🧬 biology

Probing the sequence variability tolerance in a de novo α-helical barrel biocatalyst

This study demonstrates that combining AlphaFold2-guided RosettaRemodel and ProteinMPNN enables the successful engineering of a de novo α-helical barrel biocatalyst with a truncated structure and significantly improved soluble yield, while retaining its core architecture, thermal stability, and catalytic activity despite substantial sequence variability.

Original authors: Wael Elaily, Markus Braun, David Stoll, Morakot Chakatok, Matteo Aleotti, Birgit Grill, Aleksandar Bijelic, Mélanie Hall, Gustav Oberdorfer

Published 2026-10-27

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Wael Elaily, Markus Braun, David Stoll, Morakot Chakatok, Matteo Aleotti, Birgit Grill, Aleksandar Bijelic, Mélanie Hall, Gustav Oberdorfer

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 ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed.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

Enzymes are nature's master chemists, tiny protein machines that speed up chemical reactions with incredible precision. For centuries, scientists have tried to mimic this power by designing their own proteins from scratch, creating entirely new structures that do not exist in the natural world. One promising shape for these artificial proteins is the alpha-helical barrel, a structure made of twisted strands of protein wound together like a rope to form a hollow tube. These tubes are prized because they are incredibly sturdy, able to withstand high heat, and their simple, repeating patterns make them easier for computers to redesign. However, a major question has lingered: just how much can you change the instructions inside these artificial proteins before the whole structure falls apart? If a computer designs a perfect protein, can a scientist then rewrite large parts of its code to make it easier to produce in a factory, or to change its shape, without breaking its ability to work?

Researchers at Graz University of Technology and the University of Graz set out to answer this question by testing the limits of a specific, computer-designed protein called 6H5L. This protein is a six-stranded barrel that acts as a catalyst for a specific chemical reaction. The team used two different computer strategies to modify it. First, they used a powerful artificial intelligence tool to completely rewrite the surface code of the protein, hoping to make it easier to grow in bacteria. Second, they used a classic physics-based computer method to cut the protein shorter, creating a smaller version of the barrel. The goal was to see if these drastic changes would destroy the protein's shape or its ability to function, or if the structure was robust enough to survive the transformation.

The results showed that the artificial protein was far more resilient than expected. When the team used the artificial intelligence tool to rewrite the surface code, they found one version that worked remarkably well. This new variant, which had a completely different sequence of building blocks on its surface, was produced in bacteria at ten times the yield of the original design. Despite having a different chemical makeup, it folded into the correct barrel shape and remained stable even when heated to near boiling. It also kept its ability to speed up the chemical reaction, though it grabbed onto its target molecule slightly less tightly than the original. This suggests that the protein's core structure is flexible enough to tolerate massive changes in its outer instructions, a finding that could make it much easier to manufacture these artificial enzymes for industrial use.

In the second part of the study, the researchers tested whether they could shorten the barrel without breaking it. They used a computer model to trim the ends of the protein, creating a version that was significantly smaller. When they looked at the crystal structure of this shortened protein, it matched the computer prediction almost perfectly, with the six strands still forming a stable tube. Like the redesigned version, this truncated protein remained heat-stable and active. It even showed a slight improvement in how fast it could process its target molecule. The team also tested a second, even shorter version, but this one failed to fold correctly, indicating that there is a limit to how much the structure can be altered before it collapses.

The study concludes that these de novo protein scaffolds are highly engineerable. They can withstand dramatic changes in their genetic code and significant changes in their physical length while maintaining their structural integrity and catalytic power. By proving that these artificial barrels can be both redesigned for better production and reshaped for new functions, the researchers have opened the door to creating a wider variety of robust, custom-built enzymes. This work suggests that the future of biocatalysis may not just rely on copying nature, but on building entirely new, adaptable protein machines that can be tailored to solve specific chemical challenges in green chemistry and sustainable manufacturing.

Technical Summary: Probing the Sequence Variability Tolerance in a De Novo α-Helical Barrel Biocatalyst

Problem Statement
De novo-designed enzymes have demonstrated high catalytic activity and thermostability in novel protein scaffolds, yet a fundamental question remains regarding the "engineerability" of these designed backbone-sequence pairs. Specifically, it is unclear how much sequence variation and structural reengineering a functional de novo scaffold can withstand without losing its structural integrity, stability, or catalytic function. Furthermore, improving soluble protein production in E. coli for these designs remains a significant challenge. This study investigates these limitations using the catalytically active, six-helical barrel scaffold 6H5L as a model system.

