🦠 microbiology

Structure-guided generative design of peptides targeting the FtsQBL divisome complex inhibit Escherichia coli cell division.

This study demonstrates that combining interpretable interface mapping with generative AI design enables the creation of structure-guided peptides that mimic native interactions to disrupt the FtsQBL divisome complex, thereby inhibiting cell division and growth in *Escherichia coli*.

Original authors: Remont, P., Liu, X., Croci, F., Mechaly, A., Karimova, G., Nguyen, M.-H., Guijarro, J. I., Davi, M., Guyon, C., Ciambur, C. B., Agou, F., Boucharlat, A., Ahmed, H., Chiaravalli, J., Ladant, D., Speran

Published 2026-03-01

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Remont, P., Liu, X., Croci, F., Mechaly, A., Karimova, G., Nguyen, M.-H., Guijarro, J. I., Davi, M., Guyon, C., Ciambur, C. B., Agou, F., Boucharlat, A., Ahmed, H., Chiaravalli, J., Ladant, D., Sperandio, O.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine a bacterial cell as a bustling factory that needs to split in half to create a new factory. To do this, it relies on a massive, complex construction crew called the divisome. This crew is made up of many different workers (proteins) who must hold hands and link up perfectly to build a wall that cuts the factory in two.

One of the most critical links in this chain is a trio of workers: FtsQ, FtsB, and FtsL. Think of FtsQ as a central hub or a docking station. FtsB and FtsL are two specialized workers who must zip their "Velcro" strips onto FtsQ to keep the construction crew together. If they can't zip on, the factory falls apart, the wall never gets built, and the bacteria can't reproduce.

For decades, scientists have struggled to stop bacteria like E. coli because they are hard to reach, and the "Velcro" spots where these proteins stick together are flat and wide, making them impossible to block with traditional tiny drug molecules.

Here is how this paper solves that problem using Artificial Intelligence:

1. The AI Architect (RFdiffusion)

Instead of trying to invent a tiny key to fit a lock, the researchers asked an AI architect named RFdiffusion to design a custom "mimic."

The Goal: Create a tiny, synthetic piece of string (a peptide) that looks and feels exactly like the natural "Velcro" strips of FtsB and FtsL.
The Strategy: The AI was told, "Here is the exact shape of the docking station (FtsQ). Here is the specific pattern of the natural workers (FtsB/FtsL). Now, design a new, short string that can snap onto that spot just as tightly, but is made of different materials."
The Result: The AI generated hundreds of designs. It found that the best designs were shaped like a hairpin (a folded loop), which allowed them to lock perfectly into the FtsQ docking station.

2. The "Trojan Horse" Peptides

The researchers synthesized these AI-designed strings (peptides) and tested them.

The Test: They put these strings inside bacteria. Because the strings were designed to look exactly like the natural workers, they successfully "stuck" to the FtsQ hub.
The Effect: By occupying the docking spot, the peptides blocked the real FtsB and FtsL workers from attaching. It was like putting a piece of gum in the keyhole of a door. The construction crew couldn't assemble, the factory couldn't split, and the bacteria grew into long, tangled chains (filaments) before dying.

3. The "Permeability" Problem and the Solution

There was a catch: Bacteria have a tough outer shell (like a fortress wall) that keeps big things out. The initial peptides were too "negative" in charge to get through this wall.

The Fix: The researchers tweaked the peptides, adding positive charges (like giving them a magnetic pull) to help them sneak through the fortress wall.
The Breakthrough: One specific peptide, Bd5.1-7, was a superstar. It could penetrate the bacterial wall, find the FtsQ hub, jam the works, and stop the bacteria from growing. It worked at very low doses and didn't hurt human cells (tested on human lung cells), meaning it's a safe candidate for a new antibiotic.

