DE NOVO DESIGN OF ANTIBODY

Description

[0001] The present invention relates to computational design of antibodies that will bind to a target epitope.

[0002] Targeting the correct epitope is a critical step in selection of a monoclonal antibody to achieve the desired mechanism of action. Current approaches for the discovery of novel antibodies for therapeutic and diagnostic use rely on raising antibodies against a target protein in immunised animals, or on in vitro screening of naive or immunised libraries using display technologies. Neither method allows complete control over affinity, specificity, epitope and binding mode.

[0003] Sormanni et al. (Sormanni, P., Aprile, F.A., Vendruscolo, M. Rational design of antibodies targeting specific epitopes within intrinsically disordered proteins. Proc. Natl. Acad. Sci. USA. 112, 9902-9907 (2015)), Robinson et al. (Robinson, L.N., et al. Structure-guided design of an anti-dengue antibody directed to a non-immunodominant epitope. Cell 162, 493-504 (2015)), Lippow et al. (Lippow, S.M., Wittrup, K.D. & Tidor, B. Computational design of antibody-affinity improvement beyond in vivo maturation. Nat. Biotechnol. 25, 1171-1176 (2007)), and Kuroda et al. (Kuroda, D., Shirai, H., Jacobson, M.P. & Nakamura, H. Computer-aided antibody design. Protein Eng. Des. Sel., 25, 507-521 (2012)) have demonstrated some success in attempts to engineer rationally antibodies but also that the computational design of antibodies targeting pre-selected epitopes on target proteins remains a challenging problem.

[0004] Computational antibody design has enabled rational engineering of antibodies to enhance affinity and stability by in silico scanning of interfacial CDR sequence spaces (see Lippow et al. above and Jordan et al. (Jordan, A.L., et al. Structural understanding of stabilization patterns in engineered bispecific Ig-like antibody molecules. Proteins 77, 832-841 (2009))). Recent development of general antibody design approaches like OptMAVEn (Li, T., Pantazes, R.J., Maranas, C.D. OptMAVEn-a new framework for the de novo design of antibody variable region models targeting specific antigen epitopes. PLoS One. 9, e105954 (2014)) and AbDesign (Lapidoth, G.D. et al. AbDesign: An algorithm for combinatorial backbone design guided by natural conformations and sequences. Proteins 83, 1385-1406 (2015)) are based on protein-protein docking to sample the possible binding poses of artificial antibody scaffolds, followed by the generation of combinatorial backbone configurations and sequence space scanning. US 2009 / 130102 A1 discloses methods for humanising antibodies and humanised antibodies made thereby. Kashmiri et al. (Kashmiri SV, et al. SDR grafting - a new approach to antibody humanization. Methods (San Diego, Calif.). 2005 May;36(1):25-34. DOI: 10.1016 / j.ymeth.2005.01.003. PMID: 15848072) discloses an approach to antibody humanisation. However without ultimate proof of experimental validation of designed antibodies from these methods so far, the computational design of high-affinity antibodies targeting precise epitopes remains a largely unsolved problem. The development of computational methods for the design of antibodies binding with high affinity at pre-selected epitopes would have wide-ranging applications, such as achieving epitope-dependent mechanism of actions and accessing immunisation blind spots which are often biologically relevant, conserved orthosteric sites.

[0005] It is an object of the invention to provide an alternative framework for computational design of antibodies.

[0006] According to the invention, there is provided a computer-implemented method of designing an antibody that will bind to a target epitope, as defined in claim 1.

[0007] The present inventors have demonstrated that is possible based on the above framework to design novel antibodies binding at naturally occurring protein-binding sites, guided by pre-identified hotspot-mediated interactions. The novel computational approach offers the potential for structure-based rational design of novel antibodies with precise control of binding mode for therapeutic and diagnostic application.

[0008] The binding affinities of the designed antibodies are optionally further optimised by in silico swap and redesign of the CDR sequences. Exemplification has been achieved through computational design of antibodies with nanomolar-level binding affinities to Kelch-like ECH-associated protein 1 (Keap1) at the nuclear factor-like 2 (Nrf2) binding site. An X-ray co-crystal structure of one of the designed antibodies shows atomic-level agreement with the corresponding computational model, demonstrating successful application of an experimentally validated computational design of antibodies targeting a pre-selected epitope.

[0009] In an embodiment the selection of candidate antibody structures from the database is performed using a preselection based on matching distances between characteristic atoms, followed by a further selection based on determining whether at least three of the matching residue sub-structure characteristic atoms can be superimposed on the corresponding at least three hotspot sub-structure characteristic atoms with the spatial deviation between each pair of superimposed atoms averaged over all pairs being less than the predetermined threshold. This two step approach enables the candidate antibody structures to be selected from the database particularly efficiently. This increase in efficiency is expected to become increasingly important as available databases of antibody structures get larger.

[0010] In an embodiment the generating of the designed antibody further comprises iteratively swapping one or more CDR loops of the candidate antibody structure with CDR loops from a database of CDR loops to increase a predicted affinity between the candidate antibody structure and the target epitope. The inventors have found that this step advantageously provides additional conformational degrees of freedom which allows improved affinity to be achieved between the designed antibody and the target epitope. In the absence of this step the relatively limited number of antibody structures available from databases means that it can be challenging to find high-affinity antibodies bearing CDRs that form optimal shape / electrostatic complementarity to the selected epitope on target proteins. CDR loop swap leverages the large number of sequences and experimentally determined CDR configurations from other antibody structures to construct new chimeric antibody models. Combining CDR loop swap with the other steps of the invention allows fast generation of high affinity antibodies targeting the selected binding site.

[0011] Embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings in which corresponding reference symbols represent corresponding parts, and in which: Figure 1 depicts steps in an example method of designing an antibody that will bind to a target epitope; Figure 2 depicts example implementation of a step of selecting candidate antibody structures from a database; Figure 3 depicts a pre-selection process for multiple matching residues; Figure 4 depicts a schematic example geometry for calculating a first set of distances for three hotspot sub-structures; Figure 5 depicts a pre-selection process for a single matching residue; Figure 6 depicts a schematic example geometry for calculating a first set of distances in a hotspot sub-structure having three characteristic atoms involved in superimposition; Figure 7 depicts a schematic example geometry for calculating a first set of distances in a hotspot sub-structure having four characteristic atoms involved in superimposition; Figure 8 depicts an example procedure for determining when characteristic atoms superimpose with an average spatial deviation within the predetermined threshold; Figure 9 depicts an example procedure for refining a designed antibody where geometrical clashing is detected; Figure 10 depicts an example workflow of hotspots-guided antibody scaffold graft design in anti-Keap1 antibodies targeting Nrf2 binding site; Figure 11 depicts SPR kinetic profiles for G54.1 / keap1 and G85 / keap1 interaction, where titrations of keap1 are flowed over chip surfaces comprising immobilized anti-keap1 Fabs and where the Fab designs have been derived from grafted Nrf2 hotspots; Figure 12 depicts sequence alignments of the V H regions of two best hotspots graft designs, G54 (a) and G85 (b), with corresponding original PDB scaffold structures and variants from in silico mutagenesis. Residues labelled "M" represent amino acids that differ from the scaffolds after hotspots graft and alanine mutation to reduce the clashes. "G" indicates residues that are introduced during in silico mutagenesis to yield variants G54.1 and G85.1, respectively. The three Nrf2-inspired hotspots being grafted are marked with asterisks; Figure 13 depicts SPR kinetic profiles for G54.1 / keap1 and G85 / keap1 interaction, in the presence of competing titrations of the cognate high affinity Nrf2 peptide segment that interacts with the keap1 binding site, thus demonstrating specific binding of designed antibody to the Nrf2 binding site of keap1; Figure 14 depicts the modelled binding poses of designed antibodies G54.1 (Left) and G85 (Right) in complex with Keap1 in anti-Keap1 antibodies targeting Nrf2 binding site; three hotspot residues (depicted as sticks) backbones on CDRH2 loops and Nrf2 peptide are superimposed; Figure 15 depicts workflow of CDRH3 loop swap design in affinity improvement of G54.1 antibody; Figure 16 depicts CDR-loop-wise Rosetta G scores decomposition of designed G54.1 antibody; the individual CDR loop's contributions to the Rosetta G scores between G54.1 and Keap1 were estimated by truncating each CDR loop from the Fv fragment of modelled G54.1 / Keap1 complex structure; Figure 17 depicts sequence alignment of CDRH3 loop in the CDRH3-swap variants of G54.1 in affinity improvement of G54.1 antibody by CDRH3 loop swap; Figure 18 depicts relative binding affinity improvements of designed CDRH3-swap variants over parental G54.1 Fab; Figure 19 depicts computationally modelled CDRH3 conformations and interaction modes of parental G54.1 and four highest affinity-improved CDRH3-swap designs with Keap1 in affinity improvement of G54.1 antibody by CDRH3 loop swap; key contact residues in CDRH3 loops are depicted as sticks; Figure 20 depicts close-ups of conformations and interaction modes of isolated modelled CDRH3 loops of four highest affinity-improved CDRH3-swap designs with Keap1: a, LS171; b, LS145; c, LS168; d, LS146; the conformations and interaction modes of CDRH3 (lighter grey) with Keap1 (darker grey) are show from top (Left) and side view (Right); the key contact residues in CDRH3 loops are depicted as sticks; it is clearly shown that V H 99L and V H 100Y in LS171, V H 97W in LS168 and V H 97Y in LS146 occupy the interfacial void between antibodies and Keap 1 that is not occupied by LS145 or G54.1 (see also Figure 19 for comparison); Figure 21 depicts a crystal structure of LS146-scFv / Keap1 complex showing the precision of the computational design; Figure 22 depicts crystal packing in LS146-scFv / Keap1 complex; the asymmetric unit contains two copies of LS146-scFv / Keap1 complexes; Figure 23 depict grafted hotpots binding site 2F-F maps of LS146-scFv / Keap1 complex; 2F o -F c omit map electron densities (grey meshes, contoured at 1.0 σ) of grafted hotspots and other LS 146-scFv CDRH2 residues interacting with Keap1 for the two molecules in the asymmetric unit; the crystal waters are shown as grey spheres; Figure 24 depicts a crystal structure of LS 146-scFv / Keap1 complex confirming occupation of Nrf2 binding site in Keap1; Figure 25 depicts a crystal structure of LS146-scFv / Keap1 complex showing the precision of the computational design - close-up of LS146-scFv epitopes of CDRH2, with the key contact residues depicted as sticks, and hydrogen bonds depicted as dot lines; Figure 26 depicts a close-up of LS146-scFv epitopes which are mapped onto Keap 1 molecular surface coloured in terms of contacting CDRs; Figure 27 depicts a crystal structure of LS146-scFv / Keap1 complex showing the precision of the computational design - close-up of LS146-scFv epitopes of CDRH3, with the key contact residues depicted as sticks, and hydrogen bonds depicted as dot lines; Figure 28 depicts a crystal structure of LS146-scFv / Keap1 complex showing the precision of the computational design - close-up of LS146-scFv epitopes of CDRH1, with the key contact residues depicted as sticks, and hydrogen bonds depicted as dot lines; Figure 29 depicts a crystal structure of LS146-scFv / Keap1 complex showing the precision of the computational design - close-up of LS146-scFv epitopes of V H framework 3 (FR3), with the key contact residues depicted as sticks, and hydrogen bonds depicted as dot lines; Figure 30 depicts a crystal structure of LS146-scFv / Keap1 complex showing the precision of the computational design - comparison of the binding modes of crystal LS146-scFv with modelled LS146-Fab by superimposing onto the Keap1 side; Figure 31 depicts a crystal structure of LS146-scFv / Keap1 complex showing the precision of the computational design - comparison of backbone conformations and sidechain orientations of CDRH2 loops (the hotspots acceptor) from crystal (Left) and modelled (Right) structures of LS146 Fv region; the key CDRH3 residues are depicted as sticks, and hydrogen bond that affects V H 52D's conformation from predicted model is depicted as dot lines; Figure 32 depicts a crystal structure of LS146-scFv / Keap1 complex showing the precision of the computational design - comparison of residues packing at V H / V L interface from crystal (Left) and modelled (Right) structures; the key packing residues that undergo apparent conformational change from prediction are depicted as sticks; Figure 33 depicts a comparison of potency of LS 146-scFv versus -Fab in Biacore competition assay; IC 50 values were calculated by fitting to the logarithm concentration versus normalized response / variable slope model: Y = 100 1 + 10 log IC 50 − X × S Hill ; Figure 34 depicts combined hotspot residues from TGFBR1 & 2 and Fresolimumab in pan-TGFb blocking Fab fragment design by transferring combined receptors- and Fresolimumab-inspired hotspot residues example; Figure 35 depicts SPR kinetics profiles for Fab 184 / TGFβs complexes with designed antibody Fab immobilized on the chips in pan-TGFb blocking Fab fragment design by transferring combined receptors- and Fresolimumab-inspired hotspot residues example; Figure 36 depicts neutralisation of TGFβs-receptors binding by titration of Fab184 TGFβs in HEK Blue reporter gene cell assay in pan-TGFb blocking Fab fragment design by transferring combined receptors- and Fresolimumab-inspired hotspot residues example; Figure 37 depicts comparison of the binding modes of crystal Fab184 with modelled one by superimposing onto the TGFβ1 side in pan-TGFb blocking Fab fragment design by transferring combined receptors- and Fresolimumab-inspired hotspot residues example.

[0012] There is provided a computer-implemented method of designing an antibody that will bind to a target epitope. Figures 1-9 schematically show example aspects of the method in flow chart form.

[0013] The method comprises a) identifying one or more hotspot residues that will each bind to a corresponding one of one or more hotspot sites on the target epitope (step 100 in Figure 1). Each hotspot residue comprises a hotspot sub-structure. The hotspot sub-structure comprises one or more hotspot sub-structure characteristic atoms. The hotspot sub-structure characteristic atoms are atoms that will be used for matching of residues that are potentially different to the hotspot residue (i.e. derived from a different amino acid). The characteristic atoms are thus atoms which are common to residues of different amino acid type.

[0014] The method further comprises b) selecting from a database of antibody structures one or more candidate antibody structures (step 200 in Figure 1). The antibody structures or relevant portions of the antibody structures may be referred to as antibody scaffolds. The selection is performed to find antibody structures or scaffolds that are capable of being modified to bear residues matching the hotspot residues (as described below). The nature or origin of the database is not particularly limited. The database entries may be filtered or reformatted as required. For example, in an embodiment, only database entries representing structures which have been solved by X-ray crystallography are used. In an embodiment if multiple crystal copies are available for the same antibody structure with different chain identifiers, only the first copy which appears in the PDB file may be retained for use. In an embodiment only the Fv regions are kept from the Fab structures. In an embodiment the Abnum procedure (Abhinandan, K.R. & Martin, A.C. R. Analysis and improvements to Kabat and structurally correct numbering of antibody variable domains. Mol. Immunol. 45, 3832-3839 (2008)) is used to renumber the residues in the Fv structures according to Chothia numbering scheme (Al-Lazikani, B., Lesk, A.M. & Chothia, C. Standard conformations for the canonical structures of immunoglobulins. J. Mol. Bio. 273, 927-948 (1997)). In an embodiment any structures with broken polypeptide CDR loops are discarded.

