EA039658B1 - Cd3/cd38 t cell retargeting hetero-dimeric immunoglobulins and methods of their production - Google Patents

Abstract

The invention relates to hetero-dimeric immunoglobulins that target both a component of the human CD3 antigen and a component of the human CD38 antigen and methods of making the same. The invention also relates to antibodies which bind to the human CD38 antigen and derivatives thereof for use as therapeutic or diagnostic reagents and methods of making the same.

Description

The present invention relates to heterodimeric immunoglobulins that target both the human CD3 antigen component and the human CD38 antigen component, and methods for their preparation. The present invention also relates to antibodies that bind to the human CD38 antigen and derivatives thereof for use as therapeutic or diagnostic reagents, and methods for their preparation.

State of the art

The use of monoclonal antibodies (MAT) in vivo is becoming one of the main directions of routine clinical practice in the treatment of various human diseases, and targeted immunotherapy using MAT is becoming critical for the successful treatment of many forms of cancer. An example of this is rituximab, a chimeric anti-CD20 antibody that has revolutionized the treatment of several B-cell malignancies such as follicular lymphoma. However, B cells lose CD20 expression after terminal differentiation into plasma cells, and accordingly, rituximab is of very limited benefit in the treatment of multiple myeloma, among other cancers.

The CD38 molecule is an attractive target due to its expression pattern and dual role as receptor and ectoenzyme. In particular, CD38 targeting has been proposed as a treatment for multiple myeloma and chronic lymphocytic leukemia (CLL).

Redirected killing by T cells is a desirable mechanism of action in many therapeutic areas. Various bispecific antibody formats have been shown to mediate T cell retargeting in both preclinical and clinical studies (May C et al., (2012) Biochem Pharmacol, 84(9): 1105-12; Frankel SR & Baeuerle PA, (2013) Curr Opin Chem Biol, 17(3): 385-92). All T cell retargeting bispecific antibodies or fragments thereof are designed to contain at least two antigen binding sites, with one site binding antigen on the surface of the target cell and the other site binding antigen on the surface of the T cell. Among antigens on the surface of T cells, the human CD3 epsilon subunit from the TCR protein complex has been most commonly used to redirect killing by T cells.

Many bispecific antibody formats have been used to retarget T-cell killing, which include mainly tandem scFv fragments and diabody formats, with only a few examples of Fc-based bispecific antibody formats described (Moore PA et al., (2011) Blood , 117(17): 4542-51; May C et al., (2012), supra; Frankel SR & Baeuerle PA, (2013), supra). Bispecific formats that include a human Fc region will have a longer circulating half-life, possibly resulting in improved efficacy and/or lower dosing frequency. Among the possible Fc-based bispecific formats, one preferred format for retargeting T cell killing is the so-called heavy chain heterodimer format. This format is of particular interest because it prevents the aggregation of multiple copies of human CD3 molecules on the surface of T cells, thereby completely preventing T cell inactivation (Klein C et al., (2012) MAbs, 4(6): 653- 63).

The first described method for constructing heavy chain heterodimers is the method known as the ridge-to-trough method (PCT Publication No. WO199627011; Merchant AM et al., (1998) Nat Biotechnol, 16(7): 677-81). Recently, a chemical method, known as the Fab exchange method, has been described in which two antibodies are combined into one bispecific antibody by in vitro reconstitution and shuffling of immunoglobulin halves (PCT Publication Nos. .); Labrijn AF et al. (2013) Proc Natl Acad Sci USA, 110(13): 5145-50).

Brief description of the invention

The present invention provides an antibody or fragment thereof that binds to human CD38, comprising a heavy chain hypervariable region 1 (CDR1) comprising the amino acid sequence of SEQ ID NO: 25 or SEQ ID NO: 208, and/or a heavy chain CDR2 comprising the amino acid sequence SEQ ID NO: 26 or SEQ ID NO: 209 and/or a heavy chain CDR3 comprising the amino acid sequence of SEQ ID NO: 27 or SEQ ID NO: 210; and/or containing a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 28 or SEQ ID NO: 211 or SEQ ID NO: 214 and/or a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 29 or SEQ ID NO : 212, and/or a light chain CDR3 comprising the amino acid sequence of SEQ ID NO: 30 or SEQ ID NO: 213; or wherein the epitope-binding region that binds the CD38 protein complex comprises a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 31, a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 32, and a heavy chain CDR3 comprising the amino acid sequence of SEQ ID NO: 32 NO: 33, and a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 34, a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 35, and a light chain CDR3 comprising the sequence ami

- 1 039658 no acids SEQ ID NO: 36.

In accordance with this aspect of the present invention, said anti-CD38 antibody or fragment thereof is a mouse antibody, a chimeric antibody, a humanized antibody, or a fully human antibody.

In particular, the antibody or fragment thereof contains a heavy chain variable region sequence comprising an amino acid sequence consisting of SEQ ID NO: 52, SEQ ID NO: 58, SEQ ID NO: 60, or a sequence that is at least 80% identical to non related to the CDR region of any of the specified heavy chain variable region sequences, and/or the specified antibody or fragment contains a light chain variable region sequence comprising the amino acid sequence of SEQ ID NO: 55, SEQ ID NO: 59, SEQ ID NO: 61, or a sequence that is at least 80% identical to the non-CDR portion of any of said light chain variable region sequences.

In particular, the antibody or fragment thereof contains a heavy chain sequence comprising an amino acid sequence consisting of SEQ ID NO: 73, SEQ ID NO: 75, SEQ ID NO: 76, SEQ ID NO: 145, SEQ ID NO: 147, SEQ ID NO: 149, or a sequence that is at least 80% identical to the non-CDR region of any of said heavy chain variable region sequences, and/or said antibody or fragment thereof contains a light chain sequence comprising the amino acid sequence of SEQ ID NO: 74 , SEQ ID NO: 77, SEQ ID NO: 146, SEQ ID NO: 148, SEQ ID NO: 150, or a sequence that is at least 80% identical to the non-CDR region of any of the indicated light chain variable region sequences.

In particular, said antibody or fragment thereof comprises human heavy and/or light chain constant regions, wherein the human heavy chain constant region is selected from the group of human immunoglobulins consisting of IGHG1, non-fucosylated IGHG1 and IGHG4.

The present invention also provides novel anti-human CD3/CD38 bispecific antibodies.

In accordance with the present invention, the target-binding arm portion of a human CD3-binding antibody preferably comprises a single chain variable fragment (scFv) of SP34 or OKT3. In particular, if the arm of the antibody that binds human CD3 is SP34, it will have a better expression profile than scFv-Fc containing the heavy and light chain variable regions encoded by the sequences of SEQ ID NOs: 1 and 2, while maintaining its CD3 binding properties.

In particular, when expressed as a scFv-Fc, at least a two-fold increase in expression is observed compared to an SP34 chimera in scFv-Fc format containing the heavy and light chain variable regions encoded by the sequences of SEQ ID NOs: 1 and 2. Preferably, an improved SP34 scFv show at least a six-fold increase in expression compared to the scFv format SP34 chimera containing the heavy and light chain variable regions encoded by the sequences of SEQ ID NOs: 1 and 2, and most preferably a twelve-fold increase in expression compared to the scFv format SP34 chimera containing the variable regions of the heavy and light chains encoded by the sequences of SEQ ID NO: 1 and 2.

In accordance with another aspect of the present invention, when expressed as a scFv in a bispecific antibody containing a FAB fragment in BEAT format, at least a two-fold increase in expression is observed compared to an SP34 chimera in scFv format containing variable regions of the heavy and light chains , encoded by the sequences of SEQ ID NOs: 1 and 2. Preferably, the improved SP34 scFv as a component of the bispecific BEAT antibody is observed to have at least a five-fold increase in expression compared to the SP34 chimera in scFv format containing the heavy and light chain variable regions encoded by the sequences of SEQ ID NO: 1 and 2, as a component of the BEAT bispecific antibody.

In accordance with the present invention, each of the two binding arms of an anti-human CD3 bispecific antibody contains an immunoglobulin constant region, wherein said first arm or polypeptide binds to protein A and said second arm or polypeptide does not bind to protein A.

According to a further aspect of the present invention, the epitope-binding region of the polypeptide that binds the CD3 protein complex comprises a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 3, a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 4, and a heavy chain CDR3 comprising the sequence amino acids of SEQ ID NO: 5 or SEQ ID NO: 9, and a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 6, a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 7 or SEQ ID NO: 11, and a light chain CDR3 , including the listed amino acid sequences: SEQ ID NO: 8, SEQ ID NO: 10 or SEQ ID NO: 12; or wherein the epitope-binding region that binds the CD3 protein complex comprises a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 163 and a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 164 and a CDR3

- 2 039658 heavy chain, comprising the amino acid sequence of SEQ ID NO: 165; and contains a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 166 and a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 167 and a light chain CDR3 comprising the amino acid sequence of SEQ ID NO: 168.

According to a further aspect of the present invention, the epitope-binding region of the polypeptide that binds the CD38 protein comprises a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 13, a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 14, and a heavy chain CDR3 comprising the amino acid sequence SEQ ID NO: 15, and a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 16, a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 17, and a light chain CDR3 comprising the amino acid sequence of SEQ ID NO: 18; or wherein the epitope-binding region that binds the CD38 protein complex comprises a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 19, a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 20, and a heavy chain CDR3 comprising the amino acid sequence of SEQ ID NO: : 21, and a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 22, a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 23, and a light chain CDR3 comprising the amino acid sequence of SEQ ID NO: 24; or an epitope-binding region of a polypeptide that binds the CD38 protein comprises a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 25, a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 26, and a heavy chain CDR3 comprising the amino acid sequence of SEQ ID NO: 27 and a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 28, a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 29, and a light chain CDR3 comprising the amino acid sequence of SEQ ID NO: 30; or wherein the epitope-binding region that binds the CD38 protein complex comprises a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 31, a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 32, and a heavy chain CDR3 comprising the amino acid sequence of SEQ ID NO: : 33, and a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 34, a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 35, and a light chain CDR3 comprising the amino acid sequence of SEQ ID NO: 36.

The use of these new anti-human CD3/CD38 bispecific antibodies is not limited, but includes the treatment of various types of cancer, autoimmune and inflammatory diseases in humans. The specific destruction of cancer cells to a greater extent than healthy cells and tissues is a top priority in oncology. A drug that can safely redirect T cells to kill cells containing tumor-associated surface antigens could provide increased clinical efficacy. Known areas with clinically unmet treatment needs in oncology include, but are not limited to, breast cancer, metastatic breast cancer, ovarian cancer, pancreatic cancer, lung cancer, lymphomas, and multiple myeloma.

According to a further aspect of the present invention, the constant region of said second heterodimeric immunoglobulin polypeptide or fragment thereof comprises the IgG3 CH3 region.

According to a further aspect of the present invention, the constant region of said second heterodimeric immunoglobulin polypeptide or fragment thereof comprises a CH3 region other than that of IgG, and said non-IgG3 CH3 region contains at least one substitution to reduce/impair protein A binding.

According to a further aspect of the present invention, said epitope-binding region of the second heterodimeric immunoglobulin polypeptide or fragment thereof comprises a VH3 region containing at least one modification that reduces protein A binding.

The present invention also provides a method for producing heavy chain bispecific heterodimers directed against human CD3/CD38 containing at least one VH3-based antigen binding site from a recombinant mammalian host cell line, wherein the resulting bispecific antibody is readily isolated after a single chromatography step with protein A with a high degree of purity.

In particular, the modified VH3 region contains an amino acid substitution selected from the group consisting of 57, 65, 81, 82a and the combination 19/57/59 (Kabat numbering), and even more preferably, said modified VH3 region contains an amino acid substitution selected from group consisting of 57A, 57E, 65S, 81E, 82aS and combinations 19G/57A/59A (Kabat numbering).

According to a further aspect of the present invention, the heterodimeric immunoglobulin or fragment thereof may contain additional substitutions, wherein the heavy chain variable region framework region contains an amino acid substitution selected from the group consisting of: I34M, V48I, A49G, R58N/Y, I69L, A71T, and T73K ( Kabat numbering) and the light chain variable region framework contains an amino acid substitution selected from the group consisting of: M4L, V33M, A34N, L46R, L47W, T51A, R66G, F71Y, and P96F (Kabat numbering); or, in this case, the frame section of the varia

The heavy chain protein region contains amino acid substitutions 134M, A49G and A71T (Kabat numbering) and the framework region of the light chain variable region contains amino acid substitutions M4L, L46R, L47W and F71Y (Kabat numbering).

In a further embodiment, the epitope-binding region that binds to the CD3 protein complex comprises a heavy chain variable region framework that is derived from or derived from the human VH3 subclass. Preferably, said heavy chain variable region framework is derived from or derived from human IGHV3-23. More preferably, the heavy chain variable region framework is derived from or derived from human IGHV3-23*04 (SEQ ID NO: 37). The heavy chain variable region framework region comprises a modification of at least one amino acid from the corresponding heavy chain variable region framework region of the corresponding mouse antibody comprising the amino acid sequence of SEQ ID NO: 38 or SEQ ID NO: 1.

In a preferred embodiment, the epitope-binding region of the first polypeptide that binds to the CD3 protein complex comprises a light chain variable region framework that is the product of or derived from a human VK1 subclass or a human VK3 subclass. Preferably, said light chain variable region framework is derived from or derived from human VK1-39 or VK3-20. More preferably, said light chain variable region framework is derived from or derived from human IGKV1-39*01 (SEQ ID NO: 39) or IGKV3-20*01 (SEQ ID NO: 40). The light chain variable region framework region comprises a modification of at least one amino acid from the corresponding light chain variable region framework region of the corresponding mouse antibody comprising the amino acid sequence of SEQ ID NO: 41 or SEQ ID NO: 2.

In a preferred embodiment, the epitope-binding region that binds to the CD3 protein complex comprises a humanized heavy chain variable domain containing backmutations selected from the group consisting of: I34M, V48I, A49G, R58N/Y, I69L, A71T, and T73K (numbering by Kabat), and a humanized light chain variable domain containing backmutations selected from the group consisting of: M4L, V33M, A34N, L46R, L47W, R66G, F71Y and P96F (Kabat numbering). More preferably, the epitope-binding region that binds to the CD3 protein complex comprises a humanized heavy chain variable domain containing backmutations I34M, A49G and A71T (Kabat numbering) and a humanized light chain variable domain containing backmutations M4L, L46R, L47W and F71Y (Kabat numbering).

According to a further aspect of the present invention, there is provided an epitope-binding region that binds a CD3 protein complex, a heterodimeric immunoglobulin, or a fragment thereof, wherein the epitope-binding region that binds the CD3 protein complex comprises a heavy chain variable region comprising the amino acid sequence of SEQ ID NO: 42, and a light chain variable region comprising the amino acid sequence of SEQ ID NO: 46; or wherein the epitope-binding region that binds the CD3 protein complex comprises a heavy chain variable region comprising the amino acid sequence of SEQ ID NO: 44 and a light chain variable region comprising the amino acid sequence of SEQ ID NO: 47; or wherein the epitope-binding region that binds the CD3 protein complex comprises a heavy chain variable region comprising the amino acid sequence of SEQ ID NO: 45 and a light chain variable region comprising the amino acid sequence of SEQ ID NO: 47; or wherein the epitope-binding region that binds the CD3 protein complex comprises a heavy chain variable region comprising the amino acid sequence of SEQ ID NO: 45 and a light chain variable region comprising the amino acid sequence of SEQ ID NO: 48; or wherein the epitope-binding region that binds the CD3 protein complex comprises a heavy chain variable region comprising the amino acid sequence of SEQ ID NO: 45 and a light chain variable region comprising the amino acid sequence of SEQ ID NO: 49.

The CD3 protein complex contains many subunits, such as delta, epsilon, and gamma. In a preferred embodiment, the epitope-binding region that binds to the CD3 protein complex binds to the CD3 epsilon subunit.

The epitope-binding region described herein includes a combination of one or more heavy chain variable domains and one or more complementary light chain variable domains, which together form a binding site that allows specific binding of a heterodimeric immunoglobulin, or fragment thereof, to one or more epitopes. In some embodiment, the epitope-binding region of said first polypeptide comprises FAB and the epitope-binding region of said second polypeptide comprises scFv. Alternatively, the epitope-binding region of the first polypeptide comprises scFv and the epitope-binding region of the second polypeptide comprises FAB.

The epitope-binding region that binds to CD38 includes a heavy chain variable region framework that is derived from or derived from the human VH3 subclass,

- 4 039658 preferably human VH3-23, more preferably human IGHV3-23*04 (SEQ ID NO: 37). The heavy chain variable region framework region comprises a modification of at least one amino acid from the corresponding heavy chain variable region framework region of the corresponding mouse antibody comprising the amino acid sequence of SEQ ID NO: 50, or 51, or 52. The epitope-binding region further comprises a light variable region framework region a chain that is the product of or derived from a subclass of human VK1, preferably human VK1-39, more preferably human IGKV1-39*01 (SEQ ID NO: 39). The light chain variable region framework region comprises a modification of at least one amino acid from the corresponding light chain variable region framework region of the corresponding mouse antibody comprising the amino acid sequence of SEQ ID NO: 53 or 54 or 55.

In particular, the CD38 binding polypeptide comprises a heavy chain variable domain and light chain variable domain pair encoded by the sequences of SEQ ID NOs: 56/57, 58/59, 60/61 and 62/63.

Anti-CD3 antibodies have been found to trigger toxic effects through direct and indirect mechanisms. Indirect mechanisms are mediated by the Fc region of an anti-CD3 antibody, which interacts with Fc receptor-expressing immune cells and results in transient activation of T cells and release of cytokines. Therefore, in order to improve the safety of said heterodimeric immunoglobulins or fragments thereof described herein, the immunoglobulin constant region of the first and/or second polypeptide binds or does not bind to effector immune cells and/or complement C1q to a lesser extent. Preferably, the immunoglobulin constant region is designed to disrupt binding to the Fc receptor in the lower hinge region. More preferably, the immunoglobulin constant region of the first and/or second polypeptide contains the substitution(s) L234A and/or L235A (EU numbering). Most preferably, the immunoglobulin constant region of the first and/or second polypeptide contains the substitutions L234A and L235A (EU numbering).

In another aspect of the disclosure, the present invention also provides a heterodimeric immunoglobulin or fragment thereof, wherein the epitope-binding region binds to the epsilon CD3 subunit of the CD3 protein complex and contains a FAB whose thermal stability is higher than that of the FAB chimera SP34 containing the heavy chain variable domain with the sequence amino acids shown in SEQ ID NO: 1, and a light chain variable domain with the amino acid sequence shown in SEQ ID NO: 2, as measured by differential scanning calorimetry (DSC), as described in table. 1. This increased thermal stability would mean that these improved SP34 binding antibody arms would have increased in vivo and in vitro stability, meaning better performance as a therapeutic agent and also in terms of its stability/storage/shelf life.

In a preferred embodiment, the present invention provides a heterodimeric immunoglobulin or fragment thereof that binds to:

i) a protein complex of CD3 and CD38, wherein the amino acid sequence of said first polypeptide is shown in SEQ ID NO: 65, and it joins a cognate light chain with the amino acid sequence shown in SEQ ID NO: 66 and binds CD38, and the sequence the amino acids of said second polypeptide are shown in SEQ ID NO: 67, SEQ ID NO: 68, SEQ ID NO: 69 and SEQ ID NO: 70, and it binds CD3 epsilon;

ii) a protein complex of CD3 and CD38, wherein the amino acid sequence of said first polypeptide is shown in SEQ ID NO: 71, and it joins a cognate light chain with the amino acid sequence shown in SEQ ID NO: 72 and binds CD38, and the sequence the amino acids of said second polypeptide are shown in SEQ ID NO: 67, SEQ ID NO: 68, SEQ ID NO: 69 and SEQ ID NO: 70, and it binds CD3 epsilon;

iii) a protein complex of CD3 and CD38, wherein the amino acid sequence of said first polypeptide is shown in SEQ ID NO: 73, and it joins a cognate light chain with the amino acid sequence shown in SEQ ID NO: 74 and binds CD38, and the sequence the amino acids of said second polypeptide are shown in SEQ ID NO: 67, SEQ ID NO: 68, SEQ ID NO: 69 and SEQ ID NO: 70, and it binds CD3 epsilon;

In a further embodiment, the present invention provides a heterodimeric immunoglobulin or fragment thereof that binds to the protein complex of CD3 and CD38, wherein the amino acid sequence of said first polypeptide is shown in SEQ ID NO: 75 or 76 and is assembled with a light chain with the amino acid sequence shown in SEQ ID NO: 77, and binds CD38, and wherein the amino acid sequence of said second polypeptide is selected from the group consisting of SEQ ID NO: 67, SEQ ID NO: 68, SEQ ID NO: 69, and SEQ ID NO: 70, and it binds CD3 epsilon; protein complex of CD3 and CD38, wherein the amino acid sequence of said first polypeptide is shown in SEQ ID NO: 78, and it assembles to a light chain with the amino acid sequence shown in SEQ ID NO: 66 and binds CD38, and the amino acid sequence of said second polypeptide is selected from the group

- 5 039658 comprising the sequences of SEQ ID NO: 67, SEQ ID NO: 68, SEQ ID NO: 69 and SEQ ID NO: 70, and it binds CD3 epsilon.

