B cell epitopes and predictions Pernille Haste Andersen, Ph.d. student Immunological Bioinformatics CBS, DTU

Outline What is a B-cell epitope? How can you predict B-cell epitopes? B-cell epitopes in rational vaccine design. Case story: Using B-cell epitopes in rational vaccine design against HIV.

What is a B-cell epitope? Antibody Fab fragment B-cell epitopes Accessible structural feature of a pathogen molecule. Antibodies are developed to bind the epitope specifically using the complementary determining regions (CDRs). Antibodies are developed to specifically bind epitopes by their CDR regions forming a paratope. In order to determine the structure of antibody-epitope complexes, it was discovered that antibodies had to be cleaved by proteases in order to crystallize. The crystallizing fragment binding epitopes is called a FAB fragment, and it contains two chains: The heavy chain and the light chain.

The binding interactions Binding strength Salt bridges Hydrogen bonds Hydrophobic interactions Van der Waals forces Salt bridges are formed by charged side chains and are the strongest non-covalent interactions in binding. Hydrogen bonds a formed between polar side chains and between backbone atoms. Hydrophobic interactions occur because hydrophobe sidechains are less interactive with water. The most energetically favorable state is when the hydrophobic parts are hidden from water. Van der Waals forces are weak attractive forces that occur when atoms in close proximity induce polarities between one another. The salt bridges and hydrogen bonds are the interactions responsible for the specificity of the interactions.

B-cell epitopes are dynamic Many of the charged groups and hydrogen bonding partners are present on highly flexible amino acid side chains. Most crystal structures of epitopes and antibodies in free and complexed forms have shown conformational rearrangements upon binding. “Induced fit” model of interactions. When thinking of protein structure, people tend to think of them as rigid. But protein structures are dynamic and this is also the case for B-cell epitopes. There is evidence of conformational changes upon antibody binding in a high number of cases.

B-cell epitope classification B-cell epitope – structural feature of a molecule or pathogen, accessible and recognizable by B-cells Linear epitopes One segment of the amino acid chain Discontinuous epitope (with linear determinant) Historically B-cell epitopes were classified into linear and discontinuous. Most early theoretical and computational methods for epitope identification were biased towards linear epitopes, since they are much easier to identify and do not require 3D structural knowledge. the classification is somewhat arbitrary, since most linear epitopes are not truly linear, they only have a “linear determinant” or highly protruding part, but antibody interaction also involves neighboring part. even long flexible loops (true linear epitope) can interact with antibodies in different modes, which is almost never completely independent of the rest of the protein. Discontinuous epitope Several small segments brought into proximity by the protein fold

Binding of a discontinuous epitope Antibody FAB fragment complexed with Guinea Fowl Lysozyme (1FBI). Black: Light chain, Blue: Heavy chain, Yellow: Residues with atoms distanced < 5Å from FAB antibody fragments. Hydrophobic Saltbrides Hydrogen bonds Guinea Fowl Lysozyme KVFGRCELAAAMKRHGLDNYRGYSLGNWVCAAKFESNFNSQNRNTDGS DYGVLNSRWWCNDGRTPGSRNLCNIPCSALQSSDITATANCAKKIVSDG GMNAWVAWRKCKGTDVRVWIKGCRL

B-cell epitope annotation Linear epitopes: Chop sequence into small pieces and measure binding to antibody Discontinuous epitopes: Measure binding of whole protein to antibody The best annotation method : X-ray crystal structure of the antibody-epitope complex

B-cell epitope data bases Databases: AntiJen, IEDB, BciPep, Los Alamos HIV database, Protein Data Bank Large amount of data available for linear epitopes Few data available for discontinuous

