JP2010088451A - Generation and selection of protein library in silico - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a methodology for efficiently generating and screening protein libraries for optimized proteins with desirable biological functions (such as improved binding affinity towards biologically and/or therapeutically important target molecules). <P>SOLUTION: The method for constructing a library of desired proteins comprises steps of: comparing a lead sequence with a plurality of tester protein sequences; selecting at least two peptide segments that have at least 15% sequence identity with the lead sequence from the plurality of tester protein sequences (the selected peptide segment forms a hit library); and forming the library of designed proteins by substituting the lead sequence with the hit library. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

Background of the Invention
CROSS-REFERENCE TO RELATED APPLICATIONS This application U.S. Patent Application No. 10/153 and 159 (May 20, 2002 application, entitled "structure-based selection and affinity maturation of an antibody library") is a continuation-in-part application of, This is a continuation-in-part of patent application No. 10 / 153,176 (filed on May 20, 2002, titled “In silico (Computer-Based) Generation of Antibody Library and Affinity Maturation”). Patent application No. 10 / 125,687 (filed Apr. 17, 2002, continuation-in-part of the title “Structure-Based Construction of Human Antibody Library”, which is also a provisional application of US Provisional Application No. 60 / 284,407 ( Claims the benefit of the application entitled 17 April 2001, entitled “Structure-Based Construction of Human Antibody Libraries”, which are hereby incorporated by reference.

FIELD OF THE INVENTION The present invention relates generally to computer-aided design of proteins with binding affinity for target molecules, and more particularly to diverse sequences by combining computer-predicted and experimental screening of antibody-biased libraries. And an antibody (or immunoglobulin) screening method and identification method having high affinity for a target antigen.

Description of Related Art Antibodies are made in vertebrates in response to various internal and external stimuli (antigens). Antibodies are synthesized only by B cells, producing millions of types, each with a different amino acid sequence and a different binding site for an antigen. Collectively these are called immunoglobulins (abbreviated as Ig) and are one of the most abundant protein components in the blood and constitute 20% by weight of total plasma protein.

  Naturally occurring antibody molecules are composed of two identical “light” (L) protein chains and two identical “heavy” (H) protein chains, all linked by hydrogen sulfide and precisely located disulfide bonds. Yes. Chothia et al. (1985) J. Mol. Biol. 186: 651-663; and Novotny and Haber (1985) Proc. Natl. Acad. Sci. USA 82: 4592-4596. Together, the N-terminal domains of the L and H chains form the antigen recognition site for each antibody.

  The mammalian immune system has evolved a unique genetic mechanism that combines almost individual countless different light and heavy chains together by joining separate gene segments before they are transcribed. Allows to generate. For each type of Ig chain (kappa light chain, lambda light chain, and heavy chain), there is a separate pool of gene segments where a single peptide chain is ultimately synthesized. Each pool is on a different chromosome and usually contains a large number of gene segments encoding the V region of the Ig chain and a small number of gene segments encoding the C region. During B cell growth, the complete coding sequence for each of the two Ig chains synthesized is assembled by site-specific genetic recombination to bring together the entire coding sequence of the V region and the coding sequence of the C region. To do. In addition, the V region of the light chain is encoded by a DNA sequence assembled from two gene segments (V gene segment and short junction or J gene segment). The V region of the heavy chain is encoded by a DNA sequence assembled from three gene segments (V gene segment, J gene segment, and diversity or D segment).

  The large number of inherited V, J and D gene segments available to encode the Ig chain itself contributes greatly to antibody diversity, but the combined binding of these segments greatly increases this contribution. Let Furthermore, inaccurate binding of gene segments and somatic mutations introduced during VDJ segment binding at the pre-B cell stage greatly increase V region diversity.

  After immunization against an antigen, the mammal undergoes a process known as affinity maturation to produce antibodies with high affinity for the antigen. Such antigen-directed somatic hypermutations are probably due to the accumulation of point mutations specifically in the heavy and light chain V region coding sequences and due to the selective expansion of B cell clones with high affinity antibodies. Fine tune the antibody response to an antigen.

Structurally, the various functions of antibodies are limited to discontinuous protein domains (regions). The site that recognizes and binds to the antigen consists of three hypervariable regions or complementarity determining regions (CDRs) within the variable (V _H and V _L ) regions at the N-terminus of the two heavy chains and the two light chains. . Constant domains are not directly involved in antibody binding to antigen, but are involved in various effector functions (eg, antibody participation in antibody-dependent cytotoxicity).

  Natural light and heavy chain domains have the same general structure, each domain has four framework regions, the sequences of which are linked by three CDRs and are conserved to some extent. The four framework regions mainly take a β-sheet conformation, and the CDR connects β-sheet structures, and in some cases forms a loop that forms part of it. The CDRs of each chain are held in the vicinity by the framework region and contribute to the formation of an antigen binding site together with the CDRs of other chains.

In general, all antibodies adopt a characteristic “immunoglobulin fold”. Specifically, both the variable and constant domains of the antigen-binding fragment (Fab, consisting of V _L and C _L of light chain and V _H and C _H 1 of light chain) are two twisted opposite parallel β It consists of a sheet, which constitutes a β sandwich structure. The constant region has triple and quadruple β-sheets arranged in Greek key-like motifs, and the variable region further has two short β-strands to create a 5-fold β-sheet.

The V _L and V _H domains interact through a five-fold β sheet to form a nine-fold β cylinder with a radius of about 8.4 mm, and the domain boundary chains are tilted about 50 ° from each other. Domain pairing brings a CDR loop in the vicinity. The CDR itself forms approximately 25% of the V _L / V _H domain boundary.

  Six CDRs (CDR-L1, -L2 and -L3 for the light chain and CDR-H1, -H2 and -H3 for the heavy chain) are supported on the β cylindrical framework to form the antigen binding site. Its sequence is hypervariable compared to the rest of the immunoglobulin structure, but some of the loops show a relatively high degree of sequence and structural conservation. In particular, CDR-L2 and CDR-H1 are highly conserved in the conformation.

  Chothia and co-workers found that a limited number of main-chain conformations (CDRs) in which five of the six CDR loops (all except CDR-H3) were separated by analysis of the conserved key residues. It is proved to be called standard structure). Chotia and Lesk (1987) J. Mol. Biol. 196: 901-917; Chothia et al. (1989) Nature (London) 342: 877; and Chothia et al. (1998) J. Mol. Biol. 278: 457-479. The structure taken depends on both the CDR length and the body of some major amino acid residues, both in the framework that contacts the CDR involved in the packing. The standard conformation is determined by specific packing of only these major residues that act as structure determinants, hydrogen bonding interactions, and stereochemical constraints.

  Various methods have been developed to model the three-dimensional structure of the antigen binding site of an antibody. In addition to X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy has been used with computer model construction to study atomic details of antibody-ligand interactions. Dwek et al. (1975) Eur. J. Biochem. 53: 25-39. Dwek and co-workers used a spin-label hapten to estimate the binding site of the MoPC315 myeloma protein for dinitrophenyl. Similar analysis using an anti-spin-labeled monoclonal antibody (Anglister et al. (1987) Biochem. 26: 6958-6064) and an anti-2-phenyloxazolone Fv fragment (McManus and Riechmann (1991) Biochem. 30: 5851-5857) Has been done.

  Computational analysis and modeling of antibody binding sites (or antigen binding sites) is based on the target antibody sequences and known structures or structural motifs in existing databases (eg, Brookhaven Protein Data Bank). Based on homology analysis comparing antibodies with Using such homology-based modeling methods, an approximate three-dimensional structure of the target antibody is constructed. Early antibody modeling was based on the assumption that CDR loops with the same length and different sequences may adopt similar conformations. Kabat and Wu (1972) Proc. Natl. Acad. Sci. USA 69: 960-964. A typical segment matching algorithm is as follows: when there is a loop sequence, search the Protein Data Bank for a short homologous backbone fragment (eg, a tripeptide) and then assemble it Create a new binding site model on your computer.

  More recently, standard loop concepts have been incorporated into computer-aided structural modeling of antibody binding sites. For many common types, the standard structure concept is (1) Sequence changes other than the standard position are independent of loop conformation, (2) Standard loop conformation is essentially independent of loop-loop interaction And (3) the number of standard motifs present is limited and these are well documented in the currently known antibody crystal conformation database. Based on this concept, Chothia predicted that all six CDR loop conformations were in the lysozyme binding antibody D1.3 and that five standard loop conformations were in the other four antibodies. Chothia (1989), supra. It is also possible to combine homology-based modeling with conformational search methods to improve CDR modeling of antibody structures. Martin, A.C.R. (1989) PNAS 86: 9268-72.

In addition to modeling specific antibody structures, attempts have been made to create artificial (or synthetic) libraries of antibodies and screen them against specific target antigens. A fully synthetic combinatorial antibody library has been designed based on a modular consensus framework and trinucleotide randomized CDRs. Knappik et al. (2000) J. Mol. Biol. 296: 57-86. In this study, the human antibody repertoire was analyzed in terms of structure, amino acid sequence diversity, and germline usage. Modular consensus framework sequences with 7 V _H and 7 V _L covering 95% of various germline families were derived and optimized for E. coli. After all 49 combinations of genes were cloned into the phagemid vector, a set of antibody phage display libraries were created that would be 2 × 10 ⁹ members in the library.

  Phage display technology is widely used to create large libraries of antibody fragments, taking advantage of the ability of bacteriophages to express and display biologically functional protein molecules on their surface. Yes. Combinatorial libraries of antibodies have been generated in a bacteriophage lambda expression system, which are screened as bacteriophage plaques or as lysogenic colonies (Huse et al. (1989) Science 246: 1275; Caton and Koprowski (1990) Proc. Natl. Acad. Sci. USA 87: 6450; Mullinax et al. (1990) Proc. Natl. Acad. Sci. USA 87: 8095; Persson et al. (1991) Proc. Natl. Acad. Sci. USA 2432). Various examples of bacteriophage antibody display libraries and lambda phage expression libraries have been described (Kang et al. (1991) Proc. Natl. Acad. Sci. USA 88: 4363; Clackson et al. (1991) Nature 352: 624; McCafferty et al. (1990) Nature 348: 552; Burton et al. (1991) Proc. Natl. Acad. Sci. USA 88: 10134; Hooggenboom et al. (1991) Nucleic Acids Res. 19: 4133; Chang et al. (1991) J. Immunol. 147: 3610; Britling et al. (1991) Gene 104: 147; Marks et al. (1991) J. Mol. Biol. 222: 581; Barbas et al. (1992) Proc. Natl. Acad. Sci. USA 89: 4457; Hawkins and Winter. (1992) J. Immunol. 22: 867; Marks et al. (1992) Biotechnology 10: 779; Marks et al. (1992) J. Biol. Chem. 267: 16007; Lowman et al. (1991) Biochemistry 30: 10832; Lerner et al. (1992). ) Science 258: 1313). See also the review by Rader, C. and Barbas, C.F. (1997) “Phage display of combinatorial antibody libraries”, Curr. Opin. Biotechnol. 8: 503-508.

Generally phage library, a library or cDNA library of random oligonucleotides encoding antibody fragments such as V _L and V _H, are prepared by inserting the gene 3 of the M13 or fd phage. Each inserted gene is expressed at the N-terminus of the gene 3 product (small phage coat protein). As a result, a peptide library containing various peptides can be constructed. The phage library is then affinity screened for the target molecule of interest (eg, antigen), and the specifically bound phage particles are recovered and infected by infection into Escherichia coli host cells. Amplified. Typically, a target molecule of interest such as a receptor (eg, a polypeptide, carbohydrate, glycoprotein, nucleic acid) is immobilized to a chromatographic resin by covalent bonds, and reactive phage particles are obtained by affinity chromatography. Concentrated and / or labeled for plaque or colony lift screening. This method is called biopanning. Finally, high affinity phage colonies can be amplified and sequenced for estimation of specific peptide sequences.

  A method for humanizing antibodies using computer modeling has been developed by Queen et al. US Patent No. 5,693,762. The structure of the non-human donor antibody (eg mouse monoclonal antibody) is predicted based on computer modeling, and the major amino acids in the framework are necessary to retain the shape and thus retain the binding specificity of the CDR It is predicted that. These few major mouse donor amino acids are selected based on position and nature within certain defined categories and are substituted into the human acceptor antibody framework along with the donor CDRs. For example, category 1: amino acid positions are in the CDRs as defined by Kabat et al. Kabat and Wu (1972) Proc. Natl. Acad. Sci. USA 69: 960-964. Category 2: If the amino acid in the framework of the human acceptor immunoglobulin is not common, and if the donor amino acid at that position is typical of a human sequence, then the donor amino acid is selected rather than the acceptor. Category 3: A donor amino acid is selected instead of an acceptor amino acid at a position adjacent to one or more of the three CDRs in the primary sequence of a humanized immunoglobulin chain. Based on these criteria, a thorough selection of individual amino acids from the donor antibody is performed. The resulting humanized antibody typically contains about 90% human sequences. Humanized antibodies designed by computer modeling are tested for antigen binding. Experimental results such as binding affinity are fed back into the computer modeling process to fine-tune the structure of the humanized antibody. The redesigned antibody can then be tested for improved biological function. Such an iterative fine tuning process is laborious and unpredictable.

SUMMARY OF THE INVENTION The present invention makes protein libraries efficient for optimized proteins with favorable biological functions (eg, improved binding affinity for biologically and / or therapeutically important target molecules). Provide innovative ways to create and screen creatively. This method is computationally fast by examining an ever-expanding database of protein sequences of all living organisms (especially humans). Protein evolution data is used to expand both the structure and the structural space of protein libraries for functional screening in vitro or in vivo. By using the method of the present invention, an extended but functionally biased library of proteins such as antibodies based on in silico computer evaluation of structures that are functionally related to a wide variety of protein sequences Can be built.

  In one embodiment of the invention, a method for designing and selecting a protein having a preferred function is provided. This method is preferably performed on a computer by in silico selection of protein sequences based on the amino acid sequence of the target structure / function motif or domain in the lead protein (hereinafter referred to as “lead sequence”). The lead sequence is used to search a database of protein sequences. The choice of database depends on the specific functional requirements of the desired motif. For example, if the lead protein is an enzyme and the target motif includes the active site of the enzyme, a database of proteins / peptides of a particular origin, organism, species, or combinations thereof can be searched using various search criteria to A list of hits, each of which can replace a target motif in the lead protein. A similar approach is used to design other motifs or domains of the lead protein. The designed sequences of individual motifs / domains are combined to create a library of designed proteins. In order to further reduce the immunogenicity of the engineered protein for human applications (eg therapeutics or diagnostics), preferably a database of proteins of human origin or humanized proteins is searched, especially structurally Alternatively, a hit list of sequences is obtained for motifs obtained from sites of the lead protein that are not functionally critical. The library of designed proteins can be tested experimentally to obtain proteins with improved biological function relative to the lead protein.

In certain embodiments, the method includes the following steps:
Providing an amino acid sequence obtained from the lead protein (the amino acid sequence is referred to as the lead sequence);
Comparing the lead sequence with a plurality of tester protein sequences; and selecting from the plurality of tester protein sequences at least two peptide segments having at least 15% sequence identity with the lead sequence (the selected peptide segments are hit libraries) And the lead sequence is replaced with a hit library to form a library of designed proteins.

Optionally, the method further comprises the following steps:
Create an amino acid position variant profile of the hit library;
The amino acid variants in the hit library are combined to generate a combination of hit variants that forms a hit variant library; and a protein with a preferred function is selected from the hit variant library.
Optionally, the method further comprises the following steps:
A scoring function is used to determine whether the members of the hit library or hit variant library are structurally compatible with the three-dimensional structure of the lead sequence or lead protein; Choose members that are equal or better.
Optionally, the method further comprises the following steps:
Constructing a hit library, a hit variant library or a nucleic acid library comprising a DNA segment encoding the amino acid sequence of a member selected based on the structural evaluation described above;
A nucleic acid library is expressed to create a library of recombinant proteins; and a protein having the desired function is selected from the library of recombinant proteins.
Optionally, the method further comprises the following steps:
Create an amino acid position variant profile of the hit library;
Converting the amino acid position variant profile of the hit library into a nucleic acid position variant profile by back-translating the amino acid position variant into the corresponding gene codon;
Combining nucleic acid position variants in a combinatorial manner to construct a degenerate nucleic acid library of DNA segments;
A degenerate nucleic acid library is expressed to create a library of recombinant proteins; and a protein having a desired function is selected from the library of recombinant proteins.

Optionally, gene codons are suitable for expression in cells of a particular organism (eg, mammalian cell, insect, plant, yeast, or bacterium). In some cases, the gene codon is within a diversity range that the diversity of the degenerate nucleic acid library of the DNA segment can be experimentally covered without undue experimental effort (eg, less than 1 × 10 ⁷ , preferably 1 × 10 ^{7 The} selected size can be reduced to be less than ⁶ ).

  A lead protein is a protein that preferably improves or modifies its function (preferably biological function) in vitro or in vivo. The lead protein may be a full-length protein, oligopeptide or peptide, or may be a non-natural protein or peptide. Optionally, the lead protein may also be a known protein fragment or domain, including, but not limited to, structural and / or functional domains such as enzymatic domains, binding domains, and smaller fragments or motifs such as turns, helices and loops. Good. In addition, protein variants, ie non-naturally occurring protein analog structures, may be used.

  The lead protein is preferably a protein used in the industry as a therapeutic and / or diagnostic agent. Lead protein types may be ligands, cell surface receptors, antigens, antibodies, cytokines, hormones, transcription factors, signaling molecules, cytoskeletal proteins and enzymes.

  Specific enzyme types include, but are not limited to, hydrolases such as proteases, carbohydrases, lipases; isomerases such as racemases, epimerases, or tautomerases; transferases, kinases, oxidoreductases, and phosphatases. Specific examples of enzymes are listed in the Swiss-Prot enzyme database.

  Other examples of lead protein cytokines include, but are not limited to, IL-1, IL-2, IL-3, IL-4, IL-5, IL-6, IL-8, IL-10, IFN-β , IFN-γ, IFN-α-2a; IFN α-2b, TNF-α; CD40 ligand (chk), human obesity protein liptin, granulocyte-macrophage colony-stimulating factor (GMCSF), bone morphogenetic protein-7, ciliary body Neurotrophic factor, granulocyte macrophage colony stimulating factor, monocyte chemoattractant protein 1, macrophage migration inhibitory factor, human glycosylation inhibitor, human Lantes, human macrophage inflammatory protein 1 beta, human growth hormone, leukemia inhibitory factor, Human melanoma growth stimulating activity, neutrophil activating peptide-2, Cc-chemokine Mcp-3, platelet factor M2, neutrophil activating peptide 2, eotaxin, stromal cell-derived factor-1, insulin, insulin-like proliferation Factor I, insulin-like growth factor II , Transforming growth factor B1, transforming growth factor B2, transforming growth factor B3, transforming growth factor A, vascular endothelial growth factor (VEGF), acidic fibroblast growth factor, basic fibroblast growth factor, endothelial growth Factor, nerve growth factor, brain-derived neurotrophic factor, ciliary neurotrophic factor, platelet-derived growth factor, human hepatocyte growth factor, glial cell-derived neurotrophic factor, erythropoietin; coagulation factor, although not particularly limited, TPA and Factor VIIa; receptor, but not particularly limited, extracellular region of human tissue factor, cytokine binding region of Gp130, G-CSF receptor, erythropoietin receptor, fibroblast growth factor receptor, TNF receptor, IL-1 Receptor, IL-1 receptor / IL1ra complex, IL4 receptor, receptor α chain, MHC class I, MHC class II, T cell receptor, insulin receptor, insulin There are phosphorus receptor tyrosine kinases and human growth hormone receptors.

  In yet another aspect of the invention, methods are provided for in silico design and selection of protein sequences based on lead structure templates. A set of different sequences with a structure that is substantially similar to the structural template is used as a lead sequence to search the protein structure database for distant homologues of the lead sequence that have low sequence identity but similar structure. The Using this method, a library of diverse protein sequences is constructed and screened experimentally in vitro or in vivo for protein variants with improved or desired functions.

In a specific embodiment of the invention, the methods of the invention are used to design antibodies that are diverse in sequence and are functionally related to each other. Based on the designed antibody sequence, a library of antibodies containing various sequences in the complementarity determining region (CDR) and / or humanized framework (FR) of a non-human antibody can be constructed at high speed. . This antibody library can be screened against a wide range of target molecules for new or improved functions.
In yet another aspect of the invention, a method is provided for in silico selection of an antibody sequence (hereinafter referred to as “lead sequence”) based on the amino acid sequence of a region in the lead antibody. The lead sequence is used to search a database of protein sequences. The choice of database depends on the specific functional requirements of the designed motif. For example, to design variable chain framework regions for therapeutic applications, except for some structurally critical sites, fully human immunoglobulin sequences and human germline immunity A population of evolutionarily related protein sequences such as globulin sequences should be used. This will reduce the immune response while preserving the origin of the sequence by introducing as few foreign variants as possible into this highly conserved region (for the framework region). On the other hand, to improve the binding affinity for antigens in this highly variable region, a variety of databases (eg, various species of immunoglobulin sequences or unrelated sequences in GenBank) are used. CDR can be designed. Using this method, a library of diverse antibody sequences can be constructed and experimentally screened in vitro or in vivo for antibody variants with improved or desired functions.

In certain embodiments, the method includes the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to a plurality of tester protein sequences; and selecting from the plurality of tester protein sequences at least two peptide segments having at least 15% sequence identity with the lead sequence (the selected peptide segment is a hit library Form).

The method further includes the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed.

Optionally, the present invention further comprises the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into corresponding gene codons; and combinatorially combining nucleic acid position variants into a degenerate nucleic acid library of DNA segments Build up.

In some cases, gene codons are suitable for expression in bacteria. In some cases, the gene codons are within a diversity range that allows the experiment to cover the diversity of a degenerate nucleic acid library of DNA segments without undue experimental effort (eg, less than 1 × 10 ⁷ , preferably 1 × 10 ^{7 The} selected size can be reduced to be less than ⁶ ).

In another embodiment, the method comprises the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identify the amino acid sequence in the CDR and FR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing a first amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a CDR lead sequence);
Comparing the CDR lead sequence with a plurality of tester protein sequences; and selecting from the plurality of CDR tester protein sequences at least two peptide segments having at least 15% sequence identity with the CDR lead sequence (the selected peptide segments are Forming a CDR hit library);
Select one FR in the _VH or _VL region of the lead antibody;
Providing a second amino acid sequence comprising at least three consecutive amino acid residues in the selected FR (the selected amino acid sequence is the FR lead sequence);
Comparing the FR lead sequence to a plurality of FR tester protein sequences; and selecting from the plurality of FR tester protein sequences at least two peptide segments having at least 15% sequence identity with the FR lead sequence (selected peptide segments Forms an FR hit library); and
A hit library is formed by combining the CDR hit library and the FR hit library.

In this method, the plurality of CDR tester protein sequences may comprise the amino acid sequence of a human or non-human antibody.
Also in the present invention, the plurality of FR tester protein sequences are amino acid sequences of human origin, preferably human or humanized antibodies (eg, at least 50% human sequences, preferably at least 70% human sequences in _VH or _VL). More preferably at least 90% human sequences, and most preferably at least 95% human sequences), more preferably fully human antibodies, and most preferably human germline antibodies.

  In the present invention, at least one of the plurality of CDR tester protein sequences is different from the plurality of FR tester protein sequences.

  Also in the present invention, the plurality of CDR tester protein sequences are human or non-human antibody sequences, and the plurality of FR tester protein sequences are human antibody sequences, preferably human germline antibody sequences.

The method further includes the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed.

Optionally, the method further comprises the following steps:
Create an amino acid position variant profile of the CDR hit library;
Converting the amino acid position variant profile of the CDR hit library into a nucleic acid position variant profile by back-translating the amino acid position variants into the corresponding gene codons; and combing the nucleic acid position variants in a combinatorial manner Build a rally.

In some cases, the gene codon is suitable for expression in bacteria. In some cases, the gene codon is within a diversity range that the diversity of the degenerate nucleic acid library of the DNA segment can be experimentally covered without undue experimental effort (eg, less than 1 × 10 ⁷ , preferably 1 × 10 ^{7 The} selected size can be reduced to be less than ⁶ ).

In another embodiment, the method comprises the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the FR of the lead antibody;
Select one FR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in the selected FR (the selected amino acid sequence is the first FR lead sequence);
Comparing the first lead FR sequence to a plurality of FR tester protein sequences; and selecting at least two peptide segments from the plurality of FR tester protein sequences having at least 15% sequence identity with the first FR lead sequence (Selected peptide segments form an FR hit library).

The method further includes the following steps:
Providing a second amino acid sequence comprising at least three consecutive amino acid residues in a FR different from the selected FR (the selected amino acid sequence is the second FR lead sequence);
Comparing the second FR lead sequence with a plurality of FR tester protein sequences; and selecting from the plurality of FR tester protein sequences at least two peptide segments having at least 15% sequence identity with the second FR lead sequence (The selected peptide segment forms a second FR hit library); and the first FR hit library and the second FR hit library are combined to form a hit library.

In the present invention, the lead CDR sequence comprises at least 5 consecutive amino acid residues in the selected CDR. The selected CDR is selected from the group consisting of V _H CDR1, V _H CDR2, V _H CDR3, V _L CDR1, V _L CDR2, and V _L CDR3 of the lead antibody.

In the present invention, the lead FR sequence also contains at least 5 consecutive amino acid residues in the selected CDR. The selected FR is selected from the group consisting of the lead antibodies V _H FR1, V _H FR2, V _H FR3, V _H FR4, V _L FR1, V _L FR2, V _L FR3, and V _L FR4.

The method further includes the following steps:
A degenerate nucleic acid library containing a DNA segment encoding the amino acid sequence of the hit library is constructed.
In another aspect of the invention, a method is provided for in silico selection of antibody sequences based on the amino acid sequence of a region in the lead antibody (ie, “lead sequence”). The structure of the lead sequence is used to search a database of protein structures for segments with similar 3D structures. These segments are aligned to provide a sequence profile (hereinafter referred to as the “lead sequence profile”). The lead sequence profile is used to search a database of protein sequences of distant homologues of a low lead sequence identity but similar structure. Using this method, a library of diverse antibody sequences can be constructed and experimentally screened in vitro or in vivo for antibody variants with improved or desired functions.

In certain embodiments, the method includes the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Providing a three-dimensional structure of the lead sequence;
Creating a lead sequence profile based on the structure of the lead sequence;
Comparing the lead sequence profile to a plurality of tester protein sequences; and selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments are hit live Forming a rally).

  In the present invention, the three-dimensional structure of the lead array may be a structure obtained from X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy, or theoretical structure modeling.

In the present invention, the step of creating a lead sequence profile includes the following:
Comparing the structure of the lead sequence to the structure of multiple tester protein segments;
Determining the root mean square of the difference in the main chain conformation of the lead sequence and the tester protein segment;
Selecting a tester protein segment having a root mean square difference root mean square of less than 5, preferably less than 4, more preferably less than 3, and most preferably less than 2; and the amino acids of the selected tester protein segment The sequence is aligned with the lead sequence to create a lead sequence profile.

Optionally, the structure of the plurality of tester protein segments is retrieved from the protein data bank.
Optionally, the step of creating a lead sequence profile includes:
Comparing the structure of the lead sequence to the structure of multiple tester protein segments;
Determining the Z-score of the main chain conformation of the lead sequence and tester protein segment;
Selecting a segment of the tester protein segment with a Z score greater than 2, preferably greater than 3, more preferably greater than 4, and most preferably greater than 5; and the amino acid sequence of the selected tester protein segment as the lead sequence Align to create a lead sequence profile.

Optionally, the step of creating a lead sequence profile is performed by an algorithm selected from the group consisting of CE, MAPS, Monte Carlo, and 3D clustering algorithm.
The method further includes the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed.

Optionally, the method further comprises the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into the corresponding trinucleotide codons; and combining nucleic acid position variants in a combinatorial manner Build a rally.

Any of the above methods further includes the following steps:
Introducing a DNA segment of a nucleic acid or degenerate nucleic acid library into a host organism cell;
Expressing a DNA segment in the host cell such that a recombinant antibody containing the amino acid sequence of the hit library is produced in the host organism cell; and
Recombinant antibodies that bind to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ , preferably 10 ⁷ M ⁻¹ , more preferably 10 ⁸ M ⁻¹ , and most preferably 10 ⁹ M ⁻¹ are selected.

  In yet another embodiment of the invention, a method is provided for in silico selection of antibody sequences based on the 3D structure of a lead antibody. The sequence profile from the lead sequence or the specific region of the lead antibody used is used to search the protein sequence database for distant homologues of the lead sequence with low sequence identity but similar structure. These distant homologues form a hit library. The sequences in the hit library are evaluated for structural compatibility with the 3D structure of the lead antibody (hereinafter referred to as the “lead structure template”). A sequence in the hit library that is structurally compatible with the lead structural template is selected and screened experimentally in vitro or in vivo for improved or desired antibody variants.

In certain embodiments, the method includes the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure defined as a lead structural template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence profile to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
A scoring function is used to determine whether the hit library members are structurally compatible with the lead structural template; and the hit library members with scores equal to or better than the lead sequence Select.

  In the present invention, the scoring function is an energy score selected from the group consisting of electrostatic interaction, van der Waals interaction, electrostatic solvation energy, solvent accessible surface solvation energy, and conformational entropy. Function.

  In some cases, the scoring functions are Amber Forcefield (forcefiled), Charmm Forcefield, Discover cvff Forcefield, ECEPP Forcefield, GROMOS Forcefield, OPLS Forcefield, MMFF94 Forcefield, Tripos Forcefield, MM3 Forcefield, Dreiding Forcefield , And UNRES force fields, and other knowledge based statistical force fields (mean fields) and force fields selected from the group consisting of structure based thermodynamic potential functions.

In the present invention, the step of selecting members of the hit library includes:
ΔE _total = E _vdw + E _bond + E _angel + E _{electrostatics} + E _solvation
Select members of the hit library that have a total energy that is lower or equivalent to the lead sequence calculated based on the formula:

In the present invention, the step of selecting members of the hit library includes an improved scoring function ΔG _b = ΔG _MM + ΔG _sol −TΔS _SS
(Where
ΔG _MM = ΔG _ele + ΔG _vdw (1)
ΔG _sol = ΔG _ele-sol + ΔG _ASA (2))
Is used to select members of the hit library that have a binding free energy less than the lead sequence calculated as the difference between bound and unbound states.

The method further includes the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed.
Optionally, the method further comprises the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into the corresponding trinucleotide codons; and combining nucleic acid position variants in a combinatorial manner Build a rally.

  In yet another embodiment of the present invention, there is provided a method for in silico selection of antibody sequences based on a 3D structure or a structural assembly of lead antibodies, or a structural assembly of a plurality of antibodies (hereinafter collectively referred to as lead structure templates). Provided. A lead sequence or sequence profile from a specific region of a lead antibody used to search a database of protein sequences of distant homologues of a lead sequence with low sequence identity but similar structure. These distant homologues form a hit library. The amino acid position variant profile (AA-PVP) of the hit library is constructed based on the frequency of amino acid variants that appear at each position of the lead sequence. Based on AA-PVP, a hit variant library is constructed by combinatorially combining amino acid variants at each position of the read sequence with or without low-frequency variants. The sequences in the hit variant library are subjected to evaluation for structural compatibility with the lead structural template. Sequences in the hit library that are structurally compatible with the lead structural template are selected and screened experimentally in vitro or in vivo for antibody variants with improved or desired function.

In certain embodiments, the method includes the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known 3D structure defined as a lead structure template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR of the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
Creating an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
Combining the amino acid variants in the hit library to create a combination of hit variants that forms a hit variant library;
A scoring function is used to determine whether members of the hit variant library are structurally compatible with the lead structural template; and a hit variant library with a score equal to or better than the lead sequence Select members.

In this method, combining the amino acid variants in the hit library includes the following:
Select amino acid variants with an appearance frequency greater than 4, preferably greater than 6, more preferably greater than 8, and most preferably greater than 10 (2% to 10% of the cutoff frequency, preferably Hit variants that combine the selected amino acid variants in the hit library to form a hit variant library, including 5% if lost after cut-off, including some amino acids from the lead sequence); Generate a combination of

In the present method, the scoring function is an energy score selected from the group consisting of electrostatic interaction, van der Waals interaction, electrostatic solvation energy, solvent accessible surface solvation energy, and conformational entropy. Function.
In some cases, the scoring functions are Amber forcefield (forcefiled), Charmm forcefield, Discover cvff forcefield, ECEPP forcefield, GROMOS forcefield, OPLS forcefield, MMFF94 forcefield, Tripos forcefield, MM3 forcefield, Dreiding forcefield , And UNRES force fields, and other knowledge based statistical force fields (mean fields) and force fields selected from the group consisting of structure based thermodynamic potential functions.

The method further includes the following steps:
A nucleic acid library is constructed that includes DNA segments that encode the amino acid sequences of selected members of the hit library.
Optionally, the method includes the following steps:
Splitting selected members of the hit variant library into at least two sub-hit variant libraries;
Select a sub-hit variant library;
Creating an amino acid position variant profile of the selected sub-hit variant library;
Converting the amino acid position variant profile of the selected sub-hit variant library into a nucleic acid position variant profile by back-translating the amino acid position variants into the corresponding trinucleotide codons; and combining the nucleic acid position variants in a combinatorial manner Construct a degenerate nucleic acid library.

Degrading the hit variant library includes the following:
Randomly select 10-30 members of the hit variant library that have a score equal to or better than the lead sequence (the selected members form a subvariant library).

Optionally, decomposing the hit variant library includes the following:
Create an amino acid position variant profile of the hit variant library to obtain a hit variant profile;
The hit variant profile is broken down into sub-variant profile segments based on the Cα or Cβ or heavy atom contact map of the structure or assembly of lead sequences within a certain distance cutoff (8 Å to 4.5 Å). Structural model or lead structure template within 4.5 mm, preferably 5 mm, more preferably 6 mm, and most preferably 8 mm.

In another embodiment, the method comprises the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure);
Providing a 3D structure of one or more antibodies having different sequences in the _VH or _VL regions other than the lead antibody;
Combining the lead antibody with one or more antibody structures to form a structure assembly (a structure assembly is defined as a lead structure template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
Creating an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
Combining amino acid variants in the hit library to generate a combination of hit variants that form a hit variant library;
A scoring function is used to determine whether members of the hit variant library are structurally compatible with the lead structural template; and a hit variant library with a score equal to or better than the lead sequence Select members.

  [Route VII. Claim a continuous process using the structure to functional space lead sequence from the sequence shown in FIG. 2B]

In a specific embodiment, the method includes the following steps:
a) providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure);
b) identify the amino acid sequence in the CDR of the lead antibody;
c) select one CDR in the _VH or _VL region of the lead antibody;
d) providing an amino acid sequence comprising at least 3 consecutive amino acid residues in the selected CDR (the selected amino acid sequence is defined as the lead sequence);
e) compare the lead sequence to multiple tester protein sequences;
f) selecting at least two peptide segments having at least 10% sequence identity with the lead sequence from a plurality of tester protein sequences (the selected peptide segments form a hit library);
g) create an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
h) combining the amino acid variants in the hit library to generate a combination of hit variants that forms a hit variant library;
i) using a scoring function to determine whether members of the hit variant library are structurally compatible with the lead structure template;
j) Select a member of the hit variant library that has a score equivalent to or better than the lead sequence;
k) constructing a degenerate nucleic acid library comprising a DNA segment encoding the amino acid sequence of a selected member of the hit variant library;
l) determining the diversity of nucleic acid libraries, if diversity is greater than 1 × 10 ⁶ is the step j) to l) until the diversity of the nucleic acid library is less than or equal to 1 × 10 ⁶ repetition;
m) introducing the DNA segment in the degenerate nucleic acid library into the cells of the host organism;
n) expressing a DNA segment in the host cell so that a recombinant antibody containing the amino acid sequence of the hit library is produced in the host organism cell;
o) selecting a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ ; and
p) If no recombinant antibody is found that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ , repeat steps e) to o).

In another specific embodiment, the method comprises the following steps:
a) providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure defined as a lead structure template);
b) identify the amino acid sequence in the CDR of the lead antibody;
c) select one CDR in the _VH or _VL region of the lead antibody;
d) providing an amino acid sequence comprising at least 3 consecutive amino acid residues in the selected CDR (the selected amino acid sequence is defined as the lead sequence);
e) mutating the lead sequence by substituting one or more amino acid residues of the lead sequence with one or more different amino acid residues to create a lead sequence variant library;
f) using a first scoring function to determine if the lead sequence variant library is structurally compatible with the lead structural template;
g) select a lead sequence variant with a score equal to or better than the lead sequence;
h) comparing the lead sequence to multiple tester protein sequences;
i) selecting at least two peptide segments having a sequence identity of at least 10% with the lead sequence from the plurality of tester protein sequences (the selected peptide segments form a hit library);
j) create an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
k) combining amino acid variants in a hit library to generate a combination of hit variants;
l) combining a selected sequence variant with a combination of hit variants to produce a hit variant library;
m) Use a second scoring function to determine whether members of the hit variant library are structurally compatible with the lead structural template;
n) Select a member of the hit variant library that has a score equivalent to or better than the lead sequence;
o) constructing a degenerate nucleic acid library comprising DNA segments encoding the amino acid sequences of selected members of the hit variant library;
p) determining the diversity of nucleic acid libraries, if diversity is greater than 1 × 10 ⁶ is the step n) ~p) to the diversity of the nucleic acid library is less than or equal to 1 × 10 ⁶ repetition;
q) introducing the DNA segment in the degenerate nucleic acid library into the cells of the host organism;
r) expressing a DNA segment in the host cell such that a recombinant antibody containing the amino acid sequence of the hit library is produced in the host organism cell;
s) selecting a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ ; and
t) If a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ is not found, repeat steps e) to s).

In yet another aspect of the invention, a computerized method for constructing a library of mutant antibodies based on lead antibodies is provided. In certain embodiments, the method includes:
Taking as input an amino acid sequence comprising at least three consecutive amino acid residues in the CDR region of the lead antibody (the amino acid sequence is the lead sequence);
Compare the lead sequence to multiple tester protein sequences using computer-implementable logic;
From the plurality of tester protein sequences, select at least two peptide segments having at least 15% sequence identity with the lead sequence; and, as an output, create selected peptides that form a hit library.
In any of the above methods, the length of the lead sequence is preferably 5 to 100 aa, more preferably 6 to 80 aa, and most preferably 8 to 50 aa.

In any of the above methods, the step of identifying the amino acid sequence in the CDR is performed using Kabat criteria or Chothia criteria.
In any of the above methods, the lead sequence may be a specific sequence within the _VH or _VL of the lead antibody, CDR1, CDR2 or CDR3, or a combination of CDR and FR (eg, CDR1-FR2, FR2-CDR2-FR3). Region, and the amino acid sequence from the full length _VH or _VL sequence. The lead sequence is preferably at least 6 consecutive amino acid residues in the selected CDR, more preferably at least 7 consecutive amino acid residues in the selected CDR, and also preferably all in the selected CDR Of amino acid residues.

Also in any of the above methods, the lead sequence further comprises at least one amino acid residue immediately adjacent to the selected CDR.
Also in any of the above methods, the lead sequence further comprises at least one FR that flanks the selected CDR.

  Also in any of the above methods, the lead sequence further comprises one or more CDRs or FRs adjacent to the C-terminus or N-terminus of the selected CDR.

  Also in any of the above methods, the lead structure template is a fully assembled lead antibody 3D structure, or a heavy or light chain variable region of a lead antibody (eg, CDR, FR and combinations thereof).

  Also in any of the above methods, the plurality of tester protein sequences comprises, in particular, a framework region, preferably an antibody sequence, more preferably a human antibody sequence, most preferably a human germline antibody sequence (V database).

Also in any of the above methods, a plurality of tester protein sequences are retrieved from a gene bank of the NIH or Swiss-Prot database or Kabat database for antibody CDRs.
In any of the above methods, the step of comparing the lead sequence with a plurality of tester protein sequences is performed by an algorithm selected from the group consisting of BLAST, PSI-BLAST, profile HMM, and COBLATH.

Also in any of the above methods, the sequence identity between the selected peptide segment in the hit library and the lead sequence is preferably at least 25%, preferably at least 35%, and most preferably at least 45%.
Also in any of the above methods, the method further comprises the following steps:
Introducing a recombinant antibody containing the amino acid sequence of the hit library encoded by the nucleic acid or degenerate nucleic acid library into the cells of the host organism; and
Recombinant antibodies that bind to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ , preferably 10 ⁷ M ⁻¹ , more preferably 10 ⁸ M ⁻¹ , and most preferably 10 ⁹ M ⁻¹ are selected.

The recombinant antibody may be a fully assembled antibody, Fab fragment, Fv fragment, or single chain antibody.
Host organisms include any organism or cell line thereof that can express a transferred foreign gene sequence, including but not limited to yeast, plants, insects and mammals.

