Methods for humanizing antibodies

Abstract

Provided herein is a method for obtaining humanized antibodies based on non-human antibodies with affinity for an antigen of interest, starting from experimental or model antibody structures, grafting non-human CDRs into a library of human frameworks, and energetically relaxing and ranking the designs using constrained / constrained atomistic simulations.

Description

[Technical field]

[0001] Related Applications This application claims the benefit of priority to U.S. Provisional Patent Application No. 63 / 230,089, filed August 6, 2021, the contents of which are incorporated herein by reference in their entirety.

[0002] Sequence Listing A 162,676 byte ASCII file entitled txt_93036.txt, created on August 4, 2022, submitted concurrently with the filing of this application, is hereby incorporated by reference.

[0003] FIELD OF THEINVENTION The present invention, in some embodiments thereof, relates to computational methods for designing antibodies, and more particularly, but not exclusively, to methods for humanizing antibodies. [Background technology]

[0004] Antibodies are the largest segment of protein-based therapeutics, with over 100 in clinical use or under regulatory review. As many as 40% of these antibodies were isolated from animal sources, mostly murine / mouse, and humanized prior to clinical application. Antibody humanization is essential to achieve three important therapeutic goals: recruiting the immune system through Fc effector functions; extending blood circulation half-life; and reducing immunogenicity if antibodies are to be used for long-term treatment.

[0005] Despite the important role of antibody humanization, it is a repetitive and often frustrating process. The first step is to chimerize the animal variable domain (Fv) with a human constant domain. Next, the Fv, which contains more than 200 amino acids, is humanized. This step is often complicated by the fact that the Fv contains the so-called complementarity determining regions (CDRs) involved in antigen recognition. Thus, the mainstream humanization strategy grafts CDRs from animal sources onto a human framework, typically resulting in an Fv with more than 80% sequence identity with the human germline (compared to 50-70% for mouse Fvs). To increase the chance that the grafted CDR is compatible with the human framework, it is usually selected from those that show the highest homology with the parent antibody. Other approaches to antibody humanization use structural similarities in the CDR regions, rather than sequence homology, and either humanize only predicted immunogenic segments in the parent framework or graft fragments from the human framework into the animal antibody.

[0006] However, despite these advances, humanization of Fvs usually leads to significant, sometimes orders of magnitude, reductions in expression levels, stability and affinity. The reduction in the biophysical properties of antibodies is particularly detrimental for antibodies intended for clinical use, as it can result in reduced efficacy and lead to undesirable complications in drug formulation and delivery. Generally, therefore, the third step, "backmutation," mutates positions in the humanized antibody back to their parental identity through repeated design-and-experiment cycles.

[0007] The reasons underlying the decrease in affinity and stability due to humanization are structural and energetic. Structural analysis has pointed out the importance of a region of the framework called the Vernier zone, which underlies the CDR. Despite the relatively high conservation of the framework, the Vernier zone contains about 30 sequence determinants that vary even between homologous frameworks; these determinants are essential for the structural integrity and relaxation of the CDR. Therefore, most backmutation attempts use structural modeling to select mutations that reconstruct some Vernier zone positions found in animal antibodies. This process can restore the affinity and stability of the parent antibody at the cost of a lower degree of humanization and long repeats.

[0008] U.S. Patent No. 8,343,489 describes the use of three-dimensional structural information to guide the process of modifying antibodies with amino acids from one or more templates or surrogates such that the antigen-binding properties of the parent antibody are maintained and the potential for immunogenicity is reduced when administered as a therapeutic in humans.

[0009] International Publication No. WO2019 / 025299 provides a method for humanizing non-human antibodies using a structure-based scoring matrix that can be used to determine the requirements and suitability of specific backmutations of amino acid residues at defined positions of a selected human germline sequence. The scoring matrix takes into account the shape, three-dimensional structure and interactions of each residue and change, thereby determining the effect of specific amino acid residue changes on antigen binding. Summary of the Invention

[0010] Humanization is an essential step in developing animal-derived antibodies into therapeutics, and approximately 40% of FDA-approved antibodies are humanized. Traditional humanization methods graft the complementarity determining regions (CDRs) of an animal antibody onto several dozen homologous human frameworks. However, this process often dramatically reduces stability and antigen binding, and requires iterative mutational fine-tuning to restore the properties of the original antibody. Provided herein is a method called computational human antibody (hUMan AntiBody) design ("CUMAb"), which starts from an experimental or model antibody structure, grafts animal CDRs onto thousands of human frameworks, and uses Rosetta atomic-level simulations to rank the designs by energy and structural completeness.

[0011] Thus, the present disclosure provides a method for designing antibodies that are compatible for use in humans even though they originate from another species, i.e., a method for humanizing antibodies, which is based on a ranking based on structure and energy, rather than the commonly used sequence homology. Starting from an experimentally determined or calculated model structure of a non-human Ab and a database of human antibody germline sequences, the method involves generating a large number of grafted structures, and then using computational atomic structure design, as provided in the Rosetta package, to moderate, score and rank the designed variants to be humanized and grafted based on their energetic stability scores. Crucially, automation allows the method to scale humanization from a few dozen homology frameworks to as many as 20,000 different ones. Proof-of-concept (POC) experiments showed that some of the top-ranked designs against three unrelated targets (for which classical humanization methods have failed) showed expression levels and binding properties comparable to the parent animal antibodies, without the need for an iterative backmutation process. In all cases used in POC, the experimentally best performing humanized designs were derived from human frameworks that were not necessarily the most homologous to the parent Ab, suggesting that energy-based humanization may solve the problems seen with traditional homology-based humanization. The method provided herein is also referred to as Computational human Antibody design (CUMAB).

[0012] Thus, according to an aspect of some embodiments of the present invention, there is provided a method for designing and generating a humanized antibody having affinity for an antigen of interest, the method comprising: i) providing a structural model of a non-human antibody (parent Ab) having affinity for an antigen of interest and identifying amino acid residues of at least one complementarity determining region (CDR) in the structural model; ii) generating all combinations of antibody segments derived from a plurality of human antibody germline sequences and substituting corresponding amino acid residues in each of the combinations with amino acid residues of the CDRs, thereby obtaining a library of grafted human antibody sequences; iii) threading each of the grafted human antibody sequences onto the structural model, and subjecting each of the threaded grafted human antibody structures to a constrained energy minimization (constrained structural relaxation), thereby obtaining a plurality of threaded grafted human antibody structures; iv) ranking multiple relaxed grafted human antibody structures by energy score; v) clustering the plurality of relaxed grafted human antibody structures according to V / J gene families to thereby obtain an energy-ranked, gene-family clustered library of humanized antibody designs; and vi) expressing at least one humanized antibody design from at least one cluster of humanized antibody designs and selecting at least one humanized antibody design having affinity to the antigen of interest, thereby obtaining a humanized antibody having affinity to the antigen of interest. This is achieved by:

[0013] In some embodiments, the method further comprises subjecting the structural model to energy minimization (constrained structural relaxation) prior to the threading step.

[0014] In some embodiments, the antibody segment is selected from the group consisting of a heavy chain variable (V) gene segment, a light chain variable (V) gene segment, a heavy chain joining (J) gene segment, a light chain joining (J) gene segment, a kappa gene segment, and a lambda gene segment.

[0015] In some embodiments, the method further comprises removing (filtering out) sequences that display three or more cysteines outside the CDRs from the library of human antibody sequences to be grafted.

[0016] In some embodiments, the method further comprises removing (filtering out) sequences that display an Asn-Gly or Asn-X-Ser / Thr (wherein X is not Pro) motif from the library of human antibody sequences to be grafted.

[0017] In some embodiments, the method further comprises removing (filtering out) structures from the plurality of relaxed grafted human antibody structures that exhibit an RMSD of more than 0.5 Å in the main chain atoms of the CDRs compared to the structural model of the non-human antibody.

[0018] In some embodiments, multiple human antibody germline sequences can be obtained from a human gene database.

[0019] In some embodiments, the human genetic database is the IMGT database of immunogenetics and immunoinformatics.

[0020] In some embodiments, the non-human antibody is a murine antibody.

[0021] As used herein, the term "about" refers to ±10%.

[0022] The terms "comprises, comprising, includes, including," "having," and their cognates mean "including but not limited to."

[0023] The term "consisting of" means "including and limited to."

[0024] The term "consisting essentially of" means that a composition, method, or structure may include additional ingredients, steps, and / or moieties only if the additional ingredients, steps, and / or moieties do not materially alter the basic and novel characteristics of the claimed composition, method, or structure.

