Abstract
Epitope-based DNA vaccine development is one application of bioinformatics or in silico studies, that is, computational methods, including mathematical, chemical, and biological approaches, which are widely used in drug development. Many in silico studies have been conducted to analyze the efficacy, safety, toxicity effects, and interactions of drugs. In the vaccine design process, in silico studies are performed to predict epitopes that could trigger T-cell and B-cell reactions that would produce both cellular and humoral immune responses. Immunoinformatics is the branch of bioinformatics used to study the relationship between immune responses and predicted epitopes. Progress in immunoinformatics has been rapid and has led to the development of a variety of tools that are used for the prediction of epitopes recognized by B cells or T cells as well as the antigenic responses. However, the in silico approach to vaccine design is still relatively new; thus, this review is aimed at increasing understanding of the importance of in silico studies in the design of vaccines and thereby facilitating future research in this field.
Introduction
The development of new drugs is time-consuming and costly. For example, the prerequisite average costs for the development of an active substance accepted by the Food and Drug Administration (FDA) are $648.0 million (US dollars; range: $157.3–1950.8 million) and the average time required to develop a new drug is 7.3 years (range: 5.8–15.2 years). 1 Furthermore, around 53% of newly developed drugs fail to reach the preclinical phase, mainly due to intolerable side effects, unacceptable toxicological effects, and unpredictable drug interactions. For these reasons, the development of new drugs is not a top priority for the pharmaceutical industry, which instead largely focuses on the development and production of pre-existing active compounds to increase profits and reduce losses. Nevertheless, many effective treatments have yet to be discovered, especially for multifactorial diseases such as degenerative diseases.2,3
When vaccines were first created, they were intended to prevent diseases caused by infectious agents. However, as vaccine technology has developed, vaccines have been expanded in an effort to combat noninfectious diseases such as autoimmune diseases,4,5 cancer,6–8, and degenerative diseases.9–11
Conventional vaccines are produced by inactivating or attenuating some part of an infectious agent and exposing it to the body’s immune system. This approach has been so successful that vaccines are considered one of the major successes of the modern world. In particular, this approach has been effective against infectious agents with low antigen variations such as polio, smallpox, measles, and rubella.12,13 However, for diseases with mechanisms involving complex immune reactions, this approach is often ineffective; thus, new strategies for vaccine development are required. 14 Conventional vaccination methods also often trigger side effects including fever and hypersensitivity reactions. Therefore, it is necessary to develop a new generation of vaccines, such as epitope-based vaccines, with high effectiveness and minimal side effects.
The evolution of epitope-based vaccines is one of the most promising developments to arise from bioinformatics-based research. 15 Bioinformatics or in silico studies, that is, computational methods that include mathematical, chemical, and biological approaches, are widely used in drug development. For example, in silico studies are often utilized to analyze the bioavailability of drug compounds,16,17 pharmacokinetic–pharmacodynamic processes, 18 interactions among drug compounds, and the toxic effects of drug compounds.19,20
During vaccine design, in silico studies are performed to predict epitopes that can trigger both T-cell and B-cell reactions, which in turn produce cellular and humoral immune responses. 21 Immunoinformatics is the branch of bioinformatics in which the relationship between immune responses and predicted epitopes is studied. 22 The rapid development of immunoinformatics has been characterized by the creation of tools used for the prediction of epitopes recognized by B cells and T cells and for antigen responses.23,24 Nevertheless, the in silico approach to vaccine design is still relatively new; therefore, it is necessary to conduct in-depth studies that will increase general understanding of the importance of in silico research to the vaccine design process.