The recombinant antibody may be a fully assembled antibody, Fab fragment, Fv fragment, or single chain antibody. For example, recombinant antibodies are expressed in bacterial cells and displayed on the surface of phage particles. The recombinant antibody displayed on the phage particle may be a double-stranded heterodimer formed by _VH and _VL . Heterodimerization of _VH and _VL chains is facilitated by heterodimers formed between two non-antibody polypeptide chains fused to _VH and _VL chains, respectively. For example, these two non-antibody polypeptides are derived from the heterodimeric receptors GABA _B R1 (GR1) and R2 (GR2), respectively.

Alternatively, the recombinant antibody displayed on the phage particle may be a single chain antibody containing _VH and _VL linked by a peptide linker. The display of single chain antibodies on the surface of the phage particles is facilitated by heterodimers formed by a fusion of the single chain antibody and GR1 and a fusion of the phage pIII capsid protein and GR2.

Target antigens screened against them include small molecules and macromolecules (eg, proteins, peptides, nucleic acids and polycarbohydrates).
In yet another aspect of the invention, a computer readable medium is provided. Computer media has logic to construct a library of mutant antibodies based on lead antibodies, which includes the following:
As an input, comprising an amino acid sequence comprising at least three consecutive amino acid residues in the CDR of the lead antibody (the amino acid sequence is the lead sequence);
Comparing the lead sequence to multiple tester protein sequences;
Logic that selects, from a plurality of tester protein sequences, at least two peptide segments having at least 15% sequence identity with the lead sequence; and, as an output, creates selected peptide segments that form a hit library.

In yet another aspect of the invention, monoclonal antibodies are provided that are capable of binding to human vascular endothelial growth factor (VEGF) with a binding affinity higher than 10 ⁶ M ⁻¹ . This monoclonal antibody may be a fully assembled antibody, Fab fragment, Fv fragment, single chain antibody (scFv).

In certain embodiments, the heavy chain CDR3 of the monoclonal antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 36-48 and 63-125.
In another embodiment, the heavy chain CDR1 of the monoclonal antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 19-30.

In yet another embodiment, the heavy chain CDR2 of the monoclonal antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 31-35.
Optionally, the heavy chain CDR3 of the monoclonal antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 36-48 and 63-125, and the heavy chain CDR1 of the monoclonal antibody is selected from the group consisting of SEQ ID NOs: 19-30 Amino acid sequence.

  In some cases, the heavy chain CDR3 of the monoclonal antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 36 to 48 and 63 to 125, and the heavy chain CDR2 of the monoclonal antibody is from a group consisting of SEQ ID NOs: 31 to 35. Contains a selected amino acid sequence.

  In some cases, the heavy chain CDR1 of the monoclonal antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 19 to 30, and the heavy chain CDR2 of the monoclonal antibody is an amino acid selected from the group consisting of SEQ ID NOs: 31 to 35 Contains an array.

In another embodiment, the heavy chain variable region (V _H ) of a monoclonal antibody against VEGF comprises the amino acid sequence of SEQ ID NO: 126, and the light chain variable region (V _L ) of the monoclonal antibody against VEGF comprises the amino acid sequence of SEQ ID NO: 127. Including.

In yet another embodiment, the heavy chain variable region (V _H ) of a monoclonal antibody against VEGF comprises the amino acid sequence of SEQ ID NOs: 126, 128, 129, 130, and 131, and the light chain variable region (V V) of the monoclonal antibody against VEGF. _L ) comprises the amino acid sequence of SEQ ID NO: 127.

  Antibodies designed using the methods of the invention are not particularly limited, but include cancer, autoimmune diseases (eg, multiple sclerosis, rheumatoid arthritis, systemic lupus erythematosus, type I diabetes, and myasthenia gravis). Of various diseases, including graft versus host reaction disease, cardiovascular disease, viral infection (eg, HIV, hepatitis virus, and herpes simplex virus), bacterial infection, allergy, type II diabetes, blood disease (eg, anemia) Used for diagnosis or treatment. The antibodies can also be used as conjugates bound to diagnostic or therapeutic residues, or in combination with chemotherapeutic agents or biologics. The antibodies are also prepared for administration by a variety of administration routes. For example, antibodies can be oral, topical, parenteral, intraperitoneal, intravenous, intraarterial, transdermal, intranasal, inhalation, vaginal, intraocular, topical delivery (eg chemical or stent), subcutaneous, adipose tissue Administered intra-, intra-articularly or intrathecally.

  In any of the above embodiments, the designed protein (eg, antibody) is synthesized or expressed in the cells of any organism, including but not limited to bacteria, yeast, plants, insects and mammals. . Specific types of cells are not particularly limited, but include Drosophila, Drosophila melanogaster cells, Saccharomyces cerevisiae and other yeasts, E. coli, Bacillus subtilis SF9 cells, C129 cells, 293 cells, Neurospora, BHK, CHO, COS and HeLa cells, fibroblasts, schwannoma cell lines, immortalized mammalian myeloid and lymphoid cell lines, Jurkat cells, There are mast cells, and other endocrine and exocrine cells, and nerve cells. Examples of mammalian cells include, but are not limited to, all types of tumor cells (especially melanoma, myeloid leukemia, lung, breast, ovary, colon, kidney, prostate, pancreas and testicular cancer), cardiomyocytes, Endothelial cells, epithelial cells, lymphocytes (T cells and B cells), mast cells, eosinophils, vascular stromal cells, hepatocytes, leukocytes (including mononuclear lymphocytes), stem cells (eg hematopoietic stem cells), nerve cells There are skin cells, lung cells, kidney cells, liver cells, and cardiac cell stem cells, osteoclasts, chondrocytes and other connective tissue cells, keratinocytes, melanocytes, liver cells, kidney cells, and adipocytes.

  Preferably, the designed protein is purified or isolated after expression according to methods known to those skilled in the art. Examples of purification methods include electrophoresis, molecular methods, immunological methods, and chromatographic methods (ion exchange chromatography, hydrophobic chromatography, affinity chromatography, reverse phase HPLC chromatography, and chromatofocusing). Included). The degree of purification required will vary depending on the intended use of the protein. In some cases, there is no need for purification.

  Also in any of the above embodiments, the designed protein has a desired function, preferably biological function (eg, binding to a known binding partner), physiological activity, stability profile (pH, thermal, (Buffer conditions), substrate specificity, immunogenicity, toxicity, etc. can be screened.

In screening using cell-based assays, the designed protein is preferably selected based on the phenotype of the cell that has been detectably and / or measurably altered. Examples of phenotypic changes include, but are not limited to, large physical changes such as cell morphology, cell proliferation, cell viability, adhesion to substrates or other cells, and cell density; one or more RNAs, proteins, Changes in the expression of lipids, hormones, cytokines or other molecules; changes in the equilibrium (ie, half-life) of one or more RNAs, proteins, lipids, hormones, cytokines or other molecules; one or more RNAs, proteins, Changes in biological or specific activity of lipids, hormones, cytokines or other molecules; changes in secretion of ions, cytokines, hormones, growth factors, or other molecules; changes in cell membrane potential, polarization, integrity or transport; viruses And changes in infectivity, susceptibility, latency, adhesion, and uptake of bacterial pathogens.
In any of the above embodiments, the engineered protein (eg, antibody) is synthesized or expressed as a fusion protein with a tag protein or peptide. A tag protein or peptide is used to identify, isolate, signal, increase flexibility, increase degradation, secretion, migration, increase intracellular retention, or enhance expression of the engineered protein. The

Four embodiments of methods that can be used in the present invention to select for proteins having the desired function are illustrated. The leads in FIGS. 1A-D are lead sequences or sequence profiles from an alignment based on multiple structures. Hit libraries and hit variant libraries I and II are defined in the definition part. Four embodiments of methods that can be used in the present invention to select for proteins having the desired function are illustrated. The leads in FIGS. 1A-D are lead sequences or sequence profiles from an alignment based on multiple structures. Hit libraries and hit variant libraries I and II are defined in the definition part. Four embodiments of methods that can be used in the present invention to select for proteins having the desired function are illustrated. The leads in FIGS. 1A-D are lead sequences or sequence profiles from an alignment based on multiple structures. Hit libraries and hit variant libraries I and II are defined in the definition part. Four embodiments of methods that can be used in the present invention to select for proteins having the desired function are illustrated. The leads in FIGS. 1A-D are lead sequences or sequence profiles from an alignment based on multiple structures. Hit libraries and hit variant libraries I and II are defined in the definition part. Four possible embodiments of the method that can be used in the present invention to select for proteins with the desired function are illustrated. Here, lead means a structure or structural model or structure assembly or profile (multiple overlapping structure), and then the corresponding sequence or sequence profile from the lead structure or structure assembly is structure-based screening. Can be used to screen all possible sequences or random combinations for hit sequence libraries. The resulting hit variant library can be used for direct experimental screening or compared to a sequence hit profile obtained from the corresponding lead sequence or sequence profile (see FIGS. 2A-C). Structural template refers to a structure, an assembly of structures (greater than two structures) from experimental measurements and / or modeling. Four possible embodiments of the method that can be used in the present invention to select for proteins with the desired function are illustrated. Here, lead means a structure or structural model or structure assembly or profile (multiple overlapping structure), and then the corresponding sequence or sequence profile from the lead structure or structure assembly is structure-based screening. Can be used to screen all possible sequences or random combinations for hit sequence libraries. The resulting hit variant library can be used for direct experimental screening or compared to a sequence hit profile obtained from the corresponding lead sequence or sequence profile (see FIGS. 2A-C). Structural template refers to a structure, an assembly of structures (greater than two structures) from experimental measurements and / or modeling. Four possible embodiments of the method that can be used in the present invention to select for proteins with the desired function are illustrated. Here, lead means a structure or structural model or structure assembly or profile (multiple overlapping structure), and then the corresponding sequence or sequence profile from the lead structure or structure assembly is structure-based screening. Can be used to screen all possible sequences or random combinations for hit sequence libraries. The resulting hit variant library can be used for direct experimental screening or compared to a sequence hit profile obtained from the corresponding lead sequence or sequence profile (see FIGS. 2A-C). Structural template refers to a structure, an assembly of structures (greater than two structures) from experimental measurements and / or modeling.

Four possible embodiments of the method that can be used in the present invention to select for proteins with the desired function are illustrated. Here, lead means a structure or structural model or structure assembly or profile (multiple overlapping structure), and then the corresponding sequence or sequence profile from the lead structure or structure assembly is structure-based screening. Can be used to screen all possible sequences or random combinations for hit sequence libraries. The resulting hit variant library can be used for direct experimental screening or compared to a sequence hit profile obtained from the corresponding lead sequence or sequence profile (see FIGS. 2A-C). Structural template refers to a structure, an assembly of structures (greater than two structures) from experimental measurements and / or modeling. 1 is a schematic diagram of an in silico protein evolution system provided by the present invention. A triangular relationship between sequences, structures, and functional space is shown to illustrate the path of movement from a lead structure / lead sequence profile or from a lead sequence / lead sequence profile to a candidate sequence through the sequence, structure, and functional space. It is. In sequence space, a lead sequence or profile is used to search a specific database of evolutionarily related sequences. A sequence profile based on the structural alignment of the lead structure can be used to search for distant homologues of the lead sequence. The hit library variant profile explains the location frequency and entropy of the amino acid sequence. The variant profile is filtered and re-profiled to give an evolutionarily suitable variant profile. This operation is repeated using various search methods on the relevant sequence database. In structural space, in silico variant profiles are generated using structure-based screening of random or evolutionarily pooled sequence libraries. The variant profile is filtered and modified to give a structurally suitable variant profile. This method is iteratively improved with better scoring functions and representative structure aggregates. Variant profiles created using evolutionary or structure based approaches can be generated sequentially (2B: sequence to structure to functional space; 2C: structure to sequence to functional space) or parallel (sequence specific to functional Can be used in space and from structural space to functional space to give an overall variant profile or library of amino acids. The resulting amino acid variant library is back-translated into the nucleic acid library using suitable or optimized codons. This operation can be repeated using different filtering and partitioning methods to adjust the library size to an experimentally manageable range. In order to select for functional variants in the functional space, the synthesized nucleic acid library is introduced into the vector by transformation, eg, functionally expressed or displayed on phage particles. A round of selection and enrichment for immobilized antigen is performed. All or part of the method can be repeated and refined until the desired candidate is experimentally selected. Schematic representation of an embodiment of the method provided in the present invention for antibody library design. The sequential method moves first from “array” to “structure” and then to “functional” space. Design starts with a lead sequence or from a sequence profile (multiple aligned sequences from structure-based alignments). A hit library is created by searching the sequence database. A hit profile obtained with a hit library at a certain cut-off gives a hit variant library. A hit library or hit variant library is screened in a computer using the lead structure or assembly of structures as the template structure. The resulting sequence library is ranked based on compatibility with the template structure or structure aggregate. Sequences with scores that are superior or equivalent to the lead sequence are selected and profiled to create a nucleic acid (NA) library. If the in silico NA library size is assessed and the library size is acceptable, move on to oligonucleotide synthesis. Otherwise, the hit variant library is subdivided into smaller segments, creating a smaller NA library. In functional space, nucleic acid libraries are screened experimentally and positive sequences are returned to the computational cycle for library improvement. Strong positive clones are transferred to candidates for further evaluation and treatment development. If no experimental screening hits, the lead or its new lead profile is selected for the target system and the operation is repeated.

Schematic representation of an embodiment of the method provided in the present invention for antibody library design. An alternative sequential method moves first from “structure” to “array” and then to “functional” space. The design starts with a lead structure or a structure assembly. Combinations of random mutations at the target location are screened computationally for their compatibility with the structural template. A variant profile of the sequence with a score better or equal to the lead sequence is created. This variant profile is compared and / or combined with that provided by searching a sequence database. New variants are included or excluded based on the consensus frequency shown in the sequence and structure space to create a nucleic acid library. The rest of the operation is similar to that described in FIG. 2B. This approach emphasizes the importance of finding new variants by structure-based computer screening without relying on evolutionary sequence information. Sequence profiles from database searches will help evaluate variant profiles obtained from computer screening depending on the accuracy of the scoring function as well as the test algorithm used. Figure 3 illustrates in silico hit library construction by database search using a single lead or lead profile based on structural alignment. The search results are classified and duplicate sequences (even with different backgrounds) are removed to create a list of unique sequences in the hit library. The effects of the lead sequence / sequence profile, sequence search methods, and various databases are shown in FIGS. 2 illustrates the construction of hit variant library I based on variant profiles from hit libraries used to analyze amino acid evolutionary position selection. Filtering based on the frequency, variation entropy, and energy score of the amino acid variant at each position yields an improved variant profile. From the improved variant profile, a hit variant library II is calculated combinatorially. 2 illustrates a method for structural evaluation and selection of hit variant library I or II to generate a structurally screened version of hit variant library II. Computer scoring uses a simple and custom energy function to score and rank the hit variant library I or II applied to the lead structure template. For each sequence, side chains are created using a backbone-dependent rotamer library, and the side chains and backbone are minimized in energy relative to the template background to mitigate local strain. Matches of hit variant library I or II in the template structure are scored and ranked using a simple and custom energy function. To construct a new hit variant library II for translation into a nucleic acid (NA) library, several sets of “optimal” sequences are selected. Selection criteria include sequence set, structural considerations or functional considerations. To create a nucleic acid library within experimentally manageable limits, a collection of amino acid sequences is re-profiled (Figure 6).

The method for constructing a nucleic acid (NA) library by reverse translation from hit variant library II is illustrated. The reverse translation of amino acids into nucleic acids is to maintain the size of the nucleic acid library within experimentally manageable limits while optimizing the preferred codon usage. The size of the nucleic acid library is calculated and maintained within experimental limits, or the hit variant profile is modified by reducing the number of variants or divided into shorter segments. The division is performed using a structure correlated segment or a series of overlapping consecutive correlated segments. Fig. 2 is a schematic of a strategy for testing a library in several areas of fitness landscape. If the combinatorial amino acid or its degenerate nucleic acid library can be designed to test a larger functional space, the matching background of the selected peptide sequence can be extended to cover the larger matching background. . Strategic testing from designed libraries leads to duplication and diversity expansion, which can include large evolutionary jumps in functional space matching background. The modular elements of a typical library plasmid for antibody manipulation are shown. A framework and a library of CDR sequences are designed either repeatedly or combinatorially. FR = Framework area. CDR = complementarity determining region. RE = restriction enzyme site.

FIG. 9A is a sequence comparison of a parent VEGF antibody and a mature anti-VEGF antibody in a V _H CDR. “C” indicates that the atoms of the antigen-antibody complex are contacted within 4.5 mm in the X-ray structure. Bold highlights amino acid differences between parent and mature anti-antibodies in V _H CDRs (CDR1 and CDR3). V _H CDR numbering follows Kabat's rules and sequential scheme (100, 100a, etc. 100, 101). FIG. 9B is a sequence comparison of a parent anti-VEGF antibody and a mature anti-VEGF antibody in V _H CDR3 with flanking regions. The sequence from the parent antibody (SEQ ID NO: 5) is the lead sequence used for database searches. V _H CDR numbering is a sequential scheme used again with Kabat. The frequency distribution of the hit library is shown relative to its sequence identity (in%) relative to the V _H CDR3 lead sequence of the parent anti-VEGF antibody. The lead sequence is shown in FIG. 9B and the Kabat database was searched using the profile HMM (HAMMER 2.1.1) (Johnson, G and Wu, TT (2001) Nucleic Acids Research 29: 205-206). In order to show the phylogenetic diversity of the hit library obtained from the database search of FIG. 10A, the phylogenetic tree of the hit library sequence shown in FIG. 10A is illustrated. A variant profile of 107 sequences from a hit library generated based on the V _H CDR3 read sequence of the parent anti-VEGF antibody is shown. The upper part shows a table listing the amino acid frequencies of the 20 amino acids at each position of the lead sequence. The lower variant profile shows the diversity of amino acid positions. A complete description of a combinatorial library without selective control of amino acid diversity (shown in the lower left of the figure) would require a library size on the order of 10 ¹⁹ . The lower right part of the figure shows the filtered variant profile obtained using a cut-off frequency of 10. All position amino acids present in 10 or fewer of the 107 members of the hit list are filtered. This filtered variant profile can be further screened in a computer to reflect the structural compatibility ranking order if only antibody structures are used, or if antibody-antigen complex structures are used, The binding affinity with the antigen can be reflected. The variant profile does not show a correlation with the contact sites of antigen and antibody, as shown in FIG. 9A. The parent (1bj1) antibody structure and mature (1cz8) antibody in the absence (A) and presence (B) of the VEGF antigen using the Amber94 Forcefield total energy scoring function performed at CONGEN, respectively. A typical plot of the score of the anti-VEGF antibody variant library in the structure is shown. The scores for mature (M) and parent (P) sequences are indicated by arrows. The score of the mature sequence is better than the parent sequence in both template structures in the absence and presence of antigen. FIG. 12c shows the correlation of variant library scores in the presence and absence of antigen. The simple scoring function used here also correlated entirely with the improved scoring function of the hit library (Figures 10 and 11) using the mature antibody (lcz8) template structure, but the correlation plot Some of the variances suggest that some terms, including solvation, etc. should be added to a simple scoring function to improve the correlation. FIG. 13A shows from a computer screen of an anti-VEGF V _H CDR3 hit variant library for experimental screening to demonstrate that the method of the invention can select diverse functional sequences that differ from the parental or mature sequence. It shows that the top 10 sequences of can be selected. Describes the amino acid variant profile in the degenerate nucleic acid and the corresponding variant library. The energy schematic in the upper right of the figure is selected from the left-to-right energy distribution of 10 sequences selected from computer screening, their variant amino acid combinatorial libraries, nucleic acid combinatorial libraries, and in vitro experimental screening. Positive clones are shown. Sequence libraries corresponding to each of the sequence pools shown in the energy schematic are indicated by arrows.

10 sequences selected from computer screening of V _H CDR1 and CDR2 variant libraries, their amino acid variant profiles, and corresponding variant libraries in the degenerate nucleic acids of the V _H CDR1 and CDR2 libraries of anti-VEGF antibodies Indicates. 10 sequences selected from computer screening of V _H CDR1 and CDR2 variant libraries, their amino acid variant profiles, and corresponding variant libraries in the degenerate nucleic acids of the V _H CDR1 and CDR2 libraries of anti-VEGF antibodies Indicates. Shown are UV readings of ELISA positive clones identified in round 1 and round 3 selection of functional anti-VEGF ccFv antibodies using V _H CDR3 (FIG. 13A) encoded by the designed nucleic acid library. The lower number indicates the column number in the 96 well (8 × 12) ELISA plate. Different shaded bars indicate different columns. The V _H CDR3 sequences of positive clones from round 1 and round 3 selections are shown via phage display of the nucleic acid library shown in FIG. 13A. It is clear that many diverse sequences that differ from the parent and mature anti-VEGF antibody V _H CDR3 (FIGS. 9A and B) are selected with great variation at several positions. The phylogenetic tree of the positive clone which shows the diversity of the screened sequence is illustrated. The sequence identity of selected positive clones from the V _H CDR3 shown in FIGS. 14A and B is 57-57 relative to the parent V _H CDR3 sequence, including the N-terminal CAK and C-terminal WG residues (see FIG. 9B). It was in the range of 73%. Pie charts showing the origin of the first and third round screened sequences classified into 3 groups: designed amino acid sequences, combinatorial amino acid sequences from designed sequences, and synthesized A novel combinatorial amino acid sequence encoded by a degenerate nucleic acid library. A: V _H CDR3 clones from the first round of screening in vitro with a distribution of sequences experimentally selected from three library positive clones. B: V _H CDR3 clone from a third round of screening in vitro with a distribution of sequences experimentally selected from three library positive clones. Since only a few positive clones are selected from each round for sequence analysis, the figure is only to show the selected sequence from the designed sequence, its combinatorial amino acids, and a rough percentage of the nucleic acid library used. 14 is a table that describes experimentally selected amino acid sequences from the V _H CDR1, CDR2, and CDR3 libraries of the degenerate nucleic acids shown in FIGS. 13A-C. FIG. 16B shows the distribution of sequence identity of selected sequences from the V _H CDR1, CDR2 and CDR3 libraries against the corresponding parental sequences of anti-VEGF V _H CDR1, 2 and 3, respectively. It is clear that a variety of functional sequences different from the corresponding parent sequence can be selected.

Diagram between four different libraries (designed amino acid sequences, combinatorial libraries of amino acid variants of the designed sequences, and degenerate and fully degenerate nucleic acid libraries that encode unique amino acid sequences) And the distribution of experimentally selected positive clones indicated by X. The innermost (hatched) circle indicates the designed amino acid sequence library selected, for example, based on the hit variant library energy score. Shaded circles indicate a combinatorial amino acid library of sequences selected from computer screening of hit variant libraries. The third (hatched) circle is a combinatorial amino acid library that encodes a unique combinatorial amino acid library. The outermost circle represents a degenerate nucleic acid library of all amino acid sequences obtained from back translation of the amino acid library. The relative size of the outermost circle versus the third (hatched) circle depends on the efficiency of the reverse translation method from amino acid to nucleic acid sequence, taking into account other factors such as codon usage. “X” indicates an experimentally selected sequence. For example, the anti-VEGF V _H CDR3 library from the third round is shown here (see table in FIG. 17B). Distribution in different libraries depends on selection conditions, the effectiveness of the library design, the relative size of the selected clones versus the library, or the number of clones sequenced.

A table showing the relationship between the four libraries (FIG. 17A) and the distribution of experimentally selected sequences of positive clones of the anti-VEGF V _H CDR1, 2, and 3 libraries is shown. The column “AA_Seq / Comb” indicates the number of amino acid sequences selected by computer screening (designed library I) and the number of recombinant sequences of the selected sequence (variant library II). The “NN_seqs / peptide seq” column indicates the number of nucleic acid sequences of the degenerate nucleic acid library and the unique amino acid sequences encoded by the degenerate nucleic acid library. The “exp_seq” column indicates the number of unique sequences experimentally selected from positive clones. The “Distribution of Selected Sequences” column indicates the number of designed amino acid sequences, their combinatorial libraries of amino acid variants, and libraries of degenerate nucleic acids that encode unique peptide sequences. From left to right, the evolution of the sequence match score of anti-VEGF V _H CDR3 at various stages of manipulation is shown: sequences selected from lead sequence, hit library, hit variant library I, computer screen (shadowed) Band), combinatorial library of selected sequences (hit variant library II), combinatorial nucleic acid library encoding combinatorial amino acid sequences, and experimentally selected sequences. The lead sequence was used to identify an evolutionary hit library from a sequence database. An in silico combinatorial library was designed based on the diversity of hit libraries. A subset of sequences screened by computers with better scores than reads were used to create a combinatorial amino acid library. Using a degenerate nucleic acid synthesis strategy to expand diversity, a degenerate nucleic acid library encoding a combinatorial amino acid library was created. Screening the library with experiments yielded sequences that may have improved function. Figure 3 shows a lead profile generated from a structure-based multiple sequence alignment. The structural motif of the lead sequence is used to search the protein structure database (PDB databank) for similar structures within a range of cutoffs. Five structures are overlaid using the Cα atom of V _H CDR3. The root mean square (RMSD) between each structure and the V _H CDR3 structural motif (colored blue) is about 2 mm. The corresponding multiple sequence alignment is shown to the right along with its PDB ID and the corresponding color. FIG. 5 shows a variant profile of 251 unique sequences of a hit library generated based on the read sequence profile of the V _H CDR3 of the parent anti-VEGF antibody. The upper part shows a table listing the amino acid frequencies of 20 amino acids at each position of the lead sequence. The lower part of the figure shows the filtered variant profile obtained using a 5% cut-off of frequency or 12 in this case. Remove all positional solid amino acids present in 12 or fewer of the 251 members of the hit list. This filtered variant profile can be further screened computationally using the structural assembly. Shown is the distribution of sequences from the hit library against the parent V _H CDR3 sequence (FIG. 9B). Circles indicate that up to 36% sequence identity can be identified using a single parent sequence for HMM searches. Triangles indicate that even lower sequence identities up to ~ 20% can be found using lead sequence profiles from structure-based multiple sequence alignments. The sequence search method used here can find diverse hits with distant homology (20% lower) to the lead sequence. Demonstrates a general strategy for creating a focused library within the intersection of sequence, structure and functional space. As shown in FIGS. 19A-C, hit sequence diversity is enhanced using structure-based multiple alignment. Diversity can be expanded in both sequence and structure space, and good hits can be identified at the intersection of all three spaces.

It is a schematic diagram which shows various antigen binding unit (Abu) orientation. Note two novel display systems used in the method of the present invention: ccFv system, heterodimeric coiled-coil stabilized Fv with disulfide bond between GR1 and GR2, and GMCT system, adapter-mediated scFv display system. The nucleotide and amino acid sequences of GABAb receptors 1 and 2 used to construct the subject ccFv Abu are shown. Coiled-coil sequences are obtained from a human GABA _b -R1 and GABA _b -R2 receptors. The coding amino acid sequence from the GABA _b receptor is shown in bold. A flexible GlyGlyGlyGly spacer was added to the amino terminus of the R1 and R2 heterodimerization sequences to promote functional Fv heterodimer formation. To further stabilize the heterodimer, we introduced a ValGlyGlyGly spacer to immobilize the heterodimeric coiled-coil pair by disulfide bonds. An additional SerArg coding sequence at the N-terminus of the GGGG spacer provides an XbaI or XhoI site for fusion of the GR1 and GR2 domains, respectively, at the carboxy terminus of the V _H and _VL fragments. The nucleotide sequence and amino acid sequence of the anti-VEGF ccFv antibody AM2 are shown. The nucleotide sequence and amino acid sequence of the anti-VEGF ccFv antibody AM2 are shown. FIG. 3 is a schematic diagram of a fazimide vector pABMD12. The sequence of the pABMD12 vector is shown. A comparison of the ability of phage-displayed AM2 ccFv and scFv to bind to immobilized VEGF antigen is shown. The results show that ccFv can be assembled and displayed on phage particles. Results of ELISA using AM2-ccFv from model library panning are shown. The results show enrichment of phage displaying AM2-ccFv antibody in the panning of the model library. Test sequences shows the PCR results from 1/10 ⁷ model library panning indicating that can be selected from the model library. The ELISA backbone using phage from library panning is shown. The results show that VEGF-binding phages were selected from the V _H CDR1 and CDR2 libraries (see FIG. 14A for V _H CDR3). FIG. 16 is a table describing the amino acid sequences of experimentally selected clones encoding designed anti-VEGF V _H CDR1, CDR2, and CDR3 libraries (same as FIG. 16A). Figure 2 shows a sequence library of a composite anti-VEGF V _H CDR3 library. Because the library size is too large to be covered by one or several degenerate nucleic acid libraries, the variant profile is broken down into three segments with the variant profile shown in FIG. 28A. The segment is decomposed based on the contact map of the Cα atoms in 8Å shown on the right side of FIG. 28A. FIG. 28A also shows the ribbon schematic of anti-VEGF V _H CDR3 and the contact distance of Cα atoms within 8 cm. This approach provides a general way to decompose large variant profiles into smaller segments based on the structural topology. Segmentation of sequences to capture distant covariants in the primary sequence (eg, N- and C-terminal residues close together in the loop) requires only structural constraints from topological features Thus, low resolution structures or structural models serve this purpose. FIG. 28B covers the N-terminus and C-terminus, which may contain bound variants (1-3). The variant profiles of both the amino acid library and the nucleic acid library are described, along with the combinatorial size of the library and the last synthesized degenerate oligonucleotide. FIG. 28C contains segment (4). All three segments are covered by a nucleic acid library of size less than 10 ⁶ : (1-3) in FIG. 28B is targeted by three degenerate nucleic acid libraries, and in FIGS. 28C-D ( 4) and (5) are targeted by another degenerate nucleic acid library.

FIG. 28D contains another segment (5). All three segments are covered by a nucleic acid library of size less than 10 ⁶ : (1-3) in FIG. 28B is targeted by three degenerate nucleic acid libraries, and in FIGS. 28C-D ( 4) and (5) are targeted by another degenerate nucleic acid library. Summarize the procedures and conditions used to pan the ccFv library L14, as well as for the enrichment factors from each panning. The L14 library is constructed in FIGS. 28A-D, pooling together all five degenerate oligonucleotides shown in FIGS. 28B-D. The amino acid sequences of V _H CDR3 variants selected by panning libraries L14 5 and 7 using the ccFv display platform are shown. Note that after 5 pannings, all variants are in position 101. Only two variants (S101R and S101T) are selected after round 7. HR (H97, S010R) phage enrichment from panning of library L14 for V _H CDR3. The enrichment of HR and parent antibody WT (see also FIG. 9B) in rounds 0, 5, and 7 is highlighted. 1 shows a simple schematic diagram of a novel coiled-coil domain interaction mediated display (CDIM) adapter command display system for single chain antibody libraries. Transgenic infection of the expression vector pGDH1 alone in E. coli allows the expression and production of soluble proteins fused with GR1 in the peribacterium space. Superinfection of the same bacteria with an UltraHelper phage vector expressing a genetically engineered coat protein fused with GR2 and other phage proteins, followed by the synthesis of phage particles in the peribacterium space, followed by antibody fragments on the surface of filamentous phage ( Or display other proteins). FIG. 33A shows a map of the GMCT-UltraHelper phage plasmid. This construct comprises a nucleotide sequence encoding an additional copy of genetically engineered gene III fused to adapter GR2 and myc protein in the KO7kpn phage vector, and a ribosome binding sequence OmpA leader sequence adjacent to the WT gene III sequence. contains. FIG. 33B shows the genetically modified region of KO7Kpn to produce GMCT-UltraHelper phage at the nucleotide and amino acid sequence level. Shows protein expression vector map (A) and complete nucleotide sequence (B) of pABMX14, which includes the ampicillin resistance gene (Amp), plasmid origin of replication (ColE1 ori), f1 phage origin of replication for antibiotic selection (F1 ori), a lac promoter / lacO1 regulatory protein expression cassette (plac-RBS-pelB-GR1-DH), and a restriction endonuclease site. NcoI / XbaI or NcoI / NotI or XbaI / NotI restriction sites can be used to insert nucleotide sequences encoding the protein of interest. Shows protein expression vector map (A) and complete nucleotide sequence (B) of pABMX14, which includes the ampicillin resistance gene (Amp), plasmid origin of replication (ColE1 ori), f1 phage origin of replication for antibiotic selection (F1 ori), a lac promoter / lacO1 regulatory protein expression cassette (plac-RBS-pelB-GR1-DH), and a restriction endonuclease site. NcoI / XbaI or NcoI / NotI or XbaI / NotI restriction sites can be used to insert nucleotide sequences encoding the protein of interest. FIG. 35A summarizes the procedures and conditions used to pan the scFv library L17 along with each round of enrichment factor (A). The sequence of the L17 library in the V _H CDR3 region is exactly the same as L14 (see FIGS. 28A-D). FIG. 35B shows a flowchart of the panning process.

The amino acid sequences of V _H CDR3 variants selected from library L17 by off-rate panning from two parallel steps 4 and 5, respectively, using an adapter-mediated phage display system are shown. For off-rate panning 4, sequences were selected using a variant located at position 97 and / or 101 (100a in Kabat nomenclature). For off-rate panning 5, sequences were selected using variants located at 101 (100a) and / or 102 (100b) and / or 103 (100c). Two important mutants YS (H97Y-S101) and HT (H97-101T or H97-S100aT) in the mature sequence were selected separately from panning 4 and panning 5. The combination of variants at these two positions may give the mature sequences H97Y and S100aT in V _H CDR3 (FIG. 9B). However, this combination is consciously avoided in the disassembled segment (see FIGS. 28A-D). HR (H97-S100aR) was again proven at a higher frequency (3/1) than the mature sequence (Figure 9B) HT (H97-S100aT), which is similar to Panning 7 in Figure 30 (7 / Match 3). Affinity data for four antibodies containing the anti-VEGF antibody VH CDR3 (FR123) selected via the ccFv display format from a designer library using a BIAcore biosensor. By measuring the change in SPR units (y-axis) versus time (x-axis) when the purified antibody binds to its antigen (VEGF) immobilized on a CM5 biochip at 25 ° C, the measurement is Done. Both on-rate and off-rate changes were determined from data fitting using a 1: 1 Langmuir coupling model. X50 is in the ccFv format and contains the VH and VL parental sequences shown in FIGS. 22A and 22B. X63 contains H97Y and S101T in VH CDR3 with a Kd improvement of 6.3 times (see FIG. 9B), the rest being the same as X50. X64 contains the S101R variant in VH CDR3 with a 2.5-fold improvement over reference X50; this improvement is mostly due to increased on-rate. X65 contains H97Y and S101R and shows a 10-fold improvement over X50 using the ccFv format under the same conditions, which is the most reported variant combination of affinity matured VH CDR3 sequences It has stronger binding affinity than X63 (H97Y and S101T) (see Chen et al., Supra (1999), J. Mol. Biol. 293: 865-881). FIG. 38A shows the heavy chain variable region framework region FR123 defined based on Kabat nomenclature, with a random library used for humanization reported for comparison (Baca et al., Supra, 1997). The mouse anti-VEGF VH framework FR123 sequence shown in A4.6.1 is shown in FIG. 9B. Humanized antibodies (hereinafter referred to as “humanized anti-VEGF antibodies”) used here as parent and reference framework fr123 have been reported in the literature (Presta et al., Supra, 1997). The above SEQ ID NO of FR123 is based on Kabat nomenclature (kabataa) and its sequential order includes amino acids in the CDR. FIG. 38B shows a variant profile of a hit library generated using human VH germline sequences based on the lead sequence of mouse anti-VEGF antibody VH FR123. The lower variant profile shows the positional diversity of amino acids. The lower part of the figure shows the filtered variant profile obtained using cut-off frequencies 5 and 13, respectively. All position amino acids that occur 5 times or less (13 or less) of the members of the hit list are filtered. FIG. 38B (continued), where the re-profiled variant profile of the hit library was generated using the human VH germline sequence based on the VH FR123 lead sequence of the mouse anti-VEGF antibody without cut-off, Each position variant is shown to be ranked based on its structural compatibility with the antibody structure using total energy or van der Waals energy. This ranking emphasizes that some low-occurrence amino acids are structurally important in stabilizing the framework scaffolds that are maintained for optimization. FIG. 38C shows the variant profile of a hit library generated using the human VH sequence from Kabat based on the lead sequence of the mouse anti-VEGF antibody VH FR123 with a cut variant 19 and a filtered variant profile. The mouse VH FR123 sequence is indicated by the dotted line at the position numbered consecutively for reference. All variants of amino acids are listed under the dotted line. Points in the variants indicate the same amino acids as in the reference.

FIG. 38D shows a designer library using a filtered variant profile from human VH germline sequence with cutoff 5 (see FIG. 38B). The above SEQ ID No. attached to FR123 is based on Kabat nomenclature (kabataa) and sequential order including amino acids in the CDR. The filtered variant profile is further screened by computer to reflect the structural compatibility ranking order when only antibody structures are used. Two amino acids F70 (F69) and L72 (L71), which are missing from the variant profile filtered at cutoff 5, are the most preferred amino acid companions at these positions based on structure-based scoring Therefore, these were also included. Libraries finally submitted for the top 100 ranked sequences from structure-based screening are also F70 (F69), L72 (L71), S77 (S76) and K98 (K94) (in parentheses) Numbers indicate SEQ ID NOs based on Kabat nomenclature) because some amino acids such as R have L72 (L71) and K98 as already discussed for K94R during VH CDR3 affinity maturation. This is because the calculation for (K94) is overestimated. Figure 39A shows in column 1 using 1bj1 (upper panel) and 1cz8 (lower panel) as template structures in the absence (leftmost column) and presence (middle column) of VEGF antigen. The distribution of scoring diagrams of mouse anti-VEGF VH framework fr123 hit sequences using human VH germline sequences in a relatively dense blue strip is shown on the x-axis, together with relatively sparse in column 0 The mouse and human framework fr123 (Presta et al., See above) sequence in the blue strip and the widely used human VH germline DP47 are shown on the x-axis. Sequence scores in the presence and absence of antigen are correlated (rightmost column), and because they have minimal contact with antigen, framework optimization is sufficient for most framework optimizations The antibody structure of is shown. The scoring diagram of the combinatorial sequence library is not shown. FIG. 39B reported (see Presta et al., Supra, 1997 and Chen et al., Supra, 1999) mouse VH FR123, humanized VH FR123, and the top 200 designer sequences and human VH3 germline (referred to as DP47). For the widely used VH human germline, the left panel shows the ranking score based on the difference between the sequence in the library and the reference mouse VH FR123 sequence, and the phylogenetic distance on the x-axis (the linkage of these) Distance (see also Fig. 14C)). The top 200 ranked sequences from a structure-based screen of one human germline variant profile (AA-PVP) assemble with the human VH3 germline family in a phylogenetic analysis (red circles), while leading The mouse antibody framework is phylogenetic, although its phylogenetic distance is genetically distant from the humanized sequence designed from 1bj1 (Presta et al., Supra) (contains only human germline VH sequences). The distance will vary slightly by including amino acids with relatively low incidence such as F89 (F69) and K98 (K94) (FIGS. 42C and D). The y-axis shows that most designed framework VH fr123 has good structural compatibility with the structure compared to the mouse reference and humanized framework VH fr123 (close to DP47). These support the human-like features of framework optimization of the methods of the invention described herein, as defined in part by the database used. The nucleic acid and amino acid sequences of overlapping oligos, anti-VEGF heavy chain variable region (VH) libraries used for library assembly are shown. The degenerate positions of DNA sequences are S (C or G), R (A or G), M (A or C), Y (C or T), K (G or T), W (A or T), respectively. And the corresponding amino acid residue encoded is described as “X”. CDR regions are shown in bold. HindIII and StyI are the upstream and downstream cloning sites of the library, respectively. The nucleic acid and amino acid sequences of overlapping oligos, anti-VEGF heavy chain variable region (VH) libraries used for library assembly are shown. The degenerate positions of DNA sequences are S (C or G), R (A or G), M (A or C), Y (C or T), K (G or T), W (A or T), respectively. And the corresponding amino acid residue encoded is described as “X”. CDR regions are shown in bold. HindIII and StyI are the upstream and downstream cloning sites of the library, respectively.