[0015] Each candidate antibody structure has one or more matching residues. Each of the matching residues matches a corresponding one of the hotspot residues (in the sense explained below). Each matching residue comprises a matching residue sub-structure. Each matching residue sub-structure comprises one or more matching residue sub-structure characteristic atoms. The selection is performed such that the relative positions of the matching residue sub-structure characteristic atoms within the antibody structure and the relative positions of the hotspot sub-structure characteristic atoms when bound to the target epitope are such that at least three of the matching residue sub-structure characteristic atoms can be superimposed computationally on a corresponding at least three hotspot sub-structure characteristic atoms with a spatial deviation between each pair of superimposed characteristic atoms averaged over all pairs being less than a predetermined threshold. The averaging may be achieved for example by computing a spatial separation between each pair of superimposed characteristic atoms and calculating a mean average or root mean square average of the spatial separations. Each of the corresponding matching residue sub-structure characteristic atoms and hotspot sub-structure characteristic atoms are generally of the same characteristic atom type (e.g. alpha carbon, backbone carbon derived from the carboxyl group, backbone nitrogen, backbone oxygen, beta carbon of the side chain, etc.). A matching residue is thus matched with a hotspot residue when corresponding characteristic atoms from each of the two residues can be superimposed over each other with relatively high precision (such that, overall, the average deviation satisfies the predetermined threshold as described above). The matching residue does not need to be of the same amino acid type as the hotspot residue (i.e. with the same side chain). The matching depends only on whether the two residues have characteristic atoms in the sub-structure that can be superimposed with relatively high precision. An example approach for determining whether this requirement is met for a given antibody structure is described below with reference to Figure 8.

[0016] The matching using at least three matching residue sub-structure characteristic atoms and a corresponding at least three hotspot sub-structure characteristic atoms constrains the position and orientation of the candidate antibody structure relative to the target epitope to at least partially retain functionally relevant aspects of the paratope / epitope interaction geometry of the one or more identified hotspot residues and the target epitope. Matching more than three characteristic atoms and / or matching using more than one matching residue will tend to increase the geometrical constraints and retain the paratope / epitope interaction geometry more closely (see examples below).

[0017] In an embodiment the selection of step (b) is performed by looking for matching residues exclusively within an interaction site on the antibody structure, the interaction site consisting of the CDR loops or the CDR loops and any region on the surface of the antibody Fv domain.

[0018] The method further comprises c) generating a designed antibody using one of the candidate antibody structures selected in step b) (step 300 in Figure 1). Either the candidate antibody structure is modified by replacing at least one of the matching residues with a different residue such that a predicted affinity between the designed antibody and the target epitope is higher than a predicted affinity between the candidate antibody structure and the target epitope, or the candidate antibody structure is output as a designed antibody structure, without modification at this stage, in the case where each of the matching residues is already a residue of the same amino acid as the hotspot residue which the matching residue matches. The different residue may be a residue of the same amino acid type as the corresponding hotspot residue for example. The replacing of a matching residue with a different residue may be referred to as grafting of the different residue. The designed antibody structure produced according to any of the procedures discussed above may be modified in a subsequent step to further improve an affinity between the designed antibody and the target epitope.

[0019] In an embodiment the predetermined threshold used in step (b) is 2.0 Angstroms, optionally 1.75 Angstroms, optionally 1.5 Angstroms, optionally 1.25 Angstroms, optionally 1.0 Angstroms. There is some freedom for choosing the predetermined threshold. Choosing a relatively high threshold may lead to more candidate antibody structures being selected from the database. This may increase the chances of finding a designed antibody structure with high affinity but will tend to increase demands on further processing steps used for example to assess the potential of the selected candidate antibody structures (e.g. by assessing real or predicted affinity and the extent to which further modifications may improve affinity). Choosing a relatively low threshold may result in fewer candidate antibody structures being selected from the database but these selected structures may on average be of greater potential. This may allow further processing steps to be more focussed and thereby potentially find high affinity novel antibody structures more quickly.

[0020] In an embodiment, in step (c) the modifying comprises replacing each of at least one of the matching residues with a residue of the same amino acid as the hotspot residue which the matching residue matches. In many cases this will result in the designed antibody structure achieving relatively high affinity by presenting at least one residue that is identical to a hotspot residue in terms of side chain and which is positioned and oriented in a very similar manner to the hotspot residue when the hotspot residue is bound to the target epitope (which by definition occurs with high affinity). However it is not essential that all matching residues are replaced with residues of the same amino acid type as the corresponding hotspot. In some cases, for at least a subset of the matching residues, a higher affinity may be obtained by not replacing the matching residue or by replacing the matching residue with a residue of an amino acid type which is not the same as the corresponding hotspot residue.

[0021] The characteristic atoms (either of the hotspot residue sub-structures or the matching residue sub-structures) may comprise one or more of the following: the alpha carbon, the backbone carbon atom derived from the carboxyl group, the backbone nitrogen, the backbone oxygen, and the beta carbon of the side chain.

[0022] In an embodiment, the alpha carbon atom of at least one of the matching residues is in one of the pairs of superimposed characteristic atoms.

[0023] In an embodiment, the pairs of superimposed characteristic atoms comprise the alpha carbon and at least one of the backbone carbon atom derived from the carboxyl group, the backbone nitrogen, the backbone oxygen, and the beta carbon of the side chain of each of at least one of the matching residues. Thus in this embodiment at least one of the matching residues has two characteristic atoms involved in the superimposition process. This provides relatively good matching in terms of position and orientation without overly constraining the selection process.

[0024] In an embodiment, the pairs of superimposed characteristic atoms comprise the alpha carbon and at least two of the backbone carbon atom derived from the carboxyl group, the backbone nitrogen, the backbone oxygen, and the beta carbon of the side chain of each of at least one of the matching residues. Thus in this embodiment at least one of the matching residues has three characteristic atoms involved in the superimposition process. This provides a relatively high degree of matching of position and orientation of the residue.

[0025] In an example useful for understanding the present invention, the one or more matching residues consists of a single matching residue only. In such an example each of the pairs of superimposed characteristic atoms will comprise a different characteristic atom from the single matching residue. In an example of this type, the at least three of the matching residue sub-structure characteristic atoms that can be superimposed on the corresponding at least three hotspot sub-structure characteristic atoms optionally comprise the alpha atom of the matching residue and at least two of the backbone carbon derived from the carboxyl group of the matching residue, the backbone nitrogen of the matching residue, the backbone oxygen of the matching residue, and the beta carbon of the side chain of the matching residue.

[0026] In an embodiment the one or more matching residues consists of a first matching residue and a second matching residue. In an example of an embodiment of this type the first matching residue comprises at least two of the matching residue sub-structure characteristic atoms that can be superimposed on the corresponding hotspot sub-structure characteristic atoms and the second matching residue comprises at least one of the matching residue sub-structure characteristic atoms that can be superimposed on the corresponding hotspot sub-structure atoms.

[0027] The one or more matching residues consists of a first matching residue, a second matching residue and a third matching residue (optionally a first matching residue, a second matching and a third matching residue only). Each of the first matching residue, second matching residue and third matching residue comprises three of the matching residue sub-structure characteristic atoms that can be superimposed on the corresponding hotspot sub-structure characteristic atoms. This approach imposes a relative high constraint on the relative positions and orientations of the three matching residues, thereby providing a relatively focussed selection of candidate antibody structures having a relatively high average affinity (relative to less restrictive selections of candidate antibody structures) even without further modifications to improve affinity further. In a particular embodiment the three of the matching residue sub-structure characteristic atoms in each of the three matching residues that are involved in the superimposition comprise the alpha carbon atom, the backbone carbon atom and the backbone nitrogen atom. The inventors have found this combination to be particularly effective, as demonstrated in the detailed Keap 1 example discussed below.

[0028] As shown in Figure 2, in an embodiment the selection of the one or more candidate antibody structures (step 200 in Figure 1) comprises a pre-selection of a subset of antibody structures (step 210 in Figure 2) followed by a further selection (step 220 in Figure 2).

[0029] In an embodiment the pre-selection (step 210) comprises the steps set out in Figure 3 and explained below with reference to the schematic example geometry depicted in Figure 4. The pre-selection comprises (step 211A) determining a first set of distances representing separations between all possible pairings between identical characteristic atoms in different sub-structures of the hotspot residues. This is illustrated schematically, simplified into a two dimensional view, in Figure 4. Figure 4 shows the hotspot residue sub-structure characteristic atoms for three different hotspot residues: circles A1-A3 represent the characteristic atoms for a first hotspot residue, circles B1-B3 represent the characteristic atoms for a second hotspot residue, and circles C1-C3 represent the characteristic atoms for a third hotspot residue. The broken lines connect together all possible pairs of characteristic atoms of the same characteristic atom type (e.g. alpha carbon, backbone carbon derived from the carboxyl group, backbone nitrogen, backbone oxygen, beta carbon of a side chain, etc.). The lengths of all the broken lines represents the first set of distances: {s11, s12, s13, s21, s22, s23, s31, s32, s33}.

[0030] The pre-selection further comprises (step 212A) determining a second set of distances representing separations between all possible pairings between identical characteristic atoms in different sub-structures of the matching residues. This process is the same as the process of step 211A except that characteristic atoms of the matching residues are used instead of the hotspot residues. The second set of distances will take the same form as the first set of distances (e.g. a set comprising 9 numbers). In an embodiment, the numbers are expressed to a predetermined level of accuracy (e.g. rounded up to the nearest Angstrom). In an embodiment the first and second sets of distances are expressed as a sequence of numbers in a canonicalized form to allow easy comparison between sequences obtained from different antibody structures. The sequence of numbers may be used as an index for searching the database of antibody structures (see Keap1 example discussed below).

[0031] The pre-selection further comprises (step 213A) comparing the first set of distances to the second set of distances to determine if a match has been obtained within a predetermined separation threshold. For example, a sequence of numbers representing the first set, expressed to the predetermined level of accuracy (which effectively defines the predetermined separation threshold - a lower level of accuracy will correspond to a larger predetermined separation threshold and vice versa), is compared with a sequence of numbers representing the second set, expressed to the same predetermined level of accuracy. If YES, the process proceeds to step 215A and the antibody structure is output for further processing. If NO, the process loops through steps 214A, 212A and 213A to iteratively repeat the determination of the second set of distances and the comparison with the first set of distances until a match is obtained. The process may also loop through steps 214A, 212A and 213A after the output step 215A in order to select multiple antibody structures for further processing.

[0032] In an embodiment the pre-selection (step 210) comprises the steps set out in Figure 5 and explained below with reference to the schematic example geometries depicted in Figures 6 and 7. In this embodiment the pre-selection comprises (step 211B) determining a first set of distances representing separations between all possible pairings between different characteristic atoms of the sub-structure of a single hotspot residue. This is illustrated schematically, simplified into two dimensional views, for different example hotspot sub-structures in Figures 6 and 7. Figure 6 shows an example hotspot sub-structure in which three characteristic atoms A1, A2 and A3 are involved in the superimposition with a corresponding matching residue sub-structure (having a corresponding three characteristic atoms of corresponding type). Figure 7 shows an alternative example hotspot sub-structure in which four characteristic atoms A1, A2, A3 and A4 are involved in the superimposition with a corresponding matching residue sub-structure (having a corresponding four characteristic atoms of corresponding type). In Figures 6 and 7 the broken lines connect together all possible pairs of characteristic atoms in the hotspot residue. By definition each pair will involve a pairing between characteristic atoms of different type to each other because they are in the same residue. The lengths of all the broken lines represents the first set of distances: {d1, d2, d3} for Figure 6 and {d1, d2, d3, d4, d5, d6} for Figure 7.

[0033] The pre-selection further comprises (step 212B) determining a second set of distances representing separations between all possible pairings between different characteristic atoms of the sub-structure of the matching residue. This process is the same as the process of step S211B except that the characteristic atoms of the matching residue are used instead of the characteristic atoms of the hotspot residue. The second set of distances will take the same form as the first set of distances (e.g. a set comprising 3 or 6 numbers for the particular geometries shown in Figures 6 and 7). In an embodiment, the numbers are expressed to a predetermined level of accuracy (e.g. rounded up to the nearest Angstrom). In an embodiment the first and second sets of distances are expressed as a sequence of numbers in a canonicalized form to allow easy comparison between sequences obtained from different antibody structures.

[0034] The pre-selection further comprises (step 213B) comparing the first set of distances to the second set of distances to determine if a match has been obtained within a predetermined separation threshold. For example, a sequence of numbers representing the first set, expressed to the predetermined level of accuracy (which effectively defines the predetermined separation threshold - a lower level of accuracy will correspond to a larger predetermined separation threshold and vice versa), is compared with a sequence of numbers representing the second set, expressed to the same predetermined level of accuracy. If YES, the process proceeds to step 215B and the antibody structure is output for further processing. If NO, the process loops through steps 214B, 212B and 213B to iteratively repeat the determination of the second set of distances and the comparison with the first set of distances until a match is obtained. The process may also loop through steps 214B, 212B and 213B after the output step 215B in order to select multiple antibody structures for further processing.

[0035] In an embodiment, the further selection step 220 of Figure 2 comprises determining whether at least three of the matching residue sub-structure characteristic atoms can be superimposed on the corresponding at least three hotspot sub-structure characteristic atoms with the spatial deviation between each pair of superimposed atoms averaged over all pairs being less than the predetermined threshold. Figure 8 depicts an example approach for determining when this requirement is met for a given antibody structure.

[0036] In step 221 of Figure 8, the matching residue sub-structure characteristic atoms are computationally superimposed (i.e. overlaid) over the hotspot sub-structure characteristic atoms in the relative position or positions they occupy when bound to the target epitope. The way in which this initial superimposition is performed is not particularly limited. In step 222 a spatial deviation is calculated for each pair of identical characteristic atoms in each pair of matching residue and corresponding hotspot residue. An average of these spatial deviations is then obtained, for example by calculating a mean average or a root mean square average. If the characteristic atoms are all exactly superimposed then the average spatial deviation will be zero. Otherwise, the average spatial deviation will be a measure of the extent to which the set of pairs of characteristic atoms superimpose for the particular relative positions and orientations of the antibody structure for this iteration. In step 223 it is determined whether the average spatial deviation is below a predetermined threshold. This determination tests whether the fit is sufficiently close to be satisfactory. If YES, it is concluded that the antibody structure is a candidate antibody structure and the result is output for further processing (step 227). If NO, the process loops through steps 224, 225, 222 and 223 where the antibody structure is shifted relative to the hotspot residues and the average spatial deviation is recalculated and compared with the threshold. The process continues until either a sufficiently good match is obtained (by reaching step 227) or a predetermined maximum number of iterations has been achieved, in which case the YES branch of step 224 is followed to step 226 and the process starts again from step 221 with a different antibody structure.