[0025] As used herein, the phrases "substantially devoid of" and / or "essentially devoid of" in the context of a particular substance refer to a composition that is completely free of that substance or that contains about 5, 1 or 0.1% of that substance by total weight or volume of the composition. Alternatively, the phrases "substantially devoid of" and / or "essentially devoid of" in the context of a process, method, property or characteristic refer to a process, composition, structure, article that is completely free of a particular process / method step, or a particular property or characteristic, or a particular process / method step is performed at less than about 5, 1 or 0.1% compared to a given standard process / method, or is a property or characteristic that is characterized by less than about 5, 1 or 0.1% compared to a given standard.

[0026] As used herein, the term "substantially maintaining" when applied to an original property, or a desired property, or a resulting property of an object or composition, means that the property does not change by more than 20%, 10% or 5% in the treated object or composition.

[0027] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments and / or to exclude the incorporation of features from other embodiments.

[0028] The phrases "optionally" or "alternatively" are used herein to mean "is provided in some embodiments and is not provided in other embodiments." Any particular embodiment of the invention may include multiple "optional" features unless such features are inconsistent.

[0029] As used herein, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise. For example, the term "a compound" or "at least one compound" can include a plurality of compounds, including mixtures thereof.

[0030] Throughout this application, various embodiments of the present invention may be presented in a range format. It should be understood that the description in range format is merely for convenience and brevity, and should not be construed as an inflexible limitation on the scope of the present invention. Thus, the description of a range should be considered to have all possible subranges specifically disclosed, as well as individual numerical values ​​contained within that range. For example, the description of a range such as 1-6 should be considered to have specifically disclosed subranges (e.g., 1-3, 1-4, 1-5, 2-4, 2-6, 3-6, etc.), as well as individual numerical values ​​contained within that range (e.g., 1, 2, 3, 4, 5, and 6). This is true regardless of the breadth of the range.

[0031] Whenever a numerical range is given herein, it is intended to include any recited numbers (fractional or integer) within the given range. The phrases "ranging between" a first given number and a second given number and "ranging from" a first given number to a second given number are used interchangeably and are intended to include the first given number, the second given number, and all fractional and integer numbers therebetween.

[0032] The terms "process" and "method" as used herein refer to manner, means, techniques, and procedures for accomplishing a given task, and include, but are not limited to, such manner, means, techniques, and procedures that are known to a practitioner in the chemical, pharmacological, biological, biochemical, and medical arts fields or that are readily developed by a practitioner in the chemical, materials, mechanical, computational, and digital arts fields from known manner, means, techniques, and procedures.

[0033] Unless otherwise specified, all technical and / or scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.Although methods and materials similar or equivalent to those described herein can be used to carry out or test embodiments of the present invention, exemplary methods and / or materials are described below.In case of conflict, this patent specification, including definitions, will take precedence.In addition, the materials, methods and examples are only illustrative and are not intended to be limiting in any way.

[0034] Some embodiments of the present invention are described herein, by way of example only, with reference to the accompanying drawings and images. With regard to specific reference herein to detailed drawings and images, it should be emphasized that the details shown are exemplary and are intended to provide an illustrative discussion of embodiments of the present invention. In this regard, the description given by the drawings and images will make apparent to one skilled in the art how embodiments of the present invention may be implemented. The description of the accompanying drawings is as follows: [Brief description of the drawings]

[0035] [Figure 1A-C]shows structural aspects of CDR grafting, FIG. 1A shows a schematic diagram representing the domains of an antibody, the inset shows the crystal structure of an antibody variable region (4IJ3; SEQ ID NO: (heavy / light): 43 / 42) with the heavy chain in blue and the light chain in pink and the CDRs highlighted in light blue and light pink depending on the chain, while the schematic diagram at the bottom of the inset shows the break of the V and J genes within each chain, FIG. 1B shows a table comparing the CUMAB CDR definitions with those from Kabat using Kabat numbering, and FIG. 1C shows the distribution of energies of sequences generated in a typical CDR grafting experiment - indicated by lines are the energies of the parental sequences after the same protocol, the top 5 clustered designs, and the remaining combinations of light and heavy chain sequences found in the top 5 designs. [Figure 2A-C] Shown is an initial attempt using the currently provided method for Ab humanization against an anti-QSOX1 antibody: FIG. 2A. Dot blot of 10 designs (SEQ ID NOs (heavy / light): 48 / 50, 48 / 51, 48 / 52, 48 / 53, 48 / 54, 49 / 50, 49 / 51, 49 / 52, 49 / 53, 49 / 54 (design labels begin with "K", some shown in duplicate) as well as chimeras 3 and 18 from AbLift stabilization of chimeric antibodies, and several failed designs (beginning with "L") from a previous initial effort, expressed as full-length IgG1 in HEK293 cells; FIG. 2B. Dot blot of designs and two chimeras from an anti-QSOX1 antibody, QSOX1, QSOX1 inhibition was tested by incubating with 55 kDa substrate with 2 cysteines and 55 kDa substrate with 2 cysteines. The reaction was stopped by adding PEG maleimide 5k, which adds an apparent mass of about 30 kDa per reduced cysteine ​​pair to the protein. If QSOX1 is present and active, the cysteines are oxidized and only the 55 kDa band is present. If QSOX1 is absent or inhibited, the cysteines are reduced and the 85 kDa band is also present. Figure 2C. Electrophoretic analysis of the 10 designs and two chimeras under denaturing conditions. Bands are predicted to be at approximately 25 kDa (light chain) and 50 kDa (heavy chain); [Figure 3A-F]Humanization of anti-QSOX1 antibodies: Figure 3A. Fifteen humanized designs were expressed and expression levels were measured by dot blot (Table 4: SEQ ID NOs (heavy / light): 1 / 5, 1 / 6; 1 / 7, 1 / 8, 1 / 9, 2 / 5, 2 / 6, 2 / 7, 2 / 8, 2 / 9, and 3 / 5, 3 / 6, 3 / 7, 3 / 8, 3 / 9). Chim18 (SEQ ID NOs: 46 / 47) is an antibody chimeric version with additional stabilizing mutations that was previously shown to have the same expression levels as the parent antibody (SEQ ID NOs: 57 / 58). Figure 3B. The designs were run on a non-denaturing gel. The predicted size of each antibody is approximately 150 kDa. Figure 3C. The designs were screened for QSOX1 inhibition by incubating the antibody, QSOX1, and the QSOX1 substrate ZG16. The reaction was stopped by the addition of PEG maleimide 5k, which added an apparent mass of about 30 kDa when the cysteine ​​was reduced. QSOX1 activity was observed at a slower electrophoretic mobility (higher band) of ZG16, whereas its inhibition was manifested at a faster mobility (lower band). Figure 3D. The four best designs from the preliminary screen were expressed in larger cultures, purified, and run on a non-denaturing gel (SEQ ID NOs (heavy / light): 1 / 6, 2 / 8, 3 / 6, 1 / 6 (repeat), and 3 / 8). Figure 3E. QSOX1 inhibition was measured by incubating antibodies (SEQ ID NOs (heavy / light): 79 / 80, 2 / 8, and 3 / 8; various concentrations) and QSOX1 (25 nM). The reaction was activated by adding DTT and then stopped by adding DTNB. The absorbance at 412 nm was measured and reflects the number of free thiols from DTT reacted with DTNB. Plotted is percent inhibition = absorbance / (absorbance without QSOX-absorbance without antibody). Figure 3F. Dot blot analysis of anti-QSOX1 CDRs grafted onto the most homologous consensus framework IGKV1-IGHV4 (SEQ ID NO:56 / 55) compared to mock transfected and well expressed abLIFT design 3 (SEQ ID NO:44 / 45); [Figure 4A-E]Humanization of anti-PSA clone 10: FIG. 4A. Four anti-PSA antibodies were expressed as chimeras and expression was measured by ELISA of the supernatants (SEQ ID NOs: 83 / 84, 85 / 86, 87 / 88, 81 / 82). Clone 3 (heavy and light chains are represented by SEQ ID NOs: 68 and 67, respectively) showed strong expression, while clones 9, 10, and 56 (heavy and light chains are represented by SEQ ID NOs: 84 / 83, 86 / 85, and 88 / 87, respectively) showed no expression. FIG. 4B. Five humanized designs for clone 10 were measured by ELISA of the supernatants. All five designs showed high expression levels (PSA1-PSA5 with heavy and light chains SEQ ID NOs: 70 / 69, 72 / 72, 74 / 73, 76 / 75, and 78 / 77, respectively). Figure 4C. Binding of recombinant five humanized designs (PSA1-PSA5) to clone 10 (after purification with protein G) to PSA was measured using ELISA. Figure 4D. Immunoprecipitation of five humanized designs to clone 10 (PSA1-PSA5 with heavy and light chain sequence numbers (heavy / light): 70 / 69, 72 / 72, 74 / 73, 76 / 75, and 78 / 77, respectively) with protein G after incubation with recombinant PSA. Membranes were developed with a commercially available anti-PSA antibody. Recombinant PSA is known to be approximately 35 kDa, with a protein degradation product of approximately 25 kDa. Figure 4E. Antibodies (Ayelet79 and PSA4-sequence number: 76 / 75) were incubated overnight with PSA (Rec or native in LNCaP supernatant) and immunoprecipitated with protein G. Membranes were developed with a commercially available anti-PSA antibody. Native PSA is approximately 35 kDa with an additional higher band of pre-PSA; [Figure 5A-D]SDR-grafting of anti-lysozyme antibodies: FIG. 5A. Histograms of binding from a yeast display screen of six designs (SEQ ID NOs (heavy / light): 27 / 22, 21 / 22, 23 / 24, 25 / 26, 27 / 28, 27 / 29) as well as a negative control (G6) and a positive control (D44.1, SEQ ID NOs: 103 / 102) using a lysozyme concentration of 240 nM. FIG. 5B. Denaturing gel of D44.1 and Des6 (SEQ ID NOs: 106 / 107) expressed as full-length IgG1 after purification with Protein G. Three elutions were performed for each sample. Bands are predicted to be at approximately 25 kDa (light chain) and approximately 50 kDa (heavy chain). FIG. 5C. Denaturing gel of Des1 (SEQ ID NOs: 19 / 20) expressed as full-length IgG1 after purification with Protein G. Four elutions were performed and are shown from left to right on the gel. The bands are predicted to be at approximately 25 kDa (light chain) and 50 kDa (heavy chain). Figure 5D. Dual layer interferometry traces for Des1 (SEQ ID NO: 19 / 20). Lysozyme was at concentrations of 1000, 250, 100, 25, and 10 nM, and 1 μg / mL of antibody was loaded onto the sensor. Kd was calculated to be 11 nM with an error of 0.31 nM; [Figure 6] 1 shows a schematic flow chart of the methods provided herein, in accordance with some embodiments of the present invention. [Figure 7] A comparative bar graph is shown comparing the activity (affinity) analysis results performed on expressed antibodies 492 (SEQ ID NO: 79 / 80), chimera, h2bk4 (SEQ ID NO: 2 / 8), h3k4 (SEQ ID NO: 3 / 8), h3newK2 (SEQ ID NO: 4 / 6) and h3newK4 (SEQ ID NO: 4 / 8). DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0036] The present invention, in some embodiments thereof, relates to computational methods for designing antibodies, and more particularly, but not exclusively, to methods for humanizing antibodies.

