Researchers at the Broad Institute of Harvard and MIT have combined artificial intelligence with laboratory protein evolution to redesign the enzyme behind Botox, according to a press release issued by the institute. Led by David Liu, the team used the AI model ProteinMPNN—developed by Nobel laureate David Baker and colleagues at the University of Washington—to generate significantly more stable variants of natural botulinum neurotoxin proteases. These AI-redesigned enzymes proved far more stable and specific at cutting a target protein linked to neurodegeneration than their natural counterparts, overcoming a long-standing evolutionary bottleneck in synthetic biology.
The Evolutionary Bottleneck of Natural Proteins
Proteins serve as the body’s chemical workhorses by speeding up vital reactions, converting nutrients into energy, breaking down toxins in the liver, and powering gene editing tools. To build or improve these enzymes, scientists traditionally rely on directed evolution, a tedious laboratory process that nudges natural proteins through successive generations of mutations. However, natural enzymes often collapse and lose stability during this process. As proteins mutate to gain new functions, their complex structures frequently warp, causing them to clump together uselessly rather than docking precisely with their intended molecular targets, according to Liu’s lab team.
While researchers have previously attempted workarounds such as adding protein-folding chaperones or pre-evolving natural enzymes for stability, these methods introduce heavy layers of cost, time, and labor. “Laboratory evolution requires the commitment of time and resources. So what you start with is incredibly important as a major determinant of what you end up with,” Liu said in a press release. Without stable starting points, scientists repeatedly hit walls where evolving new therapeutic capabilities causes the underlying protein scaffolding to fail.
Did you know? In 2011, Liu’s lab reported a high-throughput continuous evolution system called PACE, which grows bacteriophages in vessels to run dozens of rounds of protein evolution a day without human intervention.
Redesigning Molecular Scissors with AI
To bypass traditional instability, Liu and his colleagues turned to AI models capable of predicting and designing protein structures from raw molecular sequences in seconds. The team focused on natural botulinum neurotoxin proteases, the molecular scissors that paralyze muscles by snipping specific proteins and serve as the active component in Botox. Using ProteinMPNN, the researchers generated 58 distinct designs predicted to possess higher structural stability than wild-type proteins.
When produced in E. coli bacteria, the top three AI-generated candidate designs proved highly soluble, avoiding the intracellular clumping that plagues natural variants. Some even demonstrated higher baseline activity than their natural counterparts. According to study author Nicholas Krasnow, “If you start with a more stable protein, it has more stability to spare, so it can afford larger changes in pursuit of new functions.” This structural flexibility allows engineered enzymes to tolerate a much higher rate of beneficial mutations without losing their functional integrity.
Targeting Neurodegeneration and ALS Proteins
The team fed their AI-redesigned enzymes into the PACE continuous evolution system, pushing the molecules to slice away a mutated region of a protein associated with neuron health. In neurodegenerative conditions like amyotrophic lateral sclerosis (ALS), repetitive stretches of proteins expand, form toxic aggregates, and gradually damage neurons. Naturally occurring enzymes have historically struggled to cut these mutant proteins before toxic clumps form.
Compared to enzymes evolved directly from wild-type botulinum neurotoxins, the AI-descended variants achieved remarkable performance improvements. According to the research findings, the redesigned enzymes were nearly 80 times more efficient at cutting the target protein and over 56 times more selective for the intended disease-associated region. Mathematical mapping of their evolutionary pathways confirmed that the AI starting points successfully tolerated larger genetic changes to acquire entirely new therapeutic functions.
Frequently Asked Questions
What is directed evolution in protein engineering?
Directed evolution is a laboratory process that mimics natural selection to nudge enzymes toward new, tailored properties through successive rounds of mutation and screening.
Why do natural enzymes fail during laboratory evolution?
As natural enzymes mutate to acquire new functions, their structures often warp and lose stability, causing the proteins to clump together and stop working properly, according to the research team.
How does AI help stabilize proteins?
AI models like ProteinMPNN analyze underlying molecular sequences to dream up new protein sequences that preserve an enzyme’s overall folding structure while altering its building blocks for enhanced stability.
What did the Broad Institute researchers achieve with AI-designed enzymes?
According to the published study, the team redesigned botulinum neurotoxin proteases using AI to create variants that were nearly 80 times more efficient at cutting target proteins linked to neurodegeneration.
Explore More: Dive deeper into our coverage of synthetic biology advancements and gene editing breakthroughs. Subscribe to our newsletter to receive weekly updates on cutting-edge biotechnology research directly in your inbox.
Keep reading