AbDesigner3D: a structure-guided tool for peptide-based antibody production

Abstract

Summary

We present AbDesigner3D, a new tool for identification of optimal immunizing peptides for antibody production using a peptide-based strategy. AbDesigner3D integrates 3D structural data from the Protein Data Bank (PDB) with UniProt data, which includes basic sequence data, post-translational modification sites, SNP occurrences and more. Other features, such as uniqueness and conservation scores, are calculated based on sequences from UniProt. The 3D visualization capabilities allow an intuitive interface, while an abundance of quantitative output simplifies the process of comparing immunogen peptides. Important quantitative features added in this tool include calculation and display of accessible surface area (ASA) and protein-protein interacting residues (PPIR). The specialized data visualization features of AbDesigner3D will greatly assist users to optimize their choice of immunizing peptides.

Availability and implementation

AbDesigner3D is freely available at http://sysbio.chula.ac.th/AbDesigner3D or https://hpcwebapps.cit.nih.gov/AbDesigner3D/.

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

Antibodies have been vital tools in many medical and research applications for decades. For instance, cancer immunotherapy using monoclonal antibodies (mAbs) has shown great promise, and many antibody–drug conjugates (ADCs) are currently in the approval process (Beck et al., 2017; Perez et al., 2014; Scott et al., 2012). In basic science research, antibodies are standard tools for labeling or quantifying specific proteins. These techniques include western blotting (WB), immunoprecipitation (IP), enzyme-linked immunosorbent assays (ELISA), quantitative immunofluorescence (QIF) and immunohistochemistry (IHC) (Bordeaux et al., 2010). Conventionally, antibodies are used to investigate one or several protein targets at a time in reductionist studies. Despite the ubiquity of antibody use, however, it is still not unusual for researchers to suffer from antibody performance issues. Such difficulties typically stem from weak or absent binding to epitopes or, at the other extreme, non-specific binding.

Given the time and costs involved in procuring suitable antibodies, a means of selecting optimal epitopes would be highly desirable. To this end, we previously created AbDesigner, an online tool for identification of optimal immunizing peptides for antibody production using a peptide-based strategy (Pisitkun et al., 2012, 2014). AbDesigner has been widely used in the antibody research and development community, including many academic institutes and commercial companies. The intuitive, practical interface makes the tool an excellent teaching device, in addition to its widespread use in research environments.

In some cases, however, it is necessary to take into account the 3D structure of target proteins, a function that AbDesigner did not perform. Here we describe a new software program, AbDesigner3D, which combines the features of AbDesigner with the specialized display of 3D structural information from the Protein Data Bank (PDB), allowing more accurate determination of highly exposed regions on a protein (Haste Andersen et al., 2006; Novotny et al., 1986; Thornton et al., 1986). New features include the calculation of an accessible surface area (ASA) for each amino acid residue and the determination of protein-protein interacting residues (PPIR). These features will help scientists to optimize their choice of immunogen peptides to assure that they are located at the surface of the target protein, thus avoiding regions that may be inaccessible to antibodies in applications where the proteins are still in their native conformation. Alternatively, the PPIR feature could be used to design antibodies that block interaction.

2 Implementation and usage

AbDesigner3D is implemented using the Java development kit (JDK 1.8 update 121). Installation files and databases are available online at http://sysbio.chula.ac.th/AbDesigner3D or https://hpcwebapps.cit.nih.gov/AbDesigner3D/. The user input can be either gene symbol, UniProt accession number, or UniProt entry name of the following seven species: Homo sapiens, Rattus norvegicus, Mus musculus, Drosophila melanogaster, Caenorhabditis elegans, Saccharomyces cerevisiae and Arabidopsis thaliana. AbDesigner3D displays 3D protein structures using source code from the Jmol plugin (http://www.jmol.org) with minor modifications to enable communication with other functionalities of our tool.