[0037] In an embodiment the generating of the designed antibody comprises one or more further processing steps to modify the candidate antibody structure to further improve a predicted affinity with the target epitope (e.g. by iteratively mutating residues or iteratively swapping CDR loops - see below) or to discard antibody structures which will not work (for example due to clashing - see below). These further processing steps comprise computationally modifying the candidate antibody structure while the designed antibody is in a binding position defined by the matching to the identified hotspot residues. The sub-structure atoms of the antibody structure that correspond to the sub-structure characteristic atoms used in the superimposition of the selecting step (b) discussed above are therefore positioned relative to the target epitope at the same positions as the corresponding hotspot sub-structure characteristic atoms. In this way the superimposition process not only assists with selecting the most suitable candidate antibody structures from the database but also in providing an efficient reference for fixing the antibody structures in a way which conserves the critical paratope / epitope interaction geometry, therefore enabling the further processing steps to be performed in an efficient and effective way.

[0038] In an embodiment the generating of the designed antibody comprises detecting geometrical clashing. Geometrical clashing is where one or more atoms are predicted to occupy positions that are closer together than is physically possible when a candidate antibody structure is computationally bound to the target epitope. An example procedure for dealing with geometrical clashing is depicted in Figure 9.

[0039] In step 301 it is determined whether geometrical clashing has occurred and, if so, which atoms are involved in the geometrical clashing. If NO, the process proceeds to step 306 where the candidate antibody structure is output for further processing. If YES, the process proceeds to step 302.

[0040] In step 302 it is determined whether the geometrical clashing involves a backbone of any candidate antibody residue. If YES, the process proceeds to steps 304 and 301, whereby the candidate antibody structure is discarded and the process is repeated with a different candidate antibody structure. If NO, the process proceeds to step 303.

[0041] In step 303 it is determined whether the geometrical clashing involves a beta carbon atom of any candidate antibody residue. If YES, the process proceeds to steps 304 and 301, whereby the candidate antibody structure is discarded and the process is repeated with a different candidate antibody structure. If NO the process proceeds to step 305.

[0042] In step 305 it is determined whether the geometrical clashing is with a side chain of a residue of the candidate antibody structure. If YES, the process proceeds to step 307 where the side chain is modified. The modification may involve swapping the side chain for a side chain of a different amino acid, for example a smaller amino acid. For example, the side chain may be modified to an alanine side chain, a glycine side chain, a valine side chain, a serine side chain, a threonine side chain, or homo-alanine side chain. The process then proceeds to step 301 where it is determined whether there is still a geometrical clash. If NO at step 305 the process proceeds to step 306 where the candidate antibody structure is output for further processing.

[0043] In an embodiment the generating of the designed antibody further comprises iteratively mutating the amino acid types of residues in the candidate antibody structure to increase a predicted affinity between the designed antibody and the target epitope. This process may be referred to as in silico mutagenesis. In an embodiment the selection of residues that are iteratively mutated is constrained so that the hotspot residues are not mutated. In other embodiments the selection of residues is not constrained to avoid mutation of the hotspot residues. Subject to the potential constraint mentioned above, the iterative mutation may comprise singly mutating all residues in a region on the candidate antibody that is expected to participate significantly in the interaction with the target epitope (e.g. an interfacial region), for example to all other amino acid types (excluding glycine, proline, and cysteine). The skilled person would be aware of various algorithms for performing computational analyses involving iterative mutations of residues to reduce a free energy associated with binding of a protein to a target. For example, the Rosetta software suite may be used (https: / / www.rosettacommons.org / ).

[0044] In an embodiment the generating of the designed antibody further comprises iteratively swapping each of one or more of the CDR loops of the candidate antibody structure with CDR loops from a database of CDR loops to increase a predicted affinity between the candidate antibody structure and the target epitope. The affinity may be predicted for example using publically available software such as the Rosetta software suite. Swapping CDR loops greatly increases freedom of design, effectively increasing the number antibody structures that can be tested relative to the number of antibody structures available in the original database.

[0045] In an embodiment the swapping of the CDR loops is constrained so that all of the hotspot residues are retained. The inventors have found that this approach allows a designed residue of high affinity to be obtained without placing excessive demands on computing resource.

[0046] In an alternative embodiment the swapping of the CDR loops is constrained so that at least one of the hotspot residues is retained. The inventors have found that this approach provides more freedom of mutation than embodiments in which all hotspot residues are required, potentially allowing antibodies with higher affinity to be found, without demand on computing resource being increased too much.

[0047] In an embodiment the swapping of the CDR loops comprises swapping at least one of the CDRH3 loop and CDRL3 loop. These loops show the most variability. Focussing on swapping these loops allows affinity to be improved most efficiently.

[0048] In an embodiment the swapping of the CDR loops comprises swapping at least the CDRH3 loop. This loop is the most variable. Focussing on swapping this loop allows affinity to be improved even more efficiently.

[0049] In an embodiment the swapping of the CDR loops further comprises iteratively mutating the amino acid types of residues in the swapping CDR loops to increase predicted affinity between the candidate antibody structure and the target epitope. This step enables affinity to be increased still further.

[0050] One or more of the hotspot residues themselves may be identified (step 100 in Figure 1) in a variety of different ways. In an embodiment the hotspot residues are identified from a cognate protein binder known to bind to the target epitope. This approach provides hotspots residues with a high level of reliability and predictable affinity. However, the range of hotspot residues that can be identified in this way is limited. In the Keap 1 example discussed in detail below the hotspot residues were determined based on the known cognate binding partner, Nrf2.

[0051] Alternatively or additionally, one or more of the hotspot residues may be identified using a numerical method to iteratively find residues that are predicted to provide an interaction with the target epitope consistent with providing a disproportional amount of a binding energy between an antibody comprising the residues and the target epitope.

[0052] The term "hotspot" in the context of protein binding is well known in the art. The skilled person would understand that each pair of hotspot residue and corresponding hotspot site on the target epitope define an interaction between a hotspot residue and the target epitope consistent with providing a disproportional amount of a binding energy between an antibody comprising the hotspot residues and the target epitope. See for example: Fleishman, S.J. et al. Computational design of proteins targeting the conserved stem region of influenza hemagglutinin. Science 332, 816-821 (2011); Liu, S. et al. Nonnatural protein-protein interaction-pair design by key residues grafting. Proc. Natl Acad. Sci. USA, 104, 5330-5335 (2007); and Fleishman, S.J. et al. Hotspot-centric de novo design of protein binders. J. Mol. Biol. 413, 1047-1062 (2011).

[0053] In one embodiment multiple designed antibodies are obtained and a preferred designed antibody is selected based on its real affinity for the target epitope, determined for example using surface plasmon resonance.

[0054] Any or all of the steps of embodiments of the invention may be performed using computing apparatus known to the skilled person in combination with appropriate software and / or firmware. The software may be provided as a signal from an external source or recorded in a memory or computer readable media.FURTHER DETAILS, SPECIFIC EXAMPLES AND RESULTSKeap1 example

[0055] In a specific example, embodiments of the invention were applied to design antibodies binding to Keap1, a BTB-Kelch substrate adaptor protein that regulates steady-state levels of Nrf2, a bZIP transcription factor, in response to oxidative stress. Nrf2 binds to Keap1 in a 1:2 stoichiometric ratio through two hairpin loop motifs with binding affinities of 5 µM and 5 nM, respectively. Three interactional patterns, derived from hotspot residues Glu79, Thr80 and Glu82 in the higher affinity Nrf2 loop (see Supplementary Table 1), were grafted into designed antibodies' binding interfaces and ranked by computed binding energy (Figure 10 and Supplementary Table 2). Five designs were selected and subjected to in silico mutagenesis to identify extra potential interfacial point mutations in CDR loops with improved binding energies to Keap1, leading to the generation of variants of original designs. The ten designed antibodies, before and after in silico mutagenesis, were expressed in the Fab format, and their binding affinities were measured by surface plasmon resonance (SPR). Eight of the ten selected antibody Fab designs showed detectable binding against Keap1, with the best two (G54.1 and G85) showing binding affinities in the low-to-mid nanomolar range (Figure 11, Figure 12 and Supplementary Tables 2-4). Binding was reduced when a cognate Nrf2 peptide binder was added as a competitor (Figure 13), suggesting that the epitope of the designed antibodies on Keap1 overlapped with that of Nrf2. The original antibody scaffolds of G54.1 and G85 (Protein Data Bank (PDB) accession codes 31VK and 2JB5, respectively) did not bind to Keap1, and none of the corresponding native antigens were biologically associated with Keap 1 or Nrf2 (Supplementary Table 4), strongly suggesting that the Keap1 binding of both antibodies was mediated via the computationally designed interfaces. Modelled structures suggested that the three Nrf2 hotspots grafted onto CDRH2 loops of the two antibody scaffolds presented similar conformations to the Nrf2 peptide and completely occupied the Nrf2 binding sites on Keap1 along with CDRH1 and CDRH3 loops (Figure 14).

[0056] A barrier to designing high affinity antibodies is that current approaches treat their scaffolds as rigid structures with minimal perturbation of their backbone degrees of freedom. However there is an experimentally validated precedent for transplanting CDR loops into different antibody frameworks due to the structural conservation of different loop types, thus providing alternative, additional conformation degrees of freedom that have so far been untapped by rigid-scaffold design methods. See the following publications for example: Clark, L.A. et al. An antibody loop replacement design feasibility study and a loop-swapped dimer structure. Protein Eng. Des. Sel. 22, 93-101 (2009); Söderlind, E. et al. Recombining germline-derived CDR sequences for creating diverse single-framework antibody libraries. Nat. Biotechnol. 18, 852-856 (2000); North, B., Lehmann, A., Dunbrack, R.L. A New clustering of antibody CDR loop conformations. J. Mol. Bio. 406, 228-256 (2011).

[0057] In order to improve further the binding affinity, a computational method was developed to swap the CDRH3 loop of G54.1 with ones from a curated CDRH3 loop fragment structure library (Figure 15), given that CDRH3 is known as the most diverse antibody loop in terms of length and conformation among the six CDRs, and does not host any hotspot residues in this case (Figure 14 and Figure 16). The CDRH3 sequences of generated chimeric Fv fragments in complex with Keap1 were further optimised using RosettaDesign (Kuhlman, B. et al. Design of a novel globular protein fold with atomic-level accuracy. Science 302, 1364-1368 (2003)) and ranked by computed binding energy. Nineteen CDRH3-swap variants of G54.1 were selected (Figure 17 and Supplementary Tables 5-7), four of which show obviously improved affinities, with the best affinities of 4.1 and 5.4 nM measured from LS171 and LS145, representing respectively a 30- and 23-fold improvement of affinity over parental G54.1 (Figure 18 and Supplementary Table 7), and rivalling the affinity of cognate Nrf2. LS148 and LS146, albeit with weaker affinities, show respectively 13- and 6-fold improvement. These four CDRH3 swap designs possessed completely new CDRH3 loops (sequences and lengths, Figure 17) with different conformations from G54.1, presenting improved shape complementarity scores with Keap 1 (Supplementary Tables 5). As shown in the modelled structures (Figure 19 and Figure 20), these affinity-improved G54.1 variants either adopt aromatic residue substitutions in shorter CDRH3 (10 vs. 13 of G54.1) to fill a void between G54.1 and the Keap 1 surface (like VH99L and V H 100Y in LS171, V H 97W in LS168 and V H 97Y in LS146), or bear larger CDRH3 contact surface areas with Keap1 (like 2734 Å 2 of LS168 vs. 2583 Å 2 of G54.1).

[0058] A high resolution (1.85 Å) crystal structure of Keap1 in complex with LS146 (Figures 21-23) was then solved, due to the failure of crystallographic trials with the other three highest-affinity CDRH3-swap designs to yield diffraction-quality crystals. LS146, formatted as a single chain Fv (scFv), binds almost exactly as designed in the Nrf2 binding site on Keap1 (Figure 24), with CDRH2 making the most extensive contacts (interfacial hydrogen bond networks) to Keap 1 residues (Figures 25 and 26). The 12-mer long CDRH3 loop folds into a hairpin-like conformation and interacts with the loops at the end of two Keap1 propeller blades as predicted (Figure 27). Other CDR loops involved in binding are CDRH1 (Figure 28) and part of V H framework 3 (Figure 29). The structure of LS146-scFv bound to Keap 1 shows general atomic-level agreement with the design model (interfacial-Cα-atom root-mean-square deviation (RMSD) = 2.5 Å, with the two complex structures superimposed on the Keap1 side; Figure 30). The three grafted hotspots adopt nearly identical side chain orientations as predicted (heavy-atom RMSD = 1.6 Å; Figure 31), with the exception of a flipped sidechain of V H 52D due to an unexpected intramolecular hydrogen bond with backbone amide of V H 53E. An obvious conformational drift occurs at the tip of CDRH3 loop led by sidechain c reorganisation of V H 96Y, V H 100 Y, V L 49Y, and V L 55Y (Figure 32), which changes the torsional angle between CDRH1 and L1 and detaches V L completely from Keap1 (Supplementary Table 8). It is known that conversion to scFv can lead to variation in V H / V L orientation and a subsequent loss in affinity, which may explain why the potency of LS146- scFv is three-fold lower than that of its Fab form (Figure 33).

[0059] Although CDRH2 of LS146 displays a similar structural configuration (Cα-atom RMSD = 0.27 Å) as well as high sequence identity (83%) with hotspot residue donor Nrf2 'DEETGE' peptide segment, CDRH2 was not the only hotspot residue acceptor identified in antibody scaffold grafting. Because the triplet hashing (see below) was performed against all the surface CDR residues, CDRH3 was also found hosting Nrf2-inspired hotspots in some designs, albeit of much weaker affinities (Supplementary Table 4). Comparison of the properties of strong and weak binding designs suggests that more favourable computed Keap1-antibody binding energies, larger interfacial surface areas, and fewer buried unsaturated polar atoms are the most important factors (Figure 11 and Supplementary Table 2). These are reminiscent of the well-known challenges of computational antigen-antibody interface design (large, polar binding surfaces dominated loop interactions). Rational swapping of CDR configurations enables exploration of alternative shapes and chemical complementarities that are untapped by hotspot-guided grafting design, which relies on a limited number of scaffold structures (Supplementary Table 5). The tested loop swap designs, with distinctive CDRH3 backbone conformations and sequences, show improved binding affinities by targeting the same epitope, suggesting that use of the computational CDR swap strategy described enables optimisation of in silico designed antibodies for experimental selection of higher-affinity variants.