In accordance with a further aspect of the present invention, there is provided a heterodimeric immunoglobulin, or fragment thereof, comprising (a) a first polypeptide comprising an immunoglobulin constant region with a CH2 domain and a first engineered CH3 domain;

(b) a second polypeptide comprising an immunoglobulin constant region with a CH2 domain and a second engineered CH3 domain, said two polypeptides being heterodimerized by the engineered CH3 domains; wherein the first polypeptide does not bind to protein A or protein G, and the second polypeptide binds only to protein A and/or protein G via an immunoglobulin constant region; wherein the first engineered CH3 domain contains substitutions at one or more of the following residues 20, 22, 26, 79, 85.1, 86, 88, 90 and the second engineered CH3 domain contains substitutions at least one residue selected from group 3 , 5, 20, 22, 26, 27, 81, 84, 85.1, 86, 88, and said second strand also contains a substitution in residue 84.4 according to IMGT numbering.

According to a further aspect of the present invention, the substitution at position 84.4 is selected from the group 84.4Q, 84.4V, 84.4A, 84.4S, 84.4T, 84.4N, 84.4G, 84.4L, 84.4L, 84.4Y, 84.4F, 84.4M.

In accordance with a further aspect of the present invention, the first engineered CH3 domain contains substitutions at least the following residues 22, 86, 88, and the second engineered CH3 domain contains substitutions at least the following residues 7, 84.4, 85.1, 86 according to IMGT numbering.

In accordance with a further aspect of the present invention, the first engineered CH3 domain contains substitutions at least the following residues 20, 22, 79, 86, 88, and the second engineered CH3 domain contains substitutions at least the following residues 7, 26, 84, 84.4, 85.1, 86 according to IMGT numbering.

In accordance with a further aspect of the present invention, the first engineered CH3 domain contains substitutions in at least the following residues 20, 22, 26, 79, 85.1, 86, 88, 90, and the second engineered CH3 domain contains substitutions in at least the following residues 3, 5, 7, 20, 22, 26, 81, 84, 84.2, 84.4, 85.1, 86, 88, 90.

In accordance with a further aspect of the present invention, the first engineered CH3 domain contains substitutions in at least the following residues 3, 20, 22, 26, 79, 85.1, 86, 88, 90, and the second engineered CH3 domain contains substitutions in at least , in the following residues 3, 5, 7, 20, 22, 26, 81, 84, 84.2, 84.4, 85.1, 86, 88, 90.

In accordance with a further aspect of the present invention, there is provided a heterodimeric immunoglobulin which comprises a first chain containing an IgG3 isotype Fc region that will contain a BTA CH3 domain and a non-VH3 variable domain or a VH3-based variable domain that has been devoid of the ability to bind to protein A (e.g., using G65S or N82aS substitutions), or will not contain a variable domain and therefore cannot bind to protein A, and a second strand that binds protein A containing the BTB CH3 domain of D401Q (e.g., derived from the human IgG1 isotype ) and either a non-VH3 variable domain, or a VH3 variable domain that has been deprived of the ability to bind to protein A (eg, using G65S or N82aS substitutions), or does not contain a variable domain.

Brief description of the figures

Fig. 1A-F: All of these figures refer to humanization of OKT3 on stable human scaffolds.

Fig. 1A-C: Brief description of humanized candidate antibodies in human IgG1 antibody format. HPB-ALL staining compared to OKT3 chimeric antibody: (-) indicates no binding, (+) weaker binding, (++) moderate binding, and (+++) similar binding.

Fig. 1D: DSC profiles of selected candidate antibodies.

Fig. 1E: Brief description of humanized candidate antibodies in scFv-Fc fusion format. HPB-ALL staining compared to OKT3 chimeric antibody: (-) indicates no binding, (+) weaker binding, (++) moderate binding, and (+++) similar binding.

Fig. 1F: DSC profiles of selected candidate scFv-Fcs.

Fig. 2A-B: All of these figures refer to humanization of SP34 on stable human scaffolds.

Fig. 2A: Brief description of humanized candidate antibodies in human IgG1 antibody format.

Fig. 2B: Brief description of humanized candidate antibodies in scFvFc fusion protein format (Fc from the human IgG1 isotype). Surface plasmon resonance (SPR) results compared with SP34 chimeric antibody to human and cynomolgus monkey CD3-epsilon 1-26_Fc fusion proteins: (-) indicates no binding, (+) weaker binding, (++) moderate binding, strong , but not similar binding (+++), and (++++) similar binding.

In FIG. 3 shows the relative expression levels after format change from IgG1 to scFv-Fc for the SP34 chimera as well as for SP34 H1L21, where a sharp decrease in expression was observed.

- 6 039658

In FIG. 4 shows the effect of alanine scanning on the expression level of SP34 H1L21 ScFv-Fc at positions: T27, G27a, V27c, T28, T29, S30, N31, Y32, N52, K53, R54, P56, L90, Y92, S93, N94, and L95.

In FIG. 5A shows the effect on expression level of a random mutation at position 29 in SP34 H3L23 ScFv-Fc; in fig. 5B shows the effect on expression level of a random mutation at position 30 in SP34 H3L23 ScFv-Fc; in fig. 5C shows the effect on expression level of a random mutation at position 95 in SP34 H5L23 ScFv-Fc.

In FIG. 6 shows normalized expression levels for several humanized SP34s.

Fig. 7A: Antibody-antigen interaction measured by SPR between chimeric HB-7 antibody and human CD38 antigen. Protein G was covalently linked to the CM5 sensor chip and 200 response units (RU) of the HB-7 chimeric antibody were captured. Human CD38 protein (human CD38 extracellular domain with polyhistidine tag) was injected at a concentration of 125, 31, 7.8, 3.9, 1.9, 1 and 0.5 nM at a flow rate of 30 μl/min in HBS-P.

Fig. 7B: Antibody-antigen interaction measured by SPR between humanized best-fit HB-7 antibody and human CD38 antigen. Protein G was covalently linked to the CM5 sensor chip and 200 RU of the humanized best-match HB-7 antibody was captured. Human CD38 protein (extracellular domain of human CD38 polyhistidine tagged) was injected at a concentration of 50, 25, 12.5, 6.25 and 0.39 nM at a flow rate of 30 μl/min in HBS-P.

Fig. 7C: Antibody-antigen interaction measured by SPR between humanized best-matched 9G7 antibody and human CD38 antigen. Protein G was covalently linked to the CM5 sensor chip and 200 RU of the humanized best match 9G7 antibody was captured. Human CD38 protein (extracellular domain of human CD38 polyhistidine tagged) was injected at concentrations of 25, 12.5, 6.25, 3.12, 1.56, 0.78, 0.39, 0.19, and 0.1 nM at a flow rate of 30 µl/min in HBS-P.

Fig. 7D: Antibody-antigen interaction measured by SPR between humanized 9G7 scaffold best antibody and human CD38 antigen. Protein G was covalently linked to the CM5 sensor chip and 200 RU of the humanized antibody with the best framework 9G7 was captured. Human CD38 protein (the extracellular domain of human CD38 polyhistidine tagged) was injected at concentrations of 50, 25, 12.5, 6.25, 3.12, 1.56, 0.78, 0.39, 0.19, and 0. 1 nM at a flow rate of 30 µl/min in HBS-P.

Fig. 7E: Antibody-antigen interaction measured by SPR between human antibody 767 and human CD38 antigen. Protein G was covalently linked to the CM5 sensor chip and 200 RU of human 767 antibody was captured. Human CD38 protein (the extracellular domain of human CD38 polyhistidine tagged) was injected at a concentration of 500, 250, 125, 62.5, 31.25, and 15.6 nM at a flow rate of 30 μl/min in HBS-P. Affinity was obtained from a plot of the equilibrium response (Req) versus analyte concentration (C) according to the following equation: Req=KAxCxRjnax/(KAxCxn+1), concentration at 50% saturation was KD. All GNR results are expressed as the number of response units (RU for short; y-axis) versus time (x-axis).

Fig. 7F: DSC profiles of the most relevant chimeric HB-7 and humanized HB-7 antibodies.

Fig. 7G: DSC profiles of the most relevant chimeric 9G7 and humanized 9G7 antibodies.

Fig. 7H: DSC profiles of the humanized antibody with the best scaffold 9G7.

Fig. 7I: DSC profiles of human antibody clone 767.

Fig. 7J: Summary table for humanized 9G7 antibodies.

Fig. 8: Schematic representation of the most appropriate BEAT CD38-HB7/CD3 15 (format A) and BEAT CD3 8-767/CD3 (format B) antibodies. [(A+)] means a functional protein A binding site. [(A-)] means a non-functional protein A binding site.

Fig. 9A: Antibody-antigen interaction measured by NIR between the most appropriate BEAT CD38-HB7/CD3 antibody and human CD38 antigen. Protein G was covalently linked to a 20 CM5 sensor chip and 200 RU of the most appropriate BEAT CD38-HB7/CD3 antibody was captured. Human CD38 protein (polyhistidine labeled protein) was injected at concentrations of 50, 25, 12.5, 6.25 and 0.39 nM at a flow rate of 30 μl/min in HBS-P. Results are expressed as response units (abbreviated RU; y-axis) versus time (x-axis).

Fig. 9B: DSC profile 25 of the most appropriate BEAT CD38-HB7/CD3 antibody.

Fig. 10: Example of retargeting with the most appropriate BEAT CD38-HB7/CD3 T cell killing antibody.

Data readout: RDL-FACS method. Effector cells: purified T cells 30 human. The ratio of effector cells to target cells was 10:1. Mean of two donors after 48 hours of incubation. Target cells: RPMI 8226. Antibody concentration is shown in nM.

Fig. 11: Example of BEAT antibody retargeting CD38-767/CD3(SP34) to kill T cells. Data readout: RDL-FACS method. Effector cells: human peripheral blood mononuclear cells (PBMCs). The ratio of effector cells to target cells was 10:1. Mean of three donors after 24 hours of incubation. Target cells: Daudi. The antibody concentration is shown in nM.

- 7 039658

Fig. 12: Schematic representation of the most appropriate BEAT CD38-HB7/CD3(SP34) (format A) and the most appropriate BEAT CD38-9G7/CD3(SP34-kappa2) (format B) antibodies. [(A+)] means the functional binding site of protein A.

Fig. 13: Example of retargeting with the most appropriate BEAT CD38-HB7/CD3(SP34) T cell killing antibody. Data readout: RDL-FACS method. Effector cells: human PBMC. The ratio of effector cells to target cells was 10:1. Mean of three donors after 24 hours of incubation. Target cells: Daudi cells. The antibody concentration is shown in nM.

Fig. 14: Antibody-antigen interaction measured by NIR between the most appropriate BEAT CD38-9G7/CD3(SP34-kappa2) antibody and human_Fc fusion protein CD3 epsilon 1-26. 500 RU of CD3 epsilon 1-26 human_Fc fusion protein was covalently bound to the CM5 sensor chip. The most appropriate BEAT CD38-9G7/CD3(SP34-kappa2) antibody was injected at concentrations of 50, 25, 12.5, 6.2, 3.1, 0.8 and 0.4 nM at a flow rate of 30 μl/min in HBS -P. The results are expressed as the number of response units (abbreviated as RU; y-axis) versus time (x-axis).

Fig. 15: Example of retargeting of T cell killing by the BEAT CD38/CD3(SP34kappa2) antibody. Data readout: RDL-FACS method. Effector cells: human PBMC. The ratio of effector cells to target cells was 10:1.

Mean of three donors after 24 hours of incubation. Target cells: Daudi cells. The antibody concentration is shown in nM.

In FIG. 16 shows the performance in RDL assay of several CD38/CD3 bispecific antibodies in which the CD3 binding arm of the antibody contains the original mouse SP34 in scFv format (SEQ ID NO: 137) or modified humanized scFv SP34 containing the indicated heavy/light chain combinations. : H1/L21 (SEQ ID NO: 67), H5/L32 (SEQ ID NO: 68), H5/L65 (SEQ ID NO: 69) and H5/L67 (SEQ ID NO: 70).

Fig. 17: Non-reducing PAAG/SDS analysis of BTB homodimers with and without protein A-binding scFv (lane group 1) and without it (lane group 2). Purification using affinity chromatography: eluted and neutralized fractions were analyzed after protein A chromatography (A) or protein G chromatography (G). For group 2, flow through or unbound fractions were collected and showed no binding to protein A or G (abbreviated as A ft and G ft, respectively). Elution was carried out at pH 3.0. [(A+)] denotes the protein A binding site. Molecular weight markers (kDa) are indicated.

Fig. 18A: Non-reducing PAAG/SDS analysis of a BEAT heterodimer containing only one protein A binding site in the BTB chain bearing the D401Q substitution. Purification using affinity chromatography: eluted and neutralized fractions were analyzed after protein A chromatography (A) or protein G chromatography (G). Elution was carried out at pH 3.0. [(A+)] denotes the protein A binding site. Molecular weight markers (kDa) are indicated; 18B: PAAG/SDS analysis of a BEAT heterodimer containing only one protein A binding site in the BTB chain that does not have a D401Q substitution. Purification using affinity chromatography: eluted and neutralized fractions were analyzed after protein A chromatography (A) or protein G chromatography (G). Elution was carried out at pH 3.0. [(A+)] denotes the protein A binding site. Molecular weight markers (kDa) are indicated.

Fig. 19: The graphs show the number of cells per µl of blood from the CD4+ population (solid line) and the CD8+ population (dashed line) for both animals injected with CD3/CD38 BEAT. The upper charts show the amounts for animals dosed at 1 μg/kg. The lower charts show amounts for animals dosed at 10 μg/kg. Numbers at time zeros correspond to samples collected on day 1 before dosing.

Fig. 20: The graphs show the number of cells per μl of blood from the CD14+CD38+ monocyte population for both animals injected with CD3/CD38 BEAT. The upper charts show the amounts for animals dosed at 1 μg/kg. The lower charts show amounts for animals dosed at 10 μg/kg. Numbers at time zeros correspond to samples collected on day 1 before dosing.

Fig. 21: The graphs show the mean fluorescence intensity (MFI) of the fluo-4 dye as a function of time in seconds. Each diagram shows a condition with isotype control. After collecting background data for 40 seconds, the indicated antibodies were added to the samples and data collection was resumed. The isotype control was a human IgG1 antibody. An increase in MFI indicates the mobilization of calcium into the cytoplasm of cells.

Fig. 22: Epitope mapping of the humanized 9G7 antibody by SPR using peptide-Fc fusion proteins derived from the extracellular domain of human CD38. The results are expressed as the number of response units (abbreviated as RU; y-axis) versus time (x-axis).

Fig. 23: 3D rendering of the extracellular domain of human CD38. The F6 peptide sequence is shaded in gray, positions M110 and T148 are shaded in black.

Fig. Figure 24: Epitope mapping of humanized 9G7 and SAR650984 antibodies by SPR with

- 8 039658 using the extracellular domains of human and cynomolgus CD38. The results are expressed as the number of response units (abbreviated as RU; y-axis) versus time (x-axis).

Fig. 25: Epitope mapping of human antibody 767 by SPR using peptide-Fc fusion proteins derived from the extracellular domain of human CD38. The results are expressed as the number of response units (abbreviated as RU; y-axis) versus time (x-axis).

Fig. 26: 3D rendering of the extracellular domain of human CD38. The F3 peptide sequence is shaded in gray, positions E76 and H79 are shaded in black.

Fig. 27: Epitope mapping of the humanized SAR650984 antibody by SPR using peptide-Fc fusion proteins derived from the extracellular domain of human CD38. The results are expressed as the number of response units (abbreviated as RU; y-axis) versus time (x-axis).

Detailed description of the invention

The present invention generally relates to novel heterodimeric immunoglobulins that bind to the CD3 protein complex and to the CD38 antigen.

Moreover, these heterodimeric immunoglobulins have reduced or prevented protein A binding and can therefore be purified to very high purity using affinity chromatography.

For the purpose of explaining the present application, the following definitions will be used, and whenever applicable, terms used in the singular will also include the plural, and vice versa. It should be obvious that the terminology used in this text is for the sole purpose of describing particular implementations and is not intended to be limiting.

The terms polypeptide and protein refer to a polymer of amino acid residues in which amino acids are combined through peptide bonds to form a chain of amino acids that have been linked to each other by synthesis by dehydrogenation. Polypeptides and proteins can be synthesized by chemical synthesis or recombinant expression and are not limited by minimum amino acid length.

In accordance with the present invention, a group of polypeptides includes proteins, provided that these proteins consist of a single polypeptide chain. The polypeptides can further form multimers such as higher order dimers, trimers and oligomers, i.e. consisting of more than one polypeptide molecule. The polypeptide molecules forming such dimers, trimers, etc. may be identical or non-identical. The corresponding structures of such higher order multimers are therefore called homo- or heterodimers, homo- or heterotrimers, etc. An example of a heteromultimer is an antibody molecule which, in its naturally occurring form, consists of two identical light polypeptide chains and two identical heavy polypeptide chains. The terms polypeptide and protein also refer to naturally modified polypeptides/proteins in which the modification is carried out, for example, by post-translational modifications such as glycosylation, acetylation, phosphorylation, and the like. Such modifications are well known in the art. Moreover, for the purposes of the present invention, a polypeptide refers to a protein that contains modifications such as deletions, insertions and substitutions (which may be conservative in nature) in the native sequence. These modifications may be intentional, eg introduced by site-directed mutagenesis, or may be accidental, eg resulting from mutations in hosts that produce said proteins or errors in polymerase chain reaction (PCR) amplification.

The term CD3 complex in this text refers to a protein complex known as the CD3 T cell co-receptor (differentiation cluster 3) (Wucherpfennig KW et al. (2010) Cold Spring Harb Perspect Biol, 2(4): a005140). The CD3 protein complex consists of four distinct chains. In mammals, this complex includes a CD3y chain, a CD3δ chain, and two CD3ε chains. These chains bind to a molecule known as the T cell receptor (TCR) and to the ζ chain, resulting in the generation of an activating signal in T lymphocytes (van der Merwe PA & Dushek O (2011) Nat Rev Immunol, 11(1 ): 4755). The TCR, ζ chain, and CD3 molecules together make up the TCR complex. The CD3y, CD3δ and CD3ε chains are highly related cell surface proteins of the immunoglobulin superfamily containing a single immunoglobulin extracellular domain. The intracellular tails of CD3 molecules contain a single conserved motif, known as the immunoreceptor tyrosine activating motif, or ITAM for short, which is required for the TCR's signaling ability. Because CD3 is essential for T cell activation, drugs (often monoclonal antibodies) that target CD3 have been and are being investigated as immunosuppressive drugs.

The term immunoglobulin as used herein may be used interchangeably with the term antibody. Immunoglobulin includes full-length antibodies and any antigen-binding antibody fragment or individual antibody chains. Immunoglobulins may be homodimeric

- 9 039658 or heterodimeric. Immunoglobulins, and in particular naturally occurring antibodies, are glycoproteins that exist as one or more copies of Y-shaped units consisting of four polypeptide chains. Each Y-shaped unit contains two identical copies of the heavy (H) chain and two identical copies of the light (L) chain, named according to their relative molecular weights. Each light chain is paired with a heavy chain and each heavy chain is paired with another heavy chain. Covalent disulfide bonds between chains and non-covalent interactions bind these chains to each other. Immunoglobulins and, in particular, naturally occurring antibodies contain variable regions, which are two copies of the antigen binding site. Papain, a proteolytic enzyme, cleaves the Y-shaped unit into three separate molecules: two so-called Fab or FAB fragments (Fab = antigen-binding fragment) and one so-called Fc fragment or Fc region (Fc = crystallizable fragment). The Fab fragment consists of the entire light chain and part of the heavy chain. The heavy chain contains one variable region (VH) and either three or four constant regions (CH1, CH2, CH3 and CH4, depending on the class or isotype of the antibody). The region between the CH1 and CH2 regions is called the hinge region and provides flexibility between the two Y-shaped FAB fragments of an antibody molecule, allowing them to open and close to allow space for binding to two antigenic determinants separated by a fixed distance. The hinge region referred to in this text is a 6-62 amino acid sequence found only in IgA, IgD and IgG, which includes cysteine residues that bridge two heavy chains. Each of the heavy chains of IgA, IgD and IgG contains four regions, ie, one variable region (VH) and three constant regions (CH1-3). IgE and IgM contain one variable and four constant regions (CH1-4) on the heavy chain. Immunoglobulin constant regions can mediate binding to host tissues or factors, including various cells of the immune system (eg, effector cells) and the first component (C1q) of the classical complement pathway. Each light chain is usually linked to the heavy chain by one covalent disulfide bond. Each light chain contains one variable region (VL) and one light chain constant region. The light chain constant region is the kappa light chain constant region, referred to herein as IGKC, or is the lambda light chain constant region, referred to herein as IGLC. IGKC is used in this text equivalent to Sk or CK and the term has the same meaning. IGLC is used in this text equivalent to Cλ or CL and the term has the same meaning. The term IGLC region in this text refers to all lambda light chain constant regions, eg all lambda light chain constant regions selected from the group consisting of IGLC1, IGLC2, IGLC3, IGLC6 and IGLC7. The VH and VL regions can be further subdivided into regions of hypervariability called complementarity-determining regions (CDRs), which alternate with regions that are more conserved, called framework regions (FRs or FWs). Each VH and VL consists of three CDRs and four FRs, arranged from amino to carboxy in the following order: FR1, CDR1, FR2, CDR2, FR3, CDR3, FR4. The variable regions of the heavy and light chains contain an epitope-binding region that interacts with an antigen. Constructed immunoglobulins can include various formats of epitope binding regions, such as scFv, FAB, or dAb fragments. These fragments are typically fused to form an antibody-like structure by genetic fusion to the Fc region of an IgG. Engineered immunoglobulins can be engineered as homo- or heterodimers, with or without heterodimerization enhancement techniques, and can have mono- or bispecific binding properties.

