Prediction of hypervariable CDR-H3 loop structures in antibodies

Homology Modeling of Antibody Variable Regions: Methods and Applications

Adaptive immunity specifically protects us from antigenic challenges. Antibodies are key effector proteins of adaptive immunity, and they are remarkable in their ability to recognize a virtually limitless number of antigens. Fragment variable (F_V), the antigen-binding region of antibodies, can be split into two main components, namely, framework and complementarity determining regions. The framework (FR) consists of light-chain framework (FR_L) and heavy-chain framework (FR_H). Similarly, the complementarity determining regions (CDRs) comprises of light-chain CDRs 1-3 (CDRs L1-3) and heavy-chain CDRs 1-3 (CDRs H1-3). While FRs are relatively constant in sequence and structure across diverse antibodies, sequence variation in CDRs leading to differential conformations of CDR loops accounts for the distinct antigenic specificities of diverse antibodies. The conserved structural features in FRs and conformity of CDRs to a limited set of standard conformations allow for the accurate prediction of F_V models using homology modeling techniques. Antibody structure prediction from its amino acid sequence has numerous important applications including prediction of antibody-antigen interaction interfaces and redesign of therapeutically and biotechnologically useful antibodies with improved affinity. This chapter summarizes the current practices employed in the successful homology modeling of antibody variable regions and the potential applications of the generated homology models.

Antibody complementarity determining region design using high-capacity machine learning

Motivation

The precise targeting of antibodies and other protein therapeutics is required for their proper function and the elimination of deleterious off-target effects. Often the molecular structure of a therapeutic target is unknown and randomized methods are used to design antibodies without a model that relates antibody sequence to desired properties.

Results

Here, we present Ens-Grad, a machine learning method that can design complementarity determining regions of human Immunoglobulin G antibodies with target affinities that are superior to candidates derived from phage display panning experiments. We also demonstrate that machine learning can improve target specificity by the modular composition of models from different experimental campaigns, enabling a new integrative approach to improving target specificity. Our results suggest a new path for the discovery of therapeutic molecules by demonstrating that predictive and differentiable models of antibody binding can be learned from high-throughput experimental data without the need for target structural data.

Availability and implementation

Sequencing data of the phage panning experiment are deposited at NIH’s Sequence Read Archive (SRA) under the accession number SRP158510. We make our code available at https://github.com/gifford-lab/antibody-2019.

Supplementary information

Supplementary data are available at Bioinformatics online.

Length-independent structural similarities enrich the antibody CDR canonical class model

Complementarity-determining regions (CDRs) are antibody loops that make up the antigen binding site. Here, we show that all CDR types have structurally similar loops of different lengths. Based on these findings, we created length-independent canonical classes for the non-H3 CDRs. Our length variable structural clusters show strong sequence patterns suggesting either that they evolved from the same original structure or result from some form of convergence. We find that our length-independent method not only clusters a larger number of CDRs, but also predicts canonical class from sequence better than the standard length-dependent approach.

To demonstrate the usefulness of our findings, we predicted cluster membership of CDR-L3 sequences from 3 next-generation sequencing datasets of the antibody repertoire (over 1,000,000 sequences). Using the length-independent clusters, we can structurally classify an additional 135,000 sequences, which represents a ∼20% improvement over the standard approach. This suggests that our length-independent canonical classes might be a highly prevalent feature of antibody space, and could substantially improve our ability to accurately predict the structure of novel CDRs identified by next-generation sequencing.

Computational design of an epitope-specific Keap1 binding antibody using hotspot residues grafting and CDR loop swapping

Therapeutic and diagnostic applications of monoclonal antibodies often require careful selection of binders that recognize specific epitopes on the target molecule to exert a desired modulation of biological function. Here we present a proof-of-concept application for the rational design of an epitope-specific antibody binding with the target protein Keap1, by grafting pre-defined structural interaction patterns from the native binding partner protein, Nrf2, onto geometrically matched positions of a set of antibody scaffolds. The designed antibodies bind to Keap1 and block the Keap1-Nrf2 interaction in an epitope-specific way. One resulting antibody is further optimised to achieve low-nanomolar binding affinity by in silico redesign of the CDRH3 sequences. An X-ray co-crystal structure of one resulting design reveals that the actual binding orientation and interface with Keap1 is very close to the design model, despite an unexpected CDRH3 tilt and V_H/V_L interface deviation, which indicates that the modelling precision may be improved by taking into account simultaneous CDR loops conformation and V_H/V_L orientation optimisation upon antibody sequence change. Our study confirms that, given a pre-existing crystal structure of the target protein-protein interaction, hotspots grafting with CDR loop swapping is an attractive route to the rational design of an antibody targeting a pre-selected epitope.

