Why structural models of antibody targets are needed

This article describes the state-of-the-art in silico Antibody Structure Modeling workflow used in BIOVIA Discovery Studio® software and illustrates key reasons why structural models of antibody targets are needed.

Antibodies are becoming increasingly significant in medical diagnostics and in the treatment of various diseases, including cancer, inflammation, and auto-immune illnesses. Antibodies also enable the targeted delivery of conventional drugs using antibody Drug Conjugates (ADCs). In contrast to standard chemotherapeutic drugs, ADCs target and destroy cancer cells while sparing healthy cells.

Many key features of antibodies cannot be predicted by sequencing alone or can be predicted more accurately with access to a 3D antibody structure. Some examples of the usefulness of quality 3D structure models are:

  • Humanization
  • Maturation
  • Detecting exposed and buried residues to evaluate likely antigenicity
  • Formulations
  • Predict changes in stability or antigen binding (avidity) with in silico mutations (including f(pH) or f(temperature)
  • Rank targets by developability index or identify patches with high predicted aggregation propensity (Figure 1)

The Aggregation prediction of a target before and after mutations designed to maintain affinity and reduce aggregation (aggregation-prone regions in red).

Figure 1. The Aggregation prediction of a target before and after mutations designed to maintain affinity and reduce aggregation (aggregation-prone regions in red). Image Credit: BIOVIA, Dassault Systèmes

Understanding these properties as early as possible in formulation and development will affect speed and total cost to market. In instances where 3D structures are unavailable from NMR or X-Ray, 3D in silico models produced by BIOVIA Discovery Studio are the solution.

Method

BIOVIA Discovery Studio has all of the capabilities required by both experts and non-experts to develop a quality antibody structure depending on sequence input (Figure 2).

The generation of a structure proceeds from sequence using multiple templates for framework residues and CDR loop residues.

Figure 2. The generation of a structure proceeds from sequence using multiple templates for framework residues and CDR loop residues. Image Credit: BIOVIA, Dassault Systèmes

The modeling workflow comprises three principal stages, detailed below.

1. Identify framework templates

  • Locate quality crystal structures with suitable framework region sequence similarity.
  • Optionally, results can contain complementarity-determining region (CDR) similarity and filter by organism.

2. Generate 3D model based on Framework Templates

  • Multiple Framework Templates can be used as input. Using homology modeling software (Modeler), conformations are reduced to restraints and then utilized to anneal the target conformation.

3. Refine Antibody Loops to predict CDR conformations

  • Refines the extremely sequence variable CDR loops based on homology to multiple quality CDR loop templates. If none exist, it will apply de novo force field-based techniques.

Results

This article is based on Dassault Systèmes’ involvement in Antibody Model Assessment II2 (referred to as AMA-II henceforth), a “blinded” exercise in which conformations for eleven antibody Fv sequences were predicted and assessed in relation to known but unreleased crystal structures.

In the analysis of the predicted structures, several criteria were assessed for their ability to identify the highest quality model structures. This analysis revealed that backbone RMSD is not a sensitive measure of structural quality. The crucial backbone orientation is not captured by that conventional metric.

According to the findings of this study, backbone root-mean-square deviation (RMSD) is not a sensitive measure of structural quality. That conventional metric does not capture the critical backbone orientation.

The AMA-II2 study advises using RMSD based on the amino acid carbonyl group atoms C and O, therefore, the company employed this metric (RMSDCO) below. In this study, numerous workflows were tested.1 These briefly included modeling the framework regions (FR) with a single template, separating optimal templates for the L and H chains, or using multiple templates for the L and H FR.

The annotation method used determines the CDR length for CDR loops, and filtering based on canonical loop type prediction is an option. The 11 models were then subjected to automated structure determinations to more rigorously evaluate the relative merits of all these methods. The results are summarized in Figure 3.