The term full-length antibody in this text includes a structure that is a natural biological form of an antibody containing variable and constant regions. For example, in most mammals, including humans and mice, a full-length IgG antibody is a tetramer and consists of two identical pairs of two immunoglobulin chains, each pair containing one light and one heavy chain, each light chain containing immunoglobulin VL regions and a constant region light chain and each heavy chain contains immunoglobulin VH, CH1 (Cy1), CH2 (Cy2), CH3 (Cy3), and CH4 (Cy4) regions, depending on the class or isotype of the antibody). In some mammals, such as camels and llamas, IgG antibodies may consist of only two heavy chains, with each heavy chain containing a variable region attached to an Fc region.

Antibodies are grouped into classes, also called isotypes, which are determined genetically by a constant region. The human light chain constant regions are classified as kappa (CK) and lambda (Cλ) light chains. Heavy chains are classified as mu (μ), delta (δ), gamma (γ), alpha (α) or epsilon (ε) and define the antibody isotype as IgM, IgD, IgG, IgA and IgE, respectively. Thus, the isotype in this text means any of the classes and/or subclasses of immunoglobulins, which was determined by the chemical and antigenic properties of the constant regions. Known isotypes of human immunoglobulins are IGHG1 (IgG1), IGHG2 (IgG2), IGHG3 (IgG3), IGHG4 (IgG4),

- 10 039658

IGHA1 (IgA1), IGHA2 (IgA2), IGHM (IgM), IGHD (IgD), and IGHE (IgE). The so-called human immunoglobulin gamma IGHGP pseudogene is an additional human immunoglobulin heavy chain constant region gene that has been sequenced, but which does not encode a protein due to an altered switch region (Bensmana M et al., (1988) Nucleic Acids Res, 16(7): 3108). Despite the altered switch region, the human immunoglobulin gamma IGHGP pseudogene contains open reading frames for all heavy chain constant regions (CH1 - CH3) and the hinge. All open reading frames for its heavy chain constant regions encode protein regions that align well with all human immunoglobulin constant regions with predicted structural features. This additional pseudo-gamma isotype is referred to herein as IgGP or IGHGP. Other immunoglobulin pseudogenes have been described, such as the P1 and P2 pseudogenes of the human immunoglobulin heavy chain epsilon constant region (IGHEP1 and IGHEP2). The IgG class is most commonly used for therapeutic purposes. In humans, this class includes the IgG1, IgG2, IgG3, and IgG4 subclasses. In mice, this class includes the IgG1, IgG2a, IgG2b, IgG2c, and IgG3 subclasses.

The term immunoglobulin fragments in this text includes, but is not limited to: (i) a region, including, for example, a CH1, CH2, or CH3 region, (ii) a Fab fragment consisting of VL, VH, CL, or CK and CH1 regions, including Fab' and Fab'-SH, (ii) an Fd fragment consisting of VH and CH1 regions, (iii) a dAb fragment (Ward ES et al. (1989) Nature, 341(6242): 544-6) which consists of one variable region, (iv) F(ab') ₂ fragments, bivalent fragments containing two linked Fab fragments, (v) single chain Fv (scFv) fragments in which the VH region and the VL region are linked by a peptide linker that allows the two regions to connect to form an antigen binding site (Bird RE et al. (1988) Science, 242(4877): 423-6; Huston JS et al. (1988) Proc Natl Acad Sci USA, 85(16): 5879-83) , (vi) diabodies or tribodies, polyvalent or multispecific fragments constructed by gene fusion (Holliger P et al. (1993) Proc Natl Acad Sci USA, 90(14): 6444-8; Tomlin son I & Holliger P, (2000) Methods Enzymol, 326: 461-79), (vii) scFv, diabody or antibody region fused to an Fc region, and (viii) scFv fused to the same or different antibody.

The term variable region refers to regions or domains that mediate antigen binding and determine the specificity of a particular antibody for a particular antigen. In naturally occurring antibodies, the antigen binding site consists of two variable regions that determine specificity: one located on the heavy chain, referred to herein as the heavy chain variable region (VH), and the other located on the light chain, referred to herein as the variable region. light chain region (VL). In humans, the heavy chain variable region (VH) can be subdivided into seven subgroups or subclasses: VH1, VH2, VH3, VH4, VH5, VH6, and VH7. In some cases, specificity may be confined to only one of the two specified regions, as in single-domain antibodies from the heavy chains of antibodies found in camelids. V regions are typically approximately 110 amino acids in length and consist of relatively unchanged stretches of amino acid sequence called framework regions (FRs or non-hypervariable regions) of 15-30 amino acids separated by shorter highly variable regions called hypervariable regions that are 717 amino acids. The variable domains of native heavy and light chains contain four FRs that mostly adopt a beta-sheet configuration connected by three hypervariable regions that form loops. The hypervariable regions in each chain are held in close proximity by the FR and, together with the hypervariable regions from the other chain, contribute to the formation of an antibody antigen binding site (see Kabat EA et al., supra). The term hypervariable region in this text refers to the amino acid residues of an antibody that are responsible for antigen binding. A hypervariable region typically contains amino acid residues from a complementarity determining region or CDR that has the highest sequence variability and/or is involved in antigen recognition. All variable regions are numbered according to Kabat (Kabat EA et al., supra).

Many definitions of CDRs are used, and all of them are included in the scope of this application. The Kabat definition is based on sequence variability and is the most commonly used (Kabat EA et al., supra). The Chothia definition refers instead to the location of structural loops (Chothia & Lesk J. (1987) Mol Biol, 196: 901-917). The AbM definition is a compromise between the Kabat and Chothia definitions and is used in Oxford Molecular AbM antibody modeling software (Martin ACR et al., (1989) Proc Natl Acad Sci USA 86: 9268-9272; Martin ACR et al., ( 1991) Methods Enzymol, 203: 121-153; Pedersen JT et al. (1992) Immunomethods, 1: 126-136; Rees AR et al., (1996) In Sternberg M. J. E. (ed.), Protein Structure Prediction, Oxford University Press, Oxford, 141-172). A contact-based definition has recently been introduced (MacCallum RM et al., (1996) J Mol Biol., 262: 732-745) and is based on the analysis of complex structures available in the Protein Databank. The definition of CDR by IMGT® (International Immunogenetics Information System®, (http://www.imgt.org)) is based on the IMGT numbering of V-regions from all immunoglobulins and T-cell receptors from all

- 11 039658 species (IMGT®, International Immunogenetics Information System®; Lefranc MP et al. (1999) Nucleic Acids Res, 27(1): 209-12; Ruiz M et al., (2000) Nucleic Acids Res, 28 (1): 219-21; Lefranc MP (2001) Nucleic Acids Res, 29(1): 207-9; Lefranc MP (2003) Nucleic Acids Res, 31(1): 307-10; Lefranc MP et al., (2005) Dev Comp Immunol, 29(3): 185-203; Kaas Q et al., (2007) Briefings in Functional Genomics & Proteomics, 6(4): 253-64). All complementarity determining regions (CDRs) referred to in the present invention are preferably defined as follows (numbering according to Kabat EA et al., supra): LCDR1: 24-34, LCDR2: 50-56, LCDR3: 89-98, HCDR1: 26-35, HCDR2: 50-65, HCDR3: 95-102.

Non-CDR regions of the variable domain are known as framework regions (FRs). The non-CDR regions of the VL region herein include the amino acid sequences 1-23 (FR1), 35-49 (FR2), 57-88 (FR3) and 99-107 (FR4). Non-CDR regions of the VH region in this text include amino acid sequences: 1-25 (FR1), 36-49 (FR2), 66-94 (FR3) and 103-113 (FR4).

The CDRs of the present invention may include extended CDRs that are based on the above definitions and contain variable domain residues as described below: LCDR1: 24-36, LCDR2: 46-56, LCDR3: 89-97, HCDR1: 26-35, HCDR2 : 47-65, HCDR3: 93-102. These extended CDRs are also numbered according to Kabat et al., supra. The unexpanded portion of the VL region CDR in this text includes amino acid sequences: 1-23 (FR1), 37-45 (FR2), 57-88 (FR3) and 98-approximately 107 (FR4). The unexpanded portion of the VH region CDR in this text includes the amino acid sequences: 1-25 (FR1), 37-46 (FR2), 66-92 (FR3) and 103-approximately 113 (FR4).

The term Fab or FAB or Fab region or FAB region in this text includes polypeptides that contain VH, CH1, VL and immunoglobulin light chain constant regions. A Fab may refer to a given region as isolated, or to a given region in the context of a full length antibody or antibody fragment.

The term Fc or Fc region as used herein includes a polypeptide comprising an antibody heavy chain constant region excluding the first immunoglobulin constant region. Thus, Fc refers to the last two IgA constant regions, IgD and IgG, or the last three IgE and IgM constant regions, and the flexible hinge located on the N-side of these regions. For IgA and IgM, Fc may include the J chain. For IgG, Fc includes the immunoglobulin regions C-gamma-2 and C-gamma-3 (Cy2 and Cy3) and the hinge between C-gamma-1 (Cy1) and C-gamma2 (Cy2) . Although the boundaries of the Fc region may vary, the human IgG heavy chain Fc region is generally considered to comprise C226 or P230 residues at the carboxyl terminus, using EU index numbering. Fc may refer to a given region as isolated, or to a given region in the context of an Fc polypeptide, such as an antibody.

The term immunoglobulin constant region as used herein refers to the constant regions of an immunoglobulin heavy chain or antibody from a human or animal species, and is intended to include all isotypes. Preferably, the immunoglobulin constant regions are derived from a human and are selected from the group consisting of the following constant regions, but are not limited to IGHG1 CH1,

IGHG2 CHI, IGHG3 CHI, IGHG4 CHI, IGHA1 CHI, IGHA2 CHI, IGHE CHI, IGHEP1

CHI, IGHM CHI, IGHD CHI, IGHGP CHI, IGHG1 CH2, IGHG2 CH2, IGHG3 CH2, IGHG4

CH2, IGHA1 CH2, IGHA2 CH2, IGHE CH2, IGHEP1 CH2, IGHM CH2, IGHD CH2, IGHGP

CH2, IGHG1 CH3, IGHG2 CH3, IGHG3 CH3, IGHG4 CH3, IGHA1 CH3, IGHA2 CH3,

IGHE CH3, IGHEP1 CH3, IGHM CH3, IGHD CH3, IGHGP CH3, IGHE CH4 and IGHM CH4.

Preferred immunoglobulin constant regions are selected from the group consisting of IGHE CH2, IGHM CH2, IGHG1 CH3, IGHG2 CH3, IGHG3 CH3, IGHG4 CH3,

Human IGHA1 CHS, IGHA2 CHH, IGHE CHH, IGHM CHH, IGHD CHH and IGHGP CHH.

More preferred immunoglobulin constant regions are selected from the group consisting of

IGHG1 SNZ, IGHG2 SNZ, IGHG3 SNZ, IGHG4 SNZ, IGHA1 SNZ, IGHA2

CHS, IGHE CHH, IGHM CHH, IGHD CHH, and IGHGP CHH in humans.

The term epitope-binding region includes a polypeptide or fragment thereof having the minimum amino acid sequence that allows specific binding of an immunoglobulin molecule to one or more epitopes. Naturally occurring antibodies contain two epitope-binding regions, which are also known as antigen-binding or combining sites, or paratopes. Epitope-binding regions in naturally occurring antibodies are located within the CDR regions of the VH and/or VL domains that contain amino acids that mediate binding

- 12 039658 epitope. In addition to naturally occurring antibodies, artificial VH domains or VL domains, or fragments and combinations thereof, can be engineered to provide epitope binding regions (Holt LJ et al., (2003) Trends Biotechnol, 21(11): 484-490; Polonelli L et al., (2008) PLoS ONE, 3(6): e2371). Examples of non-immunoglobulin-derived epitope-binding regions can be found in artificial protein domains used as a scaffold for constructing epitope-binding regions (Binz HK et al. (2005) Nat Biotechnol, 23(10): 1257-1268), or peptidomimetics ( Murali R & Greene MI (2012) Pharmaceuticals, 5(2): 209-235). Preferably, the term epitope-binding region includes a combination of one or more heavy chain variable domains and one or more complementary light chain variable domains, which together form a binding site that allows specific binding of an immunoglobulin molecule to one or more epitopes. Examples of an epitope binding region provided in the present invention include scFv and FAB.

As used herein, the term epitope includes a fragment of a polypeptide or protein or non-protein molecule having antigenic or immunogenic activity in an animal, preferably a mammal, and most preferably a human. An epitope with immunogenic activity is a fragment of a polypeptide or protein that elicits a humoral immune response in an animal. An epitope with antigenic activity is a fragment of a polypeptide or protein to which an antibody or polypeptide specifically binds, as determined by any method well known to one of skill in the art, such as immunoassays. Antigenic epitopes need not be immunogenic. Preferably, the term epitope as used herein refers to a polypeptide sequence of at least about 3 to 5, preferably about 5 to 10, or up to 15, but not more than about 1000 amino acids (or any whole number of amino acids in between) , which specifies a sequence that, by itself or as part of a longer sequence, binds to an antibody that has been produced in response to such a sequence. There is no critical upper limit to the length of this fragment, which may contain almost a full-length protein sequence, or even a fusion protein containing one or more epitopes. An epitope for use in the present invention is not limited to a polypeptide containing the exact sequence of a portion of the original protein from which it was derived. Thus, the scope of the term epitope includes sequences that are identical to the native sequence, as well as modified native sequences, for example, by deletion, insertion and substitution (generally conservative in nature). Epitopes of protein antigens are classified into two categories: conformational epitopes and linear epitopes, based on their structure and interaction with the epitope binding site (Goldsby R et al., (2003) Antigens (Chapter 3) Immunology (Fifth edition, ed.), New York: W. H. Freeman and Company, pp. 57-75, ISBN 0-7167-4947-5). A conformational epitope consists of discontinuous stretches of the amino acid sequence of an antigen. The interaction of these epitopes with the paratope is based on the three-dimensional surface features and the shape or tertiary structure of the antigen. Most epitopes are conformational. In contrast, the interaction of linear epitopes with the paratope is based on their primary structure. A linear epitope is formed by a continuous sequence of amino acids from an antigen.

The term heterodimeric immunoglobulin, or heterodimeric fragment, or heterodimer, or heterodimer of heavy chains in this text includes an immunoglobulin molecule or part thereof, containing at least the first and second polypeptide, like the first and second sections, while the second polypeptide differs in amino acid sequence from the first polypeptide.

Preferably, the heterodimeric immunoglobulin contains two polypeptide chains, wherein the first chain contains at least one site that is not identical to that of the second chain, and both chains are connected, i. interact through non-identical sites. More preferably, the heterodimeric immunoglobulin can specifically bind to at least two different ligands, antigens, or binding sites, ie. it is bispecific. Heterodimeric immunoglobulin as used herein includes, but is not limited to, full-length bispecific antibodies, bispecific Fab, bispecific F(ab') ₂ , bispecific scFv fused to an Fc region, diabody fused to an Fc region, and an antibody domain fused to an Fc region.

The term homodimeric immunoglobulin, or homodimeric fragment, or homodimer, or heavy chain homodimer, as used herein, includes an immunoglobulin molecule, or portion thereof, comprising at least a first and a second polypeptide, similar to the first and second regions, wherein the second polypeptide is identical in amino acid sequence to the first polypeptide . Preferably, the homodimeric immunoglobulin contains two polypeptide chains, with the first chain containing at least one site identical to that of the second chain, and both chains are connected, i. interact through identical sites. Preferably, the homodimeric immunoglobulin fragment contains at least two regions, with the first region identical to the second region, and both regions are connected, i. interact through the protein-protein interface.

- 13 039658

For all constant regions of immunoglobulin included in the scope of the present invention, the numbering can be carried out according to IMGT® (IMGT®; above).

For all human immunoglobulin CH1, CH2, CH3 heavy chain constant regions selected from the group consisting of IGHG1, IGHG2, IGHG3, and IGHG4, numbering may follow the EU numbering system (Edelman GM et al., (1969) Proc Natl Acad Sci USA , 63(1): 78-85). Full correspondence for human constant regions CH1, hinge, CH2 and CH3 from IGHG1 can be found in the IMGT database (IMGT®; supra).

For the human immunoglobulin kappa (IGKC) light chain constant region, numbering can be according to the EU numbering system (Edelman GM et al., supra). A complete match for the human CK region can be found in the IMGT database (IMGT®; above).

For human lambda immunoglobulin light chain constant regions (IGLC1, IGLC2, IGLC3, IGLC6 and IGLC7), numbering can be done according to the Kabat numbering system (Kabat EA et al., supra). Full correspondence for human IGLC regions can be found in the IMGT database (IMGT®; supra).

The human IGHG1 heavy chain constant regions referred to in this text have the following boundaries: CH1 region (EU numbering: 118-215), γ1 hinge region (EU numbering: 216-230), CH2 region (EU numbering: 231-340) and the CH3 region (EU numbering: 341-447). The human CK region referred to in this text covers residues 108 to 214 (EU numbering). Human IGLC1, IGLC2, IGLC3, IGLC6 and IGLC7 regions referred to in this text cover residues 108-215 (Kabat numbering).

The terms amino acid or amino acid residue in this text include both natural amino acids and non-natural amino acids. Preferably natural amino acids are included.

The term amino acid modification or modification in this text includes substitution, insertion and/or deletion of an amino acid in a polypeptide sequence. The terms substitution, or amino acid substitution, or amino acid residue substitution in this text refers to the replacement of the first amino acid residue in a sequence of amino acids with a second amino acid residue, wherein the first amino acid residue is different from the second amino acid residue, i.e., the substituting amino acid residue is different from the amino acid , which has been replaced. For example, an R94K substitution polypeptide refers to a polypeptide variant in which the arginine at position 94 is replaced by a lysine. For example, 94K indicates a change at position 94 to lysine. For the purposes of the present invention, multiple substitutions are usually separated by a slash or a comma. For example, R94K/L78V or R94K, L78V refers to a dual variant containing R94K and L78V replacements. By amino acid insertion or insertion in this text is meant the addition of an amino acid at a specific position in the original polypeptide sequence. For example, insertion -94 refers to an insertion at position 94. An amino acid deletion or deletion in this text refers to the removal of an amino acid at a specific position in the original polypeptide sequence. For example, R94- denotes a deletion of arginine at position 94.

In some embodiments, the terms reduce, reduce, or decrease protein A binding refers to an overall reduction in binding of at least 25%, 30%, 40%, 50%, 60%, 70%, 80%, 85%, 90%, 95%, 97%, or 99% up to 100%. % (elimination) of the modified immunoglobulin or its fragment with protein A, which is detected using standard methods known in this field, such as those described in this text, compared with the original, i.e., unmodified, immunoglobulin, or wild-type IgG , or an IgG containing a wild-type human IgG Fc region. In some embodiments, these terms may alternatively refer to an overall reduction of 10 times (i.e., 1 log), 100 times (2 log), 1000 times (or 3 log), 10,000 times (or 4 log) or 100,000 times (or 5 log).

The terms eliminate, disrupt, eliminate, or disrupt protein A binding refer to an overall 100% reduction in the binding of a modified immunoglobulin or fragment thereof to protein A, i.e., the complete loss of binding of the modified immunoglobulin or fragment thereof to protein A, which is detected with using standard methods known in the art, such as those described in this text, compared with the original, ie, unmodified, immunoglobulin, or wild-type IgG, or IgG containing the wild-type human IgG Fc region.