AbDesigner3D retains all features from the previous version. These features include commonplace measures, such as conservation, hydropathy, possible modification sites, transmembrane domains and more. More unconventional outputs would be the calculation of an ‘immunogenicity-score’, which combines hydropathy and secondary structure prediction, and a novel ‘uniqueness’ calculation, which surveys the entire peptidome of a species to aid the user in minimizing off-target effects (Pisitkun et al., 2012, 2014). Our new features were motivated by feedbacks from existing users who requested an interface that reflects the 3D nature of proteins. This capability is now the centerpiece of AbDesigner3D. However, in addition to the obvious, intuitive appeal of 3D protein structural manipulation, AbDesigner3D adds two new quantitative outputs. Specifically, ASA and PPIR are evaluated (Fig. 1A) (see Supplementary Material for more information). First, the ASA indicates the exposed surface area of each amino acid residue calculated using the ‘asa package’ (org.biojava.nbio.structure.asa) in Biojava 3.10 (Prlic et al., 2012). The exposed surface area is known to be positively correlated with antigenicity of epitopes (Novotny et al., 1986; Thornton et al., 1986). The ASA is calculated separately for each amino acid residue by summing all ASA values from all atoms corresponding to that residue. Second, the PPIR feature highlights all residues that contact other residue(s) in a different polypeptide chain of the same protein structure complex. Protein structures used for all AbDesigner3D features are downloaded from the PDB (https://www.rcsb.org/).

Fig. 1.

Two novel outputs in AbDesigner3D, ASA and PPIR (A). User interface of AbDesigner3D (result section) showing panels of interactive heat map (B1), specialized protein structure viewer (B2) and custom options (B3). A residue selected in (B1), as shown in the red rectangle, is highlighted along with its neighboring residues as yellow circles in (B2) (Color version of this figure is available at Bioinformatics online.)

The resulting heat map itself is interactive. When the user clicks on any residue in the heat map, the protein structure viewer will automatically re-center and highlight that selected residue and also its neighboring residues (see Fig. 1B). The specialized protein structure display as well as the interactive behavior of AbDesigner3D outputs will greatly assist scientists to easily identify optimal immunizing peptide regions based on direct color rendering of the ASA and the PPIR outputs on both 3D structural surfaces and their corresponding heat maps. For the ASA, greater color intensities (in red) indicate more exposed surface area values; therefore these values help suggest the regions that can be readily accessible by antibodies. For the PPIR, display of colored interacting residues helps users either to avoid or to select particular regions as immunizing peptides according to their design requirements (e.g. avoiding PPIR highlighted regions could minimize chances of epitope blocking).

In all, users can optimize which region should be used as an immunizing peptide by considering trade-offs between all features displayed in AbDesigner3D.

3 Validation of new features

The ASA and PPIR features have obvious and logical utility. Nevertheless some degree of empirical validation of ASA and PPIR could be useful. In the case of PPIR, Supplementary Figure S1 illustrates three protein-protein interactions (PD-1/PD-L1, PD-1/Pembrolizumab and PD-1/Nivolumab), showing how this feature enables the user to focus on sites at which an antibody blocks an interaction. AbDesigner3D clearly shows that the antibodies Pembrolizumab and Nivolumab block the PD-1/PD-L1 interaction at the specific PPIR involved in this interaction. The PPIR feature conveniently aids the user in identifying residues that are common to all three protein-protein interactions. Specifically, we note that K131 of PD-1 stands out in all three cases, an observation that has gone unnoticed in the literature to our knowledge.

In the case of ASA, the literature lacks rigorous validation of the possibility that antibodies would tend to bind at regions of high ASA values. Therefore, we probed this question ourselves. Our simple hypothesis is that known antibody-binding residues in crystal structures should have particularly high ASA values compared to other non-binding residues. To test this, we examined 32 randomly chosen antibody/target complexes (excluding structures in which the target protein size was smaller than 15 amino acids). While not exhaustive, the result (Supplementary Figure S2) clearly shows that average ASA values from known antibody-binding residues is significantly higher than those of non-binding residues in 24 out of 32 cases. The probability of this level of success was evaluated by the binomial test, with P-value < 0.004. Thus, in addition to the simple logic of the ASA concept, we present novel evidence for its utility.

4 Conclusions

AbDesigner3D offers visualization of the information relevant to selection of immunogen peptide sequences. The specialized display of 3D structural information from the PDB is available in AbDesigner3D, allowing users to view relevant information extracted from UniProt and PDB databases interactively. AbDesigner3D is available at http://sysbio.chula.ac.th/AbDesigner3D or https://hpcwebapps.cit.nih.gov/AbDesigner3D/. More information and examples also can be found at this website.