[0037] Before describing at least one embodiment of the invention in detail, it is to be understood that the invention is not necessarily limited in its application to the details set forth in the following description or illustrated by way of examples, as the disclosure is intended to encompass other embodiments or to be practiced or carried out in various ways.

[0038] As discussed herein above, antibody humanization is essential for developing animal-derived antibodies into therapeutics, and about 40% of clinically approved antibodies have been humanized.The traditional method for humanization is to graft the complementarity determining region (CDR) of an animal antibody onto several dozen homologous human frameworks.Despite the success and importance of this method, grafting often substantially reduces stability, expression level and binding affinity or specificity, and requires repeated fine-tuning of mutations to reproduce the properties of the parent antibody.

[0039] The present disclosure provides a computational method using experimental or model structures as input, grafting animal CDRs onto thousands of human frameworks, and, for example, Rosetta atomic simulations to relax and rank designs by energy. Implementations of the methods provided herein have resulted in designs that exhibit the same affinity as animal antibodies, even when traditional antibody humanization has failed to produce expressible antibodies, much less high affinity binders. The successful design of a large number of mutations in antibody variable domains suggests that other important antibody design methods may be amenable to design automation. The methods can be made publicly accessible via web servers, streamlining the antibody design pipeline.

[0040] Unlike other computational methods for antibody (Ab) humanization, which require experimentally derived parent (original) antibody (Ab) structures, the method presented herein (also referred to as "CUMAB") does not rely on such a starting point. Computational methods for Ab humanization, such as those presented in U.S. Pat. No. 8,343,489, are performed by aligning the parent antibody structure with structures in the Protein Data Bank to find the closest match, and then performing the complete antibody structure grafting method and the EPU method for each of the CDRs and frameworks separately. The method presented herein, according to some embodiments of the present invention, does not rely on other antibody structures; this is an advantage, since relying on other antibody structures can be very limiting and misleading.

[0041] Computational methods for Ab humanization, such as those presented in U.S. Pat. No. 8,343,489 and WO 2019 / 025299, rely on sequence homology to choose from previously selected structures, and CDR definitions can also bias the final result. Furthermore, such methods require many intensive visual inspection steps of the antibodies, making such methods incompatible with high-throughput efforts.

[0042] The method provided herein (CUMAb) is based on the fundamental insight that the stability and activity of an antibody are determined by both the CDRs and the amino acid positions where the CDRs are located. This insight may help address other important challenges in antibody design leading to a general, reliable and automated antibody design strategy.

[0043] Some independent antibody CUMAb designs show similar affinity to animal antibodies, even when conventional antibody humanization fails to produce expressible antibodies. Low-energy but non-homologous frameworks are often preferred over the most homologous ones, and some CUMAb designs that code multiple mutations to each other are functionally equivalent. Surprisingly, some designs show significant improvements in stability and expressibility compared to the parent antibody. Thus, CUMAb represents a general and modern approach to optimize antibody stability and expressibility while increasing the degree of humanization.

[0044] Instead of using only a few dozen homology frameworks as in most conventional humanization strategies, the method presented here (CUMAb) uses all possible combinations of human gene segments (more than 20,000 for each antibody) and ranks them by energy. Genes belonging to a single subgroup are similar to each other, but they contain mutations, including in vernier positions, that may stabilize specific CDRs of the parent antibody. Structural analysis shows that low-energy designs retain important framework-CDR interactions that may be removed in homology-based humanization. Thus, the lowest energy designs from this large space of possible frameworks may retain stability, expression, and binding affinity, and even improve stability and expressibility, more than the most homologous designs. CUMAb may thus provide a strategy to improve antibody stability, including antibodies of human origin, while maintaining or increasing the degree of humanization.

[0045] Crucially, the energy-based strategy eliminates recurrent back mutations even when starting from an Fv model structure. CUMAb therefore broadens the scope of antibody humanization to, in principle, all animal antibody sequences. Moreover, CUMAb is automated and requires experimental screening of fewer than a dozen constructs, substantially reducing time and costs. It is therefore applicable at the scale of many antibodies in parallel, including those for which no structure is available. We also observed that in the three cases tested, several unique humanization alternatives exhibited stability and binding properties comparable to those of animal antibodies. These designs provide an alternative to select the best-behaving humanized antibody variants in downstream experiments, because they differ from each other by a large number of surface mutations.

[0046] The method provided herein (CUMAb) relies on readily available input data, it can be set up for full automation, and it can be easily scaled to many antibodies and used by non-skilled personnel.

[0047] Energy-based ranking of humanized antibodies: A computational workflow was developed to model and energy rank structures in which the Fv framework region is replaced with all compatible human frameworks. The frameworks are encoded by two gene segments, V and J, on both light and heavy chains (Figure 1A) [Janeway, CA; Travers, P.; Walport, M.; Shlomchik, MJ Immunobiology: The Immune System in Health and Disease, 6th edition; Garland Science: New York, 2005]. Recombination of all human V and J segments on both light and heavy chains generates tens of thousands of unique frameworks (63,180 including kappa light chains and 48,600 for lambda light chains) [Lefranc, MP Nomenclature of the Human Immunoglobulin Kappa (IGK) Genes. Exp. Clin. Immunogenet. 2001, 18(3), 161-174]. The working hypothesis of this study is that energy-based ranking of a large number of different human frameworks can yield more stable and functional humanized designs than those from traditional homology-based humanization.