4. Proof in the Pudding (X-ray and NMR)

To make sure the AI didn't just guess right but actually built the right shape, the scientists used high-tech cameras:

X-ray Crystallography: They took a 3D photo of the peptide stuck to the protein. It showed the peptide was sitting exactly where the AI predicted, locking into place with extra "grip" that the natural proteins didn't even have.
NMR (Magnetic Resonance): They checked the peptide in a liquid solution and found it was already folded into the perfect shape before it even touched the bacteria. This means it doesn't have to waste energy folding itself once it arrives; it's ready to fight immediately.

Why This Matters

This paper is a blueprint for the future of fighting superbugs.

Old Way: Try millions of random chemicals to see if one works (like throwing darts in the dark).
New Way: Use AI to read the blueprint of a bacteria's weakness, design a custom "molecular key" to jam it, and verify it with super-precise cameras.

The researchers even showed they could customize these keys for different types of bacteria (like Pseudomonas), suggesting we could create "narrow-spectrum" antibiotics that kill only the bad bacteria without hurting the good ones in our gut.

In short: They used AI to design a tiny, custom-made "molecular wedge" that jams the gears of a bacterial factory, stopping it from multiplying. It's a powerful new tool in the fight against antibiotic resistance.

1. Problem Statement

The emergence of antimicrobial resistance (AMR) in Gram-negative bacteria, particularly Escherichia coli, poses a critical global health threat. There is an urgent need for novel antibiotics with new mechanisms of action.

Target: The bacterial divisome, specifically the FtsQBL complex (FtsQ, FtsB, and FtsL), is essential for cell division and lacks eukaryotic counterparts.
Challenge: The FtsQBL complex assembles via a $\beta$ -strand-mediated interface in the periplasm. Disrupting this Protein-Protein Interaction (PPI) is difficult because these interfaces are large, flat, and lack deep catalytic pockets, making them poor targets for conventional small molecules.
Goal: To design peptide mimetics that can specifically bind to the FtsB-binding site (FbBS) on FtsQ, disrupting the FtsQ-FtsB/FtsL interaction and inhibiting cell division.

2. Methodology

The study employed a hybrid approach combining interpretable structural analysis with AI-driven generative design (RFdiffusion).

Target Identification & Constraint:
- The researchers used InDeep (a tool for interface mapping) to identify the critical $\beta$ -strand motif TFYRL within the native FtsQ-FtsB interface.
- This motif was used as a structural "seed" or constraint to guide the generative model.
Generative Design (RFdiffusion):
- Motif-Scaffolding: The TFYRL motif was pre-positioned in the FtsQ binding groove. RFdiffusion was then used to generate diverse peptide backbones (scaffolds) flanking this fixed motif.
- Sequence Design: ProteinMPNN was used to assign sequences to the generated backbones while keeping the FtsQ receptor fixed.
- Filtering: Designs were filtered using AlphaFold2 to ensure low predicted alignment error (pAE < 10), lack of steric clashes, and continuous $\beta$ -strand pairing.
- Initial Selection: 8 candidates were selected: 3 manually designed mimetics (Bd1–Bd3) and 5 RFdiffusion-generated designs (Bd4–Bd8).
Experimental Validation Pipeline:
- In Vivo: Bacterial Two-Hybrid (BACTH) assays to test interactions with full-length membrane-bound FtsQ and the isolated periplasmic domain.
- In Vitro: Biolayer Interferometry (BLI) and Fluorescence Polarization (FP) to measure binding affinity ( $K_D$ and $IC_{50}$ ).
- Structural Validation: X-ray crystallography of the peptide-FtsQ complex and NMR spectroscopy to determine solution structures.
- Functional Assays: Overexpression of peptides in E. coli to observe morphological changes (filamentation) and growth inhibition assays on the permeable lptD4213 strain.
Optimization:
- A second round of RFdiffusion was performed on the best initial hits (Bd4, Bd5, Bd8) to generate optimized variants (e.g., Bd5.1, Bd8.3).
- Charge engineering was applied to improve bacterial uptake (modifying net charge from negative to positive).