Results for the BIOVIA (Accelrys) models submitted for AMA-II are shown as red (model 1), blue (model 2), and green (model 3) bars. The results for different automated approaches in the post-experiment analysis are shown as purple (single template), orange (chimeric template) and yellow (top five templates) bars. In each panel, the box plots in the background indicate the distribution for the models submitted by AMA-II participants. The thick black bar inside the boxes indicates the median, the top and bottom boundaries of the boxes indicate the first and third quartiles (i.e., 25th and 75th percentiles). The tails indicate the highest/lowest RMSDCO values that fall within a factor of 1.5 times the interquartile distance of the box boundaries. Any outliers falling in the regions beyond the tails are drawn as black circles. (a) Plots RMSDCO of the model structures compared to the X-ray structure calculated over b-core of the VL region. (b) Plots the same data for the VH region.

Figure 3. Results for the BIOVIA (Accelrys) models submitted for AMA-II are shown as red (model 1), blue (model 2), and green (model 3) bars. The results for different automated approaches in the post-experiment analysis are shown as purple (single template), orange (chimeric template) and yellow (top five templates) bars. In each panel, the box plots in the background indicate the distribution for the models submitted by AMA-II participants. The thick black bar inside the boxes indicates the median, the top and bottom boundaries of the boxes indicate the first and third quartiles (i.e., 25th and 75th percentiles). The tails indicate the highest/lowest RMSDCO values that fall within a factor of 1.5 times the interquartile distance of the box boundaries. Any outliers falling in the regions beyond the tails are drawn as black circles. (a) Plots RMSDCO of the model structures compared to the X-ray structure calculated over b-core of the VL region. (b) Plots the same data for the VH region. Image Credit This figure is extracted from Fasnacht et al., 2014

Figure 3 demonstrates consistently high-quality results that, on average, outperform other currently used methods. It also shows that completely automated structure development may provide models with accuracy on par with those generated by expert modelers while consuming less CPU power and time restraints.1

Discussion

The antibodies in AMA-II included some difficult targets. Ma2-01 (model assessment-II, structure 01), for instance, was derived from Rabbit, which has considerably fewer templates than Human or Mouse. Ma2-05 has a lambda light chain, which is much rarer in the crystal database than kappa. Ma2-10 was found to contain a long H3 loop (16 residues using IMGT).

When executed in a semi-automated workflow, the protocols created and available through BIOVIA Discovery Studio are built so that the output delivers suggested templates in ranked order according to the similarity score and template resolution, allowing for easy selection.

On average, the best results were obtained when numerous templates were used to generate the framework areas and to determine the ideal conformation of the CDRs utilizing the Model Antibody Loops procedure.

This is due to the Modeler solution engine’s ability to refine the target conformation using a weighted combination of restraints derived from all templates where the degree of sequence homology differs relative to the template, or different conformations are present for templates with identical sequences.

Figure 3 depicts the VL and VH RMSDCO values relative to crystal structure for the three user-submitted models and three computer-generated models for each target in AMA-II. The boxes depicted in this figure relate the quality of the predictions to those of the structures presented by other participants of AMA-II (See Figure 3 caption).

When the similarity difference between the different single templates is sufficiently large, the framework RMSDCO data benefits from the flexibility to choose separate heavy and light chain templates.

Figure 3 indicates that separate VL and VH framework templates have, at worst, no influence and, at best, allow for a significant improvement in RMSDCO when compared to a single framework template (target 05, 07, and 08). The CDR loop conformations achieved by Model Antibody Loops and BIOVIA Discovery Studio’s de novo loop models provide state-of-the-art fidelity to reference crystal structures, as demonstrated in Figure 4.

Average peptide carbonyl RMSD values for each group participating in AMA-II. Averages are for model 1 from each group, excluding the rabbit target, Ab01.2 ACC is BIOVIA (Accelrys). RMSDCO values shown are light chain only (VL), heavy chain only (VH), the five canonical CDRs, and CDR H3. Tilt refers to the deviation in the angle between the VL and VH chains relative to the crystal structure.

Figure 4. Average peptide carbonyl RMSD values for each group participating in AMA-II. Averages are for model 1 from each group, excluding the rabbit target, Ab01.2 ACC is BIOVIA (Accelrys). RMSDCO values shown are light chain only (VL), heavy chain only (VH), the five canonical CDRs, and CDR H3. Tilt refers to the deviation in the angle between the VL and VH chains relative to the crystal structure. Image Credit: BIOVIA, Dassault Systèmes

Conclusions

With a sixfold increase in global sales since 2003, antibody-based diagnostics and therapies have already established themselves as a prominent part of the life sciences commercial sector.3 Reducing product time to market and improving the ability to “fail early” in the Discovery phase are critically important. The traditional workflow is in vitro optimization after the development of hybridomas.

