TL;DR
Researchers have developed a novel approach combining protein language models (pLLMs) and homology searches to discover highly active kynureninases, enzymes that can degrade immunosuppressive metabolites in the tumor microenvironment. This method led to the identification of a new enzyme, K3, which demonstrates significantly improved catalytic activity and anti-tumor efficacy compared to previously known enzymes, offering promising potential for cancer immunotherapy.
What Problem Does This Paper Address?
Cancer immunotherapies, while promising, often struggle with low response rates for solid tumors. A key challenge is the immunosuppressive tumor microenvironment (TME), which hinders the effectiveness of immune cells. One major contributor to this immunosuppression is the metabolite L-kynurenine (L-KYN). Reducing L-KYN levels in the TME is a potential therapeutic strategy. While enzymes called kynureninases (KYNases) can degrade L-KYN, finding highly efficient and therapeutically viable ones has been difficult. Traditional methods of searching for homologous enzymes in vast protein databases are time-consuming and require extensive experimental validation. This paper aims to address this by developing a more efficient and predictive method for identifying potent KYNases.
Methods
The study employed a two-pronged approach:
- Homology Search and pLLM-based Prediction:
- They began by searching a massive protein database (3.2 billion sequences) for homologs of a known, highly active KYNase (Pf-K) using MMseqs2. This yielded over 10,000 potential candidates.
- To manage this large set and predict their catalytic efficiency (kcat/KM), they trained a regression model. This model used embeddings generated by the ESM pre-trained protein language model, which captures rich information about protein sequences.
- The model was trained on a dataset of 159 experimentally measured kcat/KM values.
- This combined approach allowed them to rank the identified homologous sequences based on their predicted enzymatic activity.
- Experimental Validation:
- The top-ranked predicted KYNases were then experimentally validated. This involved:
- Catalytic Activity Assays: Measuring the consumption of L-KYN to determine actual kinetic parameters (kcat and kcat/KM).
- Molecular Dynamics (MD) Simulations: Investigating the molecular interactions between the enzymes and L-KYN to understand the basis of their catalytic activity and binding affinity.
- In vitro T Cell Activation Assays: Assessing the ability of the enzymes to reverse L-KYN-induced immunosuppression on T cells.
- In vivo Anti-tumor Efficacy Studies: Testing the therapeutic potential of the most promising enzymes in mouse models of melanoma and colon cancer.
Datasets
- Training Data for pLLM Regressor: 159 protein sequences with experimentally measured kcat/KM values. This dataset included sequences from Homo sapiens, Mus musculus, Acinetobacter baumannii, Cyclobacterium marinum, Chlamydophila pecorum, and Pseudomonas fluorescens, as well as modified H. sapiens sequences from a patent.
- Homology Search Database: A database containing 3.2 billion protein sequences from cultivated and metagenomic sources (BFD and Metaclust2).
- Cell Lines: B16-F10 melanoma and CT26 colon cancer cell lines.
- Animal Models: C57BL/6 mice and BALB/c mice.
Key Insights and Ideas
- pLLMs as powerful enzyme discovery tools: The study demonstrates that protein language models, when combined with homology searches, can effectively predict and prioritize enzymes with high catalytic activity, even with limited experimental training data. This significantly accelerates the discovery process.
- The TME and L-KYN as a therapeutic target: The paper reinforces the importance of targeting the immunosuppressive TME by degrading L-KYN as a viable strategy for cancer immunotherapy.
- K3: A superior KYNase candidate: The discovery of K3, a novel KYNase, with significantly higher catalytic activity and in vivo anti-tumor efficacy than previously known enzymes, is a major breakthrough.
- Structural basis for enhanced activity: MD simulations revealed that K3's active site is more tightly packed, leading to improved substrate binding and more stable enzyme-substrate interactions, explaining its enhanced catalytic performance.
- PEGylation for therapeutic enhancement: PEGylation of K3 (PEG-K3) was shown to maintain its activity and potentially shield it from immune recognition, making it a more suitable candidate for therapeutic development.
Key Results
- The pLLM-based predictor achieved a Spearman correlation of 0.813 with experimentally determined kcat/KM values, indicating high predictive accuracy.
- Four top-ranked KYNases were identified and experimentally validated.
- The most active KYNase, K3, exhibited approximately a 2-fold higher turnover number (kcat) compared to Pf-K.
- In vivo studies showed that PEGylated K3 (K3-PEG) significantly inhibited tumor growth in mouse models:
- Melanoma model: Tumor growth inhibition (TGI) of 63.9% for K3-PEG compared to 37.8% for Pf-K-PEG.
- Colon cancer model: Similar promising results were observed.
- K3-PEG treatment did not cause significant body weight changes in mice, suggesting a good safety profile.
Limitations and Future Work
- Immunogenicity Risk: The authors acknowledge that K3, being of bacterial origin, could pose an immunogenicity risk in vivo. However, they suggest that PEGylation may mitigate this risk.
- Acidic pH Activity: K3 shows lower catalytic activity at acidic pH conditions found in tumors compared to neutral pH. While it retains sufficient activity, this could be a factor in its therapeutic efficacy.
- Further Optimization: While K3 shows great promise, further optimization of its properties, such as stability and delivery, might be explored.
- Broader Application: The developed pLLM-based prediction framework could be applied to discover other highly active enzymes for various therapeutic applications.
Summary updated