Panning summary of phage display library of anti-VEGF VH. P1 to P8 indicate the first to eighth rounds of panning. The VEGF concentration for coating and the amount of phage in the library (input) decreased with the progress of panning. All wash conditions began with 10 simple washes in PBST and ended with 10 simple washes in PBS before elution of bound phage occurred. In all cases the incubation was carried out at 37 ° C. for 2 hours. In the eighth round of panning, the library was mixed with competitor phage in a ratio of 5 during the incubation. Full length sequence of hit clones from panning of phage display library of anti-VEGF VH. Sequencing data was isolated from the 7th and 8th panning of the phage display library, respectively. The sequence of the CDR regions employed (CDR1, 2 and 3) remained the same as in the mouse anti-VEGF antibody sequence (see FIG. 9B) during library construction as described in the text. The hit rate is the incidence of a particular clone at the described panning stage. Panning summary of phage display library of anti-VEGF VH. The letter indicates the amino acid residue at a particular position (indicated by the number after the letter, which is variable in the heavy chain of anti-VEGF as illustrated by Figure 38A in both sequential and Kabat nomenclature. Based on the linear order of the amino acid sequence of the region). The published mouse sequence of anti-VEGF VH and its corresponding humanized form are shown in the left first and second columns, respectively, aligned with the dominant residues at the same position of human immunoglobulin family III. Sequencing data was obtained from clones isolated from the 5th, 6th, 7th, and 8th pannings of the phage display library, respectively. The number before the letter indicates the hit rate (in%) for the particular residue under test (* caused by PCR error). Phylogenetic analysis of the above hit VH sequences from panning of anti-VEGF phage display library, as annotated, along with human germline VH3 family, mouse anti-VEGF VH framework FR123 and humanized VH framework fr123 Show. As shown in FIG. 42C, the human germline VH3 family assembles at a phylogenetic distance as expected. The selected optimized VH framework also assembles with the humanized VH sequence (as noted in the annotation), which is very close to the phylogenetic distance from the human germline VH3 family, while the mouse VH The framework is far from the optimized VH framework and the human germline, which is a framework that has fully human or human-like sequences of antibodies that the method of the invention has optimized in design. We support the conclusion that was optimized depending on the delicate balance between human-like and structural template or assembly structure or compatibility with template from structure average. FIG. 42D is annotated for some well-characterized sequences D36, D40 and D42 and related sequences and shows the phylogenetic distance of these sequences in a different tree view. D36 is human or slightly better than the humanized sequence reported at phylogenetic distance. Using the ccFv phage display system (see description of FIGS. 23-25 above), the sequence of the optimized VH framework (FR123) of the anti-VEGF antibody selected from the designer VH optimized library Show. VH fr123 of D36, D40 and D42 together with the original murine antibody VH FR123 and a humanized sequence (Presta et al., Supra) with the same CDRs from the murine antibody. The dots in the bottom panel indicate that the amino acids are the same as the reference (mouse VH framework fr123).

Five antibodies of the anti-VEGF antibody, parent antibody (X50), and optimized framework (D36) selected from the designer library using the BIAcore biosensor (see Figure 43A and see Figure 43B for its sequence) , D40, D41 and D42) are shown. When the purified antibody binds to its antigen (VEGF) immobilized on a CM5 biochip at 25 ° C, the measurement is performed by measuring the change in SPR units (y-axis) versus time (x-axis). . Both on-rate and off-rate changes were determined from data fitting using a 1: 1 Langmuir coupling model. The two humanized frameworks D36 and D40 have framework-optimized binding affinity (in ccFv format) almost 4 times higher than the parent / reference anti-VEGF antibody sequences (humanized anti-VEGF antibody frames reported in the literature) See Figures 22A and B for workpieces (Presta LG, Chen H, O'Connor SJ, Chisholm V, Meng YG, Krummen L, Winkler M, Ferara N (1997) Cancer Res. 57: 4593-4599)), D42 Is almost the same as the reference antibody. Since the reported humanized anti-VEGF antibodies (FIGS. 22A and B) are almost 2-fold weaker than the corresponding mouse antibodies, these two humanized antibodies have a binding affinity almost 2-fold higher than the corresponding mouse antibodies upon humanization. Should have. Shows increased stability of optimized VH frameworks (D36 and D40). The y-axis represents the proportion of antibody immobilized VEGF using BIAcore at 25 ° C after incubation for 17 hours at 4, 37 and 42 ° C with purified antibody for parental X50 and optimized framework (D36 and D40) Shows that binding to the antigen remains active. This indicates that the optimized framework has a higher stability than the reported humanized VH framework (Presta et al., Supra, 1997). Shows improved expression of optimized _VH framework. The optimized framework (D36, D40 and D42) also improved expression compared to the parent / wild-type antibody (X50) as shown by the yield expression detected by SDS-PAGE / Coomassie blue staining Indicates that The VH and VL amino acid sequences of selected antibodies against human VEGF are shown.

Definitions Structural clusters: based on several empirically selected cutoff values (eg, of Cα atoms of aligned residues) and statistical significance (Z score) of root mean square (RMSD) A group of structures collected. These values are determined empirically after an overall comparison between the target structures. Several programs can be used to search the structural cluster. For the CE (combinatorial extension) algorithm (Shindyalov IN, Bourne PE (1998) Protein Engineering 11: 739-747), the criterion used is RMSD <2Å and Z score> 4. MAPS (Multiple Alignment of Protein Structures) is an automated program for comparing multiple protein structures. The program automatically superimposes 3D models of common structural similarity, detects which residues are structurally equivalent in all structures, and provides residue-to-residue alignment. Residues of equivalent structure are defined according to the approximate position of the main chain and side chain atoms of all proteins. According to structural similarity, the program calculates a structural diversity score, which can be used to construct a phylogenetic tree (Lu, G (1998) “Approach for multiple alignment of protein structures”). In a structural set, members within a structural cluster are analyzed to understand common information regarding the distribution of structural templates within the family and constraints on sequences or sequence profiles within the structural family.

  Aggregate structure: In a structure measurement by NMR (nuclear magnetic resonance), a collection of structures (possibly several members) rather than a single structure (all of which match the NMR data and retain good conformation) It is known that it is stored in a protein data bank. Comparison between models in this set provides some information about how protein conformation is measured due to NMR constraints. It should be pointed out that all sequences corresponding to the aggregate structure measured by NMR have the same sequence (one protein with variable conformation). Structural aggregates also have sequence and / or length variations, but have similar backbone conformations other than these structures (eg, from NMR measurements or from molecular dynamics simulations) and are of the same sequence Some have different structures due to natural shape variations.

  Aggregated sequence: A collection of sequences that statistically define certain properties (eg, stability or binding affinity) of a target protein.

  Aggregate average or representative structure: if all members in a structural cluster have the same length of amino acids, the positions of the atoms in the main chain atoms of all structures are averaged, and then the average model is measured by NMR As with the average structure, it is adjusted to follow normal bond distances and angles (“inhibition minimization”). If the amino acid lengths of all members in the structural cluster are different, a member that is representative of the average characteristics of all other members in the cluster will be selected as the representative structure.

  Standard structure: The normal backbone conformation of the hypervariable region.

Structural repertoire: A collection of all structures occupied by a class of proteins, such as the modular and standard structures observed for antibody frameworks and CDRs.
Sequence repertoire: A collection of sequences from a protein family.

  Functional repertoire: A collection of all the functions a protein exhibits, for example for antibodies, associated with a variety of functional CDRs that can bind to different antigens.

  Germline gene segment: refers to genes from the germline (the haploid germ and diploid cells from which they are generated). Germline DNA contains multiple gene segments that encode a single immunoglobulin heavy or light chain. These gene segments are carried in germ cells, but cannot be transcribed or translated into heavy and light chains until they are prepared into functional genes. During B cell differentiation in the bone marrow, these gene segments are randomly shuffled by a dynamic gene system that can provide more than 108 specificities. Most of these gene segment sequences are available from germline databases. Variable heavy and light chains, referred to as V gene databases, are classified into subfamilies based on sequence homology.

  Realigned immunoglobulin sequences: functional immunoglobulin gene sequences in the heavy and light chains that are generated by transcription and translation of germline gene segments during the B cell differentiation and maturation process. Most of the rearranged immunoglobulin sequences used here are from the Kabat-Wu database.

  BLAST: Basic Local Alignment Search Tool for paired sequence analysis. BLAST uses a heuristic algorithm that uses position-independent scoring parameters to detect similarities between two sequences. The default parameters are Expect 10, Word Size 3, Scoring matrix BLOSUM62, exostence. Used in Gap costs 11, extension 1.

  PSI-BLAST: A position-specific repeated BLAST or PSI-BLAST program uses a sequence found in one round of search to perform an iterated search that builds a score model for the next round of search. In PSI-BLAST, the algorithm is not related to a specific score matrix. Traditionally, this is done using an A × A substitution matrix (where A is the alphabet size). Instead, PSI-BLAST uses a QxA matrix, where Q is the length of the sequence in question; at each point the cost of the letter depends on the position relative to the sequence in question and the letter in the subject sequence. Two PSI-BLAST parameters are adjusted; the pseudocount constant default has been changed from 10 to 7, and the E value threshold for including matches in the PSI-BLAST model has been changed from 0.001 to 0.002.

  Energy background: An energy distribution whose peaks and valleys define the molecular state of aggregation. The energy background can give a complete description of the folding process as well as a description of the local structure state, and a common optimization or minimization structure can be derived from a collection of many possible states within the local energy minimum. Only the structural species will be described.

  Match / match score: a measure of the experimentally observable properties (eg, stability, activity and affinity) of a molecule.

  Fitness landscape: A distribution of match scores defined by other intrinsic parameters of the molecule (eg, sequence).

Sequence space: See sequence repertoire.
Structural space: see structural repertoire.
Functional space: see functional repertoire.

Lead sequence: A sequence used to search a sequence database.
Variant Profile / Sequence Profile / Position Variant Profile (PVP): A description of the amino acid entropy at each position for a set of peptide sequences. This includes both the range and frequency of amino acids (AA-PVP) or nucleic acids (NA-PVP).

  Hit library / hit list: A collection of sequences found by searching a sequence database using a lead sequence or sequence profile.

  Hit variant library / library I: An in silico amino acid sequence library obtained from combinatorial calculation of variant profiles of hit libraries.

  Hit Variant Library II / Library II / Designed Amino Acid Library / Improved Amino Acid Library: An in silico amino acid sequence library obtained from Hit Variant Library I as a result of reprofile or specific design. Variant re-profiling is done by: 1) By selecting sequence cluster-based energy ranking using a sequence window containing specific cut-off values or major amino acid residues, 2) functional By including residues at specific positions identified by screening, and / or 3) using any other method available to those skilled in the art to make such measurements, or including residues or sequence clusters By eliminating.

  Hit variant library III / library III: an amino acid sequence library expressed in vitro by a degenerate oligonucleotide library (described below) for functional screening. Library III expands the sequence space of Library II for reverse translation, optimized codon usage, recombination at the nucleotide level, and expression of the resulting combinatorial nucleic acid library.

  Degenerate nucleic acid / oligonucleotide library: A library of mixed oligonucleotides used to target amino acid variant profiles corresponding to a designed amino acid library (Library II above). This is obtained from combinatorial calculations of the corresponding nucleic acid position variant profiles that are back-translated from the amino acid position variant profiles in Library II using optimized codons.

  Combinatorial amino acid / peptide library: A library created from a complete combinatorial calculation of amino acid position variant profiles. Library I and library II are such libraries.

  Combinatorial nucleic acid / oligonucleotide library: A library created from a complete combinatorial calculation of nucleic acid position variant profiles.

  DNA shuffling: A method in which recombinant oligonucleotides are made from a mixture of parental sequences by multiple repetitions of oligonucleotide fragmentation and homologous recombination (Stemmer WP (1994) Nature 370: 389-391).

  In silico rational library design: a digital amino acid or nucleic acid library that captures evolutionary, structural and functional data to efficiently define a collection of samples in sequence and structure space to identify those with the desired match How to design.

  Profile Hidden Markov Model (Profile HMM): A statistical model of primary structural consensus of sequence families based on the sequence profile of proteins. This detects distant sequence homologues based on statistical descriptions of consensus of multiple sequence alignments, using amino acid and position specific scores to open and extend insertions and deletions. Multiple sequence alignment is provided by multiple sequence alignment programs such as ClustalW or structure-based multiple sequence alignment provided by structure sets.

  Threading: A method of assigning protein folds by connecting a sequence of proteins to a library of candidate structural templates using a scoring function that captures the sequence as well as local parameters (eg secondary structure and solvent exposure). The threading process starts with the prediction of the secondary structure of the amino acid sequence and the solvent accessibility of each residue in the sequence in question. The resulting one-dimensional (1D) profile of the predicted structure is linked to each member of the library of known 3D structures. Optimal threading for each sequence-structure pair is obtained using dynamic programming. The overall optimal sequence-structure pair constitutes the predicted 3D structure of the sequence in question.

  Reverse threading: A method of searching for optimal sequences from sequence databases by connecting to certain target structures and / or structure clusters. Various scoring functions are used to select for optimal sequences from libraries containing protein sequences of various lengths.

  Side chain rotamer: Amino acid side chain conformation defined by the dihedral angle or chi angle of the side chain.

  Rotamer library: Skeletal dihedral phi and phi (called backbone-dependent rotamer library) or backbone-independent dihedral angles for all amino acids obtained from side chain conformation analysis in protein structure databases Side chain rotamer distribution based on (called backbone independent rotamer library).

  See Dunbrack RL and Karplus M (1993) JMB 230: 543-574.

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to protein live for optimized proteins with improved biological function, eg improved binding affinity for biologically and / or therapeutically important target molecules. Provide systems and methods for efficiently creating and screening rallies. This method is processed at high speed by a computer by examining an ever-growing database of protein sequences of all organisms, especially humans. Combining a database search of evolutionary sequences from nature with the computational design of variants that are related to the structure of the native sequence, the method of the present invention can be used in other methods in the computational design and functional screening of protein libraries. A clear leap from the method.

  By using this innovative method, it is possible to construct a library of proteins, such as antibodies, in silico based on computational evaluation of structures that are functionally related to extremely diverse protein sequences. it can. This set-based statistical method of in silico library construction and screening efficiently matches protein sequence and structural space and distribution of energy background, a goal that in vitro or in vivo screening has virtually failed to achieve To create. After in silico screening, an expanded nucleic acid library based on the sequence encoding the selected protein is constructed and introduced into the expression system and screened for proteins that have been improved or have novel functions in vitro or in vivo .

  FIG. 1 is a series of flowcharts outlining various embodiments of the method of the present invention. Based on lead proteins using known sequences and / or structures, a library of proteins can be constructed and screened for candidates with the desired function after at least four pathways (routes I-IV) shown in FIG. .

In certain embodiments, the method includes the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to a plurality of tester protein sequences; and selecting from the plurality of tester protein sequences at least two peptide segments having at least 15% sequence identity with the lead sequence (the selected peptide segment is a hit library Form).
The method further includes the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed.

  Route I in FIG. 1A schematically illustrates this embodiment. In this embodiment, a lead protein (eg, antibody) having a known sequence and structure is provided. A rich pool of protein sequences (eg, a human antibody repertoire) is screened for varying degrees of identity with selected segments of the lead protein (hereinafter referred to as “lead sequences”). From this screen, a list of protein sequences with different degrees of homology is selected using a shaded Markov model or sequence alignment method such as HMM (hereinafter referred to as a “hit library”). The amino acid sequence of the hit library is then profiled against the lead sequence to demonstrate the variation of amino acid residues at each position of the lead sequence. As detailed in Section 7 below, some or all of the profiled sequences are selected and back-translated into a library of nucleic acids for functional screening in vitro or in vivo.

Optionally, the present invention further comprises the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into corresponding gene codons; and combinatorially combining nucleic acid position variants into a degenerate nucleic acid library of DNA segments Build up.

  Route II of FIG. 1B schematically illustrates this embodiment. In this embodiment, after the amino acid sequence of the hit library is profiled to the lead sequence, a combinatorial library (hereinafter referred to as the amino acid position variant profile or AA-PVP) is used based on the frequency of amino acids at each residue position (also referred to as amino acid position variant profile or AA-PVP). (Called "Hit Variant Library I" or "Library I"). Using this approach, the hit variant library I is substantially larger than the hit library. Based on what is observed more frequently (indicating evolutionary selection), AA-PVP can be modified (eg, filtered) to bias the preferred variant at each position, resulting in a reduced variant profile Then, the hit variant library II is obtained by the combinatorial calculation. The hit variant library II profile is back-translated into a library of nucleic acids for functional screening in vitro or in vivo.

In some cases, the gene codon is suitable for expression in bacteria. In some cases, the gene codon is within a diversity range that the diversity of the degenerate nucleic acid library of the DNA segment can be experimentally covered without undue experimental effort (eg, preferably less than 1 × 10 ⁷ , more preferably The selected size can be reduced to be less than 1 × 10 ⁶ ).

In another embodiment, the method comprises the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identify the amino acid sequence in the CDR and FR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing a first amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a CDR lead sequence);
Comparing the CDR lead sequence with a plurality of tester protein sequences; and selecting from the plurality of CDR tester protein sequences at least two peptide segments having at least 15% sequence identity with the CDR lead sequence (the selected peptide segments are Forming a CDR hit library);
Select one FR in the _VH or _VL region of the lead antibody;
Providing a second amino acid sequence comprising at least three consecutive amino acid residues in the selected FR (the selected amino acid sequence is the FR lead sequence);
Comparing the FR lead sequence to a plurality of FR tester protein sequences; and selecting from the plurality of FR tester protein sequences at least two peptide segments having at least 15% sequence identity with the FR lead sequence (selected peptide segments Forms an FR hit library); and
A hit library is formed by combining the CDR hit library and the FR hit library.

  In this method, the plurality of CDR tester protein sequences may comprise the amino acid sequence of a human or non-human antibody.

Also in the present invention, the plurality of FR tester protein sequences are amino acid sequences of human origin, preferably human or humanized antibodies (eg, at least 50% human sequences, preferably at least 70% human sequences in _VH or _VL). More preferably at least 90% human sequences, and most preferably at least 95% human sequences), more preferably fully human antibodies, and most preferably human germline antibodies.

  In the present invention, at least one of the plurality of CDR tester protein sequences is different from the plurality of FR tester protein sequences.

  In the present invention, the plurality of CDR tester protein sequences are human or non-human antibody sequences, and the plurality of CDR tester protein sequences are human antibody sequences, preferably human germline antibody sequences.

The method further includes the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed.
Optionally, the method further comprises the following steps:
Create an amino acid position variant profile of the CDR hit library;
Converting the amino acid position variant profile of the CDR hit library into a nucleic acid position variant profile by back-translating the amino acid position variants into the corresponding gene codons; and combing the nucleic acid position variants in a combinatorial manner Build a rally.

In some cases, the gene codon is suitable for expression in bacteria. In some cases, the gene codon is within a diversity range that the diversity of the degenerate nucleic acid library of the DNA segment can be experimentally covered without undue experimental effort (eg, less than 1 × 10 ⁷ , preferably 1 × 10 ^{7 The} selected size can be reduced to be less than ⁶ ).

In another embodiment, the method comprises the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the FR of the lead antibody;
Select one FR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in the selected FR (the selected amino acid sequence is the first FR lead sequence);
Comparing the first lead FR sequence to a plurality of FR tester protein sequences; and selecting at least two peptide segments from the plurality of FR tester protein sequences having at least 15% sequence identity with the first FR lead sequence (Selected peptide segments form an FR hit library).

The method further includes the following steps:
Providing a second amino acid sequence comprising at least three consecutive amino acid residues in a FR different from the selected FR (the selected amino acid sequence is the second FR lead sequence);
Comparing the second FR lead sequence with a plurality of FR tester protein sequences; and selecting from the plurality of FR tester protein sequences at least two peptide segments having at least 15% sequence identity with the second FR lead sequence (The selected peptide segment forms a second FR hit library); and the first FR hit library and the second FR hit library are combined to form a hit library.

In this method, the lead CDR sequence comprises at least 5 consecutive amino acid residues in the selected CDR. The selected CDR is selected from the group consisting of V _H CDR1, V _H CDR2, V _H CDR3, V _L CDR1, V _L CDR2, and V _L CDR3 of the lead antibody.

In the present invention, the lead FR sequence also contains at least 5 consecutive amino acid residues in the selected CDR. The selected FR is selected from the group consisting of the lead antibodies V _H FR1, V _H FR2, V _H FR3, V _H FR4, V _L FR1, V _L FR2, V _L FR3, and V _L FR4.

The method further includes the following steps:
A degenerate nucleic acid library containing a DNA segment encoding the amino acid sequence of the hit library is constructed.

  In another aspect of the invention, a method is provided for in silico selection of antibody sequences based on the amino acid sequence of a region in the lead antibody (ie, the lead sequence). The structure of the lead sequence is used to search a database of protein structures for segments with similar 3D structures. These segments are aligned to provide a sequence profile (hereinafter referred to as the “lead sequence profile”). The lead sequence profile is used to search a database of protein sequences of distant homologues of a low lead sequence identity but similar structure. Using this method, a library of diverse antibody sequences can be constructed and experimentally screened in vitro or in vivo for antibody variants with improved or desired functions.

In certain embodiments, the method includes the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Providing a three-dimensional structure of the lead sequence;
Creating a lead sequence profile based on the structure of the lead sequence;
Comparing the lead sequence profile to a plurality of tester protein sequences; and selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments are hit live Forming a rally).

  In the present invention, the three-dimensional structure of the lead array may be a structure obtained from X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy, or theoretical structure modeling.

In the present invention, the step of creating a lead sequence profile includes the following:
Comparing the structure of the lead sequence to the structure of multiple tester protein segments;
Determining the root mean square of the difference in the main chain conformation of the lead sequence and the tester protein segment;
Selecting a tester protein segment having a root mean square difference root mean square of less than 5, preferably less than 4, more preferably less than 3, and most preferably less than 2; and the amino acids of the selected tester protein segment The sequence is aligned with the lead sequence to create a lead sequence profile.

Optionally, the structure of the plurality of tester protein segments is retrieved from the protein data bank.
Optionally, the step of creating a lead sequence profile includes:
Comparing the structure of the lead sequence to the structure of multiple tester protein segments;
Determining the Z-score of the main chain conformation of the lead sequence and tester protein segment;
Selecting a segment of the tester protein segment with a Z score greater than 2, preferably greater than 3, more preferably greater than 4, and most preferably greater than 5; and the amino acid sequence of the selected tester protein segment as the lead sequence Align to create a lead sequence profile.

  Optionally, the step of creating a lead sequence profile is performed by an algorithm selected from the group consisting of CE, MAPS, Monte Carlo, and 3D clustering algorithm.

The method further includes the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed.

Optionally, the method further comprises the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into the corresponding trinucleotide codons; and combining nucleic acid position variants in a combinatorial manner Build a rally.

Any of the above methods further includes the following steps:
Introducing a DNA segment of a nucleic acid or degenerate nucleic acid library into a host organism cell;
Expressing a DNA segment in the host cell such that a recombinant antibody containing the amino acid sequence of the hit library is produced in the host organism cell; and
Recombinant antibodies that bind to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ , preferably 10 ⁷ M ⁻¹ , more preferably 10 ⁸ M ⁻¹ , and most preferably 10 ⁹ M ⁻¹ are selected.

In certain embodiments, the method includes the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure defined as a lead structural template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence profile to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
A scoring function is used to determine whether the hit library members are structurally compatible with the lead sequence profile; and hit library members with scores equal to or better than the lead sequence Select.

  In the present invention, the scoring function is an energy score selected from the group consisting of electrostatic interaction, van der Waals interaction, electrostatic solvation energy, solvent accessible surface solvation energy, and conformational entropy. Function.

  In some cases, the scoring functions are Amber Forcefield (forcefiled), Charmm Forcefield, Discover cvff Forcefield, ECEPP Forcefield, GROMOS Forcefield, OPLS Forcefield, MMFF94 Forcefield, Tripos Forcefield, MM3 Forcefield, Dreiding Forcefield , And UNRES force fields, and other knowledge based statistical force fields (mean fields) and force fields selected from the group consisting of structure based thermodynamic potential functions.

In the present invention, the step of selecting members of the hit library includes:
ΔE _total = E _vdw + E _bond + E _angel + E _{electrostatics} + E _solvation
Select members of the hit library that have a total energy that is lower or equivalent to the lead sequence calculated based on the formula:

In the present invention, the step of selecting members of the hit library includes an improved scoring function ΔG _b = ΔG _MM + ΔG _sol −TΔS _SS
(Where
ΔG _MM = ΔG _ele + ΔG _vdw (1)
ΔG _sol = ΔG _ele-sol + ΔG _ASA (2))
Is used to select members of the hit library that have a binding free energy less than the lead sequence calculated as the difference between bound and unbound states.

The method further includes the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed.

  Route III in FIG. 1C schematically illustrates this embodiment. In this embodiment, the sequence of the hit library is constructed in the 3D structure of the lead protein by substituting side chains from the rotamer database, and the 3D structure of the lead protein (hereinafter referred to as “lead structure template”) Scored for structural compatibility. Based on the structure evaluation, the hit library is reprofiled by ranking according to the energy function score. A portion of the hit library sequence having the desired energy function is selected and back-translated into a library of nucleic acids for in vitro or in vivo functional screening. In this embodiment, there is no amino acid sequence combinatorial step.

Optionally, the method further comprises the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into the corresponding trinucleotide codons; and combining nucleic acid position variants in a combinatorial manner Build a rally.

In yet another embodiment of the present invention, the method comprises the following steps:
In certain embodiments, the method includes the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known 3D structure defined as a lead structure template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR of the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
Creating an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
Combining the amino acid variants in the hit library to create a combination of hit variants that forms a hit variant library;
A scoring function is used to determine whether members of the hit variant library are structurally compatible with the lead structural template; and a hit variant library with a score equal to or better than the lead sequence Select members.

In this method, combining the amino acid variants in the hit library includes the following:
Select amino acid variants with an appearance frequency greater than 4, preferably greater than 6, more preferably greater than 8, and most preferably greater than 10 (2% to 10% of the cutoff frequency, preferably Hit variants that combine the selected amino acid variants in the hit library to form a hit variant library, including 5% if lost after cut-off, including some amino acids from the lead sequence); Generate a combination of

  In the present method, the scoring function is an energy score selected from the group consisting of electrostatic interaction, van der Waals interaction, electrostatic solvation energy, solvent accessible surface solvation energy, and conformational entropy. Function.

  In some cases, the scoring functions are Amber Forcefield (forcefiled), Charmm Forcefield, Discover cvff Forcefield, ECEPP Forcefield, GROMOS Forcefield, OPLS Forcefield, MMFF94 Forcefield, Tripos Forcefield, MM3 Forcefield, Dreiding Forcefield , And UNRES force fields, and other knowledge based statistical force fields (mean fields) and force fields selected from the group consisting of structure based thermodynamic potential functions.

The method further includes the following steps:
A nucleic acid library is constructed that includes DNA segments that encode the amino acid sequences of selected members of the hit library.

  Route IV in FIG. 1D schematically illustrates this embodiment. In this embodiment, after profiling the amino acid sequence of the hit library against the lead sequence, a combinatorial library of hit variants, ie variant library I, the hit variant library II is responsible for the occurrence of the amino acid at each residue position. Constructed based on frequency (as in Route III). The sequence of hit variant library II is built into the 3D structure of the template protein by substituting side chains from the rotamer database and scored for structural compatibility with the lead structural template. Based on the structure evaluation, hit variant library II is reprofiled by ranking according to the score in the energy function. A portion of the hit library II sequence having the desired energy function is selected and back-translated into a library of nucleic acids for in vitro or in vivo functional screening. Additional modifications to the library II variant profile will be applied based on other selective factors determined by those skilled in the art. That is, Library II is a desired library based on evolutionary, structural, and / or functional data.

  Based on the selected hit list or hit variant library II sequences generated in silico, a synthetic library of antibodies is constructed in the laboratory and screened against the target antigen. A variety of biological assays can be used for rapid screening such as phage display (Smith and Scott (1993) Meth. Enzymol. 217: 228-257), ribosome display (Hanes and Pluckthun (1997) Proc. Natl. Acad. Sci. USA 94: 4937-4942), yeast designation (Kieke et al. (1997) Protein Eng. 10: 1303-1310), and other extracellular or intracellular expression systems.

In another embodiment, the method comprises the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure);
Providing a 3D structure of one or more antibodies having different sequences in the _VH or _VL regions other than the lead antibody;
Combining the lead antibody with one or more antibody structures to form a structure assembly (a structure assembly is defined as a lead structure template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
Creating an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
Combining amino acid variants in the hit library to generate a combination of hit variants that form a hit variant library;
A scoring function is used to determine whether members of the hit variant library are structurally compatible with the lead structural template; and a hit variant library with a score equal to or better than the lead sequence Select members.

  Such processes (ie, computer prediction of digital antibody libraries and experimental screening of synthetic antibody libraries) can be repeated to improve the binding affinity of selected antibodies. After the first round of screening, the three-dimensional structure of the selected antibody is computer modeled. In addition, the structure is modified by expanding the sequence and conformation space, and this is subjected to soft docking with a target antigen to create a second generation digital antibody library. The second generation digital antibody library is then screened experimentally to select antibodies that have a higher affinity than the first generation selected antibody. Such an iterative process of modification and screening for antigen mimics the natural process of antibody maturation in vertebrates.

  The conceptual framework and practical application of the present invention are described in detail in the following sections.

1. Conceptual framework of the present invention The present invention provides innovative solutions to problems that have long existed in the field of molecular biology, especially protein folding and design. The approach we have developed combines the best ideas of protein folding and design into a powerful integrated system that develops new protein formulations for high-speed processing and low-cost practical applications. Can do.
We believe that a central challenge in molecular biology is to describe the functional repertoire of biopolymers, such as proteins, RNA and DNA molecules, in their sequence and structure. The functional repertoire of biopolymers is created by a complex interaction of selective pressures during the evolution process and partial constraints on biopolymer folding and stability under various environmental conditions. What is the difference between natural biopolymers and random polymers? What is the best strategy to create new biopolymers with stable structure and correct biological function, utilizing the rich diversity of naturally occurring biopolymer functions, sequences and structural spaces? The answers to these questions are particularly important for molecular design and evolution, especially the discovery of new proteins with increased binding and catalytic activity.

  The present invention addresses this problem in three steps: 1) Considers a general conceptual framework that is the basis for protein folding and evolution, and provides the basic knowledge necessary to understand the present invention. 2) Explain the current experimental and theoretical methods used for protein folding and design and the problems associated with these approaches; and 3) To solve some of the longstanding problems of protein design and engineering The approach of the present invention is outlined.

1) Protein folding and evolution Proteins are the basic molecules for performing diverse biological functions. A protein acquires its biological function by folding its linear array into a unique three-dimensional structure. Predicting protein structure from sequence is an open question. However, significant progress has been made in understanding the mechanism of protein folding, especially with the advent of statistical interpretations of the assembly of intermediates and the transition states of the folding pathway.

  The dynamic nature of protein conformation in solution is well explained in both experimental and theoretical studies. Dynamic changes in protein conformation are allosteric regulation of enzyme activity (Monod, J., Wyman, J. and Changeux, JP (1965) J. Mol. Biol., 12: 88-118), protein-protein and protein -Essential for carrying out some biological functions such as nucleic acid interactions and conformation gating (Zhou, HX, Wlodek, ST, McCammon, JA (1998) PNAS 95: 9280-9283) .

  The continuous assembly approach not only gives a more realistic view of biopolymers compared to static X-ray structures, but also explains the increasing experimental observations that are otherwise impossible to explain The continuous assembly approach is superior to the classical discontinuous approach to explain protein folding mechanisms (Hong Qian (2002) Protein Science 11: 1-5). This view emphasizes the importance of using the statistical nature of the continuous distribution of conformational sets in the energy context to understand the biological function of macromolecules (Baldwin RL (1995) 5: 103 -109 J. Biomol. NMR; Pande VJ et al. (1998) Curr. Opin. Struct. Biol. 8: 68-79).

  The random energy model (REM) used to study the freezing and design of heteropolymers is an excellent approximate physical model for protein folding and design (Vijay S. Pande, Alexander Yu. Grosberg, and Toyoichi Tanhaka) , Review of Modern Physics, Vol. 72, No. 1, 2000 and references therein). Much has been learned from quantitative studies of simple models of protein folding and design based on the statistical nature of the freezing transition of heteropolymers. The phase transitions between a set of conformational states distributed in a continuous energy spectrum are associated with protein folding compared to the traditional view of several discrete states in a set of well-defined energy wells. And give a more realistic explanation of connectivity. The REM background shows that the necessary and sufficient conditions for a given sequence to fold into a kinetically stable conformation show a continuous energy spectrum in the upper part and in the lower part. It suggests that the energy distribution shows a remarkable energy minimum (Vijay S. Pande, Alexander Yu. Grosberg, and Toyoichi Tanhaka, Review of Modern Physics, Vol. 72, No. 1, 2000 and references therein; Shakhnovich And Gutin, 1993 PNAS, 90: 7195-7199). Therefore, the array should be designed to widen the energy gap between the ground state of the specified array and the bottom of the REM continuous energy spectrum. The energy gap is widened by reducing the energy of the native conformation of the sequence (positive design for stability) or boosting the energy of the alternative conformation of the sequence (negative design for specificity) Is done.

  The general rules derived from this simple model of protein folding were strictly followed in protein design by recent new computers: amino acid composition does not change, but energy is minimized (Koehl P and Levitt M ( 1999) J Mol Biol 293: 1161-1181). It is said that defining the collective properties of sequences that conform to a certain structure is more important than finding specific optimal sequences (Koehl P and Levitt M (1999) J Mol Biol 293: 1183-1193) . Multiple alignments of designed sequences define the sequence space measured by information entropy; a subset of this sequence space is similar in size to that derived from the same structural alignment observed in nature (Koehl P And Levitt M (2001) PNAS 1-6). This study defines a folded sequence space that is topological and stable, and shows that a subset of the sequence space can be defined by functional agreement. However, this method places too large a limit on the choice of amino acids at each position by not changing the amino acid composition.

The dynamic nature of protein evolution has been actively studied by theoretical and evolutionary biologists (Maynard-Smith, J (1970) Nature, 225: 563-564). Mapping (genotype) sequences to values that measure congruent background is a central issue in evolutionary biology. The relationship between genotype and phenotype is generally too complex to analyze in a quantitative manner, but this relationship is simplified to the relationship between sequence (genotype) and structure (phenotype) and can therefore be used to Sequence matches can be scored into certain biopolymer forms as follows:
Genotype (sequence) ← Match score → Phenotype (structure)

  Proteins observed in nature have evolved under selective pressure to perform specific functions. Interestingly, the functional protein identity background is mapped and simulated using similar means such as the protein folding field. The matching background is mapped in sequence space to define a collection of variants that enhance the functional properties of the protein. Statistical properties of sequence sets have been used to describe neutral networks in the sequence space of target proteins (Stadler PF. Journal of Molecular Structure (Theochem) 463, 7-19 (1999); J Theor Biol 2001 , 212: 35-46).

  The background theory embeds three basic components: a set of orientations; a matching function assigned to each orientation; and a relationship between orientations or orientations that define the distance. A matching function is broadly defined as the nature of a protein (eg, binding affinity between two proteins (receptor and ligand; antigen and antibody), catalytic activity of an enzyme, or structural stability of a target scaffold).

  From an evolutionary point of view, the coincidence background generated by mapping the sequence-structure relationships of natural RNA and proteins predicts the existence of a neutral network of sequence space evolved under a partially correlated background, leading to a new coincidence function. Provides an efficient path to match evolution. In contrast, random sequences that have evolved with a coarse coincidence background without neutral neighbors are captured in the local optimum and lead to a local population in the sequence space. Natural sequences have undergone evolutionary optimization under selective pressure in the mountaineering process. An effective route to new matching functions by sequence modification is to follow a neutral network in space rather than random mutation (Stadler P F. Journal of Molecular Structure (Theochem) 463, 7-19 (1999); J Theor Biol 2001, 212: 35-46; Aderonke Babajide et al. (1997) Folding & Design 2: 262-269). Searching for matching backgrounds via point mutations versus the relative efficiency of genetic recombination in the protein space can be simulated and compared using REM as well as heteropolymer-based models (Bogarad L, Deem MW (1999) PNAS 96: 2591-2595; Cui Y, Wong WH, Bornber-Bauer E, Chan HS (2002) 99: 809-814).

  The above theoretical study of protein folding and evolution using a simplified model provided information about the statistical properties of protein structure and sequence aggregation during folding and evolution. We believe that the theory that combines the concepts of molecular biology, spin glass physics, and heteropolymer physics gives a unified framework of biopolymer dynamics. The question here is how to translate such a conceptual framework based on a protein model into a practical approach that maps the functional background of a protein in both sequence and structure space.

2) Modern experimental and theoretical methods of protein sequence design in the field and the problems underlying them The main goal of protein engineering is to create proteins with new or improved functions. For this, two approaches have been used to obtain proteins (mainly enzymes) with the desired properties: in vitro directed molecular evolution and structure-based computational design. In vitro directed evolution approaches use homologous sequences, random mutagenesis and gene shuffling to create a diverse sequence library. Variants with the desired properties are selected and reshuffled by rapid screening. This method is repeated until the desired level of functional enhancement is achieved.

  The first rule of command evolution “getting screened” emphasizes the importance of screening methods in assessing functional identity of protein libraries (Wintrode, P and Arnold, FH (2000) Adv Protein Chem. 55: 161-226). The availability of high-throughput enzymatic screening and improved sensitivity have led to some success in directed evolution. Compared to rational engineering, directed evolution requires little or no additional information (eg, the structure of the target enzyme) and can be screened for biological activity from a large pool of molecules under a defined selective pressure. it can.

  The dependence on screening ability places an upper limit on the size of the combinatorial library that is created, and thus places an upper limit on the size of the functional space to be tested. Random mutagenesis using error-prone PCR is a biased and inadequate process for creating diverse libraries, so the potential for significant functional improvement from a single random mutation is small. It decreases rapidly for multiple simultaneous random mutations. In addition, it is difficult to simultaneously create several variants at a single codon position at the nucleic acid level.

  Furthermore, the dependence of DNA shuffling on homologous recombination of sequences with high homology (> 70%) limits the sequence space that the resulting library can cover. As a result, each successive iteration of shuffling and screening leads to a test in the shrinking local sequence space. This is efficient for identifying new homologous sequences with enhanced properties, but may not be appropriate for identifying truly novel sequences with greater functional improvements.

  Nevertheless, by incorporating random mutagenesis, effective amino acid substitutions are made and identified. Accumulation of effective point mutations has been successfully used to evolve and screen many important enzymes with the desired properties. In addition to simple random mutagenesis strategies, DNA shuffling (including a family shuffling approach that combines genes from multiple parents of the same or different species) produces a highly improved biocatalyst (Ness JE Del Cardayre, SB Minshul, J & Stemmer, WPC (2000) Adv Protein Chem 55: 261-292).

  As a problem closely related to protein folding, protein design is considered as a reverse folding problem (Drexler, KE (1981) PNAS 78: 5f275-5278; Pabo, C. (1983) Nature 301: 200) (target structure) Find an array that gives Designing protein sequences that provide a target scaffold is considered an important step in manipulating proteins with improved properties for a wide range of applications.

  A major challenge associated with reverse folding protocols is the need to maintain a rigid protein backbone. Because of the large conformational space that needs to be tested, for practical reasons, the static X-ray structure of proteins is still used as a starting point for rational structure-based protein or drug design . The reverse protein folding approach seeks to calculate an optimal sequence that fits the protein structure based on semi-empirical all-atom energy functions describing interactions between amino acids. Native proteins are known to tolerate small variations due to robust conformational adaptation, but the computational ground state of the rigid protein backbone is sufficient to allow small variations in the protein backbone or side chain rotamers. It is unacceptable and does not provide a sufficient measure of stability.

  To tackle these challenges, several attempts have been made to display skeletal parameters by adjusting the relative orientation between regular secondary structures (Harbury, PB, Tidor B. & Kim, PS (1995) Protein Science 92, 8408-8412; Su A & Mayo SL (1997) Prot Sci. 6, 1701-1707; Harbury PB, Plecs JJ, Tidor B, Alber T, Kim PS (1998) Science 282, 1562- 1467). We find that a simple but efficient solution to alleviate local constraints is the present invention for protein loops (which are irregular and whose backbone movements are generally difficult to parameterize). As evidenced, it is considered to be the minimization of energy including the backbone and side chains for any type of protein structure (Keating AE, Malashkevich VN, Tidor B, Kim PS (2001) PNAS 98, 14825-30).

  Apart from some cases, such as regular secondary structure (see below), most protein design strategies are defolding protocols in sequence selection to reduce the enormous task of searching the conformational space. Strictly follow. A powerful search algorithm (in order to find the best solution to the empirical energy function that incorporates various factors to stabilize proteins assembled from protein side chain rotamer libraries, even in the case of immobilized scaffolds ( Stochastic Monte Carlo methods or genetic algorithms and deterministic dead end exclusion are required (Ponder, JW & Richards, FM (1983) J. Mol. Biol. 193, 775-791; Hellinga, HW, Richards , FM (1994) PNAS 91, 5803-5807; Desjarlais, JR & Handel, TM (1995) Prot Sci. 4, 2006-2018; Dahiyat, BI & Mayo, SL (1996) Prot. Sci. 5, 895-903 ).