Antibody recognition of carbohydrate epitopes†

Carbohydrate antigens are valuable as components of vaccines for bacterial infectious agents and human immunodeficiency virus (HIV), and for generating immunotherapeutics against cancer. The crystal structures of anti-carbohydrate antibodies in complex with antigen reveal the key features of antigen recognition and provide information that can guide the design of vaccines, particularly synthetic ones. This review summarizes structural features of anti-carbohydrate antibodies to over 20 antigens, based on six categories of glyco-antigen: (i) the glycan shield of HIV glycoproteins; (ii) tumor epitopes; (iii) glycolipids and blood group A antigen; (iv) internal epitopes of bacterial lipopolysaccharides; (v) terminal epitopes on polysaccharides and oligosaccharides, including a group of antibodies to Kdo-containing Chlamydia epitopes; and (vi) linear homopolysaccharides.

The origin of CDR H3 structural diversity

Antibody complementarity determining region (CDR) H3 loops are critical for adaptive immunological functions. Although the other five CDR loops adopt predictable canonical structures, H3 conformations have proven unclassifiable, other than an unusual C-terminal "kink" present in most antibodies. To determine why the majority of H3 loops are kinked and to learn whether non-antibody proteins have loop structures similar to those of H3, we searched a set of 15,679 high-quality non-antibody structures for regions geometrically similar to the residues immediately surrounding the loop. By incorporating the kink into our search, we identified 1,030 H3-like loops from 632 protein families. Some protein families, including PDZ domains, appear to use the identified region for recognition and binding. Our results suggest that the kink is conserved in the immunoglobulin heavy chain fold because it disrupts the β-strand pairing at the base of the loop. Thus, the kink is a critical driver of the observed structural diversity in CDR H3.

A complete, multi-level conformational clustering of antibody complementarity-determining regions

Classification of antibody complementarity-determining region (CDR) conformations is an important step that drives antibody modelling and engineering, prediction from sequence, directed mutagenesis and induced-fit studies, and allows inferences on sequence-to-structure relations. Most of the previous work performed conformational clustering on a reduced set of structures or after application of various structure pre-filtering criteria. In this study, it was judged that a clustering of every available CDR conformation would produce a complete and redundant repertoire, increase the number of sequence examples and allow better decisions on structure validity in the future. In order to cope with the potential increase in data noise, a first-level statistical clustering was performed using structure superposition Root-Mean-Square Deviation (RMSD) as a distance-criterion, coupled with second- and third-level clustering that employed Ramachandran regions for a deeper qualitative classification. The classification of a total of 12,712 CDR conformations is thus presented, along with rich annotation and cluster descriptions, and the results are compared to previous major studies. The present repertoire has procured an improved image of our current CDR Knowledge-Base, with a novel nesting of conformational sensitivity and specificity that can serve as a systematic framework for improved prediction from sequence as well as a number of future studies that would aid in knowledge-based antibody engineering such as humanisation.

Antibody modeling assessment II. Structures and models

To assess the state-of-the-art in antibody structure modeling, a blinded study was conducted. Eleven unpublished Fab crystal structures were used as a benchmark to compare Fv models generated by seven structure prediction methodologies. In the first round, each participant submitted three non-ranked complete Fv models for each target. In the second round, CDR-H3 modeling was performed in the context of the correct environment provided by the crystal structures with CDR-H3 removed. In this report we describe the reference structures and present our assessment of the models. Some of the essential sources of errors in the predictions were traced to the selection of the structure template, both in terms of the CDR canonical structures and VL/VH packing. On top of this, the errors present in the Protein Data Bank structures were sometimes propagated in the current models, which emphasized the need for the curated structural database devoid of errors. Modeling non-canonical structures, including CDR-H3, remains the biggest challenge for antibody structure prediction.

Computer-aided antibody design

Recent clinical trials using antibodies with low toxicity and high efficiency have raised expectations for the development of next-generation protein therapeutics. However, the process of obtaining therapeutic antibodies remains time consuming and empirical. This review summarizes recent progresses in the field of computer-aided antibody development mainly focusing on antibody modeling, which is divided essentially into two parts: (i) modeling the antigen-binding site, also called the complementarity determining regions (CDRs), and (ii) predicting the relative orientations of the variable heavy (V(H)) and light (V(L)) chains. Among the six CDR loops, the greatest challenge is predicting the conformation of CDR-H3, which is the most important in antigen recognition. Further computational methods could be used in drug development based on crystal structures or homology models, including antibody-antigen dockings and energy calculations with approximate potential functions. These methods should guide experimental studies to improve the affinities and physicochemical properties of antibodies. Finally, several successful examples of in silico structure-based antibody designs are reviewed. We also briefly review structure-based antigen or immunogen design, with application to rational vaccine development.