3. Key Contributions

Proof of Concept for Generative PPI Inhibition: Demonstrated that RFdiffusion, when constrained by known structural hotspots, can successfully design $\beta$ -hairpin peptides that mimic native $\beta$ -augmentation interfaces.
Structural Validation: Provided atomic-resolution evidence (X-ray and NMR) that AI-designed peptides adopt the predicted binding geometry and pre-organized $\beta$ -hairpin conformations in solution.
Mechanistic Insight: Showed that the designed peptides act as "cumulative mimics," binding to a composite interface normally formed by two distinct proteins (FtsB and FtsL) and introducing new stabilizing interactions (e.g., a tryptophan residue in a hydrophobic pocket) not present in the native complex.
Species Selectivity: Demonstrated that the design pipeline can be adapted to generate binders specific to E. coli FtsQ versus Pseudomonas aeruginosa FtsQ, suggesting a path toward narrow-spectrum antibiotics.

4. Key Results

Binding Specificity:
- RFdiffusion-derived peptides (Bd5, Bd6, Bd7, Bd8) successfully bound to the FtsQ periplasmic domain in BACTH assays, whereas manual mimetics (Bd1–Bd3) did not.
- Mutagenesis confirmed that the peptides bind specifically to the C-terminal $\beta$ -strand of FtsQ. Alanine substitution of key aromatic residues (F15/Y16 in Bd8, corresponding to F84/Y85 in FtsB) abolished binding.
- Competition assays showed that the native FtsB/FtsL dimer, but not individual proteins, effectively competed with the peptides for FtsQ binding.
Biophysical Affinity:
- Initial peptides (Bd4, Bd5, Bd8) showed micromolar affinity ( $K_D \approx 0.3–1.4 \mu M$ ).
- Optimized peptides (Bd5.1, Bd8.3) showed improved affinity ( $IC_{50} \approx 0.1–0.5 \mu M$ ).
- Charge-optimized peptide (Bd5.1-7): By increasing the net positive charge, this variant achieved an $IC_{50}$ of 0.109 $\mu M$ and a Minimum Inhibitory Concentration (MIC) of 6.25 $\mu M$ in the permeable E. coli lptD4213 strain.
Structural Insights:
- X-ray Crystallography: The structure of Bd4.3 bound to FtsQ confirmed the $\beta$ -strand augmentation geometry. Crucially, a designed Tryptophan residue inserted into a hydrophobic pocket adjacent to the native interface, providing extra stability not found in the native complex.
- NMR: Peptides Bd4.3 and Bd5.1-7 adopted pre-organized antiparallel $\beta$ -hairpin structures in solution, minimizing the entropic penalty upon binding. The solution structure of Bd4.3 matched the crystal structure with an RMSD of 1.2 Å.
Biological Activity:
- Overexpression of optimized peptides (Bd5.1, Bd8.3) in E. coli caused severe filamentation (failure to divide), phenocopying divisome disruption.
- Synthetic peptide Bd5.1-7 inhibited growth of the permeable E. coli strain and induced a chaining phenotype.
- Selectivity: Bd5.1-7 showed no cytotoxicity against human lung fibroblasts (MRC5) and did not bind to P. aeruginosa FtsQ.

5. Significance

New Antibiotic Paradigm: This work establishes a viable pipeline for targeting "undruggable" PPIs in Gram-negative bacteria using AI-generated peptide mimetics.
Beyond Small Molecules: It demonstrates that $\beta$ -strand augmentation interfaces, previously considered difficult to target, can be effectively disrupted by designed peptides that mimic and extend native interactions.
Closed-Loop Design: The study integrates predictive modeling (RFdiffusion/AlphaFold), atomic-resolution structural validation (X-ray/NMR), and functional cellular assays, creating a robust "design-to-function" framework.
Future Potential: While current peptides require permeabilized strains for activity, the success of this approach paves the way for developing stabilized peptidomimetics or delivery systems to treat resistant Gram-negative infections. The ability to generate species-specific binders offers a strategy to spare the human microbiome.