In silico approach for vaccine design
The aim of vaccination is to stimulate the memory of the adaptive immune system to ensure that it responds immediately to the next antigen exposure. The adaptive immune system consists of two classes, namely the humoral immune system mediated by antibodies produced by B lymphocyte cells and the cellular immune system mediated by T lymphocytes. The humoral and cellular immune systems are stimulated when a receptor recognizes a particular part of an antigen known as an epitope.21,25
Conventional vaccination approaches that use weakened or activated antigens are ineffective against several types of diseases, especially those involving complex immunity such as HIV, tuberculosis, cancer, and atherosclerosis. Thus, to increase specificity, effectiveness, and safety, bioinformatics methods are used during epitope-based vaccine development.10,26–28 This approach minimizes resource use and time costs because the initial screening is conducted in silico to increase the efficiency of the vaccine candidate search. 23 In comparison to conventional vaccines, epitope-based vaccines are typically well-tolerated and have fewer side effects.15,24 At present the approach has been demonstrated to be effective designing vaccine, including for COVID-19 vaccines29–31 and other infectious diseases.32,33
During vaccine design, the bioinformatics approach is based on the availability of data and epitope predictions. 34 In broad terms, the steps involved in vaccine design using the in silico method include searching antigen protein databases, analyzing protein interactions, characterizing the epitopes recognized by both T cells and B cells, and analyzing antigenicity and homology. This approach usually requires massive amounts of reliable antigenic protein data. The prediction of epitopes recognized by B cells and T cells is based on sequences and structure and not on pathological mechanisms. 35
Protein databases and protein interaction analysis
Massive, reliable protein databases are required for the design of epitope-based vaccines. Currently, several such protein databases can be easily accessed including those provided by the National Center of Biotechnology Information (NCBI), UniProt, or Protein Data Bank (PDB).
From the NCBI website (www.ncbi.nlm.nih.gov), various data related to proteins can be accessed. Some programs for analysis can also be operated from this website including a search and retrieval system that provides users with integrated access to sequence, mapping, taxonomy, and structural data. Moreover, the website provides sequence similarity search tools and can be used to identify genes and genetic features. 36
The PDB (www.rcsb.org) provides access to protein-related data including protein sources, crystallographic data, chemical structures, peptide sequences, and protein structures based on nuclear magnetic resonance (NMR). In addition, this database enables various protein visualizations and protein analyses such as sequencing, prediction of protein structure, and protein symmetry analysis. Furthermore, the PDB provides a domain-based structural alignment method. It also includes structure depositions that have been determined using several techniques including macromolecular crystallography, three-dimensional (3D) electron microscopy (EM), powder diffraction, and fiber diffraction.34,37,38
UniProt (www.uniprot.org) provides almost complete information on proteins including data on functions, names and taxonomies, subcellular locations, related pathologies, post-translational modifications (protein processing), expressions, interactions, structures, sequences, families and domains, reference information, and similar proteins. Moreover, Uniprot includes improved metagenomic assembly and binning tools that provide high-quality metagenomic assembled genomes. In addition, Uniprot provides the UniRef databases, which cluster sequence sets at various levels of sequence identity, and the UniProt Archive (UniParc), which delivers a complete set of known sequences.39,40
Many new databases and tools have been developed as accessible repositories for storing and analyzing large amounts of immunology-related biological data. Most of these databases have been listed as public repositories to make it easier for researchers to find the databases they need. These public repositories provide access to up-to-date annotated lists of immunoinformatic resources, ensuring the quality and relevance of these databases and tools. 41 The three main public repositories containing information on available databases and tools related to immunoinformatics are (1) Nucleic Acids Research Database Annual Issue, (2) Canadian Bioinformatics Links Directory, and (3) Immune Epitope Database and Analysis Resources (IEDB).
Antigenicity and epitope prediction
After protein data is analyzed, 3D protein structure analysis or modeling can be conducted. In general, homology modeling or comparative modeling methods are performed because not all studied proteins have known 3D structures. Using these methods, protein structure can be predicted based on alignment results with one or more other proteins for which the structure is known. The program largely applied in protein modeling is MODELER, in which information from an input target-template alignment is used to create a series of homology-derived spatial restraints that act on the atoms of the 3D protein model. Sigma values of homology-derived distance restraints define the acceptable amount of conformational freedom for the model based on its templates. 42
Analysis of protein interactions is performed using protein docking, which predicts the formation of protein complexes or ligands based on binding models and surface free energy. Protein docking can be divided into two processes: sampling and scoring. Sampling is a method used to determine which parts of a protein are relevant to conformational or binding orientations. The sampling process can involve the use of a binding orientation algorithm (rigid-body sampling) or may be based on protein conformation (conformational sampling). Once sampling is completed, scoring is conducted to assess each binding model. Moreover, each binding model is sorted, with the binding model possessing the highest score suggested as a protein complex formation model. 43
Molecular docking is then performed to ensure that a candidate epitope vaccine could generate a stable immune response. This is achieved by measuring interactions between the candidate and target immune cell receptors such as Toll-like receptor 2 (TLR2), TLR3, and TLR4.44–46 Several studies have included molecular docking of vaccines with the human leukocyte antigen molecule or major histocompatibility complex I (MHC I) and MHC II receptors.45,47 After molecular docking is completed, the protein–protein interactions among docked molecules are also analyzed.