[0060] Although not a conventional target for therapeutic antibodies given that it is an intracellular protein, the Keap1-Nrf2 interaction features readily identifiable hotspot residues that provide an ideal proof-of-concept system for structure-based design of novel antibodies targeting pre-selected epitopes to directly block the cognate protein-protein interactions, or alternatively to capture predicted transition states, circumventing the need to isolate or stabilise transient conformations. With further improvements in computational accuracy and parallel probing of designed sequence space, using modern oligonucleotide assembly methods, such as focused display library design, and next generation sequencing, for efficient selection of stronger binding variants, the structure-based design method offers the potential for rapid generation of antibodies for therapeutic and diagnostic applications.Further details of computational methods applied to Keap1 example General computational methods

[0061] Anti-Keap1 antibodies targeting Nrf2 binding site were designed by a residue-based triplet hashing method to search for antibody scaffold crystal structures that are able to accommodate Nrf2 hotspots-mediated interaction patterns in the geometrically matched positions in CDRs, followed by CDRH3 swap to explore alternative loop configurations of the selected design. RosettaDesign was utilised to optimize the CDR loops' sequences of the designs during these two stages to improve the predicted binding energy to Keap1. The pseudo codes for hotspots graft, CDRH3 swap, and RosettaScripts design protocols used are provided at the end of the description.Hotspots graft

[0062] The triplet-based hashing method is an example of the process described above with reference to step 200 in Figure 1, in the case where three matching residues are used, each having three sub-structure characteristic atoms involved in the superimposition. Further information about performing triplet hashing more generally may be found in the following publication: Wolfson, H.J. & Rigoutsos, I. Geometric hashing: an overview. J. Comput. Sci. Eng. 4, 10-21 (1997). The triplet hashing was implemented to search for antibody structures ("scaffolds") that were able to host hotspots-mediated interaction patterns from 1417 antibody crystal structures in SAbDab database (Supplementary Table 9). A 'triplet' was defined as consisting of three virtual triangles that connected three residues' backbone Cα, N and C atoms, respectively. Any three Nrf2 hotspots were compiled into a triplet and indexed with a unique key for looking up. All possible triplets of the CDRs residues in antibody scaffold structures were enumerated and indexed in the same way. The identical triplets from hotspots and antibody scaffolds were identified by comparing the respective index keys. The antibody scaffolds were grafted onto the hotspots by superimposing the scaffold triplet onto the corresponding identical hotspots one to minimise the RMSD between two sets of nine vertexes in the three triplet triangles. The three scaffold triplet residues were replaced with corresponding hotspots ones. The designed structures after triplet superimposition and hotspots graft were discarded if the backbone atoms of any residues in the grafted antibody scaffolds clashed with Keap1.CDRH3 loop swap

[0063] All the exogenous CDRH3 loops were dissected from the 1417 antibody scaffold structures aforementioned. The original CDRH3 loop was removed from G54.1 / Keap1 complex structure in the same way, onto which each exogenous CDRH3 loop was grafted by superimposing the backbone atoms of the anchor residues, and then ligated onto G54.1 framework by connecting the new CDRH3 anchor residues with the adjacent G54.1 framework residues. The designed structures were discarded if the backbone atoms of the new CDRH3 loop clashed with either original G54.1 Fv or Keap1.Rosetta sequence design

[0064] Two rounds of Rosetta sequence design were used, aiming for optimising the computed binding energies for the designs obtained from hotspots graft and CDRH3 loop swap, respectively. During the first round, starting from the five designed antibody structures that accommodated the three Nrf2 hotspots-mediated interaction patterns, each interfacial position in antibody side was singly mutated to all other amino acid types (excluding glycine, proline, and cysteine). Each mutation structure was optimized by repack and minimization of all the interfacial residues. The changes of computed binding energies for each point mutation (termed ΔΔG) were evaluated in Rosetta full-atom scoring terms with the long-range electrostatics correction (see Fleishman, S.J. et al. RosettaScripts: A scripting language interface to the Rosetta macromolecular modelling suite. PLoS ONE 6, e20161 (2011)). Maximum five top ranked single point mutations in terms of lowest ΔΔG scores were selected for manual incorporation into a combined mutant variant of each original design. During the second round, all CDRH3 residues in CDRH3-swap variants of G54.1 were allowed to mutate into all other amino acid types (excluding glycine, proline, and cysteine) simultaneously, with the backbone conformation of all interfacial residues on CDRs and Keap1 locally perturbed using backrub method, which has been reported to help improving mutant side-chains prediction (Smith, C.A., Kortemme, T. Backrub-like backbone simulation recapitulates natural protein conformational variability and improves mutant side-chain prediction. J. Mol. Biol. 380, 742-756 (2008). Three iterations of sequence design were used to increase the likelihood that higher-affinity interactions could be found, starting with a soft-repulsive potential, and ending with the default standard van-der-Waals parameters.Design scoring

[0065] Designs were evaluated by computed binding energy (Rosetta G score), buried solvent accessible surface area (SASA), and shape complementarity (Sc) score (see Lawrence, M.C., Colman, P.M. Shape complementarity at protein / protein interfaces. J. Mol. Biol. 234. 946-950 (1993)). High shape complementarity was enforced by rejecting designs with Sc < 0.5 in hotspots graft and Sc < 0.6 in CDRH3 swap. Rosetta total energy for each designed complex structure, and number of buried unsaturated polar atoms (Stranges, P.B. & Kuhlman, B. A comparison of successful and failed protein interface designs highlights the challenges of designing buried hydrogen bonds. Protein Sci. 22, 74-82 (2013)) were used as the reference of the design quality evaluation as well.General experimental methods

[0066] Detailed procedures for the Keap 1 protein as well as antibodies expression, cloning, purification, crystallization are given below and Supplementary Tables 10, 11.Binding analysis

[0067] Surface plasmon resonance (SPR) experiments were carried out on a Biacore 3000 system (GE Healthcare) and detailed experimental details are given below. Briefly, supernatant containing expressed Fab (or sham transfected supernatant control) was injected over immobilized anti-human F(ab') 2 polyclonal on a CM5 chip. A second injection of a Keap1 titration or a zero analyte control allowed association and dissociation kinetics to be monitored. Chip regeneration completed each sensorgram cycle. Sensorgrams were corrected for baseline drift, caused by slow dissociation of captured Fab, by subtraction of an adjacent zero analyte control cycle. Non-specific binding of Keap1 at each concentration was corrected for by subtraction of the equivalent, baseline corrected, control supernatant cycle sensorgram. Biaevaluation ™< software was used to fit association and dissociation kinetics and hence determine affinity constants (K D ). Specificity of Fab binding to Keap1 was assessed by the same protocol by titration of an Nrf2 peptide analogue against a constant concentration of Keap1.Supplementary information Nrf2 hotspots identification

[0068] Three Nrf2 hotspot residues dominating the binding to Keap1 were identified using Rosetta in silico alanine scanning script AlaScan.xml (see Das, R., Baker, D. Macromolecular modeling with Rosetta. Annu. Rev. Biochem. 77, 363-382 (2008)). The binding energy of Nrf2 and Keap1 in the complex structure (PDB accession code 2FLU - see Lo, S.C., Li, X., Henzl, M.T., Beamer, L.J. & Hannink, M. Structure of the Keap1:Nrf2 interface provides mechanistic insight into Nrf2 signalling. Embo J. 25, 3605-3617 (2006)) was predicted by calculating the Rosetta total energy difference using default all-atom forcefield (score12 weights) between bound and unbound structures, referred as Rosetta G scores hereafter. Each Nrf2 residue was in silico mutated into alanine, and the top ranked three Nrf2 residues (Glu79, Thr80, and Glu82) with the Rosetta G scores decreased by at least 0.8 Rosetta energy unit (REU) after alanine mutation were confirmed as hotspots (Supplementary Table 1). The hotspots conformations were diversified by generation of inverse rotamers starting from their side chain atoms nearest to the Keap 1 surface using the Rosetta script InverseRotamers.xml. Extra rotamer sampling (two half step standard deviations) was performed around all side chain torsion angles.Antibody V-region scaffold structures

[0069] The antibody V-region scaffold structures with at least one paired V H / V L stored in PDB were extracted from SabDab (http: / / opig.stats.ox.ac.uk / webapps / sabdab) database in 2014. Only the structures solved by X-ray crystallography were used, including Fab and scFv formats. If multiple crystal copies were available for the same antibody structure with different chain identifiers, only the first copy which appeared in the PDB file was kept. Only the Fv regions were kept from the Fab structures. Abnum (Abhinandan, K.R. & Martin, A.C. R. Analysis and improvements to Kabat and structurally correct numbering of antibody variable domains. Mol. Immunol. 45, 3832-3839 (2008)) was used to renumber the residues in the Fv structures according to Chothia numbering scheme (Al-Lazikani, B., Lesk, A.M. & Chothia, C. Standard conformations for the canonical structures of immunoglobulins. J. Mol. Bio. 273, 927-948 (1997)). Any structures with broken polypeptide CDR loops were discarded. Finally 1417 antibody Fv scaffold structures were kept for hotspots graft design (Supplementary Table 8).Graft Nrf2 hotspots onto antibody scaffold structures

[0070] The residue-based triplet hashing method was implemented to search for the best antibody scaffold structures to graft the three Nrf2 hotspots onto, while maintaining the hotspots original interaction patterns with Keap1. We defined a 'residue triplet' as consisting of three virtual triangles that connected three residues' backbone Cα, N and C atoms, respectively. The triplet is characterised by nine vertexes (Vα1, Vα2, Vα3, VN1, VN2, VN3, VC1, VC2 and VC3, corresponding to the positions of nine backbone Cα, N, and C atoms of the three residues consisting of the triplet) and nine edges (Eα1, Eα2, Eα3, EN1, EN2, EN3, EC1, EC2 and EC3, corresponding to the edges from the three triangles). On the hotspots side, any three inverse rotamers were enumerated from the three Nrf2 hotspot residues (Glu79, Thr80, and Glu82) and compiled into a residue triplet. Each triplet was canonicalized by ensuring that the longest and second longest Cα edges always corresponded to Eα1 and Eα2, respectively. Each triplet was indexed into a unique string key by concatenating six edges' round-off (RO) lengths in order. For example, for a given triplet with Eα1=6.32, Eα2=4.67, Eα3=8.8, EN1=4.3, EN2=3.93, EN3=7.21, EC1=5.28, EC2=5.4 and EC3=9.82 the key is expressed as: Key = Concatenate RO E = 6594475510

[0071] All of the non-redundant index keys of hotspots' triplets were stored into a lookup table for fast access to corresponding hotspot triplet's information, including vertex residue types and atomic coordinates to facilitate later grafting onto the CDRs of antibody scaffold structures.

[0072] On the antibody scaffold side, any three CDR residues were enumerated and compiled into a triplet. The index key lookup table was generated in the same way as for hotspots triplet. To find the antibody scaffold structures which are able to accommodate the three hotspot residues in the geometrically matched positions in CDRs, the identical hotspots and antibody scaffold triplets were identified by directly comparing the respective index keys. The antibody scaffolds were grafted onto the hotspots by superimposing the scaffold triplet onto the corresponding identical hotspots one to minimise the RMSD between two sets of nine vertexes of the three triplet triangles. The three scaffold triplet residues were replaced with corresponding hotspots' ones by fitting the hotspots backbone atoms onto those of antibody triplet ones.

[0073] For each antibody designs obtained from hotspots graft, the sidechains of interfacial residues in antibody scaffolds clashing with Keap1 atoms were mutated into alanine to reduce clashes. The heavy-atom RMSD of the hotspots sidechain atoms before and after replacement was calculated. All residues were repacked and minimised using the Rosetta ppk.xml script. Several filters described below were applied to triage the designs: The heavy-atom RMSD of the hotspots before and after replacement onto the antibody scaffold was smaller than 2.0 Å. The buried solvent accessible surface area (SASA) upon binding was greater than 1200 Å (Hu, Z., Ma, B., Wolfson, H. & Nussinov, R. Conservation of polar residues as hot spots at protein interfaces. Proteins 39, 331-342 (2000). Shape-complementarity (Sc) score was greater than 0.5. The Rosetta G score (binding energy) was lower than 0.0 REU.

[0074] The surviving designs that passed the filtering rules were finally ranked by Rosetta G scores.CDRH3 loop swap

[0075] The individual CDR loop's contributions to the Rosetta G scores of G54.1 were calculated by truncating each CDR loop from the Fv region of modelled G54.1 / Keap1 complex structure (Figure 16). The Rosetta G scores of each CDR truncation mutant were re-calculated. Individual CDR's contribution to binding was estimated by computing the Rosetta G scores difference between each CDR truncation mutant and the original G54.1 antibody.

[0076] All the exogenous CDRH3 loops from the antibody scaffold crystal structures used in previous hotspots graft stage were dissected at the positions from V H 93 to V H 103 (according to Chothia numbering scheme) of Fv structures and labelled as the CDRH3 anchor residues. To graft an exogenous CDRH3 loop onto G54.1, the original CDRH3 loop of G54.1 was removed at the positions from V H 94 to V H 102, leaving V H 93 and V H 103 as the Fv anchor residues. Each exogenous CDRH3 loop was fitted onto the G54.1 Fv structure by superimposing the backbone atoms from two sets of anchor residues. The Fv anchor residues of G54.1 were later removed and the grafted exogenous CDRH3 loop was ligated onto G54.1 Fv by connecting the CDRH3 anchor residues with the neighbouring G54.1 residues (V H 92 and V H 104). The resulting structures were discarded if the backbone atoms of the new CDRH3 loop clashed with original G54.1 / Keap1 complex structure. Any CDRH3 residue sidechains clashing with G54.1 / Keap1 residues were mutated to alanine to reduce clashes. The final structures obtained from CDRH3 swap were repacked and minimised using Rosetta ppk.xml script as in Step 2 and ranked by Rosetta G scores.Rosetta sequence design

[0077] Two rounds of Rosetta sequence design were performed to optimise the binding affinities of the designed antibodies from hotspots graft and CDRH3 swap, respectively.

[0078] During the first round, starting from the five designed antibody structures that accommodated the three Nrf2 hotspots-mediated Keap1 interaction patterns, each interfacial CDR residue in the antibody side was mutated into other amino acid types (except cysteine, glycine and proline) to probe the mutation effect on Rosetta G scores in order to identify mutants that were potentially able to improve the computed binding energies of designed antibodies with Keap1. The Rosetta script MutationScanPB.xml for computing change in binding free energy during in silico mutagenesis using the scoring function with the modified electrostatics scoring term was used to generate the single point mutants list. The point mutations were ranked by calculating the change of Rosetta G scores, or, between each mutant and corresponding wild type structures. The top ranked single point mutations were selected and combined (maximum 5 mutations) to generate a variant of the original antibody graft.

[0079] During the second round, all residues of the swapped CDRH3 loops on G54.1 were allowed to mutate into all other amino acid types (excluding glycine, proline, and cysteine) simultaneously, with the backbone conformation of all interfacial residues on CDRs and Keap1 locally perturbed using backrub method, using the Rosetta flexbb-interfacedesign.xml script. Explicit electrostatics was not used in the scoring function. Three iterations of redesign and minimization were used to increase the likelihood that higher-affinity interactions could be found, starting with a soft-repulsive potential (soft_rep weights), and ending with the default all-atom forcefield (score12 weights). Similar filter rules previously described for hotspots grafting designs were used to triage and rank the resulting CDRH3-swap designed structures: The buried SASA upon binding was greater than 2000 Å. The Rosetta G score was lower than -20.0 REU. Sc score was greater than 0.6. Design scoring

[0080] All the previously described computational features used for filtering or ranking the designs (Supplementary Table 2, 5) were calculated by Rosetta3.4 InterfaceAnalyzer application: Rosetta G score, or binding energy was defined as the difference between the total system energy in the bound and unbound states. In each state, interface residues were allowed to repack. Rosetta total energy of the modelled complex structures. Buried solvent accessible surface areas (SASAs) were defined as the difference between the total system SASAs in the bound and unbound states. Shape-complementarity (Sc) score of the modelled antibody / Keap1 complex structures. Buried unsaturated polar atoms.