Similarly, the terms reduce, reduce, or decrease binding to an affinity reagent refer to an overall reduction of at least 25%, 30%, 40%, 50%, 60%, 70%, 80%, 85%, 90%, 95%, 97%, or 99% up to 100% (elimination) of binding. a modified immunoglobulin or fragment thereof with an affinity reagent, which is detected using standard methods known in the art, such as those described in this text, compared to the original, i.e., unmodified, immunoglobulin, or wild-type IgG, or IgG, containing the Fc region of human IgG wild type. In some embodiments, these terms may alternatively refer to an overall reduction of 10 times (i.e., 1 log), 100 times (2 log), 1000 times (or 3 log), 10,000 times (or 4

- 14 039658 log), or 100000 times (or 5 log).

The terms eliminate, disrupt, eliminate, or disrupt binding to an affinity reagent refer to an overall 100% reduction in the binding of a modified immunoglobulin or fragment thereof to an affinity reagent, i.e., the complete loss of binding of a modified immunoglobulin or fragment thereof to an affinity reagent, which is detected with using standard methods known in the art, such as those described in this text, compared with the original, ie, unmodified, immunoglobulin, or wild-type IgG, or IgG containing the wild-type human IgG Fc region.

Bispecific antibodies are monoclonal antibodies that are capable of specifically binding to at least two different antigens. In some embodiments, the bispecific antibodies are bispecific antibodies with one or more amino acid modifications in the VH region compared to the parent antibody. In some embodiments, the bispecific antibodies may be human or humanized antibodies. Bispecific antibodies can also be used to deliver cytotoxic agents to cells that express the target antigen. These antibodies contain an arm that binds the target antigen and an arm that binds a cytotoxic agent such as, for example, saporin, an anti-interferon-alpha antibody, a vinca alkaloid, a ricin A chain, methotrexate, or a radiolabelled hapten. Bispecific antibodies can be obtained as full length antibodies or antibody fragments. Methods for producing bispecific antibodies are known in the art. Typically, recombinant production of bispecific antibodies is based on the co-expression of two pairs of immunoglobulin heavy chains and light chains, with the two heavy chains having different specificities (Milstein and Cuello, (1983) Nature, 305: 537-40). Due to the random distribution of immunoglobulin heavy and light chains, these hybridomas (quadromas) form a potential mixture of different antibody molecules, of which only one has the correct bispecific structure. Purification of the correct molecule, which is usually carried out using affinity chromatography steps, is rather laborious and product yields are low. Similar procedures are described in WO 1993/08829 and Traunecker et al. (1991) EMBO J, 10:3655-9. In a different approach, antibody variable regions with desired binding specificities (antibody and antigen binding sites) are fused to immunoglobulin constant region sequences. The fusion is, for example, carried out with an immunoglobulin heavy chain constant region containing at least a portion of the hinge and CH2 and CH3 regions. In some embodiments, a first heavy chain constant region (CH1) containing a site required for light chain binding is present in at least one of the fusion proteins. DNA encoding immunoglobulin heavy chain fusion proteins and, optionally, immunoglobulin light chain, are inserted into separate expression vectors and co-transfected with them into a suitable host organism. This provides the flexibility to adjust the mutual ratios of these three polypeptide fragments in embodiments in which unequal ratios of the three polypeptide chains are used for construction, providing optimal yields. However, it is possible to insert the coding sequences for two or all of these three polypeptide chains into a single expression vector if expression of at least two polypeptide chains in equal ratios results in high yields or if the ratios are not particularly important.

Bispecific antibodies include crosslinked or heteroconjugated antibodies. For example, one of the antibodies in the heteroconjugate may be coupled to avidin and the other to biotin. Such antibodies, for example, have been proposed for targeting cells of the immune system to unwanted cells (US 4676980) and for the treatment of HIV infection (WO 1991/00360, WO 1992/00373 and EP 03089). Heteroconjugated antibodies can be made using any convenient crosslinking technique. Suitable crosslinking agents are well known in the art (see US 4,676,980), along with a variety of crosslinking techniques. Antibodies with more than two valences are also provided. For example, trispecific antibodies can be prepared (see Tutt A et al. (1991) J. Immunol. 147:60-9).

Protein A: Protein A is a cell wall component produced by several strains of Staphylococcus aureus that consists of a single polypeptide chain. The protein A gene product consists of five homologous repeats attached in tandem to the pathogen cell wall. These five domains are approximately 58 amino acids in length, and the domains are designated EDABC, each of which exhibits immunoglobulin binding activity (Tashiro M & Montelione GT (1995) Curr. Opin. Struct. Bid., 5(4): 471-481). Five homologous immunoglobulin-binding domains form a three-stranded bundle. Each domain is capable of binding to proteins from many mammalian species, most notably IgG (Hober S et al. (2007) J. Chromatogr. In Analyt. Technol. Biomed. Life Sci., 848(1): 40-47). Protein A binds to the heavy chain of most immunoglobulins within the Fc region, but also within the FAB region in the case of human VH3 families (Jansson B et al., (1998) FEMS Immunol. Med. Microbiol., 20(1): 69-78 ). Protein A binds to IgG from various species including human, mouse, rabbit, and guinea pig, but does not bind to human IgG3 (Hober S et al., (2007)

- 15 039658 above). The inability of human IgG3 to bind block A can be explained by the H435R and Y436F substitutions in the Fco region of human IgG3 (EU numbering, Jendeberg et al., (1997) J. Immunol. Methods, 201(1): 25-34). In addition to IgG, protein A also interacts with IgM and IgA.

Among the human VH subclasses, VH3 is the only subclass that binds to protein A (Graille M et al. (2000) Proc. Natl. Acad. Sci. USA 97(10): 5399-5404) and all five domains are known to A proteins bind to this subclass of variable domains (Jansson B et al., (1998) FEMS Immunol. Med. Microbiol., 20(1): 69-78). VH3-based immunoglobulins or fragments thereof are of great importance in the biotechnology industry. VH3-based molecules are designed on a large scale because their ability to bind protein A facilitates functional pre-screening, and for this reason many synthetic or donor-derived phage display libraries or transgenic animal techniques used for antibody discovery are based on the VH3 subclass. . Additionally, VH3-based molecules are often chosen due to their good expression and stability compared to other known heavy chain variable domain subclasses.

The ability of protein A to bind antibodies with such high affinity is the main motivation for its industrial scale application in biopharmaceuticals. Protein A, used to make antibodies in biopharmaceuticals, is usually produced recombinantly in E. coli and has essentially the same functions as native protein A (Liu HF et al., (2010) MAbs, 2(5): 480 -499).

Typically, recombinant protein A is coupled to a stationary phase, a chromatography resin, to purify antibodies. Optimal binding occurs at pH 8.2, although binding is also good under neutral or physiological conditions (pH 7.0-7.6). Elution is usually achieved by shifting towards a more acidic pH (glycine-HCl, pH 2.5-3.0). This results in efficient dissociation of most protein-protein and antibody-antigen interactions, without irreversible impact on protein structure. However, some antibodies and proteins are damaged by low pH and it is best to neutralize them immediately after preparation by adding 1/10 volume of an alkaline buffer such as 1 M Tris-HCl, pH 8.0 to minimize the length of time under low pH conditions. .

There are various commercially available protein A chromatography resins. The main differences between these media are the type of support matrix, protein A ligand modification, pore size, and particle size. Differences in these factors lead to differences in compressibility, chemical and physical stability, diffusion resistance and binding capacity of these adsorbents (Hober S et al., (2007), supra). Examples of Protein A chromatography resins include, but are not limited to, MabSelect SuRe™ Protein A Resin and MabSelect™ Protein A Resin from GE Healthcare, which are used in the examples.

The term chromatography refers to protein liquid chromatography and includes rapid protein liquid chromatography (FPLC), which is a form of liquid chromatography often used to analyze or purify mixtures of proteins. As with other chromatography formats, separation is possible because the different components of a mixture have different affinities for two materials: a moving liquid (mobile phase) that passes through a porous solid (stationary phase). In FPLC, the mobile phase is an aqueous solution or buffer. The flow rate of the buffer can be controlled by allowing the liquid to flow under gravity or controlled by a positive displacement pump, which is usually kept at a constant speed, while the composition of the buffer can be changed by supplying liquids in various ratios from two or more external tanks. The stationary phase is a resin consisting of beads, usually cross-linked agarose, packed into a cylindrical glass or plastic column. FPLC resins are available in a wide range of bead sizes and surface ligands depending on the application.

The affinity chromatography process involves the use of an affinity reagent as ligands that are cross-linked to a stationary phase and that have a binding affinity for a specific molecule or class of molecules. The ligands may be biomolecules, such as protein ligands, or may be synthetic molecules. Both types of ligands generally show good specificity. The most commonly used protein ligand in manufacturing is the protein A affinity reagent. In affinity chromatography, when a solution (e.g., crude cell supernatant containing a protein of interest) is loaded onto a column, the target protein is usually adsorbed, allowing impurities (other proteins, lipids, carbohydrates, DNA, dyes, etc.) to flow through the column. The adsorbent is usually packed into a chromatographic column; however, the adsorption step can be carried out using a stirred slurry adsorbent in batch binding mode. The next step after adsorption is a washing step in which the adsorbent is washed to remove residual impurities. The bound protein is then eluted semi-pure or pure. Elution is usually carried out by changing the composition of the buffer or salt so that the protein can no longer interact with the immobilizo.

- 16 039658 bath ligand and released. In some cases, the protein of interest may not bind to the affinity resin and affinity chromatography aims to bind undesired impurities and therefore the unbound fraction is collected to isolate the protein of interest. Affinity chromatography can be carried out in a fixed bed or in a fluidized bed.

The term gradient chromatography refers to a chromatography method in which the percentage of elution buffer (buffer B) is increased from 0 to 100% in a gradual or stepwise manner.

The terms capture-elution chromatography or capture-elution purification or capture-elution purification refers to a chromatography method in which the proportion of elution buffer (buffer B) is not increased from 0 to 100% in a gradual or stepwise manner, but instead directly used at 100% concentration after capture and possibly a running buffer (buffer A) wash step.

Development of heterodimeric immunoglobulins targeting CD3 and CD38

The present invention provides an epitope-binding region that binds a CD3 protein complex, comprising the heavy and light chain CDRs as described above, and further comprising a heavy chain variable region framework region that is a product of or derived from the human IGHV3-23*04 gene (SEQ ID NO: 37). The heavy chain variable region framework region comprises a modification of at least one amino acid from the corresponding heavy chain variable region framework region of the corresponding mouse OKT3 antibody comprising the amino acid sequence of SEQ ID NO: 38. Preferably, said amino acid modification is an amino acid substitution. Typically, no more than seven, preferably no more than six, preferably no more than five, preferably no more than four, more preferably no more than three, even more preferably no more than two, most preferably no more than one amino acid modifications are made to the framework region. In some embodiments, provided herein is an epitope-binding region that binds to a CD3 protein complex, wherein the modification of an amino acid in the heavy chain variable region framework regions comprises a substitution of an amino acid at a position selected from the group consisting of: 34, 48, 49, 58 , 69, 71 and 73, and the position of the amino acid of each member of the group is indicated according to the Kabat numbering. Preferably, the amino acid substitutions in the heavy chain variable region framework regions are selected from the group consisting of: 134M, V48I, A49G, R58N, R58Y, I69L, A71T and T73K. Preferred amino acid substitutions in the heavy chain variable framework regions are at amino acid positions selected from the group consisting of 34, 49 and 71. More preferred amino acid substitutions in the heavy chain variable framework regions are selected from the group consisting of I34M, A49G and A71T.

In a further aspect, the epitope-binding region of the first polypeptide that binds the CD3 protein complex comprises a light chain variable region framework region that is the product of or derived from a human gene selected from the group consisting of: IGKV1-39*01 (SEQ ID NO: 39 ) and IGKV3-20*01 (SEQ ID NO: 40). The light chain variable region framework region comprises a modification of at least one amino acid from the corresponding light chain variable region framework region of the corresponding mouse OKT3 antibody comprising the amino acid sequence of SEQ ID NO: 41. Preferably, said amino acid modification is an amino acid substitution. Typically, no more than eight, preferably no more than seven, preferably no more than six, preferably no more than five, preferably no more than four, more preferably no more than three, even more preferably no more than two, most preferably no more than one amino acid modifications are made to the framework region. In some embodiments, provided herein is an epitope-binding region that binds to a CD3 protein complex, wherein amino acid modification in the framework regions of the light chain variable region sequence comprises substitution of an amino acid at a position selected from the group consisting of: 4, 33, 34, 46, 47, 66, 71, and 96. Preferably, the amino acid substitutions in the light chain variable framework regions are selected from the group consisting of: M4L, V33M, A34N, L46R, L47W, R66G, F71Y, and P96F. Preferred amino acid substitutions in the light chain variable framework regions are at amino acid positions selected from the group consisting of 4, 46 and 47. More preferred amino acid substitutions in the light chain variable framework regions are selected from the group consisting of M4L, L46R, L47W and F71Y. In some embodiments, the epitope-binding region of the first polypeptide that binds to the CD3 protein complex may include the amino acid modifications in the framework regions of the heavy chain variable region sequence described above and the amino acid modifications in the framework regions of the variable light chain sequence described above.

Also provided herein is an antibody or fragment thereof that binds to a CD3 protein complex that includes a heavy chain sequence selected from the group consisting of SEQ ID NOs: 79 to 90, 91-95 and 64, preferably consisting of SEQ ID NO: 64. Also provided herein is an antibody or fragment thereof that binds

- 17 039658 are associated with a CD3 protein complex which includes a light chain sequence selected from the group consisting of SEQ ID NOs: 96 to 104, 105 to 126, 127, 128 and 129, preferably consisting of SEQ ID NOs: 127 .

Because each of these heavy and light chain variable region sequences can bind to the CD3 protein complex, the heavy and light chain variable region sequences can be mixed and matched to produce the CD3 binding molecules of the present invention. The binding of CD3 to such mixed and matched antibodies can be examined using the binding assays described, for example, in the examples.

Construction of an immunoglobulin constant region to promote heterodimer formation rather than homodimer formation

Methods for producing heterodimeric immunoglobulins are known in the art, and one of the simplest methods is based on the expression of two different immunoglobulin chains in one cell (WO95/33844, Lindhofer H & Thierfelder S). Without such design, this direct method is limited in that it produces homodimeric species rather than the heterodimeric ones of interest (Kufer P et al. (2004) Trends Biotechnol., 22(5): 238-244). By using additional techniques that increase heavy chain heterodimerization (Merchant AM et al. (1998) Nat. Biotechnol., 16(7): 677-681), more heterodimer production can be achieved, but a significant amount of unwanted homodimers will still be formed. (Jackman J et al. (2010) J Biol Chem. 285(27):20850-9, Klein C et al. (2012) MAbs. 4(6):653-63). The present invention therefore uses the described BEAT® method (PCT Publication No. WO 2012/131555), which is based on a unique biomimicry concept that provides better heterodimerization than prior art methods. The BEAT technique is based on the exchange of surface regions between naturally occurring pairs of immunoglobulin homo- or heterodimeric domains to generate new heterodimers that can be used as building blocks to create Fc-based bispecific antibodies.

In one aspect of the present invention, a heterodimeric immunoglobulin or fragment thereof is provided, comprising a first and a second polypeptide comprising an engineered immunoglobulin constant region with a modified CH3 domain containing a protein-protein contact surface, wherein the protein-protein contact surface of the first polypeptide contains an amino acid substitution at the position, selected from the group consisting of: 3, 5, 7, 20, 22, 26, 27, 79, 81, 84, 84.2, 85.1, 86, 88 and 90 (IMGT® numbering) and the protein-protein interface the second polypeptide contains an amino acid substitution at position 84.4 and at a position selected from the group consisting of 3, 5, 7, 20, 22, 26, 27, 79, 81, 84, 84.2, 85.1, 86, 88 and 90 (IMGT numbering ®).

In a further embodiment, the present invention provides a heterodimeric immunoglobulin, or fragment thereof, wherein the first and second polypeptides comprise an engineered immunoglobulin constant region with a modified CH3 domain containing a protein-protein contact surface, wherein the protein-protein contact surface of the first polypeptide contains an amino acid substitution in position 88 and at a position selected from the group consisting of: 3, 5, 7, 20, 22, 26, 27, 79, 81, 84, 84.2, 85.1, 86 and 90 (IMGT® numbering), and the surface the protein-protein contact of the second polypeptide contains an amino acid substitution at position 85.1 and/or 86 and at a position selected from the group consisting of 3, 5, 7, 20, 22, 26, 27, 79, 81, 84, 84.2, 84.4, 88 and 90 (IMGT® numbering), wherein the amino acid residue substituted at position 88 in the first engineered immunoglobulin constant region interacts with the amino acid residue substituted at position 85.1 and/or 86 in the second const a ruled immunoglobulin constant region, where the amino acid position of each member of the group is indicated according to the IMGT® numbering.

Preferably, the amino acid residue that is substituted at the protein-protein interface of the first engineered immunoglobulin constant region at position 88 is 88W and conservative amino acid substitutions at that position, with the amino acid position indicated according to IMGT® numbering. More preferably, the amino acid residue that is substituted at the protein-protein interface of the first immunoglobulin engineered constant region at position 88 is 88W, and the additional amino acid residue substituted at the protein-protein interface of the first engineered immunoglobulin constant region is selected from the group, consisting of 3A, 20V, 20T, 20A, 20N, 20Q, 20E, 20S, 20K, 20W, 22A, 22G, 22T, 22L, 22I, 22V, 26R, 26Q, 26T, 26K, 26V, 26S, 26N, 26E, 79Y, 85.1T, 85.1M, 85.1A, 85.1S, 85.1R, 85.1H, 85.1K, 85.1F, 85.1C, 85.IN, 85.1W, 86S, 86I, 86T, 86H, 86Q, 86V, 86W, 86Y, 86F and 90N, with the amino acid position indicated according to the IMGT® numbering.

Preferably, the amino acid residue that is substituted at positions 85 and 86 at the protein-protein interface of the second engineered immunoglobulin constant region is selected from the group consisting of 85.1A, 85.1S, 85.1C and 86S and conservative amino acid substitutions at that position (IMGT® numbering ). More preferably, the amino acid residue that is substituted at the protein-protein interface of the second engineered immunoglobulin constant region is selected.

- 18 039658 from the group consisting of 85.1A, 85.1S, 85.1C and 86S, and wherein the additional amino acid residue substituted at the protein-protein interface of the second engineered immunoglobulin constant region is selected from the group consisting of 3E, 5A, 7F , 20T, 22V, 26T, 81D, 84L, 84.2E, 88R and 90R and conservative amino acid substitutions at this position (IMGT® numbering).

In a preferred embodiment, the amino acid residue that is substituted at the protein-protein interface of the first immunoglobulin engineered constant region at position 88 is 88W, and the additional amino acid residue substituted at the protein-protein interface of the first immunoglobulin engineered constant region is 3A, 20K, 22V, 26T, 79Y, 85.1S, 86V and 90N and wherein the amino acid residues that are substituted at the protein-protein interface of the second engineered immunoglobulin constant region at positions 85.1 and 86 are 85.1A, 85.1S or 85.1A and 86S, and wherein the additional amino acid residue substituted at the protein-protein interface of the second engineered immunoglobulin constant region is 3E, 5A, 7F, 20T, 22V, 26T, 81D, 84L, 84.2E, 84.4Q, 88R and 90R (IMGT® numbering).

In an alternative embodiment, the present invention provides a heterodimeric immunoglobulin, or fragment thereof, wherein the first and second polypeptides comprise an engineered immunoglobulin constant region with a modified CH3 domain containing a protein-protein contact surface, wherein the protein-protein contact surface of the first polypeptide contains an amino acid substitution in position 20, and at a position selected from the group consisting of: 3, 5, 7, 22, 26, 27, 79, 81, 84, 84.2, 85.1, 86, 88, and 90 and wherein the protein-protein interface of the second polypeptide contains an amino acid substitution at position 26 and at a position selected from the group consisting of: 3, 22, 27, 79, 81, 84, 85.1, 86, and 88, while the amino acid residue substituted at position 20 in the first constructed constant region of an immunoglobulin interacts with an amino acid residue substituted at position 26 in a second engineered immunoglobulin constant region, where the amino acid position each representative of the group is indicated according to the IMGT® numbering.

Preferably, the amino acid residues that are substituted at the protein-protein interface of the first engineered immunoglobulin chain include amino acid residues at positions 20 and 22, and optionally an additional amino acid residue at a position selected from the group consisting of 3, 5, 7, 26, 27, 79, 81, 84, 84.2, 84.4, 85.1, 86, 88 and 90, wherein the amino acid residues that are substituted at the protein-protein interface of the second engineered immunoglobulin chain include amino acid residues at positions 26 and at an additional position selected from a group consisting of 3, 5, 7, 20, 22, 27, 79, 81, 84, 84.2, 84.4, 85.1, 86, 88 and 90, where the amino acid position of each member of the group is indicated according to IMGT® numbering. Preferably, the amino acid residues that are substituted at the protein-protein interface of the first engineered immunoglobulin chain include amino acid residues at positions 20 and 22 and optionally an additional amino acid residue at a position selected from the group consisting of 3, 5, 7, 26, 27, 79 , 81, 84, 84.2, 84.4, 85.1, 86, 88 and 90, and wherein the amino acid residues that are substituted at the protein-protein interface of the second engineered immunoglobulin chain include amino acid residues at positions 26 and 86 and optionally at an additional position, selected from the group consisting of 3, 5, 7, 20, 22, 27, 79, 81, 84, 84.2, 84.4, 85.1, 88 and 90, where the amino acid position of each member of the group is indicated according to IMGT® numbering.