Antigenicity prediction
Antigenicity prediction is used to determine the peptides that have high antigenicity and can be developed as vaccine candidates. The many tools used to predict antigenicity are based on various antigenicity determination methods, for example, the Kolaskar–Tongaonkar method 48 and Welling method. 49
The Kolaskar–Tongaonkar method is based on experimental research indicating that hydrophobic residues, such as cysteine, leucine, and valine, on the surface of proteins tend to have antigenic characteristics. Based on these experimental data, a semi-empirical method was developed to assess whether a peptide is or is not antigenic. The method has an accuracy of around 75% and has the advantage of being simple to use as it only requires one parameter. 48 The Welling method is used to determine an antigenicity value based on a comparison between the percentage of specific amino acids on the antigenic side and the percentage of these amino acids in the protein. 49
B-cell epitope mapping
Structural epitope mapping could be conducted using X-ray crystallography, nuclear magnetic resonance (NMR), 50 EM, 51 or cryoelectron microscopy (CM).52,53 X-ray crystallography is believed to be the most precise method for structural epitope mapping. However, the quality of cocrystals and the antibody’s electron density limit X-ray crystallography. 54 NMR offers peptide mapping based on the difference in the NMR signal of the free antigen or the antibody-bound antigen to determine the epitopes. NMR epithelial mapping provides more detailed information than mutagenesis or peptide mapping and can be much faster than X-ray crystallography.55,56 CM is another technique used to determine macromolecular structures with resolution comparable to X-ray crystallography. Because the samples are flash frozen in CM, crystallization is not required. Typically, fewer samples are required, but they must still be relatively homogeneous in purity. CM provides higher resolution information for larger molecules and less information for smaller molecules. 57
B-cell epitope prediction
B cells mediate the humoral immune system through antibody secretion that neutralizes antigens. B cells are stimulated when the antigen receptor, which is part of the paratope, recognizes antigenic epitopes. Most available epitope mapping methods (structural and functional approaches) are costly, time-consuming, and frequently fail to detect all epitopes. The protein structure including residues in direct contact with an antibody is interpreted using structural epitope mapping methods, although these methods frequently fail to identify the role of amino acids in binding strength. The goal of functional epitope mapping techniques is to identify and characterize residues critical for binding within structurally specified antigenic determinants. 24
Epitopes recognized by B cells can be classified into two types: continuous and discontinuous epitopes. Continuous epitopes (also referred to as linear or sequential epitopes) are short peptide fragments (about 15 amino acids in size) of an antigen protein that are specifically identified by certain antibodies. Discontinuous epitopes consist of amino acid residues that are not sequential in their primary structure but involve a folding mechanism that forms into a region that is close together. However, the folding mechanism increases the complexity of epitope prediction; the classification is not rigid because several continuous epitopes could form certain conformations that are recognized by antibodies and discontinuous epitopes can also contain several sequential linear peptide sequences. 58 Because of their complexity, the prediction of B-cell epitopes is often less accurate than the simpler prediction of T-cell epitopes.
Linear epitope prediction
Sequence-based prediction can be used to predict continuous epitopes based on the propensity scale method, which is used to assess and compare the tendency of amino acids to become epitopes recognized by B cells relative to amino acids that form antigens. To determine the propensity value for a residue, i, a central residue in a window chosen with size n, we would use the formula i − (n − 1)/2. The value of residue i is the average value for amino acids in a predetermined window range. In general, 5–7 amino acids are used to determine an epitope. The assessment is based on the physical characteristics of these amino acids, for example, hydrophilicity, flexibility, solvent accessibility, or protein helixes.59,60
