18.2 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, together with a variable heavy chain domain constituting an Fv that binds CLDN18.2 as outlined herein, and an scFv domain that binds CD3; a) a common light chain comprising a variable light chain domain and a constant light chain domain, and b) a variable heavy chain domain comprising the variable light chain domain and the constant light chain domain, together with the variable light chain domain and the constant light chain domain.

[0171] 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 CLDN18.2 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 CLDN18.2 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.

[0172] 2+1 Fab2-scFv-Fc format One heterodimeric scaffold of particular use in the present invention is the "2+1 Fab2-scFv-Fc" format (also referred to as the "central scFv format" in earlier related applications) shown in Figure 13B and Figure 42F. 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 CLDN18.2 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.

[0173] 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 other monomer is a standard Fab side. This embodiment further utilizes a common light chain comprising a variable light domain and a constant light domain that associates with the heavy chain to form two identical Fabs that bind to CLDN18.2. For many of the embodiments described herein, these constructs include scubariants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0174] The present invention provides a "2+1 Fab2-scFv-Fc" format, in which the CD3 binding domain sequence is as shown in FIG. 12 and the anti-CLDN 18.2 sequence is as shown in FIG.

[0175] Additionally, the Fc domain of the 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:S ... 5, wherein the heavy chain comprises a pI variant (including those shown in FIG. 2), 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).

[0176] In some embodiments, the central scFv format includes scFv variants, pI variants, and truncation variants. Thus, some embodiments include a) a first monomer comprising a scFv domain that binds to CD3, together with a scFv variant S364K / E357Q, a deletion variant E233P / L234V / L235A / G236del / S267K, and a variable heavy chain domain that constitutes an Fv that binds to CLDN18.2 as outlined herein, together with a variable light chain domain of the light chain; b) a first monomer comprising a scFv domain that binds to CD3, together with a scFv variant L368D / K370S, a pI variant E233P / L234V / L235A / G236del / S267K, and a variable light chain domain of the light chain; a) a second monomer comprising a variable heavy chain domain constituting an Fv that binds to CLDN18.2 as outlined herein together with variants N208D / Q295E / N384D / Q418E / N421D, deletion variants E233P / L234V / L235A / G236del / S267K, and the variable light chain domain of the light chain; and c) a central scFv format comprising a light chain comprising a variable light chain domain and a constant light chain domain.

[0177] In some embodiments, the central scFv format includes scFv variants, pI variants, truncation variants, and FcRn variants. Thus, some embodiments include a) a first monomer comprising a scFv variant S364K / E357Q, a deletion variant E233P / L234V / L235A / G236del / S267K, an FcRn variant M428L / N434S, and a variable heavy chain domain constituting an Fv that binds CLDN18.2 as outlined herein together with a variable light chain domain of the light chain, and an scFv domain that binds CD3; b) a first monomer comprising a scFv variant L368D / K370S, a second monomer comprising the variable heavy chain domain constituting an Fv that binds to CLDN18.2 as outlined herein together with the pI variants N208D / Q295E / N384D / Q418E / N421D, the deletion variants E233P / L234V / L235A / G236del / S267K, the FcRn variant M428L / N434S, and the variable light chain of the light chain; and c) a central scFv format comprising a light chain comprising a variable light chain domain and a constant light chain domain.

[0178] Central Fv One heterodimeric scaffold of particular use in the present invention is the central Fv format shown in Figure 42I. In this embodiment, the format relies on the use of an inserted Fv domain (i.e., a central Fv domain) thus forming a third antigen-binding domain where the Fab portions of the two monomers bind to CLDN18.2 and the "central Fv domain" binds to CD3. The scFv domain is inserted between the Fc domain and the CH1-Fv region of the monomer thus providing a third antigen-binding domain where each monomer comprises components of an scFv (e.g., one monomer comprises a variable heavy chain domain and the other comprises a variable light chain domain).