  For amino acids exposed on the surface, evolutionary pressure plays a greater role in determining sequence selection than in the core region where filling constraints lead to conserved amino acid selection. However, low surface physical constraints, highly variable charge and polar solvation interactions pose difficult design issues for exposed side chains. This limitation almost limits the protein design method to the protein core, since steric constraints are a major determinant in designing amino acids at the core position of the protein.

  Some algorithms attempt to divide proteins into discrete regions (eg, cores, boundaries and surface residues) and have different scoring functions for different parts of the protein structure (Dahiyat, BI & Mayo, SL (1996) Prot. Sci. 5, 895-903). However, for protein-protein interactions, critical residues are likely located on the surface of the protein and above the loop of the protein (the most difficult or irregular structural class of proteins). Interactions between proteins can cause some of the interacting residues to be buried or halved, making it difficult to model the interaction as a specific class of residues in discontinuous regions of the protein. ing. Although we are widely involved in protein loops to mediate protein-protein interactions (eg, the interaction of antibody CDRs with antigens or cytokines and their receptors), The method believes that, unless a good homology model and database information are combined, it will not be possible to accurately predict protein loop structure interactions using only a force-field-based approach ( van Vlijmen HW, Karplus M (1997) J Mol Biol 267, 975-1001).

  Because the current force field is insufficient to predict protein folding, the persistent problem of protein folding and design captures and predicts all factors known to contribute to protein stability. Is to develop an energy function that agrees well with experimental data. No matter how sophisticated this operation is, it is inherently difficult and error prone to calculate the small difference between two large numbers of stability in the protein's folded and unfolded states. If the region of interest is at the boundary between two proteins with positive and charged residues (the force field parameter is currently under active investigation for accurate assessment), this difficulty It gets bigger. The scoring function may also overfit the experimental feedback from a particular test system. Briefly, compared to core packing in proteins, accurate calculation of interactions between polar and charged residue dominant proteins is still a difficult task in this field. The inventor believes that the side chain placing algorithm that has proven very effective in packing the hydrophobic core of a protein may not be an effective answer to this continuing problem. thinking.

  We emphasize that the use of fixed scaffolds in the reverse folding protocol also over-limits the position of the side chain rotamers and the steric repulsion between them. Such rigid limitations on side chain rotamers are impractical. The actual protein will accept side chain mutations or rotamers due to dynamic variations in solution reminiscent of a modified set of conformational states. Note that the parametric representation between the regular secondary structural elements has been used to advance the overall folding of the protein backbone (Harbury, PB, Tidor B. & Kim, PS (1995); Su & Mayo (1997) Prot Sci .; Harbury PB, etc (1998) Science 282, 1562-1467). However, it is still difficult to use such an approach for non-regular secondary structural elements (eg loops) to account for the varying collective states.

  Due to the limitations of computational methods, tedious evolutionary protein designers have chosen to avoid any rational structure-based approach and have invented a set of powerful experimental tools. However, no matter how powerful it is, it is extremely inefficient to create a diverse library by random mutagenesis and to screen them experimentally. On the other hand, recombination of homologous genes by DNA shuffling only allows limited testing of sequence and structural space.

  The present inventors believe that a computer method without physical constraints in advance can search a much larger sequence space. Furthermore, the main advantage and main driving force of the rational approach is that the sequence can be designed and controlled at all stages prior to experimental screening. This allows protein designers to make larger practical jumps in protein sequence space that test larger distances that may lead to the discovery of new sequences with little or no homology to the starting sequence To do. Furthermore, the actual size and direction of these “jumps” can be controlled consistent with experimental feedback to follow the functional background to the new peak. This capability is expected to rise dramatically with the enhancement of computer capabilities and the development of new algorithms and new software tools.

  Obviously, the computer's ability itself does not make computer protein design superior to in vitro protein evolution experiments unless small but important structural changes are understood and captured. For example, effective mutations are generally not localized at the catalytic site, but have been shown to be distributed in large parts of proteins with variable protein backbones (Spiller B, Gershenson A, Arnold FH, Stevens R. (1999) PNAS 96, 12305-12310).

  With current technology, experimental screening for biological activity is still the only reliable approach available to evaluate the biological function of molecules controlled by complex competitors under experimental conditions. It is extremely difficult to pinpoint the answers without capturing extensive details at the same time using computer methods and without extensive experimental testing. Furthermore, most scoring functions can only calculate stability, not activity or specificity.

  Some statistical-based approaches have been developed that shed light on evolutionary sequence design. Using a simplified model that resembles a random energy model in protein folding, Bogarad and Deem show that DNA exchange of non-homologous DNA segments with low energy structure is more than genetic recombination of homologous DNA by DNA shuffling. It was far more efficient in searching for protein space matching backgrounds, which proved superior to point mutations (Bogarad L, Deem MW (1999) PNAS 96, 2591-2595). Recently, heteropolymer-based models have been used to successfully map consensus background sequence-structure relationships in a structure-based evolutionary approach (Cui Y, Wong WH, Bornberg-Bauer E, Chan HS (2002) 99, 809-814). Point mutations have been found to lead to evolutionary background diffusive walks, where crossing occurs through a reduced identity barrier. The smoothness of the energy or matching background, along with the ratio of crossover and point mutation rate, determines the effectiveness of the crossover in protein sequence and structural space testing. That is, the inventors believe that evolutionary sequence design should not be limited to point mutations and homologous genetic recombination.

  Experimental feedback also indicates a predicted improvement in protein properties and is essential to improve the agreement between theoretical predictions and experiments (Desjarlais, JR & Handel, TM (1995) Prot Sci. 4, 2006-2018; Dahiyat, BI & Mayo, SL (1996) Prot. Sci. 5, 895-903; Keating AE, Malashkevich VN, Tidor B, Kim PS (2001) PNAS 98, 14825-30). That is, the present inventors have confirmed that experimental values and computer values agree (Keating AE, Malashkevich VN, Tidor B, Kim PS (2001) PNAS 98, 14825-30). Unless proven (including various regions of polar and charged residues), we believe that experimental libraries should not be limited to sequences around global optimization or sub-optimization solutions from computers. Instead, an experimental library should be constructed that covers a wide range of dispersal to an energy background that is as good or better than the lead sequence.

  In vitro command evolution and computerized sequence design are beginning to converge. For example, newly designed enzymes based on structures are usually not very active (Benson, DE, Wisz, MS & Hellinga HW (2000) PNAS 97, 6292-6297; Bolon DN, Mayo SL (2001) PNAS 98, 14274- 14279). However, these new designs of sequences in different scaffolds are the starting point and undergo directed evolution for improved activity (Altamirano, MM, Blackburn, JM, Aguayo C, Fersht AR (1000) Nature 403, 617-622). Conversely, structure-based computational methods can be used to identify possible sites of point mutations enriched in evolutionary design to reduce the search space in directed evolution, but these The site is known to be different from that of sequence profiling (Voigt CA, Mayo S, Arnold, FH & Wang ZG (2001) PNAS 98, 3778-3783).

  However, the present inventors believe that the strategy for command evolution should be analyzed and quantitatively measured before tedious laboratory work. Several steps have been taken to computer simulate DNA shuffling in order to optimize possible experimental conditions and possible limits for augmentation (Moore, GL, Maranas CD, Lutz S, Benkovic S (2001) PNAS 98, 3226-3231). Because there is a huge protein space that can be searched by different approaches, it is important to determine the optimal path for the specific problem at hand by comparing the efficiencies and limitations inherent in each experimental or computational approach. is there.

  We also believe that for structure-based protein design, the core of the problem is a deterministic approach to complex problems using unrealistic processes. It is known that the interactions that stabilize proteins are very complex. The static structure used in the design is the collective average of dynamic variations in solution that are observed to interact with or change with other proteins or ligands. Thus, the idea of finding the optimal solution for a target function is an interesting theoretical task, but may not be very relevant or realistic for actual biological problems. Strict limitations, or both, using energy function defects or rigid frameworks will contaminate the "optimal solution" to the design problem. That is, again, we should limit the experimental library to an array around a globally optimal or suboptimal answer from a calculation that may be biased by the process and parameters used in the calculation. I don't think so. Instead, sequences that cover a suitable range (eg, superior or equivalent score to the lead sequence) should be used for experimental screening.

  For evolutionary protein design, the current approach to designing proteins as biocatalysts (eg enzymes) is still art rather than science. However, some methods are robust enough and can be applied directly to solve the public problem of commercial biocatalyst design. DNA recombination by DNA shuffling and random mutagenesis has provided a diverse protein library for functional screening, but more efficient methods for library creation should be explored and the process Will be predictable and routine rather than relying solely on the final screening results. So far, command evolution has been best applied to the design of biocatalysts because it is easy to perform high-speed screening for enzyme activity that can easily detect chemical reactions.

  However, the inventors believe that the unexpected answer provided by directed evolution in which mutations are distributed across the entire protein sequence also raises the problem of evolving some pharmacologically related proteins. In therapeutic antibody design, mutations that need to be limited to a few regions (eg, modifications to CDRs and previously inactive framework regions) may render the protein immunogenic. Such undesired variants during experimental shuffling must be minimized or eliminated by the reverse cross method; if possible, removal of these immunogenic variants can result in improved activity resulting from rigorous experimental attempts. I hope you never waste.

  Rational structure-based protein design has evolved rapidly in its development and has begun to give inspiring results. Over the years, repacking the hydrophobic core (Malakauskas, SM & Mayo, SL (1998) Nature Struct. Biol. 5, 470-475) and discovering new scaffolds not observed in nature (Harbury PB et al. ( 1998) Science 282, 1462-1467) by computer of protein variants with target scaffold (Dahiyat, BI & Mayo, SL (1997) Science 278, 82-87) and significantly improved thermal stability There have been exciting advances in design. For biological activity and affinity design, this rational approach can be extended to bind sites in three different conformational states: open, apo, and closed ligand binding states that can modulate binding activity via allosteric action on the binding site. An interesting development has been made to influence binding affinity by designing residues around (Marvin, JS & Hellinga HW (2001) Nat Struct Biol 8, 795-798). However, for most proteins of biological and medical interest, the structural information necessary for such a design is not yet available or has a low resolution that is insufficiently low for such a design. It is thought to increase the structural information at an accelerating rate.

3) Approach of the present invention The present invention provides an innovative approach to efficiently map the distribution of coincidence and energy background in protein space and structure space by using a set-based statistical method.

  Due to inadequate knowledge of the underlying principles of protein folding and design, a set-based statistical approach to protein combinatorial libraries fits a structure or family of structures and has an energy background that scores better than the lead sequence. Try to design an array set that covers the distribution. This is statistical because it is not a specific optimal solution to a fixed structure that is designed, but a distribution of sequences or structures. This is a collection-based because it is the structure / sequence collection targeted by the nucleic acid library, not the specific sequence or structure.

  The present inventors believe that the division of the energy distribution function into different sets in the sequence space allows for effective testing by subsequent experimental methods. This statistical approach for mapping the functional space of selected protein sequences provides a means of selecting the actual protein sequence of biological interest in terms of the above consensus background. By defining the statistical properties of a set from a single optimized sequence or a group of sub-optimal sequences, protein designers can be trapped or misled by biased solutions caused by limitations inherent in current computational methods It is likely to avoid moving in the direction.

  The approach of the present invention is developed by combining knowledge gained from theoretical studies of simple models of protein folding and evolution based on our understanding of problems associated with existing methods in the field. Has been. Through research and experimentation, the inventors have developed realistic solutions to the problem of purification folding, manipulation, and design, particularly in the exciting field of antibody manipulation.

  FIG. 2A outlines the in silico biopolymer evolution system developed by the inventors. As shown in FIGS. 2A-C, the pathway from the initial target biopolymer (eg protein) to the final candidate sequence with the desired function passes through three biologically important spaces: sequence, structure And functional space.

  In sequence space, lead sequences are used to search a database of evolutionarily related sequences. This search is applied to the structure space, and if structure alignment is used, more distant sequences are obtained. The hit library variant profile lists the amino acid frequency and variant at each position.

  In structural space, hit variant libraries are created in silico based on reduced variant profiles and partitioning (FIGS. 1C, 1D and 2A-C) or complete sequence libraries, or random combinations thereof (FIG. 1E). ~ H, 2A and C). This hit variant library or random / complete sequence library is scored using structural templates, a suitable set of sequences is selected, re-profiled, and an extended nucleic acid (NA) library is generated in silico. Generate. If the size of the in silico NA library is evaluated and the library size is acceptable, proceed to oligonucleotide synthesis. Otherwise, the hit variant library is subdivided into small segments and a small NA library with overlapping sequences is generated to maintain sequence and structural correlation between the resulting libraries (Example column). And see Figures 28A-C).

  The NA library is screened experimentally in the functional space. Positive sequences are entered back into the computer cycle to improve the library. Strong positive clones are moved to further evaluation and potential drug development. If none of the experimental screening hits, a new set of lead sequences and / or variant profiles for structure-based scoring are selected for the target system and the process is restarted.

  As is apparent from the description of FIG. 2A, the key difference between the approach described herein and other evolutionary sequence designs from other methods in the computer field is that the present invention makes the best of both worlds. In combination, the matching background is searched continuously, and the structure space is searched more efficiently. Our approach combines evolutionary information in a protein sequence database with protein physical constraints (eg, compatibility with the 3D structure of the sequence). The biological function of a protein can be assessed computationally by examining a limited set of sequences that satisfy both evolutionary selection in sequence space and physical constraints in structural space.

  In certain applications of the methods of the invention, antibodies are used as model systems for both experimental and computer testing. Antibodies are widely used in research, diagnostics, and medical applications. Antibodies bind to various targets with good specificity and affinity. Catalytic antibodies that catalyze chemical reactions have also been developed.

  In more specific applications, antibody hypervariable loops or complementarity determining regions (CDR) as well as framework regions (FR) are targeted. CDRs determine antibody antigen binding and specificity, and the framework regions provide a scaffold where the CDRs are correctly positioned for biological function. Antibody molecules are well suited for manipulation because of their molecular structure, and CDRs and framework regions are defined in a sequence and structure.

As outlined in FIG. 1A (Route I), polypeptide segments in the expressed protein database are computer screened against a specific region of the lead antibody to be optimized (eg, V _H CDR3) and read Those having a sequence pattern that matches the antibody are selected. The selected sequences form a hit library.

  As further outlined in FIG. 1B (Route II), a variant profile can be created by listing amino acid variants at each sequence position from the hit library, along with the number of occurrences in the hit library. The combinatorial calculation of this profile shows hit variant library I. This variant profile includes amino acids from the lead sequence or sequence profile at the corresponding position (where they are missing from the hit library) or exclude amino acid variants below a certain cutoff frequency Edited by or both. The resulting variant profile defines the hit variant library II (designated library).

  As outlined in FIGS. 1C and 1D, each member of the hit variant library I or II is “grafted” onto the corresponding region of the lead template structure or model (if available) and is scored using a scoring function. Selected for those that are structurally compatible with the rest of the 3D structure. Optionally, the hit variant library can be evaluated in the presence or absence of the target antigen. Antibodies with a favorable score are selected and screened experimentally for actual binding affinity to the antigen in the laboratory. As described in the Examples section, this approach is used to select a number of antibodies against human vascular endothelial growth factor (VEGF) and prove that they can bind to the target antigen VEGF. Some of these show higher affinity than lead antibodies (Figures 30 and 36).

  As will become more apparent in the further disclosure in the following section, the approach provided by the present invention is not only conceptually distinguishable from that in the art, but also has many practical advantages in antibody manipulation.

  By utilizing expressed protein sequences collected in protein databases, this approach not only effectively mimics the natural process of affinity maturation in silico, but also evolves proteins with improved binding affinity. Can be dramatically accelerated. For example, using any set of amino acid sequences (including, but not limited to, sequences of immunological interest from various species), a variety of library profiles for lead sequences for CDR affinity maturation Sex can be maximized. However, to minimize the possibility of immunogenicity, sequences of human germline and / or human origin are used to profile against the lead sequence of framework regions of humanized or framework designs Should. That is, selection of a database based on its application, size and origin of species (eg, human, mouse, etc., or all available species) allows flexibility and control of the designed protein.

  Further, this approach involves modeling protein variants from time to time in the presence of target molecules (eg, lead antibody antigens) if complex structures or models are available. By including the antibody-antigen interaction in the calculation, the screening process may more closely mimic the natural process of affinity maturation as an antigen-directed process, and the calculated binding affinity may correlate well with experimental values. Absent.

  In addition, the methods of the present invention have computerized prediction of antibody libraries (which is biased for specific target molecules or antigens if complex structures or structural models are available) and have high binding affinity for antigens Combined with experimental screening of libraries to select ones. Such a process can be repeated to improve the binding affinity of the selected antibody. If high-affinity complex structures are available as templates, hit variant libraries are pre-screened by computers to reduce the library size and traditionally created by complete randomization of amino acids at each position of the lead antibody. Higher focus can be maintained compared to the library. Predicting and building hit variant libraries in silico can accelerate the entire process of protein evolution and effectively mimic the natural process of antibody affinity maturation at high speeds.

  In a preferred embodiment, the lead protein is an antibody or an immunoglobulin and the target molecule is an antigen that binds to the template antibody. The lead protein is any protein, preferably a protein having a known three-dimensional structure that is resolved by X-ray crystallography or nuclear magnetic resonance spectroscopy. Alternatively, the 3D structure or assembly of template proteins is provided by computer modeling using algorithms known in the art.

4) Comparison of the method of the invention with other methods in antibody selection and manipulation The selection of antibodies from a very diverse library makes it possible to cover a wide range of sequences, thus increasing the chance of finding the optimal sequence . However, for example, for antibody sequences obtained from random mutagenesis of a lead antibody in a CDR, not all structures of the randomized CDR are compatible with the 3D structure of the lead antibody. Using a protein sequence expressed opposite to the protein sequence from random mutagenesis and filtering incompatible sequences using the method of the invention, a smaller number of sequences are selected. As a result, the sequence space of the antibody being screened is reduced in size without losing sequences that are highly relevant to affinity binding maturation and stabilization of the mutant antibody.

  In contrast, current methods in the art for constructing antibody libraries include immunohuman antibody gene pools, inexperienced B cell Ig repertoires, or in-vivo of cDNA libraries from specific germline sequences. Includes in vitro isolation. Barbas and Burton (1996), supra; De Haard et al. (1999), supra; and Griffiths et al. (1994), supra. These libraries are very large and the antibody sequences are very diverse. Such conventional approaches attempt to mimic an immunological response to an antigen in vivo by creating a library of as diverse a antibody as possible. Typically, large libraries of these antibodies are displayed on the phage surface and screened for antibodies with high binding affinity to the target molecule. Such “fishing in a large pond” or “finding needles in a large haystack” approach can result in a simple increase in the size of the sequence repertoire that lifts antibodies that can bind to the target antigen with high affinity. High, but based on the assumption that it is inefficient due to insufficient testing, insufficient diversity, and indeterminate library composition.

The inventors believe that there are several problems associated with such conventional approaches. A simple increase in the size of a sequence library may not necessarily correlate with an effective increase in functional diversity. Furthermore, because of the physical limitations to creating extremely large experimental libraries, it may be very difficult to construct libraries with diversity exceeding 10 ¹¹ in vitro. A library that is actually screened experimentally provides only a portion of the sequence repertoire with a theoretically predicted size. Furthermore, due to the difficulties and under-representation associated with handling and handling very large libraries in vitro, attempts to increase the size of the library lost time and money, yet functional diversity was significant. There is a natural concern that it will not increase.

Another approach that exists in the art is to engineer artificial antibody libraries and then construct synthetic antibody libraries that are expressed in bacteria. Knappik et al. Artificial antibody libraries were designed based on consensus sequences for each subgroup of heavy and light chain sequences according to the germline family. The consensus was automatically weighted according to usage frequency. By searching against a population of realigned sequences, the most homologous realigned sequence of each consensus sequence was identified, and all positions where the consensus differed from this closest realigned sequence were examined. In addition, models were created for 7 _VH and 7 _VL consensus sequences and analyzed for their structural properties.

  However, as far as the therapeutic application of the selected antibody is concerned, there are several problems with such an approach. Although the definition of consensus sequence is too liberal, such artificial sequences as defined may not represent natural functional structures, but experimentation and structural analysis eliminate some unfavorable amino acid combinations. Consensus sequences are designed primarily to cover human germline sequences that are often used in realigned human sequences, but this is a limited consensus sequence library that has been contacted so far in the course of evolution. There may be a bias towards a specific number of antigens. These library construction methods primarily focus on finding lead antibodies or hits from large antibody libraries, but for affinity maturation, most of the above approaches are still very limited for antibody affinity maturation. Yes. More classical approaches (eg, CDR walking, random mutagenesis, stepwise saturation mutagenesis at each position of the CDR, etc.) have been used for antibody affinity maturation. The present invention is designed to design a library with an affinity maturation bias.

  We believe that examining the functional space by mapping structures from different species covers a wide range of functional CDRs of an antibody library and can extend the range of antigens that it can bind. . This approach will be very important in the design of antibody libraries to target novel antigens. The methods of the invention typically rely on structural constraints from antibodies or other natural sources. In the present invention, the complete sequence space of all available proteins (preferably antibodies), including those from humans and other species, is analyzed by fitting each library sequence to the 3D structural framework of the lead antibody. be able to.

  Based on this analysis, the resulting mutant antibody is not only novel in sequence, but also has a higher affinity than the lead antibody. As shown in the Examples section below, a number of mutant antibodies are selected using the method of the present invention and experimentally shown to bind to human VEGF with an affinity equal to or stronger than lead anti-VEGF antibodies. Proven to.

2. Detailed Description of Methods Used to Implement the Protein Design Strategies of the Invention This method involves exploring sequences, structures and functional spaces and evaluating relationships between them (FIGS. 1A-D, 1E-H, 2A-C). The starting point is the lead structure and / or the lead sequence, if available. The method systematically examines both sequence space and structural space to identify variant profiles that are optimized for functional screening. There are three modes of information exchange: i) separate evaluation and then combination of information in the sequence and / or structure space, ii) sequence to structure, or structure to sequence, or iii) sequence or structure alone Continuous evaluation of. Sequence design is examined separately in sequence space and structure space (two separate cycles), but variant profiles from these two separate cycles can be compared and combined to produce strong candidates in functional screening An optimal overall variant profile with a good and good consensus variant profile can be reached.

  The two starting points are functionally related to each other, either as a result of comparing the target sequence with a homologous sequence, or by a structural alignment of known homologous structures, resulting in a sequence profile. Sequence profiles are also obtained from mutation data that suggest functional or structural information. Similarly, structural aggregates are created by molecular dynamic simulation, but can also be obtained from sequence alignment of known structures or from homologous based modeling.

  The variant profiles that arrive at each cycle are being compared for further refinement and / or transferred to other cycles, so the two filtering and refinement cycles in the sequence and structure space are in the filtering and evaluation process. More relevant. For sequence-derived variant profiles, this is structurally evaluated for known templates in the structure space to rank and improve variant profiles. Conversely, structure-derived variant profiles are moved into sequence space for comparison and partitioning to control whether they belong to the same superfamily of hit or variant libraries, or the final library size. Can be evaluated.

1) Sequence space In sequence space, the goal is to determine a variant profile that is optimized for the target function. This cycle begins with the identification of hit libraries by database sequence search and alignment using sequence profiles. This is a probabilistic approach like a simple BLAST search or profile HMM. Based on variation within the hit library, the sequence is filtered and partitioned. This is done by evaluating the amino acid frequency and distribution at each position. Residues that usually have the highest frequency at each position as well as residues from the target sequence are included in the variant profile. Variant frequency distributions, or cut-off values (5% or more, respectively) that depend on amino acids that are ranked relatively higher at each position can be included in the variant profile.

  Partitioning may be necessary to place practical limits on the final size of the oligonucleotide library. Splitting is determined by calculating the size of the oligonucleotide library as a function of the degenerate nucleic acid library of various variant profile segments. That is, highly variable variant profiles can be resolved such that the size of the resulting oligonucleotide library can be set within the limits for effective and efficient experimental synthesis, transformation and screening.

  Another partitioning scheme is to use structural correlation information. Since peptides that fold in three dimensions interact with distant segments in order, structural templates or models can be used to assign structurally correlated sequences for partitioning. For example, the ends of the loop are correlated, but the vertices themselves do not interact much with the ends. In such cases, the variant profile is divided into at least two profiles: one for the two ends and one for the vertices.

  One or both approaches are used to highly divide the variant profile. When splitting, there should be at least 2, preferably 3 or more residue overlaps between segments so that some structural correlation is maintained between adjacent segments. One or both approaches are used to achieve a functionally optimized oligonucleotide library size.

  Once the sequence variant profile is determined, the library is screened by computer using known structural templates or homology-based models and scoring functions (described below). This ranking filters and reduces the variant profile by identifying preferred variants while filtering out unwanted variants, thus simultaneously concentrating and reducing the size of the experimental library.

2) Structure space In structure space, the goal is to determine the variant profile that is optimized for the target function, but start with one structure or set of structures and then based on the average of the set of structures To score the sequence. This cycle begins with a set of structures and related sequences that can be screened by a computer and evaluated using a scoring function.

  For a theoretical ideal scoring function that can account for all physicochemical variables, the energy score ranking will fully correlate with the functional ranking. This is not possible and not practical on a computer, and an incomplete scoring function must be used where the structure or sequence is poorly correlated with function. Since the goal of the design protocol is to identify a set of possible sequences that have the desired function, a scoring function that is incomplete but correlates sequence and structure with function can be used.

  Such scoring functions include any combination of computer terms and correlate or map functional values to sequence or structure values. A simple example is the van der Waals energy that correlates the hydrophobic packing function with a sequence having the appropriate density of aliphatic or aromatic side chains. Another example is an enzymatic hydrolysis activity that correlates with the presence of a nucleophilic side chain group at a particular position in the sequence.

  In general, the scoring function will be based on the sum of thermodynamic energy, including some or all of the contribution terms that correlate with the structural stability and function of the protein. Most commonly, these will include electrostatic solvation energy, nonpolar solvation energy, and side chain and backbone entropy. MM-PBSA or MM-GBSA solves standard terms calculated using molecular mechanics (MM) force field, solves Poisson-Boltzmann (BP) equation, or generalized Born (GB) approximation. The solvation term, including electrostatic solvation using the continuous solvent model calculated using, and the solvation term accessible by the solvent based on its ratio to the surface area (SA), conformation entropy (skeleton and side With the contribution from (including chain). Good correlations have been reported between experimental values and values calculated by MM-PBSA based on aggregate structures obtained from molecular dynamic simulations (Wang, W, Donini O, Reyes CM, Kollman PA (2001) Annu Rev Biophys Biomol Struct 30, 211-43). This improved scoring function based on MM-PBSA is a simple scoring function based on the total energy of the Amber94 force field performed at CONGEN used to scan the sequence library for compatibility with the template structure Was used to evaluate (see, eg, FIG. 12). A comparison of the simple scoring function used here with the improved scoring function for the lead sequence hit library using one template structure (1cz8) (Figures 12D and E) Suggests that the function correlates with the improved scoring function, but that significant variance in the correlation map can be an improvement over a simple scoring function to improve matching with the improved scoring function To suggest.

  Compared to other scoring functions used in protein and drug design, MM-PBSA or MM-GBSA is a better physical model for scoring and handles various problems equally, It is computationally expensive because it requires multiple orbits from the explicit molecular dynamics simulation of water to calculate the collective average. This method is useful for studying some difficult mutants beyond simple scoring methods and serves as a control to evaluate the methods used in high-speed computer screening.

3) Optimized variant profile The first result of the design protocol is the optimized variant profile. This embodies the results of both sequence and structure evaluation so that evolutionary and structural choices are incorporated into the design. Subsequent steps in the functional space aim to evaluate and improve this profile, but if necessary, modify the previous step to cycle through the resulting library at various steps in the design protocol. Concentration is performed.

In a preferred embodiment, the method comprises:
The method includes the following steps:
a) providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure);
b) identify the amino acid sequence in the CDR of the lead antibody;
c) select one CDR in the _VH or _VL region of the lead antibody;
d) providing an amino acid sequence comprising at least 3 consecutive amino acid residues in the selected CDR (the selected amino acid sequence is defined as the lead sequence);
e) compare the lead sequence to multiple tester protein sequences;
f) selecting at least two peptide segments having at least 10% sequence identity with the lead sequence from a plurality of tester protein sequences (the selected peptide segments form a hit library);
g) create an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
h) combining the amino acid variants in the hit library to generate a combination of hit variants that forms a hit variant library;
i) using a scoring function to determine whether members of the hit variant library are structurally compatible with the lead structure template;
j) Select a member of the hit variant library that has a score equivalent to or better than the lead sequence;
k) constructing a degenerate nucleic acid library comprising a DNA segment encoding the amino acid sequence of a selected member of the hit variant library;
l) determining the diversity of nucleic acid libraries, if diversity is greater than 1 × 10 ⁶ is the step j) to l) until the diversity of the nucleic acid library is less than or equal to 1 × 10 ⁶ repetition;
m) introducing the DNA segment in the degenerate nucleic acid library into the cells of the host organism;
n) expressing a DNA segment in the host cell so that a recombinant antibody containing the amino acid sequence of the hit library is produced in the host organism cell;
o) selecting a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ ; and
p) If no recombinant antibody is found that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ , repeat steps e) to o).

  As shown in FIG. 2B, the method starts with a target sequence or sequence profile based on structure-based multiple alignment, searches for a variant profile based on an evolutionary enriched sequence database, and then matches that with a structural template or assembly. Suitability is assessed and then a set of sequences that can be experimentally targeted is selected. This method is illustrated in our example. First, it makes use of evolutionary information (including expression, folding) encoded in sequences or combinations thereof that have not yet been captured by theoretical calculations. Second, after removing many irrelevant random sequences, structure-based screening for the resulting library is subjected to improved computer screening. Alternatively, improved computer scoring such as MM-PBSA can be applied to a portion of the aggregate structure. We believe that this method provides a highly improved sequence library for experimental screening, with significant time and cost savings.

FIG. 2C illustrates another embodiment of the method. The method includes the following steps:
a) providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure defined as a lead structure template);
b) identify the amino acid sequence in the CDR of the lead antibody;
c) select one CDR in the _VH or _VL region of the lead antibody;
d) providing an amino acid sequence comprising at least 3 consecutive amino acid residues in the selected CDR (the selected amino acid sequence is defined as the lead sequence);
e) mutating the lead sequence by substituting one or more amino acid residues of the lead sequence with one or more different amino acid residues to create a lead sequence variant library;
f) using a first scoring function to determine if the lead sequence variant library is structurally compatible with the lead structural template;
g) select a lead sequence variant with a score equal to or better than the lead sequence;
h) comparing the lead sequence to multiple tester protein sequences;
i) selecting at least two peptide segments having a sequence identity of at least 10% with the lead sequence from the plurality of tester protein sequences (the selected peptide segments form a hit library);
j) create an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
k) combining amino acid variants in a hit library to generate a combination of hit variants;
l) combining a selected sequence variant with a combination of hit variants to produce a hit variant library;
m) Use a second scoring function to determine whether members of the hit variant library are structurally compatible with the lead structural template;
n) Select a member of the hit variant library that has a score equivalent to or better than the lead sequence;
o) constructing a degenerate nucleic acid library comprising DNA segments encoding the amino acid sequences of selected members of the hit variant library;
p) determining the diversity of nucleic acid libraries, if diversity is greater than 1 × 10 ⁶ is the step n) ~p) to the diversity of the nucleic acid library is less than or equal to 1 × 10 ⁶ repetition;
q) introducing the DNA segment in the degenerate nucleic acid library into the cells of the host organism;
r) expressing a DNA segment in the host cell such that a recombinant antibody containing the amino acid sequence of the hit library is produced in the host organism cell;
s) selecting a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ ; and
t) If a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ is not found, repeat steps e) to s).

4) Functional space In functional space, the goal is to express and screen libraries derived from optimized variant profiles. There are two components that contain a functional cycle. An important engineering component for protein expression, which may not directly affect function, is oligonucleotide optimization. Determining realistic limits on the size of the oligonucleotide library is used as a guide to variant sequence resolution and reprofile.

  The other component is a functional screen that directly reflects the results of all previous steps and is the final evaluation part of the design strategy. The results of experimental functional screening determine whether library candidates can be transferred for further evaluation or can be used to enrich and refine libraries from previous steps. For example, a set of sequences exhibiting various levels of function can be used to narrow the variant profile or weight different residues at the described positions. Furthermore, sequence space jumps through the use of degenerate oligonucleotide designs lead to the identification of new functional variants, which can be used to further enrich the optimized variant profile. Alternatively, the frequency of a particular set of amino acids may reflect functional preference or expression preference. In the latter case, sequences that are low in expression but show good function may facilitate modification of codon usage to improve expression levels while maintaining function. Since selecting only the most frequent variants only approaches consensus and can lead to an “average” functional sequence, some second or third “stage” variants, ie less frequently It is important to select the variant that exists. Exceptional variants may come from combinations that are not observed in nature. We are using natural evolutionary patterns as a guide, but looking for combinations that are not observed in nature, which are useful on our immediate application, although these are not preferred on the evolution time scale Or perhaps nature hasn't tried these yet. In this sense, structure-based screening of random mutants or combinations thereof may give structurally favorable mutants that have not yet been observed in nature, but this Imposes strict requirements on accuracy and computer speed.

5) Repetition, refinement, and enrichment The design protocol is divided according to the different spaces to be evaluated, but all the operating cycles are interrelated and integrated, information can be exchanged and any space is freely Cycle to continuously improve and enrich the library based on the optimized variant profile. As a result, the path from the target sequence or structure to the candidate sequence is not a single path, but a series of oscillations between three cycles, each improving the selection of the optimized variant profile.

Furthermore, the functionality evaluation and repeatability of the design protocol not only improves variant selection, but also improves the accuracy of the scoring function for at least the range and sequence of the examined range. The failure of the prediction indicates that the template is not compatible. This also indicates that certain contributions (eg, skeletal entropy in glycine preference in functional screening) should be weighted more heavily. Certain charged residues (eg, Arg vs. Lys in V _H CDR3) may be preferred due to their role in orienting a particular conformation (see below).

6) Sequence re-profiling by scoring and ranking As noted above, sequences in the hit variant library should be evaluated based on structural compatibility with the lead antibody in the presence and absence of antigen. Can do. According to the scores and rankings obtained from the structure evaluation, the sequences in the hit variant library are reprofiled to optimize the sequence and structure space tests for functional sequences. This step involves selecting a subpopulation of hit variant libraries that score better than the lead sequence and re-profile them to create an optimized library. One option is to reprofile all sequences that score better than reads. However, this makes the library for experimental screening too large. A preferred method is to select a sequence subset within a low energy window or several such subsets (FIG. 7), which is experimental as outlined in the columns below and FIG. The final size of the nucleic acid library will be reduced. Combining rational selection and design, this process should enrich libraries with good-scoring sequences.

Profile modification and optimization must take into account the final size of the physical nucleic acid library (Figure 6). One strategy is to reprofile 10-20% of the maximum score of the hit variant library to limit the number of positional variants within limits that can be easily targeted in experiments (preferably degenerate nucleic acids). <10 ^{6 for} libraries. Similarly, we may select a set of low energy sequences containing the desired amino acid at a position.

7) Partitioning the sequence into fragments Another strategy to control the size is to partition the sequence based on fragments that do not correlate with structurally correlated fragments in the structure space. These split sequences with smaller variant profiles can be used to create several small libraries. This explanation suggests that, as a first approximation, dissociated segments are often irrelevant, so that large variations can be processed independently, but fragments that bind to each other in space are simultaneously analyzed by combinatorial nucleic acid libraries. It should be targeted. In the case of a loop, the sequence forming the base of the loop is generally associated with the closure of the loop, but the top is often uncorrelated with the base of the loop. In such a case, the amino acid subvariant profile is divided into three segments, the first and third segments (the base of the loop) are used for profile and library design, and the second segment (the top of the loop) is Used for second profile and library design. There should be 2 or 3 position overlaps between the fragments to maintain a small level of structural correlation between the resulting libraries. Similarly, longer profiles can be divided into overlapping segment chains in order to span the length of the library corresponding to the sequence. Simple criteria such as Cα or Cβ distance matrix can be examined to identify correlated segments (FIG. 28A). In some cases, a more detailed interaction matrix can be mapped to examine the number and type of interactions, but the basic principle is the same as identifying correlated segments.

  The resulting reprofile is further modified and enhanced based on observed experimental or structural criteria. These include various positions of known hydrogen bonds with additional polar amino acids, areas of high van der Waals contact with large aliphatic or aromatic groups, or areas that benefit from increased flexibility with glycine. . In experimental feedback, variants may be added based on assay results from early screening as a basis for subsequent design improvements. A more sophisticated analysis may take into account the bonding of amino acid groups within the sequence (eg, salt bonding or hydrogen bonding). An additional design constraint is the surface area of the nonpolar group of the protein that is accessible to the solvent.

  With a modified and optimized profile, we created a new amino acid sequence library, or library group (hit variant library IIA, IIB, IIC, etc.) called “Hit Variant Library II” Score using the energy function. Because variant recombination and profile modifications are thought to extend to the covered sequence and structural space, the energy distribution should extend outside the original energy window (FIGS. 7, 13A, 17A, and 18). .

  Various embodiments of the method of the present invention are described in detail below.

3. Construction of Hit Antibody Library in Silico As illustrated in FIG. 1A, a hit library can be constructed in silico based on the lead sequence from the region of the lead antibody. Sequences from protein sequence databases (eg, NIH GenBank, or antibody CDR Kabat database) are searched using various sequence alignment algorithms based on alignment with the lead sequence.

  FIG. 3 shows an example of a method for constructing a hit library, which begins with a search of a protein sequence database of different identity with the lead sequence or sequence profile. Lead sequence profiles are created by aligning sequences within structural motifs of the same family. This lead sequence profile can be used to construct an HMM to search a sequence database for hit libraries with low homology to the lead sequence. This approach is performed to find a rich pool of diverse hit sequences (ie, a hit library) to ensure that all available variants of the read sequence from the database are included.

  The database screened against the lead sequence preferably includes expressed protein sequences (including sequences from all organisms). More preferably, the protein sequence may be derived from mammals (including humans) or rodents that target the framework. Optionally, the protein sequence may be from a particular species or from a particular means of the same species. For example, protein sequences taken from a human immunoglobulin sequence database can be used to construct a library of polypeptide segments. Compared to the traditional method of constructing a library using fully random protein sequences, this approach of the present invention takes advantage of the sequence information obtained from the evolution of the protein, thus allowing antibody generation and affinity maturation. More closely mimics the natural process.

  Depending on the region / domain of the protein being designed, a database of proteins with different evolutionary origins is utilized. For example, to reduce the human immunogenicity of a designed antibody, sequences of human origin, more preferably germline sequences, are used for design purposes. On the other hand, extensive sequence search and selection from a wide range of databases and / or structure-based design methods is utilized to increase structural and / or functional diversity to increase diversity in CDRs. With such sequence and structure-based selection, rare combinations of sequences are found in the CDRs, but the sequences in the framework regions are kept as close as possible to the human sequence family.

  In addition, some combinations of amino acid residues from the sequences of various species (including but not limited to human or other non-human species, such as mouse, rabbit, etc.) can be combined into certain regions (eg, CDRs in antibodies and May be preferred at framework boundaries). This approach is done to maintain or optimize the relative orientation between the various motifs.

  A number of sequence alignment methods can be used to align sequences from a database with a lead sequence (or lead sequence profile) ranging from high to low sequence identity. Many sequence-based alignment programs have been developed and include, but are not limited to, Smith-Waterman algorithm, Needleman-Wunsch algorithm, Fasta, Blast, Psi-Blast, Clustalx, and profile shadow Markov models.

  In some cases, a simple sequence search method such as BLAST (Basic Local Alignment Search Tool) to search closely related sequences (eg> 50% sequence homology) Can be used. BLAST is a heuristic algorithm that uses position-independent scoring parameters (such as BLOSUM62) to detect similarity between two sequences and is widely used in routine sequence alignment (Altschul SF, Gish W, Miller W, Myers EW, Lipman Dj (1990) J Mol Biol 215, 403-410). However, BLAST analysis is too limited to detect distant homologues of the lead sequence. To search for distant homologues of the lead sequence, more advanced tools for sequence alignment can be used.

  Profile-based sequence alignment methods are used to search for lead sequence variants, such as PSI-BLAST (position specific repeat BLAST) and HMM. These profile-based sequence alignment methods can detect distant homologues of the lead sequence (Altshul, SF, Madden, TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ (1997) Nucleic Acids Res 25, 3389-3402; Krogh, A, Brown M, Mian SI, Sjolander Km Haussler D (1994) J. Mol. Biol 235, 1501-1531).