[0048] CDRs were defined based on a combination of definitions provided elsewhere [MacCallum, RM; Martin, AC; Thornton, JM Antibody-Antigen Interactions: Contact Analysis and Binding Site Topography. J. Mol. Biol. 1996, 262(5), 732-745; Dondelinger, M.; Filee, P.; Sauvage, E.; Quinting, B.; Muyldermans, S.; Galleni, M.; Vandevenne, MS Understanding the Significance and Implications of Antibody Numbering and Antigen-Binding Surface / Residue Definition. Front. Immunol. 2018, 9, 2278] (Figure 1A-B).

[0049] The use of the methods provided herein begins by replacing the amino acid sequences in the regions outside the CDRs with all combinations of human V and J sequences obtained from the ImMunoGeneTics (IMGT) database. The light chain is humanized with either lambda or kappa light chains depending on the light chain class of the animal antibody. Genes containing Asn-Gly or Asn-X-Ser (X is not Pro) sequence motifs were removed, as these may result in undesired post-translational modifications. In addition, all sequences showing three or more cysteines outside the CDRs are excluded to reduce the possibility of antibody misfolding or aggregation. These constraints preserve the majority of possible human gene combinations, resulting in >20,000 unique frameworks per antibody.

[0050] For the grafting step, human and animal V and J genes were aligned and the parental (non-human, animal) CDR amino acids were replaced with their human counterparts. The result is a library of grafted human antibody sequences.

[0051] Starting with the structure of the parent (non-human) antibody Fv, each humanized design is modeled using Rosetta all-atom calculations by threading the sequence of the humanized design onto the structure of the parent antibody [Leaver-Fay, A.; Tyka, M.; Lewis, SM; Lange, OF; Thompson, J.; Jacak, R.; Kaufman, K.; Renfrew, PD; Smith, CA; Sheffler, W.; Davis, IW; Cooper, S.; Treuille, A.; Mandell, DJ; Richter, F.; Ban, Y.-EA; Fleishman, SJ; Corn, JE; Bradley, P. ROSETTA3: An Object-Oriented Software Suite for the Simulation and Design of Macromolecules. Methods Enzymol. 2011, 487, 545-574]. The result of this step is a multi-threaded grafted human antibody structure.

[0052] The resulting model structures are relaxed by iterative minimization of the side chains and harmonically constrained backbone and combinatorial side chain packing throughout the Fv. Each model is ranked using the ref2015 energy function governed by van der Waals interactions, hydrogen bonding, electrostatics, and potential solvation [O'Meara, MJ; Leaver-Fay, A.; Tyka, MDM; Stein, A.; Houlihan, K.; DiMaio, F.; Bradley, P.; Kortemme, T.; Baker, D.; Snoeyink, J.; Kuhlman, BA Combined Covalent-Electrostatic Model of Hydrogen Bonding Improves Structure Prediction with Rosetta. J. Chem. Theory Comput. 2015, 11(2), 609-622]. The result of this step is multiple relaxed grafted human antibody structures.

[0053] Models in which any of the CDR backbone structures deviate from the parent structure by more than 0.5 Å are removed. To select a diverse set of sequences for experimental testing, the top ranked designs were clustered according to V gene subgroups (7, 6 and 10 clusters for heavy chains V, kappa and lambda), determined according to sequence homology. This clustering resulted in a shortlist of diverse low energy models for experimental testing, also referred to herein as an energy-ranked and gene family clustered library of humanized antibody designs.

[0054] Remarkably, unlike conventional CDR grafting methods, the approach taken by the method provided herein does not rely on homology between mouse and human frameworks and is scalable, designing and ranking 20,000 different humanized constructs on a 500 CPU cluster within a few hours.

[0055] The next step may be to express at least one humanized antibody design from at least one cluster of the resulting humanized antibody designs and to select at least one humanized antibody design. Criteria for successful use of the methods provided herein may include expression level assays, inferences regarding Ab protein stability, as well as assaying the affinity of the expressed Ab for the antigen of interest and comparing this affinity to that of the parent Ab.

[0056] Crystal structure-based CDR grafting: As an exemplary humanization target, an antibody generated by mouse immunization targeting human Quiescin Sulfhydryl Oxidase 1 (QSOX1) was selected [Grossman, I.; Alon, A.; Ilani, T.; Fass, D. An Inhibitory Antibody Blocks the First Step in the Dithiol / disulfide Relay Mechanism of the Enzyme QSOX1. J. Mol. Biol. 2013, 425(22), 4366-4378]. This antibody is challenging to humanize because chimerization of its mouse Fv with the human IgG1 constant region leads to a complete loss of expression in HEK293 cells. This challenge has been addressed previously using the AbLIFT method, which uses atomic-level design calculations to improve intermolecular interactions between Fv light and heavy chain domains [Warszawski, S.; Borenstein Katz, A.; Lipsh, R.; Khmelnitsky, L.; Ben Nissan, G.; Javitt, G.; Dym, O.; Unger, T.; Knop, O.; Albeck, S.; Diskin, R.; Fass, D.; Sharon, M.; Fleishman, SJ: Optimizing Antibody Affinity and Stability by the Automated Design of the Variable Light-Heavy Chain Interfaces. PLoS Comput. Biol. 2020, 16(10), e1008382]. One design, AbLIFT18, rescued HEK293 expression and QSOX1 inhibition but provided an incompletely humanized antibody.

[0057] Using the preliminary humanization workflow provided herein, ten designs were experimentally tested in the following combinations: SEQ ID NO: (heavy / light): 48 / 50, 48 / 51, 48 / 52, 48 / 53, 48 / 54, 49 / 50, 49 / 51, 49 / 52, 49 / 53, and 49 / 54, including the five lowest energy designs after clustering and all combinations of heavy and light chains observed in these five; all of these combinations were within the top 2,700 of 26,000 designs (as ranked using the updated method). The difference between this preliminary protocol and the current one was that the H1 CDR definition in the preliminary version removed two N-terminal amino acid positions compared to the final protocol. In addition, in a preliminary use of the method provided herein, the CDR mutations implemented in the AbLIFT18 design were incorporated into the mouse antibody. These 10 designs (heavy / light chain sequences: 48 / 50, 48 / 51, 48 / 52, 48 / 53, 48 / 54, 49 / 50, 49 / 51, 49 / 52, 49 / 53, and 49 / 54) were formatted as IgG1 full-length antibodies and expressed in HEK293 followed by protein G affinity purification. Qualitative dot blot analysis showed that many designs expressed similarly to AbLIFT18, but none showed comparable levels of QSOX1 inhibition (Figures 2A and 2B). Electrophoretic mobility analysis in denaturing conditions revealed that the apparent molecular weights of the designs were heterogeneous and different from that of the parent antibody, suggesting that these designs were aggregated or misfolded (Figure 2C). Visual inspection of the design models to find the source of these stability issues revealed that one of the humanizing mutations, heavy chain Val24Phe (Kabat numbering), which was present in almost all designs, was structurally incompatible with the CDR H1 configuration in the parent antibody, suggesting that the preliminary CDR definitions failed to include amino acids critical for humanization.

[0058] The design calculations were repeated using the CDR definitions of the final version of CUMAB, incorporating Kabat positions 24 and 25 in the H1 definition (Figure 1A-B). Implementation of the method started directly from the parent mouse antibody, without including mutations in the CDRs from design AbLIFT18. Fifteen designs were experimentally tested (top 5 ranked designs and light and heavy chain combinations from these 5; Table 4), 12 of which showed expression levels comparable to design AbLIFT18 in dot blot analysis (Figure 3A; SEQ ID NOs (heavy / light): 1 / 5, 1 / 6, 1 / 7, 1 / 8, 1 / 9, 2 / 5, 2 / 6, 2 / 7, 2 / 8, 2 / 9, 3 / 6, and 3 / 8).

[0059] Unlike the first humanization attempt, electrophoretic analysis revealed that seven designs showed expression levels comparable to AbLIFT18, with no significant evidence of misfolding or aggregation (Figure 3B; SEQ ID NOs: 1 / 6, 1 / 8, 2 / 6, 2 / 7, 2 / 8, 3 / 6, 3 / 8). Notably, two design antibodies, designated H2B-K4 and H3-K4 (SEQ ID NOs: 2 / 8 and 3 / 8, respectively), showed QSOX1 inhibition levels comparable to the parental murine antibody (492, SEQ ID NOs: 79 / 80) when added at stoichiometric ratios, and one showed significant levels of inhibition even at substoichiometric ratios (Figure 3E). Thus, these two designs recapitulated parental murine expression and activity using the humanized Fv framework. These results also highlight the importance of structurally appropriate CDR definition for successful humanization.