[0179] 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 chain 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 chain 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.

[0180] This embodiment further utilizes a common light chain comprising a variable light domain and a constant light domain that associates with the heavy chain to form two identical Fabs that bind to CLDN18.2. For many of the embodiments described herein, these constructs include scubariants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0181] The present invention provides a central Fv format, in which the CD3 binding domain sequence is as shown in FIG. 12 and the CLDN18.2 binding domain sequence is as shown in FIG.

[0182] For the central Fv format, CD3 binding domain sequences of particular use in these embodiments 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, and anti-CD3 H1.31_L1.47, as shown in FIG. 12.

[0183] OneArmCentral scFv One heterodimeric scaffold of particular use in the present invention is the one-arm central scFv format shown in Figure 42C. In this embodiment, one monomer contains only the Fc domain, while the other monomer uses an inserted scFv domain, thus forming the second antigen-binding domain. In this format, the Fab portion binds to CLDN18.2 and the scFv binds to CD3, or vice versa. The scFv domain is inserted between the Fc domain and the CH1-Fv region of one of the monomers.

[0184] 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 using a domain linker between the C-terminus of the CH1 domain of the heavy chain constant domain and the N-terminus of the first Fc domain. The second monomer comprises an Fc domain. This embodiment further utilizes a light chain comprising a variable light domain and a constant light domain that associates with the heavy chain to form a Fab. For many of the embodiments described herein, these constructs include scavariant, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0185] The present invention provides a central Fv format, in which the CD3 binding domain sequence is as shown in FIG. 12 and the CLDN18.2 binding domain sequence is as shown in FIG.

[0186] In addition, the Fc domain of the one-arm central scFv format generally comprises a scFv variant (e.g., a set of amino acid substitutions as 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 ... / E357L, K370S:S364K / E357Q, T366S / L368A / Y407V:T366W, and T366S / L368A / Y407V / Y349C:T366W / S354C), optionally a deletion variant (including those shown in Figure 3), optionally a charged scFv linker (including those shown in Figure 5), and the heavy chain comprises a pI variant (including those shown in Figure 2).

[0187] In some embodiments, the one-arm central scFv format includes scFv 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 CLDN18.2 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 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.

[0188] 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 CLDN18.2 as outlined herein, together with a variable light chain domain of the light chain, and a sFv that binds CD3. a) a first monomer comprising a cFv domain; b) a second monomer comprising an Fc domain having the scivariant 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.

[0189] For the one-arm central scFv format, 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, and anti-CD3 H1.31_L1.47, as shown in FIG. 12.

[0190] One-arm scFv-mAb One heterodimeric scaffold of particular use in the present invention is the one-arm scFv-mAb format shown in Figure 42D. 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-[optional domain linker]-CH1-hinge-CH2-CH3 or (in the opposite orientation) VL-scFv linker-VH-[optional domain linker]-CH1-hinge-CH2-CH3. In this format, the Fab portions each bind to CLDN18.2 and the scFv binds to CD3. This embodiment further utilizes a light chain comprising a variable light domain and a constant light domain that associates 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.

[0191] The present invention provides a one-arm scFv-mAb format in which the CD3 binding domain sequence is as shown in FIG. 12 and the CLDN18.2 binding domain sequence is as shown in FIG.

[0192] In addition, the Fc domain of the one-arm scFv-mAb format generally comprises a scFv variant (e.g., a set of amino acid substitutions as shown in Figures 1 and 4, particularly useful scFv variants are S364K / E357Q:L368D / K370S, L368D / K370S:S364K, L368E / K370S:S364K, T411T / E360E / Q362E:D401K, L368D ... 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).

[0193] 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 CLDN18.2 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 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.

[0194] 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 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 CLDN18.2 as outlined herein, together with a variable light chain domain of the light chain, and a sFv that binds CD3. a) a first monomer comprising a cFv domain; b) a second monomer comprising an Fc domain having the scivariant 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.