  PSI-BLAST is a new generation BLAST program belonging to the profile-based sequence search method (Altshul, SF, Madden, TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ (1997) Nucleic Acids Res 25, 3389-3402). PSI-BLAST automatically combines the statistically significant alignments created by BLAST into a position-specific matrix to score sequence alignments in the database. The newly searched sequence is incorporated into a position-specific scoring matrix to initiate another round of sequence search in the database. This method is repeated until there is no hit or until a preset criterion is met. PSI-BLAST is not as sensitive as a profile-shaded Markov model (HMM), but can be used in the present invention due to its speed and ease of operation in the absence of a pre-made motif profile.

  Profile-shaded Markov model or HMM is a statistical model of the primary sequence consensus of a sequence or sequence alignment family. A sequence family is defined as a multiple sequence alignment resulting from a corresponding multiple sequence and / or structural alignment. The formal probabilistic base that is the basis of the HMM allows the use of Bayesian probability theory to guide the setting of scoring parameters based on the aligned sequence profile. This same feature also allows HMM to use a unified approach using position-dependent scores, scoring both amino acid and gap alignments. These features of HMM make it a powerful way to search distant homologues compared to classical heuristic methods (Eddy S.R (1996) Curr Opin Struct. Biol 6, 361-365). Patterns in the primary sequence can be detected by a pattern recognition algorithm, and thus more related to the target sequence (when only one sequence is used) or sequence profile (when multiple sequence alignments are used) Can be used to retrieve many members. In order to capture higher order correlations in sequences, or amino acid interactions in three dimensional space, multiple sequence alignments resulting from multiple structural alignments are preferably used in the present invention to create hit libraries. It is a simple method.

  In some cases, structure-based sequence alignment may be used to search a highly diverse hit library. This is because it is a standard method that can be used to compare various multiple sequence alignments in the absence of detectable sequence homology (Sauder JM, Arthur JW, Dunbrack RL Jr (2000) Proteins 40, 6 -twenty two). Multiple structural alignments directly provide corresponding multiple sequence alignments. Alternatively, these closely related structures can be used as structural templates for sequence threading to create multiple sequence alignment profiles (Jones DT (1999) J Mol Biol 1999, 797-815). It has been reported that a combination of multiple sequences and structural alignments explains the structural and functional properties of known protein sequences (Al-Lazikani B, Sheinerman FB, Honig B (2001) PNAS 98, 14796-14801). ).

  Also, in some cases, a highly diverse hit library may be searched using a reverse threading process. The reverse threading process is the opposite of the threading process. Threading uses a scoring function that captures sequence side chain interactions and local parameters (eg secondary structure and solvent exposure) to thread that sequence (ie, the sequence in question) into a library of structural template candidates. Process. The threading process begins with the prediction of the secondary structure of the amino acid sequence of the extracted residue of the sequence in question and the accessibility of the solvent. The resulting one-dimensional (1D) profile of the predicted structure is threaded to each member of a library of known 3D structures. The optimal threading of each sequence-structure pair constitutes the predicted 3D structure in question.

  In contrast, reverse threading is the process of threading a sequence into a target structure or a set of structural clusters of target structures to retrieve the optimal sequence from a sequence database. Various scoring functions are used to select optimal sequences from libraries containing protein sequences of various lengths.

For example, an amino acid sequence from a human germline immunoglobulin database can be threaded onto the 3D structure of a lead antibody to search for sequences with acceptable scores. The selected sequences constitute a hit library. The reverse threading process is the opposite of the threading process, where the former seeks to find the optimal sequence that matches the target structure template, while the latter finds the optimal 3D structure that matches the target structure profile.
In addition, above the sequence hits found for the lead antibody, reverse threading multiple amino acids at each position with a combinatorial approach to select the optimal “consensus” combinatorial sequence compatible with the 3D structure of the lead antibody. To be profiled. This process of searching for consensus sequences differs from the method using simple sequence averaging at each position described by Knappik et al. (2000). The consensus sequence of the present invention is created based on the searched sequence using a structure-based reverse manipulation approach using all possible combinations of amino acids possible at each position and its fit with the structural template. Optimized by scoring gender.

  In addition to the method used by sequence alignment, the database corresponding to the sequence motif used for sequence alignment is also critical in the method of the present invention. A sequence or sequence profile as used herein is defined based on structural analysis of protein function for antibody regions (eg, CDR motifs for antigen binding (CDR1, CDR2, and CDR3), to support antibody scaffolds Framework regions (FR1, FR2, FR3, and FR4), for example, using GeneBank and Kabat database to search for sequence hits from various species and hit libraries that match antibody CDRs While the human or human germline sequence database is preferably designed, the immunogenicity of the framework's non-human origin Used to search for sequence hits in the framework design to reduce the chance of creating epitopes. , Especially considering the final therapeutic applications of the antibody designed to permit control and maximum flexibility of the array origin of design.

The hit library can be further refined by eliminating duplicate sequences and re-profiling to obtain a more accurate HMM or PSI-BLAST profile. Read V _H CDR3 according to the Kabat classification (and structural motif) of human anti-VEGF antibodies with or without some residues flanking at the N-terminus or C-terminus, as described in detail in the Examples section Used as an array. HMM model using HMMER 2.1.1 software package utilities (hmmbuild, hmmcalibrate, hmmsearch, hmmalign) with default settings (Eddy S in http: //hmmer.wust1.eduhttp: //hmmer.wust1.edu ) Was constructed, the HMM model was calibrated against the synthesized random sequences, the database was searched for hit sequences, and they were aligned. One hit sequence of the same length as the lead sequence is used for alignment and variant profiles. Insertions or deletions in the aligned sequence should also be used to profile variants at aligned positions. Can do.

As shown in FIG. 3, searching the Kabat database using one lead sequence of the V _H CDR3 sequence of an anti-VEGF antibody as an HMM has a sequence identity in the range of 40-100% to the lead sequence. Unique sequences were found (Figures 10A and 19C). Searching the same Kabat database using the multiple aligned sequence profile of this lead sequence as an HMM found 251 unique sequences with sequence identity in the range of 15-100% to the lead sequence (FIG. 19C). ). These results indicate that the profile HMM can find sequences with homology far from the lead sequence. That is, the sequence profile obtained from the multiple structure alignment will extend the diversity of the hit library.

  The sequence of the hit library also depends on the database used. For example, using Genpept instead of the Kabat database above, hits different from those in the Kabat database were found when using a single read sequence as the HMM or when using a structure-based sequence profile as the HMM.

  Sequences in hit libraries constructed by searching the database are analyzed (eg, by profiling based on the frequency of each amino acid residue) and screened for the desired function in vitro or in vivo Can be used directly. See path I in FIG. 1A and FIG.

  Optionally, the sequences in the hit library are profiled and used to construct hit variant library I, which is then screened in vitro or in vivo for the desired function. See Path II in FIG. 1B and FIG.

  Also, optionally, the hit library is filtered based on scoring compatibility with the lead structure template using methods such as reverse threading or force field-based full atom presentation. Based on the resulting score ranking, hit variant library II is selected for screening in vitro or in vivo for the desired function. See path III in FIG. 1C and FIG.

  Also optionally, the hit library I is filtered based on scoring compatibility with lead structure templates using methods such as threading or force field based full atom presentation. Based on the relative ranking of hits, hit variant library II is selected for screening in vitro or in vivo for the desired function. See Path IV in FIG. 1D and FIG.

4. Construction of a hit variant library To further investigate the abundance of diversity encoded in the protein structure and sequence space, hits selected based on sequence alignment are used to create a variant profile. Profiled at amino acid positions. Hit variant libraries are calculated combinatorially using this variant profile. FIG. 4 is an example of a method for constructing a hit variant library. Variant profiles generated from hit libraries (ie, sequence hits or filtered sequence hits) are described based on the frequency of amino acids appearing at each position in the hit sequence (FIGS. 11 and 19B). Profiled variants provide an excellent starting point for building combinatorial libraries.

  To reduce the size of this hit variant library, apply several cutoff values and / or computational results based on the frequency of amino acids at each position (eg 5% or more) or the preferred variant (See the lower part of FIG. 11 for a cut-off of 10% of the total number of hits; FIG. 19B uses 5%). Variants based on these highly suitable amino acid residues at each position should provide an excellent pool of recombinant sequences to lift sequences with high affinity or other desired functions.

  Information sequence entropy calculated based on the frequency of variants at each position provides a quantitative means to measure how much the identity of the residues in the aligned sequences deviates from the random distribution of amino acid residues. provide. To account for the highly variable mutagenesis probability of sequences containing protein variants, the present invention can use relative entropy (Plaxo KW, Larson S, Ruczinski, Riddle DS, Thayer EC, Buchwitz B, Davidson AR, Baker D (2000) J Mol Biol 298, 303-312). Since the relative site entropy is based on true evolutionary data from the database of expressed proteins, we have a good guide to the locations and variants to be targeted for computational and experimental screening. I think it will be.

  Relative site entropy is a measure of the diversity of each position of amino acid residues accumulated during evolution while maintaining the structure and function of the hit sequence. These sites are selected to recombine computer screening and experimental screening. The resulting combinatorial hit variant library size is much smaller than that generated by a random combination of all 20 amino acids at each position, so more accurate and detailed computer screening or direct experimental screening can do.

  The sequence entropy generated by the hit library of the present invention is the site entropy used by other researchers in the field to measure structure tolerance for amino acid substitutions using force field-based computational methods. Irrelevant (Voigt CA, Mayo SL, Arnold FH, Wang ZG (2001) PNAS 98, 3778-3783). Force field-based methods will provide some new variants that have not been tested by evolution, but site entropy derived from evolutionary sequences (ie, sequence entropy) depends on incorporated structural, dynamic All information, including biological, expression and biological activity should be used to provide more significant statistics for changes at each position and suitable variants. While this may be important for targeting difficult structures such as antibody loop regions that are not well understood or predicted by force field-based methods, they are Can be used and modeled with some confidence. Homology-based methods that rely on evolutionary information are still one of the most reliable methods for modeling loop structures that can be enhanced using force field-based simulation.

As described in detail in the Examples section, the variant profile of anti-VEGF antibody (lead antibody) was searched using several different approaches. Based on the V _H CDR3 sequence of this lead antibody, variant profiles of the hit list from Kabat, genpept, and non-redundant databases combining Kabat, genpept, imgt, etc. are described. In affinity matured sequences from this antibody, important variants observed by other investigators also appear frequently in variant profiles searched using the methods of the present invention. For example, the single most important variant is H97 in the lead sequence replaced by Y97 in the mature sequence (Figure 9B), which is almost 50% amino acid variant at this position (Figure 11). The above method of the present invention has several advantages in protein design and manipulation. In any recombinant library, diversity is not necessarily limited by the ability to be screened, but this is an important factor in creating a library in which diversity is assigned and thus the design is functionally related Means that. This method is an in silico rational design of proteins, especially antibodies. This begins with the selection of functionally similar “natural” polypeptide fragments from a database of expressed proteins to form a hit library. Analysis of specific position changes in “naturally occurring” peptide fragments gives evolutionary data (variant profiles) for preferred residues and positions. A critical analysis of the variants can identify key residues and combinations. Combinatorial calculation of a reduced set of selected variants creates a hit variant library that focuses on functionally related sequences.

  Starting with a variant profile, the in silico rational library design of the present invention generates a focused library of protein fragments based on functional and structural data. To some extent, in silico recombination is similar in principle to DNA shuffling of a family of homologous sequences. However, the approach of the present invention is a very efficient sequence recombination method for a family of protein sequences with widely distributed sequence homology. Furthermore, in the present invention, recombination can be placed at the amino acid level and localized to specific functional regions to create a library whose members are recombined randomly. This is not constrained by homology requirements and can be selectively modified according to structural or experimental data. For example, the sequences in the hit library have 100-20 or less sequence identity to the lead sequence, depending on the search method and the database used. In comparison, DNA shuffling is a process of DNA recombination between closely related sequence homologues and there are stringent requirements for sequence homology between recombined nucleic acid sequences; DNA shuffling is an effective mutation Creating somatic recombination is inefficient and subject to random mutations during experimental recombination.

5. Structure-based evaluation of antibody variant libraries As described above, hit variant libraries obtained from recombination of hit libraries or variant profiles from hit libraries are evaluated based on structural compatibility with the lead protein. May be. For structure-based evaluation of antibody variant libraries, the present invention addresses the following issues: i) How to model the conformation of a non-standard loop in the presence of an antigen that forms a protein complex with an antibody. (Ii) how to place side chains in the CDR loop backbone to optimally match the antibody and / or antigen structure; (iii) to form a stable antibody-antigen complex with high affinity. And how to combine the CDR loop with the optimal framework model. The method of implementation is detailed below.

1) Antibody structure and structural model The structural template of the lead antibody is taken directly from the X-ray or NMR structure or modeled using the structural computer engine described below. As shown in the Examples section, the structural template for the anti-VEGF antibody is taken from the PDB database (1BJ1 for the parent antibody and 1CZ8 for the mature antibody). Both templates were used in the presence and absence of antigen VEGF. The scoring described in the examples is from 1CZ8 in the presence of the antigen VEGF.

2) Evaluation based on lead antibody structural template As an example, an antibody having a known 3D structure functions as a lead protein. Alternative methods (eg, homology-based modeling) can be applied to create a well-defined template structure for the target protein being manipulated, so that well-defined structures (eg those obtained by X-ray crystallography) ) This requirement is not absolute. Creation of a hit variant library requires the determination, modification, and optimization of amino acid position variant profiles. The lead sequences and the sequences in the hit and hit variant libraries are scored in terms of the 3D structure of the lead antibody and scored to obtain a ranking distribution for these sequences. Note that the scoring in the Examples column is based on an empirical total atomic energy function, but any sequence that can be handled by any computer can be applied to structurally evaluate these sequences.

  FIG. 5 shows an example of a method for structural evaluation of sequences from reads, hit libraries, and hit variant libraries. For scoring and ranking, these sequences are built into the lead structure template by substituting side chains from the backbone-dependent / independent rotamer library (Dunbrack RL Jr, Karplus M (1993) J Mol Biol 230: 543-574). The side chains and backbone of the substituted segment are then locally minimized in energy to mitigate local strains. Each structure is scored using a custom energy function that measures the relative stability of the sequences in the lead structure template.

  Comparison of the energy of the sequences from the lead, hit library and hit variant library shows the degree of structural compatibility of the various sequences with the lead structural template. It is not unreasonable to obtain a very broad distribution with many sequences having better or worse scores than the lead sequence. The focus is not to identify a specific sequence (although it is acceptable), but to identify a population of sequences or a set of sequences that have a score equal to or better than the lead sequence and use a degenerate nucleic acid library Sharing the nature of aggregates in sequences that can be targeted simultaneously. Amino acid sequence aggregates are sequence spaces that may exhibit better structural compatibility with a binding site and orientation for better epitope recognition than a single specific sequence. A combinatorial library of sequence ensembles distributed around the statistical ensemble average should be targeted experimentally to increase the chances of finding good candidates with improved affinity.

3) Evaluation based on lead structure template in the presence of its ligand. In the presence, it can be evaluated based on the lead structure template. This approach is useful when the structure of the complex formed by the lead protein and its ligand is known or can be easily identified.

  In the presence of antigen, the complete thermodynamic cycle of antibody-antigen complex formation is included in the calculation. In particular, antibody conformation at the binding site is modeled on the basis of individual CDR loop conformations from that standard family with suitable side chain rotamers as well as interactions between CDR loops. A wide range of conformations (including those in the side chains of amino acid residues, and those in the CDR loop of the antigen binding site) can be tested and incorporated into the main framework (or scaffold) of the antibody. In the presence of antigens, such conformational modeling uses physico-chemical force fields and semi-empirical and knowledge-based parameters to achieve higher physical relevance, a natural process of antibody production than scoring. Ensuring better presentation, and maturation in the body.

4) Correlation of antibody sequence scores in the presence and absence of antigen. Focus on sequences that have a complex structure between the antigen and its antibody and have a good probability of binding the antibody library to the antigen. Is preferred. For the remaining antibodies of biomedical interest, the antibody-antigen complex structure is not yet available.

We have identified that many sequences suitable for stabilizing the target antibody scaffold also stabilize specific antibody-antigen complexes, even for V _H CDR3 that are directly involved in binding to the antigen. Has been found to be one of the selected candidates. Correlation analysis shows that there is a general correlation trend in antibody sequence scores in the presence and absence of antigen (FIG. 12C). Furthermore, a large population of sequences selected with a good score is preferred to stabilize a binding motif scaffold such as the anti-VEGF V _H CDR3 used herein.

  Note that even without complex structures, antibody structures alone can provide a population of sequences that stabilize the target scaffold while having the correct binding site for the antigen. Although conformational changes due to antigen binding have been observed, it is unclear whether conformational changes are the only one of many possible solutions or the absolute need for antigen-antibody interactions. The goal is to identify a set of sequences that can form a functional protein so that the binding structure is not a requirement unless it undergoes this conformational shift. This is a preferred assumption based on the available structure of the antibody in bound and unbound states. As long as they belong to the same family assembly structure, the approach taken here (see 19A) allows at least some structural variations.

  Alternatively, if the lead antibody structure is not available, the template may be created by modeling. Antibody structures or structural motifs are some of the best known examples of proteins for which structural models are relatively confidently created using homology modeling. That is, it is possible to target a sequence library of lead sequences without using a lead structure template. As shown in the Examples section, a stretch of sequence library covering the target motif can be synthesized and used to screen for antibodies with high affinity without depending on the structure of the lead antibody.

5) Structural computer engine Many programs are available to model and evaluate libraries against lead structural templates. For example, molecular mechanics software is used for these purposes, and further examples include, but are not limited to, CONGEN, SCWRL, UHBD, GENPOL, and AMBER.

  CONGEN (CONformation GENerator) is a program for performing conformational search on protein segments (RE Bruccoleri (1993) Molecular Simulations 10, 151-174 (1993); RE Bruccoleri, E. Haber, J. Novotny, ( 1998) Nature 335, 564-568 (1988); R. Bruccoleri, M. Karplus. (1987) Biopolymers 26, 137-168). This is best suited to the problem when it is necessary to build loops or segments that are not determined in a known structure (ie, homology modeling). This program is a modified version of CHARMM version 16 and has most of the functions of the CHARMM version (Brooks BR, Bruccoleri BE, Olafson BD, States DJ, Swaminathan S, Karplus M. (1983) J. Comput. Chem. 4 , 187-217).

  The fundamental energy function used includes terms for electrostatic interactions with binding, angle, torsional angle, singular angle, van der Waals and distance-dependent dielectric constant, and an Amber94 force field that can be measured using CONGEN. use. (See Example section).

  The CONGEN program is used to search for low energy conformers that are close to or correspond to the naturally occurring structure with the lowest energy (Bruccoleri and Karplus (1987) Biopolymers 26: 137-168; and Bruccoleri And Novotny (1992) Immunomethods 96-106). With an accurate Gibbs function and a short loop sequence, all the stereochemically acceptable structures of the loop can be created and its energy calculated. Those with low energy are selected.

  This program can be used to perform both conformational searching and structural evaluation using basic or improved scoring functions. The program can calculate other properties of the molecule, such as solvent accessible surface and conformational entropy, certain steric constraints, and the like. Each of these properties can be used in combination with other properties described below to score a digital library.

In the present invention, a defined standard structure of five CDRs (V _L CDR1, 2, and 3, and V _H CDR1, and 2) other than V _H CDR3. V _H CDR3 is known to vary greatly in length and conformation, but because more antibody structures are available in the PDB (Protein Data Bank) database, The modeling of the conformation is progressing. CONGEN is used to create a conformation of the loop region (eg, V _H CDR3) and replace the side chain of the template sequence with the corresponding side chain rotamer of the target amino acid when the standard structure is not available . Third, the model can be further optimized by energy minimization or molecular dynamics simulation or other protocols to mitigate steric mismatch and strain of the structural model.

  SCWRL is a side chain configuration program that can create side chain rotamers and combinations of rotamers using a backbone-dependent rotamer library (Dunbrack RL Jr, Karplus M (1993) J Mol Biol 230: 543- 574; Bower, MJ, Cohen FE, Dunbrack RL (1997) J Mol Biol 267, 1268-1282). This library provides a list of Cai 1 -Cai 2 -Cai 3 -Cai 4 values and the relative probabilities of residues at a certain Phai sai value. The program can also examine these conformations to reduce side chain-backbone mismatches or side chain-side chain mismatches. Once the steric mismatch is minimized, the side chains and backbone of the substituted segment can minimize energy and use CONGEN to mitigate local strain (Bruccoleri and Karplus (1987) Biopolymers 26 : 137-168).

  Several automated programs specifically developed to construct antibody structures are used in the structural modeling of the antibodies of the present invention. The ABGEN program is an automatic antibody structure creation algorithm for obtaining a structure model of an antibody fragment. Mandal et al. (1996) Nature Biotech. 14: 332-328. ABGEN utilizes homology-based scaffolding techniques, invariant strictly conserved residues, known Fab structural motifs, standard features of hypervariable loops, twist distortion of residue substitutions, and key residues Including the use of interactions. Specifically, the ABGEN algorithm consists of two main modules (ABalign and ABbuild). ABalign provides an alignment of antibody sequences with all V region sequences of antibodies whose structure is known and calculates the alignment score. The library sequence with the highest score is considered the best fit for the test sequence. ABbuild then uses this optimal fit model output by ABalign to create a three-dimensional structure and provide Cartesian coordinates for the desired antibody sequence.

  WAM (Whitelegg NRJ and Rees, AR (2000) Protein Engineering 13, 819-824) is an improved version of ABM, combining algorithms (Martin, ACR, Cheetham, JC, and Rees AR (1989) PNAS 86, 9268- 9272) to model a CDR conformation using a standard conformation of CDR loops from an X-ray PDB database and a loop conformation created using CONGEN. Briefly, the modularity of antibody structure allows the structure to be modeled using a combination of protein homology modeling and structure prediction.

  In a preferred embodiment, the following method is used to model antibody structure. Since antibodies are one of the most conserved proteins, both in sequence and structure, antibody homology models are relatively simple, but are still within existing standard structures or structures with insertions or deletions. Exclude some CDR loops that have not been determined. However, these loops can be modeled using algorithms that combine homology modeling and conformational search (eg, CONGEN can be used for such purposes).

A standard structure defined for five of the CDRs (L1, 2, 3 and H1, 2) is used. H3 in the variable heavy chain (ie, V _H CDR3) is known to vary greatly in length and conformation, but modeling its conformation as more antibody structures become available Is progressing. Modeling methods include protein structure prediction methods such as threading and comparative modeling, which aligns an unknown structure sequence with at least one known structure based on a similarity modeling sequence. This de novo or ab initio method also makes it promising to predict structure from sequences alone. Unknown loop conformations can be tested using CONGEN when standard structures are not available (Bruccoleri, E. Haber, J. Novotny, (1998) Nature 355, 564-568). Or use the ab initio method (including but not limited to Rosetta ab initio method) without depending on the folding level similarity between the modeled sequence and any known structure The antibody CDR structure can be predicted (Bonneau R, Tsai J, Ruczinski I, Chivian D, Rohl C, Strauss CE, Baker D (2001) proteins Suppl 5, 119-126). Improve and select native-like structures from models created from CONGEN or Rosseta ab initio methods, using more accurate methods using the latest explicit solvent molecular dynamics and indeterminate solvent-free energy calculations (Lee MR, Tsai J, Baker D, Kollman PA (2001) J Mol Biol 313, 417-430).

  The X-ray structure used here (1BJ1 and / or 1CZ8) or the modeled structure described above can be used as a structural template for designing antibody libraries for screening by experiments described below.

6) Scoring function for structural evaluation In one embodiment of the present invention, computational analysis is used for structural evaluation of selected sequences from the sequence evaluation process described in Sections 3 and 4 above. This assessment is based on an empirical and parameterized scoring function and aims to reduce the number of subsequent in vitro screens required.

  This approach uses an existing structural template to score all created amino acid libraries. Using a known structure as a template for assessing antibody-antigen interaction assumes the following: (i) The structure of the antibody and antigen molecule does not change significantly in the bound and free state, (ii ) Mutations in CDRs do not significantly alter overall and local structure, and (iii) The energy effects of mutations in CDRs are localized and can be scored to evaluate functions directly associated with the mutation it can. The advantage of having a known structure as a template is that it can be a good starting point for design improvement compared to the more difficult approach using modeled structures. The energy distribution of these sequence hits should reveal how well they cover the matching function of the target scaffold in terms of structural compatibility with its target.

  The above assumptions always introduce errors due to the uncertainty in the structure of the mutant, so if the mutant changes its structure, a sophisticated scoring function may not be able to make a meaningful prediction. As shown in the Examples section, a general but well-tested force field (described below) was used in the first calculations in the model system for anti-VEGF antibodies. This can generally avoid bias introduced into a specific system if suitable regions of matching background can be examined by examining experimentally implemented collection sequences. However, the present invention does not exclude the use of sophisticated scoring functions for structure evaluation.

  A number of energy functions can be used to score the fitness between the sequence and the structure. Typically four types of energy functions can be used: (1) Empirical physical chemical force fields, eg standard molecular mechanical force fields described below from simple model compounds; (2) Protein structure Knowledge-based statistical force field extracted from the so-called structure-based sequence profiling potential or threading score of average force; (3) parameters by fitting force field parameters using an experimental model system (4) A combination of one or several terms from (1) to (3) with various weighting factors in each term.

  The following are some well-tested physical-chemical force fields that can be used or incorporated into the scoring function. For example, amber94 was used in CONGEN to score sequence-structure compatibility in the examples below. Force fields include, but are not limited to, the following force fields, which are widely used by those skilled in the art: Amber 94 (Cornell, WD, Cieplak P, Bayly CI, Gould IR, Merz KM Jr, Ferguson DM, Spellmeyer DC. Fox T, Caldwell JW and Kollman PA. JACS (1995) 117, 5179-5197 (1995); CHARMM (Brooks, BR, Bruccoleri, RE, Olafson, BD, States, DJ. Swaminathan, S., Karplus, M. (1983) J. Comp. Chem. 4, 187-217 .; MacKerell, AD; Bashford, D; Bellott, M; Dunbrack, RL; Eva seck, JD; Field, MJ; Fischer, S; Gao, J Guo, H; Ha, S; JosephMcCarthy, D; Kuc nir. L; Kuczera, K; Lau, FTK; Mattos, C; Michnick, S; Ngo, T; Nguyen, DT; Pro hom, B; Reiher, WE Roux, B; Schlenkrich, M; Smith, JC; Stote, R; Straub. J; Watanabe, M; WiorkiewiczKuczera, J; Yin, D; Karplus, M (1998) J. Phys. Chem., B 102, 3586 -3617); Discover CVFF (Dauber-Osguthorpe, P .; Roberts.VA; Osguthorpe, DJ; Wolff, J .; Genest, M .; Hagler, AT (1988) Proteins: Structure, Function and Genetics, 4, 31-47.); ECEPP (Momany, FA, McGuire, RF. Burgess, AW, & Scheraga, HA, (1975) J. Phys. Chem. 79, 2361-2381 .; Nemethy G., Pottle, MS, & Scheraga, HA, (1983) J. Phys. Chem. 87, 1883-1887); GROMOS (Hermans. J., Berendsen, HJC, van Gunsteren, WF, & Postma, JPM, (1984) Biopolymers 23, 1); MMFF94 (Halgren, TA (1992) J. Am. Chem. Soc. 114, 7827-7843 .; Halgren, TA (1996) J. Comp. Chem 17, 490-519 .; Halgren, TA (1996) J. Comp. Chem. 17, 520-552 .; Halgren, TA (1996) J. Comp. Chem. 17, 553-586 .; Halgren, TA, and Nachbar, RB (1996) J Comp. Chem. 17, 587-615 .; Halgren, TA (1996) J. Comp. Chem. 17, 616-641.); OPLS (Jorgensen. WL, & Tirado-Rives, J., (1988) J Am. Chem. Soc. 1lO, 1657-1666 .; Damm, W., A. Frontera, J. Tirado-Rives and W. L Jorgensen (1997) J. Comp. Chem. 18, 1955-1970.); Tripos, (Clark. M., Cramer 111, RD, van Opdenhosch, N., (1989) Validation of the General Purpose Tripose 5.2 F orce Field, J. Comp. Chem. 10, 982-1012.); MM3 (Lii, JH., & Allinger, NL (1991) J. Comp. Chem. 12, 186-199). For example, Dreiding (Mayo SL, Olafson BD, Goddard (1990) J Phy Chem 94, 8897-8909), or protein folding or UNRES (United Residue Forcefield; Liwo et al., (1993) Protein Science 2, 1697-1714; Liwo et al., (1993) Protein Science 2 , 1715-1731; Liwo et al., (1997) J. Comp. Chem. 18, 849-873; Liwo et al., (1997) J. Comp. Chem. 18: 874-884; Liwo et al., (1998) J. Comp Other general force fields used in simulations such as Chem. 19: 259-276.) Are also used.

  Statistical potentials obtained from protein structures can also be used to assess the compatibility between sequences and protein structures. These potentials include, but are not limited to, residue pair potentials (Miyazawa S, Jernigan R (1985) Macromolecules 18, 534-552; Jernrgan RL, Bahar, I (1996) Curr. Opin. Struc. Biol. 6, 195-209). The mean force potential (Hendlich et al. (1990) J. Mol. Biol. 216, 167-180) has been used to calculate the conformational set of proteins (Sippl M (1990) J Mol Biol. 213). , 859-883). However, some limitations of these force fields are also considered (Thomas PD, Dill KA (1996) J Mol Biol 257, 457-469; Ben-Naim A (1997) J Chem Phys 107, 3698-3706).

  Other methods for scoring fitness between sequences and structures are sequence profiling (Bowie JU, Luthy R, Eisenberg DA (1991) Science 253, 164-170), or threading scores (Jones DT, Taylor WR, Thornton JM (1992) Nature 358, 86-89; Bryant, SH, Lawrence, CE (1993) Proteins 16, 92-112; Rost B, Schneider R, Sander C (1997) J Mol Biol 270, 471-480; Xu Y, Xu D (2000) Proteins 40, 343-354). These statistical force fields based on quasichemical approximations or Boltzmann statistics or Bayes' theorem (Simons KT, Kooperberg C. Huang E, Baker D (1997) J Mol Biol 268, 209-225) It is used to evaluate the sequence advantages of protein design (Dima RI. Banavar JR, Maritan A (2000) Protein Science 9, 812-819).

  In addition, structure-based thermodynamic parameters related to the thermodynamic stability of protein structures are also used to assess sequence and structure agreement. In structure-based thermodynamic methods, thermodynamic quantities (eg, heat capacity, enthalpy, entropy) can be calculated based on the structure of the protein, and thermodynamic data from model compounds or protein calorimetric studies. Can be used to explain the temperature dependence of heat denaturation (Spolar RS, Livingstone JR, Record MT (1992) Biochemistry 31, 3947-3955; Spolar RS, Record MT (1994) Science 263, 777-784; Murphy KP, Freire E (1992) Adv Protein Chem 43, 313-361; Privalov PL, Makhatadze GI (1993) J Mol Biol 232, 660-679; Makhatadze GI, Privalov PL (1993) J Mol Biol 232, 639-659 ). Using structure-based thermodynamic parameters, the structural stability of mutant sequences and hydrogen exchange protection factors can be calculated using a set-based statistical thermodynamic approach (Hilser VJ, Dowdy D, Oas TG. Freire E (1998) PNAS 95, 9903-9908). Thermodynamic parameters for statistical thermodynamic models of protein secondary structure formation have also been measured using experimental model systems, with excellent agreement between predictions and experimental data (Rohl CA, Baldwin RL (1998) Methods Enzymol 295, 1-26; Serrano L (2000) Adv Protein Chem 53, 49-85).

  The combination of various terms from the molecular dynamic force field and some specific components are mostly used in protein design programs. In preferred embodiments, the force field is one or several terms (eg, vdw, hydrogen bonds, and electrostatic interactions) from standard molecular mechanics force fields (eg, Amber, Charmm, OPLS, cvff, ECEPP). ) And one or several terms that are thought to control the stability of the protein.

  In order to improve the scoring function, an additional energy term was introduced in a later step, which is a scoring function to better resolve deviations from experimental results and the effects of the specific antibody-antigen interaction of interest. Makes it possible to tweak. For example, one energy term makes arginine mutations disadvantageous, reduces the contribution to the overall score due to the uncertainty in predicting side chain conformation, and compensates for this scoring function bias that favors arginine. Other energy terms can be scored for charged and polar groups based on surface area calculations, so that the exposed surface defeats mutations that lead to charge burial.

  In fact, there are many scoring functions that can be used to score sequence suitability with a template structure or structure aggregate. The improved scoring function has several terms (contributed from electrostatic and van der Waals interactions, calculated using molecular mechanics force fields, ΔGMM, electrostatic solvation and solvent accessible (Including Sharp KA. (1998) Proteins 33, 39-48; Novotny J. Bruccoleri RE, Davis M, Sharp KA (1997) J Mol Biol 268, 401-411).

A convenient and rapid method of computational screening uses a sum or combination of energy terms, using a basic scoring function that includes terms from a molecular mechanical force field such as Amber94 as performed at CONGEN. Is to calculate.
ΔE _total = E _bond + E _angel + E _dihed + E _impr + E _vdw + E _elec + E _solvation + E _other or bond free energy is an improved scoring function ΔG _b = ΔG _MM + ΔG _sol -TΔS _SS where ΔG _MM = ΔG _ele + ΔG _vdw (1)
ΔG _sol = ΔG _ele-sol + ΔG _ASA (2)
Is used as the difference between the bound and unbound states.

ΔG _ele and ΔG _vdw (electrostatic and van der Waals interaction energies) are calculated using the amber94 parameter made in CONGEN for ΔG _MM , where ΔG _ele-sol is the non-insulating protein The city solvation energy required to transfer a uniformly distributed charge into an aqueous phase with an insulating boundary defined by the protein shape. This is calculated by solving the Poisson-Boltzmann equation for the electrostatic potential of the reference and mutant structures. ΔG _ASA (nonpolar energy) is the energy cost for transferring nonpolar solute groups into an aqueous solvent, where solvent molecules are reconstructed. This has been shown to be first-order correlated with the accessible surface of the molecular solvent (Sitkoff D, Sharp, KA, Honig B (1994) J Phys Chem 98, 1978-1988; Pascual-Ahir & Silla (1990 ) J Comp Chem 11, 1047-1060).

  The change in side chain entropy (ΔSSS) is a measure of the effect on the local side chain conformation space, especially at the binding interface. This is calculated from the ratio of the number of allowed side chain conformations in the bound and unbound states. For general scoring purposes, to test the specificity of multiple side-chain conformation species in various backbone conformations to avoid the burden of large computational burdens, separate independent side-chain approximations to mutated side chains Apply.

  The sequences in the hit library or hit variant library are evaluated for structural compatibility with the target structure and mapped onto the target folding energy background. For many anti-VEGF antibodies, the score of the antibody sequence in the presence and absence of antigen is generally relevant since many variants can stabilize the antibody scaffold (see FIG. 12C). In particular, there is a significant percentage of sequences that can bind to the target epitope. As shown in the Examples section, the CDR library sequences are ranked based on the match score, based on the relative stability of the template antibody-antigen complex (1CZ8), and experimentally selected sequences are identified. (FIG. 13A).

  It is useful and possible to measure scores in both bound and unbound antigens to eliminate highly undesired sequences in either state. By doing so, we can avoid the need to accurately score the difference between bound and unbound states while effectively reducing the search space.

  The scoring function is used to score sequences in the hit library, hit variant library I or hit variant library II, and optionally the difference between the lead sequence or the lead structure template sequence and the library sequence. Calculated to complete the thermodynamic cycle. Thus, a sequence is selected for further experimental screening based on any of the following criteria: 1) A sequence with a better score than the lead sequence is selected in stabilizing the antibody structure; 2) Antibody-antigen A sequence with a score better than the lead sequence is selected for stabilization of the complex; 3) If the scoring function is sensitive enough to distinguish small differences between large numbers, the score of the bound and unbound state will be The difference is superior to the lead arrangement. The last criterion is used only when highly improved scoring functions or high quality set-based scoring functions are available, and preferably in systems where high quality variant data is available for scoring function calibration Should.

  Sequences with better scores than the lead sequence are analyzed and classified into different clusters. The combination of clusters covers sufficient sequence and structural space, which covers the desired area in the matching background (Figure 7). This approach of clustering sequences and selecting a scoring window is considered an attempt to reduce the physical library size. Another advantage of clustering the approach is that the combination of nucleic acid libraries (eg, nucleic acid libraries I, II, III, etc., FIG. 7) after several distant scoring windows is still much better scored than the lead sequence. Is to cover the array and structure space. The favorable result of this clustering process is that each of these sequence clusters only requires a much smaller physical library size than the combinatorial library, so the nucleic acid library that encodes each of the clusters is either in vitro or Small enough for complete in vivo screening.

  In certain embodiments of the invention, hit variant library scoring is used to select a population of sequences optimized for the desired function and to prepare a starting design for hit variant library II. Is done. The resulting hit variant library II scoring is used to examine the effect of modification and design enhancement on the variant profile. Hit variant library III, derived from a nucleic acid library (detailed in section 7 below), also determines library matches, and scoring functions in mapping sequence and structure space onto molecular target match backgrounds Is scored to assess the effectiveness of.

  In a specific embodiment, the standard term from the MM term includes an electrostatic solvation term and a solvation term accessible by the solvent calculated using a continuous solvent model of electrostatic solvation. These MM-PBSA or MM-GBSA methods show a good correlation between experimental and calculated values in free energy changes, with contributions from conformational entropy involving the backbone and side chains (Wang W, Kollman P (2OOO) J Mol Biol 303, 567-582). Compared to other scoring functions used in protein and drug design, MM-PBSA or MM-GBSA is a good physical model for scoring and handles a variety of issues with a consistent approach, To calculate the collective average of a system, multiple orbitals are required from a clear molecular dynamics simulation in water, making it computationally expensive, and the continuous solvent model is computationally slow. These accurate methods provide a benchmark for calibrating the simple scoring function used for library screening, or for studying some difficult mutations that are not possible with simple calculations.

7) Force field example of protein design Computationally allowed rotamer sequences using the van der Waals (vdw) interaction, a key interaction for scoring the correct packing interaction within the protein core The protein core sequence was designed by testing (Ponder JW, Richards FM (1987) J Mol Biol 193, 775-791). A simulated evolution using a stochastic algorithm can be used to select a group of sequences under the potential function; the rank order of the selected sequence's energy for the residues in the hydrophobic core of the protein is It correlates well with activity (Hellinga HW, Richards FM (1994) PNAS 91, 5803-5807).

  A similar algorithm was used to design proteins using estimation algorithms (Desjarlais J, Handel T, (1995) Protein Science 4, 2006-2018; Kono H, Doi J (1994) Proteins, 19, 244 -255). The function of the potential function on the target scaffold design sequence is an automated protein that maintains a constant amino acid sequence composition, including van der Waals, electrostatics, and surface-dependent semi-empirical environment-free energy or a combination of terms It is evaluated by the design method. Each additional term of the energy function depends on vdw for packing, electrostatic action for folding specificity, and environmental solvation terms for embedding hydrophobic residues and for exposing hydrophilic residues. Has been shown to progressively improve the performance of designed sequences (Koehl P. Levitt M (1999) J Mol Biol 293, 1161-1811).

  A self-consistent average field approach was used to test the energy surface to find the optimal solution (Delarue M, Koehl. (1997) Pac. Symp. Biocomput. 109-121; Koehl P, Delarue M, (1994) J. Mol. Biol. 239, 249-275; Koehl P, Delarue M (1995) Nat. Struct. Biol. 2, 163-170; Koehl P. Delarue M (1996) Curr. Opin. Struct. Biol 6: 222-226; Lee J. (1994) Mol. Biol. 236, 918-939; Vasquez (1995) Biopolymers 36, 53-70). Combinations of terms from molecular fields, knowledge-based statistical force fields and other empirical corrections have also been used to design protein sequences close to the native sequence of the target scaffold (Kuhlman B, Baker D (2000) PNAS 97, 10383-l0388). In addition to the steric repulsion of the protein core design, a structure-based thermodynamic term was included (Jiang X. Farid H, Pistor E, Farid RS (2OOO) Protein Science 9, 403-416). Knowledge-based potential has been used to design proteins (Rossi A. Micheletti C, Seno F, Maritan A (2001) Biophysical Journal 80, 480-490).

  Forcefield has also been optimized specifically in combination with dead end exclusion algorithms for protein design purposes (Dahiyat BI, Mayo SL (1996) Protein Science 5, 895-903). The energy function is broken down into pairs of functional types that combine molecular mechanical energy terms with specific solvation terms and used for residues at the core, boundary and surface positions; Used to examine the combinatorial rotamer sequence of The strictness of the force field used in protein design and a robust defolding protocol with a fixed backbone inevitably gives a high rate of false negative results; if a soft energy function or a flexible backbone is possible Many sequences that would be allowed are rejected. In addition, the energy function used for protein design is very different from common force fields (eg Amber or Charmm) that are widely used and tested for protein folding or stability studies (Gordon DB, Marshall SA, Mayo SL). (1999) Curr Opin Stru Biol 9, 509-513). Care should be taken when comparing sequences designed using a particular protocol with sequences from other methods, since a direct comparison may not be possible due to the false negatives involved in protein design protocols. There is a need to.