[0060] Humanization based on Fv model structure: Experimental structure determination is a long and uncertain process. For many antibody discovery studies, it is impractical to obtain an experimentally determined structure for each antibody. Recently, structure prediction methods have reached a stage where antibody Fv structures can be predicted with near atomic accuracy, except for the solvent-exposed parts of CDR H3, where the models are still not very reliable [Almagro, JC; Teplyakov, A.; Luo, J.; Sweet, RW; Kodangattil, S.; Hernandez-Guzman, F.; Gilliland, GLSecond Antibody Modeling Assessment (AMA-II). Proteins: Struct. Funct. Bioinf. 2014, 82(8), 1553-1562]. Although H3 is a key determinant for antigen recognition, most of its solvent-exposed regions do not form direct interactions with the framework regions that are the target of humanization. Therefore, despite the strict accuracy required by atomic level design calculations, it was hypothesized by the inventors that the Fv model structure is sufficiently accurate for the purposes of humanization design according to some embodiments of the present invention.

[0061] The accuracy of modeling is high and largely comparable among several recent Fv structure prediction methods. The AbPredict method was chosen to provide the starting model [Lapidoth, G.; Parker, J.; Prilusky, J.; Fleishman, SJ, AbPredict 2: A Server for Accurate and Unstrained Structure Prediction of Antibody Variable Domains. Bioinformatics 2019, 35(9), 1591-1593]. Unlike most other structure prediction methods that rely on sequence homology, AbPredict is energy-dependent. The resulting model structures are therefore stereochemically and energetically relaxed, reducing the risk of distortion due to modeling artifacts and inaccuracies, and may result in a humanization workflow that selects designs that mitigate artificial distortions rather than those that actually stabilize the CDRs. All CDR grafting calculations start with the lowest energy AbPredict model. From this point on, all design calculations are identical to the CUMAB protocol above, except that modeling uncertainty is due to the inclusion of humanized structures regardless of deviation from the starting model structure.

[0062] To test the ability of CUMAb to humanize antibodies without relying on experimentally determined structures, we targeted an antibody produced in mice against human prostate specific antigen (PSA). As with the anti-QSOX1 antibody above, this antibody could not be expressed in HEK293 cells when its Fv was fused to a human IgG1 constant domain (Figure 4A; SEQ ID NOs (heavy / light): 84 / 83, 86 / 85, 88 / 87, 81 / 82). In contrast, all five top-ranked designs obtained by the method provided herein expressed well in HEK293 cells (Figure 4B; SEQ ID NOs (heavy / light): 10 / 11, 10 / 12, 13 / 14, 15 / 16, and 13 / 16), were purified using a protein G affinity column, and were tested for binding of recombinantly produced PSA. The resulting designs showed some level of binding to PSA, with Design 4 (SEQ ID NO: 15 / 16) showing a significantly higher level of binding than the others. The parent murine antibody was used as a benchmark since it could not be expressed, and Design 4 (SEQ ID NO: 15 / 16) showed a low dissociation constant (K D) 2-5 nM (Figure 4C) compared to previously isolated antibodies. Immunoprecipitation with recombinant PSA showed that all five designs bound to PSA, with design 4 (SEQ ID NO: 15 / 16) outperforming the others (Figure 4D). In this experimental proof-of-concept of the method provided herein, the affinity of the designs for PSA appears to be as high as the reference antibody. Finally, design 4 (SEQ ID NO: 15 / 16) and the reference antibody (Ayele79) were tested for their ability to bind to the supernatant of LNCaP cells naturally expressing PSA, showing that both antibodies also bound strongly to naturally expressed PSA (Figure 4E). Thus, despite the uncertainty of using an Fv model rather than an experimentally determined structure, CUMAB produced a soluble, high affinity antibody in a single round.

[0063] SDR port: The CDR grafting methods used in the methods provided herein typically increase sequence identity to 80-88% to human germline (Table 1). To further increase identity, another strategy called specificity-determining residue (SDR) grafting has been proposed [Kashmiri, SVS; De Pascalis, R.; Gonzales, NR; Schlom, J., SDR Grafting-A New Approach to Antibody Humanization. Methods 2005, 36(1), 25-34; De Pascalis, R.; Iwahashi, M.; Tamura, M.; Padlan, EA; Gonzales, NR; Santos, AD; Giuliano, M.; Schuck, P.; Schlom, J.; Kashmiri, SVS, Grafting of “Abbreviated” Complementarity-Determining Regions Containing Specificity-Determining Residues Essential for Ligand Contact to Engineer a Less Immunogenic Humanized Monoclonal Antibody. The Journal of Immunology 2002,169(6),3076-3084].

[0064] SDR grafting only exchanges amino acid positions that directly contact the antigen, while the remainder of the antibody, including the CDRs, is humanized. To prevent undesired changes to the CDR backbone structure, in SDR grafting, the method used only human germline genes that exhibit CDRs that match the length of the CDRs in the parent antibody, except for H3 (see the Examples section below). Furthermore, since the heavy chain J gene segment codes for part of H3, the resulting humanized designs were clustered according to their V gene subgroups as well as their heavy chain J segments. Thus, unlike the CDR grafting described above, designs that differ from each other only in their heavy chain J segments can be selected. Because SDR grafting requires precise identification of antigen-binding amino acid positions and because of the uncertainty in modeling the solvent-accessible regions in H3, its application was limited in this exemplary trial to the experimentally determined antigen-binding structure.

[0065] SDR grafting was applied to the murine antibody D44.1 (SEQ ID NO: 105 / 104), as observed in its co-crystal structure with hen egg white lysozyme (PDB entry: 1MLC). In D44.1, 28 positions out of a total of 63 CDR positions interact with the antigen. Thus, in this specific illustrative example, SDR grafting can result in V gene sequence identity to human germline of approximately 90% in the heavy chain and 93-95% in the light chain, compared to approximately 85-87% in the heavy chain and 79-88% in the light chain using the CDR grafting method. In an initial experimental screen, six designs (heavy / light chain sequences: 27 / 22, 21 / 22, 23 / 24, 25 / 26, 27 / 28, and 27 / 29) were formatted as single-chain variable fragments (scFv) and tested for their binding to lysozyme at a concentration of 240 nM using yeast cell surface display (Figure 5A) [Chao, G.; Lau, WL; Hackel, BJ; Sazinsky, SL; Lippow, SM; Wittrup, KD, Isolating and Engineering Human Antibodies Using Yeast Surface Display. Nat. Protoc. 2006, 1(2), 755-768]. This preliminary screen showed that five of the six designs (SEQ ID NOs: (heavy / light): 27 / 22, 21 / 22, 23 / 24, 27 / 28, and 27 / 29) showed high expression levels, and that designs 1 and 6 (SEQ ID NOs: (heavy / light): 21 / 22 and 27 / 29) bound to lysozyme, with design 1 being significantly better than 6. Although this single concentration experiment was only qualitative, it showed that design 1 could bind lysozyme more potently than the parent D44.1 (SEQ ID NOs: 105 / 104). To obtain quantitative binding data, designs 1 and 6 were formatted as human IgG1 antibodies and expressed in HEK293 cells in addition to D44.1 (expressed as mouse IgG1). The three antibodies expressed well (Figures 5B and 5C), with design 1 showing an affinity of 11 nM for lysozyme as measured by bilayer interferometry binding experiments (Figure 5D). Design 1 and D44.1 (SEQ ID NOs: 19 / 20 and 105 / 104) were expressed as Fabs (human for design 1 and mouse for D44.1), and design 1 showed an affinity of 41.6 nM, while D44.1 had an affinity of 7.4 nM, as measured by surface plasmon resonance. There was a slight decrease in affinity for design 1, but a large increase in the degree of humanization and expression.

[0066] Energy-based humanization utilizes non-homologous frameworks: Conventional antibody humanization strategies use human germline genes that are sequence-wise closest to those of animal antibodies. The energy-based humanization strategy implemented in the methods provided herein, in some embodiments of the present invention, ignores sequence identity and instead focuses on the structural integrity and energy stability (structural relaxation) of the humanized design. The energy-based humanization method provided herein expands the range of potential frameworks from a few dozen highly homologous ones to several thousand. Three test examples presented herein were analyzed to understand what role sequence homology has in the success of antibody humanization.