More preferably, the amino acid residue which is substituted at position 20 at the protein-protein interface of the first engineered immunoglobulin constant region is selected from the group consisting of 20V, 20T, 20A, 20N, 20Q, 20K, 20S, 20W and 20E, and the amino acid residue substituted at the protein-protein interface of the first engineered immunoglobulin constant region is selected from the group consisting of 3A, 22A, 22G, 22L, 22I, 22V, 22T, 26K, 26R, 26Q, 26T, 26V, 26S, 26N, 26E, 79Y, 85.1W, 85.1F, 85.1T, 85.1M, 85.1A, 85.1S, 85.1R, 85.1H, 85.1K, 85.1C, 85.1N, 86W, 86Y, 86S, 86I, 86H, 86Q, 86V . selected from the group consisting of 26T and 26E and conservative substitutions of amino acids at this position, while the position of the amino acid is indicated according to See IMGT® numbering.

In the most preferred embodiment, the amino acid residue that is substituted at the protein-protein interface of the first immunoglobulin engineered constant region at position 20 is 20K, and the additional amino acid residue that is substituted at the protein-protein interface of the first engineered immunoglobulin constant region is is 3A, 22V, 26T, 79Y, 85.1S, 86V, 88W, and 90N, and wherein the amino acid residue that is substituted at the protein-protein interface of the second engineered immunoglobulin constant region at position 26 is 26T, and the additional amino acid residue substituted at the protein-protein interface of the second designed con

- 19 039658 of the immunoglobulin stant region is 3E, 5A, 7F, 20T, 22V, 81D, 84L, 84.2E, 84.4Q, 85.1C/S/A, 86S, 88R and 90R (IMGT® numbering).

Examples

Example 1 Materials and Methods

Construction of Expression Vectors for Transient Expression in Mammalian Cells The cDNAs encoding the various polypeptide chains indicated were first partially or completely synthesized at GENEART AG (Regensburg, Germany) and modified using standard molecular biology techniques. The PCR products were digested with the appropriate DNA restriction enzymes, purified, and ligated into a modified pcDNA3.1 plasmid (Invitrogen AG, Zug, Switzerland) carrying the CMV promoter and a polyadenylation site (poly(A)) from bovine hormone, previously digested with the same DNA restriction enzymes. All polypeptide chains were independently ligated into this expression vector, with the mouse VJ2C leader peptide leading to secretion.

Expression of recombinant proteins

Antibodies, scFv-Fc fusion proteins, antibodies and BEAT antigens were expressed as described below unless otherwise noted. For transient expression, equal amounts of each chain engineered vector were co-transfected with suspension-adapted HEK293-EBNA cells (ATCC-LGL Standards, Teddington, UK; catalog number: CRL-10852) using polyethyleneimine (PEI; Sigma, Bux, Switzerland). ). Typically, 100 ml of cells in suspension at a density of 0.8-1.2 million cells per ml were transfected with a DNA-PEL mixture. over a period of 4 to 5 days to allow their secretion into the culture medium (EX-CELL 293, serum-free medium for HEK293 (Sigma) supplemented with 0.1% pluronic acid, 4 mM glutamine and 0.25 μg/ml geneticin ). Cell-free culture supernatants containing secreted immunoglobulins were obtained by centrifugation followed by sterile filtration and used for further analysis.

Protein A differential affinity chromatography

After cell-free supernatants were obtained, they were loaded onto a 1 ml HiTrap™ MabSelect SuRe™ protein A column pre-equilibrated with 0.2 M citrate-phosphate buffer, pH 6.0, and purified on an AKTApurifier™ chromatography system (both purchased from GE Healthcare Europe GmbH; column number in the catalog: 11-0034-93) at a flow rate of 1 ml/min. The running buffer was 0.2 M citrate-phosphate buffer, pH 6. Elution of the heterodimer of interest was performed using 20 mM sodium citrate buffer, pH 4, while homodimeric species were eluted with 0.1 M glycine, pH 3.0.

After elution, the optical density (OD) was determined at 280 nm; fractions containing the heterodimer of interest were pooled and neutralized with 0.1 volume 1 M Tris, pH 8.0 (Sigma).

The supernatant, flow-through and eluted fractions were analyzed under non-reducing conditions using PAAG/SDS (Bis-Tris 4-12% NuPAGE acrylamide gels, Invitrogen AG, Basel, Switzerland).

Differential Scanning Calorimetry (DSC)

The thermal stability of the antibodies was compared using calorimetric measurements. Calorimetric measurements were carried out on a VP-DSC differential scanning microcalorimeter (MicroCal-GE Healthcare Europe GmbH, Glattbrugg, Switzerland). The cell volume was 0.128 ml, the heating rate was 1°C/min, and the overpressure was maintained at 64 psi. All protein fragments were used at a concentration of 1-0.5 mg/ml PBS (pH 7.4). The molar heat capacity of each protein was evaluated by comparison with duplicate samples containing identical buffer from which the protein had been removed.

Partial molar heat capacity and melting curves were analyzed using standard procedures. Thermograms were corrected for background and normalized for concentration prior to further analysis using the non-two-state model in Origin software version 7.0.

Expected melting profiles for human IgG subclasses are known (Garber E & Demarest SJ (2007) Biochem Biophys Res Commun, 355(3): 751-7) and all profiles have been shown to involve three conformational unfolding transitions corresponding to independent domain unfolding. CH2, CH3 and FAB. Of the four human IgG subclasses, IGHG1 contains the most stable CH3 domain (~85°C); while the CH3 domains of the other subclasses are less stable, although none are known to melt below 70°C. Similarly, all subclasses are known to have a melting temperature of the CH2 domain of ~70°C.

Purity Evaluation by Capillary Gel Electrophoresis

Sample preparation under non-reducing conditions 40 µg of a desalted protein sample was buffered in SDS sample buffer (Beckman Coulter International S.A., Nyon, Switzerland; IgG Purity kit,

- 20 039658 catalog number: A10663) containing 5 mM iodoacetamide (Sigma). A 10 kDa internal standard was added to the samples. Mixtures with samples were heated at 70°С for 10 min.

Capillary gel electrophoresis

After preparation, the samples were run on a ProteomeLab PA 800 system (Beckman Coulter International S.A., Nyon, Switzerland) equipped with a photodiode array detector (DAD) set at 220 nm. Uncoated fused silicon capillaries with an internal diameter of 50 μm x 30.2 cm (effective length to the detector 20.2 cm) were used as separation media. The injection and separation of the samples was carried out at a constant voltage of 5 and 15 kV, respectively, with reverse polarity in SDS-(molecular weight) gel buffer. The results were recorded at a frequency of 2 Hz and the current was stable during separation. The capillaries and samples were thermostated at 25°C.

Affinity measurement with SPR

SPR analysis was used to measure association and dissociation rate constants for the binding kinetics of various antibodies (mouse and humanized antibodies). Antibody binding kinetics were measured on a BIAcore 2000 device (BIAcore-GE Healthcare Europe GmbH, Glattbrugg, Switzerland) at room temperature and analyzed using BiaEvaluation software (version 4.1, BIAcore-GE Healthcare Europe GmbH).

Measurements were performed on CM5 sensor chips (GE Healthcare Europe GmbH, catalog number: BR-1000-14) separately coupled to the ligand of interest using a commercially available amine coupling kit (GE Healthcare Europe GmbH, catalog number: BR-1000- fifty). Protein G ligand was obtained from Pierce (Thermo Fisher Scientific-Perbio Science S.A., Lausanne, Switzerland, catalog number: 21193).

The results (sensogram: fc2-fc1) were fitted with a 1:1 Langmuir model with or without mass transfer as indicated. In capture experiments, to account for experimental variations at the beginning of each measurement, the value of R _max was set to local for all approximations. The dissociation time was at least 350 s. Measurements were performed in triplicate and samples with zero concentrations were included for comparison. Both the chi-square (Chi2) and the residuals were used to assess the quality of the fit between the experimental results and the individual binding models.

Affinity measurement on HPB-ALL cells by FACS

HPB-ALL cells (DSMZ, Braunschweig, Germany, catalog number: ACC483) were used as a CD3 positive cell line for FACS staining. HPB-ALL was maintained in RPMI 1640 medium supplemented with 10% FCS and 100 U/ml penicillin and 100 μg/ml streptomycin. A 100 µl dilution series of chimeric OKT3 antibody and humanized variants were incubated with 4x105 HPBALL cells in PBS supplemented with 1% BSA and 0.1% sodium azide (called FACS buffer) for 45 min on ice. Foreign human IgG1 was used as an isotype control and OKT3 chimeric antibody was used as a positive control. After washing, cells were incubated with a 1/200 dilution of PE-labeled anti-human Fc antibody (EBioscience, Vienna, Austria) for 45 min on ice. Cells were then washed again and resuspended in 200 µl of FACS buffer. The relative mean fluorescence in each sample was measured on a FACSCalibur (BD Biosciences, Allipville, Switzerland). The results are summarized in FIG. 9 as relative staining of HBP-ALL versus OKT3 chimeric antibody.

Cell lines for in vitro assays Human cell lines positive for CD38

Human cells expressing the CD38 antigen have been described in PCT Publication Nos WO 2005103083, WO 2008047242, WO 2011154453 and WO 2012092612. The human CD38 positive cell lines used in this text were as follows: Stable recombinant CHO[CD38] cells

The gene encoding human CD38 was ordered from Source Biosciences (Berlin, Germany, catalog number: IRAU37D11, 4309086). Human CD38 was amplified using primers adding a Kozak sequence, an initiation codon followed by a signal peptide (mouse V leader) to the 5' end and a NheI restriction site to the 3' end. The amplicon was digested with NheI and HindIII and cloned into the pT1 expression cassette, a proprietary vector derived from pcDNA3.1 (Invitrogen AG). The pT1 expression cassette links expression of the gene of interest to GFP and PAC (puromycin resistance gene) expression using two internal ribosome entry sites (IRES) on polycistronic mRNA. A plasmid was generated using the Midiprep kit and the cloned CD38 open reading frame was confirmed by DNA sequencing. Suspension CHO-S cells (Invitrogen AG) were transfected using polyethyleneimine (JetPEI®, Polyplus-transfection, Illkirsch, France) in a 50 ml bioreactor format (TubeSpin 50 bioreactors, TPP, Trazadingen, Switzerland). For this purpose, cells in the exponential growth phase were seeded in OptiMEM medium (Invitrogen AG, catalog number: 31985-047). TetPET® DNA complex was added to the cells After 5 h incubation of cells with JetPEI®: DNA complex at 37°C with shaking (200 rpm) for endocytosis into cell suspension

- 21 039658 added one volume of PowerCHO2 culture medium (Lonza, distributor of RUWAG Lifescience, Betlach, Switzerland, catalog number: BE12-771Q) supplemented with 4 mM Gln. The cells were then incubated on a shaking platform at 37°C, 5% CO ₂ and 80% humidity. One day after transfection, cells were seeded in 96-well plates at various concentrations in a selective medium containing puromycin (Sigma, catalog number: P8833-25mg). After approximately 14 days of static selection, 46 populations of high GFP-expressing cells were expanded as suspension cultures using TubeSpin 50 bioreactors. After the cells had successfully adapted to the suspension, CD38 expression in the cells was analyzed by FACS. Stable CHO[CD38] clones with a homogeneous staining profile for CD38 were selected and used further.

Other lines of CD38 positive cells included:

NCI-H929 (ATCC-LGL Standards; catalog number: CRL-9068).

Namalwa (ATCC-LGL Standards; catalog number: CRL-1432)

U266 (ATCC-LGL Standards; catalog number: TIB-196)

RPMI 8226 (ATCC-LGL Standards; catalog number: CCL-155)

Culture conditions: RPMI 1640 medium supplemented with 10% heat-inactivated fetal bovine serum (FBS), 1% penicillin-streptomycin (Invitrogen AG) and 1% GlutaMAX-1 (Invitrogen AG).

Raji (ATCC-LGL Standards; catalog number: CCL-86)

Daudi (ATCC-LGL Standards; catalog number: CCL-213)

Recombinant target antigens Human CD3-gamma-epsilon-Fc fusion protein

First, a cDNA encoding the extracellular region of human CD3-gammaα was synthesized at GENEART AG (Regensburg, Germany) (UniProt accession number: P09693, residues 23-103 (SEQ ID NO: 184); UniProt Consortium (2013) Nucleic Acids Res., 41 (database edition): D43-7; http://www.uniprot.org/) coupled to the extracellular region of human POP-epsilon (UniProt accession number: P07766, residues 22-118 (SEQ ID NO: 185)) using a 26 amino acid long peptide linker (after GSADDAKKDAAKKDDAKKDDAKKDGS; SEQ ID NO: 169).

ratio ^{5 J}

This synthetic gene was coupled to the human IgG1 Fc region using standard PCR techniques with overlapping primers and a human IgG1 Fc cDNA template also obtained from Geneart AG. The resulting cDNA was cloned into the modified plasmid pcDNA3.1 mentioned above.

For transient expression of the CD3-gamma-epsilon-Fc protein (SEQ ID NO: 170), suspension-adapted HEK-EBNA (ATCCCRL-10852) cells were transfected with the recombinant vector using polyethyleneimine (PEI) as described above. The CD3-gamma-epsilon-Fc construct was then purified from the cell-free supernatant using Streamline Recombinant Protein A media (GE Healthcare Europe GmbH, Glattbrugg, Switzerland) and used for further analysis.

Human and cynomolgus monkey_CD3-epsilon-1-26 fusion proteins cDNA encoding human COP-epsilon peptide 1-26 (UniProt accession number: P07766, amino acids 23-48, SEQ ID NO: 171) and cDNA encoding peptide 1 -26 Cynomolgus CD3 epsilon (UniProt accession number: Q95LI5, amino acids 22-47, SEQ ID NO: 172) was NDP amplified from synthetic cDNA obtained from GENEART A.G. for human and cynomolgus CD3-epsilon extracellular regions, respectively . The amplified products were then fused to the human IgG1 Fc region using standard overlapping primer NDP techniques. Human IgG1 Fc cDNA template was obtained from Geneart AG. The resulting cDNA was cloned into the modified plasmid pcDNA3.1 mentioned above.

For transient expression of human and cynomolgus epsilon CD3 constructs (SEQ ID NOs: 173 and 174, respectively), suspension-adapted HEK-EBNA cells (ATCC-CRL-10852) were transfected with recombinant vectors using polyethyleneimine (PEI) as described above. . The CD3 epsilon fusion constructs were then purified from the cell-free supernatant using Streamline Recombinant Protein A media (GE Healthcare Europe GmbH, Glattbrugg, Switzerland) and used for further analysis. These two fusion proteins are referred to herein as human CD3-epsilon-1-26 α_Fc fusion proteins.

Extracellular regions of human CD38 and cynomolgus monkey cDNA for human CD38 was obtained from Source Biosciences (Erwin-Negelein-Haus, Germany, catalog number: IRAU37D11, 4309086), its extracellular region (UniProt accession number: P28907, residues 43-300) was amplified using PNR and cloned into a proprietary expression vector derived from pcDNA3.1 (Invitrogen AG). This expression vector contained a Kozak sequence and an initiation codon followed by a mouse VJ2C leader peptide 5' and a 6-His tag 3' from its multiple cloning site. The soluble extracellular region of human CD38 fused to 6-His tag (SEQ ID NO: 175) was expressed and purified as follows: one volume of RPMI 1640 medium (PAA Laboratories, catalog number: E15-039) containing HEK cells, 0.1% pluronic acid (Invitrogen AG),

- 22 039658 Expression vector and polyethyleneimine (JetPEI®, Polyplus-transfection, Illkirche, France) were incubated in a shake flask at 37° C., 5% CO 2 and 80% humidity. After 4 h, one volume of ExCell293 supplemented with 6 mM glutamine was added to the mixture, and further incubation continued for a total of 5 days. After production, the cell-free supernatant was obtained by centrifugation and filtered using filters with a pore diameter of 0.2 μm, the pH was adjusted to 7.4 (4°C) using 1 M Tris, pH 8.7. Ni-sepharose Excell beads (GE Healthcare, catalog number: 17-3712-03) were added to the solution and incubated overnight at 4° C. with stirring. The solution was loaded onto an Econo-Column (BioRad Laboratories AG, Reinach, Switzerland, catalog number: 737-4252) for gravity flow purification. The beads were washed in PBS (2x), 20 mM imidazole, and the protein was eluted in PBS, 500 mM imidazole. The eluted fractions were pooled and the buffer was changed to PBS using two steps of dialysis at 4°C. The purified extracellular region of human CD38 was filtered using syringe filters with a pore diameter of 0.22 μm.

Using the methods described above, a soluble extracellular region of cynomolgus monkey CD38 antigen fused to a 6-His tag (SEQ ID NO: 176) was cloned, expressed and purified.

In vitro T cell killing redirection assay Production of peripheral blood mononuclear cells

For the production of peripheral blood mononuclear cells (PBMCs), blood filters containing human leukocytes were obtained from the Blood Collection Center in La Chaux-de-Fonds, Switzerland (Centre de Transfusion Sanguine et Laboratoire de Serologie, rue Sophie-Mairet 29 , CH-2300). Cells were removed from the filters by backwashing with 60 ml of PBS containing 10 units. liquamine/ml (Drossapharm AG, Lucerne, Switzerland). PBMCs were then purified using 50 ml Blood-Sep-Filter tubes (Brunschwig, Basel, Switzerland) following the manufacturer's instructions. The tubes were centrifuged for 20 min at 800g at room temperature (without using a brake) and cells were collected from the boundary layer. Cells were washed 3x with Roswell Park Memorial Institute medium (RPMI, PAA Laboratories, Pasching, Austria) without addition of PBS or phosphate buffered saline (PBS). PBMCs were resuspended to a concentration of 106 cells/ml in RDL medium (RPMI supplemented with 10% heat-inactivated fetal bovine serum (FBS) and penicillin/streptomycin) and cultured overnight at 37° C. in an incubator at 5% CO2 prior to analysis.

T cell preparations

Purification of T cells was performed immediately after PBMC isolation using the Pan-T-cell isolation kit II (Myltenyi Biotec GmbH, Bergisch Gladbach, Germany, catalog number: 130-091-156), following the instructions manufacturer. After purification, T cells were resuspended to a concentration of 106 cells/ml in RDL medium and cultured overnight at 37° C. in a 5% CO ₂ incubator prior to analysis.

Reading analysis results

To quantify kill retargeting, two different data readouts were used, which gave highly comparable results.

The first was the flow cytometry method, referred to herein as the RDLFACS method, based on the fluorescence cytometry described by Schlereth B et al. ((2005) Cancer Res, 65: 2882-2889), Moore PA et al. ((2011) Blood , 117(17): 4542-51) and Friedrich M et al. ((2012) Mol Cancer Ther, 11: 2664-2673). Target cells were collected, counted, washed once and resuspended to a concentration of 5x106 cells/ml in PBS + 1 μM carboxyfluorescein succinimidyl ester (CFSE, Sigma). Cells were incubated for 15 min at 37°C with gentle shaking every 5 min. CFSE-loaded cells were washed 3x with RDL medium and resuspended to a concentration of 2x105 cells/ml in RDL medium. PBMCs were collected, counted and resuspended to a concentration of 2x106 cells/ml in RDL medium. Serial dilutions (3x solutions) of antibodies were prepared in RDL medium. Target cells (50 μl/well), T cells (50 μl/well) and 3x antibody solutions (50 μl/well) were dispensed into a 96 well flat bottom plate (TPP, Trazadingen, Switzerland). The effector:target ratio was 10:1. These plates were incubated for 48 hours in an incubator at 5% CO2 at 37°C. After incubation, the plates were centrifuged for 3 min at 300 g, the supernatants were discarded by shaking off the plates. The plates were washed once with 200 μl of PBS, centrifuged again and discarded with PBS. A pre-warmed Accutase solution (Invitrogen AG) was added and the plates were incubated for 10 min at 37°C. Detached adherent cells were resuspended by up/down pipetting after addition of 100 µl of RDL medium. The resulting solution was transferred to a 96-well U-bottom plate (TPP). U-bottom plates were centrifuged for 3 min at 300 g, supernatants were discarded and cells were resuspended in 200 µl of cold FACS buffer (PBS + 2% FBS + 10% versene) supplemented with 7-AAD (Becton Dickinson AG, Alypville, Switzerland) when diluted 1/40. These plates were read immediately on a Guava easyCyte™ flow cytometer (Millipore AG, Zug, Switzerland). For each well, the absolute number of live target cells was determined by setting a window to select for CFSE positive and 7ADD negative populations using Flowjo® software (Miltenyi Biotec GmbH, Behr

- 23 039658 Gisch-Gladbach, Germany). For each sample, the percentage of specific cytotoxicity was determined using as a background level the condition under which only target cells were incubated. EC ₅₀ values were determined using non-linear regression with variable slope using Prism software (GraphPad software, La Jolla, CA, USA). The percent specific redirected lysis (RDL) was calculated by subtracting the percent specific cytotoxicity for the no antibody condition from that for the conditions where test antibody was added.