Hydrophilic scores are determined based on amino acid retention times in high-performance liquid chromatography in the reverse phase column. In such assessments, a window consisting of seven amino acids has been used, in which for the fourth amino acid residue value is determined from the average hydrophilicity value of the seven residues. 61
The flexibility assessment by Karplus and Schulz is based on the mobility of protein segments at a factor B temperature of carbon α for 31 proteins with known structures. This flexibility calculation uses the first amino acid from a window span consisting of six amino acids. 59
Solvent accessibility scores are determined based on the probability of an amino acid being exposed to the X-ray structure of 28 proteins. The surface probability (Sn) is determined using the following formula:
where Sn is surface probability, δn is the fractional probability value of the surface, and the i value varies from 1 to 6. A hexapeptide with Sn >1 indicates an increased probability that an amino acid will be on the surface. 62 Moreover, an assessment by Chou and Fasman is based on the probability that a certain range of residues are part of a β-turn structure. 59
Studies have shown that predictions made using a single physicochemical characteristic cannot accurately predict B-cell epitopes. Therefore, some tools simultaneously use a combination of physicochemical character assessments, such as PREDITOP, PEOPLE, and BEPITOPE, to improve the accuracy of predictions.21,63,64 The machine learning method is a computational method that uses a train classifier to distinguish epitope and non-epitope antigenic structures based on data related to structural differences and physicochemical characteristics. 64
Sequence-based prediction has the advantage of not requiring any understanding of the target antigen’s 3D structure. To determine the 3D structure of a target antigen, data from X-ray crystallography studies are required; however, not all target antigens have known 3D structures. In contrast, the disadvantage of sequence-based prediction of B-cell epitopes is their relatively low accuracy, which on average is 60–70% because the majority of epitopes recognized by B cells are naturally in a discontinuous condition. 60
Furthermore, newer methods have been developed to predict continuous B-cell epitopes. The SVMTriP service predicts continuous B-cell epitopes using the support vector machine algorithm (SVM) and the Tri-peptide similarity and propensity score (SVMTriP) in order to improve prediction accuracy. 65 Another approach is using validated B-cell epitopes as well as non B-cell epitopes from Immune Epitope Database which is resulting in two types of datasets called Lbtope_Variable and Lbtope_Fixed length. 66
Confirmed epitope prediction
The prediction of a confirmed epitope was developed because 90% of the epitopes recognized by B cells are in a discontinuous condition or form specific conformations. The first discontinuous epitope prediction method developed was the conformational epitope prediction, which can predict continuous or discontinuous epitopes using 3D protein structures. The underlying algorithm employs solvent accessibility data based on the Voronoi polyhedron. Continuous epitopes are determined based on the presence of at least three sequential residues, whereas discontinuous epitopes are determined by decreasing continuous epitopes with Cα within 6A. 24
Another method, DiscoTope, uses a combination of amino acid statistics, spatial context, and amino acid surface accessibility to predict B-cell epitopes. 67 This method can detect 15.5% of the residues present in discontinuous epitopes with a specificity of up to 95%; at this level of specificity, Parker’s hydrophilicity method can only detect 11% of residues in discontinuous epitopes.24,26,67
ElliPro is a tool based on the combination of the Thornton concept, the MODELER program, and Jmol viewer. 59 For the prediction of B-cell epitopes, ElliPro uses three steps: estimation of protein structure as an ellipsoid, calculation of the residual protrusion index (PI), and grouping of residues based on PI values. PI values are defined as the percentage of protein atoms inside the ellipsoid where the first residue is outside the ellipsoid. 59 The Antigenic Epitopes Prediction with Support Vector Regression server (EPSVR) manipulates vector regression to combine the same scores as EPCES and achieves an area under the curve (AUC) of 0.597. 68 Other tools that can be used to predict discontinuous B-cell epitopes include Epitopia, 69 PEPOP, 70 EPIMAP, 60 and CBTOPE 71 (Table 1).
B-cell epitope prediction tools.
ANN, artificial neural network.