[0195] For the one-arm scFv-mAb format, 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, and anti-CD3 H1.31_L1.47, as shown in FIG. 12.

[0196] scFv-mAb A heterodimeric scaffold of particular use in the present invention is the mAb-scFv format shown in Figure 42E. 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 CLDN18.2 and the "additional" scFv domain binding to CD3.

[0197] 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 attached 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 the 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 that associates with the heavy chain to form two identical Fabs that bind to CLDN18.2. For many of the embodiments described herein, these constructs include scv variants, pI variants, deletion variants, additional Fc variants, etc., as desired and described herein.

[0198] The present invention provides an scFv-mAb format in which the CD3 binding domain sequence is as shown in FIG. 12 and the CLDN18.2 binding domain sequence is as shown in FIG.

[0199] In addition, the Fc domain of the scFv-mAb 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 / 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 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).

[0200] In some embodiments, the scFv-mAb format includes scFv variants, pI variants, and deletion variants. Thus, some embodiments include an scFv-mAb format, which includes 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 CLDN18.2 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, together with a variable heavy chain domain constituting an Fv that binds CLDN18.2 as outlined herein, and an scFv domain that binds CD3; a) a common light chain comprising a variable light chain domain and a constant light chain domain, and b) a variable heavy chain domain comprising the variable light chain domain and the constant light chain domain, together with the variable light chain domain and the constant light chain domain.

[0201] 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 includes a) a first monomer comprising a variable heavy chain domain constituting an Fv that binds CLDN18.2 as outlined herein, together with the scFv variant S364K / E357Q, the deletion variants E233P / L234V / L235A / G236del / S267K, the FcRn variants M428L / N434S, and 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 CLDN18.2 as outlined herein together with the variable light chain of the sq variant 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.

[0202] For mAb-scFv format scaffold 1 from FIG. 10 (optionally containing M428L / N434S), 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, and anti-CD3 H1.31_L1.47, as shown in FIG. 12.

[0203] Dual scFv format The invention also provides a dual scFv format, as known in the art and shown in Figure 42B. In this embodiment, the CLDN18.2xCD3 heterodimeric bispecific antibody is composed of two scFv-Fc monomers, both in either the (VH-scFv linker-VL-[optional domain linker]-CH2-CH3) format or 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.

[0204] The present invention provides a dual scFv format, in which the CD3 binding domain sequence is as shown in FIG. 12 and the CLDN18.2 binding domain sequence is as shown in FIG.

[0205] In some embodiments, the dual scFv format comprises a scFv variant, a pI variant, and a deletion variant. Thus, some embodiments comprise a dual scFv format comprising a) a first monomer comprising the scFv variant S364K / E357Q, the truncation variant E233P / L234V / L235A / G236del / S267K, and a first scFv that binds either CD3 or CLDN18.2, and b) a second monomer comprising the scFv variant L368D / K370S, the pI variant N208D / Q295E / N384D / Q418E / N421D, the truncation variant E233P / L234V / L235A / G236del / S267K, and a second scFv that binds either CD3 or CLDN18.2.

[0206] In some embodiments, the dual scFv format comprises a scuba variant, a pI variant, a deletion variant, and an FcRn variant, In some embodiments, the dual scFv format comprises a scuba 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 CLDN18.2, and b) a second monomer comprising the scFv variants 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 CLDN18.2.

[0207] For the dual scFv format, 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, and anti-CD3 H1.31_L1.47, and those shown in FIG. 12.

[0208] I. Antigen-binding domain to target antigen The bispecific antibodies of the invention have two different antigen binding domains (ABDs) that bind two different target checkpoint antigens ("target pairs") in either a bivalent or trivalent bispecific format, as generally shown in FIG. 1. It is noted that these bispecific antibodies are generally named "anti-CLDN18.2 x anti-CD3", or generally for simplification or simplicity (and thus interchangeably), "CLDN18.2 x CD3" for each pair, etc. It is noted that unless specified herein, the order of the antigen list in the name does not confer structure. That is, a CLDN18.2 x CD3 bottle opener antibody can bind scFv to CLDN18.2 or CD3, but in some cases, the structure is designated as the order is indicated.