The second was the cell viability method, referred to herein as the RDL-MTS method, based on the colorimetric evaluation of cell viability described in Biihler P et al. ((2008) Cancer Immunol Immunother, 57: 43-52, Labrijn AF et al. ((2013) Proc Natl Acad Sci USA, 110(13): 5145-50) and PCT Publication No. WO 2012143524. Target cells were collected, counted, washed once and resuspended to a concentration of 2x10 ⁵ cells/ml in RDL medium. collected, counted and resuspended to a concentration of 2x10 ⁶ cells/ml in RDL medium Serial dilutions of antibodies (3x solutions) were prepared in RDL medium Target cells (50 μl/well), T cells (50 μl/well) and 3x solutions antibodies (50 μl/well) were dispensed into a 96-well flat bottom plate (TPP) The effector:target ratio was 10:1 These plates were incubated for 48 hours in an incubator at 5% CO ₂ at 37° C. After incubation supernatants were discarded and the plates were washed three times with 200 µl of PBS to remove b PBMC, and then 100 µl of RDL medium was added to each well. Data reading was performed using a CellTiter 96® kit (Promega AG, Dübendorf, Switzerland) following the manufacturer's instructions. Briefly, 10-20 µl of MTS reagent was added to each well and the plates were incubated for 2-6 hours in an incubator at 5% CO2 at 37°C. The absorbance at 490 nm was then read on a BioTek synergy plate reader spectrophotometer (BioTek AG, Lucerne, Switzerland). Percent specific kill was calculated using the following formula: Specific kill = 100x [(SD-Sp)/(SD-MD)]. SD is the absorbance measured under the condition of spontaneous death, in which the target cells were incubated separately. Sp is the absorbance measured under each test condition (target cells + PBMC + antibody). MD is the absorbance measured under the maximum death condition at which target cells were lysed with 3 cycles of freeze and thaw. The percent specific redirected lysis (RDL) was calculated by subtracting the percent specific cytotoxicity for the no antibody condition from that for the conditions where test antibody was added. EC50 values were determined using non-linear regression with variable slope using Prism software (GraphPad software).

Example 2 Antigen binding sites that target human CD3 antigen and CD38 antigen.

2.1 Binding sites for the human CD3 antigen.

The human CD3-epsilon subunit was chosen to trigger redirected killing by T cells via bispecific capture.

2.1 A Humanized variants of mouse OKT3 antibody.

The human epsilon CD3 antigen binding site used in this application was derived from a mouse OKT3 antibody (muromonab-CD3, trade name orthoclone OKT3, which was marketed by Janssen-Cilag and subsequently discontinued; mouse heavy chain and light chain variable domains with sequences SEQ ID NO: 38 and 41, respectively). Mouse OKT3 variable domains were humanized and formatted as scFv and Fab fragments.

Humanization was performed following the method described by Jung S and Pluckthun A (1997, Protein Eng, 10(8): 959-66) to produce a highly stable humanized variant that would be suitable for conversion to both FAB and scFv formats. This method utilizes the highly stable VH/VL domain pair found in the Herceptin® antibody (rhuMAbHER2, huMAB4D5-8, trastuzumab or the trade name Herceptin®; US Pat. No. 5821337; heavy chain and light chain variable domains with sequences of SEQ ID NO: 20 and 21, respectively), and it follows the sequence of the humanization process on fixed scaffolds (Almagro JC and Fransson J (2008), Front Biosci, 13: 1619-33). Since the Herceptin antibody is originally derived from the highly stable human VH3 and VK1 families of germline scaffolds, germline scaffolds from these two families can equally be used as a source of fixed scaffolds. Alternatively, human VK3 light chain germline scaffold families can be used instead of VK1, as they also have a good stability property (Ewert S et al. (2003) J Mol Biol, 325: 531-553). In addition to mouse antibodies, human antibodies can be constructed using the fixed scaffold method to improve stability. It is preferable to use the human germline scaffold IGHV3-23*04, IGKV1-39*01 and IGKV3-20*01 with the sequence of SEQ ID NO: 37, 39 and 40, respectively (according to IMGT® (International Information System ImMunoGeneTics (Lefranc MP et al. (1999) Nucleic Acids Res, 27(1): 209-12; Ruiz M. et al. (2000) Nucleic Acids Res, 28(1): 219-21; Lefranc MP (2001) Nucleic Acids Res, 29( 1): 207-9; Lefranc MP (2003) Nucleic Acids Res, 31(1): 30710; Lefranc MP et al. (2005) Dev Comp Immunol, 29(3): 185-203; Kaas Q et al. , (2007) Briefings in Functional

- 24 039658

Genomics & Proteomics, 6(4): 253-64; http://www.imgt.org).

For this purpose, a first humanized antibody was constructed in which the specified CDRs in the variable domains of the Herceptin® antibody were appropriately substituted for the CDRs from the mouse OKT3 antibody and compared with the mouse OKT3 antibody chimera (heavy chain and light chain variable regions with sequences of SEQ ID NO: 130 and 131, referred to in this text as the OKT3 chimeric antibody).

Prototype antibodies (heavy chain and light chain variable regions of SEQ ID NOs: 79 and 96, abbreviated as VH/VL) showed increased production levels in transient expression assays and increased FAB stability as measured by differential scanning calorimetry, but not revealed bindings to HPB-ALL cells (assessed by mean fluorescence intensity in FACS experiments, see Materials and Methods), which is a human CD3 epsilon positive tumor T cell line (Fig. 1A).

Based on the 3D model of the first prototype of a pair of variable domains, a subset of backmutations (from the Herceptin® CDR-grafted prototype to the mouse OKT3 sequence) were selected and investigated: I34M, V48I, A49G, R58N, R58Y, I69L, A71T and T73K in the heavy chain variable domain and M4L, V33M, A34N, L46R, L47W, R66G, F71Y and P96F in the light chain variable domain (Kabat numbering). It is worth noting that the R58N substitution corresponds to a mutation from the Herceptin® CDR-grafted prototype in mouse OKT3, while the R58Y substitution corresponds to a germline substitution from the Herceptin® CDR-grafted prototype to human IGHV3-23*04. The design strategy for the combination of substitutions was based on the complementarity of the various substitutions in terms of their intended effect on CDR regions and/or variable domain folding and/or immunogenicity.

In the first approach, all candidate antibodies were converted to human IgG1 antibody format. The best variants were selected according to expression levels, thermal stability of the Fab fragment, and ability to bind to HPB-ALL cells as measured by FACS. In the best humanized variants, the A protein binding site present in their VH domain has been disrupted by introducing a G65S or N82aS substitution. This design step was necessary to further generate safe T-cell retargeting BEAT antibodies free of anti-CD3 homodimers.

Backmutations in VH: I34M, A49G and A71T, along with backmutations in VL: M4L, L46R, L47W and F71Y, restored affinity. The best combinations of variable domains were VH8/VL4, VH8/VL8, VH11/VL4 and VH11/VL8, as they allowed to retain most of the binding to the original cells (Fig. 1A-C). In addition, combinations of VH8/VL8 (variable domains with the sequences of SEQ ID NO: 131 and 132, respectively) and VH11/VL8 (variable domains with the sequences of SEQ ID NO: 133 and 132, respectively) showed an increased stability of FAB (+2, 8°C and +1.6°C, respectively, Fig. 1D).

Finally, the best humanized variants were also converted to scFv-Fc fusion format and transiently expressed. Variants were prioritized in terms of their relative affinity, stability, expression levels when transiently transfected in a given format (FIG. 1E). The best combinations of variable domains in the scFv-Fc fusion format were similar to those found in the antibody format: VH8-VL4 (scFv fragment of SEQ ID NO: 135) and VH8-VL8 (scFv fragment of SEQ ID NO: 136). Both scFv fragments had good thermal stability in the scFv-Fc fusion format (FIG. 1F).

2.1 B Humanized variants of mouse SP34 antibody.

The mouse SP34 antibody was first described in 1985 (Pessano S et al. (1985) EMBO J, 4(2):337-44). Produced by a hybridoma derived from mice immunized with denatured protein extracts from HPB-ALL cells, the antibody was specific for human antigens and was able to cross-react with cynomolgus monkey antigens.

Following the methods and process sequence described in this Example above, humanized VH and VL domains were constructed for mouse SP34 antibody containing a VH domain of SEQ ID NO: 1 and a VL domain of SEQ ID NO: 2 by grafting CDRs onto germline scaffolds. VH3-23 and VK3 respectively. The resulting VH3-based variable domains can then be rendered incapable of binding to protein A using G65S or N82aS substitutions (Kabat numbering), depending on their use in the BEAT antibody format.

For this purpose, a first humanized antibody was constructed in which the hypervariable regions in the variable domains of a human antibody containing a heavy chain VH3 germline domain and a light chain VK3 germline domain were respectively replaced with hypervariable regions from mouse SP34 antibody. The resulting humanized antibody was used as a starting point for further affinity improvement and compared to the SP34 antibody chimera (having the heavy chain and light chain sequences of SEQ ID NOs: 137 and 138, respectively, referred to herein as the SP34 chimeric antibody).

Prototype antibody (heavy chain and light chain variable regions with sequences

- 25 039658

SEQ ID NOS: 91 and 105, and abbreviated as VH1/VL1) binds with lower affinity to the CD3-epsilon 1-26 human_Fc fusion protein (assessed by NIR, see Materials and Methods and Figure 2A).

Based on the 3D model of the first pair of prototype variable domains, a subset of substitutions was selected that corresponded to any of the backmutations between the human CDR-grafted germline VH3/VK3 prototype and the mouse SP34 sequence (human-to-mouse or mouse-to-human substitutions), or rationally designed amino acid substitutions. . The following substitutions were introduced and tested in various combinations: W100eF, and W100eY in the heavy chain variable domain, and A2I, S25A, T27A, G27aA, V27cA, T28A, T29A, S30A, N31A, Y32A, E38Q, F44P, G46L, T51A, N52A , K53A, R54A, P56A, L66G, D69T, F87Y, Q89A, W91F, Y92A, S93A, N94A and Q100G in the light chain variable domain (Kabat numbering; see Fig. 2A). The design strategy for the combination of substitutions was based on the complementarity of the various substitutions in terms of their intended effect on CDR regions and/or variable domain folding and/or immunogenicity and/or effects on transient expression in mammalian cells.

In the first approach, all candidate antibodies were converted to human IgG1 antibody format, and later further examined in scFv-Fc fusion protein format (Figure 2B), with some variants having a G65S disrupted protein A binding site present in the VH domain. The best humanized candidate antibodies were selected based on expression levels and ability to bind to human and cynomolgus Fc CD3-epsilon 1-26 fusion proteins by SPR.

Surprisingly, when the format was changed to scFv, there was a dramatic decrease in the expression of the SP34 and H1L21 chimera compared to the IgG1 format (FIG. 3). However, scFv-Fc expression was upregulated after the combination of W100eY substitutions in VH and W91F B VL (FIG. 2B).

In order to further improve the feasibility of producing CD3-based bispecific antibodies, a humanized SP34 scFv was further constructed. In order to identify important positions influencing SP34 scFv-Fc expression, alanine scanning of the VL21 CDR was performed. The following replacements were introduced: T27A, G27aA, V27cA, T28A, T29A, S30A, N31A, Y32A, N52A, K53A, R54A, P56A, L90A, Y92A, S93A, N94A and L95A. Interestingly, only the substitutions at positions 29, 30 (VH1/VL26 SEQ ID NO: 139) and 95 (VH1/VL34 SEQ ID NO: 140) led to a significant increase in scFv-Fc expression (Fig. 4), moreover, these mutants were the only ones that retained binding to human and cynomolgus monkey CD3 epsilon.

Based on these results, additional substitutions at positions 29, 30 and 95 in the light chain were investigated.

Since there is hypervariability at positions 29 and 30 across all the different families of light chain variable regions, these two positions were randomly mutated by site-directed PCR. Of all the resulting mutants, only the substitutions T29A, T29E, T29S, S30A and S30D significantly improved the level of transient expression (FIGS. 5a and 5b) while retaining binding to human CD3epsilon.

In an excellent approach, the L95 residue, which is known in the art as the canonical structure residue, has been replaced with residues most commonly found at this very position in the lambda family of variable regions. The following replacements have been introduced: L95A, L95G, L95T, L95S, L95D and L95N. Surprisingly, only L95G and L95T significantly improved expression while maintaining target binding (FIG. 5c).

As a result of this work, several humanized SP34-based BEATs were developed and their expression and binding to CD3 epsilon was investigated. Among all these constructs, the bispecific antibody containing scFv with H5L65 (SEQ ID NO: 69) and H5L67 (SEQ ID NO: 70) showed a 2-fold increase in expression compared to the scFv-based H5L32 BEAT control (Fig. 6).

H5L65 and H5L67 were additionally subjected to DSC analysis as either IgG1 or scFv-Fc and compared with H5L32, H1L21 and SP34 chimera. H5L65 showed better thermal stability as IgG1, but more interestingly, an increase of 5-2°C compared to other humanized variants in scFv-Fc format. (Table 1). The in vivo stability of the improved antibodies was consequently improved, as well as in vitro.

- 26 039658

Table 1

fab IgG (Fab) scFvFc SP34 options Tm Tm Tm CO (°C) (°C) Chimera SP34 66.6 66.7 X A hSP.U III 1.21 75.1 73.1 56.7 I1SP34 115 1.32 77.5 75.9 59.7 hSP34 H 5 1.65 ^8.1 77.3 61.6 hSP34 115 1.67 77.4 76.5 54 9

Preferred combinations of heavy chain and light chain variable domains with respect to antigen binding and recombinant expression are described below VH1 (SEQ ID NO: 42) or VH2 (SEQ ID NO: 43) or VH3 (SEQ ID NO: 44) or VH5 ( SEQ ID NO: 45) paired with the light chain domains of VL21 (SEQ ID NO: 46), VL32 (SEQ ID NO: 47), VL65 (SEQ ID NO: 48) and VL67 (SEQ ID NO: 49).

2.2 Binding sites for the human CD38 antigen.

CD38 is a type II transmembrane glycoprotein that is commonly found on hematopoietic cells and in hard tissues. CD38 is also expressed in various hematological malignancies. Bispecific antibodies that will redirect T cells to kill CD38-positive cancer cells will be useful in the treatment of various malignant hematological diseases, including multiple myeloma, B-cell chronic lymphocytic leukemia, B-cell acute lymphocytic leukemia, Waldenström macroglobulinemia, primary systemic amyloidosis, mantle cell lymphoma, prolymphocytic/myelocytic leukemia, acute myeloid leukemia, chronic myeloid leukemia, follicular lymphoma, NK cell leukemia and plasma cell leukemia. Several anti-CD38 antibodies have been described as research reagents or therapy candidates (PCT Publication No. WO2006099875). Among the best studied antibodies to human CD38 are mouse hybridomas OKT-10 and HB-7 (Hoshino S et al. (1997) J Immunol, 158(2): 741-7).

In a first approach, binding sites for the human CD38 antigen can be obtained from mouse hybridomas OKT 10 (heavy chain and light chain variable regions with sequences of SEQ ID NOs: 50 and 53, respectively) or HB-7 (heavy chain and light chain variable regions with sequences of SEQ ID NO: 51 and 54, respectively) and their humanized variants, which can be further converted to FAB or scFv fragment format. Following the methods and process sequence described in Example 2.1, humanized VH and VL domains for HB-7 hybridoma can be constructed by grafting CDRs onto VH3-23 and VK1 germline scaffolds, respectively.

In a second approach, following the so-called best-match humanization method described by Almagro JC & Fransson J (Front Biosci, (2008) 13: 1619-33), the best-fit humanized VH and VL domains for HB-7 hybridoma were constructed by grafting CDRs onto human IGHV4-59*03 and IGKVl-NLl*01 germline scaffolds, respectively (according to IMGT®, supra). Humanized VH and VL variants with varying numbers of backmutations were examined in silico, and one preferred selected humanized VH and VL variant was transiently expressed as a human IgGl format and referred to herein as humanized HB-7 with the most appropriate VH domains (SEQ ID NO: 56) and VL (SEQ ID NO: 57). The following back substitutions were made for mouse residues: (VH)-S35H, I37V, I48L, V67L, V71K, T73N, F78V, Y91F, and (VL)-M4L, L48I, Y49S, T69K (Kabat numbering).

Staining with humanized most appropriate HB-7 antibody (with heavy chain sequence of SEQ ID NO: 141 and light chain sequence of SEQ ID NO: 142) of recombinant CHO[CD38] cells was analyzed by FACS (results not shown). The binding affinity for the CD38 extracellular region of the humanized most appropriate HB-7 antibody was similar to that of the chimeric HB-7 antibody (with the heavy chain sequence of SEQ ID NO: 143 and with the light chain sequence of SEQ ID NO: 144) when analyzed by SPR (KD were 3.6 and 2.5 nM, respectively, Fig. 7A (chimeric) and Fig. 7B (humanized)). Surprisingly, a significant increase (+14.6°C) in Fab fragment stability was observed for the humanized HB-7 best match compared to that of the chimeric HB-7 antibody, as judged by calorimetry profiles (76.4°C (chimeric) by compared to 91.0°C (humanized), Fig. 7F).

In a third approach, mice immunized with the extracellular domain of human CD38 and human CD38+ cells were used to generate novel anti-human CD38 hybridoma candidates. Methods for producing hybridomas are known and the methods used in this application were similar

-27039658 by the methods described in PCT Publication No. WO 2013008171. Mouse 9G7 candidate antibody had high affinity for both human and cynomolgus CD38 (heavy chain and light chain variable regions of SEQ ID NOS: 52 and 55, respectively). This mouse antibody was first humanized according to the methods described in this example above. Using a best fit approach, IGHV2-5*09 VH germline scaffold and IGKV1-33*01 VK scaffold (according to IMGT® above) were chosen as the starting points for the humanization process. After CDR inoculation, strong binding of the first prototype antibody (in human IgG1 isotype format, heavy chain sequence of SEQ ID NO: 145 and light chain sequence of SEQ ID NO: 146) to human CD38 was observed to be only three times lower than original mouse antibody), as judged by SPR (9G7 chimeric antibody with heavy chain sequence of SEQ ID NO: 147 and with light chain sequence of SEQ ID NO: 148; KD was 0.3 and 1 nM for chimeric 9G7 antibody (results not presented) and the first humanized prototype (results not shown), respectively). Affinity increased two-fold after the introduction of the F36Y reverse mutation in the light chain variable domain of the antibody (Kabat numbering) (the resulting antibody is herein referred to as the humanized best match antibody 9G7 with the heavy chain sequence of SEQ ID NO: 145 and with the light chain sequence SEQ ID NO: 74; KD was 0.5 nM to human CD38, Figure 7C). The humanized best match antibody 9G7 also showed high affinity for cynomolgus CD38 antigen (KD 3.2 nM, results not shown) and increased thermal stability of FAB (Tm FAB obtained by DSC scan) compared to chimeric 9G7 antibody. (94°C compared to 82.2°C for the humanized best match 9G7 antibody and chimeric 9G7 antibody, respectively; see Fig. 7G). The humanized 9G7 best match antibody contained a heavy chain variable domain of SEQ ID NO: 58 and a light chain variable domain of SEQ ID NO: 59.

In addition, mouse 9G7 antibody was humanized following the best scaffold approach by grafting CDRs onto VH3-23 and VK1 germline scaffolds. Humanized VH and VL variants with varying numbers of backmutations were examined in silico, and one preferred humanized VH and VL variant was transiently expressed as a human IgG1 antibody (the resulting antibody is referred to in this text as the humanized antibody with the best frame 9G7 with the heavy sequence chain of SEQ ID NO: 149 and with the light chain sequence of SEQ ID NO: 150). The following mouse back substitutions were made: (VH) A24F, V37I, V48L, S49A, F67L, R71K, N73T, L78V, and K94R and (VL) F36Y (Kabat numbering). Strong binding of this antibody to human CD38 and cynomolgus CD38 was found with affinity constants similar to those of the humanized best match 9G7 antibody (KD was 0.4 and 1 nM for human and cynomolgus CD38, respectively; Fig. 7D). The thermal stability of the FAB (Tm FAB obtained by DSC scan) was also very similar to that of the most appropriate humanized variant of 9G7 with F36Y (89.2° C., see FIG. 7H). In FIG. 7J summarizes the various humanized 9G7 antibodies described above. The best backbone humanized antibody 9G7 contained a heavy chain variable domain of SEQ ID NO: 60 and a light chain variable domain of SEQ ID NO: 61.