B cell epitope prediction

Sequence-based methods for prediction of linear epitopes Protein hydrophobicity – hydrophilicity algorithms Parker, Fauchere, Janin, Kyte and Doolittle, Manavalan Sweet and Eisenberg, Goldman, Engelman and Steitz (GES), von Heijne Protein flexibility prediction algorithm Karplus and Schulz Protein secondary structure prediction algorithms GOR II method (Garnier and Robson), Chou and Fasman, Pellequer Protein “antigenicity” prediction : Hopp and Woods, Welling A number of algorithms were developed in 70’s and 80’s to predict B-cell epitopes from protein sequence gAll these methods tried to correlate some sequence features (flexibility, hydrophobicity, hydrophilicity) with a propensity of this sequence to be a B-cell epitope. Naturally these studies demonstrated correlations, and some confirmed experimentally. At this time, there was a lot of debates which method is better. There is two problems with this approach: 1. All the features in these methods in fact correlate with protein surface accessibility, which is the most accurate predictor 2. All the methods are able to predict only so called “linear” or “continuous” epitopes. TSQDLSVFPLASCCKDNIASTSVTLGCLVTGYLPMSTTVTWDTGSLNKNVTTFPTTFHETYGLHSIVSQVTASGKWAKQRFTCSVAHAESTAINKTFSACALNFIPPTVKLFHSSCNPVGDTHTTIQLLCLISGYVPGDMEVIWLVDGQKATNIFPYTAPGTKEGNVTSTHSELNITQGEWVSQKTYTCQVTYQGFTFKDEARKCSESDPRGVTSYLSPPSPL

Propensity scales: The principle D 2.46 E 1.86 N 1.64 S 1.50 Q 1.37 G 1.28 K 1.26 T 1.15 R 0.87 P 0.30 H 0.30 C 0.11 A 0.03 Y -0.78 V -1.27 M -1.41 I -2.45 F -2.78 L -2.87 W -3.00 The Parker hydrophilicity scale Derived from experimental data Hydrophilicity

Propensity scales: The principle ….LISTFVDEKRPGSDIVEDLILKDENKTTVI…. (-2.78 + -1.27 + 2.46 +1.86 + 1.26 + 0.87 + 0.3)/7 = 0.39 Prediction scores: 0.38 0.1 0.6 0.9 1.0 1.2 2.6 1.0 0.9 0.5 -0.5 Epitope

Evaluation of performance A Receiver Operator Curve (ROC) is useful for finding a good threshold and rank methods

Turn prediction and B-cell epitopes Pellequer found that 50% of the epitopes in a data set of 11 proteins were located in turns Turn propensity scales for each position in the turn were used for epitope prediction. 1 4 2 Pellequer et al., Immunology letters, 1993 3

Blythe and Flower 2005 Extensive evaluation of propensity scales for epitope prediction Conclusion: Basically all the classical scales perform close to random! Other methods must be used for epitope prediction

BepiPred: CBS in-house tool Parker hydrophilicity scale Hidden Markow model Markow model based on linear epitopes extracted from the AntiJen database Combination of the Parker prediction scores and Markow model leads to prediction score Tested on the Pellequer dataset and epitopes in the HIV Los Alamos database

Hidden Markow Model A C D ………… G S T V W Pos 1 7.28 Pos 2 9.39 ….LISTFVDEKRPGSDIVEDLILKDENKTTVI…. 2.46+1.86+1.26+0.87+0.3 = 6.75 Prediction value

ROC evaluation Evaluation on HIV Los Alamos data set

BepiPred performance Pellequer data set: HIV Los Alamos data set Levitt AROC = 0.66 Parker AROC = 0.65 BepiPred AROC = 0.68 HIV Los Alamos data set Levitt AROC = 0.57 Parker AROC = 0.59 BepiPred AROC = 0.60

BepiPred BepiPred conclusion: On both of the evaluation data sets, Bepipred was shown to perform better Still the AROC value is low compared to T-cell epitope prediction tools! Bepipred is available as a webserver: www.cbs.dtu.dk/services/BepiPred

Prediction of linear epitopes Pro easily predicted computationally easily identified experimentally immunodominant epitopes in many cases do not need 3D structural information easy to produce and check binding activity experimentally Con only ~10% of epitopes can be classified as “linear” weakly immunogenic in most cases most epitope peptides do not provide antigen-neutralizing immunity in many cases represent hypervariable regions Pro Slide a window across sequence, add propensities. Chop protein sequence up and see which part bind the antibody Immunodominant No need of 3d structure Synthesize peptide and add radioactive label, to make a traditional binding assays Con A small peptide does not give a good response in the immunesystem ….. Linear epitopes are often found in regions which are less conserved. The virus will escape a vaccine based on the epitope.

