Evolution naturally produces the fittest living things for a given environment, but labs can speed up that process to uncover how cells respond to specific pressures. These directed evolution experiments typically rely on genetically simple organisms like bacteria or yeast, but a new system pushes that paradigm into mammalian cells—notoriously difficult targets for new drugs and therapeutics.
Scientists at the University of Sydney’s Neely Lab and the Centenary Institute spent three years building a virus-inspired biological machine, PROTEUS (PROTein Evolution Using Selection), an open-source platform that could enable more effective gene-editing tools, mRNA-based medicines with greater specificity, and therapeutic proteins engineered to avoid side effects. Each cycle takes 24 hours of diversification, selection, and amplification, shortening the protein evolution process from months or years to just weeks.
“PROTEUS can produce mammalian-specific adaptations that we wouldn’t have predicted or evolved in other systems,” says University of Sydney research fellow Christopher Denes, one of the study’s authors. The team evolved a transcription factor—a protein that activates the expression of a gene target—with mutations only beneficial in mammalian cells, not bacteria.
“The general method of directed evolution is cyclic, where we generate diversity in a gene of interest, select for fitness against a challenge, and amplify successful genetic variants to repeat the process. This is all done with the goal of directing a gene towards a desired outcome,” says Denes. “PROTEUS enables all three steps in a single round of evolution. We repeat these cycles over and over, sometimes playing with the intensity of selective pressure to really challenge our protein.”
In May, the team published a study in the open-access journal Nature Communications, making their data and sequences public as an academic resource.
How PROTEUS Works
PROTEUS resembles biological artificial intelligence, a relatively new field combining bioengineering with machine learning principles to create dynamic biological systems. But Denes conceptualizes PROTEUS more as a “biological trial-and-error machine” that adapts to the user’s demands. Unlike AI, the system doesn’t need a defined problem or goal to evolve. It naturally generates genetic diversity, thanks to its foundation in error-prone RNA viruses.
“Think of this as similar to how SARS-CoV-2 [the virus that causes COVID-19] adapted and evolved through variants from Alpha through to Delta and now Omicron, but the selection pressure applied by our genetic problem filters out any ‘bad’ variants, amplifying the good,” Denes says. “In the next round, these good variants can continue to become better or might even introduce a single bad mutation that cancels out the benefit.”
This tube contains 15 billion virus-like vesicles—small sacs of noninfectious particles replicated in a lab. Each carries a mutated gene.Tian Du
PROTEUS starts by engineering a single gene sequence for the protein the researcher wants to evolve into the genome of a virus. The genome is then introduced into mammalian cells with a shell-like packaging element to produce virus-like vesicles (VLVs), small sacs of noninfectious particles. The VLVs are then added to cells that carry a defined synthetic circuit, or a genetic problem the protein has been tasked to solve.
“Within 24 hours, these VLVs copy their genome in readiness to grow, but this copying step frequently makes mistakes, introducing mutations along the gene we want to evolve. Our gene is converted into protein and then challenged within cells by the genetic problem at hand,” Denes explains. “If it succeeds in solving this puzzle, it’ll produce more of itself. We’ve effectively linked together protein fitness with VLV survival, pretty much enforcing the survival of the fittest.”
Across hundreds of thousands to millions of cells, Denes says PROTEUS can quickly produce millions of possible genetic variants—all selected for in parallel. It can also be scaled for more cells, providing broader diversity.
Why Are Mammalian Cells So Difficult?
Many proteins evolved in typical directed evolution environments—yeast or bacteria—don’t translate well to human cells.
Mammalian cells are tricky evolution hosts. They grow slowly, contain large genomes, and regulate protein behavior through unpredictable “post-translational modifications”—diverse chemical changes that occur after a protein is built from mRNA. That complexity makes conventional tools too imprecise to isolate and manipulate their genetic material.
Mammalian directed evolution also requires massive cell-population sizes to cover all the potential mutations necessary to test a desired variant, says Kate Adamala, an associate professor in the University of Minnesota’s Department of Genetics, Cell Biology, and Development. “Here, [with PROTEUS] they managed to circumvent mammalian cells’ natural intolerance to high mutation rates, which is important because if you want to test a lot of variants, you need a huge mutation rate.”
As a general-purpose synthetic biologist practicing artificial evolution herself, though not with mammalian cells, Adamala was excited about the findings. “I’ve been following it because I’m looking for ways to make artificial evolution more biomedically applicable, and that means we have to start poking at complex mammalian systems,” Adamala says. “If my lab had PROTEUS, I’d start with membrane proteins, because that’s a huge area and incredibly attractive from a human health perspective and as a foundational research question of understanding how drugs interact with membranes.”
This is the predicted 3D structure of an evolved protein, featuring mutation sites [red], displaced functional groups [blue], and the drug-binding area [green circle].
Alexander Cole, Christopher Denes, Cesar Moreno, et al.
Many existing diseases and drugs are related to membrane proteins, like opioids, weight-loss medications, and viral resistance. “There are a lot of targets on a membrane’s surface, and evolving those proteins has been a pain in the lower back because they’re difficult to work with and existing processes are slow,” Adamala says.
Through the not-for-profit plasmid repository Addgene, PROTEUS will be available to any lab worldwide with the appropriate infrastructure, virology training, and skills in molecular biology. According to Denes, the evolution process costs a few thousand dollars and is cost-effective in campaigns for multiple genes in parallel.
PROTEUS’s lead researchers filed a provisional patent application in Australia. Denes says the team is exploring commercialization pathways.
Neely Lab is interested in applying the technology to gene editing, having previously used CRISPR to understand the mechanisms behind how venoms cause cell death and pain, and how proteins bind to cells, including the crucial spike protein in SARS-CoV-2.
“Evolved CRISPR tools would be really valuable for both research and medicine,” Denes says, citing a recent study that used CRISPR-based gene editing in a human baby to treat a rare genetic disease within eight months from diagnosis at birth.
“The gene editor they used was actually derived through a bacteria-based directed-evolution method, so there’s huge potential in evolving these editors further in mammalian cells with PROTEUS to produce proteins with boosted activity in human disease treatment,” Denes adds.