[0081] Finally, 10 designs in 5 unique scaffolds after hotspots graft (Supplementary Table 3) and 19 CDRH3-swap variants of G54.1 were chosen for experimental testing (Supplementary Table 6).Keapl expression & purification

[0082] The gene encoding the Kelch domain of Keap 1 was cloned into the expression vector pET-28a in frame with an N-terminal His tag and a TEV protease cleavage site. The construct was transformed into E. Coli strain BL21 (DE3), which was subsequently cultured in 2TY medium containing 25ug / ml kanamycin at 37 °C. Protein production was induced with 0.3 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) at an O.D.600 of 4. Glycerol-based feed (50 mM MOPS, 1 mM MgSO4 / MgCl2, 2 % glycerol) was added to the culture immediately after addition of IPTG, and the cultured was incubated further at 17 °C overnight. Cells were harvested by centrifugation and lysed in a buffer containing 50 mM Tris pH 8.5, 50 mM NaCl, 10% glycerol, 0.5% tritom-X100, 20 mM imidazole and sufficient amount of protease inhibitors (Roche). The lysate, pre-cleared by centrifugation, was filtered with a 0.2 µM filter and then mixed with Ni-NTA beads (Qiagen). The beads were washed with 50 mM Tris pH 8, 150 mM NaCl, 50 mM imidazole and 1 mM DTT before Keap 1 was eluted with the former buffer supplemented with imidazole to a concentration of 250 mM. After the His tag was cut off, the sample was applied to a Ni-NTA (Qiagen) column to remove any Ni-binding contaminating proteins. The flow-through was collected and further purified by size exclusion (Superdex 75, GE Healthcare) and, if necessary, ion exchange (Mono Q, GE Healthcare) chromatography. The purified keap1 was concentrated and stored in 20 mM Tris pH 7.5 and 5 mM DTT at -80 °C.Antibody cloning & expression

[0083] Heavy and light chain variable region genes designed in silico were chemically synthesized by DNA2.0, Inc. Transcriptionally active PCR (TAP) was employed to separately amplify the heavy and light chain variable regions and subsequently introduce DNA sequences encoding the hCMV promotor sequence, human γ1 C H 1 and C κ (Km3 allotype) constant regions and poly(A) tail. The resultant constructs contained all of the required components for transient cellular expression. To generate Fab fragments for SPR analysis, HEK-293 cells were transiently transfected with TAP products using 293Fectin lipid transfection (Life Technologies, according to the manufacturer's instructions).

[0084] Crystallographic trials with the top four high affinity CDRH3-swap antibodies in Fab formats failed to yield diffraction-quality crystals in complex with Keap1. To convert LS146 from a Fab to a scFv construct, a gene encoding V H fused to V L through a (Gly 4 Ser) 4 linker, a His 10 tag along with a TEV protease cleavage site was synthesized and cloned into a UCB proprietary expression vector by DNA2.0, Inc. The amino acid sequence of the gene product is given in Supplementary Table 10. CHO-S XE cells, a CHO-K1 derived cell line were transiently transfected with plasmid DNA using electroporation. Cells were removed by centrifugation and scFv-TEV-His tagged protein was purified by IMAC. Supernatant was filtered with a 0.2uM filter and then loaded into a HisTrap excel column (GE healthcare). The column was washed with 50 mM Tris pH 8, 150 mM NaCl, 45 mM imidazole before the antibody was eluted with 50 mM Tris pH 8, 150 mM NaCl, 250 mM imidazole. After the His tag was removed, the sample was applied to the HisTrap excel column again to remove the Ni-binding contaminating proteins. The flowthrough was collected and further purified by size exclusion (Superdex 75, GE Healthcare) chromatography. Purified antibody was concentrated, in 50 mM HEPES pH 7.5, 150 mM NaCl, 5% glycerol, and stored in aliquots at -80 °C until required.Binding analysis

[0085] Surface plasmon resonance (SPR) experiments were carried out on a Biacore 3000 system (GE Healthcare) using reagents from the same manufacturer. Fabs were captured on the surface of CM5 sensor chips via affinity purified goat polyclonal F(ab') 2 fragment specific to anti-human F(ab') 2 (Jackson 109-006-097). The latter was immobilised to the activated carboxymethyl dextran surface via amine coupling as follows: a fresh mixture of 50 mM N-hydroxysuccimide and 200 mM 1-ethyl-3-(3- dimethylaminopropyl)-carbodiimide was injected for 5 minutes at a flow rate of 10 µl / min, followed by 50 µg / ml anti-human F(ab') 2 in 10 mM acetate pH 5.0 buffer for 5 min at the same flow rate. Finally the surface was deactivated with a 10 minute pulse of 1 M ethanolamine•HCl pH 8.5. Reference flow cell was on the chip was prepared by omitting the protein from the above procedure, thus in the following experiments sensorgrams were obtained as the response unit difference between anti-F(ab') 2 and reference flow cells. Initial binding of Keap1 to expressed Fabs was assessed by injecting 50 µl supernatant, diluted 1 in 5 in running buffer, over the reference and anti-F(ab') 2 flow cells at a flow rate of 10 µl / min, followed by a 150 µl injection of 0, 500 or 5000 nM Keap1 in running buffer at a flow rate of 30 µl / min. After the dissociation phase lasting at least 5 min the chip surface was regenerated with two 60 sec pulses of 40 mM HCl interspersed with a 30 sec pulse of 5 mM NaOH at the same flow rate. Association and dissociation kinetics of Keap1 binding to captured Fabs were determined by the same protocol over at least 8 values of the following concentrations: 75, 100, 150, 250, 350, 500, 750, 1000, 1500, 2500, 3500 and 5000 nM. Zero Keap1 controls were interspersed between the former cycles in order to correct for baseline drift and sham transfected supernatant was assessed at each Keap 1 concentration in order to determine and correct for non-specific binding of Keap 1 . Specificity of Fab binding to Keap 1 was assessed by competition with a high-affinity Nrf2 peptide analogue, biotin-PEG-LQLDEETGEFLP1Q-amide, corresponding to Nrf2 residues 74 to 87 that comprise the stronger Keap1 binding loop motif. Peptide Keap1 binding in the presence of peptide titrations to captured Fabs was followed using the above protocol. Using BIAevaluation ™< software all sensorgrams were first transformed by subtracting a zero Keap1 control cycle and the corresponding non-specific control cycle prior to fitting dissociation and association kinetics. Dissociation constants (K D ) were estimated as the logarithmic mean of values measured over at least 6 Keap1 concentrations. IC 50 values were calculated using GraphPad Prism ™< software by fitting to the log concentration versus normalized response / variable slope model represented by the following equation, where percent inhibition values for the three report points were treated as replicates at each concentration: Y = 100 1 + 10 log IC 50 − X × S Hill .Crystallisation

[0086] Keap1 was buffer exchanged to the storage buffer of LS146-scFv (50 mM HEPES pH 7.5, 150 mM NaCl and 5% glycerol) prior to complex formation. This removed DTT from Keap1 storage buffer and prevented it from breaking the disulphide bonds in the antibody. Keap 1 was then mixed with LS146-scFv at a molar ratio of 1:1.5 and incubated at room temperature for 30 minutes. The complex was purified by size exclusion chromatography (Superdex 75 ™< 26 / 60, GE Healthcare) and concentrated to 5 mg / ml. Initial crystallisation trials, with 200 nl protein solution plus 200 nl reservoir solution (Qiagen) in sitting-drop vapor-diffusion format, produced crystals in two conditions. Reproduction and optimization of one of the hit crystallization conditions (0.2 M sodium acetate and 20% PEG3500), using seed crystals obtained from the initial screening, generated diffraction quality crystals. The crystals were cryoprotected in mother liquor, supplemented with PEG 3350 to 35% (w / v), and vitrified in liquid nitrogen prior to data collection.Crystallographic data collection and processing

[0087] Datasets from crystals LS146-scFv / Keap1 complex was collected at the Diamond Light Source synchrotron facility (Didcot, United Kingdom) on beamline 104-1 at a wavelength of 0.917 Å. Molecular replacement was performed using program PHASER 9< in the CCP4 software suite 10,11< using Keap1 (PDB accession code 1X2J 12< ), V H and V K frameworks without CDR loops (PDB accession code 3IVK 13< ) as the models. See: McCoy, A.J. et al. Phaser crystallographic software. J. Appl. Crystallogr. 40, 658-674 (2007); Potterton, E., Briggs, P., Turkenburg, M., & Dodson, E. A graphical user interface to the CCP4 program suite. Acta Crystallogr. Sect. D 59, 1131-1137 (2003); Winn, M.D. et al. Overview of the CCP4 suite and current developments. Acta Crystallogr. Sect. D 67, 235-242 (2011); Padmanabhan, B. et al. Structural basis for defects of Keap1 activity provoked by its point mutations in lung cancer. Mol. Cell 3, 689-700 (2006); and Shechner, D.M. et al. Crystal Structure of the Catalytic Core of an RNA-Polymerase Ribozyme. Science 326, 1271-1275 (2009). The solvent content of the crystal was determined as 46.09% and there are two copied of complexes in an asymmetric unit. Solutions were found in three stages; positions of two copies of Keap1 were searched and obtained first, and then the two copies of heavy chains and the two light chains. Refinement and model building were carried out using Refmac5.4 (REFinement of MACromolecular structures) and COOT (Crystallography Object-Oriented Toolkit), respectively. The geometric quality of the final model was validated using Rampage, ProCheck, SFCheck, and the validation tools provided by the RCSB Protein Data Base. Data collection and refinement statistics for LS146-scFv / Keap1 is provided in Supplementary Table 11. See: Murshudov, G.N., Vagin, A.A. & Dodson, E.J. Refinement of macromolecular structures by the maximum-likelihood method. Acta Cryst. D53, 240-255 (1997); Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. Acta Crystallogr. Sect. D 60, 2126-2132 (2004); Lovell, C. Structure validation by Calpha geometry: phi,psi and Cbeta deviation. Proteins 50, 437-450 (2002). 17. Laskowski, R.A., MacArthur, M.W., Moss, D.S., & Thornton, J.M. PROCHECK: a program to check the stereochemical quality of protein structures. J. Appl. Crystallogr. 26, 283-291 (1993); and Vaguine, A.A., Richelle, J., & Wodak, S.J. SFCHECK: a unified set of procedures for evaluating the quality of macromolecular structure-factor data and their agreement with the atomic model. Acta Crystallogr. Sect. D 55, 191-205 (1999).Additional example - Computational design of novel pan-TGFβ blocking antibody Fab fragment by transplanting combined hotspot residues from native TGFβ receptors and a known anti-TGFβ antibody

[0088] Inspired by the success of antibody design targeting Keap1, we applied the same approaches on TGFβs to design a pan-specific anti-TGFβs antibody. TGFβ is widely expressed and has a multitude of different functions, including immune homeostasis and fibrosis regulation. TGFβs exist in a homodimer format and there are at least three homologous isoforms (TGFβ1, TGFβ2, and TGFβ3), which signal via the same receptors complex consisting of TGFβs dimer and three membrane receptors (TGFβR1, TGFβR2, and TGFβR3). TGFβR2 initially binds at the tip of the "fingers" on TGFβ and later recruits the other two receptors binding to the TGFβ dimer interface. The crystal complex structure of TGFβ1 and the extracellular domains of TGFβR1 and TGFβR2 have been solved. We attempted to design antibodies to bind at the same region as the two receptors do by transplanting five interfacial hotspot residues from two receptors, but unfortunately did not generate any experimentally validated binding. It was speculated that the receptors-inspired hotspots were not strong enough to fix the antibody scaffold templates at the desired binding site because the affinities of hotspot donors, the TGFβ receptors, are very weak (K D values of 2.5 and 0.4 µM for TGFβR1 and TGFβR2, respectively). Fresolimumab (GC-1008) is a pan-TGFβ blocking antibody with low-nanomolar affinities. The crystal structure of Fresolimumab in complex with TGFβ3 reveals that the epitopes of Fresolimumab are highly overlapped with the receptors binding sites. So it is presumed by mixing the hotspot residues from both two receptors and Fresolimumab as combined query will increase the chance to generate an antibody binder binding at the same region. Five residues from receptor 1&2 and 9 residues from Fresolimumab were selected by virtual alanine scanning and used as the mixed query hotspots. It is noted that our hotspots transplant approach is based on residue triplet hashing that each time only three out of the 14 hotspot residues are transplanted first to determine an orientation for the given antibody templates, on which the rest of the hotspots are transplanted by checking if their backbone atoms' positions are close to those of any residues on the orientation of the antibody template fixed by the current hotspots triplet. After hotspots transplant, the residues on CDR loops of the antibody templates are mutated by Rosetta to generate new sequences to stabilize the current transplant and orientation using the same method aforementioned in the Keap 1 case. Given that the highly homologous of the three TGFβs at the receptors binding site, only TGFβ1 structure from the complex with TGFβR1 and TGFβR2 was used as the antigen target to calculate the Rosetta binding energy for each designed antibody Fab structural model.

[0089] The affinities of the designed antibodies Fab fragments were measured using Biacore aforementioned. Only one designed Fab shows obvious affinities against TGFβ1 and TGFβ3 (K D s are 106 and 32.9 nM, respectively), and much weaker affinity against TGFβ2 (the biding curves were difficult to fit). The affinities are much weaker than those of the reference antibody Fresolimumab, but are slightly stronger than those of the receptors. To test if the designed antibody is able to block the receptors' binding and disrupt the initiated downstream signalling, a cellular reporter gene assay driven by TGFβs binding was developed to determine the blocking efficacy of the designed antibody. It is demonstrated that upon antibody binding, the downstream signalling initiated by all three TGFβs binding with the corresponding receptors were partly disrupted in a concentration-dependent manner. The IC 50 s were determined and displayed a correlation with the K D s from biophysical binding assay. It is indicated that the designed antibody Fab, though presenting weak affinities, is probably binding at the region overlapping with receptors and Fresolimumab's epitopes, and therefore blocks the receptors binding as expected in a pan-specific manner.

[0090] The complex of the designed Fab with TGFβ1 was crystalized and the structure was solved. As predicted, the Fab completely occupies both receptors binding site on TGF β1, and overlaid very well with the predicted binding pose. The heavy chain of the antibody occupies majority of the binding site using hydrophobic residues, including CDR H2 and H3 hosting four hotspot residues from the receptors and Fresolimumab.

[0091] Supplementary Table 12 shows binding affinities of the ordered antibody Fab designs from hotspots graft. Dissociation constants (K D ) were determined by SPR.

[0092] Supplementary Table 13 shows Fv regions' amino acid sequences of ordered antibody designs from hotspots graft.

[0093] Supplementary Table 14 shows pan-blocking IC 50 s of Fab 184 design from hotspots graft in the reporter gene assay (n = 2).