Simulated mutagenesis of the hypervariable loops of a llama VHH domain for the recovery of canonical conformations

In this work, wildtype and mutated hypervariable regions of an anti-hCG llama VHH antibody were simulated via a molecular dynamics replica exchange method (REM). Seven mutants were simulated with the goal of identifying structural determinants that return the noncanonical H1 loop of the wildtype antibody to the type 1 canonical structure predicted by database methods formulated for conventional antibodies. Two cases with three point mutations yielded a stable type 1 H1 structure. In addition, other mutants with fewer mutations showed evidence of such conformations. Overall, the mutagenesis results suggest a marked influence of interloop interactions on the attainment of canonical conformations for this antibody. On the methodological front, a novel REM scheme was developed to quickly screen diverse mutants based on their relative propensities for attaining favorable structures. This multimutant REM (MMREM) was used to successfully identify mutations that stabilize a canonical H1 loop grafted on the llama antibody scaffold. The use of MMREM and REM for screening mutants and assessing structural stability may be useful in the rational design of antibody hypervariable loops.

Hunting for "key residues" in the modeling of globular protein folding: an artificial neural network-based approach

An approach to modeling globular protein folding based on artificial neural networks (ANNs) is presented. This approach, that can be regarded as an inverse protein folding problem, investigates whether and when a protein fragment needs a specific residue in the center of its primary structure as a necessary condition to fold as observed. To perform this analysis, an ANN has been trained on a set of 55 proteins, searching for a relation between protein fragments modeled by 13alpha torsion angles and the residue corresponding to the central alpha torsion angle of the fragment. The results obtained show that only Asp, Gly, Pro, Ser and Val residues are often a necessary, even though not sufficient, condition to obtain a specific folded fragment structure, playing therefore, the role of "key residue" of this fragment.

Modeling of loops in protein structures

Comparative protein structure prediction is limited mostly by the errors in alignment and loop modeling. We describe here a new automated modeling technique that significantly improves the accuracy of loop predictions in protein structures. The positions of all nonhydrogen atoms of the loop are optimized in a fixed environment with respect to a pseudo energy function. The energy is a sum of many spatial restraints that include the bond length, bond angle, and improper dihedral angle terms from the CHARMM-22 force field, statistical preferences for the main-chain and side-chain dihedral angles, and statistical preferences for nonbonded atomic contacts that depend on the two atom types, their distance through space, and separation in sequence. The energy function is optimized with the method of conjugate gradients combined with molecular dynamics and simulated annealing. Typically, the predicted loop conformation corresponds to the lowest energy conformation among 500 independent optimizations. Predictions were made for 40 loops of known structure at each length from 1 to 14 residues. The accuracy of loop predictions is evaluated as a function of thoroughness of conformational sampling, loop length, and structural properties of native loops. When accuracy is measured by local superposition of the model on the native loop, 100, 90, and 30% of 4-, 8-, and 12-residue loop predictions, respectively, had <2 A RMSD error for the mainchain N, C(alpha), C, and O atoms; the average accuracies were 0.59 +/- 0.05, 1.16 +/- 0.10, and 2.61 +/- 0.16 A, respectively. To simulate real comparative modeling problems, the method was also evaluated by predicting loops of known structure in only approximately correct environments with errors typical of comparative modeling without misalignment. When the RMSD distortion of the main-chain stem atoms is 2.5 A, the average loop prediction error increased by 180, 25, and 3% for 4-, 8-, and 12-residue loops, respectively. The accuracy of the lowest energy prediction for a given loop can be estimated from the structural variability among a number of low energy predictions. The relative value of the present method is gauged by (1) comparing it with one of the most successful previously described methods, and (2) describing its accuracy in recent blind predictions of protein structure. Finally, it is shown that the average accuracy of prediction is limited primarily by the accuracy of the energy function rather than by the extent of conformational sampling.

Conformational sampling of CDR-H3 in antibodies by multicanonical molecular dynamics simulation

The diversity in the lengths and the amino acid sequences of the third complementarity determining region of the antibody heavy chain (CDR-H3) has made it difficult to establish a relationship between the sequences and the tertiary structures, in contrast to the other CDRs, which are classified by their canonical structures. Enhanced conformational sampling of two different CDR-H3s was performed by multicanonical molecular dynamics (multicanonical MD) simulation while restricting the base structures, with and without the other surrounding CDR segments. The results showed that the multicanonical MD sampled a much larger conformational space than the conventional MD, independent of the initial conformations of the simulations. When the other CDRs surrounding the CDR-H3 segments were included in the calculations, the predominant conformations at 300 K corresponded to the X-ray crystal structures. When only the single CDR-H3 loops were considered with the restricted base structures, a greater number of different conformations were sampled as putative loops, but only a small number of stable conformations appeared at 300 K. Analyses of the resultant conformations revealed a structural role for the glycine, when it is located at position three residues before the last residue of CDR-H3 (Gly-X-X-last residue), coincident with the statistical tendencies of many antibody crystal structures. This reflects the general consistency between the energetically stable conformations and the empirically observed conformations. The current method is expected to be applicable to the structural modeling and the design of antibodies, especially for the inherently flexible loops.

