(T1330-02-06) Enhancing Immunogenicity Risk Assessment: Leveraging ISPRI for Bispecific Antibody Analysis

Purpose: Bispecific antibodies are recombinant molecules designed to target two different epitopes simultaneously. Several engineering techniques have been developed to facilitate their production, including knob-into-holes Fc and Fab mutations. We performed a comparative retrospective immunogenicity analysis of a panel of bispecific antibodies using in silico prediction tools hosted on ISPRI (Interactive Screening and Protein Reengineering Interface) and compared our results to published data from Wen-Ting et al. Wen-Ting et al. assessed the immunogenic risk of the same panel of GMP-produced bi-specific antibodies by employing both computational and experimental methods intended to model various steps in the cellular pathway leading to the production of anti-drug antibodies (ADA). This process includes a DC internalization assay to capture the uptake of the protein therapeutic by antigen presenting cells, both an in silico prediction tool and an MHC-associated peptide proteomics (MAPPs) assay to model the presentation of the processed peptides by MHC-II, and T cell assays to assess the activation and proliferation of naïve T cells. An immunogenicity risk was assigned to each bispecific antibody and monospecific counterpart based on the results of these individual methods. Lastly, they assigned an overall immunogenicity risk considering the collective results from all methods.

Methods: To perform this retrospective analysis, we began with a global immunogenicity assessment of each bispecific antibody and monospecific counterpart. We first screened each sequence for Class II (HLA-DR) restricted HLA ligands using EpiMatrix to identify putative T cell epitopes. The EpiMatrix scores of the individual T cell epitopes were aggregated and normalized to produce global immunogenicity scores for each antibody. Antibodies and antibody-like molecules frequently contain known regulatory epitopes (aka Tregitopes). The presence of regulatory epitopes can suppress ADA responses. We discounted the scores of known Tregitopes to yield improved measures of global immunogenic potentials. Finally, we plotted the Tregitope-adjusted EpiMatrix Scores on the antibody immunogenicity scale along with a selection of well-known benchmarks for comparison. Peptide epitopes that are frequently found within the human proteome (other than Tregitopes) are less likely to drive ADA responses. Discounting the EpiMatrix Scores of peptide epitopes frequently found in the human proteome can also improve the quality of our assessment of global immunogenic potentials. To complete our global analysis, we screened all epitopes identified by EpiMatrix against the human proteome using the JanusMatrix algorithm which calculates global JanusMatrix scores for proteins based on their human-like content. Next, we proceeded with a regional immunogenicity analysis of each antibody. Using the ClustiMer algorithm, we identified epitope clusters (short segments of amino acids that contain unusually high numbers of putative T cell epitopes) contained within each sequence. Finally, we evaluated both the EpiMatrix and JanusMatrix results of each epitope cluster. Based on a collective analysis using ISPRI’s suite of tools, we assigned an immunogenicity risk to each monospecific and bispecific antibody to compare to the risk categories assigned by Wen-Ting et al. to confirm alignment between the two analysis approaches (Image 1). Using ISPRI, we generated epitope maps providing a visual of all epitope content and clusters contained within each of the sequences. We compared these to figures produced by Wen-Ting et al. highlighting all MAPPs peptide spectrum matches as well as the immunogenic peptides detected using NetMHCIIpan-4.0 and produced figures overlaying the epitope content identified using ISPRI with their results (Image 2).

Results: After reviewing the collective results from ISPRI and assigning immunogenicity risk to each antibody, we found that our risk categories aligned with those assigned by Wen-Ting et al after their comprehensive in silico and in vitro multi-fold approach to immunogenicity risk assessment for six out of the seven antibodies (Image 1). Additionally, based on the retrospective immunogenicity analysis, EpiMatrix aligned more closely with the results of the MAPPs assay as compared to NetMHCIIpan-4.0. Of the immunogenic peptides identified using ISPRI, 92% overlap with the peptides detected by the MAPPs assay compared to 59% overlap of the peptides identified using NetMHCIIpan-4.0.

Conclusion: In silico immunogenicity analysis using the ISPRI platform is a useful supplement to experimental methods which can often be extremely time-consuming and expensive. This analysis also highlights the potential for in silico immunogenicity to identify liabilities that can be further assessed in vitro prior to clinical trials.

References: Wen-Ting K. Tsai, Yinyin Li, Zhaojun Yin, Peter Tran, Qui Phung, Zhenru Zhou, Kun Peng, Dan Qin, Sien Tam, Christoph Spiess, Jochen Brumm, Manda Wong, Zhengmao Ye, Patrick Wu, Sivan Cohen & Paul J. Carter (2024) Nonclinical immunogenicity risk assessment for knobs-into-holes bispecific IgG1 antibodies, mAbs, 16:1, 2362789, DOI: 10.1080/19420862.2024.2362789

This image shows a table of the immunogenicity risk assigned by Wen-Ting et. al as well as those assigned based on our retrospective comparative analysis using ISPRI.

This figure is an overlay of our epitope maps generated by ISPRI and the figures produced by Wen Ting et al. showing both the MAPPs elution data and the NetMHCIIpan-4.0 results.