AI-Guided Structure-Aware Modeling and Thermal Proteomics Reveal Direct Demethylzeylasteral-ACLY Interaction

This study integrates thermal proteomics with a state-of-the-art graph neural network to identify Demethylzeylasteral as a direct ACLY inhibitor, validating its micromolar binding and demonstrating its therapeutic efficacy in treating psoriasis by reversing keratinocyte hyperproliferation and suppressing lipid metabolic reprogramming.

The Gist

Imagine you have a massive, chaotic library filled with millions of books (these are natural products, like plants and herbs used in medicine). You know some of these books contain life-saving stories, but the problem is that the library doesn't have a card catalog. You don't know which book talks to which specific character in the human body to create a healing effect. Finding the right book for the right character is like trying to find a specific needle in a haystack while wearing blindfold.

This paper describes a brilliant new way to solve that mystery using a "super-sleuth" team made of two parts: a high-tech detective and a super-smart AI.

The Two Super-Sleuths

1. The Heat Detective (Thermal Proteomics):
   Imagine you want to see if two puzzle pieces fit together. If you heat them up, they might fall apart. But if they are locked together tightly, they stay stuck even when things get hot.
   The scientists used a technique that gently "heats up" the proteins inside cells. If a medicine (like a natural plant extract) is holding onto a specific protein, that protein won't fall apart as easily as the others. This helps them spot which proteins are being "hugged" by the medicine.

2. The Crystal Ball (HoloGNN AI):
   While the heat detective gives clues, there are still too many possibilities. Enter HoloGNN, a super-smart computer brain (an Artificial Intelligence). Think of this AI as a master architect who has studied millions of 3D puzzles. It looks at the shape of the medicine and the shape of the proteins and predicts, "Hey, these two fit together perfectly!" It filters out the noise and points the scientists to the most likely matches.

The Big Discovery

The team used this "Heat + AI" combo to test 50 different natural medicines. They found something exciting: a plant compound called Demethylzeylasteral was directly hugging a protein called ACLY.

Think of ACLY as a factory manager in your cells who is in charge of making fat (lipids). In diseases like psoriasis (a painful skin condition), this manager goes into overdrive, making too much fat and causing inflammation.

The scientists proved that Demethylzeylasteral acts like a brake pedal for this manager. It locks onto ACLY, stops it from making too much fat, and calms the cell down.

The Real-World Test

To make sure this wasn't just a lab trick, they tested it on mice with a skin condition that looks like human psoriasis.

• Before treatment: The mice had red, angry, thick skin (hyperproliferation).
• After treatment: The skin calmed down, the inflammation went away, and the cells stopped growing out of control.

The Bottom Line

This paper is a story about how we can finally stop guessing how natural medicines work. By combining a clever heat experiment with a futuristic AI, the scientists didn't just find what a plant medicine does; they found the exact lock it fits into inside our bodies.

They discovered that a specific plant compound works by hitting the "off switch" on a fat-making machine in our cells, which helps heal skin diseases. This is a huge step forward because it turns "folk remedies" into precise, science-backed medicines with a clear map of how they work.

Technical Summary

1. Problem Statement

The identification of direct molecular targets for bioactive natural products is a persistent bottleneck in chemical biology and drug discovery. Traditional methods often struggle to distinguish direct binding events from downstream indirect effects, leading to ambiguous target deconvolution. There is a critical need for scalable, high-confidence frameworks that can systematically prioritize and validate protein-ligand interactions, particularly for structurally complex natural products.

2. Methodology

The authors developed an integrated experimental-computational closed-loop framework designed to bridge the gap between machine learning prediction and experimental validation:

• Computational Component (HoloGNN):
  • The study introduces HoloGNN, a holistic Graph Neural Network designed for structure-aware modeling.
  • Unlike traditional models, HoloGNN incorporates holistic structural information to predict protein-ligand binding affinities.
  • The model was benchmarked against the PDBbind dataset, where it demonstrated state-of-the-art (SOTA) performance in predicting binding interactions.
• Experimental Component (Matrix-Augmented Thermal Proteomics):
  • The team utilized a matrix-augmented version of Thermal Proteome Profiling (TPP). This technique monitors protein thermal stability shifts across a proteome upon ligand binding, allowing for the detection of direct interactions in a complex cellular environment.
• Validation Pipeline:
  • Prioritization: The framework screened 50 structurally diverse natural products to identify high-probability targets.
  • Orthogonal Validation: Predicted hits were subjected to rigorous biochemical assays, including binding affinity measurements and enzymatic inhibition assays.
  • In Vivo & Omics Analysis: The lead compound was tested in an imiquimod-induced psoriasis-like mouse model. Mechanistic insights were further derived using single-cell transcriptomics to analyze cellular heterogeneity and metabolic reprogramming.

3. Key Contributions

• Novel AI Framework: The development and validation of HoloGNN, which sets a new benchmark for structure-based protein-ligand interaction prediction, specifically tailored for natural product discovery.
• Integrated Strategy: The creation of a scalable, closed-loop strategy that seamlessly combines thermal proteomics (for unbiased target fishing) with deep learning (for prioritization), significantly reducing the search space for target identification.
• Target Discovery: The successful identification of Demethylzeylasteral as a direct binder of ATP Citrate Lyase (ACLY), a key enzyme in lipid metabolism, resolving a previously uncharacterized interaction for this natural product.

4. Key Results

• Model Performance: HoloGNN achieved state-of-the-art accuracy on the PDBbind benchmark, validating its utility for predicting binding in complex biological contexts.
• Target Validation:
  • The framework identified Demethylzeylasteral as a specific interactor of ACLY.
  • Orthogonal biochemical assays confirmed micromolar binding affinity and significant enzymatic inhibition of ACLY.
• Therapeutic Efficacy:
  • In the imiquimod-induced psoriasis mouse model, Demethylzeylasteral treatment significantly reduced disease severity and suppressed the expression of pro-inflammatory cytokines.
• Mechanistic Insights:
  • Single-cell transcriptomics revealed that the drug reverses keratinocyte hyperproliferation.
  • The study established a causal link between ACLY inhibition and the suppression of lipid metabolic reprogramming, which drives the inflammatory pathology in psoriasis.

5. Significance

This work represents a paradigm shift in chemical biology by demonstrating how AI-driven structure-aware modeling can be effectively coupled with proteomics to deconvolute the targets of natural products.

• Methodological Impact: It provides a reproducible, scalable blueprint for future drug discovery campaigns, moving beyond trial-and-error approaches to data-driven target identification.
• Biological Insight: It elucidates a novel therapeutic mechanism for treating inflammatory skin diseases (specifically psoriasis) via ACLY inhibition, offering a new angle on lipid metabolism as a therapeutic target.
• Clinical Potential: The identification of Demethylzeylasteral as an ACLY inhibitor opens new avenues for repurposing natural products in the treatment of metabolic and inflammatory disorders.