In a fourth approach, the antibody phage library was screened for additional anti-human CD38 scFv fragments. In this library, diversity was based on naturally occurring human V genes. This donor-derived phage display antibody library used cDNA amplified from blood lymphocytes obtained from 48 human donors, of whom 70% had an autoimmune disease (vasculitis, systemic lupus erythematosus, spondyloarthropathy, rheumatoid arthritis, and scleroderma). In constructing this library, the protocol described by Schofield et al. (2007, Genome Biol., 8(11): R254) was followed, with a total diversity of 2.53 x ^{10 10} clones. ScFv fragments recognizing human and/or cynomolgus CD38 were isolated from this donor-derived phage display library as described below. ScFv fragments were isolated by a series of repeated selection rounds on recombinantly generated human and/or cynomolgus CD38 antigens (see Materials and Methods). Methods for screening antibody phage display libraries are known (Viti F et al. (2000) Methods Enzymol, 326: 480-505). Briefly, after incubation with a library of immobilized antigen that was previously coated on a plastic immunotube (overnight in PBS at 20 µg/mL) or captured with streptavidin on the surface of the beads (using the biotin labeled form of the antigen, the antigen was captured at a concentration of 50 nM throughout selection process), bound phages were isolated, while unbound phages were washed out. Bound phages were released as described by Marks et al. (Marks JD et al. (1991) J Mol Biol, 222(3): 581-97) and the selection process was repeated three times. Over one thousand clones from the second and third rounds of panning were expressed and analyzed by enzyme-linked immunosorbent assay (ELISA) for their binding to human and cynomolgus CD38 antigens. Positive clones were subjected to DNA sequencing and supplementation.

- 28 039658 The ability of some of the unique clones to bind to cell lines expressing human CD38 was carefully analyzed. After a first round of panning on a biotin-labeled variant of human CD38 antigen immobilized on streptavidin-coated beads and a second round of panning on a biotin-labeled variant of cynomolgus monkey CD38 antigen immobilized on streptavidin-coated beads, one preferred scFv fragment (clone no. 767) containing the sequence heavy chain variable region sequence of SEQ ID NO: 62 and the light chain variable region sequence of SEQ ID NO: 63, since it was able to bind to both human and cynomolgus CD38. When converted to human IgG1 antibody format, clone 767 had a KD of approximately 300 nM to human CD38 (Figure 7E) and approximately 1.2 μM to cynomolgus CD38 (results not shown) (clone 767 of an IgG1 antibody is referred to in this text as antibody 767 human with the heavy chain sequence of SEQ ID NO: 151 and with the light chain sequence of SEQ ID NO: 152). The thermal stability of the FAB (Tm FAB obtained by DSC scanning) was 70.2° C. (FIG. 7I). The VH domain of clone 767 belongs to a subclass of VH3 domains.

Example 3 CD38/CD3 Targeted BEAT Antibodies

Anti-CD38 and anti-CD3-epsilon antibodies can be in the format of either scFv-Fc type heavy chains consisting of an scFv fragment fused to a BEAT chain, or a heavy chain consisting of a Fab fragment fused to a BEAT chain, similar to that found in nature of the antibody. Formation of a functional antigen binding site requires the association of a FAB-based heavy chain with a cognate light chain.

Substitutions L234A and L235A were introduced in the CH2 region, and residual protein A binding was disrupted using G65S or N82aS substitutions (Kabat numbering), if necessary. Examples of BEAT antibodies targeting both human CD38 antigen and human CD3 epsilon had the following format:

A first example of BEAT antibodies targeting both human CD38 antigen and human CD3 epsilon, using the most appropriate humanized HB7 VH and VL sequences, was generated in the format described below:

The BEAT anti-CD38/CD3 antibody was constructed using the combination of antigen binding sites described in Examples 2.1 and 2.2 for the anti-human CD3-epsilon and anti-human CD38 arms, respectively. The heterodimeric immunoglobulin arm directed against human CD38 consisted of the BEAT heavy chain (SEQ ID NO: 153) containing the heavy chain variable region, CH1 γ1 region, γ1 hinge region, CH2 γ1 region with substitutions L234A and L235A (EU numbering) and domain CH3 BEAT based on γ1 linked to a cognate light chain (SEQ ID NO: 72). The heterodimeric immunoglobulin arm directed against human CD38 consisted of a BEAT heavy chain (SEQ ID NO: 154) containing an scFv fragment, CH1 γ1, a γ1 hinge region, a CH2 γ3 region with substitutions L234A and L235A (EU numbering) and a BEAT CH3 domain on basis γ3. This heavy chain contained part of the Fco region of human IgG3 and therefore did not bind to protein A, but since the heavy chain used in this application contained a heavy chain variable domain derived from the VH3 framework, the VH domain was mutated to introduce the N82aS substitution, thus thereby removing any additional protein A binding sites from the heavy chain. This bispecific antibody is referred to herein as the BEAT CD38-best bet HB7/CD3 antibody (FIG. 8, Format A).

The BEAT antibody CD38-best matched HB7/CD3 was transiently expressed, purified, and tested for its in vitro affinity for CD38 and CD3-epsilon antigens, its stability, and its ability to redirect T-cell killing. The KD value was 3.2 nM for the human CD38 antigen (measured by PNR; FIG. 9A). The DSC profiles for this bispecific antibody showed good thermal stability with a Tm of approximately 68°C for the scFv portion. The Tm of the FAB portion was approximately 91° C. (FIG. 9B).

Cell lines expressing CD38 (see Materials and Methods) were used in assays to assess retargeting of killing by T cells. In FIG. 10 shows redirected T-cell killing of RPMI 8226 myeloma cells using the BEAT CD38 antibody-most appropriate HB7/CD3. It is worth noting that purified T cells were used as effector cells in this assay, with a ratio of effector cells to target cells of 10 to 1. As measured by the RDL-FACS EC ₅₀ method of BEAT CD38 antibody, the most appropriate HB7/CD3 was 2.2 pM (mean of 2 donors, incubation for 48 hours).

A second example of BEAT antibodies targeting both human CD38 antigen and human CD3 epsilon using the VH and VL sequences of human clone 767 was generated in the format described below: Anti-CD38/CD3 BEAT antibody was constructed using a combination of sites the antigen binding described in Examples 2.1 and 2.2 for the arms to human CD3 epsilon and to human CD38, respectively. The heterodimeric immunoglobulin arm directed against human CD38 consisted of the BEAT heavy chain (SEQ ID NO: 65) containing the heavy chain variable region, CH1 γ1 region, γ1 hinge region, CH2 γ1 region with substitutions L234A and L235A (EU numbering) and a BEAT CH3 domain based on γ1 linked to a cognate light chain (SEQ ID NO: 138). The heterodimeric immunoglobulin arm directed against human CD38 consisted of the heavy chain

- 29 039658

BEAT (SEQ ID NO: 155) containing an scFv fragment, CH1 γ1, a γ1 hinge region, a CH2 γ3 region with substitutions L234A and L235A (EU numbering), and a γ3-based CH3 BEAT domain. This heavy chain contained part of the human IgG3 Fc region and therefore did not bind to protein A, but since the heavy chain used in this application contained a heavy chain variable domain derived from the VH3 framework, the VH domain was mutated to introduce the G65S substitution , thereby removing any additional protein A binding sites from the heavy chain. This bispecific antibody is referred to herein as the BEAT CD38-767/CD3 antibody (FIG. 8, format B).

The BEAT CD38-767/CD3 antibody was transiently expressed, purified, and tested for its in vitro affinity for CD38 and CD3-epsilon antigens, its stability, and its ability to redirect T-cell killing. Cell lines expressing CD38 (see Materials and Methods) were used in assays to assess retargeting of killing by T cells similar to those described in Example 3.2.1. In FIG. 11 shows redirected T cell killing of Daudi cells using the BEAT CD38-767/CD3 antibody. It is worth noting that in this assay, human PBMC were used as effector cells at a ratio of effector cells to target cells of 10:1. As measured by the RDL-FACS method, the EC _{50 of} the BEAT CD38767/CD3 antibody was 244 pM (mean of 3 donors, incubation for 24 hours).

Another example of BEAT antibodies targeting both human CD38 antigen and human CD3 epsilon, using VH and VL sequences with the best humanized 9G7 scaffold, was generated in the format described below: BEAT CD38/CD3 was constructed using a combination of binding sites antigen described in examples 2.1 and 2.2 for the binding sites of human CD3epsilon and human CD38 antigens, respectively.

The heterodimeric immunoglobulin arm directed against human CD38 consisted of the BEAT heavy chain (SEQ ID NO: 75 or 155) containing the heavy chain variable region, CH1 γ1 region, γ1 hinge region, CH2 γ3 region, with substitutions L234A and L235A (EU numbering). ) and a γ3-based BEAT CH3 domain linked to a cognate light chain (SEQ ID NO: 77). This heavy chain contained part of the human IgG3 Fc region and therefore did not bind to protein A, but since the heavy chain used in this application contained a heavy chain variable domain derived from the VH3 framework, the VH domain was mutated to introduce the G65S substitution , thereby removing any additional protein A binding sites from the heavy chain.

The heterodimeric immunoglobulin arm directed against human CD38 consisted of a BEAT heavy chain (SEQ ID NO: 68) containing an scFv fragment, CH1 γ1, a γ1 hinge region, a CH2 γ1 region with substitutions L234A and L235A (EU numbering) and a BEAT CH3 domain on base γ1. This bispecific antibody is referred to herein as the BEAT CD38-best KapKac9G7/CD3(SP34-kappa2) antibody.

Another example of BEAT antibodies targeting both human CD38 antigen and human CD3 epsilon using the VH and VL sequences of human clone 767 was generated in the format described below: Anti-CD38/CD3 BEAT was constructed using a combination of binding sites the antigen described in examples 2.1 and 2.2 for the binding sites of human CD3-epsilon and human CD38 antigens, respectively.

The heterodimeric immunoglobulin arm directed against human CD38 consisted of the BEAT heavy chain (SEQ ID NO: 78) containing the heavy chain variable region, CH1 γ1 region, γ1 hinge region, CH2 γ3 region with substitutions L234A and L235A (EU numbering) and domain CH3 BEAT based on γ3 linked to a cognate light chain (SEQ ID NO: 66). This heavy chain contained part of the human IgG3 Fc region and therefore did not bind to protein A, but since the heavy chain used in this application contained a heavy chain variable domain derived from the VH3 framework, the VH domain was mutated to introduce the G65S substitution , thereby removing any additional protein A binding sites from the heavy chain.

The heterodimeric immunoglobulin arm directed against human CD38 consisted of a BEAT heavy chain (SEQ ID NO: 68) containing an scFv fragment, CH1 γ1, a γ1 hinge region, a CH2 γ1 region with substitutions L234A and L235A (EU numbering), and a BEAT CH3 domain. based on γ1. This bispecific antibody is referred to herein as the BEAT CD38-767/CD3(SP34-kappa2) antibody.

In the first set of experiments, T cell activation was detected by BEAT 9G7/SP34 when in combination with PBMC, but not isolated T cells.

Example 4 Anti-CD3/CD38 BEAT antibodies containing only one VH3 domain.

An example of a BEAT antibody targeting both human CD38 antigen and human CD3 epsilon using the most appropriate humanized HB7 VH and VL sequences was generated in the format described below: Anti-CD38/CD3 BEAT was constructed using a combination of antigen binding sites described in Examples 2.1 and 2.2 for arms against human CD3-epsilon and against human CD38, respectively. The heterodimeric immunoglobulin arm directed against human CD38 consisted of the BEAT heavy chain (SEQ ID NO: 71) containing the heavy chain variable domain, CH1 γ1, γ1 hinge region, CH2 γ3 region, with substitutions L234A and L235A (numbered 30 039658 EU) and a BEAT CH3 domain based on γ3 linked to a cognate light chain (SEQ ID NO: 72). This heavy chain did not bind to protein A because it contained part of the human IgG3 Fc region and its heavy chain variable domain was from a subclass of non-VH3 domains. The heterodimeric immunoglobulin arm directed against human CD38 consisted of a BEAT heavy chain (SEQ ID NO: 157) containing an scFv fragment, CH1 γ1, a γ1 hinge region, a CH2 γ1 region with substitutions L234A and L235A (EU numbering), and a BEAT CH3 domain on base γ1. This combination of heavy and light chains included a humanized version of the anti-human CD3 epsilon (SP34) antibody described in PCT Publication No. WO 2008119565. This BEAT antibody format is referred to herein as the BEAT CD38-best-matched HB7/CD3(SP34) antibody (Fig. 12, format A).

The ability of the BEAT CD38-most appropriate HB7/CD3(SP34) antibody to redirect T cells to kill CD38+ cells was investigated in vitro. The Daudi CD38+ B lymphoblast cell line was used in the kill assay. In FIG. 13 shows BEAT CB38 redirected most appropriate HB7/CD3(SP34) T-cell killing of Daudi cells. In these assays, human PBMC were used as effector cells at a ratio of effector cells to target cells of 10 to 1, and the RDL-FACS reading method after a 24 hour incubation period (see Materials and Methods). The results showed that the BEAT CD38-best match HB7/CD3(SP34) antibody was highly efficient in redirecting T cell line CD38+ killing of Daudi cells with an EC ₅₀ of 1.8 pM (average of 3 donors).

A second example of BEAT antibodies targeting both human CD38 antigen and human CD3 epsilon, using the most appropriate humanized 9G7 VH and VL sequences (SEQ ID NOS: 58 and 59, respectively), was prepared in the format described below: The anti-CD38/CD3 BEAT was constructed using the combination of antigen binding sites described in Examples 2.1 and 2.2 for the anti-human CD3-epsilon and anti-human CD38 arms, respectively. The heterodimeric immunoglobulin arm directed against human CD38 consisted of the BEAT heavy chain (SEQ ID NO: 73) containing the heavy chain variable domain, CH1 γ1, the γ1 hinge region, the CH2 γ3 region with substitutions L234A and L235A (EU numbering) and the CH3 domain BEAT based on γ3 linked to a cognate light chain (SEQ ID NO: 74). This heavy chain did not bind to protein A because it contained part of the human IgG3 Fc region and its heavy chain variable domain was from a subclass of non-VH3 domains. The heterodimeric immunoglobulin arm directed against human CD38 consisted of a BEAT heavy chain (SEQ ID NO: 158) containing an scFv fragment, CH1 γ1, a γ1 hinge region, a CH2 γ1 region with substitutions L234A and L235A (EU numbering) and a BEAT CH3 domain on base γ1. This arm of the bispecific antibody contained the variable domains of the humanized SP34 VH5/VL32 antibody described in Example 2.1. This format of the BEAT antibody is referred to herein as the BEAT antibody CD38-best bet 9G7/CD3(SP34-kappa2) (FIG. 12, format B). The KD value for the CD38-most appropriate 9G7/CD3(SP34-kappa2) antibody was 18 nM relative to the human CD3 l-26_Fc fusion protein (FIG. 14).

The ability of the BEAT CD38-most appropriate 9G7/CD3(SP34-kappa2) antibody to redirect T cells to kill CD38+ cells was investigated in vitro. The Daudi CD38+ B lymphoblast cell line was used in the kill assay. In FIG. 15 shows BEAT antibody redirected CD38 most appropriate 9G7/CD3(SP34-kappa2) T-cell killing of Daudi cells. In these assays, human PBMC were used as effector cells at a ratio of effector cells to target cells of 10 to 1, and the RDL-FACS reading method after a 24 hour incubation period (see Materials and Methods). The results showed that the BEAT CD38-best match 9G7/CD3(SP34-kappa2) antibody was highly efficient in redirecting CD38+ T-cell killing of Daudi cells with an EC50 of 2 pM (average of 3 donors).

Example 5: Functional equivalence of improved SP34 in scFv format.

In order to determine whether the modifications introduced into the various SP34 scFvs to improve their expression had an effect on functional properties, namely CD3 binding, they were examined in the context of an anti-CD38xCD3 bispecific antibody. The CD38-binding antibody arm, represented as FAB, contained the heavy chain variable region encoded by SEQ ID NO: 60 and the light chain variable region encoded by SEQ ID NO: 61. The CD3-binding antibody arm contained the original mouse SP34 in the format modified to scFv (SEQ ID NO: 207), or modified humanized SP34 scFv containing the indicated heavy/light chain combinations: H1/L21 (SEQ ID NO: 67), H5/L32 (SEQ ID NO: 68), H5/L65 (SEQ ID NO: 69) and H5/L67 (SEQ ID NO: 70).

Carried out transient expression and purification of each of the BEAT against CD3/CD38, which used different variants of SP34. They were investigated in vitro to compare the ability to redirect destruction by T cells. The CD38-expressing Raji cell line (see Materials and Methods) was used to evaluate redirected killing by T cells. In this assay, human PBMC were used as effector cells at a ratio of effector cells to target cells of 10:1.

- 31 039658

As measured by the RDL-FACS method, all _BEATs showed a comparable EC50 between 6 and 10 pM (mean of 2 donors, 24 h incubation), FIG. sixteen.

Example 6 The D401Q substitution disrupts the binding of proteins A and G in one of two types of homodimers.

For analytical purposes, we generated BTA and BTB homodimers of the BEAT FABxscFv antibody, in which the FAB arm (BTA homodimer) did not bind protein A because it contained a non-VH3 variable domain and a portion of the human IgG3 Fc region similar to that described above, and the scFv arm (BTB homodimer) binds to protein A via both the VH3-based variable domain and the human IgG1 Fc region. After expression and purification (described above), we found that although the BTA homodimer could be easily purified with protein G (heavy chain SEQ ID NO: 156 and light chain SEQ ID NO: 74), the BTA homodimer could only be isolated with protein A, and it did not bind to the G protein, suggesting, for example, misfolding or conformational changes in the G protein binding site in the Fc region bearing the BTB CH3 domain (SEQ ID NO: 159). Considering that protein G binds to almost the same part of the Fc region as protein A (on the border of the CH2-CH3 domains), it was surprising that BTB homodimers bind to protein A, but not to protein G. It was then suggested that protein A binding to the BTB homodimer was mediated by only one protein A binding site located in the scFv molecule, which, as mentioned above, contains a VH3-based variable domain. In order to assess the involvement of scFv in protein A binding, a construct was prepared that did not contain VH3-based scFv and found to no longer bind protein A (SEQ ID NO: 160, see FIG. 17). It was concluded that in the Fc region of the BTB homodimer, both binding sites for proteins A and G are not functional.

This latter result suggested that the BTB homodimer was indeed misfolded or altered in structure in the A/G protein binding region. However, earlier results, in which the BTB CH3 domain was used in the context of a heterodimeric immunoglobulin, suggested otherwise. In order to further explore the functionality of the BTB CH3 domain in a heterodimeric context with respect to protein A binding, we generated a BEAT FABxscFv antibody consisting of a FAB arm (BTA arm) that did not bind to protein A because it contained a non-VH3 variable domain. , a human IgG1 hinge, a human IgG3 CH2 domain, and an engineered BTA CH3 domain derived from the human IgG3 isotype (FAB arm heavy chain of SEQ ID NO: 156 and FAB arm light chain of SEQ ID NO: 74) and scFv arm ( BTB arm) containing a VH3-based scFv fragment that lacked the ability to bind protein A, a human IgG1 hinge, a human IgG1 CH2 domain, and an engineered BTB CH3 domain derived from the human IgG1 isotype (scFv arm heavy chain with sequence SEQ ID NO: 161), therefore, binding of the heterodimer to protein A was expected only in the Fc region of the scFv arm. After expression and purification (described above), this molecule was found to actually bind proteins A and G (FIG. 18A) and there were no traces of BTB homodimers. Thus, it was concluded that the A and G protein binding sites are intact in the context of a heterodimeric immunoglobulin and that only when the BTA and BTB chains are joined does a functional A/G protein binding site form.

In order to understand the molecular basis of the disruption of the A/G protein binding site in the BTB homodimer, previously constructed BTB CH3 domains derived from the human IgG1 isotype were examined. Substitution D401Q, EU numbering, comparison between EU numbering and IMGT numbering is presented on the Internet page http://www.imgt.org/IMGTScientificChart/Numbering/Hu_IGH Gnber.html, in the CH3 domain of BTB found in modeling studies, and it was introduced, to disrupt the proposed electrostatic interaction between Arg411 of the first strand of the BTB homodimer and Asp401 of the second strand of the BTB homodimer, bearing in mind that this interaction could stabilize the homodimer complex. Since this substitution was introduced only at the last step in the construction of the BTB CH3 domain, the BEAT FABxscFv antibody described above was generated without the indicated D401Q substitution in the BTB CH3 domain (FAB arm with heavy chain and light chain sequences of SEQ ID NOs: 156 and 74, respectively; scFv arm with the heavy chain sequence of SEQ ID NO: 162). After expression and purification (described above), this molecule was found to bind the A and G proteins (Fig. 18B), but revealed the presence of BTB homodimers, suggesting that the D401Q substitution was responsible for the lack of A/G protein binding seen in the homodimers. BTB thus needs to be promoted in order to reduce impurities of said BTB in the heterodimeric immunoglobulin. It is important to note that residue 401 itself is not part of the A/G protein binding site, but is located opposite at the CH3-CH3 interface. Thus, disruption of A/G protein binding does not result from a direct disruption of interactions between A/G proteins and the Fc region. Rather, it is suggested that, unexpectedly, D401Q leads to a distant conformational change at the CH2-CH3 interface as a result of the poor ability of the BTB chains to pair with each other. In the case of a heterodimer, the boundaries are perfectly aligned, and thus the correct conformation at the CH2/CH3 interface makes the binding site of the A/G proteins functional.

