A High-Throughput Human Display Screen to Identify Target-Specific Binder Proteins via Chimeric Antigen Receptors

Introduction: Many disease-driving proteins are considered intractable by small molecules due to inaccessible binding pockets and largely disordered domains.1 Generative artificial intelligence-based computational methods have accelerated the development of these protein binders for structured and disordered targets. As a recent example, the RFDiffusion model has successfully leveraged denoising diffusion probabilistic modeling to design mini protein binders to structured target proteins.2 New language model-based methods, such as PepMLM and PepPrCLIP,3,4, now offer methods for peptide binder design conditioned purely on the target sequence. The evaluation of these models to design and down-select high-affinity and specific binders is of strong interest.

Chimeric antigen receptor (CAR)-T cell therapy has produced durable clinical responses for numerous cancer indications.5 The antigen binding domain of CARs confers target antigen specificity to activatory signals on CAR-expressing T cells.5,6 While the original CAR-based therapies were primarily targeted at the CD19 receptor, genetic engineering efforts have redirected CARs to new targets and making additional diseases tractable.7

In this study, we have repurposed the CAR platform to create a human display assay to screen target-binding proteins on the surface of CARs. To do this, we genetically engineered Jurkat76 cells to express lentiviral-integrated binder libraries in CARs, and screen them against Chinese Hamster Ovary (CHO) cells expressing their intended targets. By leveraging this platform, we have observed green fluorescence upon target binding in Jurkat76 cells, thus motivating downstream applications to evaluate new binder design models and develop CARs for diverse targets of interest.

Materials and

Methods: To generate a novel mammalian cell surface binding assay, a modular CAR-derived platform was designed (Figure 1). Briefly, antigen binding domains were cloned in lieu of conventional scFvs into a lentivirus construct expressing a 3rd generation CAR construct followed by a puromycin selection and a truncated NGFR co-expression marker. Jurkat76 T cells were chosen for proof-of-concept experiments due to their ease of use, homogeneity, and availability of reporter cell lines. Importantly, the NFAT signaling machinery is highly conserved in the Jurkat76 cell line8. We produced lentivirus and used it to transduce a novel Jurkat76 derived cell line engineered to express destabilized GFP under the control of an NFAT inducible promoter thereby linking GFP fluorescence intensity to CAR-Antigen binding affinity. To create target cell lines, we cloned antigens of interest into a custom Tet-On compatible lentivirus construct and used it to transduce Tet-On CHO cells (Figure 1A). Doxycycline-inducible target cells and binding domain expressing CAR-NFAT dsGreen Jurkat76 cells were cocultured in different concentrations of the doxycycline to evaluate the effect of antigen density on antigen-antigen binding domain interactions (Figure 1B).

Results, Conclusions, and Discussions: Results
To test this platform for binding interaction, we first chose CD19 as a target. The Tet-on CD19 CHO cells were co-cultured with the CD19 scFV Jurkat76 cell line. The Jurkat76 dsGFP reporter cell line showed a distinct dosage-dependent increase in dsGFP expression level (Figure 2).To further test the peptide-library integration efficiency, we utilized the co-expressed NGFR to evaluate the CAR-library expression level by staining of NGFR-APC antibody. The result indicated that the CAR-library NGFR expression level is comparable to the positive controls (Figure 3). These results indicate that the system may serve as a strong platform for further high-throughput screening.

To develop a preliminary high-throughput screen with CAR assay, we designed 1000 peptides for a critical target in glioblastoma multiforme, EGFRvIII, by leveraging both structure-based and sequence-based models (Table 1). 2–4We synthesized a Twist oligo pool encoding these peptides and then integrated the library into a dsGFP reporter Jurkat76 cell line. Concurrently, EGFRvIII was expressed on CHO cells under the control of doxycycline supplement. As a result, target-binder interaction activated the downstream of dsGFP reporter and all dsGFP positive cells were therefore sorted. The peptide coding sequences were further amplified for final Next Generation Sequencing (Figure 4). Finally, we calculated the significance vs. fold change for each peptide between the post-sorted and pre-sorted populations. Peptide enrichment was defined using a paired two-tailed DESeq2 test with Benjamini–Hochberg correction.9 As this work is still ongoing, we anticipate that these screening results will demonstrate the binding capability and the large-scale screening provide a platform for functional validation of binders generated from generative artificial intelligence algorithms.

Conclusions and discussion
Overall, in this work, we have laid the foundation for a high-throughput, human cell-based display screen to evaluate target binders to protein of interest. We are currently screening binders derived from diverse target-conditioned generative artificial intelligence algorithms, and can rapidly extend screening to other targets, including those that are heavily disordered. We anticipate that our CAR-based screening approach can enable the design of numerous therapeutic modalities dependent on viable target binders, as well as the development of new CAR-based therapies to diverse disease indications.

Acknowledgements (Optional): References (Optional)
1. Chen, T., Hong, L., Yudistyra, V., Vincoff, S. & Chatterjee, P. Generative design of therapeutics that bind and modulate protein states. Current Opinion in Biomedical Engineering 28, 100496 (2023).
2. Watson, J. L. et al. De novo design of protein structure and function with RFdiffusion. Nature 620, 1089–1100 (2023).
3. Chen, T. et al. PepMLM: Target Sequence-Conditioned Generation of Peptide Binders via Masked Language Modeling. ArXiv (2023) doi:10.48550/ARXIV.1706.03762.
4. Bhat, S. et al. De Novo Generation and Prioritization of Target-Binding Peptide Motifs from Sequence Alone. bioRxiv 2023.06.26.546591 (2023) doi:10.1101/2023.06.26.546591.
5. Sterner, R. C. & Sterner, R. M. CAR-T cell therapy: current limitations and potential strategies. Blood Cancer J. 11, 69 (2021).
6. De Marco, R. C., Monzo, H. J. & Ojala, P. M. CAR T Cell Therapy: A Versatile Living Drug. Int. J. Mol. Sci. 24, (2023).
7. Park, J. H., Geyer, M. B. & Brentjens, R. J. CD19-targeted CAR T-cell therapeutics for hematologic malignancies: interpreting clinical outcomes to date. Blood 127, 3312–3320 (2016).
8. Liang, Y., Zhou, Y. & Shen, P. NF-kappaB and its regulation on the immune system. Cell. Mol. Immunol. 1, 343–350 (2004).
9. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).