[0209] As outlined more fully herein, these combinations of ABDs can be in a variety of formats, generally with one ABD being in a Fab format and the other in an scFv format, as outlined below: As discussed herein and shown in Figure 42, some formats use a single Fab and a single scFv (Figures 42A, C, and D), and some formats use two Fabs and a single scFv (Figures 42E, F, and I).

[0210] Antigen-binding domain As discussed herein, the subject heterodimeric antibodies comprise two antigen binding domains (ABDs), each of which binds to CLDN18.2 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 Figure 42A), or bispecific and trivalent (one antigen is bound by a single ABD and the other by two ABDs, e.g., as shown in Figure 42F).

[0211] In addition, 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.

[0212] The present invention provides several ABDs that bind to several different checkpoint proteins, 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 are then added to heavy and light chain constant domains, with the heavy chain constant domains including variants (including within the CH1 and Fc domains). The scFv sequences included in the sequence listing utilize a specific charged linker, but uncharged or other charged linkers can be used, including those shown in Figure 5, as outlined herein.

[0213] In addition, as discussed 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 1.

[0214] 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 (including amino acid substitutions specifically used), 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 can have amino acid modifications (e.g., 1, 2, 3, 4 or 5 amino acid modifications in a set of CDRs (i.e., the CDRs can be modified as 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 can be altered, e.g., 1 change in VH CDR1, 2 changes in VH CDR2, no changes in VH CDR3, etc.)), and there can be 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.

[0215] CLDN18.2 antigen-binding domain In some embodiments, one of the ABDs binds to CLDN18.2. A suitable set of six CDRs and / or VH and VL domains is shown in FIG.

[0216] As will be appreciated by those of skill in the art, a suitable CLDN18.2 binding domain can include a set of six CDRs as shown in the figures, either as underlined, or as CDRs identified using other alignments within the VH and VL sequences of those shown in Figure 10 if a different numbering scheme is used as described herein and shown in Table 1. A suitable ABD can also include these sequences and the entire VH and VL sequences shown in the figures, used as an scFv or Fab. In many of the embodiments herein that include an Fv against CLDN18.2, it is the Fab monomer that binds to CLDN18.2.

[0217] In addition to the parental CDR sets disclosed in the Figures and Sequence Listing, which form the ABD for CLDN18.2, the invention provides variant CDR sets, in one embodiment, the set of six CDRs can have 1, 2, 3, 4, or 5 amino acid changes from the parental CDRs, so long as the CLDN18.2 ABD is still able to bind to the target antigen as measured by at least one of Biacore, surface plasmon resonance (SPR) and / or BLI (biolayer interferometry, e.g., Octet assay) assays, the latter of which is particularly used in many embodiments.

[0218] In addition to the parent variable heavy and variable light domains disclosed herein that form the ABD to CLDN18.2, the present invention provides variant VH and VL domains, in one embodiment, the variant VH and VL domains may each have 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid changes from the parent VH and VL domains, so long as the ABD is still capable of binding to the target antigen as measured by at least one of Biacore, Surface Plasmon Resonance (SPR) and / or BLI (Biolayer Interferometry, e.g., Octet assay) assays, the latter of which is particularly used in many embodiments. In another embodiment, the variant VH and VL are at least 90, 95, 97, 98, or 99% identical to the respective parent VH and VL domains, so long as the ABD is still capable of binding to the target antigen, as measured by at least one of Biacore, surface plasmon resonance (SPR) and / or BLI (biolayer interferometry, e.g., Octet assay) assays, the latter of which is specifically used in many embodiments.