Methodology
The authors employed a dual-strategy approach combining deep learning (DL) and classic physics-based computational design to modify the 6H5L scaffold:

Sequence Redesign (ProteinMPNN + AlphaFold2): The researchers utilized ProteinMPNN to redesign the surface residues and the unoptimized cavity 1 of the previously optimized variant 6H5L_RA1, while keeping the active site (cavity 2) fixed. AlphaFold2 (AF2) was used to filter designs based on high pLDDT confidence scores. Two variants were selected for experimental characterization: 6H5L_mpnn1740 (pI 5.4) and 6H5L_mpnn1409 (pI 7.18).
Structural Truncation (RosettaRemodel + AlphaFold2): To test the limits of structural diversification, RosettaRemodel was guided by AF2 to generate truncated versions of the 6H5L scaffold. This involved shortening the barrel to create open termini, resulting in two variants: 6H5L_Cut (180 amino acids) and 6H5L_Cut2 (162 amino acids).
Experimental Characterization: All variants were expressed in E. coli and purified. Their properties were assessed via:
- Solubility and Stability: SDS-PAGE, Size-Exclusion Chromatography (SEC), and Circular Dichroism (CD) spectroscopy with temperature scans.
- Folding Verification: DPH fluorescence assays to probe hydrophobic cavity integrity.
- Structural Validation: Small-Angle X-ray Scattering (SAXS) and X-ray Crystallography (specifically for 6H5L_Cut, PDB: 9R2B).
- Catalytic Activity: Retro-aldolase activity assays using a fluorescence-based detection method, inhibition assays to confirm lysine integrity, and whole-cell biotransformation experiments.

Key Results

Soluble Expression: The ProteinMPNN variant 6H5L_mpnn1740 exhibited a tenfold increase in soluble protein yield (16.7 mg/L) compared to the reference 6H5L_RA1 (1.7 mg/L). In contrast, 6H5L_mpnn1409 showed poor expression and signs of misfolding. Among the truncated variants, 6H5L_Cut produced a monodisperse peak, while 6H5L_Cut2 aggregated, correlating with lower AF2 confidence scores.
Structural Integrity:
- Both 6H5L_mpnn1740 and 6H5L_Cut retained the characteristic α-helical barrel architecture and high thermal stability (stable up to 95°C).
- Crystal Structure (6H5L_Cut): The crystal structure closely matched the AF2 design model (Cα RMSD of 2.61 Å). While the overall barrel architecture was preserved, the crystal structure revealed an anisotropic deformation where helix H5 tilted significantly (6.3° vs. 1.9° in the model), expanding the bundle along one axis and constricting it along another. Crucially, the electrostatic interaction network of the central charged residues remained intact.
- DPH Assay: 6H5L_mpnn1740 showed a saturation profile similar to the parent, though with a higher dissociation constant ( $K_D$ ), suggesting partial obstruction of cavity 1. 6H5L_Cut showed altered binding behavior but maintained micromolar affinity, whereas 6H5L_Cut2 showed no binding, confirming misfolding.
Catalytic Performance:
- 6H5L_mpnn1740: Retained retro-aldolase activity but with a lower overall catalytic efficiency ( $k_{cat}/K_m$ ) than 6H5L_RA1. This was driven by a higher $k_{cat}$ (48.45 vs. 27.31 $\times 10^{-4}$ min $^{-1}$ ) but a significantly higher $K_m$ (1580 vs. 530 µM), likely due to the redesign of cavity 1 affecting substrate binding.
- 6H5L_Cut: Showed a catalytic efficiency ( $k_{cat}/K_m$ = 2.38 M $^{-1}$ min $^{-1}$ ) slightly higher than the original 6H5L (1.96 M $^{-1}$ min $^{-1}$ ), attributed to a higher $k_{cat}$ while maintaining comparable substrate affinity. The open termini appear to allow for two accessible binding pockets.
- Whole-Cell Systems: Both successful variants demonstrated catalytic activity in whole-cell biotransformation, consuming substrate and producing products at levels nearly double that of enzyme-free controls.

Significance and Claims
The paper claims that de novo α-helical barrel scaffolds possess a high degree of sequence variability tolerance. The study demonstrates that:

Substantial sequence changes (via ProteinMPNN) can be introduced to optimize soluble expression (10-fold increase) without compromising the global fold or catalytic mechanism.
Major structural reengineering (via RosettaRemodel truncation) is possible, allowing for the creation of shorter barrels with open termini that retain thermostability and catalytic function.
Synergy of Methods: Deep learning (ProteinMPNN/AF2) and physics-based methods (RosettaRemodel) can complement each other to engineer robust, high-yield de novo biocatalysts.
Predictive Limits: The failure of 6H5L_Cut2 and 6H5L_mpnn1409 suggests that while AF2 is a powerful filter, confidence scores below a certain threshold (e.g., mean pLDDT < 90 or specific low-confidence regions) may not reliably predict successful experimental folding and stability.

Ultimately, the work provides new insights into the engineering potential of de novo scaffolds, validating their ability to withstand dramatic sequence and structural changes while maintaining function, thereby expanding their applicability in biocatalysis and sustainable chemistry.

More like this