Designing new enzymes around catalytic arrays with Riff-Diff

Markus Braun, Adrian Tripp, and colleagues from the Oberdorfer lab at Graz University of Technology developed Riff-Diff, a computational pipeline that designs enzymes by scaffolding protein backbones around catalytic residue arrays. Read the full paper in Nature paper here. 

Why this research matters

Designing enzymes from scratch could help tackle unmet needs in biocatalysis, therapeutics, and environmental chemistry. But building proteins that not only fold correctly, but also catalyze a specific reaction, is a longstanding challenge. This study explores whether starting from the chemistry—specifically the catalytic residues—and designing everything else around them can produce working enzymes on the first try.

Background

Researchers often begin with a known catalytic motif, but integrating that motif into a stable protein scaffold with a functional active site is complex. Traditional design strategies rely on multiple rounds of mutation, modeling, and experimental screening. Riff-Diff aims to simplify this process by designing backbones that support the required chemistry from the outset.

The new approach

The Riff-Diff pipeline begins with a predefined catalytic array and embeds each residue into a short helical fragment to create an artificial motif. These motifs are combined and used to guide structure generation via a diffusion model, which includes a placeholder helix to help form a substrate-binding pocket.

After backbone generation, designs are refined through a series of steps—including sequence design, structure optimization, and model prediction—before selecting final candidates based on structure quality and active-site geometry.

Key results

Two enzyme classes were targeted: retro-aldolases and Morita–Baylis–Hillmanases (MBHases). Out of 36 retro-aldolase designs (RAD1–36), 32 showed activity above a negative control. Among these, RAD29 and RAD35 stood out for their catalytic efficiency and substrate affinity.

For the MBHases, 63 designs were generated using two evolved catalytic arrays. A majority showed measurable activity in endpoint assays. MBH48, one of the top performers, surpassed the activity of an 8-round evolved MBHase variant (BH32.8), though remained below the best evolved enzymes.

Validation

Biochemical assays confirmed that most designs folded properly and showed the expected dependence on active-site residues. For select enzymes, including MBH48, crystal structures closely matched the design models, both in overall fold and catalytic geometry.

Broader significance

Riff-Diff demonstrates that it’s possible to build functional enzymes by scaffolding around idealized catalytic arrays, without iterative rounds of screening or evolution. The authors highlight this approach as a way to streamline enzyme discovery and generate high-quality benchmarks for improving future predictive models.