[0094] Figures 34-37 depict pan-TGFb blocking Fab fragment design by transferring combined receptors- and Fresolimumab-inspired hotspot residues: Figure 34 - Combined hotspot residues from TGFβR1 & 2 and Fresolimumab; Figure 35 - SPR kinetics profiles for Fab184 / TGFβs complexes with designed antibody Fab immobilized on the chips; Figure 36 - Neutralisation of TGFβs-receptors binding by titration of Fab184 TGFβs in HEK Blue reporter gene cell assay; and Figure 37 - Comparison of the binding modes of crystal Fab184 with modelled one by superimposing onto the TGFβ1 side.SUPPLEMENTARY TABLES

[0095] Supplementary Table 1. Nrf2 Hotspots identification by in silico alanine scanning. Nrf2 residue Change of Rosetta ΔG scores upon in silico alanine mutation (REU) Note Glu 780.74Not used due to sidechain missing in the crystal structureGlu 793.15Strong hotspot, hydrogen bonds with Keap1 R415 and R483Thr 800.95Weak hotspotGly 8134.08Not suitable for hotspot without sidechainGlu 823.11Strong hotspot, hydrogen bonds with Keap1 S363, R380, and N382Phe 830.01Non-hotspotLeu 840.24Non-hotspot

[0096] Supplementary Table 2. Computational features of ordered antibody designs from hotspots graft. Design Rosetta ΔG (REU) Rosetta total energy (REU) Buried SASA (Å 2< ) Shape complementarity Buried unsaturated polar atoms G53-14.6-854.422750.5915G53.1-20.8-989.421750.5714G54-16.8-815.625140.619G54.1-32.3-993.125830.584G55-15.6-981.114530.573G55.1-19.7-1089.813520.512G56-14.2-973.718940.5510G56.1-23.3-1074.216500.533G85-15.8-791.026240.5919G85.1-19.5-938.427050.5619

[0097] Supplementary Table 3. Fv regions' amino acid sequences of ordered antibody designs from hotspots graft. Design Sequence V H V L G53G53.1G54G54.1G55G55.1G56G56.1G85G85.1

[0098] Supplementary Table 4. Binding affinities of the ordered antibody Fab designs from hotspots graft.

[0099] Dissociation constants (K D ) were determined by SPR. Design Scaffold 1< Hotspots positions #Mutations from scaffolds (except grafted hotspots) Fraction of Fab binding sites occupied @500nM Keap1 3< k on ( M -1< s -1< ) k off ( s -1< ) K D (nM) K D 95% CI 4< G532YSS a< V H 53E, V H 54T, V H 56E30.002ND 2< NDNDNDG53.150.0009NDNDNDNDG543IVK b< V H 53E, V H 54T, V H 56E60.01NDNDNDNDG54.190.4682.1×10 5< 2.6×10 -2< 126110-143G553TCL c< V H 102E, V H 102 A< T, V H 102 C< E10.015NDNDNDNDG55.130.016NDNDNDNDG563U4B d< V H 102E, V H 102 A< T, V H 102 C< E10.023NDNDNDNDG56.120.027NDNDNDNDG852JB5 e< V H 54E, V H 55T, V H 57E60.1792.3×10 5< 4.9×10 -2< 236137-405G85.170.1716.8×10 4< 2.3×10 -2< 341209-555 1< Original antigens in the PDB structures: a< Hen Lysozyme; b< RNA fragment; c,d< HIV-1 Envelope Glycoprotein Gp120; e< Diagnostic dye molecule. 2< ND: Not determined. 3< Limit if detection = 0.008 4< 95% confidence intervals of K D

[0100] Supplementary Table 5. Computational features of ordered CDRH3-swap variants of G54.1. Design Rosetta ΔG (REU) Rosetta total energy (REU) Buried SASA (Å 2< ) Shape complementarity Buried unsaturated polar atoms 171-43.24-1063.625900.6314145-46.25-1058.727340.6510168-46.4-1063.026560.6415146-45.6-1080.426630.6312142-46.9-1071.026280.658153-45.5-1080.525480.659144-45.1-1076.326180.6710143-45.1-1054.626430.6511151-46.8-1085.525570.6513149-39.5-1054.526150.65147-43.3-1068.825120.647152-41.7-1040.124970.6612150-38.2-1065.625070.628169-41.5-1060.524290.639175-43.3-1071.325880.649174-43.5-1060.123350.6710148-43.6-1066.326450.669170-43.4-1073.724980.6710173-45.9-1083.426800.6110

[0101] Supplementary Table 6. Fv regions' amino acid sequences of ordered CDRH3-swap variants of G54.1.

[0102] All CDRH3-swap V L sequences are identical to that of G54.1. Design V H sequence 171145168146142153144143151149147152150169175174148170173

[0103] Supplementary Table 7. Binding affinities of ordered antibody Fab fragments of CDRH3-swap variants of G54.1. Dissociation constants (K D ) were determined by SPR. Design CDRH3 donor 1< CDRH3 length #Mutations from original CDRH3 donor k on ( M -1< s -1< ) k off ( s -1< ) K D (nM) K D 95% CI 1712VDO1062.4×10 5< 9.0×10 -4< 4.13.2-5.31452ROZ1382.1×10 5< 1.1×10 -3< 5.44.9-5.91681ND01022.7×10 5< 2.5×10 -3< 9.58.6-10.41463DET1252.7×10 5< 5.2×10 -3< 19.618.6-20.51421IGC1132.3×10 5< 1.1×10 -2< 4745-501534HWE923.5×10 5< 1.9×10 -2< 5450-581442OSL1293.2×10 5< 2.9×10 -2< 9380-1071431NCD1153.1×10 5< 3.1×10 -2< 9983-1181513TT11243.1×10 5< 3.1×10 -2< 10395-1111753U9P1141.8×10 5< 2.0×10 -2< 11087-1391493NTC953.9×10 5< 4.4×10 -2< 113105-1221473GK81131.1×10 5< 1.3×10 -2< 11998-1431523UJJ1528.6×10 4< 1.0×10 -2< 119112-1261503SQO933.4×10 5< 4.1×10 -2< 122104-1431692ADG1142.0×10 5< 2.4×10 -2< 12396-1601743E8U832.8×10 5< 3.3×10 -2< 12687-1831483KYK1054.5×10 5< 7.1×10 -2< 160129-1991702V171121.6×10 5< 6.0×10 -2< 393294-5241733DVN1132.4×10 5< 9.5×10 -2< 413283-601 1< PDB antibody structures of the exogenous CDRH3 loops

[0104] Supplementary Table 8. Structural V H / V L orientation analysis using Abangle 18< . Two reference frame planes are mapped onto Fv structures. V H / V L orientation is described as equivalent to measuring the orientation between the two planes by defining a vector C and three points on each plane as described in 18. Structure HL torsion (°) 1< HC1 bend (°) 2< LC1 bend (°) 3< HC2 bend (°) 4< LC2 bend (°) 5< dc(Å) 6< Fab-LS146 model -56.5071.59123.50118.9479.8016.06scFv-LS146 X-ray structure -66.8971.89120.40117.2981.4816.10 1< torsion angle between H1 and L1; 2< bend angle between H1 and C ; 3< bend angle between H2 and C; 4< bend angle between L1 and C ; 5< bend angle between L2 and C; 6< length of C.