We do not have a high false negative rate in proteins for designing almost unlimited proteins, but only a few restriction regions are allowed to have modified sequences to improve protein function. We believe this poses a serious problem for the design of proteins for pharmaceutical applications. For example, only one or two residues of V _H CDR3 in a VEGF antibody actually improve its binding affinity, but many variants are allowed for V _H CDR3, but the frame For the work region, only a few variants are allowed for humanization. Thus, to identify these few variants in the target region, the most important for functional improvement is accuracy, not the scale or speed of the computer screening.
In some cases, molecular dynamics or other computational methods can be used to generate structural aggregates and aggregate average scores used to rank sequences (Kollman PA, Massova I, Reyes C, Kuhn B, Huo SH, Chong LT, Lee M. Lee TS, Duan Y, Wang W, Donini O, Cieplak P, Srinivasan P, Case DA, and Cheatham TE (2000) Acc. Chem Res. 33, 889-897). The average properties calculated from the aggregate structure show a better correlation with the corresponding data from experimental measurements.

6. Construction of Mutant Antibody Library Based on Lead Structure Template Alternatively, a mutant antibody library is constructed directly based on the 3D structure of the lead antibody and then screened for the desired function in vitro or in vivo. This approach takes a shortcut by avoiding the construction of hit variant libraries and directly evaluates sequences from hit libraries that are constructed by screening protein databases. This approach is described as pathway III in FIG. 1C or 1E-H.

As detailed in Section 3, there are several ways to build a hit library. One way to construct a hit library is to search a protein database for a segment whose sequence pattern matches the amino acid sequence of the region to be mutated, eg, CDR3 (CDR H3) of the heavy chain of the lead antibody. Conventional BLAST analysis may be used to search for sequences with high homology with CDR H3 sequences.
In some cases, PSI-BLAST may be used to search for CDR H3 sequence homologues of the template antibody.

  In some cases, a single target sequence and / or multiple sequence alignment may also be used to construct a profile hidden Markov model (HMM). This HMM is then used to search both near and remote human homologous pairs from protein sequence databases (eg, the Kabat database of proteins and the human germline immunoglobulin database of frameworks). A Kabat database of proteins of immunological interest from various species can be used to design diverse sequences of CDRs.

The sequences in the hit library selected using any of the above methods for sequence alignment or combinations thereof are profiled to identify the amino acid type at each position in the corresponding region (eg CDR H3) in the template antibody and The appearance frequency can be compared.
Each member of the hit library is implanted into the corresponding region in the template antibody (eg CDR H3) and tested for structural compatibility with the rest of the antibody using the scoring function described in Section 5 above. The

  Using a similar approach, a hit library was constructed based on the lead sequences from different regions of the lead antibody (eg, heavy and light chain CDR1, CDR2) for structural compatibility with the rest of the lead antibody. Can be tested. Combining these libraries allows for simultaneous mutation of different regions of the lead antibody, thus increasing the diversity of the variant antibody library.

  All mutant antibody sequences selected in these processes are pooled and screened for high affinity binding to the target antigen in vitro or in vivo.

7. Construction of nucleic acid library for experimental screening
To facilitate in vitro or in vivo functional screening, nucleic acid libraries are constructed that encode amino acid sequences selected using the above methods of the invention. The size of a nucleic acid library varies depending on the specific method for selecting and profiling amino acid sequences. For example, if too many amino acid sequences are selected and recombined, the size of the nucleic acid may reach> 10 ⁶ . In order to facilitate experimentally efficient and complete screening, amino acid sequence segmentation and reprofiles are performed to reduce the size of the nucleic acid library. As described in Section 5 above, for example, the profile used to create hit variant library II can also be used to measure the size of nucleic acid libraries for screening by in vitro or in vivo experiments. Is done.

  FIG. 6 shows an example of a method for constructing a nucleic acid library encoding the amino acid sequence of selected amino acid residues, eg, hit variant library II (FIGS. 4 and 5). To construct a nucleic acid library, variants in the amino acid profile are back-translated into the corresponding nucleic acid, taking into account the library size and codon usage (Figure 6).

For example, to obtain the simplest and smallest nucleic acid library that covers the diversity of an amino acid library, the only suitable codon used in the expression system is selected to encode the amino acid library. The corresponding nucleotide position variant profile (NT-PVP) is obtained from back translation of AA-PVP and the nucleic acid library size is determined from nucleotide combinatorial calculations. See, for example, FIGS. If this size is less than 10 ⁶ , a nucleic acid library is synthesized (eg, nucleic acid libraries I, II, III, etc., FIG. 7) and then screened by experiment. If the size is greater than 10 ^6, as described in sequence space or profile section 2, or the hit variant library II is divided into shorter library, or by re-testing the scoring distributions, new Create AA-PVP and generate small library size.

  By using NT-PVP, a degenerate nucleic acid library can be constructed without individually synthesizing selected nucleic acid sequences. In this approach, each library (eg, nucleic acid I, II, III, etc., FIG. 7) is programmed in one pass for each live library by programming an automated oligonucleotide synthesizer with a different mixture of nucleotides for each position. The rally can be combined, which saves cost and time. As a result, the sequence space of the degenerate nucleic acid library is significantly expanded to increase diversity. Although the size of a nucleic acid library (translated as hit variant library III) is larger than that encoding the designed amino acid sequence (eg hit variant library II), this approach to degenerate library construction Increases the likelihood of finding new sequences that have functions that are equivalent to or better than those originally designed.

  As a precaution, a nucleic acid library created using NT-PVP is back-translated into an amino acid sequence library to create hit variant library III, scored using an energy function, and hit variant live Assess the match between the sequence, structure space and library covered by Rally II (FIG. 13A). The final evaluation requires experimental selection data to prove the library match and the effectiveness of the scoring function in mapping sequence and structure space to match background.

8. Construction of mutant libraries for which structure is not available Mutant libraries can be constructed by dividing the sequence library into smaller components. This is also advantageous when only low resolution structures are available or no structure is available at all. A complex library is designed by dividing the sequence into overlapping sequential sequence segments. Each fragment can be targeted using a degenerate nucleic acid library. Note that variants with measured structure binding are simultaneously targeted using a degenerate nucleic acid library, even if only low resolution structures are available or no structure is available at all ( See examples below). This idea is described in Section 2 7) and is illustrated in the examples below (see FIGS. 28A-D for design and FIGS. 30 and 36 for experimental results).

  Briefly, a sequence variant library is broken down into small fragments as follows: Segments that are distant are often uncorrelated, so that widely separated mutations are processed independently, but are separated from each other in space. The binding fragments should be targeted simultaneously by the combinatorial nucleic acid library. Note that, in this case, structural information is preferred but not absolutely necessary (see examples below and FIGS. 28A-D for details).

Advantages of the invention By testing a large combinatorial space of amino acid sequences and structural motifs and scoring intermolecular interactions between proteins, a library of amino acid sequences can be screened on a computer. For the specific antibody-antigen complexes used herein, several libraries of antibodies are designed and constructed based on the lead sequence alone, the antibody structure, and the antibody-antigen complex structure, respectively. All libraries are biased towards lead antibodies in their sequence and / or structure; some are directed to specific antigens in the complex. That is, the antibody library is enriched and related from a collection of antibodies from a cDNA library or from random mutagenesis of specific antibody leads. These libraries are screened experimentally for affinity maturation using specific antigens. Various sequences are selected that differ from the lead antibody sequence in the CDR (see FIGS. 16A and 27). Some of the selected sequences exhibit a slower off-rate (suggesting higher affinity) than the lead antibody (or parent antibody). Among them, two variants (see FIGS. 30 and 36) (eg H97Y and / or S101T) are identical to critically important variants in the affinity matured V _H CDR3 sequence reported in the literature. However, one new mutant (S101R) showed better off-rate panning when measured in two independent experimental systems than S101T reported in the literature (Chen Y, Wlesmann C. Fuh G, Li B, Christinger HW, McKay P, de Vos AM (1999) J Mol Biol 293, 865-881).

  The present invention is considered advantageous in several aspects. First, this approach uses protein evolution data to extend the hit library in both sequence and structure space. Sequence search from simple BLAST to a more powerful profile-based approach (eg PSI-BLAST and / or HAMMER) to search close and distant homologues from evolutionary enriched sequence databases Law is used. The use of sequence profiles based on multiple sequence alignments of available lead structures allows more sequence space sampling than traditional multiple sequence alignment approaches. Thus, the methods used here increase diversity as well as the opportunity to find new hits or combinations of mutants with increased binding affinity.

  Second, sampling in sequence space also highlights the selection of sequence databases suitable for a particular purpose. For example, the use of diverse sequence databases to design CDRs and the use of sequences of human origin in human germline or framework regions can be used to design proteins for pharmaceutical applications where immunogenicity is a major issue. Should be used.

  Third, sequence design using existing entities from various databases is simple and highly efficient because one evolutionally enriched sequence or combination thereof is used. An improved but expensive computational scoring function can be applied to score the resulting pool of manageable sizes, which implicitly captures information including folding and expression.

  Fourth, the use of structural templates and optimized scoring functions can efficiently filter and reduce the size of combinatorial hit variant libraries prior to screening by any experiment. That is, a large virtual sequence space can be tested on a computer, and subsequent selection of a preferred set of sequences can direct the experimental synthesis of several small libraries covering diverse sequence spaces.

Fifth, control of library size (which is usually about 10 ³ to 10 ⁷ for nucleic acid libraries) facilitates direct functional screening experimentally. Since direct functional screening is the ultimate test for the effectiveness and accuracy of in silico methods, computer screening has experimentally tested several inherent limitations related to scoring functions and structural templates. The

  Sixth, the use of simple structural correlations to split long sequences allows control of library size, so that it can be experimentally managed without significant loss of diversity. This also makes it possible to design a sequence library of lead sequences with little available structural information.

  Finally, the adaptability and parameterization of the scoring function allows improvements in each experimental cycle. The clone screened in the experiment is the actual positional variant of the profile, which can be used as feedback to improve the scoring function by improving the various scoring terms.

  In summary, using computer screening in protein sequence and structure space and examining functional space by combining direct experimental screening, within experimental limits, as we show here for antibodies, It is a powerful approach to design.

Examples The method of the present invention was used for in silico construction of antibody libraries. To prove the present invention in antibody design, vascular endothelial growth factor (VEGF) is selected as the antigen of the principle proof experiment. A wealth of sequence and structural information is available for VEGF and its receptors (Muller YA, Christinger HW, Keyt BA, de Vos AM (1997) Structure 5, 1325-1338; Wiesmann C. Fuh G, Christinger HW, Eigenbrot C, Wells JA, de Vos AM (1997) Cell 91, 695-704), a complex of VEGF and its humanized antibody (Muller YA, Christinger HW, Li B, Cunningharn BC, Lowman HB, de Vos AM (1998) ) Structure 6, 1153- I167, and a complex of VEGF and its mature antibody (Chen Y, Wiesmann C, Fuh G, Li B, Christinger HW. McKay P, de Vos AM (1999) J Mol Biol 293, 865- 881) These provide a good basis for testing the method of the present invention: by using the method provided by the present invention, the antibody sequence, the structure of the antibody, the complex of the antibody and its antigen Using the increasing information from the structure, several digital libraries of anti-VEGF antibodies were designed in silico: single chain or double chain antibody binding. By two independent new phage display system of using the unit (phage display system), the high-affinity binding to VEGF, a population of antibody libraries were screened by in vitro.

1. In silico design of anti-VEGF antibody library
VEGF is a major angiogenic factor in development and is involved in solid tumor growth by stimulating endothelial cells. Mouse monoclonal antibodies were found to block VEGF-dependent cell growth and slow tumor growth in vivo (Kim KJ, Li B, Winer J, Armanini M, Gillett N, Phillips HS, Ferrara N (1993) Nature 362 , 841-844). This mouse antibody is humanized (Presta LG, Chen H, O'Connor SJ, Chisholm V, Meng YG. Krummen L, Winkler M, Ferrara N (1997) Cancer Res. 57, 4593-4599; Baca M. Presta LG, O'Connor SJ, Wells JA (1997) J Biol Chem 272, 10678-10684), and affinity matured using phage display and off-rate selection (Chen Y, Wiesmann C, Fuh G, Li B. Christinger HW, McKay P, de Vos AM (1999) J Mol Biol 293, 865-881). Complex formed between VEGF and parent antibody (Muller YA, Chen Y, Christinger HW, Li B. Cunningham, BC, Lowman HB, de Vos AM (1998) Structure 6, 1153-1167), and VEGF Of the complex formed with mature antibody (Chen Y. Wiesmann C, Fuh G, Li B, Christinger HW, McKay P, de Vos AM, Lowman HB (1999) J. Mol Biol 293, 865-881) An X-ray structure has been reported.

FIG. 9A shows the amino acid sequence of the variable region of a humanized anti-VEGF antibody (hereinafter “parent anti-VEGF antibody”) and an antibody that has been affinity matured from the humanized anti-VEGF antibody (hereinafter “mature anti-VEGF antibody”). Each of the amino acid residues in the V _H CDRs that were observed to contact the antigen are referred to below as “c”. FIG. 9B is an alignment of parent and mature anti-VEGF antibodies in V _H CDRs. Frameworks and CDRs are named according to Kabat standards (Kabat EA, RediMiller M, Perry HM, Gottesman KS (1987) Sequences of Proteins of Inununological Interest 4th edition, National Institutes of Health, Bethesda, Maryland). Differences in amino acid residues are highlighted in bold. As shown in FIG. 9B, only the mature antibody has two amino acid residues in V _H CDR1 (T28D and N31H) and V _H CDR3 (H97Y and S100aT) that differ from the parent antibody. There is no change in CDR2 after affinity maturation.

The mature anti-VEGF antibody has a 135-fold higher binding affinity for VEGF than the parent anti-VEGF antibody with four mutations (T28D, N31H, H97Y, and S100aT) in the V _H chain. Two of the mutations in V _H CDR3 individually improve the binding affinity to the parent antibody 14-fold (from H97Y) and 2-fold (from S100aT) (Chen Y, Wiesmann C, Fuh G, Li B, (See Table 6 of Christinger HW, McKay P, de Vos AM, Lowman HB (1999) J. Mol Biol 293, 865-881). A 14-fold improvement in affinity by H97Y alone in V _H CDR3 makes it the single most important mutation for affinity maturation, which results in two additional H bonds between antigen and antibody by H97Y. This is consistent with the observation of the composite structure by X-rays being created.

In the present invention, each motif such as antibody CDRs and frameworks can be targeted using a modular in silico evolutionary design approach. This modular design is described in FIG. It is understood that each CDR has only a small limited number of conformations (called standard structures). These structural features of the antibody, the extensive analysis of antibody structures, using the structured motifs of various regions of an antibody, such as the V CDRl in _L and V _H, CDR2, and CDR3, Evolution Provides an excellent system for testing dynamic array designs. These structures and sequence conservation have been observed across different species. Indeed, antibody scaffolds, or immunoglobulin folds, are one of the most abundant structures observed in nature and are highly conserved among various antibodies and related molecules.

  The present inventors believe that the parent anti-VEGF antibody described above can function as a model protein lead protein for affinity maturation of the command antibody using the method of the present invention. Mature anti-VEGF antibody (Chen et al., Supra) can serve as a reference or positive control to demonstrate the results obtained using the methods of the invention.

  Furthermore, the structural overlap revealed that the structure of the complex formed between VEGF and the parent antibody almost overlaps with the complex formed between VEGF and the mature antibody. Since the antibody structure before and after maturation is substantially the same, the structures of the parent antibody and the mature antibody were used to design a digital library of anti-VEGF antibodies using the method of the present invention. The methods of the invention can also be used to design antibodies with matches induced by antigen binding using sequence-based approaches or structural assemblies containing induced structural changes.

Using the parent anti-VEGF antibody as the lead protein and its V _H CDR3 as the lead sequence, a digital library of V _H CDR3 was constructed according to the method outlined in Figure 1D as pathway IV and in the schematic diagram in FIG.

The lead sequence contained V _H CDR3 of the parent anti-VEGF antibody and several amino acid residues from the flanking framework regions (FIG. 9B). In summary, a hit library was constructed by searching for and selecting hit amino acid sequences that have distant homology with V _H CDR3. A variant profile was created to describe all variants at each position based on the hit library and was filtered with a certain cutoff value to reduce the size of the resulting hit variant library to within computer or experimental limits. Variant profiles were also constructed to facilitate: i) sampling of sequence space covering suitable regions in the matching background; ii) splitting and synthesizing degenerate nucleic acid libraries targeting suitable peptide assembly sequences Iii) screening the antibody library by experiment for the desired function; and iv) analysis of experimental results with feedback for further design and optimization.

  From the available X-ray structure of the complex formed by VEGF and anti-VEGF antibody, a lead structure template was obtained. The complex structure of VEGF and parent anti-VEGF antibody is called 1BJ1, and the complex structure of VEGF and mature anti-VEGF antibody is called 1CZ8. The results from the 1CZ8 structural template were similar to those from 1BJ1 in the relative rank of the scanned sequence.

1) Lead arrangement
The V _H CDR3 lead sequence was taken from the parent anti-VEGF antibody according to the Kabat classification, and amino acid residues CAK and WG from adjacent framework regions flank the V _H CDR3 sequence at the N-terminus and C-terminus, respectively (FIG. 9B). . As shown in FIG. 9B, the parent antibody and mature antibody V _H CDR3 differ only in two amino acid positions. Only the parent antibody V _H CDR3 sequence was used to create an HMM for protein database searches.

2) Hit library and variant profile HMMs created using a single read sequence (SEQ ID NO: 5) (Figure 9B) were calibrated and Kabat database (Johnson, G and Wu, TT (2001) Nucleic Acids Research, 29, 205-206). All sequence hits that are larger than the predicted or E value are listed and aligned using the HAMMER 2.2.1 package. After removing duplicate and mature sequences from the hit list (ie, SEQ ID NO: 6 assuming no mature sequences are available), the remaining 107 hit sequences for the lead HMM form a hit library.

As shown in FIG. 10A, the 107 hit sequences range from 35 to 95% of the lead sequence from the Kabat database. The evolutionary distance between hits is displayed in a dendrogram in FIG. 10B using the program TreeView 1.6.5 ( http://taxonomy.zoology.gla.ac.uk/rod/rod.html ). The phylogenetic tree is obtained from the adjoining method (Saitou N, Nei M (1987) Mol Biol Evol 4, 406-425 in ClustalW 1.81 (Thompson JD, Higgins DG, Gibson TJ (1994) Nucleic Acids Research 22, 4673-4680). ).

  The variant profile at each position is shown in FIG. The AA-PVP table in FIG. 11 gives the number of occurrences of each amino acid residue at each position. The variant profile below the table list lists all variants found from the database using the lead sequence as the reference sequence in order of decreasing occurrence at each position.

The diversity of the 107 hit sequences from the hit library can be seen in the AA-PVP table showing both the frequency and change of amino acids at each position. Comparing the sequence differences between the parent anti-VEGF antibody and the mature anti-VEGF antibody in V _H CDR3, two different amino acids (H97Y and S100aT using the Kabat numbering system) are included in the variant described at each position . H97Y (Chen Y, Wiesmann C, Fuh G, Li B, Christinger HW, McKay P, de Vos AM, Lowman HB () reported to be the most important variant to increase the binding affinity of the mature sequence 1999) J. Mol Biol 293, 865-881) is easily identified as the most frequent residue (˜25%) at that position. S100aT accounts for ~ 5% of the variants identified at that position. The lower right part of FIG. 11 shows the variant profile after filtering variants that appear with a cut-off frequency of 10 or less. After filtering, it becomes clear that only a few variants are allowed at each position of the sequence; however, with such a cut-off, energy scoring will be maintained but in the mature sequence Some important mutants like S100aT may be lost.

  Variant profiles from the evolutionary selection pool provide data useful for identifying positions in the lead sequence that can be altered or fixed. This site can be divided into three categories: i) Sites that have preserved their structure remain conserved as they evolve. Frequent residues can be used to maintain the scaffold of the target motif at these positions; ii) Focused mutagenesis should be used to target variable functional hot spots; iii) A combination of i) and ii) to stabilize the target scaffold while simultaneously providing variability in functional hot spots.

  Since the set of amino acids from the functional variant has been evolutionarily selected and optimized, they should be included in the functional hotspot according to the frequency in the variant profile. In addition, to satisfy computer and experimental constraints, each position variant can be filtered or prioritized to include other useful mutant candidates or to exclude unwanted mutant candidates.

3) Structure-based evaluation of combinatorial sequences of hit libraries Variant profiles are beneficial on preferred amino acid residues, with specific variants at each position and in the preferred order, but this is a huge number The recombinant of is materialized. Certain filtering using a frequency cut-off reduces combinatorial sequences that need to be assessed by computer screening or directly targeted by an experimental library. There are many combinatorial sequences that need to be scored and evaluated in the final sequence for experimental screening, even with a cutoff applied to the variant profile (shown in FIGS. 13A-C and 28A-D).

Structure-based scoring is applied to screen the combinatorial sequences that form the hit library and the hit variant library. The side chain of the V _H CDR3 of the parent anti-VEGF antibody was replaced with the corresponding amino acid variant rotamer from the variant library at each residue position. Rotamer conformation was created and optimized using the program SCWRL® (version 2.1) using a backbone-dependent rotamer library (Bower MJ, Cohen FE, Dunbrack RL (1997) JMB 267, 1268 -82).

  Scoring uses Amber94 of CONGEN [Bruccoleri and Karplus (1987) Biopolymers 26: 137-168] in the presence and absence of the structure of the antigen VEGF to find the optimal rotamer and minimize energy by 100 steps Was done. FIGS. 12A and B show the anti-VEGF variant library based on total energy calculated using CONGEN with and without VEGF antigen, using the structures of the parent (1BJ1) and mature (1CZ8) anti-VEGF antibodies, respectively. Indicates the energy score. Parent and mature sequence scores are marked in FIGS. 12A and B. The mature sequence scored better than the parental sequence in both structures with and without antigen, suggesting that the variant of the mature sequence stabilizes both the antibody structure as well as its complex with the VEGF antigen. FIG. 12C shows that the scoring of sequences in the presence and absence of antigen is generally correlated, indicating that sequence screening based on antibody structure alone also has good binding affinity for that antigen. It is suggested that it is a good sequence candidate having

As shown in FIGS. 12A and 12B, there are many sequences in various variant libraries with higher scores than the parental and mature sequences. The energy scoring distribution in the energy schematic is 10 selected from the hit variant library of V _H CDR3, its combinatorial peptide, the degenerate nucleic acid library combinatorial library, and experimentally selected sequences The sequence is shown in FIG. 13A. For scoring, Y97 in the mature sequence is consistently better than H97, which is consistent with experimental observations (Chen Y, Wiesmann C, Fuh G, Li B, Christinger HW, McKay P, de Vos AM, Lowrnan HB (1999) J. Mol Biol 293, 865-881). T100a is preferred over S100a as in the mature sequence, while both T and S are equally preferred in the 100b position. That is, structure-based energy scoring is a separate approach to re-profile variants at each position in a hit variant library created based on evolutionary sequence profiling originally selected from protein databases. Provide an independent method.

  In order to measure the accuracy of the scoring function using the simple energy function used in CONGEN, the energy of a randomly selected set of sequences is converted to side chain entropy, nonpolar solvation energy and electrostatic solvation energy. Calculated using an improved custom scoring function including: Three energy terms were calculated: side chain entropy, nonpolar solvation energy, and electrostatic solvation energy. There was an additional option to calculate the skeleton entropy of the loop. Side chain entropy was calculated using the conformation search command CGEN in CONGEN. The option under CGEN was defined as performing an individual side chain formation tree search using strain space at each bond (section) to expand the side chain conformation tree. These have SEARCH DEPTH and SIDE choices for each side chain, with the SGRID parameter set to AUTO so that each strain angle rotates at discrete intervals. Specifically, the AUTO setting used a strain grid angle of 30 degrees for bonds with rotational symmetry, such as in phenyl, tyrosyl, carboxyl, and amino groups, and 10 degrees for all others. The MIN option was set so that rotational sampling starts at the local energy minimum for each strain. The VAVOIDO option was also included to turn on van der Waals repulsion avoidance. The MAXEVDW parameter was set to a relatively high 100 kcal / mol to mitigate van der Waals repulsion and allow more conformers to be obtained during the calculation.

  The side chain conformation search was repeated for each mutant residue side chain. The code outputs the “number of lower leaves” reached by a tree search in the conformation space (this is the number of tree searches completed). As an approximation, the side chain conformation search treats each residue independently so that the computation time is minimized. For residues that do not contact each other, this is a good approximation. For residues that may contact each other, conformational calculations tend to overestimate the number of conformations. Since we use a relatively high van der Waals repulsion to obtain more sampling, the error due to residue contacts must be reduced in the artificial measurement of conformational space. In addition, as the number of conformations increases, the meaning of errors due to residue contacts tends to decrease, which means that the relative change in entropy is the logarithm of the number of conformations in the mutant and the reference structure. Because there is.

Non-electrostatic solvation energy can be calculated using the GEPOL (Pascual-Ahuir JL, Silla E (1993) J Comput Chem 11, 1047-1060) command used in CONGEN, with a scaling constant of 70 cal / mol / A ² ( Tunon I, Sma E, Pascual-Ahuir JL (1992) Prot Eng 5, 715-716) and calculated to be proportional to the molecular surface. The NDIV that specifies the division level of the surface triangle is set to 3. Values range from 1 to 5, with 5 giving the highest accuracy but significantly increasing CPU time requirements. RGRID is set to 2.5A, which describes the spatial grid used to find neighbors.

  Electrostatic solvation energy was calculated using the Finite Difference PB (FDPB) method used in the UHBD program (Davis ME, Madura JD, Luty BA. McCammon JA (1991) Comput Phys Commun 62, 187-197). . For the area around the mutation, the focusing method was used. The automatic protocol generates three grids: a coarse, fine, and focal grid. The grid units are 1.5, 0.5 and 0.25 angstroms, respectively. The focal grid is a spherical grid that spans the Cartesian volume occupied by mutated residues. The fine grid is a spherical grid that spans the entire volume of the protein or complex. The coarse grid is a spherical grid that is set to about twice the size of the fine grid on each axis, covering about 8 times the capacity of the fine grid. A coarse grid is useful to account for long term solvent action and sets the fine grid boundary conditions. Similarly, the fine grid accounts for the electrostatic contribution within the protein and sets the boundary conditions for the focal grid. The focal grid describes more details of the localization effects due to mutation. The dielectric constant inside and outside the protein is set to 4 and 78, respectively. The temperature is set to 300 Kelvin and the ionic strength is set to 150 mM. The maximum repetition is set to 200. The calculation is repeated with a uniform dielectric constant so that the internal and external dielectric constants are set to 4, and the difference between the two energies is calculated. The latter calculation shows the energy from applying a charge to the grid.

  A custom scoring function or molecular mechanical energy using solvation terms from Amber94 force field in CONGEN and PB in UHBD used here proved to be equal to MM-PBSA or MM-GBSA. In particular, the energy function is the collective average of energy functions (Kollman PA, Massova I, Reyes C, Kuhn B, Huo SH, Chong LT, Lee M, Lee TS, Duan Y, Wang W, Donini O, Cieplak P, Srinivasan P , Case DA, and Cheatham TE (2OOO) Acc. Chem Res. 33, 889-897) to provide a more accurate method for scoring sequences and their variants to provide structural assembly by molecular dynamics calculations. When using the body, better agreement than the experimental data (Sharp KA. (1998) Proteins 33, 39-48; Novotny J, Bruccoleri RE. Davis M, Sharp KA (1997) J Mol Biol 268, 401-411) Show.

4) Reduced variant profile of hit variant library To reduce the size of the library candidate while maintaining most suitable residues, the variant profile from the hit variant library was filtered as described above. The upper part of FIG. 13A shows a reduced variant profile of the top 10 selected sequences from the hit variant library after exclusion by structure-based evaluation with less frequently occurring amino acids than the cutoff value. Show. The list was chosen as a blind method for proof of the present invention in selecting diverse sequences that can bind to the target antigen. The 10 selected sequences from one computer screen of the variant library have several common characteristics: for example using 1bj1 or 1cz8 as template structure in the presence or absence of VEGF antigen For the 200 highest ranked sequences, R94, Y97 and R100a are always better than the corresponding K94, H97 and S100a. As will be shown later in the experimental section, in fact H97Y is a good variant for affinity maturation. However. Mutations such as K94R and S100aR to arginine are an interesting example: while K94R is not a good variant of affinity maturation, K94R is at the border between CDR and framework according to the Kabat classification, and the human frame It is evolutionarily suitable for work sequences. As evidenced by the experimental selection of the present invention, K94 is preferred over R94 (FIGS. 30 and 36) because the R94 mutation increases the binding affinity of anti-VEGF antibodies (Baca M, Presta LG , O'Connor SJ, Wells JA (1997) J Biol Chem 272, 10678-10684). On the other hand, S100aR is one of the most important single mutations of V _H CDR3 maturation and is preferred over S100aT, as reported in the literature, and has been used in multiple rounds of phage display under severe washing conditions. It persists in panning (see Figures 30 and 36).

  Some residues, such as lysine from the wild type (eg K94R) may be included so that some important variants in the variant profile are not missed, Below the cut-off values used, or they are worse in score than arginine due to problems associated with computational assumptions involving charged residues with long side chains or conformational changes. Thus, for charged residues with long side chains (eg, arginine and lysine), the design library includes residues predicted at the same position and wild type residues. The reduced variant profile was used to calculate hit variant library II as a blind method of the method of the present invention used here for the design of functional libraries with diverse sequences from the lead sequence.

5) Hit variant library II-Amino acid library designed from scoring selection and optimization Select top sequences based on suitable scores and / or the presence of residues that may participate in suitable interactions A strategy was used to identify clusters of amino acid sequences for nucleic acid library design (Figure 7). As noted above, clusters of the sequences of FIGS. 13A-C (eg, 10 sequences) for V _H CDR3, CDR1 and CDR2 from computer evaluation, respectively, were selected for further experiments in vitro. The peptide sequence and variants at each position are listed in the upper left part of FIG. 13A. A combinatorial library was created based on the filtered variant profile to form hit variant library II. For the V _H CDR3 of anti-VEGF (Figure 13A), the size of the hit variant library II is 72 based on the variant profile of the top 10 selected sequences that score better than the lead sequence (variant live used) The top 10 rank array in the rally). See FIGS. 13B and C for V _H CDR1 and CDR2.

6) Construction of degenerate nucleic acid library based on hit variant library II The hit variant library constructed above was targeted using a single degenerate nucleic acid library. The lower part of FIG. 13A shows the nucleic acid sequence profile resulting from reverse translation using the optimal E. coli codon for V _H CDR3. Based on this profile, a degenerate nucleic acid library was synthesized by incorporating a mixture of bases at each degenerate position. As a result of the synthetic combinatorial action, this degenerate nucleic acid library encodes an extended amino acid library of 4608 size (referred to as “hit variant library III”). See FIGS. 13B and C for V _H CDR1 and CDR2.

The degenerate nucleic acid library constructed above was cloned into a phage display system, and a phage-displayed antibody (ccFv) was selected based on binding to immobilized VEGF coated on a 96-well plate. . As detailed in Section 2 below, with a small nucleic acid library size, perform 1-3 rounds of washing and selection (ie panning), select clones that show a positive ELISA reaction, and see Figure 14B for V _H CDR3 Sequenced as indicated. Positive clones show diverse variant profiles at the targeted location, along with the incorporation of degenerate codons into the nucleic acid library.

The results of the designed versus experimentally screened antibody sequences are analyzed in FIGS. Brief sequence of V _H CDRs 1, 2, and 3 when is designed in accordance with the method of the present invention described above for V _H CDR3. The top 10 sequences selected from libraries screened on computers for V _H CDR3, CDR2 and CDR2 and their variant profiles are shown in FIGS. FIG. 16A lists the amino acid sequences experimentally selected from V _H CDR1, CDR2 and CDR3 of the degenerate nucleic acids shown in FIGS. 13A-C. FIG. 16B shows the distribution of sequence identity of sequences selected from the V _H CDR1, CDR2 and CDR3 libraries relative to the corresponding parental sequences of anti-VEGF V _H CDR1,2,3. FIG. 17A shows the relationship between four different libraries: a designed amino acid sequence, a combinatorial library of amino acid variants of the designed sequence, and a combinatorial degenerate nucleic acid library and a fully degenerate nucleic acid library that encode a unique amino acid sequence; Using the anti-VEGF V _H CDR3 library from round 3 as an example, the distribution of experimentally selected positive clones shown in X is shown (see table in FIG. 17B). Distribution in different libraries depends on the selection conditions, the effectiveness of the library design, the relative size of the selected clones versus the library or number of clones sequenced. FIG. 17B is a table showing the relationship between the four libraries (FIG. 17A) and the distribution of experimentally selected sequences of positive clones of the anti-VEGF V _H CDR1,2,3 library.

Detailed analysis of V _H CDR3 will be described later. FIG. 14A shows UV readings of ELISA positive clones identified in round 1 and round 3 of a functional anti-VEGF ccFv antibody with V _H CDR3 encoded in a degenerate nucleic acid library (FIG. 13A). FIG. 14B shows the V _H CDR3 sequence of positive clones from rounds 1 and 3 selection by phage display of the nucleic acid library shown in FIG. 13A. It is clear that there are large variations at several positions different from the V _H CDR3 of the parent anti-VEGF antibody and mature anti-VEGF antibody (FIGS. 9B and C), and that many diverse sequences are selected. FIG. 14C illustrates a phylogenetic tree of positive clones showing screened sequence diversity. The sequence identity of selected positive clones from V _H CDR3 shown in FIG. 14B ranges from 57 to 73 percent relative to the parent V _H CDR3 sequence. 15A-B are pie charts showing the decomposition of the screening sequence into three groups in the first and third rounds: the designed amino acid sequence, the combinatorial amino acid sequence from the designed amino acid sequence, and the synthesized degenerate nucleic acid. A unique combinatorial amino acid sequence encoded by a library. Since only a limited number of positive clones are selected from each round for sequence analysis, the figure only illustrates the design, its combinatorial amino acids, and the percent of selected sequences from the nucleic acid library.

  These experiments show that by using the methods of the present invention, not only have diverse sequences and phylogenetic distances, but also related biological functions (eg, the ability to bind to a target antigen such as VEGF). It was proved that the antibody having was selected.

FIG. 18 summarizes the progressive evolution of sequence design using amino acid sequence scoring results at each stage of V _H CDR3 as an example. A schematic diagram from left to right shows a lead sequence, a hit library created from database search, a combinatorial sequence screened with a computer in hit variant library I, a selected group of designed amino acid sequences (hit variant library II), The degenerate nucleic acid library obtained from the Library II profile and the positive clones and sequence energy spectra screened in the experiment are shown. This process is repeated with feedback from the experiment until sequences with enhancement or desired properties are experimentally selected.

19A-D show a comparison of sequence homology distributions based on lead sequences or lead sequences obtained from alignments based on multiple structures. FIG. 19A shows a lead profile generated from a structure-based multiple sequence alignment. The structural motif of the lead sequence is used to search the protein structure database (PDB databank) for similar structures within a certain distance cutoff. The root mean square difference (RMSD) between each structure and the V _H CDR3 structural motif (colored in blue) is within 2 km. Corresponding multiple sequence alignments are shown to the right of FIG. 19A along with corresponding structure PDB IDs and colors.

FIG. 19B shows a variant profile of 251 unique sequences of a hit library generated based on the read sequence profile of the V _H CDR3 of the parent anti-VEGF antibody. The lower part of the figure shows the filtered variant profile obtained using a 5% cut-off of frequency (in this case 12). Interestingly, important variants (H97Y and S100aR or S100aT, see FIGS. 30 and 36) are also observed in the variant profile generated from the lead sequence profile.

FIG. 19C shows the distribution of sequences from the hit library relative to the parent V _H CDR3 sequence. Circles indicate that up to 36% sequence identity can be identified using a single parent sequence for HMM searches. The triangles indicate that using the lead sequence profile from the structure-based multiple sequence alignment, further down to nearly 20% sequence identity can be found. The sequence search strategy used here can find diverse hits with distant homology (up to 20% low) to the lead sequence.

  FIG. 19D shows the conceptual evolution of the inventive method used here to search for promising candidates in sequence, structure and functional space. The basic idea is to extend the diversity of hit and variant libraries in the sequence and structure space to find candidates with improved functionality in the functional space. The diversity and / or size of hit and variant libraries is increased, for example, by finding distant homologues of the lead sequence or sequence profile (shown in FIG. 19A), but the intersection between sequence, structure and functional space is As the probability of finding an array with improved function increases, the focus can be on a smaller area.

  It is clear that using structure-based multiple sequence alignment as a profile for creating an HMM model makes it possible to find distant homologues of the lead sequence (up to 20% sequence identity of the sequence in question) It is. The inventive methods described herein will be more powerful in designing antibody CDR libraries as the available sequence and structural information increases and the accuracy of the scoring function improves.

2. Functional screening of antibody libraries designed in vitro An antibody library designed in silico based on the lead sequence of the parent anti-VEGF antibody using the above method was developed into a new phage display system. ) Was used to test for the ability to bind to the antigen VEGF. The structure of the parent antibody or mature antibody was used for structure-based computational screening. In contrast to the general approach to screening for antibodies that take the form of single chain antibodies (scFv) (see other novel methods shown in FIGS. 20 and 32), a double chain antibody library is expressed and the surface of the bacteriophage Was displayed. Double-chain antibodies are formed by heterodimerization of _VH and _VL to functionally mimic antibody Fabs. This double chain antibody is referred to as “ccFv”. As described above, a ccFv library was constructed based on the degenerate nucleic acid library encoding the antibody sequence designed in silico.

  The reasons for designing ccFv, the construction and expression of the ccFv library, and the functional screening of the ccFv library are detailed below.

1) ccFv-Heterodimeric coiled-coil stabilized antibody Antibody Fv fragment is the smallest antibody fragment containing the entire antigen binding site. Fv fragments have very little interaction energy between the two _VH and _VL fragments and are often too unstable for many applications under physiological conditions. Of course, the V _H and V _L domains are linked by interchain disulfide bonds located in the constant domains (C _H 1 and C _L ) to form Fab fragments. It is known that V _H and V _L fragments can also form single chain Fv antibody fragments (scFv) by artificially connecting the carboxy terminus of one fragment and the amino terminus of another fragment with a short peptide linker. It is.
The present invention provides a new strategy for stabilizing _VH and _VL heterodimers. A unique heterodimerized sequence pair was designed and used to create a Fab-like functional artificial Fv fragment ccFv (FIG. 20). Each of the heterodimeric sequence pairs was derived from the heterodimeric receptor GABA _B R1 and R2. The sequence pair specifically forms a coiled-coil structure and mediates the functional heterodimerization of GABA _B -R1 and GABA _B -R2 receptors. In order to engineer the _VH and _VL heterodimers of the antibody, GABA _B R1 and GABA _B -R2 coiled coil domains (GR1 and GR2 respectively) were fused to the carboxy terminus of the V _H and _VL fragments, respectively. That is, the functional pairing ccFv (coiled coil Fv) of _VH and _VL is mediated by specific heterodimerization of GR1 and GR2. In addition, the carboxy terminus of the GR1 and GR2 domains was modified by adding a flexible spacer or flexon “SerArgGlyGlyGlyGly” [SEQ ID NO: 7] (or “GlyGlyGlyGlySer” [SEQ ID NO: 18]). In order to further stabilize the heterodimeric ccFv, a “ValGlyGlyCys” [SEQ ID NO: 8] spacer was added to the C-terminus of the GR1 and GR2 coiled coils to introduce a pair of cysteine residues, so that the coiled coil GR1 and GR2 mediated heterodimers Covalent bonds can be made with disulfide bonds (FIGS. 20 to 21). ccFv had a molecular weight of 35 kDa and was expressed in E. coli.

2) Anti-VEGF (AM2-ccFv) and its display on the phage surface The _VH and _VL sequences of the anti-VEGF antibody AM2 are shown in FIGS. This is an antibody designed by modifying a parent anti-VEGF antibody. In order to facilitate efficient cloning of the designed CDR sequence library, unique restriction sites were introduced into the _VH and _VL genes of the parent anti-VEGF antibody. Both the AM2 _VH and _VL genes were cloned into a fuzzymid vector to construct the phage display vector pABMD12. Figures 23A and 23B show the vector map and sequence [SEQ ID NO: 17], respectively. This vector will express two fusion proteins: V _H -GR1 and V _L -GR2-pIII fusion. The expressed V _H -GR1 and V _L -GR2-pIII fusions are secreted into the periplasmic space where they are heterodimerized and are stable ccFv antibodies ("AM2-ccFv") via the coiled-coil domain. Called).
To display AM2-ccFv on phage, the pABMD12 vector was transformed into bacterial TG1 cells. TG1 cells harboring the pABMD12 vector were further superinfected with KO7 helper phage. Infected TG1 cells were grown overnight at 30 ° C. in 2 × yt / Amp / Kan. Phadimide particles were precipitated from the culture supernatant with PEG / NaCl, resuspended in PBS and library selection against immobilized VEGF. After binding for 2 hours, unbound phage was washed away and bound phage was eluted and amplified for the next round of panning.

