Design of Cytotoxic T Cell Epitopes by Machine Learning of Human Degrons

Nicholas L. Truex, Somesh Mohapatra, Mariane B. Melo, Jacob Rodriguez, Na Li, Wuhbet Abraham, Deborah Sementa, Fayçal Touti, Derin B. Keskin, Catherine J. Wu, Darrell J. Irvine, Rafael Gómez‐Bombarelli, Bradley L. PenteluteACS 2024 · DOI

TL;DR

This paper introduces a machine learning model to predict how well protein fragments (epitopes) are degraded by the cell's proteasome, a crucial step for immune cells to recognize and attack them. By understanding and controlling this "degron" activity, they demonstrate a way to design more effective vaccine epitopes that can significantly boost T cell responses against cancer and pathogens.

What Problem Does This Paper Address?

Therapeutic vaccines, especially those targeting cancer and pathogens, rely on presenting specific protein fragments (epitopes) to the immune system to trigger a response from cytotoxic T cells. A major hurdle in developing effective vaccines is that these epitopes often aren't processed efficiently within the cell, limiting their presentation to T cells and thus reducing vaccine efficacy. This inefficiency stems from a lack of understanding about how the sequence of an epitope influences its degradation by the cell's proteasome, a cellular machinery responsible for breaking down proteins. The authors aim to bridge this gap by developing a predictive model for this proteasomal degradation, which they term the "C-terminal Degron Index" (CDI), and demonstrating its utility in designing more immunogenic vaccine epitopes.

Methods

The core of this work is a machine learning approach to predict proteasomal degradation. They trained a one-dimensional convolutional neural network on a dataset of human protein sequences and their experimentally determined degradation stabilities. The model takes a peptide sequence as input and outputs a C-terminal Degron Index (CDI) score, ranging from 0 to 100, where lower scores indicate higher degradation propensity.

To validate their model and explore the functional consequences of varying degradation, they employed a clever delivery system. They conjugated synthetic peptides, including known T cell epitopes, to the C-terminus of a non-toxic anthrax protein fragment (LFN). This LFN-peptide conjugate was then delivered into mammalian cells using another anthrax protein (PA), ensuring cytosolic access for the proteasome. They then used this system to evaluate T cell proliferation in response to peptides with different predicted CDI scores, both in vitro using dendritic cells and by retrospectively analyzing data from clinical trials.

Datasets

The primary dataset used for training the machine learning model consisted of 22,564 sequences from the C-termini of human proteins, along with their associated proteasomal stability data. This data was derived from previous studies that used DNA-encoded peptide libraries and measured GFP expression to infer degradation.

For experimental validation, they synthesized various peptide constructs, including:

  • LFN-OVA peptides (1-12 and 1G-8G): These peptides incorporated the ovalbumin (OVA) epitope (SIINFEKL) with different flanking sequences designed to yield a range of CDI scores.
  • CDILO and CDIHI peptides: These were selected from large randomized libraries based on their predicted low (CDILO) or high (CDIHI) degradation scores, using model antigens like OVA and gp100.
  • Clinical trial data: Retrospective analysis was performed on peptide sequences from three clinical trial studies for personalized cancer vaccines.

Key Insights and Ideas

  • Degrons as a Design Principle: The paper establishes that specific short amino acid sequences, known as "degrons," at the C-terminus of a protein fragment significantly influence its degradation by the proteasome. This provides a new lever for designing vaccine epitopes.
  • Machine Learning for Degradation Prediction: A deep learning model can accurately predict the proteasomal degradation propensity of peptide sequences, enabling the rational design of epitopes with desired degradation characteristics.
  • Low Degradation = High Immunogenicity: Counterintuitively, epitopes with low proteasomal degradation scores (meaning they are less likely to be broken down) led to higher T cell proliferation. This suggests that more intact epitopes are better presented to T cells.
  • Flanking Residues Matter: The sequences flanking the core T cell epitope play a critical role in dictating proteasomal degradation and, consequently, immunogenicity. This is a key area often overlooked in epitope design.
  • Anthrax Toxin System as a Powerful Tool: The PA/LFN delivery system proved effective for introducing peptides into the cytosol of antigen-presenting cells, allowing for direct assessment of proteasomal processing and subsequent T cell activation.

Key Results

  • The developed C-terminal Degron Index (CDI) model achieved a R² of 0.81 and a Pearson correlation of 0.90 in predicting experimental degradation scores.
  • Experiments showed that LFN-OVA peptides with lower CDI scores (indicating higher degradation) led to significantly increased T cell proliferation compared to those with higher CDI scores.
  • Analysis of clinical trial data suggested a correlation between low CDI values and CD8+ T cell activation in successful vaccine candidates.
  • The study demonstrated that by manipulating flanking residues to achieve low CDI scores, epitope immunogenicity could be enhanced, even when protein concentrations were kept constant.

Limitations and Future Work

  • In Vivo vs. In Vitro: While the study shows strong in vitro correlations, the authors acknowledge that further studies are needed to correlate these findings with in vivo antigen degradation and immune responses in whole animal models and clinical settings.
  • Vaccine Formulation: The influence of different vaccine formulations on antigen degradation in vivo remains an area for future investigation.
  • N-terminal Degrons: The current model focuses on C-terminal degrons. While N-terminal degradation pathways exist, they are more complex and were not the focus of this study.
  • Broader Epitope Types: The study primarily used model antigens like ovalbumin. Future work could explore the application of this approach to a wider range of tumor and pathogen-specific epitopes, as well as personalized neoantigens.
  • Mechanism of Presentation: While the study links degradation to T cell proliferation, further detailed mechanistic studies could elucidate the precise pathways of antigen processing and presentation influenced by these degron sequences.

Summary updated

Abstract

Antigen processing is critical for therapeutic vaccines to generate epitopes for priming cytotoxic T cell responses against cancer and pathogens, but insufficient processing often limits the quantity of epitopes released. We address this challenge using machine learning to ascribe a proteasomal degradation score to epitope sequences. Epitopes with varying scores were translocated into cells using nontoxic anthrax proteins. Epitopes with a low score show pronounced immunogenicity due to antigen processing, but epitopes with a high score show limited immunogenicity. This work sheds light on the sequence-activity relationships between proteasomal degradation and epitope immunogenicity. We anticipate that future efforts to incorporate proteasomal degradation signals into vaccine designs will lead to enhanced cytotoxic T cell priming by these vaccines in clinical settings.
BibTeX
@inproceedings{truex2024design,
  author = {Nicholas L. Truex and Somesh Mohapatra and Mariane B. Melo and Jacob Rodriguez and Na Li and Wuhbet Abraham and Deborah Sementa and Fayçal Touti and Derin B. Keskin and Catherine J. Wu and Darrell J. Irvine and Rafael Gómez‐Bombarelli and Bradley L. Pentelute},
  title = {{Design of Cytotoxic T Cell Epitopes by Machine Learning of Human Degrons}},
  booktitle = {ACS},
  year = {2024},
  doi = {10.1021/acscentsci.3c01544},
}
Figure