[0067] Strikingly, in all three cases, the experimentally best performing humanized design is not derived from the best matched human design by sequence identity (see Table 1 in the Examples section below). For both QSOX1 (SEQ ID NO: 59 / 58) and PSA exemplary targeting antibodies, none of the best designs are derived from the closest human V gene subgroup for light or heavy chains, and for the anti-lysozyme antibody, only the heavy chain is matched. Furthermore, for both the anti-PSA and anti-lysozyme antibodies, designs based on the same subgroup V genes (designs 3 and 5; SEQ ID NO: 74 / 73 and 27 / 28, respectively) were significantly inferior to the best performing humanized design. As an additional comparison with conventional humanization strategies, the anti-QSOX1 antibody (SEQ ID NO: 59 / 58) was subjected to humanization using a "consensus" approach. In this humanization approach, frameworks are taken from consensus based on the sequences of V gene subgroups (in this specific example, IGKV1 and IGHV4). However, the designed antibody (SEQ ID NO:55 / 56) could not be expressed and its binding to QSOX1 was therefore not tested (FIG. 3F).

[0068] Taken together, these results suggest that simply grafting CDRs from animal antibodies onto the closest (or consensus) human germline often fails to recapitulate the stability and functional properties of animal antibodies, as observed over decades of antibody design. In contrast, some of the top-ranked designs from energy-based humanization are well expressed, stable, and show good affinity values, and these are usually not from homologous subgroups.

[0069] Method steps: The data preparation step involves obtaining the structure of the non-human parent antibody, either an experimentally obtained crystal structure or a computational / predictive model. The data preparation step also includes compiling a database of human antibody germline sequences using any available immunogenetic and immunoinformatics information source, such as the IMGT reference database. The crystal or predicted model is subjected to energy minimization and structure refinement (as this term is defined and discussed hereinafter) to obtain an energetically stabilized and relaxed structure. For both light and heavy chains, and for kappa and lambda light chains, recombination of all representations of each of the four V and J segments is performed separately. Based on the structure, the residues to be grafted are identified. The CDRs of the germline sequences are replaced with the CDRs of the parent antibody. Any CDR grafts with post-translational modification motifs and / or extra cysteines, e.g., sequences with more than two cysteines in a chain or with Asn-Gly or Asn-X-Ser / Thr motifs outside the CDRs (where X is not Pro), are removed from further analysis. Each of the CDR-grafted germline sequences is threaded into the relaxed conformation of the parent Ab, thereby obtaining a panel of humanized Abs. Each of the threaded humanized Abs is subjected to energy minimization and structural relaxation. The relaxed humanized Abs are ranked according to their energy scores (the top ranked structure has the lowest energy). Relaxed humanized Ab structures showing significant main-chain conformation deviations (eg, greater than 0.5 Å) in any CDR are excluded from further analysis. The relaxed humanized Abs are clustered into subgroups based on their V gene and J segment lineages. Finally, several of the top-ranking humanized structures from each cluster are selected for expression and testing of affinity to the antigen.

[0070] Figure 6 shows a schematic flow chart of the method provided herein according to some embodiments of the present invention. As can be seen in Figure 6, the algorithm starts with two paths, the first path providing a database of human antibody germline sequences, and the second path providing a starting structure in the form of a crystal structure or computational model of a non-human parent Ab. The two paths converge with a step of threading each of the remaining CDR-grafted germline sequences into an energy-minimized (relaxed) starting structure. Once the recombinant structures are ranked by individual energy scores and the recombinant structures are filtered out by RMSD from the parent Ab starting structure, the remaining sequences of the top-ranked humanized structures are clustered by V-gene subgroup J segments, and some of the top-ranked humanized structures from each cluster are selected for expression, and the expressed designs are tested for affinity to the antigen.

[0071] Energy minimization and structure refinement: In some embodiments, the methods presented herein (also referred to as "CUMAb") utilize energy minimization and structural refinement to obtain an energetically relaxed structure. This structural refinement step is performed on the grafted human antibody structure and, optionally, on a structural model of the non-human antibody, also referred to herein as the parent Ab. Structural refinement is a routine procedure in computational chemistry and typically involves weight fitting based on free energy minimization subject to rules, constraints, and harmonic constraints. The structural refinement step can be performed using structural constraints and weight fitting based on any global and / or local energy minimization software.

[0072] According to some embodiments of the invention, the structural model of the non-human antibody is optionally refined by energy minimization before using its coordinates for threading, optionally fixing the structure of the CDR residues.

[0073] The term "weight fitting", according to some embodiments of the present invention, refers to one or more structure refinement procedures or operations aimed at optimizing geometric, spatial and / or energy criteria, for example by minimizing polynomial functions based on pre-determined weights, restraints and constraints (fixed) on sequence homology scores, backbone dihedral angles and / or atomic positions (variable) of the refined structure. According to some embodiments, the weight fitting procedure includes one or more adjustments of bond lengths and angles, backbone dihedral angles (Ramachandran) angles, amino acid side chain packing (rotamers) and repeated substitutions of amino acids, while the terms "adjustment of bond lengths and angles", "adjustment of backbone dihedral angles", "amino acid side chain packing" and "changes in amino acid sequence" are also used herein to refer to well-known optimization procedures and operations widely used, inter alia, in the fields of computational chemistry and biology. An exemplary energy minimization procedure, according to some embodiments of the present invention, is cyclic-coordinate descent (CCD), which can implement the default all-atom energy function in the Rosetta™ software package for macromolecular modeling. For a review of general geometry optimization and refinement techniques, see, for example, "Encyclopedia of Optimization" by Christodoulos A. Floudas and Panos M. Pardalos, Springer Pub., 2008.

[0074] According to some embodiments of the present invention, a suitable computational platform for carrying out the methods presented herein is the Rosetta™ software package platform, which is publicly available from "Rosetta@home" at Baker laboratory, University of Washington, USA. Briefly, Rosetta™ is a molecular modeling software package for understanding protein structure, protein design, protein docking, protein-DNA and protein-protein interactions. The Rosetta software includes multiple functional modules, including RosettaAbinitio, RosettaDesign, RosettaDock, RosettaAntibody, RosettaFragments, RosettaNMR, RosettaDNA, RosettaRNA, RosettaLigand, RosettaSymmetry, etc.

[0075] Weight fitting, according to some embodiments, is carried out under a set of constraints, restrictions and weights, called rules. For example, when refining the main chain atom positions and dihedral angles of any given polypeptide segment having a first structure, to drive it to a different second structure while trying to preserve as much as possible the dihedral angles observed in the second structure, the calculation procedure uses, for example, harmonic constraints that bias the Cα positions and the main chain dihedral angles from being free to depart from the angles observed in the second structure, thus allowing minimal conformational changes to occur for each structural determinant while driving the entire main chain to change to the second structure.

[0076] In some embodiments, global energy minimization is advantageous due to the difference between the energy function used to identify and refine the origin of the template structure and the energy function used by the methods presented herein. By allowing changes in the backbone structure and in rotamer structures to occur throughout the minimization, global energy minimization reduces small mismatches and small steric hindrances, thereby lowering the total free energy of some template structures by a significant amount.

[0077] In some embodiments, energy minimization may involve repeated rotamer sampling (repacking) followed by side-chain and backbone minimization. An exemplary refinement protocol is provided by Korkegian, A. et al., Science, 2005. In some embodiments, energy minimization may involve more substantial energy minimization in the backbone of the protein.

[0078] As used herein, the terms "rotamer sampling" and "repacking" refer to a particular weight-fitting procedure in which preferred side chain dihedral angles are sampled, as defined in the Rosetta software package. Repacking typically introduces larger structural changes to the weight-fitted structure compared to standard dihedral angle minimization, because the latter samples smaller changes in the structure of a residue, whereas repacking swings the side chain around a dihedral angle so that it occupies a completely different space in the protein structure.

[0079] In some embodiments, the template structure is of a homologous protein, and the query sequence is first threaded into the protein template structure using well-established computational procedures. For example, when using the Rosetta software package, according to some embodiments of the present invention, the first two iterations are performed with a "soft" energy function, where the atomic radii are defined as smaller. The use of smaller radius values ​​reduces the strong repulsive forces, resulting in a smoother energy landscape, allowing the energy barrier to be overcome. The next iteration is performed with the standard Rosetta energy function. A "coordinate constraint" condition can be added to the standard energy function to allow for significant deviations from the original Cα coordinates. The coordinate constraint condition behaves harmonically (Hooke's law) and has a weight ranging from about 0.05 to 0.4 reu (Rosetta energy units), depending on the degree of identity between the query sequence and the sequence of the template structure. During refinement, key residues are only subjected to a small range of minimization, but not to rotamer sampling.