[0105] Supplementary Table 9. List of antibody V-region scaffold structures used in this study for hotspots graft design. Each scaffold is designated as: PDB code + "_" + V H chain ID + V L chain ID. 12e8_HL 15c8_HL 1a14_HL 1a2y_BA 1a31_HL 1a3r_HL 1a4j_EBA 1a4k_BA 1a6t_BA 1a6u_HL1a6v_HL 1a6w_HL 1a7n_HL 1a7o_HL 1a7p_HL 1a7q_HL 1a7r_HL 1acy_HL 1ad0_BA 1ad9_BA1adq_HL 1ae6_HL 1afv_HL 1ahw_BA 1ai1_HL 1aif_BA 1aj7_HL 1ap2_BA 1aqk_HL 1ar1_CD1axs_BA 1axt_HL 1ay1_HL 1b2w_HL 1b4j_HL 1baf_HL 1bbd_HL 1bbj_BA 1bey_HL 1bfo_BA1bfv_HL 1bgx_HL 1bj1_HL 1bln_BA 1bog_EA 1bql_HL 1bvk_BA 1bvl_AB 1bz7_BA 1c08_BA1e12_BA 1c1e_HL 1cSb_HL 1c5c_HL 1c5d_BA 1cbv_HL 1ce1_HL 1cf8_HL 1cfn_BA 1cfq_BA1cfs_BA 1cft_BA 1cfv_HL 1cgs_HL 1cic_BA 1ck0_HL 1cl7_HL 1clo_HL 1cly_HL 1clz_HL1cr9_HL 1ct8_BA 1cu4_HL 1cz8_HL 1d5b_BA 1d5i_HL 1d6v_HL 1dba_HL 1dbb_HL 1dbj_HL1dbk_HL 1dbm_HL 1dee_BA 1dfb_HL 1d17_HL 1dlf_HL 1dn0_BA 1dqd_HL 1dqj_BA 1dql_HL1dqm_HL 1dqq_BA 1dsf_HL 1dvf_BA 1dzb_Aa 1e4w_HL 1e4x_HL 1e6j_HL 1e6o_HL 1eap_BA1egj_HL 1ehl_HL 1ejo_HL 1emt_HL 1eo8_HL 1etz_BA lezv XY 1f11_BA 1f3d_HL 1f3r_Bb1f4w_HL 1f4x_HL 1f4y_HL 1f58_HL 1f8t_HL 1f90_HL 1fai_HL 1fbi_HL 1fdl_HL 1fe8_HL1fgn_HL 1fig_HL 1fj1_BA 1fl3_AB 1fl5_BA 1fl6_BA 1fn4_DC 1fns_HL 1for_HL 1fpt_HL1frg_HL 1fsk_CB 1fvc_BA 1fvd_BA live BA 1g7h_BA 1g7i_BA 1g7j_BA 1g7l_BA 1g7m_BA1g9m_HL 1g9n_HL 1gaf_HL 1gc1_HL 1ggb_HL 1ggc_HL 1ggi_HL 1ghf_HL 1gig_HL 1gpo_HL1h0d_BA 1h3p HL 1h8n_aA 1h8o_Aa 1h8s_Aa 1hez_BA 1hh6_BA 1hh9_BA 1hi6_BA 1hil_BA1him_LH 1hin_HL 1hkl_HL 1hq4_BA 1hys_DC 1hzh_HL 1i3g_HL 1i7z_BA 1i8i_BA 1i8k_BA1i8m_BA 1i9i_HL 1i9j_HL 1i9r_HL 1iai_HL 1ibg_HL 1ic4_HL 1ic5_HL 1ic7_HL 1ifh_HL1igc_1igf_HL 1igi_HL 1igj_BA 1igm_HL 1igt_BA 1igy_BA 1ikf_HL 1il1_AB 1ind_HL1ine_HL 1iqd_BA 1iqw_HL 1it9_HL 1j05_BA 1j1o_HL 1j1p_HL 1j1x_HL 1j5o_HL 1jfq_HL1jgu_HL 1jgv_HL 1jhl_HL 1jn6_BA 1jnh_BA 1jnl_HL 1jnn_HL 1jp5_aA 1jps_HL 1jpt_HL1jrh_HL 1jv5_BA 1k4c AB 1k4d_AB 1k6q_HL 1kb5_HL 1kb9_JK 1kc5_HL 1kcr_HL 1kcs_HL1kcu_HL 1kcv_HL 1keg_HL 1kel_HL 1kem_HL 1ken_HL 1kfa_HL 1kip_BA 1kiq_BA 1kir_BA1kn2_HL 1kn4 HL 1kno_BA 1ktr_HL 1kyo_JK 117i_HL 1l7t_HL 1lk3_HL 1lo0_HL 1lo2_HL1lo3_HL 1lo4_HL 1m71_BA 1m7d_BA 1m7i_BA 1mam_HL 1mco_HL 1mcp_HL 1mex_HL 1mf2_HL1mfa_HL 1mfb_HL lmfc HL 1mfd_HL 1mfe_HL 1mh5_BA 1mhh_BA 1mhp_HL 1mim_HL 1mj8 HL1mjj_BA 1mju_HL 1mlb_BA 1mlc_BA 1mnu_HL 1mpa_HL 1mqk_HL 1mvu_BA 1n0x_HL 1n4x_HL1n5y_HL 1n64_HL 1n6q_HL 1n7m_LH 1n8z_BA 1nak_HL 1nbv_HL 1nby_BA 1nbz_BA 1nc2_BA1nc4_BA 1nca_HL 1ncb_HL 1ncc_HL 1ncd_HL 1ncw_HL 1nd0_BA 1ndg_BA 1ndm_BA 1nfd_FE1ngp_HL 1ngq_HL 1ngw_BA 1ngx_BA 1ngy_BA 1ngz_BA 1nj9_BA 1nl0_HL 1nlb_HL 1nld_HL1nma_HL 1nmb_HL 1nmc_BC 1nsn_HL 1oak_HL 1oaq_HL 1oar_HL 1oau_HL 1oax_HL 1oay_HL1oaz_HL 1ob1_BA 1ocw_HL 1om3_HL 1op3_HL 1op5_HL 1opg_HL 1org_BA 1ors_BA 1osp_HL1ots_CD 1ott_CD 1otu_CD 1p2c_BA 1p4b_HL 1p4i_HL 1p7k_BA 1p84_JK 1pg7_HL 1pkg_BA1plg_HL 1psk_HL 1pz5_BA 1q0x_HL 1q0y_HL 1g1j_HL 1q72_HL 1q9k_BA 1q9l_BA 1q9o_BA1q9w_BA 1qbl_HL 1gbm_HL 1qfu_HL 1qfw_IM 1qkz_HL 1qle_HL 1qlr_BA 1qnz_HL 1qok_aA1qyg_HL 1r0a_HL 1r24_BA 1r3i_HL 1r3j_BA 1r3k_BA 1r3l_BA 1rfd_HL 1rhh_BA 1rih_HL1riu_HL 1riv_HL 1rjl_BA 1rmf_HL 1ru9_HL 1rua_HL 1ruk_HL 1rul_HL 1rum_HL 1rup_HL1ruq_HL 1rur_HL 1rvf_HL 1rz7_HL 1rz8_BA 1rzj_HL 1rzk_HL 1s3k_HL 1s5h_BA 1s5i_HL1s78_DC 1sbs_HL 1seq_HL 1sm3_HL 1svz_aA 1sy6_HL 1t03_HL 1t04_BA 1t2q_HL 1t3f_BA1t4k_BA 1t66_DC 1tet_HL 1tjg_HL 1tjh_HL 1tji_HL 1tpx_BC 1tqb_BC 1tqc_BC 1tzg_HL1tzh_BA 1tzi_BA 1u6a_HL 1u8h_BA 1u8_BA 1u3j_BA 1u8k_BA 1u8l_BA 1u8m_BA 1u8n_BA1u8o_BA 1u8p_BA 1u8q_BA 1u91_BA 1u92_BA 1u93_BA 1u95_BA 1ua6_HL luac_HL 1ub5_AB1ub6_AB 1ucb_HL 1uj3_BA 1um4_HL 1um5_HL 1um6_HL 1uwe_HL 1uwg_HL 1uwx_HL 1uyw_HL1uz6_FE 1uz8_BA 1v7m_HL 1v7n HL 1vfa_BA 1vfb_BA 1vge_HL 1vpo_HL 1w72_HL 1wc7_BA1wcb_BA 1wej_HL 1wt5_AC 1wz1_HL 1x9q_aA 1xcq_BA 1xct_BA 1xf2_BA 1xf3_BA 1xf4_BA1xf5_BA 1xgp_BA 1xgq_BA 1xgr_BA lxgt_BA 1xgu_BA 1xgy_HL 1xiw_DC 1y0l_BA 1y18_BAlyec_HL 1yed_BA 1yee_HL 1yef_HL 1yeg_HL 1yeh_HL 1yej_HL 1yek_HL 1yjd_HL1ymh_BA 1ynk_HL 1ynl_HL 1ynt_BA 1yqv_HL 1yuh_BA 1yy8_BA 1yy9_DC 1yyl_HL 1yym_HL1z3g_HL 1za3_BA 1za6_BA 1zan_HL 1zea_HL 1zls_HL 1zlu_HL 1zlv_MK 1zlw_HL 1ztx_HL1zwi_AB 25c8_HL 2a0l_DC 2a1w_HL 2a6d_BA 2a6i_BA 2a6j_BA 2a6k_BA 2a9m_HL 2a9n_HL2aab_HL 2adf_HL 2adg_BA 2adi_BA 2adj_BA 2aep_HL 2aeq_HL 2agj_HL 2ai0_IM 2aj3_BA2ajs_HL 2aju_HL 2ajv_HL 2ajx_HL 2ajy_HL 2ajz_BA 2ak1_HL 2ap2_BA 2arj_BA 2atk_AB2b0s_HL 2b1a_HL 2b1h_HL 2b2x HL 2b4c HL 2bdn_HL 2bfv HL 2bjm_HL 2bmk_BA 2bob_AB2boc_AB 2brr_HL 2c1o_BA 2c1p_BA 2cgr_HL 2cju_HL 2ck0_HL 2cmr_HL 2d03_HL 2d7t_HL2dbl_HL 2dd8_HL 2ddq_HL 2dlf_HL 2dqc_HL 2dqd_HL 2dge_HL 2dqf_BA 2dqg_HL 2dgh_HL2dqi_HL 2dqj_HL 2dqt_HL 2dqu_HL 2dtg_AB 2dwd_AB 2dwe_AB 2e27_HL 2eh7_HL 2eh8_HL2eiz_BA 2eks_BA 2exw_CD 2exy_CD 2ez0_CD 2f19_HL 2f58_HL 2f5a_HL 2f5b_HL 2fat_HL2fb4 HL 2fbj_HL 2fd6_HL 2fec_IL 2fed_CD 2fee_IL 2fjf_BA 2fjg_BA 2fjh BA 2fl5_BA2fr4_BA 2fx7_HL 2fx8_HL 2fx9_HL 2g2r_BA 2g5b_BA 2g60_HL 2g75_AB 2gcy_BA 2gfb_BA2ghw_bB 2gjj_Aa 2gjz_BA 2gk0 HL 2gki_Aa 2gsg_BA 2gsi_HG 2h1p_HL 2h2p_CD 2h2s_CD2h8p_AB 2h9g_BA 2hfe_AB 2hfg_HL 2hg5_AB 2hh0_HL 2hjf_AB 2hkf_HL 2hkh_HL 2hlf CD2hmi_DC 2hrp_HL 2ht2_CD 2ht3_CD 2ht4_CD 2htk_CD 2htl_CD 2hvj _AB 2hvk_AB 2hwz_HL2i5y_HL 2i60_HL 2i9l_BA 2ibz_XY 2iff_HL 2ig2_HL 2igf_HL 2ih1_AB 2ih3_AB 2ipt_HL2ipu_GK 2iq9_HL 2itc_AB 2itd AB 2j4w_HL 2j5l_CB 2j6e_HL 2j88_HL 2jb5_HL 2jel_HL2jix_DG 2jk5_AB 2kh2_bB 21tq_FE 2mcp_HL 2mpa_HL 2nlj_BA 2nr6_DC 2ntf_BA 2nxy_DC2nxz_DC 2ny0_DC 2ny1_DC 2ny2_DC 2ny3_DC 2ny4_DC 2ny5_HL 2ny6_DC 2ny7_HL 2nyy_DC2nz9_DC 2o5x_HL 2o5y_HL 2o5z_HL 2ojz_HL 2ok0_HL 2op4_HL 2oqj_BA 2or9_HL 2osl_AB2otu_BA 2otw_BA 2oz4_HL 2p7t_AB 2p81_BA 2p3p_BA 2pcp_BA 2pw1_BA 2pw2_BA 2q76_BA2qBa_HL 2q8b_HL 2qhr_HL 2qqk_HL 2qql_HL 2qqn_HL 2qr0_BA 2qsc_HL 2r0k_HL 2r01_HL2r0w_HL 2r0z_HL 2r1w_BA 2r1x_BA 2r1y_BA 2r23_BA 2r29_HL 2r2b_BA 2r2e_BA 2r2h_BA2r4r_HL 2r4s_HL 2r56_HL 2r69_HL 2r8s_HL 2r9h_CD 2rcs_HL 2uud_HL 2uyl_BA 2uzi HL2v17_HL 2v7h_BA 2v7n_BA 2vc2_HL 2vdk_HL 2vdl_HL 2vdm_HL 2vdn_HL 2vdo_HL 2vdp_HL2vdq_HL 2vdr_HL 2vh5_HL 2vir_BA 2vis_BA 2vit_BA 2vl5_AB 2vq1_BA 2vwe_EC 2vxq_HL2vxs_HL 2vxt_HL 2vxu_HL 2vxv_HL 2w0f_AB 2w60_AB 2w65_AB 2w9d_HL 2w9e_HL 2wub_HL2wuc_HL 2x71 AB 2xa8_HL 2xkn_BA 2xqy_GL 2xra_HL 2xtj_DB 2xwt AB 2xza_HL 2xzc_HL2xzq_HL 2y06_HL 2y07_HL 2y36_HL 2y5t_AB 2y6s_DC 2ybr_AB 2yc1_AB 2yk1_HL 2ykl_HL2ypv_HL 2yss_BA 2z4q_BA 2z91_AB 2z92_AB 2zch_HL 2zck_HL 2zcl_HL 2zjs_HL 2zkh_HL2zpk_HL 2zuq_FE 32c2_BA 35c8 HL 3a67_HL 3a6b_HL 3a6c_HL 3aaz_AB 3ab0_BC 3auv_Aa3b2u_CD 3b2v HL 3b9k_DC 3bae_HL 3bdv_HL 3be1_HL 3bgf_BC 3bkc_HL 3bkj_HL 3bkm_HL3bky_HL 3bn9_DC 3bpc_BA 3bqu_DC 3bsz_HL 3bt2_HL 3bz4_BA 3c09_CB 3c2a_HL 3c5s_DC3c6s_BA 3cfb_BA 3cfc_HL 3cfd_BA 3cfe_BA 3cfj_BA 3cfk_BA 3ck0_HL 3cle_HL 3clf_HL3cmo_HL 3cvh_HL 3cvi_HL 3cx5_JK 3cxd_HL 3cxh_JK 3d0v_BA 3d69_BA 3d85_BA 3d9a_HL3det_CD 3dgg_BA 3dif_BA 3dsf_HL 3dur_BA 3dus_BA 3duu_BA 3dv4 BA 3dv6_BA 3dvg_BA3dvn_BA 3e8u_HL 3efd_HL 3eff BA 3ehb_CD 3ejy_CD 3ejz_CD 3eo0_BA 3eo1_BA 3eo9_HL3eoa_BA 3eob_BA 3eot_HL 3esu_fF 3esv_Ff 3et9_Ff 3etb_Ff 3eyf_BA 3eyo_DC 3eys_HL3eyu_HL 3eyv BA 3f58_HL 3f5w_AB 3f7v_AB 3f7y_AB 3fb5_AB 3fb6_AB 3fb7_AB 3fb8 AB3fct BA 3ffd_AB 3fku_Ss 3fmg_HL 3fn0_HL 3fo0_HL 3fo1_BA 3fo2_BA 3fo9_BA 3fzu_CD3g04_BA 3g5v_BA 3g5x_BA 3g5y_BA 3g5z_BA 3g6a_BA 3g6d_HL 3g6j_FE 3gb7_AB 3gbn_HL3ggw_BA 3ghb_HL 3ghe_HL 3gi8_HL 3gi9_HL 3giz_HL 3gje_BA 3gjf_HL 3gk8_HL 3gkw_HL3gkz_aA 3gm0_Aa 3gnm_HL 3go1_HL 3grw_HL 3h0t_BA 3h3b_cC 3h42_HL 3hae_HL 3hb3_CD3hc0_AB 3hc3_HL 3hc4_HL 3hfm HL 3hi1_BA 3hi5_HL 3hi6_HL 3hmw HL 3hmx_HL 3hns HL3hnt_HL 3hnv_HL 3hpl_AB 3hr5_BA 3hzk_BA 3hzm_BA 3hzv_BA 3hzy_BA 3i02_BA 3i2c_HL3i50_HL 3i75_BA 3i9g_HL 3idg_BA 3idj_BA 3idm_BA 3idn_BA 3idx_HL 3idy_BC 3iet_BA3if1_BA 3ifl_HL 3ifn_HL 3ifo_AB 3ifp_AB 3iga_AB 3ijh_BA 3ijs_BA 3ijy_BA 3ikc SA3inu_HL 3iu3_AB 3ivk AB 3ixt_AB 3iy0_HL 3iy1_BA 3iy2_BA 3iy3_BA 3iy4_BA 3iy5_BA3iy6_BA 3iy7_BA 3iyw_HL 3j1s_HL 3j2x_BA 3j2y_BA 3j2z_BA 3j30_BA 3juy_Aa 3jwd_HL3jwo_HL 3k2u_HL 3kdm_BA 3kj4_CB 3kj6_HL 3klh_DC 3kr3 HL 3ks0_HL 3kyk_HL 3kym_BA3l1o_HL 315w BA 315x_HL 315y_HL 317e_BA 317f_BA 3195_BA 3ld8_CB 3ldb_CB 3lev_HL3lex_AB 3ley_HL 31h2_JN 3liz_HL 3lmj_HL 3loh_AB 3lqa_HL 3ls4_HL 3ls5_HL 3lzf_HL3m8o_HL 3ma9_HL 3mac_HL 3mbx_HL 3mck_BA 3mcl_HL 3mj8_BA 3mj9_HL 3mlr_HL 3mlu_HL3mlw_HL 3mlx_HL 3mly_HL 3mlz_HL 3mme_AB 3mnv_BA 3mnw_BA 3mnz_BA 3mo1_BA 3moa_HL3mob_EL 3mod_HL 3mxv_HL 3mxw_HL 3n85_HL 3n9g_HL 3na9_HL 3naa_HL 3nab_HL 3nac_HL3ncj_HL 3ncy_PS 3nfp_AB 3nfs_HL 3ngb_BC 3nh7_HL 3nid_EF 3nif_EF 3nig_EF 3nn8_AB3nps_BC 3ntc_HL 3nz8__AB 3nzh_HL 3o0r_HL 3ol1_BA 3o2d_HL 3o2v_HL 3o2w_HL 3o41_AB3o45_AB 3o6k_HL 3o6l_HL 3o6m_HL 3oau_HL 3oay_HL 3oaz_HL 3ob0_HL 3ogc_AB 3ojd_BA3okd_BA 3oke_BA 3okk_BA 3ok1_BA 3okm_BA 3okn_BA 3oko_BA 3opz_IM 3or6_AB 3or7_AB3oz9_HL 3p0v_HL 3p0y_HL 3p11_HL 3p30_HL 2pgf_HL 3pho_BA 3phq_BA 3piq_CD 2pjs_BA3pnw_BA 3pp3_HL 3pp4_HL 3g1s_HL 3q3g_BA 3q6g_HL 3qa3_BA 3qct_HL 3qcu_HL 3gcv_HL3qeh_AB 3qg6_BA 3qg7_HL 3qhf_HL 3qnx_BA 3qo0_BA 3qo1_BA 3qos_HL 3qot_HL 3qpq_DC3qpx_HL 3qq9_DC 3qrg_HL 3qum_BA 3qwo_AB 3r06_BA 3r08_HL 3rlg_HL 3ra7_HL 3raj_HL3rhw_FN 3ri5 FN 3ria_FN 3rif_FN 3rkd_DC 3ru8_HL 3rvt_DC 3rvu_DC 3rvv_DC 3rvw_DC3rvx_DC 3s34_HL 3s35_HL 3s36_HL 3s37_HL 3s62_HL 3s88_HL 3s96_AB 3sdy_HL 3se8_HL3se9_HL 3sgd_HL 3sge_HL 3skj_HL 3sm5_HL 3so3_CB 3sob_HL 3sqo_HL 3stl_AB 3stz_AB3sy0_BA 3t3m EF 3t3p_EF 3t4y_BA 3t65_BA 3t77 BA 3tcl_AB 3tnm HL 3tnn AB 3tt1_HL3u0t_BA 3u0w_HL 3u30_CB 3u46_AB 3u4b_HL 3u6r_AB 3u7w_HL 3u7y_HL 3u9p_HL 3u9u_AB3uaj_CD 3ubx_GI 3uc0_HL 3uji_HL 3ujj_HL 3ujt_HL 3uls_BA 3ulu DC 3ulv_DC 3umt_Aa3uo1_HL 3utz_BA 3ux9_Bb 3uyp_Aa 3uyr_HL 2uze_Aa 3uzq_aA 3uzv_Bb 3v0v_AB 3v0w_HL3v4p_HL 3v4u_HL 3v4v_HL 3v52_HL 3v6f_AB 3v6o_CE 3v6z_AB 3v7a_EH 3ve0_AB 3vfg_HL3vg0_HL 3vg9_CB 3vga_CB 3vi3_FE 3vi4_FE 3vrl_EF 3vw3_HL 3w11_CD 3w12_CD 3w13_CD3w14_CD 3zdx_EF 3zdy_EF 3zdz EF 3ze0_EF 3ze1_EF 3ze2_EF 3zkm_CD 3zkn_CD 3ztj_GH3ztn_HL 43c9_BA 43ca_BA 4a6y_BA 4aeh_HL 4aei_HL 4ag4_HL 4al8_HL 4ala_HL 4am0_AB4amk_HL 4at6_AB 4d91_HL 4d9q_ED 4d9r_ED 4dag_HL 4dcq_BA 4dgi_HL 4dgv_HL 4dgy_HL4dke_HL 4dkf HL 4dn3 HL 4dn4 HL 4dtg_HL 4dvb_AB 4dvr_HL 4dw2_HL 4ebq_HL 4ene_CD4eow_HL 4ers_HL 4etq AB 4evn AB 4f2m_AB 4f33 BA 4f37_FK 4f3f BA 4f57 HL 4f58_HL4f9l_cC 4f9p_cC 4fab_HL 4ffv_DC 4ffw_DC 4ffy HL 4ffz_HL 4fg6_CD 4fnl_HL 4fp8_HL4fq1_HL 4fq2_HL 4fqc_HL 4fgh_AB 4fqi_HL 4fqj_HL 4fqk_EF 4fql_HL 4fqq_BA 4fqr_ab4fqv_HL 4fqy_HL 4g3y_HL 4g5z_HL 4g6a_CD 4g6f_BD 4g6j_HL 4g6k_HL 4g6m_HL 4gag_HL4gay_HL 4gms_HL 4gmt_HL 4gw4_AB 4gxu_MN 4gxv_HL 4h0g_Aa 4h0h_bB 4h0i_aA 4h20_HL4hbc_HL 4hc1 HL 4hcr_HL 4hdi_BA 4hf5 HL 4hfu_HL 4hfw BA 4hg4_JK 4hgw_BA 4hix_HL4hj0_CD 4hk0_CD 4hk3_JN 4hlz_GH 4hpo_HL 4hpy_HL 4hs6_BA 4hs8_HL 4ht1_HL 4hwb_HL4hwe_HL 4hzl_AB 4i3r_HL 4i3s_HL 4i77_HL 4i9w_ED 4idj_HL 4imk_AD 4iml_AB 4j1u_DC4j6r_HL 4j8r_BA 4jam_HL 4jan_AB 4jb9_HL 4jdv_AB 4jha_HL 4jhw_HL 4jkp_HL 4jm2_AB4jm4_HL 4jn1_HL 4jn2_HL 4jpi_HL 4jpk_HL 4jpw_HL 4jqi_HL 4jr9_HL 4jre_BC 4jy4_BA4jy5_HL 4jy6_BA 4jzn_IP 4jzo_AB 4k2u_HL 4k3d_HL 4k3e_IM 4k4m_HL 4k8r_DC 6fab_HL7fab_HL 8fab_BA 2ymx_HL 3mls_HL 3mlv_HL 3t2n_HL 3w9d_AB 3w9e_AB 3wbd_aA 3wd5_HL4fz8_HL 4fze_HL 4gq9_HL 4gsd_HL 4gw1_BA 4gw5_BA 4h88_HL 4hh9_BA 4hha_BA 4hie_BA4hih_BA 4hii BA 4hij_BA 4hjg_BA 4hkz_BA 4hxa_HL 4hxb_HL 4iof_EF 4ioi_BA 4irz_HL4jfx_HL 4jfy_HL 4jfz_HL 4jo1_HL 4jo2_HL 4jo3_HL 4jo4_HL 4jpv_HL 4k3j_HL 4k7p_HL4k94_HL 4k9e_HL 4kjp_CD 4kjq_CD 4kjw_CD 4kk5_CD 4kk6_CD 4kk8_CD 4kk9_CD 4kka_CD4kkb_CD 4kkc_CD 4kkl_CD 4kro_DC 4krp_DC 4kuc_FE 4kvc_HL 4ky1_HL 4lbe_AB 4lcu_AB4leo_AB 4lkc_BA 4llv_HL 4lmg_HL 4lou_CD 4lss_HL 4lst_HL 41su HL 4lsv_HL 4m1d_HL4m43_HL 4m48_HL 4m5y_HL 4m5z_HL 4mhh_HL 4mhj_wv 4msw_AB