  The binding of ccFv displayed on the phage particle was detected by antigen binding activity by phage ELISA. Briefly, an antigen (eg, VEGF) was first coated on an ELISA plate. After blocking with 5% milk / PBS, the phage solution was added to the ELISA plate. Phage bound to the immobilized antigen was detected by incubation with an HRP-conjugated anti-M13 antibody against the phage coat protein pVIII. The substrate ABTS [2,2'-azino-bis (3-ethylbenzthiazolone-6-sulfonic acid)] was used to measure HRP activity. This assay proved to be highly specific for AM2.

  A single chain AM2 antibody (AM2-scFv) phage was also prepared for comparison with AM2-ccFv in the phage ELISA described above. As shown in FIG. 24, the apparent binding affinity of AM2-ccFv phage to immobilized VEGF is almost an order of magnitude higher than AM2-scFv phage. That is, when displayed on phage particles, it is concluded that both AM2-ccFv and AM2-scFv are functional.

3) Enrichment of ccFv phage from model antibody library
To prove that AM2-ccFv-displayed phage can be enriched from background phage, we performed panning experiments and selected for AM2-ccFv phage from the “model library”. The model library was prepared by mixing AM2-ccFv phage with irrelevant AM1-ccFv display phage at a ratio of 1:10 ⁶ or 1:10 ⁷ . Two rounds of panning were performed on the immobilized VEGF antigen. 100 μl of 2 μg / ml VEGF was coated to each well in a 96 well plate. After blocking with 5% milk in PBS, 1 × 10 ¹² library phage in 2% milk / PBS was added to the wells and incubated for 2 hours at room temperature. The phage solution was discarded and the wells were washed 5 times with PBST (0.05% Tween 20 in PBS) and 5 times with PBS. Bound phage was eluted with 100 mM triethylamine and added to the TG1 culture for infection. Phages prepared from infected TG1 cells were used for the next round of panning and the above phage ELISA. After each round of panning, the ratio of recovered AM2-ccFv phage to AM1-ccFv phage was measured by analyzing TG1 colonies infected by PCR. Due to the sequence difference between the AM2-ccFv gene and the AM1-ccFv gene, a pair of primers that specifically amplify the AM2-ccFv gene but not the AM1-ccFv gene were designed. As shown in FIG. 25A, phage from the second round of panning give very high ELISA readings, and high AM2-ccFv phage from both 1:10 ⁶ and 1:10 ⁷ libraries after two rounds of panning Suggests that concentration has been achieved. PCR analysis confirmed that the incidence of AM2-ccFv phage was 4.4% from the 1:10 ⁷ library after the first round of panning and 100% from the second round of panning (FIG. 25B). .

4) Construction and panning of a designed ccFv antibody phage library As outlined in Figure 8, a modular evolutionary approach was used to construct antibody libraries for computational and experimental screening. Oligos encoding the designed library of CDR sequences were synthesized and amplified by PCR. The amplification primer contains a restriction site for cloning the synthetic CDR sequence into the pABMD12 vector. Three _VH libraries were prepared for AM2-ccFv using NheI and XmaI, XmaI and SpeII, and PstI and StyI restriction sites for insertion of CDR1, CDR2, and CDR3, respectively. After ligation, the DNA was transformed into TG1 cells. Phages were prepared from TG1 cells by KO7 helper phage infection. Three rounds of panning against immobilized VEGF were performed as described below. 100 μl of 2 μg / ml VEGF was coated to each well in a 96 well plate. After blocking with 5% milk in PBS, 1 × 10 ¹² library phage in 2% milk / PBS was added to the wells and incubated for 2 hours at room temperature. The phage-containing solution was discarded and the wells were washed 5 times with PBST (0.05% Tween 20 in PBS) and 5 times with PBS. Bound phage was finally eluted with 100 mM triethylamine and added to the TG1 culture for infection. Phages prepared from infected TG1 cells were used for the next round of panning. After each round of panning, 94-376 clones were picked for phage ELISA (FIGS. 26A and B). Positive clones from phage ELISA were amplified by PCR and sequenced. The DNA sequence was then translated into an amino acid sequence. The encoded amino acid sequences from the three libraries are listed in the table of FIG.

5) Library design based on sequences with and without constraints from tertiary structures or structural models
Another strategy for designing CDR libraries is to divide the CDR sequences into unrelated and related segments in the structure space and at structurally linked sites such as the N-terminal and C-terminal regions of the CDR loop. Detecting covariant variants (in most cases a low resolution structure is sufficient). For example, FIG. 28A shows a composite variant profile of anti-VEGF antibody V _H CDR3 obtained by combining the filtered hit variant profile of V _H CDR3 with other variants from experimental selection. We want to prove that variants from various sources can be combined to create a composite variant profile for library construction. In order to confirm that each small variant profile can be covered by a nucleic acid library having a diversity of about 10 ⁶ to 10 ⁷ , this variant profile can be broken down into several segments of smaller variant profiles. Note that in the degraded segment library, the combination of the V _H CDR3 mature sequence with H97Y and S101T (Kabat S100aT) was intentionally avoided (FIGS. 28A-D).

Figures 28A-D show a sequence library of anti-VEGF V _H CDR3. Decompose the library into three segments: FIG. 28D covers N- and C-termini that may contain bound variants (1-3), FIG. 28C contains segment (4), 28D contains another segment (5). All three segments are covered by a nucleic acid library of about 10 ⁶ size: (1-3) in FIG. 28B is targeted by three degenerate nucleic acid libraries, while in FIGS. 28C-D (4) and (5) are targeted by separate degenerate nucleic acid libraries.

The reasons for designing these segment libraries are as follows. Structurally distant segments are often irrelevant, so mutations that are widely separated in space can be handled independently. For CDR3 loops, the sequence is divided into three segments: the first and third segments (the base of the loop) form one profile for the library design and the top of the loop is for the library design It is broken down into two profiles of size 10 ^{6 in} a degenerate nucleic acid library. As shown in FIG. 28B, the N-terminal and C-terminal fragments that bind together in space (the sequence that forms the base of the loop generally correlates with the loop closure) are divided into three degenerate oligonucleotides (1 to 3) should be targeted simultaneously by a combinatorial nucleic acid library with only Simple criteria such as Cα or Cβ distance matrix can be examined to identify correlated segments (see Figure 28A for structure and distance contact matrix in Cα atoms within 8 Å). In some cases, a more detailed interaction matrix can be mapped to examine the number and type of interactions, but the basic principle is the same as identifying correlated segments.

The top library (eg (4) and (5) in FIGS. 28C and 28D) is often irrelevant. If each library is limited to a size range (<10 ^{6 in} FIGS. 28C-D) that can be easily managed by experiment, then the degenerate oligonucleotide libraries along the primary sequence, Targeted. In order to maintain a small level of local correlation between the resulting libraries, there must be positional overlap between the fragments. Similarly, long segments can be split into overlapping segments to span the length of the sequence and a corresponding library can be created.

  The resulting reprofile is further modified and enhanced based on observed experimental or structural or computational criteria. These include various positions of known hydrogen bonds with additional polar amino acids, areas of high van der Waals contact with large aliphatic or aromatic groups, or areas that benefit from increased flexibility with glycine. . In experimental feedback, variants may be added based on assay results from early screening as the basis for subsequent design improvements, as evidenced by the variant profile of FIG. 28A. A more sophisticated analysis may take into account the bonding of amino acid groups within the sequence (eg, salt bonding or hydrogen bonding).

6) Off-rate panning for ccFv library L14 To select high affinity antibodies, an off-rate panning process was performed to select library L14 (see FIGS. 28A-D). The strength of the interaction between the antibody fragment on the phage surface and the immobilized antigen is measured by the interactive affinity measured by the on-rate (binding rate) and the off-rate (dissociation rate). According to the test, high affinity antibodies usually have a slow off rate, while low affinity antibodies have a fast off rate, and these on rates are similar. Off-rate panning was designed to promote dissociation of antibodies with low affinity from the immobilized antigen, and the stringency of washing conditions (gradient) was gradually increased. A wash solution with increased stringency was applied to wash out the low affinity phage, leaving the phage with higher affinity (ie slow off-rate). Thus, a phage that survives stringent wash conditions will have a high affinity, and a phage whose presence predominates should have a higher affinity than a low presence. We also demonstrate comparable off-rate panning at the phage level using two independent display platforms (FIGS. 20 and 32) under various panning conditions (FIGS. 29 and 35A-B). A consensus of positive clones or clones arising from phage panning strongly suggests having an enhanced affinity for the antigen compared to a sequence or variant, the parental sequence.

L14 was prepared as an anti-VEGF V _H CDR3 library by degrading the V _H CDR3 sequence into short overlapping segments (see FIGS. 28A-D). A number of panning conditions were manipulated to distinguish slow off rates. In the first two rounds of panning, the wells were gently washed 6 times with PBST and PBS to remove low affinity phage. Starting from panning 3, the bound phages were washed for a longer time to remove those with a faster off-rate (dissociation). The duration and stringency of such dissociation periods increased with the number of pannings (Figure 29), dissociating and removing more phage; in contrast, those with slow off-rates (high affinity) bound It remains and is finally concentrated. Panning 3 was performed in PBS at 37 ° C. for 1 hour as described in FIG. 29 (PBS was used fresh every 10 minutes, during which a short wash was performed to remove dissociated phage); panning 4 was performed in PBS for 2 hours at 37 ° C; Panning 5 was performed in PBST for 1 hour at room temperature, then in PBS for 2 hours at 37 ° C; Panning 6 was performed in a large volume (20 ml) of PBS at room temperature Panning 7 further increased washing temperature (30 ° C.), volume (50 ml), and time (24 hours). As shown in FIG. 29, in addition to changing the washing stringency described above, the antigen concentration and the phage input concentration were decreased, the temperature of the binding time was increased, and dissociation was further enhanced. Clones that survived panning were randomly picked and measured by phage ELISA to confirm their ability to bind to VEGF. Both panning 5 and 7 gave 100% ELISA positive rates from the clones, suggesting that all surviving phage could bind to VEGF after panning 5 and thus washed out phage had a faster off-rate. ing. Of the clones positive by phage ELISA, 20 clones from panning 5 and 10 clones from panning 7 were randomly picked for DNA sequencing. The encoded amino acid sequence of V _H CDR3 is summarized in FIG. The frequency of wild-type and average antibodies was 20% with panning 5. After two additional rounds of off-rate panning at high stringency, the frequency of wild-type sequences decreased to zero at panning 7. In contrast, HR (H97, R101 or R100a in Kabat) mutants were enriched from 35% in panning 5 to 70% in panning 7 (Figure 30), which ultimately became the only dominant clone It was. The presence (30%) of the HT (H97, T101 or Kabat T100a) mutant did not change between panning 5 and 7. The enrichment of HR variants from P0 to P7 is shown in FIG. These data suggest that both HR and HT variants have higher affinity than wild type antibodies. The affinity of the HR mutant was reported for the HT mutant (this was reported for mature sequences (Chen Y, Wiesmann C, Fuh G, Li B, Christinger HW, McKay P, de Vos AM, Lowrnan HB (1999) J. Mol Biol 293, 865-881) should be higher than the affinity at position 101 (or threonine instead of arginine at 100a in Kabat).

8) Panning of single chain (scFv) anti-VEGF antibody library with adapter-mediated phage display system An off-rate panning strategy was further tested using the independent system described below.

  In a conventional phage display system, a target protein is fused to a phage capsid protein (eg, pIII) and displayed on the surface of the phage. The fusion protein was assembled into a phage particle, and the wild type phage protein was provided by a helper phage such as KO7. We have developed a new phage display system called the “adapter command display system”. Generally, the protein of interest is specifically fused to a pair of adapters that form heterodimers, one fused to the indicated protein in the expression vector and the other to the phage capsid protein in the helper vector. ) To be transported to the surface of the phage particle. The pair of adapters in this example is the above-described GR1 and GR2. As described in FIG. 32, the protein of interest (scFv anti-VEGF) is expressed as a fusion with the adapter (GR1) to form a scFv-GR1 construct in the expression vector (FIGS. 33A and B). GR2 was inserted into the helper phage genome to form a fusion with the pIII capsid protein (GRIII-CT of pIII, FIGS. 33A and B). As a result, helper phage with a modified genome is then named GMCT ultrahelper phage (FIGS. 34A and B). In TG1 cells, the expression vector expresses scFv-GR1, which is then secreted into the bacterial periplasmic space. Cells are further infected with GMCT ultrahelper phage, which expresses pIII GR2-CT and is also secreted into the periplasmic space of bacteria. Thus, scFv-GR1 and pIII GR2-CT specifically form heterodimers due to the coiled-coil interaction of GR1 and GR2, which ultimately assembles the scFv on the surface of the phage.

Using this system, we constructed an anti-VEGF scFv library L17 (equivalent to the above ccFv library L14) (anti-VEGF CDR3 V _H synthesis library). Off-rate panning was applied as in library L14 selection. Library DNA was transformed into TG1 cells and then rescued with GMCT ultrahelper phage. Phages were prepared according to standard protocols and tested for binding to immobilized VEGF in 96 well plates. As shown in FIG. 35A, wells from pannings 1 and 2 were first washed 10 times with PBST, then 10 times with PBS at room temperature, and then dissociated in PBST for 1 hour at room temperature (PBST 10 Every minute, it was replaced with a fresh one, during which a short wash was performed to remove the dissociated phage); in Panning 3, the dissociation time was increased to 2 hours. Using phage recovered from panning 3, two parallel pannings (Figure 35B) (panning 4 and panning 5) were performed to further promote dissociation of low affinity phage: 18 in 150 ml PBST for panning 4 Time, Panning 5 is 37 ° C. Ten ELISA-positive clones from Panning 4 and 5 to 8 clones were randomly picked for sequencing. The data is shown in FIG. In panning 4, the presence of WT sequences was 10%. The frequency of HT mutant (30%) and HA mutant (30%) was comparable. Of the 10 clones analyzed, no arginine residue was demonstrated at position 101 (100a Kabat) (Figure 36), suggesting a low incidence at this stage. In contrast, panning 5 increases dissociation stringency, so that the presence of arginine at position 101 (100a Kabat) rises to 50% (4 out of 8 clones) and panning 5 becomes dominant. By comparison, the HT variant decreases from 30% to 12.5%, and the WT decreases from 10% to 0, which is consistent with the observations in FIG. This result strongly suggests that the HR mutant has a higher affinity than the HT mutant or WT.

9) Summary of library design, diversity and affinity maturation The results shown in both FIGS. 30 and 36 show that the off-rate panning of the two independent novel phage display systems used here is the novel mutant HR (H97 This suggests that R100a) can be selected with R101 or Kabat. The HR mutant has a higher binding affinity than the corresponding HT (H97, T101 or Kabat T100a) mutant in the reported mature sequence (FIG. 9B). Furthermore, the HR mutant binds to the antigen better than the YS (Y97, S101 or Kabat S100a) mutant (see Panning 4 in FIG. 36). The YS variant was previously reported to improve binding affinity 14-fold to WT and was considered the single most important variant of the mature anti-VEGF antibody V _H CDR3 (Chen Y, Wiesmann C , Fuh G, Li B, Christinger HW, McKay P, de Vos AM, Lowrnan HB (1999) J. Mol Biol 293, 865-881). This mutant H97Y was also found to be important in the design library by database search (Figure 11) and computer screening (Figure 13A).

K94 is an interesting example and deserves some consideration. Strictly speaking, K94 does not belong to V _H CDR3 by Kabat nomenclature. However, the N-terminal sequence CAK of V _H CDR3 is included in the creation of the HMM motif because this sequence places strong constraints on the boundaries of the sequence motif. Since CAK is the border region between the framework and V _H CDR3, we consider testing the effect of mutations in this region on binding affinity. Although R94 was found to be suitable for both database searching and computer screening (Figures 11 and 13A), K94 binds more strongly than R94 in experimental screening (Figures 30 and 36). Including both K94 and R94 in the library selected only K94 (Figures 28B, 30 and 36), but R94 is still active in binding to VEGF (Figures 13A and 14B). This is because R94 in the joint region changes the orientation of V _H CDR3 in binding to the antigen by interacting with other regions of the antibody, and thus the original K94 X-ray structure used for computer screening ( This may be due to the expiration of the mature antibody. R94 has been reported to reduce the binding affinity of anti-VEGF antibodies almost 5-fold during humanization (Presta LG, O'Connor SJ, Wells JA (1997) J Biol Chem 272, 10678-10684). Several approaches can be used to avoid this problem: (1) If designing only CDRs, avoid designing border residues; (2) both parent and preferred residues (eg, 94 Combine both K and R) in the experimental library. Since R and K are the two major residues preferred at this position from a database search (approximately 90% for R94 and approximately 10% for K94), they are reasonable and straightforward in this case (see FIG. 11) ); (3) The conformation of R94 is tested at this position with a computer by molecular dynamic simulation to see if a modified structure or structure aggregate is used with R94.

In summary, it has been found that three important sites around the V _H CDR3 region of anti-VEGF antibodies have a direct effect on the binding affinity of VEGF antibodies. Two of the mutations at three positions (K94, H97 and S101) (Y97 and R101 or R100aKabat) are improved with the antigen using the parental or mature antibody structure in the presence and / or absence of the antigen. R94 could not be predicted correctly due to the possibility of structural changes induced by mutations in the joint region. Y97 has been found to be an important mutation for affinity improvement, as demonstrated by our experimental screening. R101 (R100aKabat) is a novel mutant identified by two independent phage display systems and will probably confer higher affinity than Y97.

Most of these mutants, including R94, Y97 and R101 belong to the dominant variant in the hit variant profile (see Figure 11) (> 5%). Thus, a simple sequence search will find these from the hit variant library. In the structure-based screening of variant libraries, these variants are also ranked higher in the selected sequence profile as shown in FIG. 13A. From the standpoint of aggregate sequence scoring, pooling and reprofiling sequences with higher scores than the parental sequence also resulted in 94 (88% R, 12% K), 97 (60% Y, 17% H) ), And 101 (60% R, 17% T, 13% S). Excluding problems associated with R94, the statistical selectivity for Y97 and R101 or T101 is evident in our design. We have demonstrated our library design using sequence search and / or structure-based scoring to create variant profiles. Experimental screening or selection using two independent novel phage display systems has demonstrated the utility of the methods of the invention described herein for designing sequences that differ from the parent sequence in _VH . Some of the mutants found here (eg Y97 and / or R101 or T101) have an affinity that is at least 10-fold higher than the parent sequence (Y97 accounts for a 14-fold improvement in affinity, while R101 is our experiment (See Figure 36) Extrapolated, mutant combinations (eg Y97 and R101) may have higher affinity than reported for mature sequences There is.

The binding affinity of affinity matured V _H CDR3 was measured using VEGF immobilized on a biosensor chip as shown in FIG. 37, using an SPR (surface plasma resonance) apparatus (BIAcore). The protein was expressed and purified. X50 is in the ccFv format and contains the V _H and V _L reference sequences shown in FIGS. 22A and 22B. X63 has H97Y and S101T in V _H CDR3 and Kd improved by 6.3 times compared to 14-fold improvement of the Fab format reported in literature (Chen Y, Wiesmann C, Fuh G, Li B, Christinger HW, McKay P, de Vos AM, Lowrnan HB (1999) J. Mol Biol 293, 865-881, see Table 6). X64 contains S101R in V _H CDR3 and is improved by a factor of 2.5 over the reference; almost all this improvement is due to increased on-rate. Although the importance of this new variant in improving the reverse labeling method has not been reported, comprehensive mutagenesis was performed at this position. Also, the frequency in the database at this position is low. This indicates that the approach taken here can find important variants for affinity improvement. X65 contains H97Y and S101R and shows a 10-fold improvement using the ccFv format under the same conditions, which is an affinity matured V _H CDR3 sequence (Chen Y, Wiesmann C, Fuh G, Li B, Christinger HW, McKay P, de Vos AM, Lowrnan HB (1999) J. Mol Biol 293, 865-881) has a stronger binding affinity than the best mutant combination of X63 (H97Y and S101T).

Example 2
Creation of anti-VEGF antibody library for framework optimization
VEGF is a major angiogenic factor in development and is involved in solid tumor growth by stimulating endothelial cells. Mouse monoclonal antibodies were found to block VEGF-dependent cell growth and slow tumor growth in vivo (Kim KJ, Li B, Winer J, Armanini M, Gillett N, Phillips HS, Ferrara N (1993) Nature 362 , 841-844). This murine antibody was humanized using random mutagenesis at several key framework positions after transplanting the antigen binding loop (Presta LG, Chen H, O'Connor SJ, Chisholm V, Meng YG. Krummen L, Winkler M, Ferrara N (1997) Cancer Res. 57, 4593-4599; Baca M. Presta LG, O'Connor SJ, Wells JA (1997) J Biol Chem 272, 10678-10684). Typically, after site-directed mutagenesis and selection, created by replacing human or consensus human frameworks with non-human amino acids from the parent non-human antibody at several pre-determined key positions Is done. These humanized antibodies usually bind to the cognate antigen of the parent antibody with a weaker affinity than the parent antibody (for humanized anti-VEGF, about 6 times weaker than its parent mouse antibody, Baca M, Presta LG, O ' Connor SJ, Wells JA (1997) J Biol Chem 272, 10678-10684, and other versions of humanized anti-VEGF are two times weaker, Presta LG, Chen H, O'Connor SJ, Chisholm V, Meng YG, Krummen L. Winkler M, Ferrara N (1997) Cancer Res. 57, 4593-4599; Baca M, Presta LG, O'Connor SJ. Wells JA (1997) J Biol Chem 272, 10678-10684). This loss of binding affinity may be restored by using CDR affinity maturation (Chen Y. Wiesmann C, Fuh G, Li B, Christinger HW, McKay P, de Vos AM, Lowman HB ( 1999) J. Mol Biol 293, 865-881).

  Using the method of the present invention described, we have reviewed the literature (Presta LG, Chen H, O'Connor SJ, Chisholm V. Meng YG, Krummen L, Winkler M, Ferrara N (1997) Cancer. Res. 57, 4593-4599) discovered two humanized frameworks with a binding affinity (in ccFv format) 4 times higher than the parent / reference anti-VEGF antibody framework. Since the reported humanized anti-VEGF antibodies (FIGS. 22A and B) are almost 2-fold weaker than their corresponding mouse antibodies, these two humanized antibodies bind almost 2-fold higher than the corresponding mouse antibodies upon humanization. Should have affinity.

1. In silico design of anti-VEGF antibody framework library The top panel of Figure 38A shows mouse anti-VEGF antibody (where it is called "mouse anti-VEGF antibody or a4.6.1"), humanized antibody (selected from the library) been HU2.0 and HU2.10, and amino acids that were used for humanization a major position for both V _H and _{V L (Baca M, Presta LG} , O'Connor SJ, Wells JA (1997) J Biol Chem 272, 10678-10684) shows the amino acid sequence of the framework fr123, which is based on the Kabat standard (Kabat EA, RediMiller M, Perry HM, Gottesman KS (1987) Named according to 4th edition, National Institutes of Health, Bethesda, MD, but other classifications can also be used.The bottom panel of Figure 38A shows mouse anti-VEGF antibodies (where "mouse Called anti-VEGF antibody) and here As a reference framework and reported in literature (Presta LG, Chen H, O'Counor SJ, Chisholm V, Meng YG, Krummen L, Winkler M, Ferrara N (1997) Cancer Res. 57, 4593-4599) The amino acid sequence of the framework fr123 region of a humanized antibody (referred to herein as a “humanized anti-VEGF antibody”) is not designed because framework 4 is constant, but if desired, the same approach In addition, separate segments of FR1 or FR2 or FR3 and FR4 can be individually designed and pasted together if desired. By designing each segment or combination of segments using the approach described, the positions of CDR1 and CDR2 are indicated using arrows, but not shown in the figure. , Mau Figure 38B shows a hit library generated using human V _H germline sequences based on the V _H FR123 lead sequence of the mouse anti-VEGF antibody. The variant profile is shown. The lower variant profile shows the diversity of amino acid positions. The lower part of the figure shows the filtered variant profile obtained using cut-off frequencies 5 and 13, respectively. Filter all position amino acids of the hit list members that appear 5 times or less, or 13 times or less. FIG. 38B (continued) shows a reprofiled variant profile of a hit library generated using human V _H germline sequences based on the V _H FR123 lead sequence of the murine anti-VEGF antibody without cut-off, Each position variant is ranked based on its structural compatibility with the antibody structure using total energy or van der Waals energy. Some reference amino acids have been found to be suitable at several positions based on their total energy or specific packing, but their frequency of occurrence is very low (eg, F68 (F67 marked with an arrow). ), L72 (L7l), S77 (S76) and K98 (K94)). For example, F68 and L72 are included in the library for selection. FIG. 38C shows the variant profile of a hit library generated using the Kabat-derived human _VH sequence based on the _VH FR123 lead sequence of the mouse anti-VEGF antibody with the variant profile filtered at cutoff 19. This profile highlights the importance of several amino acids that appear infrequently but are important in scaffold formation. The mouse V _H FR123 sequence is listed as a reference above the dotted line at the positions indicated using sequential numbers. All amino acid variants are listed under the dotted line. The point in the variant is the same amino acid as the reference. FIG. 38D shows a designer library using a filtered variant profile from a human _VH germline sequence with cutoff 5 (see FIG. 38B). The SEQ ID NO above the FR123 sequence is based on the Kabat nomenclature (kabataa) and the sequential order including the amino acids in its CDRs. This filtered variant profile can be further screened on a computer to reflect the rank order of structural suitability when only antibody structures are used. The two amino acids, F70 (F69) and L72 (L71), missing from the variant profile filtered at cutoff 5, are the most preferred amino acid companions at these positions based on structure-based screening, These are also included. The final submitted library of the top 100 sequences from structure-based screening is also F70 (F69), L72 (L71), S77 (S76), and K98 (K94) (numbers in parentheses are Kabat) This indicates that some amino acids such as R are over-calculated for L72 (L71) and K98 (K94) as described above for K94R of VH CDR3 affinity maturation. This is because it is predicted.

  The bottom panel of FIG. 38D shows a designer library with amino acids used for humanization of VH fr123. As shown in FIG. 38D, human versus non-human sequences differ in many positions throughout the entire VH chain, but the amino acid libraries used in other approaches are enriched in several key positions, while the present invention Targeting various positions through both the VH and VL chains, several variants of these positions are based on a designer library of starting antibodies.

In the present invention, each motif, such as frameworks FR1, FR2, FR3, and FR4, shown in FIG. Can be targeted. It is understood that there is only a limited number of conformations (called standard structures) or combinations thereof for each motif. These structural features of antibodies provide an excellent system for testing evolutionary sequence design using structured motifs in various regions of antibodies based on extensive analysis of antibody structure. Their structure and sequence conservation has been observed across different species. In fact, antibody scaffolds or immunoglobulin folds are one of the most abundant structures observed in nature and are highly conserved among various antibodies and related molecules.
The inventors believe that the parent anti-VEGF antibody described above can function as a lead protein in a model system for command antibody optimization for therapeutic and other applications using the methods of the invention. Humanized anti-VEGF antibodies (Baca et al., Supra; Presta et al., Supra) can serve as a reference or positive control to evaluate the results obtained using the methods of the invention.

  Furthermore, the structural overlap revealed that the structure of the complex formed between VEGF and the parent antibody almost overlaps with the complex formed between VEGF and the mature antibody. Since the antibody structures, particularly the framework regions, are substantially the same, the parent and mature antibody structures were used to design a digital library of anti-VEGF antibodies using the method of the present invention. The methods of the invention can also be used to design antibody frameworks using sequence-based approaches or structural assemblies containing induced structural changes.

Using the mouse anti-VEGF antibody framework as the lead protein and its V _H FR123 as the lead sequence, construct a digital library of V _H FR123 as route IV in Figure 1D and according to the scheme outlined in Figure 2 did.

In summary, a hit library was constructed by searching for and selecting hit amino acid sequences that have distant homology with V _H CDR3. A variant profile was created to describe all variants at each position based on the hit library and was filtered with a certain cutoff value to reduce the size of the resulting hit variant library to within computer or experimental limits. Variant profiles were also constructed to facilitate: i) examination of sequence space covering suitable regions in the matching background; ii) partitioning and synthesis of degenerate nucleic acid libraries targeting suitable peptide assembly sequences Iii) screening the antibody library by experiment for the desired function; and iv) analysis of experimental results with feedback for further design and optimization.

  From the available X-ray structure of the complex formed by VEGF and anti-VEGF antibody, a lead structure template was obtained. The complex structure of VEGF and parent anti-VEGF antibody is called 1BJ1, and the complex structure of VEGF and mature anti-VEGF antibody is called 1CZ8. The results from the 1CZ8 structural template were similar to those from 1BJ1 in the relative rank of the scanned sequence. Modeled structures or structure aggregates or set averages can also be used to screen sequences.

1) Lead arrangement
The V _H FR123 lead sequence is taken from a mouse anti-VEGF antibody according to the Kabat classification (FIG. 38B).

2) Hit libraries and variant profiles HMMs created using a single read sequence A4.6.1 (Figure 38A) were calibrated and Kabat database (Johnson, G and Wu, TT (2001) Nucleic Acids Research, 29, 205-206) was used to search the human heavy chain germline sequence database and / or human sequence database (including human germline and humanized sequences). All sequence hits that are larger than the predicted or E value are listed and aligned using the HAMMER 2.2.1 package. After removing duplicate sequences from the hit list, the remaining hit sequences for the lead HMM form a hit library.

The sequence identity of hit sequences from human VH germline ranges from 40 to 68% of the lead sequence, and the Kabat database (this database is a fr123 fragment to increase search sensitivity and its relative ranking The corresponding sequence identity of the hit sequence from the human immunoglobulin sequence obtained from (which contains immunoglobulin sequences of human origin, other databases may be used) is about 30-75 % Range. The evolutionary distance between hits can be analyzed using the program TreeView 1.6.5 ( http://taxonomy.zoology.gla.ac.uk/rod/rod.html ). The phylogenetic tree is obtained from the adjoining method (Saitou N, Nei M (1987) Mol Biol Evol 4, 406-425 in ClustalW 1.81 (Thompson JD, Higgins DG, Gibson TJ (1994) Nucleic Acids Research 22, 4673-4680). ).

  The AA-PVP table in FIGS. 38B and D gives the number of occurrences of each amino acid residue at each position. The variant profile below the table list lists all variants found from the database using the lead sequence as the reference sequence in order of decreasing occurrence at each position. A comparison of hit sequence identity differences between human VH germline and Kabat-derived human VH sequences reveals the difference in AA-PVP: for AA-PVP from the human germline sequence All variants are of human origin, but AA-PVP also stabilizes scaffolds such as non-human origin or low-frequency amino acids that may come from the starting non-human antibody sequence, or target antibodies during evolution It contains structurally important amino acids. For example, F70 and L72 in FIG. 42B are not identified by AA-PVP from the VH3 germline phage component (see FIG. 42, human VH3 germline only accepts I and R at these two positions. ). However, F75 and L77 are tolerated in human VH germline sequences with very low frequency appearance. These amino acids F70 and L72 appear relatively frequently in AA-PVP from the Kabat-derived human sequence. All amino acid variants are listed below the dotted line. Points in the variant indicate the same amino acid in the reference. FIG. 38D shows a designer library that uses a filtered variant profile from human VH germline sequences in core 5 (see FIG. 38B). The SEQ ID NO above the FR123 sequence is based on the Kabat nomenclature (kabataa) and the sequential order including the amino acids in its CDRs. This filtered variant profile can be further screened in a computer to reflect the rank order of structural compatibility when only antibody structures are used. Two amino acids, F70 (F69) and L72 (L71), missing from the variant profile filtered at cutoff 5, are the most preferred amino acid companions at these positions based on structure-based scoring These are also included. The final submitted library of the top 100 sequences from structure-based screening is also F70 (F69), L72 (L71), S77 (S76), and K98 (K94) (numbers in parentheses are Kabat) This indicates that some amino acids such as R are over-calculated for L72 (L71) and K98 (K94) as described above for K94R of VH CDR3 affinity maturation. This is because it is predicted.

  FIG. 42 also shows that both F and I can be identified at this position by panning and only the dominant L72 can be identified at this position. Briefly, although the use of different databases of human origin for framework optimization is diverse in amino acids for framework optimization, including humanization with improved binding affinity and stability Provide powerful choices. Due to our increased knowledge in the development of therapeutic antibodies, more antibody sequence data will be accumulated and will help our design using the present invention. No assumptions are required in advance to assume the main positions and the amino acids associated with these positions. Since this information is automatically revealed using the method of the present invention, as more data is accumulated, it will be well defined due to its increasing appearance in the database. Using structure-based criteria, variants can be reprofiled and prioritized to include other potentially useful variants (see FIG. 38B—continuation).

3) Structure-based evaluation of combinatorial sequences of hit libraries Variant profiles are specific variants at each position and in a suitable unmodified order, beneficial on preferred amino acid residues, Embody a huge number of recombinants. Scoring indicates that F70 and L72 are suitable for structure-based scoring and should be maintained in the profile, but their frequency of occurrence is lower than the cut-off used for profiles derived from database searches (FIG. 38B—continued). That is, structure-based energy scoring is a separate approach to reprofile variants at each position for hit variant libraries originally created based on the evolutionary sequence profiling selected from protein databases. Provide a way. Certain filtering using a frequency cut-off can reduce combinatorial sequences that need to be assessed by computer screening or directly targeted by an experimental library. There are many combinatorial sequences that need to be scored and evaluated on the final sequence for experimental screening (as shown in the bottom panel of FIG. 38D), even with the cutoff applied to the variant profile.

Structure-based scoring is applied to screen the combinatorial sequences that form the hit library and the hit variant library. The side chain of V _H FR123 of the anti-VEGF antibody in 1CZ8 or 1BJ1 was replaced by the corresponding amino acid variant rotamer from the variant library at each residue position. Rotamer conformation was created and optimized using the program SCWRL® (version 2.1) using a backbone-dependent rotamer library (Bower MJ, Cohen FE, Dunbrack RL (1997) JMB 267, 1268 -82).
Scoring uses the Amber94 force field of CONGEN [Bruccoleri and Karplus (1987) Biopolymers 26: 137-168] in the presence and absence of the structure of the antigen VEGF to find the optimal rotamer and minimize energy by 100 steps This was done by

  FIG. 39A shows X axis 1 using 1bj1 (upper panel) and 1cz8 (lower panel) as template structures in the absence (leftmost column) and presence (middle column) of VEGF antigen. Distribution of scoring schematics for mouse anti-VEGF VH framework fr123 hit sequences using human VH germline sequences in relatively dense blue strips in rows, and relatively sparse in column 0 on the X axis The distribution of scoring schematics for the mouse and humanized framework fr123 (Presta et al., Supra) sequences and the widely used human VH germline DP47 are shown in the blue strips. Sequence scores in the presence and absence of antigen are correlated (rightmost column), and the antibody structure for framework optimization has minimal contact with the antigen, so most of the framework optimization Indicates that it is sufficient. A scoring diagram of a combinatorial sequence library is not shown here.

  FIG. 39B shows differences between sequences in the library and reference mouse VH FR123 sequences, and reference, mouse VH FR123, reported (Presta et al., Supra, 1997, and Chen et al., Supra, 1999) humanized VH FR123, The left panel shows rank scoring based on the phylogenetic distance of the X axis for the top 200 designer sequences and the human CDR H3 germline (a widely used VH human germline called DP47) . The top 200 sequences from the structure-based screening of one variant of human germline (AA-PVP) cluster with the human VH3 germline family in a phylogenetic analysis (red cycle), while leading The murine antibody framework is genetically separated from the designed one by a phylogenetic distance (when only high-frequency germline VH sequences are included, and humanized sequences from 1bj1 (Presta et al., Supra) ) Will vary slightly by including relatively low-occurrence amino acids such as F70 (F69) and K98 (K94) (see FIGS. 42C and D.) The y-axis is Work VH fr123 has good structural compatibility with the mouse, reference and humanized framework VH fr123 (close to DP47), as these are partly defined by the database used Support the human-like features of framework optimization of the method of the invention described herein.

4) Reduced variant profile of hit variant library Hit variant library after exclusion of amino acids by screening sequences based on suitability with structural scaffolding and / or appearance frequency lower than cut-off value In order to reduce the size of the library candidates while maintaining most of the preferred residues shown in FIG. 38B, the variant profile from the hit variant library was filtered as described above. For example, some important variants (eg F70 and L72) are included in the variant profile from the wild type, but below the cut-off value used in the hit library filtering. These are evaluated using structure-based profiling and persist through multiple rounds of panning under severe wash conditions in phage display (Figure 42). The top 100 sequences from structure-based scoring were used with F70 and L72 from structure-based profiling of the original profile.

5) Construction of degenerate nucleic acid library based on hit variant library II The hit variant library constructed above was targeted using the degenerate oligonucleotide shown in FIG. 40A. The degenerate nucleic acid library constructed above was cloned into a phage display system, and a phage-displayed antibody (ccFv) was selected based on the binding to immobilized VEGF coated on a 96-well plate. The final designed humanized sequence of VH anti-VEGF is shown in FIG. 40A. About 120 amino acid residues of VH of anti-VEGF, 34 amino acids were changed as a result of computer design: 18 of them were fixed (bold and underlined) and 16 using the described ccFv system Placed as a result of measurement by phage display library screening (labeled "X"). Therefore, a degeneracy of the DNA sequence corresponding to the 16 positions was created to create multiple choices of suitable amino acid residues during screening. The theoretical density of the library is about 2.6 × 10 ⁵ . The library was placed in the phage display vector pABMD12 where the anti-VEGF VH was replaced by the library. As a result, VL and various VHs displaced by the library will pair to form a functional ccFv of anti-VEGF. The phage display library was then used for further panning against the immobilized VEGF protein antigen.

  In order to create a library that covers a wide distribution of degenerate positions, multiple overlapping degenerate DNA oligos were synthesized using degenerate positions at the site where the library was designed. The assembly process consisted of two PCR reactions, assembly PCR, and amplification PCR. The assembly oligo was designed as a 35-40mer, with 15-20 bases overlapping at an average melting temperature of about 60 ° C. A pair of additional amplification oligo primers (Amp93 and Amp94) were created for final amplification of the designed product. Thus, assembly PCR includes: equal amounts of assembly oligo primers, final total concentration of 8 μM, 0.8 μM dNTP, 1 × pfu buffer (Stratagene), and 2.5 units of pfu turbo (Stratagene ( Stratagene)). Thermal cycling was performed as follows: 30 cycles of 94 ° C x 45 ", 58 ° C x 45", 72 ° C x 45 ", and final extension for 10 minutes at 72 ° C. PCR product diluted 10-fold and amplified Used as a template for PCR, where all reagents were the same except that amplification primers were added at a final concentration of 1 μM. Thermal cycling was performed as follows: 94 ° C x 45 ", 60 ° C x 45", 30 cycles of 72 ° C x 45 "and final extension at 72 ° C for 20 minutes. The final product (VH library) was purified and digested with HIndIII and StyI (FIG. 26) and finally subcloned into the vector pABMD12 to replace the original mouse VH. Using this library, TG1 cells were electrically transformed (electroporation?), Which was then amplified and rescued with helper phage KO7 (Amersham), followed by standard methods. Library phage were produced overnight at 0 ° C.

6) Panning anti-VEGF humanized V _H phage display library To screen the constructed library described in the above example, purified homodimeric VEGF protein (Calbiochem) was coated with coating buffer (0.05M NaHCO 3). ₃ and diluted at the specified concentration in pH 9.6) and immobilized in Maxisorb wells (Nunc) at 4 ° C. overnight. The coated wells were then blocked in 5% milk at 37 ° C. for 1 hour, and then a phage library diluted in PBS was applied to the wells and incubated at 37 ° C. for 2 hours. The incubation mixture also routinely contained 2% milk to minimize nonspecific binding. At the end of the incubation, the wells were washed and then bound phage was eluted with 1.4% triethylamine before infecting TG1 cells and then rescued with KO7 helper phage for amplification. To amplify the phage, the infected and rescued TG1 cells were then grown overnight at 30 ° C. in the presence of carbenicillin and kanamycin, and then the phage library was collected. Amplified phage was used as a supply library for the next round of panning. The panning method is summarized in FIG. On the other hand, individual clones from the 5th and subsequent panning were randomly sampled for phage ELISA to confirm specific binding to immobilized VEGF and showed 100% positive from the 5th to 7th panning. Finally, isolated clones grown on 2 × YT / carbenicillin (100 μg / ml) / kanamycin (70 μg / ml) plates were sampled for sequencing starting with the fifth round of panning (P5). The hit position and hit sequence for the design were defined.