T-cell epitope prediction
Compared with B-cell epitope predictions, T-cell epitope predictions are generally easier and more accurate because the structures of epitopes identified by T cells are simpler, that is, short, linear peptides (9–15 amino acids in length). Epitopes are recognized by the T-cell receptors (TCRs) in the form that is presented by the MHCs, that is, MHC class I or class II. It is important to consider both epitope and MHC bonding and epitope–MHC and TCR complex bonds during the prediction of epitopes recognized by T cells. Epitopes bind to specific parts of MHCs, known as grooves, which are usually formed from two α helices and one β sheet and are then presented to T cells. 72 Peptides are bound to MHCs through hydrogen bonds, electrostatic interactions, and van der Waals interactions. In general, peptides that bind to MHC class I have 8–11 amino acids sizes, whereas peptides that bind to MHC class II are 12–25 amino acids in length and protrude from the MHC groove but have at least 9 amino acids in the core. 73 However, other studies have shown that some larger peptides can also bind to the MHC but have a lower immunogenic potential.23,74
Some of the methods used to predict epitopes that are recognized by T cells are the motif-based system, matrix, SVM, empirical scoring, and molecular dynamics (MDs) methods. 73 The motif-based system was the first T-cell epitope prediction method developed. In this method, amino acid sequences that have a high tendency to bind to the MHC groove or so-called motif are predicted. The amino acid sequence is then compared with the data in a library motif, where the previously determined binding peptide sequence and the nonbinding MHC-binding motif are collected. The accuracy of this method can reach 60–70% because not all peptides have known motifs. 23
Other motif-based system has been developed based on machine learning algorithms (MLAs). For instance, based on MLAs, peptide-binding motifs can be determined according to certain classifications, for example, a positive value for a peptide binder and a negative value for a nonpeptide binder. MLAs can also be used for several classifications at the same time. Artificial neural networks are one of the types of MLAs most widely used to determine the motifs for introducing peptides to MHCs.22,75
Prediction of T-cell epitopes can also be performed by simulating MDs, in which free binding energy is calculated for a molecular system. MDs can be used to explain the movement of atoms individually or collectively in a molecular system; thus, MDs provide a dynamic picture. The advantage of MDs relative to other methods is that they are not based on data alone but based on de novo predictions of all parameters that construct the structure of the receptor ligand complex. 76 A summary of the tools that can be used to predict epitopes recognized by T cells is shown in Table 2.
T-cell epitope prediction tools.
In silico studies offer a new solution for cutting costs and time, which is important for drug development. By using in silico studies, we can predict the effectiveness of a drug compound and thereby design an ‘ideal’ drug. Indeed, with the aid of in silico technology, before a new drug is developed, its effectiveness, side effects, potential for contractions, and toxic effects can be determined in advance. Thus, the time and costs required for development can be reduced.
During vaccine development, in silico studies can provide huge benefits. Conventional vaccines that use all or part of a weakened or inactivated pathogen often cause severe side effects such as fever and hypersensitivity reactions. The imperfect inactivation process also allows active pathogens to enter the body and cause symptoms of the disease. The development of recombinant protein-based subunit vaccines requires a long time because it must include the most potent antigen screening process among many other antigen proteins. 23 The development of recombinant protein-based vaccines is expensive because the production process must be sterile. Recombinant protein stability is also relatively low; therefore, vaccines must be stored at a certain temperature, which increases difficulties with their distribution and storage. Therefore, epitope-based vaccines are considered to be a solution to the problems of conventional vaccines, including vaccines for infectious diseases,28,29,77,78 and even metabolic disorders or inflammatory diseases. 10
The development of informatics has given rise to new programs with respective advantages. The emergence of epitope prediction tools, antigenicity prediction, protein modeling, and docking analysis has made it possible to design epitope-based vaccines with maximum efficacy. In addition, the development of X-ray crystallization technology, NMR spectrophotometry, and CM has revealed the 3D structure of an increasing number of proteins, which in turn has facilitated the analysis of protein interactions. For proteins with unknown dimensions, 3D modeling and docking analysis methods have enabled predictions of protein interactions including prediction of bonds between antibodies and antigens with the highest affinities. 79
The application of in silico studies to the design of epitope-based vaccines is also relatively simple and does not require complex skills. The necessary tools used are also widely available for free and can be accessed easily. The immune epitope database (IEDB) provides tools for the prediction of epitopes that are recognized by B cells and T cells, as well as for analyzing epitope characteristics for more complete and reliable prediction results. This database and its associated tools have often been used in studies in which epitopes were predicted for vaccine development,22,63 perhaps because the resource is easy to use. The main disadvantage of using in silico studies to develop epitope-based vaccines is that all predictions are computationally based on approaches involving mathematics, chemistry, and biology; thus, the accuracy never reaches 100%. In a biological system, there can be unpredictable interactions since proteins are dynamic macromolecular complexes. Protein 3D conformations are prone to changes in the physical environment, such as changes in charge and pH, which disrupt the structure and activity of the protein including its binding with other proteins. 80 Antibodies are also proteins that are specific to certain antigens; changes in one residue alone prevent recognition by these antibodies. To improve the accuracy of epitope prediction, it is necessary to analyze MDs to validate the binding of antibodies to receptors. By improving accuracy in this manner, the effectiveness of vaccines is also expected to be improved.
Conclusion
Based on our reviews, the immunoinformatic tools are very valuable tools for predicting and evaluating the epitopes for vaccine candidate development. These tools undeniably are becoming the most informative and advantageous device for vaccine design.
Footnotes
Author contributions
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Conflict of interest statement
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.