[0219] A specific preferred embodiment comprises the H1L1 and H2L1 CLDN18.2 antigen binding domains as "Fabs" contained within any of the bottle opener scaffolds of FIG.

[0220] A specific preferred embodiment comprises the H1L1 and H2L1 CLDN18.2 antigen binding domains as "Fab" contained within any of the 2+1 format scaffolds in FIG.

[0221] CD3 antigen-binding domain In some embodiments, one of the ABDs binds to CD3. Suitable sets of six CDRs and / or VH and VL domains and scFv sequences are shown in Figures 12 and 13 and in the Sequence Listing. Particularly useful CD3 binding domain sequences 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, and anti-CD3 H1.31_L1.47, as shown in Figure 12.

[0222] As will be appreciated by one of skill in the art, a suitable CD3 binding domain can comprise the set of six CDRs as shown in Figure 12, either as underlined, or as CDRs identified using other alignments within the VH and VL sequences of those shown in Figure 12 if a different numbering scheme is used as described herein and shown in Table 1. A suitable ABD can also comprise these sequences and the entire VH and VL sequences 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 binds to CD3.

[0223] In addition to the parental CDR sets disclosed in the Figures and Sequence Listing, which form the ABD to CD3, the invention provides variant CDR sets, in one embodiment, the set of six CDRs can have 1, 2, 3, 4, or 5 amino acid changes from the parental CDRs, so long as the CD3 ABD is still able to bind to the target antigen as measured by at least one of Biacore, Surface Plasmon Resonance (SPR) and / or BLI (Biolayer Interferometry, e.g., Octet assay) assays, the latter of which is particularly used in many embodiments.

[0224] In addition to the parent variable heavy and variable light domains disclosed herein that form the ABD to CD3, the present invention provides variant VH and VL domains. In one embodiment, the variant VH and VL domains may each have 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 amino acid changes from the parent VH and VL domains, so long as the ABD is still able to bind to the target antigen as measured by at least one of Biacore, surface plasmon resonance (SPR) and / or BLI (biolayer interferometry, e.g., Octet assay) assays, the latter of which are particularly used in many embodiments. In another embodiment, the variant VH and VL are at least 90, 95, 97, 98, or 99% identical to the respective parent VH and VL domains, so long as the ABD is still able to bind to the target antigen as measured by at least one of Biacore, surface plasmon resonance (SPR) and / or BLI (biolayer interferometry, e.g., Octet assay) assays, the latter of which are particularly used in many embodiments.

[0225] J. Useful Embodiments In one embodiment, a particular combination of scFv variants and pI variants for use in the present invention is T366S / L368A / Y407V:T366W (optionally including a bridging disulfide, T366S / L368A / Y407V / Y349C:T366W / S354C), with one monomer comprising Q295E / N384D / Q418E / N481D and the other a positively charged domain linker (format includes scFv domains). As understood in the art, the "knobs-in-holes" variant does not change the pI and therefore can be used with either monomer.

[0226] K. Nucleic Acids of the Invention The present invention further provides nucleic acid compositions encoding the anti-CLDN18.2 antibodies provided herein, including, but not limited to, anti-CLDN18.2 x anti-CD3 bispecific antibodies and CLDN18.2 monospecific antibodies.

[0227] As will be appreciated by those skilled in the art, the nucleic acid composition depends 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-Fc format (e.g., a first amino acid monomer comprising an Fc domain and an 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., the dual scFv format disclosed in Figure 1), only two nucleic acids are required, again, which can be incorporated into one or two expression vectors.

[0228] As is known in the art, the nucleic acid encoding the components of the present invention can be incorporated into an expression vector, as is known in the art, depending on the host cell used to produce the heterodimeric antibody of the present invention. In general, the nucleic acid is operably linked to any number of regulatory elements (promoter, origin of replication, selectable marker, ribosome binding site, inducer, etc.). The expression vector can be an extrachromosomal vector or an integrating vector.