This unexpected result can be used to develop a new strategy effectively.

- 32 039658 to isolate heterodimeric immunoglobulins without homodimer contaminants using the following particular design: the heterodimeric immunoglobulin contains a first strand containing an Fc region of an IgG3 isotype that will contain a BTA CH3 domain and a non-VH3 variable domain or a VH3-based variable domain that has been lacks the ability to bind to protein A (for example, using G65S or N82aS substitutions), or will not contain a variable domain and, therefore, will not be able to bind to protein A, and a second chain that binds protein A containing the CH3 domain of BTB D401Q (for example derived from the human IgG1 isotype) and either a non-VH3 variable domain, or a VH3 variable domain that has been deprived of the ability to bind to protein A (for example, using G65S or N82aS substitutions), or does not contain a variable domain. In this situation, both homodimers did not contain binding sites for protein A. The first homodimer was a dimer of the first chain and did not bind protein A, since it lacked a binding site for protein A, while the second homodimer was a dimer of the second chain with a binding site for protein A. protein A, which was non-functional. The resulting heterodimer contained only one protein A binding site located in the second strand (the protein A binding site of the BTB chain is functional only when connected to the BTA chain). Therefore, the heterodimer will be the only molecular species that will bind to the Protein A chromatography resin, while the unwanted BTA and BTB chain homodimers will be washed out.

Example 7 BEAT activity against CD3/CD38 in vivo in cynomolgus monkey.

Cynomolgus monkeys (macaca fascicularis, 1 male, 1 female per dose) were dosed by intravenous bolus injection with an anti-CD3/CD38 antibody (containing the H5/L65 SP34 heavy/light chain combination in scFv format with the sequence of SEQ ID NO: 69 against CD3 and variable the heavy chain region encoded by SEQ ID NO: 60; and the light chain variable region encoded by SEQ ID NO: 61 in anti-CD38 FAB format), referred to in this example as anti-CD3/CD38 BEAT.

Immunophenotyping of all animals was performed on blood samples (EDTA) taken before treatment and at various time points after each administration of anti-CD3/CD38 BEAT. Evaluation of lymphocyte populations was performed using flow cytometry using the following markers: marker of all T-lymphocytes (CD4 and CD8), marker of monocytes (CD14) and CD38. Samples were analyzed on a FACSCanto™ II flow cytometer. The results were analyzed using the FACSDiva™ analysis software as follows: First, an electron window (gate) was drawn based on forward and side scatter properties to select lymphocyte or monocyte populations (base population). The absolute counts per µl of whole blood of the T lymphocyte population (CD4+ T cells and CD8+ T cells) and the CD38-positive monocyte population (CD14+CD38+) were calculated from their relative percentage obtained for the initial population window and the total number of cells in the original population, obtained on a reliable hematology analyzer, according to the formula: absolute number in the population (x10 / μl) = (relative % in the population x total number of cells in the original population) / 100.

There was a clear early decrease in T cell numbers, both CD4-positive and CD8-positive, after the 1 and 10 μg/kg dose (FIG. 19). Compared to cell numbers obtained 1 day prior to dosing, there was a 48% and 78% decrease in CD4+ T cells after the 1 µg/kg dose and more than 90% after the 10 µg/kg dose at time 4 hours. The numbers of CD8+ T cells decreased even more significantly, falling to 86% after a dose of 1 μg/kg. CD4+ T cell counts recovered rapidly and after 48 hours were generally close to pre-dose numbers. However, CD8+ T cell counts recovered more slowly and reached near pre-dose levels after 1 week. This transient decrease in peripheral blood T cells was expected and reflected the activation of these populations by the BEAT anti-CD3/CD38 antibody. These observations suggest that T lymphocytes, and in particular cytotoxic CD8+ T cells, are involved in the destruction of specific cells in these tissues.

There was also a clear decrease in the population of circulating CD38+ monocytes (Fig. 20). There was a very rapid disappearance and then recovery of the level between 24 and 48 hours. This recovery of the level was usually followed by a second significant decrease in the numbers of CD38+ monocytes, indicating that this population was likely to be targeted and destroyed by the cytotoxic action of T- lymphocytes.

Example 8 9G7-based BEAT antibodies showed increased killing activity compared to OKT10xOKT3 BEAT.

The activities of two different 9G7-based BEAT antibodies were compared with the OKT10-based BEAT antibody in RDL assays.

The bispecific antibody referred to herein as BEAT mouse CD38-9G7/CD3 is a BEAT anti-CD38/CD3 antibody derived from a combination of the antigen binding sites described in Example 2 (scFv VH11/VL8 OKT3 for arm against human CD3 epsilon and FAB chimera

- 33 039658 th mouse-human 9G7 antibody for the shoulder against human CD38). The anti-human CD38 arm of the bispecific antibody consisted of a BEAT heavy chain (SEQ ID NO: 177) coupled to a cognate light chain (SEQ ID NO: 178). The anti-human CD3 epsilon arm of the bispecific antibody consisted of the BEAT heavy chain (SEQ ID NO: 179) containing the anti-human CD3 epsilon scFv fragment. The indicated bispecific antibody was transiently expressed and purified as described above.

The bispecific antibody referred to herein as BEAT CD38-OKT10mouse/CD3 is a BEAT anti-CD38/CD3 antibody derived from a combination of the antigen binding sites described in Example 2 (scFv VH11/VL8 OKT3 for the arm against human CD3-epsilon and FAB chimeric mouse-human OKT10 arm anti-human CD38 antibody). The anti-human CD38 arm of the bispecific antibody consisted of a BEAT heavy chain (SEQ ID NO: 180) coupled to a cognate light chain (SEQ ID NO: 181). The anti-human CD3 epsilon arm of the bispecific antibody consisted of the BEAT heavy chain (SEQ ID NO: 179) containing the anti-human CD3 epsilon scFv fragment. The indicated bispecific antibody was transiently expressed and purified as described above.

The BEAT antibody, which is referred to herein as the BEAT CD38-best match 9G7/CD3(SP34-kappa2) antibody, has been described above in Example 4.

The cytotoxic potential of various anti-CD38 MAT 9G7 or OKT10 BEAT antibodies was assessed in the Daudi CD38+ Target Cell Kill Assay (ATCC). To determine the number of dead target cells, data was read based on flow cytometry. A similar flow cytometric readout was used in Moore et al. Blood of April 28, 2011; 117(17):4542-51; Friedrich et al. Mol Cancer Ther 2012; 11:2664-2673; Schlereth et al. Cancer Res 2005; 65:2882-2889. Effector cells were unstimulated PBMCs from healthy donors. The effector:target ratio was typically 10:1 and the incubation time was 48 hours. Briefly, target cells (Daudi) were labeled with a fluorescent cytoplasmic dye such as CFSE and dispensed into 96-well plates (TPP). Serial dilutions of antibodies (3x solutions) and PBMCs (50 μl/well) were then distributed to the appropriate wells. These plates were incubated for 48 hours in a thermostat at 5% CO ₂ and 37°C. Cells were resuspended by pipetting and transferred to 96-well U-bottom plates (TPP). U-bottomed plates were centrifuged for 3 min at 300 g, supernatants were discarded and cells were resuspended in 200 µl of cold FACS buffer (PBS + 2% FBS + 10% Versene) supplemented with 7-AAD (Becton Dickinson) at a dilution of 1/40 . These plates were read immediately on a Guava easyCyte™ flow cytometer (Millipore). For each well, the absolute number of living target cells was determined by setting a window to select dye-positive (CFSE) and 7ADD-negative populations using Flowjo software (Treestar). The % specific cytotoxicity for each sample was determined using the condition under which only target cells were incubated as background level or using the condition under which target cells were mixed with PBMC (no antibody condition). EC ₅₀ values were determined using non-linear regression with variable slope using Prism software (GraphPad Software).

The results presented in table. 2 show EC ₅₀ values obtained for various BEAT designs. 9G7-based BEAT antibodies (BEAT CD38-9G7 mouse/CD3 and BEAT CD38-most appropriate 9G7/CD3(SP34-kappa2)) exhibited killing activity with a markedly lower EC ₅₀ than the BEAT CD38-OKT10 mouse/CD3 antibody (3 ,2 and 1.9 pM for 9G7-based BEAT antibodies compared to 125.6 pM for OKT10-based BEAT antibody), indicating a higher cytotoxic potential of 9G7-based BEAT antibodies against CD38/CD3.

Table 2. 9G7-based BEAT antibodies showed increased killing activity compared to OKT10xOKT3 BEAT____________

BEAT CD38- 9O7mouse/POPs BEAT CD38- most suitable 9G7/CD3(SP34kappa2) BEAT CD38- OCT1 Mice/ CD3 EC50 (pM)* 3.2 1.9 126

*EC ₅₀ indicate the average values of the results of the destruction analyzes performed for various donors of PBMC. N = 8 for 9G7xOKT3, N = 2 for OKT10xOKT3 and N = 1 for 9G7xSP34(GV2)

Example 9 Antibodies 9G7 and 767 showed no agonism, in contrast to the antibody HB7

Since CD38 was capable of signaling, the agonist properties of 9G7 and 767 antibodies were examined compared to clone HB7 in a calcium influx assay in the Jurkat human T-cell lymphoma cell line, which naturally expresses CD38. Jurkat cells were loaded with Fluo-4 dye at a concentration of 2 μM (Invitrogen) for 1 h at 37°C. Cells were washed and resuspended in PBS supplemented with 1 mM Ca2 ⁺ and 1 mM Mg2 ⁺ . Samples were analyzed on a FACScalibur flow cytometer (Becton Dickinson) for 40 s to determine the background level. Then, test antibodies were added to the samples at a concentration of 10 μg/ml and data collection was resumed up to 7 min. Figure G shows the mean fluorescence intensity (MFI) of the dye fluo-4 (FL-1 channel), which is the detected calcium mobilization into the cell cytoplasm as a function of time. Although the isotype control (human IgG1 antibody) did not result in any influx of calcium, the anti-CD38 antibody HB7 produced a strong signal. Strikingly, neither the 9G7 antibody nor the 767 antibody was able to elicit a calcium mobilization signal, indicating that these antibodies, in contrast to the HB7 antibody, do not have CD38 agonistic properties.

The same set of experiments also showed that 9G7 and 767 specifically bind to cells expressing CD38.

Example 10 Epitope Mapping.

Mapping of the h9G7 epitope.

Linear peptide mapping was used to determine the epitope of the humanized 9G7 (best match). The extracellular domain of human CD38 (residues 43-300) was divided into 19 consecutive 13 amino acid peptides and a final 11 amino acid peptide. The peptides were linked to the human IgG1 Fc region (CH2-CH3) with a two amino acid linker (Ala-Ser) between the Fc and the peptide sequence. These constructs were named F1-F20 (sequence SEQ ID NOs: 182-201) and their binding was analyzed by SPR. The humanized 9G7 antibody (heavy chain SEQ ID NO: 145 and light chain SEQ ID NO: 146) was covalently immobilized on a CM5 sensor chip (approximately 1800 RU). Peptide-Fc fusion proteins were applied at a concentration of 15 nM for 240 seconds. The extracellular domain of human CD38 was served as a control. Only F6 and control received a binding signal (FIG. 22). All fragments were reloaded at 200 nM and the strongest signal was obtained for F6 (FIG. 23), while the remaining fragments showed no binding (results not shown). Since F6 was part of the epitope of the SAR650984 antibody described in PCT Publication No. WO 2008047242 (heavy chain with the sequence of SEQ ID NO: 202 and light chain with the sequence of SEQ ID NO: 203), their interaction was checked using SPR and confirmed (results not presented). Thus, h9G7 and SAR650984 compete for binding to human CD38. Moreover, when the F6 sequence was mapped to the three-dimensional surface of the human CD38 extracellular domain, it overlapped with the SAR650984 epitope. However, it was shown that SAR650984 did not cross-react with cynomolgus CD38, while humanized 9G7 bound to cynomolgus CD38 with high affinity (Figure 24), suggesting that the two antibodies have overlapping but distinct epitopes.

In the F6 sequence, residue M110 is a valine in cynomolgus CD38. Thus, in the context of the full-length extracellular domain of human CD38, we mutated M110 to valine (SEQ ID NO: 204) and compared h9G7 and SAR650984 binding to it. Decreased binding of SAR650984 was found compared to binding to wild-type human CD38, while binding of humanized 9G7 was not affected (results not shown). Based on the crystal structure of SAR650984 in complex with the extracellular domain of human CD38 (PDB, access code 4CMH, Deckert et al., 2014 Gin. Cancer Res. 20: 4574), we found an additional residue that is potentially important for SAR650984 binding and that differs between CD38 human and cynomolgus monkey. This residue was T148 and was the methionine in cynomolgus CD38. Mutation T148 to methionine in the extracellular domain of human CD38 (SEQ ID NO: 205) significantly reduced SAR650984 binding, while humanized 9G7 binding was not affected (results not shown). A double mutation in the extracellular domain of human CD38, including mutations M110V and T148M (SEQ ID NO: 206, see Fig. 23), disrupted the binding of SAR650984, while the binding of humanized 9G7 was not affected (Fig. 24). Thus, it can be concluded that although humanized 9G7 and SAR650984 compete for binding to human CD38, the residues that are bound by these antibodies are not identical.

Epitope Mapping 767

The 767 antibody epitope was mapped using the same set of linear peptide-Fc fusion proteins as used in the epitope mapping experiments described above. Antibody 767 (heavy chain of SEQ ID NO: 151 and light chain of SEQ ID NO: 152) was covalently immobilized on a CM5 sensor chip (approximately 700 RU). Peptide-Fc fusion proteins (F1-F20, SEQ ID NOs: 182-201) were applied at a concentration of 1000 nM for 240 seconds. The extracellular domain of human CD38 was served as a control. Only F3 and control received a binding signal (FIG. 25). When the F3 sequence was mapped to the three-dimensional surface of the extracellular domain of human CD38 (FIG. 26), it overlapped with the SAR650984 epitope, in particular with two residues: E76 and H79. However, 767 cross-reacted with cynomolgus CD38 (lower affinity), while SAR650984 did not, suggesting that these two antibodies have overlapping but distinct epitopes. Moreover, SAR650984 was not bound by the construct

- 35 039658

F3, which was revealed with the help of SPR (Fig. 27). Thus, it can be concluded that although 767 and SAR650984 compete for binding to human CD38, the residues that are bound by these antibodies are not identical.

CLAIM

Claims (12)

CLAIM 1. An antibody or fragment thereof that binds to human CD38 containing a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 25 or SEQ ID NO: 208 and a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 26 or SEQ ID NO : 209, and a heavy chain CDR3 comprising the amino acid sequence of SEQ ID NO: 27 or SEQ ID NO: 210; and containing a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 28 or SEQ ID NO: 211 or SEQ ID NO: 214 and a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 29 or SEQ ID NO: 212 and CDR3 a light chain comprising the amino acid sequence of SEQ ID NO: 30 or SEQ ID NO: 213. 2. An antibody or fragment thereof according to claim 1, wherein said antibody or fragment thereof is a mouse antibody, a chimeric antibody, or a humanized antibody. 3. An antibody or fragment thereof according to claim 1 or 2, wherein said antibody or fragment thereof comprises a heavy chain variable region sequence comprising an amino acid sequence selected from the group consisting of SEQ ID NO: 52, SEQ ID NO: 58, SEQ ID NO: 60, or a sequence that is at least 80% identical to the non-CDR portion of any of said heavy chain variable region sequences, and/or said antibody or fragment thereof comprises a light chain variable region sequence comprising an amino acid sequence selected from a group consisting of SEQ ID NO: 55, SEQ ID NO: 59, SEQ ID NO: 61, or a sequence that is at least 80% identical to the non-CDR region of any of the indicated light chain variable region sequences. 4. An antibody or fragment thereof according to claim 1 or 2, wherein said antibody or fragment thereof comprises a heavy chain sequence comprising an amino acid sequence selected from the group consisting of SEQ ID NO: 73, SEQ ID NO: 75, SEQ ID NO: 76, SEQ ID NO: 145, SEQ ID NO: 147, SEQ ID NO: 149, or a sequence that is at least 80% identical to the non-CDR region of any of the specified heavy chain variable region sequences, and/or the specified antibody or a fragment thereof contains a light chain sequence comprising an amino acid sequence selected from the group of SEQ ID NO: 74, SEQ ID NO: 77, SEQ ID NO: 146, SEQ ID NO: 148, SEQ ID NO: 150, or a sequence that at least 80% identical to the non-CDR region of any of the indicated light chain variable region sequences. 5. An antibody or fragment thereof according to claims 1 to 3, wherein said antibody or fragment comprises human heavy and/or light chain constant regions and the human heavy chain constant region is selected from the group of human immunoglobulins consisting of IGHG1, non-fucosylated IGHG1 and IGHG4. 6. A bispecific antibody that binds to human CD38 and human CD3, comprising an anti-CD38 antibody fragment according to any one of claims 1 to 3 and an anti-CD3 antibody fragment, said anti-CD3 antibody fragment comprising a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO : 3, and a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 4, and a heavy chain CDR3 comprising the amino acid sequence of SEQ ID NO: 5 or SEQ ID NO: 9; and contains a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 6 and a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 7, SEQ ID NO: 11, and a light chain CDR3 comprising the amino acid sequence of SEQ ID NO: 8, SEQ ID NO: 10, SEQ ID NO: 12; or said anti-CD3 antibody fragment comprises a heavy chain CDR1 comprising the amino acid sequence of SEQ ID NO: 163 and a heavy chain CDR2 comprising the amino acid sequence of SEQ ID NO: 164 and a heavy chain CDR3 comprising the amino acid sequence of SEQ ID NO: 165; and contains a light chain CDR1 comprising the amino acid sequence of SEQ ID NO: 166 and a light chain CDR2 comprising the amino acid sequence of SEQ ID NO: 167 and a light chain CDR3 comprising the amino acid sequence of SEQ ID NO: 168. 7. The bispecific antibody of claim 6, wherein said anti-CD38 antibody fragment is selected from the group consisting of Fab, Fab', Fab'-SH, Fd, Fv, dAb, F(ab') ₂ , scFv, single chain Fv bispecific dimers , diatel, triatel and scFv. 8. A bispecific antibody according to any one of claims 6, 7, further comprising an anti-CD3 antibody fragment selected from the group consisting of Fab, Fab', Fab'-SH, Fd, Fv, dAb, F(ab') ₂ , scFv , bispecific dimers of single-chain Fvs, diabodies, triabodies, and scFvs. 9. A composition for the treatment of a hematological disease, comprising an antibody or fragment thereof that binds to human CD38 according to any one of claims 1 to 5 or a bispecific antibody that binds to human CD38 and human CD3 according to any one of claims 6 to 8 and a pharmaceutically acceptable carrier. - 36 039658 10. An immunoconjugate for delivering a therapeutic agent to a cell that expresses a target antigen containing an antibody or fragment thereof that binds to human CD38, according to any one of claims 1 to 5, or a bispecific antibody that binds to human CD38 and human CD3, according to any of claims 6-8 conjugated to a therapeutic agent. 11. The use of an antibody or fragment thereof according to any one of claims 1 to 5 as a drug for the treatment of a disease selected from the group consisting of malignant hematological diseases, including multiple myeloma, B-cell chronic lymphocytic leukemia, B-cell acute lymphocytic leukemia, macroglobulinemia Waldenström, primary systemic amyloidosis, mantle cell lymphoma, prolymphocytic/myelocytic leukemia, acute myeloid leukemia, chronic myeloid leukemia, follicular lymphoma, NK cell leukemia, and plasma cell leukemia. 12. The use of a bispecific antibody according to any one of claims 6 to 8 as a drug for the treatment of a disease selected from the group consisting of malignant hematological diseases, including multiple myeloma, B-cell chronic lymphocytic leukemia, B-cell acute lymphocytic leukemia, Waldenström macroglobulinemia, primary systemic amyloidosis, mantle cell lymphoma, prolymphocytic/myelocytic leukemia, acute myeloid leukemia, chronic myeloid leukemia, follicular lymphoma, NK cell leukemia, and plasma cell leukemia.