De novo designed cyclic MC4R peptide agonist reduces food intake in mice

This study establishes a generalizable end-to-end workflow for de novo peptide drug discovery by using deep learning to design, optimize, and validate a novel cyclic MC4R agonist that achieves high potency and significantly reduces food intake in mice.

The Gist

Imagine you are trying to build a custom key to open a very specific, complicated lock (the MC4R receptor) that controls hunger in the body. Usually, scientists try to copy keys they already know exist in nature. But this paper is about building a brand-new key from scratch using a super-smart computer program, without copying anything that nature has already made.

Here is how the researchers did it, step-by-step:

1. The Computer Architect
First, the team used a powerful AI tool (called AlphaFold2) to act like a digital architect. They asked the computer to "hallucinate" or imagine over 5,000 different shapes for a key. Some were straight lines, and some were shaped like loops (cyclic). The goal was to find shapes that would fit perfectly into the "lock" of the hunger receptor.

2. The First Test Drive
They picked a small group of these computer-designed keys to test in the lab. It was a bit like a talent show:

• 74% of the straight-line keys actually worked.
• 23% of the loop-shaped keys worked.
  One of the loop-shaped keys was a star performer. It was able to turn the lock and send a signal, even though it looked completely different from the natural keys the body usually uses. It was a bit weak at first, but it proved the concept worked.

3. The "Tuning" Workshop
Once they had a working prototype, they didn't stop there. They treated it like a race car that needed tuning. They ran thousands of tiny experiments to swap out different parts of the key (changing one letter in the code at a time) to see which changes made it faster and stronger. They also added "stabilizers" to help the key last longer in the body.

Through this process, they discovered a new secret pattern (a specific sequence of letters called "APWR") that made the key work better. They found one specific change—swapping a part for a "Proline" piece—that made the key incredibly powerful. This new version, called E5P, was about 50 times stronger than the first working prototype.

4. The Real-World Test
Finally, they took this super-charged key and tested it on mice. They gave the mice a small dose of the key directly into their brains. The result? The mice ate significantly less food right away.

The Big Picture
This paper isn't just about one specific drug for hunger. It's a demonstration of a new "assembly line." The researchers showed that you can go from a computer idea to a real, working medicine in a mouse, all without relying on nature's existing designs. They proved that with the right tools, we can invent entirely new types of keys to unlock biological processes that we couldn't reach before.

Technical Summary

Technical Summary: De Novo Designed Cyclic MC4R Peptide Agonist

Problem Statement
While deep learning-based structure prediction has enabled the design of peptide ligands independent of natural scaffolds, the field faces a significant bottleneck: most computationally generated peptides are not advanced beyond initial activity measurements. Consequently, the critical pathways for drug-like optimization and in vivo validation remain underexplored. There is a need for an end-to-end workflow that bridges the gap between de novo design and the generation of optimized, pharmacologically active peptides.

Methodology
The authors established a comprehensive workflow for the discovery and maturation of peptide agonists, utilizing the melanocortin-4 receptor (MC4R) as a model target. The process involved the following stages:

1. De Novo Design: Using an AlphaFold2-based hallucination protocol implemented in ColabDesign, the team generated over 5,000 candidate peptides, comprising both linear and head-to-tail cyclic structures, specifically directed toward the MC4R orthosteric pocket.
2. Initial Screening: A prioritized subset of these candidates underwent functional screening to assess activity.
3. Systematic Maturation: For active candidates, the team employed a multi-pronged optimization strategy:
   • Deep Mutational Scanning: To map sequence-activity relationships.
   • Half-life Extender Conjugation Scanning: To assess stability and pharmacokinetic potential.
   • Combinatorial Optimization Library: To further refine potency.
   • Data-Driven Analysis: Used to interpret the results of the above scans and guide further design.
4. In Vivo Validation: The most potent variant was administered centrally to mice to evaluate its effect on acute food intake.

Key Results

• Hit Rate: Functional screening revealed measurable activity in 74% of the linear peptides and 23% of the cyclic peptides.
• Initial Discovery: The team identified a cyclic agonist with an EC₅₀ of 340 nM against the human MC4R (hMC4R). Notably, this peptide lacked the canonical melanocortin activation motif.
• Motif Identification: Optimization efforts identified an alternative activation motif centered on an APWR segment.
• Potency Optimization: Through systematic mutagenesis, a single-site variant with a proline substitution at position 5 (E5P) was generated. This variant demonstrated a significantly improved EC₅₀ of 6.7 nM against hMC4R.
• In Vivo Efficacy: Central administration of the E5P variant (10 nmol) successfully reduced acute food intake in mice, providing proof of concept for the in vivo activity of the de novo designed peptide.

Significance and Claims
The paper claims to demonstrate a generalizable "design-to-validation" strategy. This approach successfully converts de novo peptide designs into optimized, pharmacologically active peptides, addressing the gap between computational generation and therapeutic application. Furthermore, the work expands the known chemical space of MC4R agonists by identifying chemotypes that function beyond the constraints of endogenous melanocortins, specifically utilizing non-canonical activation motifs. The study validates the utility of deep learning hallucination protocols coupled with rigorous experimental maturation for drug discovery.