This optimization seeks to increase avidity, lessen immunogenicity, and enhance stability by avoiding undesirable post-translational modifications (PTM) and minimizing viscosity or aggregation.

In silico structure predictions produced by the aforementioned techniques allow for the prediction of key bulk and molecular level properties that could not be predicted by sequence alone. Conformational models predict solvent accessibility and charge distribution at the molecule’s surface, which are key components in estimating the critical qualities necessary to evaluate prospective targets.

Such structural models can be obtained using single crystal X-Ray crystallography or NMR spectroscopy; however, these analytic procedures are costly and time-consuming. The alternative is modeling based on homology and de novo approaches, detailed in this article.

Anti-Gastrin Fv model structure with predicted docked conformation of gastrin predicted and rendered by BIOVIA Discovery Studio.

Figure 5. Anti-Gastrin Fv model structure with predicted docked conformation of gastrin predicted and rendered by BIOVIA Discovery Studio. Image Credit: BIOVIA, Dassault Systèmes

Based on the RMSDCO comparison metric established by the independent ABA-II assessment researchers,2 the blinded AMA-II research findings1,2 demonstrated that antibody models generated by BIOVIA Discovery Studio have a superior fidelity to the gold standard high-resolution crystal structures.

References

  1. Marc Fasnacht, Ken Butenhof, Anne Goupil-Lamy, Francisco Hernandez-Guzman, Hongwei Huang, and Lisa Yan, Proteins, 2014, in press, “Automated antibody structure prediction using BIOVIA tools: Results and best practices.”
  2. Almagro JC, Teplyakov A, Luo J, Sweet W, Kondagantlil S, Hernadez-Guzman F, Stanfield, Gilliland GL. Second antibody modeling assessment (AMA-II). Proteins, in press.
  3. Monoclonal Antibodies, 2010, Datamonitor Ltd, Report code: HC0029-002.

About BIOVIA, Dassault Systèmes

BIOVIA™ provides global, collaborative product lifecycle experiences to transform scientific innovation. Our solutions create an unmatched scientific management environment that can help science-based organizations create and connect biological, chemical and material innovations to improve the way we live.

The industry-leading BIOVIA portfolio integrates the diversity of science, experimental processes and information requirements, end-to-end, across research, development, QA/QC and manufacturing. Capabilities include Scientific Informatics, Molecular Modeling/Simulation, Data Science, Laboratory Informatics, Formulation Design, BioPharma Quality & Compliance and Manufacturing Analytics.

BIOVIA is committed to enhancing and speeding innovation, increasing productivity, improving quality and compliance, reducing costs and accelerating product development for customers in multiple industries.


Sponsored Content Policy: News-Medical.net publishes articles and related content that may be derived from sources where we have existing commercial relationships, provided such content adds value to the core editorial ethos of News-Medical.Net which is to educate and inform site visitors interested in medical research, science, medical devices and treatments.

Last updated: Oct 11, 2022 at 9:46 AM

Citations

Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    BIOVIA, Dassault Systèmes. (2022, October 11). Why structural models of antibody targets are needed. News-Medical. Retrieved on December 03, 2022 from https://www.news-medical.net/whitepaper/20221011/Why-structural-models-of-antibody-targets-are-needed.aspx.

  • MLA

    BIOVIA, Dassault Systèmes. "Why structural models of antibody targets are needed". News-Medical. 03 December 2022. <https://www.news-medical.net/whitepaper/20221011/Why-structural-models-of-antibody-targets-are-needed.aspx>.

  • Chicago

    BIOVIA, Dassault Systèmes. "Why structural models of antibody targets are needed". News-Medical. https://www.news-medical.net/whitepaper/20221011/Why-structural-models-of-antibody-targets-are-needed.aspx. (accessed December 03, 2022).

  • Harvard

    BIOVIA, Dassault Systèmes. 2022. Why structural models of antibody targets are needed. News-Medical, viewed 03 December 2022, https://www.news-medical.net/whitepaper/20221011/Why-structural-models-of-antibody-targets-are-needed.aspx.

Other White Papers by this Supplier