Sequence based prediction methods Linear methods for prediction of B cell epitopes have low performances The problem is analogous to the problems of representing the surface of the earth on a two-dimensional map Reduction of the dimensions leads to distortions of scales, directions, distances The world of B-cell epitopes is 3 dimensional and therefore more sophisticated methods must be developed The linear prediction methods for continuous epitopes are not very efficient. The problem of reducing the three dimensional epitopes to a one dimensional problem is similar to the problem of making maps of the earth without distortions. You end up with distortions of scales, directions, distances and surfaces. Regenmortel 1996, Meth. of Enzym. 9.

So what is more sophisticated? Use of the three dimensional structure of the pathogen protein Analyze the structure to find surface exposed regions Additional use of information about conformational changes, glycosylation and trans-membrane helices So what is more sophisticated? Don’t try to reduce the dimensions, take into account the three dimensional nature of the world and use protein structures when analyzing epitopes. Consider the dynamical nature of proteins and use information about flexibility and conformational changes in addition to knowledge of hydrophilicity, surface electrostatics and glycosylation.

How can we get information about the three-dimensional structure? Structural determination X-ray crystallography NMR spectroscopy Both methods are time consuming and not easily done in a larger scale Structure prediction Homology modeling Fold recognition Less time consuming, but there is a possibility of incorrect predictions, specially in loop regions

Protein structure prediction methods Homology/comparative modeling >25% sequence identity (seq 2 seq alignment) Fold-recognition/threading <25% sequence identity (Psi-blast search/ seq 2 str alignment) Ab initio structure prediction 0% sequence identity Which kind of protein structure prediction method to use is determined by the similarity of sequence to sequences of protein with a known structure. In general, homology modeling or comparative modeling is leading to the structural models closest to the natural structure. The idea is to align protein sequences, and apply the local template structure to the aligned region in the query. Fold-recognition methods or threading is used when the sequence similarity is lower. It is based on the idea that structures are more conserved than sequences. By using alignments tools like PSI-BLAST, distant relatives are found and used as templates. The query sequence is dragged through the template structure in order to find a likely conformation. Ab initio methods are used when there is no sequence similarity to any protein of known structure. It is a very complicated task to build a protein structure from sequence information only, and the risk of ending up with a model which is very different from the natural conformation is high.

Annotation of protein surface Look at the surface of your protein Analyse for turns and loops Analyse for amino acid composition Find exposed Bepipred epitopes Then it is time to analyse the model with respects to features that are important for antibody binding. As you know by now, the best B cell epitopes are surface exposed and sticking out of the structure ready to be grapped. So the model is used to build a contour map of the protein surface. By using a model, discontinuous regions brought in proximity on the protein surface can be identified. Surface prediction is not easily done by using propensity scale methods or even neural networks.

Q: What can antibodies recognize in a protein? probe Antibody Fab fragment Protrusion index One of simple methods to identify epitope when 3D structure is available, is to roll a large (~10A) ball on a protein surface. This simple structure-based approach yields much more reliable predictions than sequence methods. A: Everything accessible to a 10 Å probe on a protein surface Novotny J. A static accessibility model of protein antigenicity. Int Rev Immunol 1987 Jul;2(4):379-89

Use the CEP server Conformational epitope server http://202.41.70.74:8080/cgi-bin/cep.pl Uses protein structure as input Finds stretches in sequences which are surface exposed

Use the DiscoTope server CBS server for prediction of discontinuous epitopes Uses protein structure as input Combines propensity scale values of amino acids in discontinuous epitopes with surface exposure Will be available soon (not published yet) Contact me for more information

Rational vaccine design >PATHOGEN PROTEIN KVFGRCELAAAMKRHGLDNYRGYSLGNWVCAAKFESNF Rational Vaccine Design

Rational B-cell epitope design Protein target choice Structural analysis of antigen Known structure or homology model Precise domain structure Physical annotation (flexibility, electrostatics, hydrophobicity) Functional annotation (sequence variations, active sites, binding sites, glycosylation sites, etc.) Model In the rational design of vaccines, you would often to start by choosing a protein which is essential for the pathogen. Then follows a structural analysis of the protein, where as many features as possible are mapped to the structure. Known 3D structure

Rational B-cell epitope design Protein target choice Structural annotation Epitope prediction and ranking Surface accessibility Protrusion index Conserved sequence Glycosylation status Once, you have enough information about the antigen, you start to analyze it for epitopes. Epitope prediction can involve the use of linear methods or the use of three dimensional models/structures for predictions of surface accessability and protrusion index. Also knowledge about regions of conservation and glycosylation can be included. So the best epitope has the following features: -Highly exposed -In a region sticking out of the structure -No conformational changes can hide the epitope -In a region with conserved sequence -The epitope is no hidden by glycosylations