[0229] The nucleic acids and / or expression vectors of the invention are then transformed into any number of different types of host cells known in the art, including mammalian, bacterial, yeast, insect and / or fungal cells, with mammalian cells (e.g., CHO cells) being used in many embodiments.

[0230] In some embodiments, the nucleic acids encoding each monomer and, depending on the format, the optional nucleic acid encoding the light chain are each contained in a single expression vector, generally under different or the same promoter control. In particularly useful embodiments of the present invention, each of these two or three nucleic acids is contained on a different expression vector. As shown in this specification and in 62 / 025,931, which is incorporated herein by reference, different vector ratios can be used to promote heterodimer formation. That is, surprisingly, although the protein contains the first monomer: the second monomer: the light chain in a ratio of 1:1:2 (in many of the embodiments of the present specification having three polypeptides comprising a heterodimeric antibody), these are not the ratios that give the best results.

[0231] The heterodimeric antibodies of the present invention 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 pI of each monomer so that each monomer has a different isoelectric point (pI) and the heterodimer also has a distinct pI, facilitating isoelectric purification (e.g., anion exchange column, cation exchange column) of "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).

[0232] L. Biological and Biochemical Functions of Heterodimeric Bispecific Antibodies In general, the bispecific CLDN18.2xCD3 antibody of the invention is administered to a patient 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.

[0233] M.Treatment Once made, the antibody composition of the present invention can be used in several applications. CLDN18.2 is highly expressed in gastric tumors. Therefore, the heterodimer composition of the present invention can be used to treat such CLDN18.2 positive cancers.

[0234] Antibody Compositions for In Vivo Administration Formulations of antibodies used according to the invention are prepared for storage by mixing the antibody having the desired purity with optional pharma- ceutically acceptable carriers, excipients, or stabilizers, in the form of a lyophilized formulation or an aqueous solution (Remington's Pharmaceutical Sciences 16th edition, Osol, A. Ed.

[1980] ). Acceptable carriers, excipients, or stabilizers are non-toxic to recipients at the dosages and concentrations employed, and include buffers such as phosphate, citrate, and other organic acids, antioxidants including ascorbic acid and methionine, preservatives (e.g., octadecyldimethylbenzylammonium chloride, hexamethonium chloride, benzalkonium chloride, benzethonium chloride, phenol, butyl or benzyl alcohol, alkyl parabens such as methyl or propyl paraben, catechol, resorcinol, cyclohexanol, 3-pentanol, and m-cresol), low molecular weight (less than about 10 residues) polypeptides, serum albumin, and the like. The surfactants may include, for example, proteins such as glycine, gelatin, or immunoglobulins, hydrophilic polymers such as polyvinylpyrrolidone, amino acids such as glycine, glutamine, asparagine, histidine, arginine, or lysine, monosaccharides, disaccharides, and other carbohydrates including glucose, mannose, or dextrins, chelating agents such as EDTA, sugars such as sucrose, mannitol, trehalose, sorbitol, salt forming counterions such as sodium, metal complexes (e.g., Zn-protein complexes), and / or non-ionic surfactants such as, for example, TWEEN™, PLURONICS™, or polyethylene glycol (PEG).

[0235] Administration Modality The antibodies and chemotherapeutic agents of the present invention are administered to a subject according to known methods, such as intravenously as a bolus or by continuous infusion over a period of time.

[0236] Treatment Modalities In the method of the present invention, treatment is used to provide a positive therapeutic response for a disease or condition. By "positive therapeutic response" is intended an improvement in a disease or condition and / or an improvement in symptoms associated with the disease or condition. For example, a positive therapeutic response may refer to one or more of the following improvements in a disease: (1) a decrease in the number of tumor cells; (2) an increase in tumor cell death; (3) an inhibition of tumor cell survival; (5) an inhibition (i.e., some slowing down, preferably halting) of tumor growth; (6) an increase in patient survival; and (7) some relief from one or more symptoms associated with a disease or condition.