[0080] The structure refinement and energy relaxation steps used in the methods provided herein can be implemented using, for example, “Protein Repair One Stop Shop” or PROSS [Goldenzweig A, Goldsmith M, Hill SE, Gertman O, Laurino P, Ashani Y, Dym O, Unger T, Albeck S, Prilusky J, Lieberman RL, Aharoni A, Silman I, Sussman JL, Tawfik DS, Fleishman SJ., Automated Structure- and Sequence-Based Design of Proteins for High Bacterial Expression and Stability. Mol Cell. 2016 Jul 21;63(2):337-346], and / or FuncLib [Khersonsky O, Lipsh R, Avizemer Z, Ashani Y, Goldsmith M, Leader H, Dym O, Rogotner S, Trudeau DL, Prilusky J, Amengual-Rigo P, Guallar V, Tawfik DS, Fleishman SJ. Automated Design of Efficient and Functionally Diverse Enzyme Repertoires. Mol Cell. 2018 Oct 4;72(1):178-186].

[0081] The same tools used for energy minimization and structural refinement can be used to rank refined structures by their individual port-refinement (final) minimization energy scoring.

[0082] Conclusion: Antibody humanization is a fundamental technique that has been applied to numerous antibodies as a necessary step before their clinical use. However, regardless of its importance, humanization often results in a significant reduction in antibody expression, stability, or specificity. Thirty years ago, Foote and Winter concluded that amino acid positions in the framework (Vernier positions) are critical for the structural and energetic integrity of the Fv and must be considered for successful humanization. This understanding led to the establishment of an iterative rule of thumb in which the first humanized antibody is mutated at the same Vernier positions in the parent antibody to recapitulate the binding or expression properties of the animal antibody.

[0083] Instead of using only a few dozen homology frameworks as in most conventional humanization strategies, the method provided herein uses virtually all combinations of possible human gene segments. Although genes belonging to a single subgroup are similar to each other, they contain mutations, including in Vernier positions, that can stabilize specific CDRs of the parent antibody. Results obtained using the method provided herein show that selecting the lowest energy design from this large space of possible frameworks leads to a large increase in stability, expression, and binding affinity compared to homology-based antibody humanization strategies. Remarkably, CUMAB can start directly from the antibody sequence and use existing software to model the Fv structure. This ability paves the way for humanizing a large set of antibodies obtained by animal immunization without the need for crystallographic analysis. The fundamental insight that antibody stability and activity are determined by both the CDRs and the amino acid positions where the CDRs are located can help address other important challenges in antibody design and develop general and automated design strategies.

[0084] It is anticipated that many related computational methods for humanizing antibodies will be developed during the life of the patent maturing from this application, and the scope of the phrase "computational method for humanizing an antibody" is intended a priori to encompass all such new technologies.

[0085] It will be understood that certain features of the invention that are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the invention that are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable subcombination, or as appropriate in other described embodiments of the invention. Certain features described in the context of various embodiments should not be construed as essential features of those embodiments, unless the embodiment is inoperable without those elements. Various embodiments and aspects of the present invention as described hereinabove and as claimed in the claims section below are found experimentally supported in the following examples.

[0086] Working Example Reference is now made to the following examples, which together with the above descriptions illustrate some non-limiting embodiments of the invention.

[0087] Example 1 Computer-based methods: Database construction of antibody germline sequences: A database of antibody germline sequences was obtained by searching antibody germline sequences from the IMGT reference database 17 (downloaded on July 29, 2020). For each gene, only the first allele annotated as functional was retrieved. Furthermore, genes had to be annotated as neither partial nor reverse complement. If allele 1 contained three or more cysteines, another allele with two cysteines was retrieved if possible. This sorting scheme yielded 54 heavy chain V gene sequences, 6 heavy chain J gene sequences, 39 light chain kappa gene sequences, 5 light chain kappa J gene sequences, 30 lambda V gene sequences, and 5 light chain kappa J gene sequences.

[0088] Subsequent all-to-all recombination of these four segments: V and J for both light and heavy chains, and for kappa and lambda light chains, was performed separately, resulting in 63,180 sequences for kappa light chains and 48,600 sequences for lambda light chains.

[0089] Computational CDR grafting: CDRs were defined based on a combination of definitions provided elsewhere [MacCallum, RM; Martin, AC; Thornton, JM Antibody-Antigen Interactions: Contact Analysis and Binding Site Topography. J. Mol. Biol. 1996, 262(5), 732-745; Dondelinger, M.; Filee, P.; Sauvage, E.; Quinting, B.; Muyldermans, S.; Galleni, M.; Vandevenne, MS Understanding the Significance and Implications of Antibody Numbering and Antigen-Binding Surface / Residue Definition. Front. Immunol. 2018, 9, 2278] and visual inspection of antibody crystal structures. Starting from the published atomic structure of the parent antibody (sourced from PDB), HMMer was used to identify segments of sequence corresponding to the variable regions and classify the light chains as kappa or lambda. For each germline sequence corresponding to the light chain classification, the CDRs of the germline sequence were replaced with the CDRs of the parent antibody. Any sequences containing Asn-Gly or Asn-X-Ser / Thr (where X is not pro) outside the CDRs were removed, resulting in over 20,000 unique sequences per parent antibody.

[0090] Energy ranking of CDR-grafted sequences: If a crystal structure is provided, as a first step the parent antibody crystal structure is relaxed by iterative side-chain harmonically constrained main-chain minimization and combinatorial side-chain packing throughout the Fv (see Relax.xml RosettaScript38). If a bound structure is provided, residues at the interface of the Fv and antigen are identified using Rosetta and held fixed during relaxation (see Interface.xml). If a structure is provided in complex with antigen, the entire antibody Fv-antigen complex is relaxed.

[0091] Each CDR-grafted germline sequence was then threaded into a relaxed Fv structure and relaxed in the same manner with fewer iterations (see Thread_relax.xml). Sequences were ranked according to all atom energies using the ref2015 scoring function21. Any models with a Cα-carbonyl O RMSD of 0.5 Å or more in any of the CDRs were removed from further consideration. Sequences were clustered by V-gene subgroup as defined by IMGT, meaning that only one sequence was obtained from each V-gene combination. Sequences were visually inspected and in some cases the highest-ranked representative of a cluster was replaced with a slightly lower-ranked one for reuse in a different cluster, thus minimizing cloning.

[0092] When starting from a computed (in silico) model rather than an experimental structure, the pipeline was nearly the same, the only difference being that sequences were not removed if they deviated from the model structure due to uncertainties in the modeling.

[0093] Computational SDR porting: The parent antibodies were classified as having kappa or lambda light chains as described above. Rosetta was used to identify residues at the interface between the antibody Fv and the antigen (see interface.xml). Antibody germline sequences were selected using the following criteria: the sequence must have the same CDR length in all CDRs except H3. In addition, the sequence must be the same or shorter than the length of H3 compared to the parent antibody. If the length of H3 is shorter than the parent antibody, a number of residues equal to the difference in length of the two H3s from the parent antibody are inserted into the germline sequence. The germline sequence is then threaded and relaxed as described above. The sequences were clustered according to V gene subgroups and heavy chain J gene subgroups.

[0094] Example 2 Scripts, xml, flag files and command line Initial refinement:

number

number

number

number

[0095] Example 3 The sequence identity between the parent antibody (the first in each block) and the designs to human germline is presented in Table 1 below. [Table 1]

[0096] Table 2 below provides the light and heavy chain sequences of constructs, clones and designs 1-5. [Table 2]

[0097] Example 4 The next example tracks the concentration of dissolved oxygen as a measure of QSOX1 activity.

[0098] A Clark-type oxygen electrode was used to monitor changes in dissolved oxygen concentration as a measure of QSOX1 activity. Antibodies were mixed with QSOX1 and the reaction was initiated by injection of the model substrate dithiothreitol (DTT). QSOX1 and DTT were fixed at concentrations of 25 nM and 200 μM, respectively, and the antibody concentration was varied. The initial slope of the dissolved oxygen concentration was recorded for each antibody concentration. Reactions were performed in duplicate and the results for each antibody concentration were averaged. Relative activity compared to the uninhibited reaction was plotted against antibody concentration and fitted to the Morrison Ki equation for tight binding competitive inhibitors to obtain the inhibition constant (Ki).

number

[0099] Measurements were performed at 25° C. in 50 mM potassium phosphate buffer, pH 7.5, 65 mM NaCl, and 1 mM ethylenediaminetetraacetic acid (EDTA).