Rational B-cell epitope design Protein target choice Structural annotation Epitope prediction and ranking Optimal Epitope presentation Fold minimization, or Design of structural mimics Choice of carrier (conjugates, DNA plasmids, virus like particles) Multiple chain protein engineering Now you have to decide how to present the epitope in the vaccine to the immune system. If you produce recombinant versions of the antigen, you have to make sure that the three dimensional structure of the epitope is conserved. You can do that by minimizing the structure of the recombinant protein. *You start from the known structure????? Also you can choose to make a mimic of the epitope, which might be more stable than the original epitope. You also have to think about a carrier, are you going to inject protein directly into the body, or use DNA plasmids? In general protein engineering provides a range of methods for designing rational vaccines.

Multi-epitope protein design B-cell epitope Rational optimization of epitope-VLP chimeric proteins: Design a library of possible linkers (<10 aa) Perform global energy optimization in VLP (virus-like particle) context Rank according to estimated energy strain T-cell epitope Viral coat proteins assemble spontaneously to produce higly immunogenic particles. This feature can be used in the design of vaccines encompassing both accessible B-cell epitopes and T-cell epitopes. By using recombinant technologies, viral coat proteins are fused to regions of an antigen encoding for T-cell epitopes and a B-cell epitope. An optimal linker is found by evaluation of the energy strains of the construct. The less strain, the more possible is the right conformation of the B-cell epitope.

Using B-cell epitopes in the rational vaccine design against HIV Case story Using B-cell epitopes in the rational vaccine design against HIV The gp120-CD4 epitope

HIV gp120-CD4 epitope Binding of CD4 receptor Conformational changes in gp120 Opens chemokine-receptor binding site New highly conserved epitopes Kwong et al.(1998) Nature 393, 648-658 Some diseases are very difficult to develop vaccines against. HIV is one of them. HIV starts an infection by entering the cells of the immunesystem expressing the CD4 protein. It expresses the glycoprotein GP120 on the viral surface, which binds CD4 receptors and thereby attach itself on cells. The binding of CD4 leads to conformational changes in GP120, and a conserved chemokine receptor binding site is exposed. So HIV protects itself against the immunesystem using several features: -glycosylation -highly variable exposed regions and hidden conserved, essential regions -conformational changes

Conformational changes in gp120 SIV gp120 no ligands Human gp120 complex with CD4 and 17b antibody Chen et al. Nature 2005 Kwong et al. Nature 1998

HIV gp120-CD4 epitope Efforts to design a epitope fusion protein Elicit broadly cross-reactive neutralizing antibodies in rhesus macaques. This conjugate is too large(~400 aa) and still contains a number of irrelevant loops Fouts et al. (2000) Journal ofVirology, 74, 11427-11436 Fouts et al. (2002) Proc Natl Acad Sci U S A. 99, 11842-7. So how do you design a HIV in a smart way which can prevent the virus from entering the helper T-cells? The first efforts were to form a single-chain analogue using recombinant proteins. The binding regions of CD4 and GP120 were fused to make an epitope that mimics the complexes of CD4 and GP120 on the helper T-cell surface. The idea was that the CD4 part of the conjugate bind the GP120 part and leads to the exposure of the conserved epitope. One of the problems are that conjugate-specific parts are immunodominant, and therefore antibodies raised upon the analogue will not be HIV protective.

HIV gp120-CD4 epitope Further optimization of the epitope: reduce to minimal stable fold (iterative) find alternative scaffold to present epitope (miniprotein mimic) Martin & Vita, Current Prot. An Pept. Science, 1: 403-430. Vita et al.(1999) PNAS 96:13091-6 Therefore, the construct is optimized in order to reduce the size and find a combination leading to neutralizing antibodies.

Conclusions Rational vaccines can be designed to induce strong and epitope-specific B-cell responses Selection of protective B-cell epitopes involves structural, functional and immunogenic analysis of the pathogenic proteins When you can: Use protein structure for prediction Structural modeling tools are helpful in prediction of epitopes, design of epitope mimics and optimal epitope presentation