[0237] 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 assessed for changes in tumor morphology (i.e., tumor burden, tumor size, etc.) using screening techniques such as magnetic resonance imaging (MRI) scans, radiography, computed tomographic (CT) scans, bone scans, endoscopy, and tumor biopsy sampling, including bone marrow aspiration (BMA) and enumeration of circulating tumor cells.

[0238] In addition to these positive therapeutic responses, subjects undergoing treatment may experience the beneficial effect of amelioration of symptoms associated with the disease.

[0239] Treatment according to the present invention involves the use of a "therapeutically effective amount" of a pharmaceutical agent. A "therapeutically effective amount" refers to an amount effective, at dosages and for periods of time necessary, to achieve the desired therapeutic result.

[0240] 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.

[0241] 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 can be evaluated in an animal model system that is predictive of efficacy in human tumors.

[0242] Alternatively, this property of the composition can be evaluated by testing the ability of the compound to inhibit cell proliferation or induce apoptosis by in vitro assays known to those skilled in the art. A therapeutically effective amount of a therapeutic compound can reduce tumor size or otherwise alleviate symptoms in a subject. One skilled 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.

[0243] 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 ph...

Claims

1. A Claudin 18 isoform A2 splice variant (CLDN18.2) antigen-binding domain comprising a variable heavy chain domain having at least 95% identity to the amino acid sequence of SEQ ID NO: 81 and a variable light chain domain having at least 95% identity to the amino acid sequence of SEQ ID NO:

84.

2. The CLDN18.2 antigen-binding domain according to claim 1, wherein the variable heavy chain domain has the amino acid sequence of SEQ ID NO: 81 and the variable light chain domain has the amino acid sequence of SEQ ID NO:

84.

3. A nucleic acid composition comprising: a) a first nucleic acid encoding the variable heavy chain domain according to claim 1; b) a second nucleic acid encoding the variable light chain domain according to claim 1.

4. An expression vector composition comprising: a) a first expression vector comprising the first nucleic acid according to claim 3; b) a second expression vector comprising the second nucleic acid according to claim 3.

5. A host cell comprising the nucleic acid composition according to claim 3 or the expression vector composition according to claim 4.

6. A method for producing a CLDN18.2 antigen-binding domain, comprising culturing the host cell according to claim 5 under conditions under which the CLDN18.2 antigen-binding domain is expressed, and recovering the CLDN18.2 antigen-binding domain.

7. An antibody comprising a Claudin 18 isoform A2 splice variant (CLDN18.2) antigen-binding domain comprising a variable heavy chain domain having at least 95% identity to the amino acid sequence of SEQ ID NO: 81 and a variable light chain domain having at least 95% identity to the amino acid sequence of SEQ ID NO:

84.

8. An antibody comprising the CLDN18.2 antigen-binding domain according to claim 7, wherein the variable heavy chain domain has the amino acid sequence of SEQ ID NO: 81 and the variable light chain domain has the amino acid sequence of SEQ ID NO:

84.

9. A nucleic acid composition comprising: a) a first nucleic acid encoding the variable heavy chain domain according to claim 7; b) a second nucleic acid encoding the variable light chain domain according to claim 7.

10. An expression vector composition comprising: a) a first expression vector comprising the first nucleic acid according to claim 9; b) a second expression vector comprising the second nucleic acid according to claim 9.

11. A host cell comprising the nucleic acid composition according to claim 9 or the expression vector composition according to claim 10.

12. A method for producing an antibody comprising the CLDN18.2 antigen-binding domain, the method comprising culturing the host cell according to claim 11 under conditions in which the antibody comprising the CLDN18.2 antigen-binding domain is expressed, and recovering the antibody comprising the CLDN18.2 antigen-binding domain.

Images