Guided selection of antibody fragments specific for human interferon gamma receptor 1 from a human VH- and VL-gene repertoire

OBJECTIVE

The guided selection strategy for isolation of human antibody (Ab) fragments specific for human interferon gamma receptor 1 (IFNGR-1) from a cloned Ab VH and VL repertoire has been investigated. In order to identify recombinant Abs binding to soluble antigen, a novel method termed affinity sedimentation was introduced here.

RESULTS AND CONCLUSIONS

The VH region of murine monoclonal Ab (IR gamma-1) against human IFNGR-1 was combined with human VL repertoire and used for selection of human VL regions. One of these human VL regions (kappa 2) possesses high homology to the murine template VL region, also in CDR3 (77%). A chimeric Fab consisting of kappa 2 and the murine IR gamma-1 VH region was highly IFNGR-1 specific and exerted the same epitope specificity and a comparable binding affinity as the parental murine Fab. In a further step, the selected human VL region kappa 2 was combined with a human VH repertoire and led by guided selection to the generation of a completely human Fab (1b5) specific for human IFNGR-1. The overall VH region homology of 1b5 compared to the parental antibody IR gamma-1 was 81%, with a rather low homology in CDR3. Binding competition studies revealed that the epitope recognized by 1b5 differs from the parental Ab IR gamma-1.

PDB-based protein loop prediction: parameters for selection and methods for optimization

An approach to loop prediction that starts with a database search is presented and analyzed. To obtain meaningful statistics, 130 loops from 21 proteins were studied. The correlation between the internal conformation of the loop and the conformation of the neighboring stem residues was examined. Distances between C(alpha) and C(beta) of the immediate neighbor residues at each end select template loops as well as more complex (e.g. three residues on either side) matching criteria. To have a high probability that the best possible loop candidate in the database is included in the set, relatively large cutoffs for matching the interatomic distances of the stem residues have to be used in the template loop selection procedure; for loops of length 5, this results in an average of 1000 loops and for loops of length 9, the number is about 1500. The required number increases only slowly with loop length, in contrast to the exponential time increase involved in direct searches of the conformational space. The best loops among the large number of candidates can be determined by ranking them with the standard CHARMM non-bonded energy function (without electrostatics) applied to the backbone and C(beta) atoms. The same representation (backbone plus C(beta)) can be used to optimize the loop orientations relative to the rest of the protein by constrained energy minimization. Target loops that have many non-bonded contacts with the protein yield better results so that analysis of the non-bonded contacts of the selected template loops is useful in determining the expected accuracy of a prediction. The method for loop selection and optimization predicted eight (out of 18) loops of up to nine residues to an RMSD better than 1.07 A relative to the crystal structure; for 17 of the 18 loops, one of the three lowest energy template loops had an RMSD of less than 1.79 A. The prediction of antibody loops from a database search is more effective than that for non-antibody loops. Provided that they belong to one of the canonical classes, very similar antibody loops are certain to exist in the database. Superposition of the stem residues for antibody loops also results in a better orientation than with arbitrary target loops because the neighboring residues tend to have a more similar beta-strand structure. Two H3 loops (for which no canonical structures have been proposed) were predicted with reasonable accuracy (RMSD of 0.49 A and 1.07 A) even though no corresponding antibody loops were in the database.

Advances in comparative protein-structure modelling

Comparative modelling of protein 3D structure can now be applied with reasonable accuracy to ten times more protein sequences than the number of experimentally determined protein structures. A protein sequence that has at least 40% identity to a known structure can be modelled automatically with an accuracy approaching that of a low resolution X-ray structure or a medium resolution NMR structure. Currently, the errors in comparative models include mistakes in the packing of sidechains, in the conformation and shifts of the core segments and loops, and, most importantly, in an incorrect alignment of the modelled sequence with related known structures. Nevertheless, the number of applications in which comparative modelling has been proven to be useful has grown rapidly.

Structural classification of CDR-H3 in antibodies

Large varieties in the lengths and the amino acid sequences of the third complementarity determining region of the antibody heavy chain (CDR-H3) have made it difficult to establish a relationship between the sequences and the tertiary structures, in contrast to the other CDRs, which are classified by their canonical structures. A total of 55 CDR-H3 segments from well determined crystal structures were analyzed, and we have derived several remarkable rules, which could partly govern the CDR-H3 conformation dependence on the sequence. Since the rules are physically reasonable, they are expected to be applicable to structural modeling and design of antibodies.