[0100] The humanized antibodies tested are summarized in Table 3. [Table 3]

[0101] FIG. 7 shows a comparative bar graph comparing activity (affinity) assay results performed on expressed antibodies 492 (SEQ ID NO: (heavy / light chain): 79 / 80), chimera, h2bK4 (SEQ ID NO: 2 / 8), h3K4 (SEQ ID NO: 3 / 8), h3newK2 (SEQ ID NO: 4 / 6) and h3newK4 (SEQ ID NO: 4 / 8).

[0102] As can be seen in Figure 7, the humanized antibodies (h2bK4, h3K4, h3newK2 and h3newK4; Table 3) have sub-nM Ki values ​​similar to the original murine antibody MAb492 (SEQ ID NO: (heavy / light): 79 / 80).

[0103] Table 4 shows the 15 anti-QSOX1 designs tested in the experiment. [Table 4]

[0104] Example 5 A murine monoclonal antibody inhibiting the enzymatic activity of human quiesin sulfhydryl oxidase 1 (QSOX1), the above-mentioned target for humanization, is an antibody in development as a promising cancer therapeutic due to its ability to block the contribution of QSOX1 to the extracellular matrix support of tumor growth and metastasis (pubmed ID:32064042). This antibody is a tough test for humanization because chimerizing its mouse Fv with the human IgG1 constant region leads to a complete loss of expressibility in HEK293 cells.

[0105] This challenge was previously addressed using the AbLIFT method, which uses atomic-level design computations to improve the intermolecular interactions between Fv light and heavy chain domains [Warszawski, S. et al. Optimizing antibody affinity and stability by the automated design of the variable light-heavy chain interfaces. PLoS Comput. Biol. 15, e1007207 (2019)]. One of the designs, AbLIFT18, rescued the expression in HEK293 cells and QSOX1 inhibition of a chimeric antibody composed of a murine Fv with a human IgG1 constant region. However, this design was an incompletely humanized antibody (66.3% and 57.3% V gene sequence identity to the closest human germline in the light and heavy chains, respectively), meaning that it was not suitable for therapeutic application in humans.

[0106] To simulate a realistic humanization scenario, we started directly with the parental mouse antibody, using AbLIFT18 and the mouse parental antibody as controls. We ordered genes encoding the five top-ranked CUMAb designs formatted as separate light and heavy chains, and experimentally tested all 15 unique pairs of light and heavy chains from the top five designs. Strikingly, 12 pairs showed comparable expression levels to AbLIFT18 in dot blot analysis, while no detectable expression was detected in the chimeric constructs containing mouse Fv and human constant domains. Furthermore, electrophoretic mobility analysis after purification revealed that seven designs showed comparable expression levels to AbLIFT18 without obvious misfolding or aggregation.

[0107] The expressible designs were purified and screened for QSOX1 inhibition. The two most successful designs showed inhibition constants similar to those of the parental antibodies. These designs share the same light chain and show 81.2% V gene sequence identity and 79.4% and 85.7% identity in the heavy chain to the closest human germline gene. These V gene sequence identities are much higher than those of the murine antibody and AbLIFT18, both of which have 66.3% sequence identity (light chain) and 57.3% sequence identity (heavy chain). Furthermore, these V gene sequence identities are in the range between FDA-approved humanized antibodies, which have an average of 84% sequence identity in the light chain and 81% sequence identity in the heavy chain. Thus, these two designs recapitulated the expression levels of the parental mouse using a human Fv framework and without the need for back mutations.

[0108] We measured the melting temperatures of the two designs as well as the parent antibody by nano-differential scanning fluorimetry (nano-DSF) and found that all three were above 70° C. Interestingly, the two designs are derived from different heavy chain V gene subgroups and have 36 mutations between them. Due to these mutations, they have very different patterns of surface charge.

[0109] Thus, CUMAbs generate functionally nearly identical antibodies but with very different surface properties. Because surface properties are associated with changes in the tendency of antibodies to self-associate or form non-specific interactions, it can be very advantageous to have multiple humanized options of antibodies with different surface properties.

[0110] We determined the co-crystal structure of one of the best-performing designs with an oxidoreductase fragment of human QSOX1 and found that the designs and parental antibody were highly similar, with a Cα RMSD between them of only 0.75 Å, despite 51 mutations between the parent and humanized antibodies. These results demonstrate the atomic accuracy of CUMAb and its ability to rapidly generate functionally similar yet more stable humanized designs, even when previous humanization efforts have failed.

[0111] Table 5 shows the sequence identity between the parent antibody (first in each block) and the design to human germline. Percent identity was calculated using IgBLAST. [Table 5]

[0112] The sequences reported in this example, such as anti-QSOX1 heavy chain H3 (hαQSOX1.1) and anti-QSOX1 heavy chain H2b (hαQSOX1.2), are listed in Table 6 below.

[0113] Example 6 Below is a list of antibodies related to the method of designing and making humanized antibodies (CUMAbs) according to an exemplary embodiment of the present invention. [Table 6] TIFF2024533982000017.tif196159TIFF2024533982000018.tif196159TIFF2024533982000019.tif198159 TIFF2024533982000020.tif197159TIFF2024533982000021.tif196159TIFF2024533982000022.tif196159 TIFF2024533982000023.tif196159TIFF2024533982000024.tif196159TIFF2024533982000025.tif196159 TIFF2024533982000026.tif196159TIFF2024533982000027.tif196159TIFF2024533982000028.tif124159

[0114] While the present invention has been described in conjunction with specific embodiments thereof, it is evident that many alternatives, modifications, and variations will be apparent to those skilled in the art. Accordingly, the present invention is intended to embrace all such alternatives, modifications, and variations that fall within the spirit and broad scope of the appended claims.

[0115] All publications, patents, and patent applications mentioned herein are incorporated herein by reference in their entirety to the same extent as if each individual publication, patent, and patent application was specifically and individually indicated to be incorporated by reference. Furthermore, citation or identification of any document in this application should not be construed as an admission that such document is available as prior art to the present invention. Section headings should not be construed as necessarily limiting to the extent that they are used. Additionally, all priority documents of this application are incorporated herein by reference in their entirety.

Claims

1. A method for producing a humanized antibody having affinity for a target antigen, comprising: i) providing a structural model of a non-human antibody having affinity for the target antigen, and identifying the amino acid residues of at least one complementarity determining region (CDR) in the structural model; ii) generating all combinations of antibody segments from a plurality of human antibody germline sequences, and substituting the corresponding amino acid residues in each of the combinations with the amino acid residues of the CDR, thereby obtaining a library of transplanted human antibody sequences; iii) threading each of the transplanted human antibody sequences onto the structural model to thereby obtain a plurality of threaded transplanted human antibody structures, and subjecting each of the threaded transplanted human antibody structures to constrained energy minimization (constrained structural relaxation) to thereby obtain a plurality of relaxed transplanted human antibody structures; iv) ranking the plurality of relaxed transplanted human antibody structures by an energy score; v) clustering the plurality of relaxed transplanted human antibody structures according to V / J gene families to thereby obtain an energy-ranked and gene family-clustered library of humanized antibody designs; and vi) expressing at least one humanized antibody design from at least one cluster of humanized antibody designs, and selecting at least one humanized antibody design having affinity for the target antigen, thereby obtaining a humanized antibody having affinity for the target antigen. A method comprising the above steps.

2. The method according to claim 1, further comprising subjecting the structural model to energy minimization (constrained structural relaxation) before said threading.

3. The method according to claim 1, wherein the antibody segment is selected from the group consisting of a heavy chain variable (V) V gene segment, a light chain variable (V) gene segment, a heavy chain joining (J) gene segment, a light chain joining (J) gene segment, a kappa gene segment, and a lambda gene segment.

4. The method according to any one of claims 1 to 3, further comprising removing sequences showing three or more cysteines outside the CDRs from the library of transplanted human antibody sequences.

5. The method according to any one of claims 1 to 3, further comprising removing sequences showing an Asn-Gly or Asn-X-Ser-Thr (X is not Pro) motif from the library of transplanted human antibody sequences.

6. The method according to any one of claims 1 to 3, further comprising removing structures showing an RMSD exceeding 0.5 Å in the backbone atoms of the CDRs compared to the structural model of the non-human antibody from the plurality of relaxed transplanted human antibody structures.

7. The method according to any one of claims 1 to 3, wherein the plurality of human antibody germline sequences can be obtained from a human gene database.

8. The method according to claim 7, wherein the human gene database is the IMGT database of immunogenetics and immunoinformatics.

9. The method according to any one of claims 1 to 3, wherein the non-human antibody is a mouse antibody.

Images