[0106] Supplementary Table 10. Amino acid sequences of Keapl and LS146-scFv constructs used for crystallisation. Protein construct Sequence Keap1 (Kelch 1-6 domains, AA 314-611)LS146-scFv

[0107] Supplementary Table 11. Crystallography data collection and structure refinement statistics. LS146-scFv / Keap1Data collection Space groupP2 1 Cell dimensionsa, b, c (Å)70.5, 69.8, 99.6α, β, γ (°)90.0, 90..2, 90.0Resolution (Å)29.69-1.85R sym or R merge 0.049I / σI11.2Completeness (%)99.1%Redundancy3.1Refinement Resolution (Å)1.85No. reflections265541R work / R free 22.1 / 26.1No. atomsProtein7905Ligand / ion0Water532B-factorsProtein25.96Ligand / ionN / AWater29.41R.m.s deviationsBond lengths (Å)0.013Bond angles (°)1.52

[0108] Supplementary Table 12. Binding affinities of the ordered antibody Fab designs from hotspots graft. Dissociation constants (K D ) were determined by SPR. Design Scaffold 1< Hotspots positions TGFβ1 K D (nM) TGFβ2 K D (nM) TGFβ3 K D (nM) 1843MXWV L 52I, V L 54V, V L 56I, V H 100L106Low binding32.91863NACV L 52I, V L 54V, V L 56I, V H 100 B< LNDNDND1873OB0V H 33I, V L 93L, V L 94VNDNDND

[0109] Supplementary Table 13. Fv regions' amino acid sequences of ordered antibody designs from hotspots graft. Design Sequence V H V L 184186187

[0110] Supplementary Table 14. Receptors' pan-blocking IC 50 s of Fab 184 design from hotspots graft in the reporter gene assay (n = 2). Design TGFβ1 IC 50 (nM) TGFβ2 IC 50 (nM) TGFβ3 IC 50 (nM) 18452.836.810.6 PSEUDO-CODESPseudo codes of hotspots grafting onto antibody scaffold structures:

[0111] Pseudo codes of CDRH3 loop swapping of G54.1:

[0112] RosettaScripts: AlaScan.xml:

[0113] <dock_design> <FILTERS> <AlaScan name=scan partner1=0 partner2=1 scorefxn=score12 interface_distance_cutoff=8.0 repeats=3 / > < / FILTERS> <MOVERS> <RepackMinimize name=intermin scorefxn _repack=score12 scorefxn_minimize=score12 interface_cutoff_distance=8.0 repack_partner1=0 repack_partner2=0 design_partner1=0 design_partner2=0 minimize_bb=0 minimize_sc=1 minimize_rb=0 / > < / MOVERS> <PROTOCOLS> <Add mover_name=intermin / > <Add filter_name=scan / > < / PROTOCOLS> < / dock_design>RosettaScripts: InverseRotamers.xml:

[0114] <dock_design> <FILTERS> <EnergyPerResidue name=energy pdb_num=79B energy _cutoff=0.0 / > <Ddg name=ddg threshold=-1.0 / > < / FILTERS> <MOVERS> <TryRotamers name=try pdb_num=79B / > < / MOVERS> <PROTOCOLS> <Add mover_name=try / > <Add filter_name=energy / > <Add filter_name=ddg / > < / PROTOCOLS> < / dock_design>RosettaScripts: ppk.xml:

[0115] <dock_design> <MOVERS> <Prepack name=ppk jump _number=0 scorefxn=score12 / > Jump_number=0 to prepack the entire structure without moving the partners apart. <MinMover name=min scorefxn=score 12 chi=1 bb=0 jump=0 / > < / MOVERS> <PROTOCOLS> <Add mover_name=ppk / > <Add mover_name=min / > < / PROTOCOLS> < / dock_design>RosettaScripts: MutationScanPB.xml:

[0116] <dock_design> <SCOREFXNS> <local_score weights=score12_full patch="pb_elec.wts_patch" / > <local_score_soft weights=soft_rep patch="pb_elec.wts_patch" / > < / SCOREFXNS> <TASKOPERATIONS> <InitializeFromCommandline name=init / > <ProteinInterfaceDesign name=pid repack_chain1=1 repack_chain2=1 design_chain1=0 design _chain2=1 interface_distance_cutoff=8 / > <ProteinInterfaceDesign name=pio repack _chain1=1 repack _chain2=1 design_chain1=0 design _chain2=0 interface_distance_cutoff=8 / > < / TASKOPERATIONS> <MOVERS> <AtomTree name=docking_tree docking_ft=1 / > <MinMover name=min_sc scorefxn=local_score bb=0 chi=1 jump=1 / > minimize sc, rb <PackRotamersMover name=pack_interface scorefxn=local_score task_operations=init,pio / > <PackRotamersMover name=pack_interface_soft scorefxn=local_score_soft task_operations=init,pio / > <ParsedProtocol name=relax_before_baseline> <Add mover-docking_tree / > <Add mover=pack_interface / > <Add mover= min_sc / > < / ParsedProtocol> < / MOVERS> <FILTERS> <Ddg name=ddg scorefxn=local_score confidence=0.0 / > <Delta name=delta_ddg filter=ddg upper=1 lower=0 range=-0.5 relax_mover=relax_before_baseline / > <FilterScan name=scan_binding scorefxn=local_score relax_mover=relax_before_baseline task_operations=pid,init filter=delta_ddg triage_filter=delta_ddg resfile_name="scan.resfile" / > <Time name=scan_binding_timer / > < / FILTERS> <PROTOCOLS> <Add mover-docking_tree / > <Add filter=scan_binding_timer / > <Add filter=scan_binding / > <Add filter=scan_binding_timer / > < / PROTOCOLS> < / dock_design> RosettaScripts: FlexbbInterfaceDesign.xml:

[0117] <dock_design> <TASKOPERATIONS> <ProteinInterfaceDesign name=pio repack _chain1=1 repack _chain2=1 design_chain1=0 design_chain2=0 interface_distance_cutoff=10 / > <ReadResfile name=resfile filename="design.resfile" / > < / TASKOPERATIONS> <FILTERS> <Ddg name=ddG scorefxn=score12 threshold=-20 repeats=2 / > <Sasa name=sasa threshold=2000 / > <CompoundStatement name=ddg_sasa> <AND filter_name=ddG / > <AND filter_name=sasa / > < / CompoundStatement> < / FILTERS> <MOVERS> <BackrubDD name=backrub partner1=0 partner2=1 interface_distance_cutoff=8.0 moves=1000 sc_move_probability=0.25 scorefxn=score12 small_move_probability=0.15 bbg_move_probability=0.25 task_operations=pio / > <RepackMinimize name=des1 scorefxn_repack=soft_rep scorefxn_minimize=soft_rep minimize_bb=0 minimize_rb=1 task_operations=resfile / > <RepackMinimize name=des2 scorefxn_repack=score12 scorefxn_minimize=score12 minimize_bb=0 minimize_rb=1 task_operations=resfile / > Design & minimization at the interface <RepackMinimize name=des3 minimize_bb=1 minimize_rb=0 task_operations=resfile / > <ParsedProtocol name=design> <Add mover _name=backrub / > <Add mover_name=des1 / > <Add mover_name=des2 / > <Add mover_name=des3 filter_name=ddg_sasa / > < / ParsedProtocol> <GenericMonteCarlo name=iterate scorefxn_name=score12 mover_name=design trials=3 / > <InterfaceAnalyzerMover name=IAM scorefxn=score 12 packstat=1 interface_sc=1 pack _input=1 pack_separated=1 tracer=0 fixedchains=H,L / > < / MOVERS> <PROTOCOLS> <Add mover=iterate / > <Add mover=IAM / > < / PROTOCOLS> < / dock_design >

Claims

1. A computer-implemented method of designing an antibody that will bind to a target epitope, comprising: a) identifying three or more residues from a cognate protein binder known to bind to the target epitope, wherein said cognate protein binder residues are predicted to provide an interaction with the target epitope consistent with providing a disproportional amount of a binding energy between an antibody comprising the residues and the target epitope, each cognate protein binder residue comprising a cognate protein binder residue sub-structure comprising sub-structure characteristic atoms, wherein said characteristic atoms of the cognate protein binder residue sub-structure comprise three or more of the following: the alpha carbon, the backbone carbon atom derived from the carboxyl group, the backbone nitrogen, the backbone oxygen, and the beta carbon of the side chain; b) selecting from a database of antibody structures one or more candidate antibody structures, each candidate antibody structure having first matching residue, a second matching residue and a third matching residue, each comprising a matching residue sub-structure comprising matching residue sub-structure characteristic atoms, wherein said characteristic atoms of the matching residue sub-structure comprise three or more of the following: the alpha carbon, the backbone carbon atom derived from the carboxyl group, the backbone nitrogen, the backbone oxygen, and the beta carbon of the side chain, wherein the selection is performed such that the relative positions of the matching residue sub-structure characteristic atoms within the antibody structure and the relative positions of the cognate protein binder sub-structure characteristic atoms when bound to the target epitope are such that at least three of the matching residue sub-structure characteristic atoms can be superimposed computationally on a corresponding at least three cognate protein binder sub-structure characteristic atoms with a spatial deviation between each pair of superimposed characteristic atoms averaged over all pairs being less than a predetermined threshold; wherein the matching residue is a residue of the same or different amino acid as the cognate protein binder residue; and c) either generating a designed antibody structure by modifying one of the candidate antibody structures, the modifying comprising replacing at least one of the matching residues with a different residue such that a predicted affinity between the designed antibody structure and the target epitope is higher than a predicted affinity between the candidate antibody structure and the target epitope, or outputting one of the candidate antibody structures as a designed antibody structure in the case where each of the matching residues is already a residue of the same amino acid as the cognate protein binder residue which the matching residue matches.

2. The method of claim 1, wherein the predetermined threshold is 2.0 Angstroms.

3. The method of claim 1 or 2, wherein in step (c) the modifying comprises replacing each of the matching residues with a residue of the same amino acid as the cognate protein binder residue which the matching residue matches.

4. The method of claim 1, wherein the at least three of the matching residue sub-structure characteristic atoms that can be superimposed on the corresponding at least three cognate protein binder residue sub-structure characteristic atoms comprise the alpha atom of the matching residue and at least two of the backbone carbon derived from the carboxyl group of the matching residue, the backbone nitrogen of the matching residue, the backbone oxygen of the matching residue, and the beta carbon of the side chain of the matching residue.

5. The method of claim 1, wherein said three of the matching residue sub-structure characteristic atoms comprise the alpha carbon atom, the backbone carbon atom and the backbone nitrogen atom.

6. The method of any one of claims 1-5, wherein the selection of the one or more candidate antibody structures from the database further comprises the following steps: determining a first set of distances representing separations between all possible pairings between identical characteristic atoms in different sub-structures of the cognate protein binder residues; determining a second set of distances representing separations between all possible pairings between identical characteristic atoms in different sub-structures of the matching residues; and comparing the first set of distances to the second set of distances for different antibody structures until a match is obtained within a predetermined separation threshold.

7. The method of any one of claims 1-5, wherein the generating of the designed antibody structure comprises detecting geometrical clashing, where one or more atoms are predicted to occupy positions that are closer together than is physically possible, between one or more atoms in the candidate antibody structure when bound to the target epitope and one or more atoms in the target epitope.

8. The method of claim 7, further comprising determining whether a detected geometrical clash is with a side chain of a residue of the candidate antibody structure and, if so, modifying the side chain.

9. The method of claim 8, wherein the side chain is modified to an alanine side chain, a glycine side chain, a valine side chain, a serine side chain, a threonine side chain, or homo-alanine side chain.

10. The method of any one of claims 7-9, further comprising determining whether a detected geometrical clash is with a backbone or beta carbon atoms of any candidate antibody residue and, if so, discarding the selected candidate antibody structure and repeating the determination for a different candidate antibody structure selected from the database.

11. The method of any one of claims 1-10, wherein the generating of the designed antibody structure further comprises iteratively mutating the amino acid types of residues in the candidate antibody structure to increase a predicted affinity between the designed antibody structure and the target epitope.

12. The method of claim 11, wherein the selection of residues that are iteratively mutated is constrained so that the cognate protein binder residues are not mutated.

13. The method of claim 11, wherein the selection of residues that are iteratively mutated is not constrained to avoid mutation of the cognate protein binder residues.

14. The method of any one of claims 1-13, wherein the generating of the designed antibody structure further comprises iteratively swapping each of one or more of the CDR loops of the candidate antibody structure with CDR loops from a database of CDR loops to increase a predicted affinity between the candidate antibody structure and the target epitope.

15. The method of any one of claims 1-14, wherein the designed antibody structure is output in any antibody format, including IgG, Fab, Fv, scFv.

16. A method of manufacturing an antibody, comprising: designing an antibody using the method of any one of claims 1-15; and manufacturing the antibody thus designed.