  A summary of the sequence analysis of hits from the above library panning is shown in Figure 42A, with the dominant residues of the human germline family III VH and hits from the library panning at the location where the library was designed. Amino acid residues were compared. Out of the 16 positions designed for measurement by phage display library screening, as described, serial numbers 1, 11, 17, 24, 70, 72, 74, 77, 78, 79, 98 Certain amino acid residues were maintained or predominated from P5 (5th panning) to the last (8th) panning, while the remaining positions showed some variation in the dominant residues. The final selection of residues at 9 of the last 16 positions (shaded in FIG. 42B) is largely consistent with residues at equivalent positions in the human immunoglobulin VH family III, , Indicating that the selected species is likely to fall into Family III.

  FIG. 42C is a phylogenetic view of the top hit VH sequences from panning of an anti-VEGF phage display library, with the described human germline VH3 family, mouse anti-VEGF VH framework FR123, and humanized VH framework fr123. Show the analysis. As shown in FIG. 42C, the human germline buoy HSP family forms clusters at phylogenetic distances as expected. The selected optimized VH framework also forms a cluster with the humanized VH sequence (see note), and the phylogenetic distance is very close to the human germline VH3 family, while the mouse VH framework may be Very far from the VH framework and the human germline. Phylogenetic analysis of hit sequences against the human immunoglobulin repertoire of VH suggests that these are most closely related to family III. Phylogenetic analysis also shows that the final hit sequence is much more closely related to human immunoglobulin family III compared to sequences from mouse origin of anti-VEGF (Y. Chen et al., 1999). In summary, the results indicated that amino acid residues of human origin were successfully measured for the majority of the 34 positions.

  In addition, 5 positions (ie, positions 6, 72, 77, 79, 98 in sequential numbers) (FIG. 42B) do not end with suitable human residues after selection, but positions 70 and 74 in sequential numbers (FIG. 42B). Was able to pick up a minority group of residues of human origin. Although few, these populations have survived a series of rigorous washings and multiple pannings, demonstrating that they actually have a high affinity for antigen. These positions did not select dominant residues of human origin. On the other hand, the presence of a minority population of human-origin residues (positions 70 and 74 in sequential numbers (FIG. 42B)) suggests that these positions may be humanized.

  This is a framework that is optimized using human or human-like sequences of optimized antibodies, depending on the delicate balance between human-like and the structural template from the aggregate structure or structure average or compatibility with the template. The method of the present invention in the design of FIG. 42B shows the phylogenetic distance of these sequences in another annotated phylogenetic view with some fully characterized sequences D36, D40 and D42 and related sequences. D36 is human or hardly as good as the humanized sequence reported at phylogenetic distance.

  The full-length sequence of the top hits from anti-VEGF VH library panning (top hits from the last two pannings, 7th and 8th pannings), mouse anti-VEGF VH (Y. Chen et al., 1999) And the human III immunoglobulin VH family III dominant sequence is described in FIG. 42A.

7) Selection of humanized VH for anti-VEGF with high affinity As summarized in Figure 41, to increase the stringency of washing without selecting high-affinity binders, increase washing time, wash volume Methods such as increase, decrease in coated VEGF concentration, decrease in input library phage were performed. All these methods promote the dissociation of relatively low affinity interactions so that high affinity interactions remain selectively. The phage clones that survive this panning are then sampled and sequenced. The full length of the top hit anti-VEGF VH sequence from this panning is described in FIG. 42A. Using our described method of the present invention, we have three (D36, D36, D) that have higher binding affinity in the ccFv format than the parent or reference anti-VEGF antibody by framework optimization (see FIGS. D40 and D42) discovered humanized frameworks (see Figures 22A and B for humanized anti-VEGF antibodies (Presta LG, Chen H, O'Counor SJ, Chisholm V, Meng YG, Krummen L, Winkler M, Ferrara N (1997) Cancer Res. 57, 4593-4599) These improvements mainly come from a large increase in on-rate and a small decrease in off-rate due to framework humanization alone, Figure 43A shows the ccFv phage display system ( (See description of FIGS. 23-25 above) shows the sequence of the optimized VH framework (FR123) of anti-VEGF antibodies selected from the designer VH optimized library: D36, D40 and D42 VH fr123 Is the same as the original mouse antibody VH FR123. Humanized sequences with CDR (Presta et al., Supra) along with. The lower panel point, showing the same amino acid as the reference (murine VH framework FR123).

  Figure 43B shows affinity data for five antibody, parent antibody (X50) and anti-VEGF antibody optimization frameworks (D36, D40, D41 and D42) selected from a designer library using a BIAcore biosensor. Shown (see notes in Figure 43A and Figure 43B for its sequence). The measurement is performed by measuring the change in SPR units (y axis) versus time (x axis) when the purified antibody binds to an antigen (VEGF) immobilized on a CM5 biochip at 25 ° C. Both on-rate and off-rate were measured by data fitting using a 1: 1 Langmuir binding model. The two humanized frameworks D36 and D40 have almost 4-fold higher binding affinity (in ccFv format) than the parent / reference anti-VEGF antibody sequence due to framework optimization (Figures 22A and B for humanized anti-VEGF antibodies). Reference (Presta LG, Chen H, O'Counor SJ, Chisholm V, Meng YG, Krummen L, Winkler M, Ferrara N (1997) Cancer Res. 57, 4593-4599), whereas D42 is almost the same as the reference antibody Since the reported humanized anti-VEGF antibodies (FIGS. 22A and B) are about 2-fold weaker than their compared mouse antibodies, these two humanized antibodies are about 2-fold more than the corresponding mouse antibodies upon humanization. It should have a high binding affinity.

  Figure 44 shows the increased stability of the optimized VH framework (days 36 and D40). The y-axis shows the binding of purified antibody to parental X50 and optimized frameworks (D36 and D40) at 4, 37 and 42 ° C for 17 hours, followed by binding to immobilized VEGF antigen at 25 ° C using BIAcore. The percentage of antibody maintained in activity is shown. This indicates that the optimization framework is more stable than the reported humanized VH framework (Presta et al., Supra, 1997).

  FIG. 45 shows improved expression of the optimized VH framework. The optimized framework (D36, D40 and D42) also improved expression relative to the parent / wild type antibody (X50) as shown by the expression yield detected in SDS-PAGE / Coomassie blue staining Indicates.

  Note that antibody libraries designed using the methods of the present invention are expressed and screened not only in bacteriophage systems but also in other organisms, including but not limited to yeast, insects, plants and mammalian cells. I want to be. Engineered antibodies (including antigen-binding fragments and other antibody types) are produced by various recombinant DNA techniques or other techniques. For example, a DNA segment encoding the designed antibody is cloned into an expression vector and known methods (which depend on the cell host type and are not particularly limited, but include calcium chloride transfection, electroporation, lipofection, and frameshift transfection. Is transferred to the host cell. Antibodies are purified by standard methods in the art, including but not limited to ammonium sulfate precipitation, affinity columns, column chromatography, gel electrophoresis, and the like. Various modifications may be made by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.

Antibodies designed using the methods of the invention can be used in various diseases [including but not limited to cancer, autoimmune diseases (eg, multiple sclerosis, rheumatoid arthritis, systemic lupus erythematosus, type 1 diabetes, and severe Myasthenia), graft-versus-host reaction disease, cardiovascular disease, frameshift infection (eg, HIV, hepatitis virus, and herpes simplex virus), bacterial infection, allergy, type II diabetes, blood disease (eg, anemia) It is used for diagnosis or treatment.
The antibodies can also be used as conjugates conjugated with diagnostic or therapeutic residues, or in combination with chemotherapeutic agents or biologics. The antibodies can also be prepared for administration via various routes of administration. For example, antibodies can be oral, topical, parenteral, intraperitoneal, intravenous, intraarterial, transdermal, sublingual, intramuscular, rectal, buccal, intranasal, inhalation, intravaginal, intraocular, topical Administration or co-administration may be administered (eg, via a catheter or stent), subcutaneously, in adipose tissue, in a joint, or subarachnoidally.

  The method of the present invention for designing a protein library in silico can be used in any form of any computational system (including but not limited to supercomputers, personal computers, personal digital assistants (PDAs), network computers, distributed computers, Or on the Internet or other microprocessor system). The methods and systems described herein can be implemented on various types of executable media other than storage elements, such as random access memory (RAM). Other types of executable media include, but are not limited to, computer readable storage media, which may be any storage device, compact disc, zip disk or floppy disk.

  The patents, patent applications and publications cited above are hereby incorporated by reference in their entirety.

Claims (156)

A method for constructing a library of antibody sequences comprising the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to a plurality of tester protein sequences; and selecting from the plurality of tester protein sequences at least two peptide segments having at least 15% sequence identity with the lead sequence (the selected peptide segment is a hit library Form).   2. The method of claim 1, wherein the length of the lead sequence is 5 to 100 aa.   2. The method of claim 1, wherein the length of the lead sequence is 6-80 aa.   2. The method of claim 1, wherein the length of the lead sequence is 8-50 aa.   2. The method of claim 1, wherein the step of identifying the amino acid sequence in the CDR is performed using Kabat criteria or Chothia criteria. The lead sequence is CDR1, CDR2, CDR3, FR1-CDR1, CDR1-FR2, FR2-CDR2, CDR2-FR3, FR3-CDR3, CDR3-FR4, FR1-CDR1-FR2, FR2-CDR2-FR3, and FR3-CDR3 2. The method of claim 1, comprising an amino acid sequence from a region within _VH or _VL of a lead antibody selected from the group consisting of -FR4.   2. The method of claim 1, wherein the lead sequence comprises at least 6 consecutive amino acid residues in the selected CDR.   2. The method of claim 1, wherein the lead sequence comprises at least 7 consecutive amino acid residues in a selected CDR.   2. The method of claim 1, wherein the lead sequence comprises all amino acid residues in the selected CDR.   2. The method of claim 1, wherein the lead sequence further comprises at least one amino acid residue immediately adjacent to the selected CDR.   2. The method of claim 1, wherein the lead sequence further comprises at least one amino acid residue in the FR that flanks the selected CDR.   2. The method of claim 1, wherein the lead sequence further comprises one or more CDRs or FRs adjacent to the C-terminus or N-terminus of the selected CDR.   The method of claim 1, wherein the plurality of tester protein sequences comprises an antibody sequence.   The method of claim 1, wherein the plurality of tester protein sequences comprises a human antibody sequence. 2. The method of claim 1, wherein the plurality of FR tester protein sequences comprises a humanized antibody sequence having at least 70% human sequence in _VH or _VL .   2. The method of claim 1, wherein the plurality of FR tester protein sequences comprises a human germline antibody sequence.   2. The method of claim 1, wherein the plurality of tester protein sequences are retrieved from a database consisting of NIH's Genebank, Swiss-Prot database, and Kabat database for antibody CDRs.   2. The method of claim 1, wherein the step of comparing the lead sequence to a plurality of tester protein sequences is performed by an algorithm selected from the group consisting of BLAST, PSI-BLAST, profile HMM, and COBLATH.   The method of claim 1, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 25%.   2. The method of claim 1, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 35%.   2. The method of claim 1, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 45%. The method of claim 1 further comprising the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed. The method of claim 1 further comprising the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into corresponding gene codons; and combinatorially combining nucleic acid position variants into a degenerate nucleic acid library of DNA segments Build up.   24. The method of claim 23, wherein the gene codon is suitable for bacterial expression. 24. The method of claim 23, wherein the genetic codon is selected such that the diversity of the degenerate nucleic acid library of the DNA segment is less than 1 × 10 ⁷ . 24. The method of claim 23, wherein the genetic codon is selected such that the diversity of the degenerate nucleic acid library of the DNA segment is less than 1 × 10 ⁶ . 24. The method of claim 23 further comprising the following steps:
Introducing a DNA segment of a degenerate nucleic acid library into a host organism cell;
Expressing a DNA segment in the host cell such that a recombinant antibody containing the amino acid sequence of the hit library encoded by the degenerate nucleic acid library is produced in the host organism cell; and
Recombinant antibodies that bind to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ are selected. 28. The method of claim 27, wherein the affinity of the selected recombinant antibody is greater than 10 < ⁸ > M- ¹ . 28. The method of claim 27, wherein the affinity of the selected recombinant antibody is greater than 10 ⁹ M ⁻¹ .   28. The method of claim 27, wherein the host organism is selected from the group consisting of bacteria, yeast, plants, insects and mammals.   28. The method of claim 27, wherein the recombinant antibody is selected from the group consisting of fully assembled antibodies, Fab fragments, Fv fragments, or single chain antibodies.   28. The method of claim 27, wherein the recombinant antibody is displayed on the surface of the phage particle. 33. The method of claim 32, wherein the recombinant antibody displayed on the surface of the phage particle may be a double-stranded heterodimer formed by _VH and _VL . Heterodimerization of V _H and V _L chains is facilitated by a heterodimer formed between two non-antibody polypeptide chain fused to the V _H and V _L chains, respectively The method of claim 33. 35. The method of claim 34, wherein the non-antibody polypeptide chain is obtained from the heterodimeric receptors GABA _B R1 (GR1) and R2 (GR2), respectively. 33. The method of claim 32, wherein the recombinant antibody displayed on the surface of the phage particle is a single chain antibody containing _VH and _VL linked by a peptide linker.   The display of single chain antibodies on the surface of the phage particle is facilitated by a heterodimer formed by a fusion of the single chain antibody and GR1 and a fusion of the phage pIII capsid protein and GR2. the method of.   28. The method of claim 27, wherein the target antigen is selected from the group consisting of small organic molecules, proteins, peptides, nucleic acids and polycarbohydrates. A method for constructing a library of antibody sequences comprising the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identify the amino acid sequence in the CDR and FR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing a first amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a CDR lead sequence);
Comparing the CDR read sequence to multiple tester protein sequences;
Selecting from the plurality of CDR tester protein sequences at least two peptide segments having at least 15% sequence identity with the CDR lead sequence (the selected peptide segments form a CDR hit library);
Select one FR in the _VH or _VL region of the lead antibody;
Providing a second amino acid sequence comprising at least three consecutive amino acid residues in the selected FR (the selected amino acid sequence is the FR lead sequence);
Comparing the FR lead sequence to a plurality of FR tester protein sequences; and selecting from the plurality of FR tester protein sequences at least two peptide segments having at least 15% sequence identity with the FR lead sequence (selected peptide segments Forms an FR hit library); and
A hit library is formed by combining the CDR hit library and the FR hit library.   40. The method of claim 39, wherein the plurality of CDR tester protein sequences comprise an amino acid sequence of a human or non-human antibody.   40. The method of claim 39, wherein the plurality of FR tester protein sequences may comprise the amino acid sequence of a human antibody. 40. The method of claim 39, wherein the plurality of FR tester protein sequences comprises a humanized antibody sequence having at least 70% human sequence in _VH or _VL .   40. The method of claim 39, wherein the plurality of FR tester protein sequences comprises a human germline antibody sequence.   40. The method of claim 39, wherein the at least one plurality of CDR tester protein sequences are different from the plurality of FR tester protein sequences.   40. The method of claim 39, wherein the plurality of CDR tester protein sequences are human or non-human antibody sequences and the plurality of FR tester protein sequences are human antibody sequences. 40. The method of claim 39, further comprising the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed. 40. The method of claim 39, further comprising the following steps:
Create an amino acid position variant profile of the CDR hit library;
Converting the amino acid position variant profile of the CDR hit library into the first nucleic acid position variant profile by back-translating the amino acid position variants into the corresponding gene codons; and combining the nucleic acid position variants in a combinatorial manner to reduce the DNA segment Build a heavy CDR nucleic acid library. A method of constructing a library antibody sequence comprising the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the FR of the lead antibody;
Select one FR in the _VH or _VL region of the lead antibody;
Providing a first amino acid sequence comprising at least three consecutive amino acid residues in the selected FR (the selected amino acid sequence is the first FR lead sequence);
Comparing the first lead FR sequence to a plurality of FR tester protein sequences; and selecting at least two peptide segments from the plurality of FR tester protein sequences having at least 15% sequence identity with the first FR lead sequence (Selected peptide segments form an FR hit library). 49. The method of claim 48, further comprising the following steps:
Providing a second amino acid sequence comprising at least three consecutive amino acid residues in a FR different from the selected FR (the selected amino acid sequence is the second FR lead sequence);
Comparing a second FR lead sequence to a plurality of FR tester protein sequences;
Select at least two peptide segments having at least 15% sequence identity with a second FR lead sequence from a plurality of FR tester protein sequences (the selected peptide segments form a second FR hit library) And combining the first FR hit library and the second FR hit library to form a hit library. The lead FR sequence was selected from the group consisting of V _H FR1, V _H FR2, V _H FR3, V _H FR4, V _L FR1, V _L FR2, V _L FR3, and V _L FR4 of the lead antibody 49. The method of claim 48, comprising at least 5 consecutive amino acid residues in the FR. 49. The method of claim 48, further comprising the following steps:
A nucleic acid or degenerate nucleic acid library containing a DNA segment encoding the amino acid sequence of the hit library is constructed.   49. The method of claim 48, wherein the plurality of FR tester protein sequences comprises an antibody sequence from which CDRs have been deleted.   49. The method of claim 48, wherein the plurality of FR tester protein sequences comprises a human antibody sequence from which CDRs have been deleted. A method for constructing a library of antibody sequences based on a lead sequence profile comprising the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody;
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Providing a three-dimensional structure of the lead sequence;
Creating a lead sequence profile based on the structure of the lead sequence;
Comparing the lead sequence profile to a plurality of tester protein sequences; and selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments are hit live Forming a rally).   55. The method of claim 54, wherein the three-dimensional structure of the lead sequence is a structure obtained from X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy, or theoretical structural modeling. 55. The method of claim 54, wherein generating the lead sequence profile comprises the following steps:
Comparing the structure of the lead sequence to the structure of multiple tester protein segments;
Determining the root mean square of the difference in the main chain conformation of the lead sequence and the tester protein segment;
A tester protein segment is selected that has a root mean square difference of less than 5 squares of the difference in backbone conformation; and the amino acid sequence of the selected tester protein segment is aligned with the lead sequence to create a lead sequence profile.   57. The method of claim 56, wherein the root mean square difference root mean square difference is less than 4%.   57. The method of claim 56, wherein the root mean square difference root-mean-square difference is less than 2%. 55. The method of claim 54, wherein generating the lead sequence profile comprises the following steps:
Comparing the structure of the lead sequence to the structure of multiple tester protein segments;
Determining the Z-score of the main chain conformation of the lead sequence and tester protein segment;
Selecting a segment of the tester protein segment with a Z score greater than 2, preferably greater than 3, more preferably greater than 4, and most preferably greater than 5; and the amino acid sequence of the selected tester protein segment as the lead sequence Align to create a lead sequence profile.   55. The method of claim 54, wherein creating the lead sequence profile is performed by an algorithm selected from the group consisting of CE, MAPS, Monte Carlo, and 3D clustering algorithms. 55. The method of claim 54, further comprising the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed. 55. The method of claim 54, further comprising the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into the corresponding trinucleotide codons; and combining nucleic acid position variants in a combinatorial manner Build a rally. A computational method for constructing a library of mutant antibodies based on a lead sequence, the method comprising the following steps:
Taking as input an amino acid sequence comprising at least three consecutive amino acid residues in the CDR region of the lead antibody (the amino acid sequence is the lead sequence);
Compare the lead sequence to multiple tester protein sequences using computer-implementable logic;
From the plurality of tester protein sequences, select at least two peptide segments having at least 15% sequence identity with the lead sequence; and, as an output, create selected peptides that form a hit library. A computer readable medium including logic for constructing a library of mutant antibodies based on a lead sequence, the logic comprising:
Taking as input an amino acid sequence comprising at least three consecutive amino acid residues of the CDR of the lead antibody (the amino acid sequence is the lead sequence);
Comparing the lead sequence to multiple tester protein sequences;
Selecting at least two peptide segments having at least 15% sequence identity with the lead sequence from a plurality of tester protein sequences; and generating as output the selected peptides that form a hit library, including logic Medium. A method of constructing a library of antibodies based on the structure of a lead antibody, the method comprising the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure defined as a lead structural template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
A scoring function is used to determine whether the hit library members are structurally compatible with the lead structural template; and the hit library members with scores equal to or better than the lead sequence Select.   66. The method of claim 65, wherein the length of the lead sequence is 5-100 aa.   66. The method of claim 65, wherein the length of the lead sequence is 6-80 aa.   66. The method of claim 65, wherein the length of the lead sequence is 8-50aa.   66. The method of claim 65, wherein the step of identifying the amino acid sequence in the CDR is performed using Kabat criteria or Chothia criteria. The lead sequence is CDR1, CDR2, CDR3, FR1-CDR1, CDR1-FR2, FR2-CDR2, CDR2-FR3, FR3-CDR3, CDR3-FR4, FR1-CDR1-FR2, FR2-CDR2-FR3, and FR3-CDR3 66. The method of claim 65, comprising an amino acid sequence from a region within _VH or _VL of a lead antibody selected from the group consisting of -FR4.   66. The method of claim 65, wherein the lead sequence comprises at least 6 consecutive amino acid residues in the selected CDR.   66. The method of claim 65, wherein the lead sequence comprises at least 7 consecutive amino acid residues in the selected CDR.   66. The method of claim 65, wherein the lead sequence comprises all amino acid residues in the selected CDR.   66. The method of claim 65, wherein the lead sequence further comprises at least one amino acid residue immediately adjacent to the selected CDR.   66. The method of claim 65, wherein the lead sequence further comprises at least one amino acid residue in the FR that flanks the selected CDR.   66. The method of claim 65, wherein the lead sequence further comprises one or more CDRs or FRs adjacent to the C-terminus or N-terminus of the selected CDR.   66. The method of claim 65, wherein the plurality of tester protein sequences comprises an antibody sequence.   66. The method of claim 65, wherein the plurality of tester protein sequences comprises a human antibody sequence. 66. The method of claim 65, wherein the plurality of FR tester protein sequences comprises a humanized antibody sequence having at least 70% human sequence in _VH or _VL .   66. The method of claim 65, wherein the plurality of FR tester protein sequences comprises a human germline antibody sequence.   66. The method of claim 65, wherein the plurality of tester protein sequences are retrieved from a database consisting of NIH's Genebank, Swiss-Prot database, and Kabat database for antibody CDRs.   66. The method of claim 65, wherein the step of comparing the lead sequence to the plurality of tester protein sequences is performed by an algorithm selected from the group consisting of BLAST, PSI-BLAST, profile HMM, and COBLATH.   66. The method of claim 65, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 25%.   66. The method of claim 65, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 35%.   66. The method of claim 65, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 45%.   The scoring function is an energy scoring function selected from the group consisting of electrostatic interaction, van der Waals interaction, electrostatic solvation energy, solvent accessible surface solvation energy, and conformational entropy. 66. The method of claim 65.   The scoring functions are Amber force field (forcefiled), Charmm force field, Discover cvff force field, ECEPP force field, GROMOS force field, OPLS force field, MMFF94 force field, Tripos force field, MM3 force field, Dreiding force field, and 66. The method of claim 65, wherein the scoring function includes a force field selected from the group consisting of UNRES force fields. The process of selecting hit library members is:
ΔE _total = E _vdw + E _bond + E _angel + E _{electrostatics} + E _solvation
66. The method of claim 65, comprising selecting a member of the hit library that has a total energy that is lower than or equal to the lead sequence calculated based on the formula: The process of selecting hit library members involves an improved scoring function ΔG _b = ΔG _MM + ΔG _sol -TΔS _SS
(Where
ΔG _MM = ΔG _ele + ΔG _vdw (1)
ΔG _sol = ΔG _ele-sol + ΔG _ASA (2))
66. The method of claim 65, comprising selecting a member of a hit library having a binding free energy less than the lead sequence calculated as the difference between the bound and unbound states using.   66. The method of claim 65, wherein the lead structural template is a fully assembled lead antibody. 66. The method of claim 65, wherein the lead structure template is a 3D structure of the _VH or _VL of the lead antibody.   66. The method of claim 65, wherein the lead structure template is a 3D structure of a CDR or FR of a lead antibody or a combination thereof.   66. The method of claim 65, wherein the lead structure template is a structure obtained from X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy, or theoretical structural modeling. 66. The method of claim 65, further comprising the following steps:
A nucleic acid library comprising a DNA sequence encoding the amino acid sequence of the hit library is constructed. 66. The method of claim 65, further comprising the following steps:
Create an amino acid position variant profile of the hit library;
Converting amino acid position variant profiles of hit libraries into nucleic acid position variant profiles by back-translating amino acid position variants into corresponding gene codons; and combinatorially combining nucleic acid position variants into a degenerate nucleic acid library of DNA segments Build up.   96. The method of claim 95, wherein the gene codon is suitable for bacterial expression. 96. The method of claim 95, wherein the gene codon is selected such that the diversity of the degenerate nucleic acid library of the DNA segment is less than 1 × 10 ⁷ . 96. The method of claim 95, wherein the gene codon is selected such that the diversity of the degenerate nucleic acid library of the DNA segment is less than 1 × 10 ⁶ . 96. The method of claim 95, further comprising the following steps:
Introducing a DNA segment of a degenerate nucleic acid library into a host organism cell;
Expressing a DNA segment in the host cell such that a recombinant antibody containing the amino acid sequence of the hit library encoded by the degenerate nucleic acid library is produced in the host organism cell; and
Recombinant antibodies that bind to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ are selected. 99. The method of claim 99, wherein the affinity of the selected recombinant antibody is higher than 10 < ⁸ > M- ¹ . 99. The method of claim 99, wherein the affinity of the selected recombinant antibody is greater than 10 < ⁹ > M- ¹ .   99. The method of claim 99, wherein the host organism is selected from the group consisting of bacteria, yeast, plants, insects and mammals.   99. The method of claim 99, wherein the recombinant antibody is selected from the group consisting of a fully assembled antibody, Fab fragment, Fv fragment, or single chain antibody.   99. The method of claim 99, wherein the recombinant antibody is displayed on the surface of the phage particle. 105. The method of claim 104, wherein the recombinant antibody displayed on the surface of the phage particle is a double stranded heterodimer formed by _VH and _VL . Heterodimerization of V _H and V _L chains is facilitated by a heterodimer formed between two non-antibody polypeptide chain fused to the V _H and V _L chains, respectively The method of claim 105. 107. The method of claim 106, wherein the non-antibody polypeptide chain is obtained from the heterodimeric receptors GABA _B R1 (GR1) and R2 (GR2), respectively. 105. The method of claim 104, wherein the recombinant antibody displayed on the surface of the phage particle is a single chain antibody containing _VH and _VL linked by a peptide linker.   109. The display of single chain antibodies on the surface of phage particles is facilitated by a heterodimer formed by a fusion of a single chain antibody and GR1 and a fusion of phage pIII capsid protein and GR2. the method of.   99. The method of claim 99, wherein the target antigen is selected from the group consisting of small organic molecules, proteins, peptides, nucleic acids and polycarbohydrates. A method for constructing a library of antibody sequences comprising the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure defined as a lead structural template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence profile to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
Creating a hit library application virus particle based on the frequency of amino acid variants appearing at each position of the lead sequence;
Combining the amino acid variants in the hit library to create a combination of hit variants that forms a hit variant library;
A scoring function is used to determine whether members of the hit variant library are structurally compatible with the lead structural template; and a hit variant library with a score equal to or better than the lead sequence Select members. 112. The method of claim 111, wherein combining the amino acid variants in the hit library comprises:
Select amino acid variants with an appearance frequency greater than 4. 112. The method of claim 111, wherein combining the amino acid variants in the hit library comprises:
Select amino acid variants with an appearance frequency greater than 6. 112. The method of claim 111, wherein combining the amino acid variants in the hit library comprises:
Select amino acid variants with an appearance frequency greater than 5% of total variants at each position. 112. The method of claim 111, wherein combining the amino acid variants in the hit library comprises:
Amino acid variants with an occurrence frequency greater than 10% of the total variant are selected at each position; and the selected amino acid variants in the hit library are combined to produce a combination of hit variants that forms a hit variant library. 112. The method of claim 111, wherein combining the amino acid variants in the hit library comprises:
Select amino acid variants with an appearance frequency greater than 5% of total variants at each position;
If the frequency of occurrence is less than or equal to 5% of the total variant at each position, select the amino acid of the lead sequence; and combine the selected amino acid variants in the hit library to form a hit variant library Generate a combination of hit variants.   The scoring function is an energy scoring function selected from the group consisting of electrostatic interaction, van der Waals interaction, electrostatic solvation energy, solvent accessible surface solvation energy, and conformational entropy. 112. The method of claim 111, wherein:   The scoring functions are Amber force field (forcefiled), Charmm force field, Discover cvff force field, ECEPP force field, GROMOS force field, OPLS force field, MMFF94 force field, Tripos force field, MM3 force field, Dreiding force field, and 112. The method of claim 111, wherein the scoring function includes a force field selected from the group consisting of UNRES force fields. 112. The method of claim 111, further comprising the following steps:
A nucleic acid library is constructed that includes DNA segments that encode the amino acid sequences of selected members of the hit library. 112. The method of claim 111, further comprising the following steps:
Decomposing selected members of the hit variant library into at least two sub-hit variant libraries;
Create an amino acid position variant profile of the sub-hit variant library;
Converting the amino acid position variant profile of the selected sub-hit variant library into a nucleic acid position variant profile by back-translating the amino acid position variants into the corresponding gene codons; and combinatorially combining the nucleic acid position variants into a DNA segment Build a heavy nucleic acid library. 121. The method of claim 120, wherein decomposing the hit variant library comprises:
Randomly select 10-30 members of the hit variant library that have a score equal to or better than the lead sequence (the selected members form a subvariant library). 121. The method of claim 120, wherein decomposing the hit variant library comprises:
Create amino acid position variant profile of hit variant library to get hit variant profile; and use Cα, Cβ or heavy atom contact of lead structure template using a certain distance cut-off (4.5 ~ 8 ~) Based on the map, break the hit variant profile into sub-variant profile segments. 121. The method of claim 120, wherein decomposing the hit variant library comprises:
Create an amino acid position variant profile of the hit variant library to obtain a hit variant profile; and
Using a 6-8 distance cut-off, the hit variant profile is decomposed into sub-variant profile segments based on the contact map of Cα, Cβ or heavy atoms in the lead structure template. A method of constructing an antibody library based on a plurality of structural assemblies of antibodies, the method comprising the following steps:
Providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure);
Providing a 3D structure of one or more antibodies having different sequences in the _VH or _VL regions other than the lead antibody;
Combining the lead antibody and one or more antibodies to form a structural assembly (the structural assembly is defined as a lead structural template);
Identifying the amino acid sequence in the CDR of the lead antibody;
Select one CDR in the _VH or _VL region of the lead antibody;
Providing an amino acid sequence comprising at least three consecutive amino acid residues in a selected CDR (the selected amino acid sequence is a lead sequence);
Comparing the lead sequence to multiple tester protein sequences;
Selecting from the plurality of tester protein sequences at least two peptide segments having at least 10% sequence identity with the lead sequence (the selected peptide segments form a hit library);
Creating an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
Combining amino acid variants in the hit library to generate a combination of hit variants that form a hit variant library;
A scoring function is used to determine whether members of the hit variant library are structurally compatible with the lead structural template; and a hit variant library with a score equal to or better than the lead sequence Select members. A method of constructing a library of antibodies based on the structure of a lead antibody, the method comprising the following steps:
a) providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure);
b) identify the amino acid sequence in the CDR of the lead antibody;
c) select one CDR in the _VH or _VL region of the lead antibody;
d) providing an amino acid sequence comprising at least 3 consecutive amino acid residues in the selected CDR (the selected amino acid sequence is defined as the lead sequence);
e) compare the lead sequence to multiple tester protein sequences;
f) selecting at least two peptide segments having at least 10% sequence identity with the lead sequence from a plurality of tester protein sequences (the selected peptide segments form a hit library);
g) create an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
h) combining the amino acid variants in the hit library to generate a combination of hit variants that forms a hit variant library;
i) using a scoring function to determine whether members of the hit variant library are structurally compatible with the lead structure template;
j) Select a member of the hit variant library that has a score equivalent to or better than the lead sequence;
k) constructing a degenerate nucleic acid library comprising a DNA segment encoding the amino acid sequence of a selected member of the hit variant library;
l) determining the diversity of nucleic acid libraries, if diversity is greater than 1 × 10 ⁶ is the step j) to l) until the diversity of the nucleic acid library is less than or equal to 1 × 10 ⁶ repetition;
m) introducing the DNA segment in the degenerate nucleic acid library into the cells of the host organism;
n) expressing a DNA segment in the host cell so that a recombinant antibody containing the amino acid sequence of the hit library is produced in the host organism cell;
o) selecting a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ ; and
p) If no recombinant antibody is found that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ , repeat steps e) to o). A method of constructing a library of antibodies based on the structure of a lead antibody, the method comprising the following steps:
a) providing the amino acid sequence of the variable region of the heavy chain (V _H ) or light chain (V _L ) of the lead antibody (the lead antibody has a known three-dimensional structure defined as a lead structure template);
b) identify the amino acid sequence in the CDR of the lead antibody;
c) select one CDR in the _VH or _VL region of the lead antibody;
d) providing an amino acid sequence comprising at least 3 consecutive amino acid residues in the selected CDR (the selected amino acid sequence is defined as the lead sequence);
e) mutating the lead sequence by substituting one or more amino acid residues of the lead sequence with one or more different amino acid residues to create a lead sequence variant library;
f) using a first scoring function to determine if the lead sequence variant library is structurally compatible with the lead structural template;
g) select a lead sequence variant with a score equal to or better than the lead sequence;
h) comparing the lead sequence to multiple tester protein sequences;
i) selecting at least two peptide segments having a sequence identity of at least 10% with the lead sequence from the plurality of tester protein sequences (the selected peptide segments form a hit library);
j) create an amino acid position variant profile of the hit library based on the frequency of amino acid variants appearing at each position of the lead sequence;
k) combining amino acid variants in a hit library to generate a combination of hit variants;
l) combining a selected sequence variant with a combination of hit variants to produce a hit variant library;
m) Use a second scoring function to determine whether members of the hit variant library are structurally compatible with the lead structural template;
n) Select a member of the hit variant library that has a score equivalent to or better than the lead sequence;
o) constructing a degenerate nucleic acid library comprising DNA segments encoding the amino acid sequences of selected members of the hit variant library;
p) determining the diversity of nucleic acid libraries, if diversity is greater than 1 × 10 ⁶ is the step n) ~p) to the diversity of the nucleic acid library is less than or equal to 1 × 10 ⁶ repetition;
q) introducing the DNA segment in the degenerate nucleic acid library into the cells of the host organism;
r) expressing a DNA segment in the host cell such that a recombinant antibody containing the amino acid sequence of the hit library is produced in the host organism cell;
s) selecting a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ ; and
t) If a recombinant antibody that binds to the target antigen with an affinity higher than 10 ⁶ M ⁻¹ is not found, repeat steps e) to s). A method for constructing a library of designed proteins comprising the following steps:
Providing an amino acid sequence obtained from the lead protein (this amino acid sequence is referred to as the lead sequence);
Comparing the lead sequence to a plurality of tester protein sequences; and selecting from the plurality of FR tester protein sequences at least two peptide segments having at least 15% sequence identity with the lead sequence (the selected peptide segments are hit live A lead library); and the lead sequence is replaced with a hit library to form a library of designed proteins.   128. The method of claim 127, wherein the length of the lead sequence is preferably 5-100 aa.   128. The method of claim 127, wherein the length of the lead sequence is preferably 6-80 aa.   128. The method of claim 127, wherein the length of the lead sequence is preferably 8-50aa.   128. The method of claim 127, wherein the lead protein is a protein of a type selected from the group consisting of enzyme receptors, cytokines, tumor suppressors, chemokines, antibodies and growth factors.   128. The method of claim 127, wherein the plurality of tester protein sequences comprises a human protein sequence.   128. The method of claim 127, wherein the plurality of tester protein sequences comprises a humanized protein sequence each having at least 70% human sequence.   128. The method of claim 127, wherein the plurality of tester protein sequences are retrieved from a protein database of GeneBank or Swiss-Prot database.   128. The method of claim 127, wherein the step of comparing the lead sequence to the plurality of tester protein sequences is performed by an algorithm selected from the group consisting of BLAST, PSI-BLAST, profile HMM, and COBLATH.   128. The method of claim 127, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 25%.   128. The method of claim 127, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 35%.   128. The method of claim 127, wherein the sequence identity between the selected peptide segment in the hit library and the lead sequence is at least 45%. 128. The method of claim 127, comprising the following steps:
A protein having a desired function is selected from the designed library of proteins.   140. The method of claim 139, wherein the desired function is an improved biological function of the lead protein.   The improved biological function is selected from the group consisting of enhanced stability, enhanced enzyme activity, enhanced binding affinity of the lead protein to the same type of ligand, and enhanced expression of a predetermined organism. 145. The method of claim 140. 128. The method of claim 127, further comprising the following steps:
A nucleic acid library comprising a DNA segment encoding the amino acid sequence of the hit library is constructed. 128. The method of claim 127, further comprising the following steps:
Create an amino acid position variant profile of the hit library;
The amino acid variants in the hit library are combined to generate a combination of hit variants that forms a hit variant library; and a protein with a preferred function is selected from the hit variant library. 145. The method of claim 143, further comprising the following steps:
A scoring function is used to determine whether the members of the hit library are structurally compatible with the lead sequence or the three-dimensional structure of the lead protein; and the score is equal or better than the lead sequence or lead protein Selected members.   145. The method of claim 144, wherein the three-dimensional structure of the lead sequence or lead protein is a structure obtained from X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy, or theoretical structural modeling.   The scoring function is an energy scoring function selected from the group consisting of electrostatic interaction, van der Waals interaction, electrostatic solvation energy, solvent accessible surface solvation energy, and conformational entropy. 145. The method of claim 144, wherein:   The scoring functions are Amber force field (forcefiled), Charmm force field, Discover cvff force field, ECEPP force field, GROMOS force field, OPLS force field, MMFF94 force field, Tripos force field, MM3 force field, Dreiding force field, and 128. The method of claim 127, wherein the scoring function includes a force field selected from the group consisting of UNRES force fields. The process of selecting members is
ΔE _total = E _vdw + E _bond + E _angel + E _{electrostatics} + E _solvation
145. The method of claim 143, comprising selecting a member having a total energy lower than or equivalent to a lead sequence or lead protein calculated based on the formula: The process of selecting members is an improved scoring function ΔG _b = ΔG _MM + ΔG _sol -TΔS _SS
(Where
ΔG _MM = ΔG _ele + ΔG _vdw (1)
ΔG _sol = ΔG _ele-sol + ΔG _ASA (2))
144. The method of claim 143, comprising selecting a member having a binding free energy less than the lead sequence or lead protein calculated as the difference between the bound and unbound states. 128. The method of claim 127, further comprising the following steps:
Constructing a nucleic acid library comprising DNA segments encoding the amino acid sequences of the library of proteins to be designed;
A nucleic acid library is expressed to create a library of recombinant proteins; and a protein having the desired function is selected from the library of recombinant proteins. 128. The method of claim 127, further comprising the following steps:
Create an amino acid position variant profile of the hit library;
Converting the amino acid position variant profile of the hit library into a nucleic acid position variant profile by back-translating the amino acid position variant into the corresponding gene codon;
Combining nucleic acid position variants in a combinatorial manner to construct a degenerate nucleic acid library of DNA segments;
A degenerate nucleic acid library is expressed to create a library of recombinant proteins; and a protein having a desired function is selected from the library of recombinant proteins. Human vascular endothelial growth, wherein the binding affinity of the antibody to VEGF is higher than 10 ⁶ M ^{−1 and} the heavy chain CDR3 of the monoclonal antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 36-48 and 63-125 Antibody to factor (VEGF).   153. The antibody of claim 152, wherein the heavy chain CDR1 of the antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 19-30.   153. The antibody of claim 152, wherein the heavy chain CDR2 of the monoclonal antibody comprises an amino acid sequence selected from the group consisting of SEQ ID NOs: 31-305.   153. The antibody of claim 152, wherein the antibody is a monoclonal antibody, Fab, Fv, or single chain antibody. The binding affinity of the antibody to VEGF is higher than 10 ⁶ M ⁻¹ and the heavy chain variable region (V _H ) of the antibody is an amino acid sequence selected from the group consisting of SEQ ID NOs: 126, 128, 129, 130, and 131 An antibody to human vascular endothelial growth factor (VEGF), wherein the light chain variable region (V _L ) of the antibody comprises the amino acid sequence of SEQ ID NO: 127.

Images